projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: always check dir caps asynchronously
[linux-2.6-block.git] / fs / ceph / mds_client.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
2f2dc053 3
496e5955 4#include <linux/fs.h>
2f2dc053 5#include <linux/wait.h>
5a0e3ad6 6#include <linux/slab.h>
54008399 7#include <linux/gfp.h>
2f2dc053 8#include <linux/sched.h>
3d14c5d2
YS		 9#include <linux/debugfs.h>
10#include <linux/seq_file.h>
3e0708b9 11#include <linux/ratelimit.h>
9ba1e224 12#include <linux/bits.h>
70c94820 13#include <linux/ktime.h>
d517b398 14#include <linux/bitmap.h>
9c2df227 15#include <linux/mnt_idmapping.h>
2f2dc053 16
2f2dc053 17#include "super.h"
3d14c5d2 18#include "mds_client.h"
2d332d5b 19#include "crypto.h"
3d14c5d2 20
1fe60e51 21#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 22#include <linux/ceph/messenger.h>
23#include <linux/ceph/decode.h>
24#include <linux/ceph/pagelist.h>
25#include <linux/ceph/auth.h>
26#include <linux/ceph/debugfs.h>
2f2dc053 27
81c5a148
YZ		 28#define RECONNECT_MAX_SIZE (INT_MAX - PAGE_SIZE)
29
2f2dc053
SW		 30/*
31 * A cluster of MDS (metadata server) daemons is responsible for
32 * managing the file system namespace (the directory hierarchy and
33 * inodes) and for coordinating shared access to storage. Metadata is
34 * partitioning hierarchically across a number of servers, and that
35 * partition varies over time as the cluster adjusts the distribution
36 * in order to balance load.
37 *
38 * The MDS client is primarily responsible to managing synchronous
39 * metadata requests for operations like open, unlink, and so forth.
40 * If there is a MDS failure, we find out about it when we (possibly
41 * request and) receive a new MDS map, and can resubmit affected
42 * requests.
43 *
44 * For the most part, though, we take advantage of a lossless
45 * communications channel to the MDS, and do not need to worry about
46 * timing out or resubmitting requests.
47 *
48 * We maintain a stateful "session" with each MDS we interact with.
49 * Within each session, we sent periodic heartbeat messages to ensure
50 * any capabilities or leases we have been issues remain valid. If
51 * the session times out and goes stale, our leases and capabilities
52 * are no longer valid.
53 */
54
20cb34ae 55struct ceph_reconnect_state {
81c5a148
YZ		 56 struct ceph_mds_session *session;
57 int nr_caps, nr_realms;
20cb34ae 58 struct ceph_pagelist *pagelist;
121f22a1 59 unsigned msg_version;
81c5a148 60 bool allow_multi;
20cb34ae
SW		 61};
62
2f2dc053
SW		 63static void __wake_requests(struct ceph_mds_client *mdsc,
64 struct list_head *head);
e3ec8d68 65static void ceph_cap_release_work(struct work_struct *work);
37c4efc1 66static void ceph_cap_reclaim_work(struct work_struct *work);
2f2dc053 67
9e32789f 68static const struct ceph_connection_operations mds_con_ops;
2f2dc053
SW		 69
70
71/*
72 * mds reply parsing
73 */
74
b37fe1f9
YZ		 75static int parse_reply_info_quota(void **p, void *end,
76 struct ceph_mds_reply_info_in *info)
77{
78 u8 struct_v, struct_compat;
79 u32 struct_len;
80
81 ceph_decode_8_safe(p, end, struct_v, bad);
82 ceph_decode_8_safe(p, end, struct_compat, bad);
83 /* struct_v is expected to be >= 1. we only
84 * understand encoding with struct_compat == 1. */
85 if (!struct_v || struct_compat != 1)
86 goto bad;
87 ceph_decode_32_safe(p, end, struct_len, bad);
88 ceph_decode_need(p, end, struct_len, bad);
89 end = *p + struct_len;
90 ceph_decode_64_safe(p, end, info->max_bytes, bad);
91 ceph_decode_64_safe(p, end, info->max_files, bad);
92 *p = end;
93 return 0;
94bad:
95 return -EIO;
96}
97
2f2dc053
SW		 98/*
99 * parse individual inode info
100 */
101static int parse_reply_info_in(void **p, void *end,
14303d20 102 struct ceph_mds_reply_info_in *info,
12b4629a 103 u64 features)
2f2dc053 104{
b37fe1f9
YZ		 105 int err = 0;
106 u8 struct_v = 0;
2f2dc053 107
b37fe1f9
YZ		 108 if (features == (u64)-1) {
109 u32 struct_len;
110 u8 struct_compat;
111 ceph_decode_8_safe(p, end, struct_v, bad);
112 ceph_decode_8_safe(p, end, struct_compat, bad);
113 /* struct_v is expected to be >= 1. we only understand
114 * encoding with struct_compat == 1. */
115 if (!struct_v || struct_compat != 1)
116 goto bad;
117 ceph_decode_32_safe(p, end, struct_len, bad);
118 ceph_decode_need(p, end, struct_len, bad);
119 end = *p + struct_len;
120 }
121
122 ceph_decode_need(p, end, sizeof(struct ceph_mds_reply_inode), bad);
2f2dc053
SW		 123 info->in = *p;
124 *p += sizeof(struct ceph_mds_reply_inode) +
125 sizeof(*info->in->fragtree.splits) *
126 le32_to_cpu(info->in->fragtree.nsplits);
127
128 ceph_decode_32_safe(p, end, info->symlink_len, bad);
129 ceph_decode_need(p, end, info->symlink_len, bad);
130 info->symlink = *p;
131 *p += info->symlink_len;
132
23c625ce
ID		 133 ceph_decode_copy_safe(p, end, &info->dir_layout,
134 sizeof(info->dir_layout), bad);
2f2dc053
SW		 135 ceph_decode_32_safe(p, end, info->xattr_len, bad);
136 ceph_decode_need(p, end, info->xattr_len, bad);
137 info->xattr_data = *p;
138 *p += info->xattr_len;
fb01d1f8 139
b37fe1f9
YZ		 140 if (features == (u64)-1) {
141 /* inline data */
fb01d1f8
YZ		 142 ceph_decode_64_safe(p, end, info->inline_version, bad);
143 ceph_decode_32_safe(p, end, info->inline_len, bad);
144 ceph_decode_need(p, end, info->inline_len, bad);
145 info->inline_data = *p;
146 *p += info->inline_len;
b37fe1f9
YZ		 147 /* quota */
148 err = parse_reply_info_quota(p, end, info);
149 if (err < 0)
150 goto out_bad;
151 /* pool namespace */
152 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
153 if (info->pool_ns_len > 0) {
154 ceph_decode_need(p, end, info->pool_ns_len, bad);
155 info->pool_ns_data = *p;
156 *p += info->pool_ns_len;
157 }
245ce991
JL		 158
159 /* btime */
160 ceph_decode_need(p, end, sizeof(info->btime), bad);
161 ceph_decode_copy(p, &info->btime, sizeof(info->btime));
162
163 /* change attribute */
a35ead31 164 ceph_decode_64_safe(p, end, info->change_attr, bad);
fb01d1f8 165
08796873
YZ		 166 /* dir pin */
167 if (struct_v >= 2) {
168 ceph_decode_32_safe(p, end, info->dir_pin, bad);
169 } else {
170 info->dir_pin = -ENODATA;
171 }
172
193e7b37
DD		 173 /* snapshot birth time, remains zero for v<=2 */
174 if (struct_v >= 3) {
175 ceph_decode_need(p, end, sizeof(info->snap_btime), bad);
176 ceph_decode_copy(p, &info->snap_btime,
177 sizeof(info->snap_btime));
178 } else {
179 memset(&info->snap_btime, 0, sizeof(info->snap_btime));
180 }
181
e7f72952
YC		 182 /* snapshot count, remains zero for v<=3 */
183 if (struct_v >= 4) {
184 ceph_decode_64_safe(p, end, info->rsnaps, bad);
185 } else {
186 info->rsnaps = 0;
187 }
188
2d332d5b
JL		 189 if (struct_v >= 5) {
190 u32 alen;
191
192 ceph_decode_32_safe(p, end, alen, bad);
193
194 while (alen--) {
195 u32 len;
196
197 /* key */
198 ceph_decode_32_safe(p, end, len, bad);
199 ceph_decode_skip_n(p, end, len, bad);
200 /* value */
201 ceph_decode_32_safe(p, end, len, bad);
202 ceph_decode_skip_n(p, end, len, bad);
203 }
204 }
205
206 /* fscrypt flag -- ignore */
207 if (struct_v >= 6)
208 ceph_decode_skip_8(p, end, bad);
209
210 info->fscrypt_auth = NULL;
211 info->fscrypt_auth_len = 0;
212 info->fscrypt_file = NULL;
213 info->fscrypt_file_len = 0;
214 if (struct_v >= 7) {
215 ceph_decode_32_safe(p, end, info->fscrypt_auth_len, bad);
216 if (info->fscrypt_auth_len) {
217 info->fscrypt_auth = kmalloc(info->fscrypt_auth_len,
218 GFP_KERNEL);
219 if (!info->fscrypt_auth)
220 return -ENOMEM;
221 ceph_decode_copy_safe(p, end, info->fscrypt_auth,
222 info->fscrypt_auth_len, bad);
223 }
224 ceph_decode_32_safe(p, end, info->fscrypt_file_len, bad);
225 if (info->fscrypt_file_len) {
226 info->fscrypt_file = kmalloc(info->fscrypt_file_len,
227 GFP_KERNEL);
228 if (!info->fscrypt_file)
229 return -ENOMEM;
230 ceph_decode_copy_safe(p, end, info->fscrypt_file,
231 info->fscrypt_file_len, bad);
232 }
233 }
b37fe1f9
YZ		 234 *p = end;
235 } else {
2d332d5b 236 /* legacy (unversioned) struct */
b37fe1f9
YZ		 237 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
238 ceph_decode_64_safe(p, end, info->inline_version, bad);
239 ceph_decode_32_safe(p, end, info->inline_len, bad);
240 ceph_decode_need(p, end, info->inline_len, bad);
241 info->inline_data = *p;
242 *p += info->inline_len;
243 } else
244 info->inline_version = CEPH_INLINE_NONE;
245
246 if (features & CEPH_FEATURE_MDS_QUOTA) {
247 err = parse_reply_info_quota(p, end, info);
248 if (err < 0)
249 goto out_bad;
250 } else {
251 info->max_bytes = 0;
252 info->max_files = 0;
253 }
254
255 info->pool_ns_len = 0;
256 info->pool_ns_data = NULL;
257 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
258 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
259 if (info->pool_ns_len > 0) {
260 ceph_decode_need(p, end, info->pool_ns_len, bad);
261 info->pool_ns_data = *p;
262 *p += info->pool_ns_len;
263 }
264 }
08796873 265
245ce991
JL		 266 if (features & CEPH_FEATURE_FS_BTIME) {
267 ceph_decode_need(p, end, sizeof(info->btime), bad);
268 ceph_decode_copy(p, &info->btime, sizeof(info->btime));
a35ead31 269 ceph_decode_64_safe(p, end, info->change_attr, bad);
245ce991
JL		 270 }
271
08796873 272 info->dir_pin = -ENODATA;
e7f72952 273 /* info->snap_btime and info->rsnaps remain zero */
b37fe1f9
YZ		 274 }
275 return 0;
276bad:
277 err = -EIO;
278out_bad:
279 return err;
280}
281
282static int parse_reply_info_dir(void **p, void *end,
283 struct ceph_mds_reply_dirfrag **dirfrag,
284 u64 features)
285{
286 if (features == (u64)-1) {
fb18a575
LH		 287 u8 struct_v, struct_compat;
288 u32 struct_len;
fb18a575
LH		 289 ceph_decode_8_safe(p, end, struct_v, bad);
290 ceph_decode_8_safe(p, end, struct_compat, bad);
b37fe1f9
YZ		 291 /* struct_v is expected to be >= 1. we only understand
292 * encoding whose struct_compat == 1. */
293 if (!struct_v || struct_compat != 1)
fb18a575
LH		 294 goto bad;
295 ceph_decode_32_safe(p, end, struct_len, bad);
296 ceph_decode_need(p, end, struct_len, bad);
b37fe1f9 297 end = *p + struct_len;
fb18a575
LH		 298 }
299
b37fe1f9
YZ		 300 ceph_decode_need(p, end, sizeof(**dirfrag), bad);
301 *dirfrag = *p;
302 *p += sizeof(**dirfrag) + sizeof(u32) * le32_to_cpu((*dirfrag)->ndist);
303 if (unlikely(*p > end))
304 goto bad;
305 if (features == (u64)-1)
306 *p = end;
307 return 0;
308bad:
309 return -EIO;
310}
311
312static int parse_reply_info_lease(void **p, void *end,
313 struct ceph_mds_reply_lease **lease,
4ac4c23e 314 u64 features, u32 *altname_len, u8 **altname)
b37fe1f9 315{
4ac4c23e
JL		 316 u8 struct_v;
317 u32 struct_len;
318 void *lend;
319
b37fe1f9 320 if (features == (u64)-1) {
4ac4c23e
JL		 321 u8 struct_compat;
322
b37fe1f9
YZ		 323 ceph_decode_8_safe(p, end, struct_v, bad);
324 ceph_decode_8_safe(p, end, struct_compat, bad);
4ac4c23e 325
b37fe1f9
YZ		 326 /* struct_v is expected to be >= 1. we only understand
327 * encoding whose struct_compat == 1. */
328 if (!struct_v || struct_compat != 1)
329 goto bad;
4ac4c23e 330
b37fe1f9 331 ceph_decode_32_safe(p, end, struct_len, bad);
4ac4c23e
JL		 332 } else {
333 struct_len = sizeof(**lease);
334 *altname_len = 0;
335 *altname = NULL;
5ea5c5e0
YZ		 336 }
337
4ac4c23e
JL		 338 lend = *p + struct_len;
339 ceph_decode_need(p, end, struct_len, bad);
b37fe1f9
YZ		 340 *lease = *p;
341 *p += sizeof(**lease);
4ac4c23e
JL		 342
343 if (features == (u64)-1) {
344 if (struct_v >= 2) {
345 ceph_decode_32_safe(p, end, *altname_len, bad);
346 ceph_decode_need(p, end, *altname_len, bad);
347 *altname = *p;
348 *p += *altname_len;
349 } else {
350 *altname = NULL;
351 *altname_len = 0;
352 }
353 }
354 *p = lend;
2f2dc053
SW		 355 return 0;
356bad:
b37fe1f9 357 return -EIO;
2f2dc053
SW		 358}
359
360/*
361 * parse a normal reply, which may contain a (dir+)dentry and/or a
362 * target inode.
363 */
364static int parse_reply_info_trace(void **p, void *end,
14303d20 365 struct ceph_mds_reply_info_parsed *info,
12b4629a 366 u64 features)
2f2dc053
SW		 367{
368 int err;
369
370 if (info->head->is_dentry) {
14303d20 371 err = parse_reply_info_in(p, end, &info->diri, features);
2f2dc053
SW		 372 if (err < 0)
373 goto out_bad;
374
b37fe1f9
YZ		 375 err = parse_reply_info_dir(p, end, &info->dirfrag, features);
376 if (err < 0)
377 goto out_bad;
2f2dc053
SW		 378
379 ceph_decode_32_safe(p, end, info->dname_len, bad);
380 ceph_decode_need(p, end, info->dname_len, bad);
381 info->dname = *p;
382 *p += info->dname_len;
b37fe1f9 383
4ac4c23e
JL		 384 err = parse_reply_info_lease(p, end, &info->dlease, features,
385 &info->altname_len, &info->altname);
b37fe1f9
YZ		 386 if (err < 0)
387 goto out_bad;
2f2dc053
SW		 388 }
389
390 if (info->head->is_target) {
14303d20 391 err = parse_reply_info_in(p, end, &info->targeti, features);
2f2dc053
SW		 392 if (err < 0)
393 goto out_bad;
394 }
395
396 if (unlikely(*p != end))
397 goto bad;
398 return 0;
399
400bad:
401 err = -EIO;
402out_bad:
403 pr_err("problem parsing mds trace %d\n", err);
404 return err;
405}
406
407/*
408 * parse readdir results
409 */
b37fe1f9 410static int parse_reply_info_readdir(void **p, void *end,
3859af9e
XL		 411 struct ceph_mds_request *req,
412 u64 features)
2f2dc053 413{
3859af9e 414 struct ceph_mds_reply_info_parsed *info = &req->r_reply_info;
38d46409 415 struct ceph_client *cl = req->r_mdsc->fsc->client;
2f2dc053
SW		 416 u32 num, i = 0;
417 int err;
418
b37fe1f9
YZ		 419 err = parse_reply_info_dir(p, end, &info->dir_dir, features);
420 if (err < 0)
421 goto out_bad;
2f2dc053
SW		 422
423 ceph_decode_need(p, end, sizeof(num) + 2, bad);
c89136ea 424 num = ceph_decode_32(p);
956d39d6
YZ		 425 {
426 u16 flags = ceph_decode_16(p);
427 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
428 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
f3c4ebe6 429 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
79162547 430 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
956d39d6 431 }
2f2dc053
SW		 432 if (num == 0)
433 goto done;
434
2a5beea3
YZ		 435 BUG_ON(!info->dir_entries);
436 if ((unsigned long)(info->dir_entries + num) >
437 (unsigned long)info->dir_entries + info->dir_buf_size) {
38d46409 438 pr_err_client(cl, "dir contents are larger than expected\n");
54008399
YZ		 439 WARN_ON(1);
440 goto bad;
441 }
2f2dc053 442
54008399 443 info->dir_nr = num;
2f2dc053 444 while (num) {
af9ffa6d
XL		 445 struct inode *inode = d_inode(req->r_dentry);
446 struct ceph_inode_info *ci = ceph_inode(inode);
2a5beea3 447 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
af9ffa6d
XL		 448 struct fscrypt_str tname = FSTR_INIT(NULL, 0);
449 struct fscrypt_str oname = FSTR_INIT(NULL, 0);
450 struct ceph_fname fname;
451 u32 altname_len, _name_len;
452 u8 *altname, *_name;
453
2f2dc053 454 /* dentry */
af9ffa6d
XL		 455 ceph_decode_32_safe(p, end, _name_len, bad);
456 ceph_decode_need(p, end, _name_len, bad);
457 _name = *p;
458 *p += _name_len;
38d46409 459 doutc(cl, "parsed dir dname '%.*s'\n", _name_len, _name);
af9ffa6d
XL		 460
461 if (info->hash_order)
462 rde->raw_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
463 _name, _name_len);
2f2dc053 464
b37fe1f9 465 /* dentry lease */
4ac4c23e 466 err = parse_reply_info_lease(p, end, &rde->lease, features,
af9ffa6d 467 &altname_len, &altname);
b37fe1f9
YZ		 468 if (err)
469 goto out_bad;
4ac4c23e 470
af9ffa6d
XL		 471 /*
472 * Try to dencrypt the dentry names and update them
473 * in the ceph_mds_reply_dir_entry struct.
474 */
475 fname.dir = inode;
476 fname.name = _name;
477 fname.name_len = _name_len;
478 fname.ctext = altname;
479 fname.ctext_len = altname_len;
480 /*
481 * The _name_len maybe larger than altname_len, such as
482 * when the human readable name length is in range of
483 * (CEPH_NOHASH_NAME_MAX, CEPH_NOHASH_NAME_MAX + SHA256_DIGEST_SIZE),
484 * then the copy in ceph_fname_to_usr will corrupt the
485 * data if there has no encryption key.
486 *
487 * Just set the no_copy flag and then if there has no
488 * encryption key the oname.name will be assigned to
489 * _name always.
490 */
491 fname.no_copy = true;
492 if (altname_len == 0) {
493 /*
494 * Set tname to _name, and this will be used
495 * to do the base64_decode in-place. It's
496 * safe because the decoded string should
497 * always be shorter, which is 3/4 of origin
498 * string.
499 */
500 tname.name = _name;
501
502 /*
503 * Set oname to _name too, and this will be
504 * used to do the dencryption in-place.
505 */
506 oname.name = _name;
507 oname.len = _name_len;
508 } else {
509 /*
510 * This will do the decryption only in-place
511 * from altname cryptext directly.
512 */
513 oname.name = altname;
514 oname.len = altname_len;
515 }
516 rde->is_nokey = false;
517 err = ceph_fname_to_usr(&fname, &tname, &oname, &rde->is_nokey);
518 if (err) {
38d46409
XL		 519 pr_err_client(cl, "unable to decode %.*s, got %d\n",
520 _name_len, _name, err);
af9ffa6d
XL		 521 goto out_bad;
522 }
523 rde->name = oname.name;
524 rde->name_len = oname.len;
525
2f2dc053 526 /* inode */
2a5beea3 527 err = parse_reply_info_in(p, end, &rde->inode, features);
2f2dc053
SW		 528 if (err < 0)
529 goto out_bad;
8974eebd
YZ		 530 /* ceph_readdir_prepopulate() will update it */
531 rde->offset = 0;
2f2dc053
SW		 532 i++;
533 num--;
534 }
535
536done:
1d3f8723
JL		 537 /* Skip over any unrecognized fields */
538 *p = end;
2f2dc053
SW		 539 return 0;
540
541bad:
542 err = -EIO;
543out_bad:
38d46409 544 pr_err_client(cl, "problem parsing dir contents %d\n", err);
2f2dc053
SW		 545 return err;
546}
547
25933abd
HS		 548/*
549 * parse fcntl F_GETLK results
550 */
551static int parse_reply_info_filelock(void **p, void *end,
14303d20 552 struct ceph_mds_reply_info_parsed *info,
12b4629a 553 u64 features)
25933abd
HS		 554{
555 if (*p + sizeof(*info->filelock_reply) > end)
556 goto bad;
557
558 info->filelock_reply = *p;
25933abd 559
1d3f8723
JL		 560 /* Skip over any unrecognized fields */
561 *p = end;
25933abd 562 return 0;
25933abd
HS		 563bad:
564 return -EIO;
565}
566
d4846487
JL		 567
568#if BITS_PER_LONG == 64
569
570#define DELEGATED_INO_AVAILABLE xa_mk_value(1)
571
572static int ceph_parse_deleg_inos(void **p, void *end,
573 struct ceph_mds_session *s)
574{
38d46409 575 struct ceph_client *cl = s->s_mdsc->fsc->client;
d4846487
JL		 576 u32 sets;
577
578 ceph_decode_32_safe(p, end, sets, bad);
38d46409 579 doutc(cl, "got %u sets of delegated inodes\n", sets);
d4846487 580 while (sets--) {
2ecd0edd 581 u64 start, len;
d4846487
JL		 582
583 ceph_decode_64_safe(p, end, start, bad);
584 ceph_decode_64_safe(p, end, len, bad);
d4f6b31d
JL		 585
586 /* Don't accept a delegation of system inodes */
587 if (start < CEPH_INO_SYSTEM_BASE) {
38d46409
XL		 588 pr_warn_ratelimited_client(cl,
589 "ignoring reserved inode range delegation (start=0x%llx len=0x%llx)\n",
590 start, len);
d4f6b31d
JL		 591 continue;
592 }
d4846487 593 while (len--) {
2ecd0edd 594 int err = xa_insert(&s->s_delegated_inos, start++,
d4846487
JL		 595 DELEGATED_INO_AVAILABLE,
596 GFP_KERNEL);
597 if (!err) {
38d46409 598 doutc(cl, "added delegated inode 0x%llx\n", start - 1);
d4846487 599 } else if (err == -EBUSY) {
38d46409
XL		 600 pr_warn_client(cl,
601 "MDS delegated inode 0x%llx more than once.\n",
d4846487
JL		 602 start - 1);
603 } else {
604 return err;
605 }
606 }
607 }
608 return 0;
609bad:
610 return -EIO;
611}
612
613u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
614{
615 unsigned long ino;
616 void *val;
617
618 xa_for_each(&s->s_delegated_inos, ino, val) {
619 val = xa_erase(&s->s_delegated_inos, ino);
620 if (val == DELEGATED_INO_AVAILABLE)
621 return ino;
622 }
623 return 0;
624}
625
626int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
627{
628 return xa_insert(&s->s_delegated_inos, ino, DELEGATED_INO_AVAILABLE,
629 GFP_KERNEL);
630}
631#else /* BITS_PER_LONG == 64 */
632/*
633 * FIXME: xarrays can't handle 64-bit indexes on a 32-bit arch. For now, just
634 * ignore delegated_inos on 32 bit arch. Maybe eventually add xarrays for top
635 * and bottom words?
636 */
637static int ceph_parse_deleg_inos(void **p, void *end,
638 struct ceph_mds_session *s)
639{
640 u32 sets;
641
642 ceph_decode_32_safe(p, end, sets, bad);
643 if (sets)
644 ceph_decode_skip_n(p, end, sets * 2 * sizeof(__le64), bad);
645 return 0;
646bad:
647 return -EIO;
648}
649
650u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
651{
652 return 0;
653}
654
655int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
656{
657 return 0;
658}
659#endif /* BITS_PER_LONG == 64 */
660
6e8575fa
SL		 661/*
662 * parse create results
663 */
664static int parse_reply_info_create(void **p, void *end,
665 struct ceph_mds_reply_info_parsed *info,
d4846487 666 u64 features, struct ceph_mds_session *s)
6e8575fa 667{
d4846487
JL		 668 int ret;
669
b37fe1f9
YZ		 670 if (features == (u64)-1 ||
671 (features & CEPH_FEATURE_REPLY_CREATE_INODE)) {
6e8575fa 672 if (*p == end) {
d4846487 673 /* Malformed reply? */
6e8575fa 674 info->has_create_ino = false;
d4846487 675 } else if (test_bit(CEPHFS_FEATURE_DELEG_INO, &s->s_features)) {
6e8575fa 676 info->has_create_ino = true;
06a1ad43
JL		 677 /* struct_v, struct_compat, and len */
678 ceph_decode_skip_n(p, end, 2 + sizeof(u32), bad);
d4846487
JL		 679 ceph_decode_64_safe(p, end, info->ino, bad);
680 ret = ceph_parse_deleg_inos(p, end, s);
681 if (ret)
682 return ret;
683 } else {
684 /* legacy */
1d3f8723 685 ceph_decode_64_safe(p, end, info->ino, bad);
d4846487 686 info->has_create_ino = true;
6e8575fa 687 }
1d3f8723
JL		 688 } else {
689 if (*p != end)
690 goto bad;
6e8575fa
SL		 691 }
692
1d3f8723
JL		 693 /* Skip over any unrecognized fields */
694 *p = end;
6e8575fa 695 return 0;
6e8575fa
SL		 696bad:
697 return -EIO;
698}
699
6ddf5f16
MC		 700static int parse_reply_info_getvxattr(void **p, void *end,
701 struct ceph_mds_reply_info_parsed *info,
702 u64 features)
703{
704 u32 value_len;
705
706 ceph_decode_skip_8(p, end, bad); /* skip current version: 1 */
707 ceph_decode_skip_8(p, end, bad); /* skip first version: 1 */
708 ceph_decode_skip_32(p, end, bad); /* skip payload length */
709
710 ceph_decode_32_safe(p, end, value_len, bad);
711
712 if (value_len == end - *p) {
713 info->xattr_info.xattr_value = *p;
714 info->xattr_info.xattr_value_len = value_len;
715 *p = end;
716 return value_len;
717 }
718bad:
719 return -EIO;
720}
721
25933abd
HS		 722/*
723 * parse extra results
724 */
725static int parse_reply_info_extra(void **p, void *end,
3859af9e 726 struct ceph_mds_request *req,
d4846487 727 u64 features, struct ceph_mds_session *s)
25933abd 728{
3859af9e 729 struct ceph_mds_reply_info_parsed *info = &req->r_reply_info;
6df8c9d8
JL		 730 u32 op = le32_to_cpu(info->head->op);
731
732 if (op == CEPH_MDS_OP_GETFILELOCK)
14303d20 733 return parse_reply_info_filelock(p, end, info, features);
6df8c9d8 734 else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
3859af9e 735 return parse_reply_info_readdir(p, end, req, features);
6df8c9d8 736 else if (op == CEPH_MDS_OP_CREATE)
d4846487 737 return parse_reply_info_create(p, end, info, features, s);
6ddf5f16
MC		 738 else if (op == CEPH_MDS_OP_GETVXATTR)
739 return parse_reply_info_getvxattr(p, end, info, features);
6e8575fa
SL		 740 else
741 return -EIO;
25933abd
HS		 742}
743
2f2dc053
SW		 744/*
745 * parse entire mds reply
746 */
d4846487 747static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,
3859af9e 748 struct ceph_mds_request *req, u64 features)
2f2dc053 749{
3859af9e 750 struct ceph_mds_reply_info_parsed *info = &req->r_reply_info;
38d46409 751 struct ceph_client *cl = s->s_mdsc->fsc->client;
2f2dc053
SW		 752 void *p, *end;
753 u32 len;
754 int err;
755
756 info->head = msg->front.iov_base;
757 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
758 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
759
760 /* trace */
761 ceph_decode_32_safe(&p, end, len, bad);
762 if (len > 0) {
32852a81 763 ceph_decode_need(&p, end, len, bad);
14303d20 764 err = parse_reply_info_trace(&p, p+len, info, features);
2f2dc053
SW		 765 if (err < 0)
766 goto out_bad;
767 }
768
25933abd 769 /* extra */
2f2dc053
SW		 770 ceph_decode_32_safe(&p, end, len, bad);
771 if (len > 0) {
32852a81 772 ceph_decode_need(&p, end, len, bad);
3859af9e 773 err = parse_reply_info_extra(&p, p+len, req, features, s);
2f2dc053
SW		 774 if (err < 0)
775 goto out_bad;
776 }
777
778 /* snap blob */
779 ceph_decode_32_safe(&p, end, len, bad);
780 info->snapblob_len = len;
781 info->snapblob = p;
782 p += len;
783
784 if (p != end)
785 goto bad;
786 return 0;
787
788bad:
789 err = -EIO;
790out_bad:
38d46409 791 pr_err_client(cl, "mds parse_reply err %d\n", err);
8b0da5c5 792 ceph_msg_dump(msg);
2f2dc053
SW		 793 return err;
794}
795
796static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
797{
2d332d5b
JL		 798 int i;
799
800 kfree(info->diri.fscrypt_auth);
801 kfree(info->diri.fscrypt_file);
802 kfree(info->targeti.fscrypt_auth);
803 kfree(info->targeti.fscrypt_file);
2a5beea3 804 if (!info->dir_entries)
54008399 805 return;
2d332d5b
JL		 806
807 for (i = 0; i < info->dir_nr; i++) {
808 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
809
810 kfree(rde->inode.fscrypt_auth);
811 kfree(rde->inode.fscrypt_file);
812 }
2a5beea3 813 free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
2f2dc053
SW		 814}
815
4868e537
XL		 816/*
817 * In async unlink case the kclient won't wait for the first reply
818 * from MDS and just drop all the links and unhash the dentry and then
819 * succeeds immediately.
820 *
821 * For any new create/link/rename,etc requests followed by using the
822 * same file names we must wait for the first reply of the inflight
823 * unlink request, or the MDS possibly will fail these following
824 * requests with -EEXIST if the inflight async unlink request was
825 * delayed for some reasons.
826 *
827 * And the worst case is that for the none async openc request it will
828 * successfully open the file if the CDentry hasn't been unlinked yet,
829 * but later the previous delayed async unlink request will remove the
830 * CDenty. That means the just created file is possiblly deleted later
831 * by accident.
832 *
833 * We need to wait for the inflight async unlink requests to finish
834 * when creating new files/directories by using the same file names.
835 */
836int ceph_wait_on_conflict_unlink(struct dentry *dentry)
837{
5995d90d 838 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dentry->d_sb);
38d46409 839 struct ceph_client *cl = fsc->client;
4868e537
XL		 840 struct dentry *pdentry = dentry->d_parent;
841 struct dentry *udentry, *found = NULL;
842 struct ceph_dentry_info *di;
843 struct qstr dname;
844 u32 hash = dentry->d_name.hash;
845 int err;
846
847 dname.name = dentry->d_name.name;
848 dname.len = dentry->d_name.len;
849
850 rcu_read_lock();
851 hash_for_each_possible_rcu(fsc->async_unlink_conflict, di,
852 hnode, hash) {
853 udentry = di->dentry;
854
855 spin_lock(&udentry->d_lock);
856 if (udentry->d_name.hash != hash)
857 goto next;
858 if (unlikely(udentry->d_parent != pdentry))
859 goto next;
860 if (!hash_hashed(&di->hnode))
861 goto next;
862
863 if (!test_bit(CEPH_DENTRY_ASYNC_UNLINK_BIT, &di->flags))
38d46409
XL		 864 pr_warn_client(cl, "dentry %p:%pd async unlink bit is not set\n",
865 dentry, dentry);
4868e537
XL		 866
867 if (!d_same_name(udentry, pdentry, &dname))
868 goto next;
869
dc32464a 870 found = dget_dlock(udentry);
4868e537 871 spin_unlock(&udentry->d_lock);
4868e537
XL		 872 break;
873next:
874 spin_unlock(&udentry->d_lock);
875 }
876 rcu_read_unlock();
877
878 if (likely(!found))
879 return 0;
880
38d46409
XL		 881 doutc(cl, "dentry %p:%pd conflict with old %p:%pd\n", dentry, dentry,
882 found, found);
4868e537
XL		 883
884 err = wait_on_bit(&di->flags, CEPH_DENTRY_ASYNC_UNLINK_BIT,
885 TASK_KILLABLE);
886 dput(found);
887 return err;
888}
889
2f2dc053
SW		 890
891/*
892 * sessions
893 */
a687ecaf 894const char *ceph_session_state_name(int s)
2f2dc053
SW		 895{
896 switch (s) {
897 case CEPH_MDS_SESSION_NEW: return "new";
898 case CEPH_MDS_SESSION_OPENING: return "opening";
899 case CEPH_MDS_SESSION_OPEN: return "open";
900 case CEPH_MDS_SESSION_HUNG: return "hung";
901 case CEPH_MDS_SESSION_CLOSING: return "closing";
4d681c2f 902 case CEPH_MDS_SESSION_CLOSED: return "closed";
44ca18f2 903 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053 904 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
fcff415c 905 case CEPH_MDS_SESSION_REJECTED: return "rejected";
2f2dc053
SW		 906 default: return "???";
907 }
908}
909
5b3248c6 910struct ceph_mds_session *ceph_get_mds_session(struct ceph_mds_session *s)
2f2dc053 911{
9f358999 912 if (refcount_inc_not_zero(&s->s_ref))
2f2dc053 913 return s;
9f358999 914 return NULL;
2f2dc053
SW		 915}
916
917void ceph_put_mds_session(struct ceph_mds_session *s)
918{
7e65624d
JL		 919 if (IS_ERR_OR_NULL(s))
920 return;
921
3997c01d 922 if (refcount_dec_and_test(&s->s_ref)) {
6c4a1915 923 if (s->s_auth.authorizer)
6c1ea260 924 ceph_auth_destroy_authorizer(s->s_auth.authorizer);
88828190 925 WARN_ON(mutex_is_locked(&s->s_mutex));
d4846487 926 xa_destroy(&s->s_delegated_inos);
2f2dc053 927 kfree(s);
4e7a5dcd 928 }
2f2dc053
SW		 929}
930
931/*
932 * called under mdsc->mutex
933 */
934struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
935 int mds)
936{
d37b1d99 937 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
2f2dc053 938 return NULL;
5b3248c6 939 return ceph_get_mds_session(mdsc->sessions[mds]);
2f2dc053
SW		 940}
941
942static bool __have_session(struct ceph_mds_client *mdsc, int mds)
943{
98cfda81 944 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
2f2dc053 945 return false;
98cfda81
CX		 946 else
947 return true;
2f2dc053
SW		 948}
949
2600d2dd
SW		 950static int __verify_registered_session(struct ceph_mds_client *mdsc,
951 struct ceph_mds_session *s)
952{
953 if (s->s_mds >= mdsc->max_sessions ||
954 mdsc->sessions[s->s_mds] != s)
955 return -ENOENT;
956 return 0;
957}
958
2f2dc053
SW		 959/*
960 * create+register a new session for given mds.
961 * called under mdsc->mutex.
962 */
963static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
964 int mds)
965{
38d46409 966 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 967 struct ceph_mds_session *s;
968
a68e564a
XL		 969 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO)
970 return ERR_PTR(-EIO);
971
b38c9eb4 972 if (mds >= mdsc->mdsmap->possible_max_rank)
c338c07c
NY		 973 return ERR_PTR(-EINVAL);
974
2f2dc053 975 s = kzalloc(sizeof(*s), GFP_NOFS);
4736b009
DC		 976 if (!s)
977 return ERR_PTR(-ENOMEM);
47474d0b
CX		 978
979 if (mds >= mdsc->max_sessions) {
980 int newmax = 1 << get_count_order(mds + 1);
981 struct ceph_mds_session **sa;
982
38d46409 983 doutc(cl, "realloc to %d\n", newmax);
47474d0b
CX		 984 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
985 if (!sa)
986 goto fail_realloc;
987 if (mdsc->sessions) {
988 memcpy(sa, mdsc->sessions,
989 mdsc->max_sessions * sizeof(void *));
990 kfree(mdsc->sessions);
991 }
992 mdsc->sessions = sa;
993 mdsc->max_sessions = newmax;
994 }
995
38d46409 996 doutc(cl, "mds%d\n", mds);
2f2dc053
SW		 997 s->s_mdsc = mdsc;
998 s->s_mds = mds;
999 s->s_state = CEPH_MDS_SESSION_NEW;
2f2dc053
SW		 1000 mutex_init(&s->s_mutex);
1001
b7a9e5dd 1002 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
2f2dc053 1003
52d60f8e 1004 atomic_set(&s->s_cap_gen, 1);
1ce208a6 1005 s->s_cap_ttl = jiffies - 1;
d8fb02ab
AE		 1006
1007 spin_lock_init(&s->s_cap_lock);
2f2dc053 1008 INIT_LIST_HEAD(&s->s_caps);
3997c01d 1009 refcount_set(&s->s_ref, 1);
2f2dc053
SW		 1010 INIT_LIST_HEAD(&s->s_waiting);
1011 INIT_LIST_HEAD(&s->s_unsafe);
d4846487 1012 xa_init(&s->s_delegated_inos);
2f2dc053 1013 INIT_LIST_HEAD(&s->s_cap_releases);
e3ec8d68
YZ		 1014 INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
1015
1cf03a68 1016 INIT_LIST_HEAD(&s->s_cap_dirty);
2f2dc053 1017 INIT_LIST_HEAD(&s->s_cap_flushing);
2f2dc053 1018
2f2dc053 1019 mdsc->sessions[mds] = s;
86d8f67b 1020 atomic_inc(&mdsc->num_sessions);
3997c01d 1021 refcount_inc(&s->s_ref); /* one ref to sessions[], one to caller */
42ce56e5 1022
b7a9e5dd
SW		 1023 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
1024 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
42ce56e5 1025
2f2dc053 1026 return s;
42ce56e5
SW		 1027
1028fail_realloc:
1029 kfree(s);
1030 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 1031}
1032
1033/*
1034 * called under mdsc->mutex
1035 */
2600d2dd 1036static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 1037 struct ceph_mds_session *s)
2f2dc053 1038{
38d46409 1039 doutc(mdsc->fsc->client, "mds%d %p\n", s->s_mds, s);
2600d2dd 1040 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5
SW		 1041 mdsc->sessions[s->s_mds] = NULL;
1042 ceph_con_close(&s->s_con);
1043 ceph_put_mds_session(s);
86d8f67b 1044 atomic_dec(&mdsc->num_sessions);
2f2dc053
SW		 1045}
1046
1047/*
1048 * drop session refs in request.
1049 *
1050 * should be last request ref, or hold mdsc->mutex
1051 */
1052static void put_request_session(struct ceph_mds_request *req)
1053{
1054 if (req->r_session) {
1055 ceph_put_mds_session(req->r_session);
1056 req->r_session = NULL;
1057 }
1058}
1059
59b312f3
XL		 1060void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
1061 void (*cb)(struct ceph_mds_session *),
1062 bool check_state)
1063{
1064 int mds;
1065
1066 mutex_lock(&mdsc->mutex);
1067 for (mds = 0; mds < mdsc->max_sessions; ++mds) {
1068 struct ceph_mds_session *s;
1069
1070 s = __ceph_lookup_mds_session(mdsc, mds);
1071 if (!s)
1072 continue;
1073
1074 if (check_state && !check_session_state(s)) {
1075 ceph_put_mds_session(s);
1076 continue;
1077 }
1078
1079 mutex_unlock(&mdsc->mutex);
1080 cb(s);
1081 ceph_put_mds_session(s);
1082 mutex_lock(&mdsc->mutex);
1083 }
1084 mutex_unlock(&mdsc->mutex);
1085}
1086
153c8e6b 1087void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 1088{
153c8e6b
SW		 1089 struct ceph_mds_request *req = container_of(kref,
1090 struct ceph_mds_request,
1091 r_kref);
07045648 1092 ceph_mdsc_release_dir_caps_async(req);
54008399 1093 destroy_reply_info(&req->r_reply_info);
153c8e6b
SW		 1094 if (req->r_request)
1095 ceph_msg_put(req->r_request);
54008399 1096 if (req->r_reply)
153c8e6b 1097 ceph_msg_put(req->r_reply);
153c8e6b 1098 if (req->r_inode) {
41b02e1f 1099 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
23c2c76e 1100 iput(req->r_inode);
153c8e6b 1101 }
9c1c2b35 1102 if (req->r_parent) {
3dd69aab 1103 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
23c2c76e 1104 iput(req->r_parent);
9c1c2b35 1105 }
23c2c76e 1106 iput(req->r_target_inode);
ec9595c0 1107 iput(req->r_new_inode);
153c8e6b
SW		 1108 if (req->r_dentry)
1109 dput(req->r_dentry);
844d87c3
SW		 1110 if (req->r_old_dentry)
1111 dput(req->r_old_dentry);
1112 if (req->r_old_dentry_dir) {
41b02e1f
SW		 1113 /*
1114 * track (and drop pins for) r_old_dentry_dir
1115 * separately, since r_old_dentry's d_parent may have
1116 * changed between the dir mutex being dropped and
1117 * this request being freed.
1118 */
1119 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
1120 CEPH_CAP_PIN);
23c2c76e 1121 iput(req->r_old_dentry_dir);
2f2dc053 1122 }
153c8e6b
SW		 1123 kfree(req->r_path1);
1124 kfree(req->r_path2);
7fe0cdeb 1125 put_cred(req->r_cred);
9c2df227
CB		 1126 if (req->r_mnt_idmap)
1127 mnt_idmap_put(req->r_mnt_idmap);
25e6bae3
YZ		 1128 if (req->r_pagelist)
1129 ceph_pagelist_release(req->r_pagelist);
2d332d5b 1130 kfree(req->r_fscrypt_auth);
24865e75 1131 kfree(req->r_altname);
153c8e6b 1132 put_request_session(req);
37151668 1133 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
428138c9 1134 WARN_ON_ONCE(!list_empty(&req->r_wait));
058daab7 1135 kmem_cache_free(ceph_mds_request_cachep, req);
2f2dc053
SW		 1136}
1137
fcd00b68
ID		 1138DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
1139
2f2dc053
SW		 1140/*
1141 * lookup session, bump ref if found.
1142 *
1143 * called under mdsc->mutex.
1144 */
fcd00b68
ID		 1145static struct ceph_mds_request *
1146lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
2f2dc053
SW		 1147{
1148 struct ceph_mds_request *req;
44ca18f2 1149
fcd00b68
ID		 1150 req = lookup_request(&mdsc->request_tree, tid);
1151 if (req)
1152 ceph_mdsc_get_request(req);
44ca18f2 1153
fcd00b68 1154 return req;
2f2dc053
SW		 1155}
1156
1157/*
1158 * Register an in-flight request, and assign a tid. Link to directory
1159 * are modifying (if any).
1160 *
1161 * Called under mdsc->mutex.
1162 */
1163static void __register_request(struct ceph_mds_client *mdsc,
1164 struct ceph_mds_request *req,
1165 struct inode *dir)
1166{
38d46409 1167 struct ceph_client *cl = mdsc->fsc->client;
e30ee581
ZZ		 1168 int ret = 0;
1169
2f2dc053 1170 req->r_tid = ++mdsc->last_tid;
e30ee581
ZZ		 1171 if (req->r_num_caps) {
1172 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
1173 req->r_num_caps);
1174 if (ret < 0) {
38d46409
XL		 1175 pr_err_client(cl, "%p failed to reserve caps: %d\n",
1176 req, ret);
e30ee581
ZZ		 1177 /* set req->r_err to fail early from __do_request */
1178 req->r_err = ret;
1179 return;
1180 }
1181 }
38d46409 1182 doutc(cl, "%p tid %lld\n", req, req->r_tid);
2f2dc053 1183 ceph_mdsc_get_request(req);
fcd00b68 1184 insert_request(&mdsc->request_tree, req);
2f2dc053 1185
7fe0cdeb 1186 req->r_cred = get_current_cred();
9c2df227
CB		 1187 if (!req->r_mnt_idmap)
1188 req->r_mnt_idmap = &nop_mnt_idmap;
cb4276cc 1189
e8a7b8b1
YZ		 1190 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
1191 mdsc->oldest_tid = req->r_tid;
1192
2f2dc053 1193 if (dir) {
3db0a2fc
JL		 1194 struct ceph_inode_info *ci = ceph_inode(dir);
1195
3b663780 1196 ihold(dir);
2f2dc053 1197 req->r_unsafe_dir = dir;
3db0a2fc
JL		 1198 spin_lock(&ci->i_unsafe_lock);
1199 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
1200 spin_unlock(&ci->i_unsafe_lock);
2f2dc053
SW		 1201 }
1202}
1203
1204static void __unregister_request(struct ceph_mds_client *mdsc,
1205 struct ceph_mds_request *req)
1206{
38d46409 1207 doutc(mdsc->fsc->client, "%p tid %lld\n", req, req->r_tid);
e8a7b8b1 1208
df963ea8
JL		 1209 /* Never leave an unregistered request on an unsafe list! */
1210 list_del_init(&req->r_unsafe_item);
1211
e8a7b8b1
YZ		 1212 if (req->r_tid == mdsc->oldest_tid) {
1213 struct rb_node *p = rb_next(&req->r_node);
1214 mdsc->oldest_tid = 0;
1215 while (p) {
1216 struct ceph_mds_request *next_req =
1217 rb_entry(p, struct ceph_mds_request, r_node);
1218 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
1219 mdsc->oldest_tid = next_req->r_tid;
1220 break;
1221 }
1222 p = rb_next(p);
1223 }
1224 }
1225
fcd00b68 1226 erase_request(&mdsc->request_tree, req);
2f2dc053 1227
3db0a2fc 1228 if (req->r_unsafe_dir) {
2f2dc053 1229 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
2f2dc053
SW		 1230 spin_lock(&ci->i_unsafe_lock);
1231 list_del_init(&req->r_unsafe_dir_item);
1232 spin_unlock(&ci->i_unsafe_lock);
4c06ace8 1233 }
bc2de10d
JL		 1234 if (req->r_target_inode &&
1235 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
68cd5b4b
YZ		 1236 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
1237 spin_lock(&ci->i_unsafe_lock);
1238 list_del_init(&req->r_unsafe_target_item);
1239 spin_unlock(&ci->i_unsafe_lock);
1240 }
3b663780 1241
4c06ace8 1242 if (req->r_unsafe_dir) {
23c2c76e 1243 iput(req->r_unsafe_dir);
3b663780 1244 req->r_unsafe_dir = NULL;
2f2dc053 1245 }
94aa8ae1 1246
fc55d2c9
YZ		 1247 complete_all(&req->r_safe_completion);
1248
94aa8ae1 1249 ceph_mdsc_put_request(req);
2f2dc053
SW		 1250}
1251
30c71233
JL		 1252/*
1253 * Walk back up the dentry tree until we hit a dentry representing a
1254 * non-snapshot inode. We do this using the rcu_read_lock (which must be held
1255 * when calling this) to ensure that the objects won't disappear while we're
1256 * working with them. Once we hit a candidate dentry, we attempt to take a
1257 * reference to it, and return that as the result.
1258 */
f1075480
DC		 1259static struct inode *get_nonsnap_parent(struct dentry *dentry)
1260{
1261 struct inode *inode = NULL;
30c71233
JL		 1262
1263 while (dentry && !IS_ROOT(dentry)) {
1264 inode = d_inode_rcu(dentry);
1265 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
1266 break;
1267 dentry = dentry->d_parent;
1268 }
1269 if (inode)
1270 inode = igrab(inode);
1271 return inode;
1272}
1273
2f2dc053
SW		 1274/*
1275 * Choose mds to send request to next. If there is a hint set in the
1276 * request (e.g., due to a prior forward hint from the mds), use that.
1277 * Otherwise, consult frag tree and/or caps to identify the
1278 * appropriate mds. If all else fails, choose randomly.
1279 *
1280 * Called under mdsc->mutex.
1281 */
1282static int __choose_mds(struct ceph_mds_client *mdsc,
c4853e97
XL		 1283 struct ceph_mds_request *req,
1284 bool *random)
2f2dc053
SW		 1285{
1286 struct inode *inode;
1287 struct ceph_inode_info *ci;
1288 struct ceph_cap *cap;
1289 int mode = req->r_direct_mode;
1290 int mds = -1;
1291 u32 hash = req->r_direct_hash;
bc2de10d 1292 bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
38d46409 1293 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053 1294
c4853e97
XL		 1295 if (random)
1296 *random = false;
1297
2f2dc053
SW		 1298 /*
1299 * is there a specific mds we should try? ignore hint if we have
1300 * no session and the mds is not up (active or recovering).
1301 */
1302 if (req->r_resend_mds >= 0 &&
1303 (__have_session(mdsc, req->r_resend_mds) ||
1304 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
38d46409 1305 doutc(cl, "using resend_mds mds%d\n", req->r_resend_mds);
2f2dc053
SW		 1306 return req->r_resend_mds;
1307 }
1308
1309 if (mode == USE_RANDOM_MDS)
1310 goto random;
1311
1312 inode = NULL;
1313 if (req->r_inode) {
5d37ca14
YZ		 1314 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
1315 inode = req->r_inode;
1316 ihold(inode);
1317 } else {
38f340cc
YZ		 1318 /* req->r_dentry is non-null for LSSNAP request */
1319 rcu_read_lock();
1320 inode = get_nonsnap_parent(req->r_dentry);
1321 rcu_read_unlock();
38d46409
XL		 1322 doutc(cl, "using snapdir's parent %p %llx.%llx\n",
1323 inode, ceph_vinop(inode));
5d37ca14 1324 }
38f340cc 1325 } else if (req->r_dentry) {
d79698da 1326 /* ignore race with rename; old or new d_parent is okay */
30c71233
JL		 1327 struct dentry *parent;
1328 struct inode *dir;
1329
1330 rcu_read_lock();
41883ba8 1331 parent = READ_ONCE(req->r_dentry->d_parent);
3dd69aab 1332 dir = req->r_parent ? : d_inode_rcu(parent);
eb6bb1c5 1333
30c71233
JL		 1334 if (!dir || dir->i_sb != mdsc->fsc->sb) {
1335 /* not this fs or parent went negative */
2b0143b5 1336 inode = d_inode(req->r_dentry);
30c71233
JL		 1337 if (inode)
1338 ihold(inode);
eb6bb1c5
SW		 1339 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
1340 /* direct snapped/virtual snapdir requests
1341 * based on parent dir inode */
30c71233 1342 inode = get_nonsnap_parent(parent);
38d46409
XL		 1343 doutc(cl, "using nonsnap parent %p %llx.%llx\n",
1344 inode, ceph_vinop(inode));
ca18bede 1345 } else {
eb6bb1c5 1346 /* dentry target */
2b0143b5 1347 inode = d_inode(req->r_dentry);
ca18bede
YZ		 1348 if (!inode || mode == USE_AUTH_MDS) {
1349 /* dir + name */
30c71233 1350 inode = igrab(dir);
ca18bede
YZ		 1351 hash = ceph_dentry_hash(dir, req->r_dentry);
1352 is_hash = true;
30c71233
JL		 1353 } else {
1354 ihold(inode);
ca18bede 1355 }
2f2dc053 1356 }
30c71233 1357 rcu_read_unlock();
2f2dc053 1358 }
eb6bb1c5 1359
2f2dc053
SW		 1360 if (!inode)
1361 goto random;
38d46409
XL		 1362
1363 doutc(cl, "%p %llx.%llx is_hash=%d (0x%x) mode %d\n", inode,
1364 ceph_vinop(inode), (int)is_hash, hash, mode);
2f2dc053
SW		 1365 ci = ceph_inode(inode);
1366
1367 if (is_hash && S_ISDIR(inode->i_mode)) {
1368 struct ceph_inode_frag frag;
1369 int found;
1370
1371 ceph_choose_frag(ci, hash, &frag, &found);
1372 if (found) {
1373 if (mode == USE_ANY_MDS && frag.ndist > 0) {
1374 u8 r;
1375
1376 /* choose a random replica */
1377 get_random_bytes(&r, 1);
1378 r %= frag.ndist;
1379 mds = frag.dist[r];
38d46409
XL		 1380 doutc(cl, "%p %llx.%llx frag %u mds%d (%d/%d)\n",
1381 inode, ceph_vinop(inode), frag.frag,
1382 mds, (int)r, frag.ndist);
d66bbd44 1383 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
5d47648f
XL		 1384 CEPH_MDS_STATE_ACTIVE &&
1385 !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
30c71233 1386 goto out;
2f2dc053
SW		 1387 }
1388
1389 /* since this file/dir wasn't known to be
1390 * replicated, then we want to look for the
1391 * authoritative mds. */
2f2dc053
SW		 1392 if (frag.mds >= 0) {
1393 /* choose auth mds */
1394 mds = frag.mds;
38d46409
XL		 1395 doutc(cl, "%p %llx.%llx frag %u mds%d (auth)\n",
1396 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44 1397 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
5d47648f 1398 CEPH_MDS_STATE_ACTIVE) {
224c7b67 1399 if (!ceph_mdsmap_is_laggy(mdsc->mdsmap,
5d47648f
XL		 1400 mds))
1401 goto out;
1402 }
2f2dc053 1403 }
5d47648f 1404 mode = USE_AUTH_MDS;
2f2dc053
SW		 1405 }
1406 }
1407
be655596 1408 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1409 cap = NULL;
1410 if (mode == USE_AUTH_MDS)
1411 cap = ci->i_auth_cap;
1412 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1413 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1414 if (!cap) {
be655596 1415 spin_unlock(&ci->i_ceph_lock);
23c2c76e 1416 iput(inode);
2f2dc053
SW		 1417 goto random;
1418 }
1419 mds = cap->session->s_mds;
38d46409
XL		 1420 doutc(cl, "%p %llx.%llx mds%d (%scap %p)\n", inode,
1421 ceph_vinop(inode), mds,
1422 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 1423 spin_unlock(&ci->i_ceph_lock);
30c71233 1424out:
23c2c76e 1425 iput(inode);
2f2dc053
SW		 1426 return mds;
1427
1428random:
c4853e97
XL		 1429 if (random)
1430 *random = true;
1431
2f2dc053 1432 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
38d46409 1433 doutc(cl, "chose random mds%d\n", mds);
2f2dc053
SW		 1434 return mds;
1435}
1436
1437
1438/*
1439 * session messages
1440 */
fba97e80 1441struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq)
2f2dc053
SW		 1442{
1443 struct ceph_msg *msg;
1444 struct ceph_mds_session_head *h;
1445
b61c2763
SW		 1446 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1447 false);
a79832f2 1448 if (!msg) {
fba97e80
XL		 1449 pr_err("ENOMEM creating session %s msg\n",
1450 ceph_session_op_name(op));
a79832f2 1451 return NULL;
2f2dc053
SW		 1452 }
1453 h = msg->front.iov_base;
1454 h->op = cpu_to_le32(op);
1455 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 1456
1457 return msg;
1458}
1459
9ba1e224
XL		 1460static const unsigned char feature_bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1461#define FEATURE_BYTES(c) (DIV_ROUND_UP((size_t)feature_bits[c - 1] + 1, 64) * 8)
b682c6d4 1462static int encode_supported_features(void **p, void *end)
342ce182 1463{
9ba1e224 1464 static const size_t count = ARRAY_SIZE(feature_bits);
342ce182
YZ		 1465
1466 if (count > 0) {
1467 size_t i;
9ba1e224 1468 size_t size = FEATURE_BYTES(count);
fea013e0 1469 unsigned long bit;
342ce182 1470
b682c6d4
XL		 1471 if (WARN_ON_ONCE(*p + 4 + size > end))
1472 return -ERANGE;
1473
342ce182
YZ		 1474 ceph_encode_32(p, size);
1475 memset(*p, 0, size);
fea013e0
LH		 1476 for (i = 0; i < count; i++) {
1477 bit = feature_bits[i];
1478 ((unsigned char *)(*p))[bit / 8] |= BIT(bit % 8);
1479 }
342ce182
YZ		 1480 *p += size;
1481 } else {
b682c6d4
XL		 1482 if (WARN_ON_ONCE(*p + 4 > end))
1483 return -ERANGE;
1484
342ce182
YZ		 1485 ceph_encode_32(p, 0);
1486 }
b682c6d4
XL		 1487
1488 return 0;
342ce182
YZ		 1489}
1490
3b4168dd
XL		 1491static const unsigned char metric_bits[] = CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED;
1492#define METRIC_BYTES(cnt) (DIV_ROUND_UP((size_t)metric_bits[cnt - 1] + 1, 64) * 8)
1493static int encode_metric_spec(void **p, void *end)
1494{
1495 static const size_t count = ARRAY_SIZE(metric_bits);
1496
1497 /* header */
1498 if (WARN_ON_ONCE(*p + 2 > end))
1499 return -ERANGE;
1500
1501 ceph_encode_8(p, 1); /* version */
1502 ceph_encode_8(p, 1); /* compat */
1503
1504 if (count > 0) {
1505 size_t i;
1506 size_t size = METRIC_BYTES(count);
1507
1508 if (WARN_ON_ONCE(*p + 4 + 4 + size > end))
1509 return -ERANGE;
1510
1511 /* metric spec info length */
1512 ceph_encode_32(p, 4 + size);
1513
1514 /* metric spec */
1515 ceph_encode_32(p, size);
1516 memset(*p, 0, size);
1517 for (i = 0; i < count; i++)
1518 ((unsigned char *)(*p))[i / 8] |= BIT(metric_bits[i] % 8);
1519 *p += size;
1520 } else {
1521 if (WARN_ON_ONCE(*p + 4 + 4 > end))
1522 return -ERANGE;
1523
1524 /* metric spec info length */
1525 ceph_encode_32(p, 4);
1526 /* metric spec */
1527 ceph_encode_32(p, 0);
1528 }
1529
1530 return 0;
1531}
1532
dbd0c8bf
JS		 1533/*
1534 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1535 * to include additional client metadata fields.
1536 */
66207de3
XL		 1537static struct ceph_msg *
1538create_session_full_msg(struct ceph_mds_client *mdsc, int op, u64 seq)
dbd0c8bf
JS		 1539{
1540 struct ceph_msg *msg;
1541 struct ceph_mds_session_head *h;
4a756db2 1542 int i;
342ce182 1543 int extra_bytes = 0;
dbd0c8bf
JS		 1544 int metadata_key_count = 0;
1545 struct ceph_options *opt = mdsc->fsc->client->options;
3f384954 1546 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
38d46409 1547 struct ceph_client *cl = mdsc->fsc->client;
9ba1e224 1548 size_t size, count;
342ce182 1549 void *p, *end;
b682c6d4 1550 int ret;
dbd0c8bf 1551
a6a5ce4f 1552 const char* metadata[][2] = {
717e6f28
YZ		 1553 {"hostname", mdsc->nodename},
1554 {"kernel_version", init_utsname()->release},
3f384954
YZ		 1555 {"entity_id", opt->name ? : ""},
1556 {"root", fsopt->server_path ? : "/"},
dbd0c8bf
JS		 1557 {NULL, NULL}
1558 };
1559
1560 /* Calculate serialized length of metadata */
342ce182 1561 extra_bytes = 4; /* map length */
d37b1d99 1562 for (i = 0; metadata[i][0]; ++i) {
342ce182 1563 extra_bytes += 8 + strlen(metadata[i][0]) +
dbd0c8bf
JS		 1564 strlen(metadata[i][1]);
1565 metadata_key_count++;
1566 }
9ba1e224 1567
342ce182 1568 /* supported feature */
9ba1e224
XL		 1569 size = 0;
1570 count = ARRAY_SIZE(feature_bits);
1571 if (count > 0)
1572 size = FEATURE_BYTES(count);
1573 extra_bytes += 4 + size;
dbd0c8bf 1574
3b4168dd
XL		 1575 /* metric spec */
1576 size = 0;
1577 count = ARRAY_SIZE(metric_bits);
1578 if (count > 0)
1579 size = METRIC_BYTES(count);
1580 extra_bytes += 2 + 4 + 4 + size;
1581
6df89bf2
XL		 1582 /* flags, mds auth caps and oldest_client_tid */
1583 extra_bytes += 4 + 4 + 8;
1584
dbd0c8bf 1585 /* Allocate the message */
342ce182 1586 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
dbd0c8bf
JS		 1587 GFP_NOFS, false);
1588 if (!msg) {
38d46409 1589 pr_err_client(cl, "ENOMEM creating session open msg\n");
b682c6d4 1590 return ERR_PTR(-ENOMEM);
dbd0c8bf 1591 }
342ce182
YZ		 1592 p = msg->front.iov_base;
1593 end = p + msg->front.iov_len;
1594
1595 h = p;
66207de3 1596 h->op = cpu_to_le32(op);
dbd0c8bf
JS		 1597 h->seq = cpu_to_le64(seq);
1598
1599 /*
1600 * Serialize client metadata into waiting buffer space, using
1601 * the format that userspace expects for map<string, string>
7cfa0313 1602 *
6df89bf2 1603 * ClientSession messages with metadata are v7
dbd0c8bf 1604 */
6df89bf2 1605 msg->hdr.version = cpu_to_le16(7);
7cfa0313 1606 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 1607
1608 /* The write pointer, following the session_head structure */
342ce182 1609 p += sizeof(*h);
dbd0c8bf
JS		 1610
1611 /* Number of entries in the map */
1612 ceph_encode_32(&p, metadata_key_count);
1613
1614 /* Two length-prefixed strings for each entry in the map */
d37b1d99 1615 for (i = 0; metadata[i][0]; ++i) {
dbd0c8bf
JS		 1616 size_t const key_len = strlen(metadata[i][0]);
1617 size_t const val_len = strlen(metadata[i][1]);
1618
1619 ceph_encode_32(&p, key_len);
1620 memcpy(p, metadata[i][0], key_len);
1621 p += key_len;
1622 ceph_encode_32(&p, val_len);
1623 memcpy(p, metadata[i][1], val_len);
1624 p += val_len;
1625 }
1626
b682c6d4
XL		 1627 ret = encode_supported_features(&p, end);
1628 if (ret) {
38d46409 1629 pr_err_client(cl, "encode_supported_features failed!\n");
b682c6d4
XL		 1630 ceph_msg_put(msg);
1631 return ERR_PTR(ret);
1632 }
1633
3b4168dd
XL		 1634 ret = encode_metric_spec(&p, end);
1635 if (ret) {
38d46409 1636 pr_err_client(cl, "encode_metric_spec failed!\n");
3b4168dd
XL		 1637 ceph_msg_put(msg);
1638 return ERR_PTR(ret);
1639 }
1640
6df89bf2
XL		 1641 /* version == 5, flags */
1642 ceph_encode_32(&p, 0);
1643
1644 /* version == 6, mds auth caps */
1645 ceph_encode_32(&p, 0);
1646
1647 /* version == 7, oldest_client_tid */
1648 ceph_encode_64(&p, mdsc->oldest_tid);
1649
342ce182
YZ		 1650 msg->front.iov_len = p - msg->front.iov_base;
1651 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1652
2f2dc053
SW		 1653 return msg;
1654}
1655
1656/*
1657 * send session open request.
1658 *
1659 * called under mdsc->mutex
1660 */
1661static int __open_session(struct ceph_mds_client *mdsc,
1662 struct ceph_mds_session *session)
1663{
1664 struct ceph_msg *msg;
1665 int mstate;
1666 int mds = session->s_mds;
2f2dc053 1667
a68e564a
XL		 1668 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO)
1669 return -EIO;
1670
2f2dc053
SW		 1671 /* wait for mds to go active? */
1672 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
38d46409
XL		 1673 doutc(mdsc->fsc->client, "open_session to mds%d (%s)\n", mds,
1674 ceph_mds_state_name(mstate));
2f2dc053
SW		 1675 session->s_state = CEPH_MDS_SESSION_OPENING;
1676 session->s_renew_requested = jiffies;
1677
1678 /* send connect message */
66207de3
XL		 1679 msg = create_session_full_msg(mdsc, CEPH_SESSION_REQUEST_OPEN,
1680 session->s_seq);
b682c6d4
XL		 1681 if (IS_ERR(msg))
1682 return PTR_ERR(msg);
2f2dc053 1683 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 1684 return 0;
1685}
1686
ed0552a1
SW		 1687/*
1688 * open sessions for any export targets for the given mds
1689 *
1690 * called under mdsc->mutex
1691 */
5d72d13c
YZ		 1692static struct ceph_mds_session *
1693__open_export_target_session(struct ceph_mds_client *mdsc, int target)
1694{
1695 struct ceph_mds_session *session;
b682c6d4 1696 int ret;
5d72d13c
YZ		 1697
1698 session = __ceph_lookup_mds_session(mdsc, target);
1699 if (!session) {
1700 session = register_session(mdsc, target);
1701 if (IS_ERR(session))
1702 return session;
1703 }
1704 if (session->s_state == CEPH_MDS_SESSION_NEW ||
b682c6d4
XL		 1705 session->s_state == CEPH_MDS_SESSION_CLOSING) {
1706 ret = __open_session(mdsc, session);
1707 if (ret)
1708 return ERR_PTR(ret);
1709 }
5d72d13c
YZ		 1710
1711 return session;
1712}
1713
1714struct ceph_mds_session *
1715ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1716{
1717 struct ceph_mds_session *session;
38d46409 1718 struct ceph_client *cl = mdsc->fsc->client;
5d72d13c 1719
38d46409 1720 doutc(cl, "to mds%d\n", target);
5d72d13c
YZ		 1721
1722 mutex_lock(&mdsc->mutex);
1723 session = __open_export_target_session(mdsc, target);
1724 mutex_unlock(&mdsc->mutex);
1725
1726 return session;
1727}
1728
ed0552a1
SW		 1729static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1730 struct ceph_mds_session *session)
1731{
1732 struct ceph_mds_info *mi;
1733 struct ceph_mds_session *ts;
1734 int i, mds = session->s_mds;
38d46409 1735 struct ceph_client *cl = mdsc->fsc->client;
ed0552a1 1736
b38c9eb4 1737 if (mds >= mdsc->mdsmap->possible_max_rank)
ed0552a1 1738 return;
5d72d13c 1739
ed0552a1 1740 mi = &mdsc->mdsmap->m_info[mds];
38d46409
XL		 1741 doutc(cl, "for mds%d (%d targets)\n", session->s_mds,
1742 mi->num_export_targets);
ed0552a1
SW		 1743
1744 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c 1745 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
7e65624d 1746 ceph_put_mds_session(ts);
ed0552a1
SW		 1747 }
1748}
1749
154f42c2
SW		 1750void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1751 struct ceph_mds_session *session)
1752{
1753 mutex_lock(&mdsc->mutex);
1754 __open_export_target_sessions(mdsc, session);
1755 mutex_unlock(&mdsc->mutex);
1756}
1757
2f2dc053
SW		 1758/*
1759 * session caps
1760 */
1761
c8a96a31
JL		 1762static void detach_cap_releases(struct ceph_mds_session *session,
1763 struct list_head *target)
2f2dc053 1764{
38d46409
XL		 1765 struct ceph_client *cl = session->s_mdsc->fsc->client;
1766
c8a96a31
JL		 1767 lockdep_assert_held(&session->s_cap_lock);
1768
1769 list_splice_init(&session->s_cap_releases, target);
745a8e3b 1770 session->s_num_cap_releases = 0;
38d46409 1771 doutc(cl, "mds%d\n", session->s_mds);
c8a96a31 1772}
2f2dc053 1773
c8a96a31
JL		 1774static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1775 struct list_head *dispose)
1776{
1777 while (!list_empty(dispose)) {
745a8e3b
YZ		 1778 struct ceph_cap *cap;
1779 /* zero out the in-progress message */
c8a96a31 1780 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
745a8e3b
YZ		 1781 list_del(&cap->session_caps);
1782 ceph_put_cap(mdsc, cap);
2f2dc053 1783 }
2f2dc053
SW		 1784}
1785
1c841a96
YZ		 1786static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1787 struct ceph_mds_session *session)
1788{
38d46409 1789 struct ceph_client *cl = mdsc->fsc->client;
1c841a96
YZ		 1790 struct ceph_mds_request *req;
1791 struct rb_node *p;
1792
38d46409 1793 doutc(cl, "mds%d\n", session->s_mds);
1c841a96
YZ		 1794 mutex_lock(&mdsc->mutex);
1795 while (!list_empty(&session->s_unsafe)) {
1796 req = list_first_entry(&session->s_unsafe,
1797 struct ceph_mds_request, r_unsafe_item);
38d46409
XL		 1798 pr_warn_ratelimited_client(cl, " dropping unsafe request %llu\n",
1799 req->r_tid);
1bd85aa6
JL		 1800 if (req->r_target_inode)
1801 mapping_set_error(req->r_target_inode->i_mapping, -EIO);
1802 if (req->r_unsafe_dir)
1803 mapping_set_error(req->r_unsafe_dir->i_mapping, -EIO);
1c841a96
YZ		 1804 __unregister_request(mdsc, req);
1805 }
1806 /* zero r_attempts, so kick_requests() will re-send requests */
1807 p = rb_first(&mdsc->request_tree);
1808 while (p) {
1809 req = rb_entry(p, struct ceph_mds_request, r_node);
1810 p = rb_next(p);
1811 if (req->r_session &&
1812 req->r_session->s_mds == session->s_mds)
1813 req->r_attempts = 0;
1814 }
1815 mutex_unlock(&mdsc->mutex);
1816}
1817
2f2dc053 1818/*
f818a736
SW		 1819 * Helper to safely iterate over all caps associated with a session, with
1820 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1821 *
f818a736 1822 * Caller must hold session s_mutex.
2f2dc053 1823 */
f5d77269 1824int ceph_iterate_session_caps(struct ceph_mds_session *session,
aaf67de7
XL		 1825 int (*cb)(struct inode *, int mds, void *),
1826 void *arg)
2f2dc053 1827{
38d46409 1828 struct ceph_client *cl = session->s_mdsc->fsc->client;
7c1332b8
SW		 1829 struct list_head *p;
1830 struct ceph_cap *cap;
1831 struct inode *inode, *last_inode = NULL;
1832 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1833 int ret;
1834
38d46409 1835 doutc(cl, "%p mds%d\n", session, session->s_mds);
2f2dc053 1836 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1837 p = session->s_caps.next;
1838 while (p != &session->s_caps) {
aaf67de7
XL		 1839 int mds;
1840
7c1332b8 1841 cap = list_entry(p, struct ceph_cap, session_caps);
874c8ca1 1842 inode = igrab(&cap->ci->netfs.inode);
7c1332b8
SW		 1843 if (!inode) {
1844 p = p->next;
2f2dc053 1845 continue;
7c1332b8
SW		 1846 }
1847 session->s_cap_iterator = cap;
aaf67de7 1848 mds = cap->mds;
2f2dc053 1849 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1850
1851 if (last_inode) {
23c2c76e 1852 iput(last_inode);
7c1332b8
SW		 1853 last_inode = NULL;
1854 }
1855 if (old_cap) {
37151668 1856 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1857 old_cap = NULL;
1858 }
1859
aaf67de7 1860 ret = cb(inode, mds, arg);
7c1332b8
SW		 1861 last_inode = inode;
1862
2f2dc053 1863 spin_lock(&session->s_cap_lock);
7c1332b8 1864 p = p->next;
d37b1d99 1865 if (!cap->ci) {
38d46409 1866 doutc(cl, "finishing cap %p removal\n", cap);
7c1332b8 1867 BUG_ON(cap->session != session);
745a8e3b 1868 cap->session = NULL;
7c1332b8
SW		 1869 list_del_init(&cap->session_caps);
1870 session->s_nr_caps--;
4f1d756d 1871 atomic64_dec(&session->s_mdsc->metric.total_caps);
e3ec8d68
YZ		 1872 if (cap->queue_release)
1873 __ceph_queue_cap_release(session, cap);
1874 else
745a8e3b 1875 old_cap = cap; /* put_cap it w/o locks held */
7c1332b8 1876 }
5dacf091
SW		 1877 if (ret < 0)
1878 goto out;
2f2dc053 1879 }
5dacf091
SW		 1880 ret = 0;
1881out:
7c1332b8 1882 session->s_cap_iterator = NULL;
2f2dc053 1883 spin_unlock(&session->s_cap_lock);
7c1332b8 1884
23c2c76e 1885 iput(last_inode);
7c1332b8 1886 if (old_cap)
37151668 1887 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1888
5dacf091 1889 return ret;
2f2dc053
SW		 1890}
1891
aaf67de7 1892static int remove_session_caps_cb(struct inode *inode, int mds, void *arg)
2f2dc053
SW		 1893{
1894 struct ceph_inode_info *ci = ceph_inode(inode);
38d46409 1895 struct ceph_client *cl = ceph_inode_to_client(inode);
6c93df5d 1896 bool invalidate = false;
aaf67de7
XL		 1897 struct ceph_cap *cap;
1898 int iputs = 0;
6c99f254 1899
be655596 1900 spin_lock(&ci->i_ceph_lock);
aaf67de7
XL		 1901 cap = __get_cap_for_mds(ci, mds);
1902 if (cap) {
38d46409
XL		 1903 doutc(cl, " removing cap %p, ci is %p, inode is %p\n",
1904 cap, ci, &ci->netfs.inode);
aaf67de7
XL		 1905
1906 iputs = ceph_purge_inode_cap(inode, cap, &invalidate);
1907 }
be655596 1908 spin_unlock(&ci->i_ceph_lock);
77310320 1909
aaf67de7
XL		 1910 if (cap)
1911 wake_up_all(&ci->i_cap_wq);
6c93df5d
YZ		 1912 if (invalidate)
1913 ceph_queue_invalidate(inode);
36e6da98 1914 while (iputs--)
a6d37ccd 1915 iput(inode);
2f2dc053
SW		 1916 return 0;
1917}
1918
1919/*
1920 * caller must hold session s_mutex
1921 */
1922static void remove_session_caps(struct ceph_mds_session *session)
1923{
6c93df5d
YZ		 1924 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1925 struct super_block *sb = fsc->sb;
c8a96a31
JL		 1926 LIST_HEAD(dispose);
1927
38d46409 1928 doutc(fsc->client, "on %p\n", session);
f5d77269 1929 ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
6f60f889 1930
c8799fc4
YZ		 1931 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1932
6f60f889
YZ		 1933 spin_lock(&session->s_cap_lock);
1934 if (session->s_nr_caps > 0) {
6f60f889
YZ		 1935 struct inode *inode;
1936 struct ceph_cap *cap, *prev = NULL;
1937 struct ceph_vino vino;
1938 /*
1939 * iterate_session_caps() skips inodes that are being
1940 * deleted, we need to wait until deletions are complete.
1941 * __wait_on_freeing_inode() is designed for the job,
1942 * but it is not exported, so use lookup inode function
1943 * to access it.
1944 */
1945 while (!list_empty(&session->s_caps)) {
1946 cap = list_entry(session->s_caps.next,
1947 struct ceph_cap, session_caps);
1948 if (cap == prev)
1949 break;
1950 prev = cap;
1951 vino = cap->ci->i_vino;
1952 spin_unlock(&session->s_cap_lock);
1953
ed284c49 1954 inode = ceph_find_inode(sb, vino);
23c2c76e 1955 iput(inode);
6f60f889
YZ		 1956
1957 spin_lock(&session->s_cap_lock);
1958 }
1959 }
745a8e3b
YZ		 1960
1961 // drop cap expires and unlock s_cap_lock
c8a96a31 1962 detach_cap_releases(session, &dispose);
6f60f889 1963
2f2dc053 1964 BUG_ON(session->s_nr_caps > 0);
6c99f254 1965 BUG_ON(!list_empty(&session->s_cap_flushing));
c8a96a31
JL		 1966 spin_unlock(&session->s_cap_lock);
1967 dispose_cap_releases(session->s_mdsc, &dispose);
2f2dc053
SW		 1968}
1969
d2f8bb27
YZ		 1970enum {
1971 RECONNECT,
1972 RENEWCAPS,
1973 FORCE_RO,
1974};
1975
2f2dc053
SW		 1976/*
1977 * wake up any threads waiting on this session's caps. if the cap is
1978 * old (didn't get renewed on the client reconnect), remove it now.
1979 *
1980 * caller must hold s_mutex.
1981 */
aaf67de7 1982static int wake_up_session_cb(struct inode *inode, int mds, void *arg)
2f2dc053 1983{
0dc2570f 1984 struct ceph_inode_info *ci = ceph_inode(inode);
d2f8bb27 1985 unsigned long ev = (unsigned long)arg;
0dc2570f 1986
d2f8bb27 1987 if (ev == RECONNECT) {
be655596 1988 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1989 ci->i_wanted_max_size = 0;
1990 ci->i_requested_max_size = 0;
be655596 1991 spin_unlock(&ci->i_ceph_lock);
d2f8bb27 1992 } else if (ev == RENEWCAPS) {
aaf67de7
XL		 1993 struct ceph_cap *cap;
1994
1995 spin_lock(&ci->i_ceph_lock);
1996 cap = __get_cap_for_mds(ci, mds);
1997 /* mds did not re-issue stale cap */
1998 if (cap && cap->cap_gen < atomic_read(&cap->session->s_cap_gen))
d2f8bb27 1999 cap->issued = cap->implemented = CEPH_CAP_PIN;
aaf67de7 2000 spin_unlock(&ci->i_ceph_lock);
d2f8bb27 2001 } else if (ev == FORCE_RO) {
0dc2570f 2002 }
e5360309 2003 wake_up_all(&ci->i_cap_wq);
2f2dc053
SW		 2004 return 0;
2005}
2006
d2f8bb27 2007static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
2f2dc053 2008{
38d46409
XL		 2009 struct ceph_client *cl = session->s_mdsc->fsc->client;
2010
2011 doutc(cl, "session %p mds%d\n", session, session->s_mds);
f5d77269
JL		 2012 ceph_iterate_session_caps(session, wake_up_session_cb,
2013 (void *)(unsigned long)ev);
2f2dc053
SW		 2014}
2015
2016/*
2017 * Send periodic message to MDS renewing all currently held caps. The
2018 * ack will reset the expiration for all caps from this session.
2019 *
2020 * caller holds s_mutex
2021 */
2022static int send_renew_caps(struct ceph_mds_client *mdsc,
2023 struct ceph_mds_session *session)
2024{
38d46409 2025 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 2026 struct ceph_msg *msg;
2027 int state;
2028
2029 if (time_after_eq(jiffies, session->s_cap_ttl) &&
2030 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
38d46409 2031 pr_info_client(cl, "mds%d caps stale\n", session->s_mds);
e4cb4cb8 2032 session->s_renew_requested = jiffies;
2f2dc053
SW		 2033
2034 /* do not try to renew caps until a recovering mds has reconnected
2035 * with its clients. */
2036 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
2037 if (state < CEPH_MDS_STATE_RECONNECT) {
38d46409
XL		 2038 doutc(cl, "ignoring mds%d (%s)\n", session->s_mds,
2039 ceph_mds_state_name(state));
2f2dc053
SW		 2040 return 0;
2041 }
2042
38d46409
XL		 2043 doutc(cl, "to mds%d (%s)\n", session->s_mds,
2044 ceph_mds_state_name(state));
6df89bf2 2045 msg = create_session_full_msg(mdsc, CEPH_SESSION_REQUEST_RENEWCAPS,
fba97e80 2046 ++session->s_renew_seq);
6df89bf2
XL		 2047 if (IS_ERR(msg))
2048 return PTR_ERR(msg);
2f2dc053
SW		 2049 ceph_con_send(&session->s_con, msg);
2050 return 0;
2051}
2052
186e4f7a
YZ		 2053static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
2054 struct ceph_mds_session *session, u64 seq)
2055{
38d46409 2056 struct ceph_client *cl = mdsc->fsc->client;
186e4f7a
YZ		 2057 struct ceph_msg *msg;
2058
38d46409
XL		 2059 doutc(cl, "to mds%d (%s)s seq %lld\n", session->s_mds,
2060 ceph_session_state_name(session->s_state), seq);
fba97e80 2061 msg = ceph_create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
186e4f7a
YZ		 2062 if (!msg)
2063 return -ENOMEM;
2064 ceph_con_send(&session->s_con, msg);
2065 return 0;
2066}
2067
2068
2f2dc053
SW		 2069/*
2070 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 2071 *
2072 * Called under session->s_mutex
2f2dc053
SW		 2073 */
2074static void renewed_caps(struct ceph_mds_client *mdsc,
2075 struct ceph_mds_session *session, int is_renew)
2076{
38d46409 2077 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 2078 int was_stale;
2079 int wake = 0;
2080
2081 spin_lock(&session->s_cap_lock);
1ce208a6 2082 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 2083
2084 session->s_cap_ttl = session->s_renew_requested +
2085 mdsc->mdsmap->m_session_timeout*HZ;
2086
2087 if (was_stale) {
2088 if (time_before(jiffies, session->s_cap_ttl)) {
38d46409
XL		 2089 pr_info_client(cl, "mds%d caps renewed\n",
2090 session->s_mds);
2f2dc053
SW		 2091 wake = 1;
2092 } else {
38d46409
XL		 2093 pr_info_client(cl, "mds%d caps still stale\n",
2094 session->s_mds);
2f2dc053
SW		 2095 }
2096 }
38d46409
XL		 2097 doutc(cl, "mds%d ttl now %lu, was %s, now %s\n", session->s_mds,
2098 session->s_cap_ttl, was_stale ? "stale" : "fresh",
2099 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
2f2dc053
SW		 2100 spin_unlock(&session->s_cap_lock);
2101
2102 if (wake)
d2f8bb27 2103 wake_up_session_caps(session, RENEWCAPS);
2f2dc053
SW		 2104}
2105
2106/*
2107 * send a session close request
2108 */
3e699bd8 2109static int request_close_session(struct ceph_mds_session *session)
2f2dc053 2110{
38d46409 2111 struct ceph_client *cl = session->s_mdsc->fsc->client;
2f2dc053 2112 struct ceph_msg *msg;
2f2dc053 2113
38d46409
XL		 2114 doutc(cl, "mds%d state %s seq %lld\n", session->s_mds,
2115 ceph_session_state_name(session->s_state), session->s_seq);
fba97e80
XL		 2116 msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_CLOSE,
2117 session->s_seq);
a79832f2
SW		 2118 if (!msg)
2119 return -ENOMEM;
2120 ceph_con_send(&session->s_con, msg);
fcff415c 2121 return 1;
2f2dc053
SW		 2122}
2123
2124/*
2125 * Called with s_mutex held.
2126 */
2127static int __close_session(struct ceph_mds_client *mdsc,
2128 struct ceph_mds_session *session)
2129{
2130 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
2131 return 0;
2132 session->s_state = CEPH_MDS_SESSION_CLOSING;
3e699bd8 2133 return request_close_session(session);
2f2dc053
SW		 2134}
2135
040d7860
YZ		 2136static bool drop_negative_children(struct dentry *dentry)
2137{
2138 struct dentry *child;
2139 bool all_negative = true;
2140
2141 if (!d_is_dir(dentry))
2142 goto out;
2143
2144 spin_lock(&dentry->d_lock);
da549bdd 2145 hlist_for_each_entry(child, &dentry->d_children, d_sib) {
040d7860
YZ		 2146 if (d_really_is_positive(child)) {
2147 all_negative = false;
2148 break;
2149 }
2150 }
2151 spin_unlock(&dentry->d_lock);
2152
2153 if (all_negative)
2154 shrink_dcache_parent(dentry);
2155out:
2156 return all_negative;
2157}
2158
2f2dc053
SW		 2159/*
2160 * Trim old(er) caps.
2161 *
2162 * Because we can't cache an inode without one or more caps, we do
2163 * this indirectly: if a cap is unused, we prune its aliases, at which
2164 * point the inode will hopefully get dropped to.
2165 *
2166 * Yes, this is a bit sloppy. Our only real goal here is to respond to
2167 * memory pressure from the MDS, though, so it needn't be perfect.
2168 */
aaf67de7 2169static int trim_caps_cb(struct inode *inode, int mds, void *arg)
2f2dc053 2170{
197b7d79 2171 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
38d46409 2172 struct ceph_client *cl = mdsc->fsc->client;
533a2818 2173 int *remaining = arg;
2f2dc053 2174 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 2175 int used, wanted, oissued, mine;
aaf67de7 2176 struct ceph_cap *cap;
2f2dc053 2177
533a2818 2178 if (*remaining <= 0)
2f2dc053
SW		 2179 return -1;
2180
be655596 2181 spin_lock(&ci->i_ceph_lock);
aaf67de7
XL		 2182 cap = __get_cap_for_mds(ci, mds);
2183 if (!cap) {
2184 spin_unlock(&ci->i_ceph_lock);
2185 return 0;
2186 }
2f2dc053
SW		 2187 mine = cap->issued | cap->implemented;
2188 used = __ceph_caps_used(ci);
979abfdd 2189 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 2190 oissued = __ceph_caps_issued_other(ci, cap);
2191
38d46409
XL		 2192 doutc(cl, "%p %llx.%llx cap %p mine %s oissued %s used %s wanted %s\n",
2193 inode, ceph_vinop(inode), cap, ceph_cap_string(mine),
2194 ceph_cap_string(oissued), ceph_cap_string(used),
2195 ceph_cap_string(wanted));
979abfdd 2196 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 2197 if (ci->i_dirty_caps || ci->i_flushing_caps ||
2198 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 2199 goto out;
2200 if ((used | wanted) & CEPH_CAP_ANY_WR)
2201 goto out;
89aa5930
YZ		 2202 /* Note: it's possible that i_filelock_ref becomes non-zero
2203 * after dropping auth caps. It doesn't hurt because reply
2204 * of lock mds request will re-add auth caps. */
2205 if (atomic_read(&ci->i_filelock_ref) > 0)
2206 goto out;
979abfdd 2207 }
5e804ac4
YZ		 2208 /* The inode has cached pages, but it's no longer used.
2209 * we can safely drop it */
525d15e8
YZ		 2210 if (S_ISREG(inode->i_mode) &&
2211 wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
5e804ac4
YZ		 2212 !(oissued & CEPH_CAP_FILE_CACHE)) {
2213 used = 0;
2214 oissued = 0;
2215 }
979abfdd 2216 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 2217 goto out; /* we need these caps */
2218
2f2dc053
SW		 2219 if (oissued) {
2220 /* we aren't the only cap.. just remove us */
197b7d79 2221 ceph_remove_cap(mdsc, cap, true);
533a2818 2222 (*remaining)--;
2f2dc053 2223 } else {
040d7860 2224 struct dentry *dentry;
5e804ac4 2225 /* try dropping referring dentries */
be655596 2226 spin_unlock(&ci->i_ceph_lock);
040d7860
YZ		 2227 dentry = d_find_any_alias(inode);
2228 if (dentry && drop_negative_children(dentry)) {
2229 int count;
2230 dput(dentry);
2231 d_prune_aliases(inode);
2232 count = atomic_read(&inode->i_count);
2233 if (count == 1)
533a2818 2234 (*remaining)--;
38d46409
XL		 2235 doutc(cl, "%p %llx.%llx cap %p pruned, count now %d\n",
2236 inode, ceph_vinop(inode), cap, count);
040d7860
YZ		 2237 } else {
2238 dput(dentry);
2239 }
2f2dc053
SW		 2240 return 0;
2241 }
2242
2243out:
be655596 2244 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 2245 return 0;
2246}
2247
2248/*
2249 * Trim session cap count down to some max number.
2250 */
e30ee581
ZZ		 2251int ceph_trim_caps(struct ceph_mds_client *mdsc,
2252 struct ceph_mds_session *session,
2253 int max_caps)
2f2dc053 2254{
38d46409 2255 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 2256 int trim_caps = session->s_nr_caps - max_caps;
2257
38d46409
XL		 2258 doutc(cl, "mds%d start: %d / %d, trim %d\n", session->s_mds,
2259 session->s_nr_caps, max_caps, trim_caps);
2f2dc053 2260 if (trim_caps > 0) {
533a2818
JL		 2261 int remaining = trim_caps;
2262
2263 ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
38d46409
XL		 2264 doutc(cl, "mds%d done: %d / %d, trimmed %d\n",
2265 session->s_mds, session->s_nr_caps, max_caps,
2266 trim_caps - remaining);
2f2dc053 2267 }
a56371d9 2268
e3ec8d68 2269 ceph_flush_cap_releases(mdsc, session);
2f2dc053
SW		 2270 return 0;
2271}
2272
8310b089
YZ		 2273static int check_caps_flush(struct ceph_mds_client *mdsc,
2274 u64 want_flush_tid)
2275{
38d46409 2276 struct ceph_client *cl = mdsc->fsc->client;
8310b089
YZ		 2277 int ret = 1;
2278
2279 spin_lock(&mdsc->cap_dirty_lock);
e4500b5e
YZ		 2280 if (!list_empty(&mdsc->cap_flush_list)) {
2281 struct ceph_cap_flush *cf =
2282 list_first_entry(&mdsc->cap_flush_list,
2283 struct ceph_cap_flush, g_list);
2284 if (cf->tid <= want_flush_tid) {
38d46409
XL		 2285 doutc(cl, "still flushing tid %llu <= %llu\n",
2286 cf->tid, want_flush_tid);
e4500b5e
YZ		 2287 ret = 0;
2288 }
8310b089
YZ		 2289 }
2290 spin_unlock(&mdsc->cap_dirty_lock);
2291 return ret;
d3383a8e
YZ		 2292}
2293
2f2dc053
SW		 2294/*
2295 * flush all dirty inode data to disk.
2296 *
8310b089 2297 * returns true if we've flushed through want_flush_tid
2f2dc053 2298 */
affbc19a 2299static void wait_caps_flush(struct ceph_mds_client *mdsc,
0e294387 2300 u64 want_flush_tid)
2f2dc053 2301{
38d46409
XL		 2302 struct ceph_client *cl = mdsc->fsc->client;
2303
2304 doutc(cl, "want %llu\n", want_flush_tid);
8310b089
YZ		 2305
2306 wait_event(mdsc->cap_flushing_wq,
2307 check_caps_flush(mdsc, want_flush_tid));
2308
38d46409 2309 doutc(cl, "ok, flushed thru %llu\n", want_flush_tid);
2f2dc053
SW		 2310}
2311
2312/*
2313 * called under s_mutex
2314 */
e3ec8d68
YZ		 2315static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
2316 struct ceph_mds_session *session)
2f2dc053 2317{
38d46409 2318 struct ceph_client *cl = mdsc->fsc->client;
745a8e3b
YZ		 2319 struct ceph_msg *msg = NULL;
2320 struct ceph_mds_cap_release *head;
2321 struct ceph_mds_cap_item *item;
92475f05 2322 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
745a8e3b
YZ		 2323 struct ceph_cap *cap;
2324 LIST_HEAD(tmp_list);
2325 int num_cap_releases;
92475f05
JL		 2326 __le32 barrier, *cap_barrier;
2327
2328 down_read(&osdc->lock);
2329 barrier = cpu_to_le32(osdc->epoch_barrier);
2330 up_read(&osdc->lock);
2f2dc053 2331
0f8605f2 2332 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 2333again:
2334 list_splice_init(&session->s_cap_releases, &tmp_list);
2335 num_cap_releases = session->s_num_cap_releases;
2336 session->s_num_cap_releases = 0;
2f2dc053 2337 spin_unlock(&session->s_cap_lock);
e01a5946 2338
745a8e3b
YZ		 2339 while (!list_empty(&tmp_list)) {
2340 if (!msg) {
2341 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
09cbfeaf 2342 PAGE_SIZE, GFP_NOFS, false);
745a8e3b
YZ		 2343 if (!msg)
2344 goto out_err;
2345 head = msg->front.iov_base;
2346 head->num = cpu_to_le32(0);
2347 msg->front.iov_len = sizeof(*head);
92475f05
JL		 2348
2349 msg->hdr.version = cpu_to_le16(2);
2350 msg->hdr.compat_version = cpu_to_le16(1);
745a8e3b 2351 }
92475f05 2352
745a8e3b
YZ		 2353 cap = list_first_entry(&tmp_list, struct ceph_cap,
2354 session_caps);
2355 list_del(&cap->session_caps);
2356 num_cap_releases--;
e01a5946 2357
00bd8edb 2358 head = msg->front.iov_base;
4198aba4
JL		 2359 put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
2360 &head->num);
745a8e3b
YZ		 2361 item = msg->front.iov_base + msg->front.iov_len;
2362 item->ino = cpu_to_le64(cap->cap_ino);
2363 item->cap_id = cpu_to_le64(cap->cap_id);
2364 item->migrate_seq = cpu_to_le32(cap->mseq);
2365 item->seq = cpu_to_le32(cap->issue_seq);
2366 msg->front.iov_len += sizeof(*item);
2367
2368 ceph_put_cap(mdsc, cap);
2369
2370 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
92475f05
JL		 2371 // Append cap_barrier field
2372 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2373 *cap_barrier = barrier;
2374 msg->front.iov_len += sizeof(*cap_barrier);
2375
745a8e3b 2376 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
38d46409 2377 doutc(cl, "mds%d %p\n", session->s_mds, msg);
745a8e3b
YZ		 2378 ceph_con_send(&session->s_con, msg);
2379 msg = NULL;
2380 }
00bd8edb 2381 }
e01a5946 2382
745a8e3b 2383 BUG_ON(num_cap_releases != 0);
e01a5946 2384
745a8e3b
YZ		 2385 spin_lock(&session->s_cap_lock);
2386 if (!list_empty(&session->s_cap_releases))
2387 goto again;
2388 spin_unlock(&session->s_cap_lock);
2389
2390 if (msg) {
92475f05
JL		 2391 // Append cap_barrier field
2392 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2393 *cap_barrier = barrier;
2394 msg->front.iov_len += sizeof(*cap_barrier);
2395
745a8e3b 2396 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
38d46409 2397 doutc(cl, "mds%d %p\n", session->s_mds, msg);
745a8e3b 2398 ceph_con_send(&session->s_con, msg);
e01a5946 2399 }
745a8e3b
YZ		 2400 return;
2401out_err:
38d46409
XL		 2402 pr_err_client(cl, "mds%d, failed to allocate message\n",
2403 session->s_mds);
745a8e3b
YZ		 2404 spin_lock(&session->s_cap_lock);
2405 list_splice(&tmp_list, &session->s_cap_releases);
2406 session->s_num_cap_releases += num_cap_releases;
2407 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 2408}
2409
e3ec8d68
YZ		 2410static void ceph_cap_release_work(struct work_struct *work)
2411{
2412 struct ceph_mds_session *session =
2413 container_of(work, struct ceph_mds_session, s_cap_release_work);
2414
2415 mutex_lock(&session->s_mutex);
2416 if (session->s_state == CEPH_MDS_SESSION_OPEN ||
2417 session->s_state == CEPH_MDS_SESSION_HUNG)
2418 ceph_send_cap_releases(session->s_mdsc, session);
2419 mutex_unlock(&session->s_mutex);
2420 ceph_put_mds_session(session);
2421}
2422
2423void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
2424 struct ceph_mds_session *session)
2425{
38d46409 2426 struct ceph_client *cl = mdsc->fsc->client;
e3ec8d68
YZ		 2427 if (mdsc->stopping)
2428 return;
2429
5b3248c6 2430 ceph_get_mds_session(session);
e3ec8d68
YZ		 2431 if (queue_work(mdsc->fsc->cap_wq,
2432 &session->s_cap_release_work)) {
38d46409 2433 doutc(cl, "cap release work queued\n");
e3ec8d68
YZ		 2434 } else {
2435 ceph_put_mds_session(session);
38d46409 2436 doutc(cl, "failed to queue cap release work\n");
e3ec8d68
YZ		 2437 }
2438}
2439
2440/*
2441 * caller holds session->s_cap_lock
2442 */
2443void __ceph_queue_cap_release(struct ceph_mds_session *session,
2444 struct ceph_cap *cap)
2445{
2446 list_add_tail(&cap->session_caps, &session->s_cap_releases);
2447 session->s_num_cap_releases++;
2448
2449 if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
2450 ceph_flush_cap_releases(session->s_mdsc, session);
2451}
2452
37c4efc1
YZ		 2453static void ceph_cap_reclaim_work(struct work_struct *work)
2454{
2455 struct ceph_mds_client *mdsc =
2456 container_of(work, struct ceph_mds_client, cap_reclaim_work);
2457 int ret = ceph_trim_dentries(mdsc);
2458 if (ret == -EAGAIN)
2459 ceph_queue_cap_reclaim_work(mdsc);
2460}
2461
2462void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2463{
38d46409 2464 struct ceph_client *cl = mdsc->fsc->client;
37c4efc1
YZ		 2465 if (mdsc->stopping)
2466 return;
2467
2468 if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
38d46409 2469 doutc(cl, "caps reclaim work queued\n");
37c4efc1 2470 } else {
38d46409 2471 doutc(cl, "failed to queue caps release work\n");
37c4efc1
YZ		 2472 }
2473}
2474
fe33032d
YZ		 2475void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2476{
2477 int val;
2478 if (!nr)
2479 return;
2480 val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
bba1560b 2481 if ((val % CEPH_CAPS_PER_RELEASE) < nr) {
fe33032d
YZ		 2482 atomic_set(&mdsc->cap_reclaim_pending, 0);
2483 ceph_queue_cap_reclaim_work(mdsc);
2484 }
2485}
2486
2f2dc053
SW		 2487/*
2488 * requests
2489 */
2490
54008399
YZ		 2491int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2492 struct inode *dir)
2493{
2494 struct ceph_inode_info *ci = ceph_inode(dir);
2495 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2496 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2a5beea3 2497 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
ad8c28a9
JL		 2498 unsigned int num_entries;
2499 int order;
54008399
YZ		 2500
2501 spin_lock(&ci->i_ceph_lock);
2502 num_entries = ci->i_files + ci->i_subdirs;
2503 spin_unlock(&ci->i_ceph_lock);
ad8c28a9 2504 num_entries = max(num_entries, 1U);
54008399
YZ		 2505 num_entries = min(num_entries, opt->max_readdir);
2506
2507 order = get_order(size * num_entries);
2508 while (order >= 0) {
2a5beea3 2509 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2941bf53
XL		 2510 __GFP_NOWARN |
2511 __GFP_ZERO,
2a5beea3
YZ		 2512 order);
2513 if (rinfo->dir_entries)
54008399
YZ		 2514 break;
2515 order--;
2516 }
2a5beea3 2517 if (!rinfo->dir_entries)
54008399
YZ		 2518 return -ENOMEM;
2519
2520 num_entries = (PAGE_SIZE << order) / size;
2521 num_entries = min(num_entries, opt->max_readdir);
2522
2523 rinfo->dir_buf_size = PAGE_SIZE << order;
2524 req->r_num_caps = num_entries + 1;
2525 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2526 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2527 return 0;
2528}
2529
2f2dc053
SW		 2530/*
2531 * Create an mds request.
2532 */
2533struct ceph_mds_request *
2534ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2535{
058daab7 2536 struct ceph_mds_request *req;
2f2dc053 2537
058daab7 2538 req = kmem_cache_zalloc(ceph_mds_request_cachep, GFP_NOFS);
2f2dc053
SW		 2539 if (!req)
2540 return ERR_PTR(-ENOMEM);
2541
b4556396 2542 mutex_init(&req->r_fill_mutex);
37151668 2543 req->r_mdsc = mdsc;
2f2dc053 2544 req->r_started = jiffies;
70c94820 2545 req->r_start_latency = ktime_get();
2f2dc053
SW		 2546 req->r_resend_mds = -1;
2547 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
68cd5b4b 2548 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2f2dc053 2549 req->r_fmode = -1;
6eb06c46 2550 req->r_feature_needed = -1;
153c8e6b 2551 kref_init(&req->r_kref);
fcd00b68 2552 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 2553 INIT_LIST_HEAD(&req->r_wait);
2554 init_completion(&req->r_completion);
2555 init_completion(&req->r_safe_completion);
2556 INIT_LIST_HEAD(&req->r_unsafe_item);
2557
668c9a61 2558 ktime_get_coarse_real_ts64(&req->r_stamp);
b8e69066 2559
2f2dc053
SW		 2560 req->r_op = op;
2561 req->r_direct_mode = mode;
2562 return req;
2563}
2564
2565/*
44ca18f2 2566 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 2567 *
2568 * called under mdsc->mutex.
2569 */
44ca18f2
SW		 2570static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2571{
2572 if (RB_EMPTY_ROOT(&mdsc->request_tree))
2573 return NULL;
2574 return rb_entry(rb_first(&mdsc->request_tree),
2575 struct ceph_mds_request, r_node);
2576}
2577
e8a7b8b1 2578static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 2579{
e8a7b8b1 2580 return mdsc->oldest_tid;
2f2dc053
SW		 2581}
2582
24865e75
JL		 2583#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
2584static u8 *get_fscrypt_altname(const struct ceph_mds_request *req, u32 *plen)
2585{
2586 struct inode *dir = req->r_parent;
2587 struct dentry *dentry = req->r_dentry;
2588 u8 *cryptbuf = NULL;
2589 u32 len = 0;
2590 int ret = 0;
2591
2592 /* only encode if we have parent and dentry */
2593 if (!dir || !dentry)
2594 goto success;
2595
2596 /* No-op unless this is encrypted */
2597 if (!IS_ENCRYPTED(dir))
2598 goto success;
2599
14e034a6
LH		 2600 ret = ceph_fscrypt_prepare_readdir(dir);
2601 if (ret < 0)
24865e75
JL		 2602 return ERR_PTR(ret);
2603
2604 /* No key? Just ignore it. */
2605 if (!fscrypt_has_encryption_key(dir))
2606 goto success;
2607
2608 if (!fscrypt_fname_encrypted_size(dir, dentry->d_name.len, NAME_MAX,
2609 &len)) {
2610 WARN_ON_ONCE(1);
2611 return ERR_PTR(-ENAMETOOLONG);
2612 }
2613
2614 /* No need to append altname if name is short enough */
2615 if (len <= CEPH_NOHASH_NAME_MAX) {
2616 len = 0;
2617 goto success;
2618 }
2619
2620 cryptbuf = kmalloc(len, GFP_KERNEL);
2621 if (!cryptbuf)
2622 return ERR_PTR(-ENOMEM);
2623
2624 ret = fscrypt_fname_encrypt(dir, &dentry->d_name, cryptbuf, len);
2625 if (ret) {
2626 kfree(cryptbuf);
2627 return ERR_PTR(ret);
2628 }
2629success:
2630 *plen = len;
2631 return cryptbuf;
2632}
2633#else
2634static u8 *get_fscrypt_altname(const struct ceph_mds_request *req, u32 *plen)
2635{
2636 *plen = 0;
2637 return NULL;
2638}
2639#endif
2640
3fd945a7
JL		 2641/**
2642 * ceph_mdsc_build_path - build a path string to a given dentry
197b7d79 2643 * @mdsc: mds client
3fd945a7
JL		 2644 * @dentry: dentry to which path should be built
2645 * @plen: returned length of string
2646 * @pbase: returned base inode number
2647 * @for_wire: is this path going to be sent to the MDS?
2648 *
2649 * Build a string that represents the path to the dentry. This is mostly called
2650 * for two different purposes:
2651 *
2652 * 1) we need to build a path string to send to the MDS (for_wire == true)
2653 * 2) we need a path string for local presentation (e.g. debugfs)
2654 * (for_wire == false)
2f2dc053 2655 *
3fd945a7
JL		 2656 * The path is built in reverse, starting with the dentry. Walk back up toward
2657 * the root, building the path until the first non-snapped inode is reached
2658 * (for_wire) or the root inode is reached (!for_wire).
2f2dc053
SW		 2659 *
2660 * Encode hidden .snap dirs as a double /, i.e.
2661 * foo/.snap/bar -> foo//bar
2662 */
197b7d79
XL		 2663char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc, struct dentry *dentry,
2664 int *plen, u64 *pbase, int for_wire)
2f2dc053 2665{
38d46409 2666 struct ceph_client *cl = mdsc->fsc->client;
4c793d4c
JL		 2667 struct dentry *cur;
2668 struct inode *inode;
2f2dc053 2669 char *path;
f77f21bb 2670 int pos;
1b71fe2e 2671 unsigned seq;
69a10fb3 2672 u64 base;
2f2dc053 2673
d37b1d99 2674 if (!dentry)
2f2dc053
SW		 2675 return ERR_PTR(-EINVAL);
2676
f77f21bb 2677 path = __getname();
d37b1d99 2678 if (!path)
2f2dc053 2679 return ERR_PTR(-ENOMEM);
f77f21bb
JL		 2680retry:
2681 pos = PATH_MAX - 1;
2682 path[pos] = '\0';
2683
2684 seq = read_seqbegin(&rename_lock);
4c793d4c 2685 cur = dget(dentry);
f77f21bb 2686 for (;;) {
3fd945a7 2687 struct dentry *parent;
2f2dc053 2688
4c793d4c
JL		 2689 spin_lock(&cur->d_lock);
2690 inode = d_inode(cur);
2f2dc053 2691 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
38d46409 2692 doutc(cl, "path+%d: %p SNAPDIR\n", pos, cur);
3fd945a7
JL		 2693 spin_unlock(&cur->d_lock);
2694 parent = dget_parent(cur);
2695 } else if (for_wire && inode && dentry != cur &&
2f2dc053 2696 ceph_snap(inode) == CEPH_NOSNAP) {
4c793d4c 2697 spin_unlock(&cur->d_lock);
d6b8bd67 2698 pos++; /* get rid of any prepended '/' */
2f2dc053 2699 break;
3fd945a7 2700 } else if (!for_wire || !IS_ENCRYPTED(d_inode(cur->d_parent))) {
4c793d4c 2701 pos -= cur->d_name.len;
1b71fe2e 2702 if (pos < 0) {
4c793d4c 2703 spin_unlock(&cur->d_lock);
2f2dc053 2704 break;
1b71fe2e 2705 }
4c793d4c 2706 memcpy(path + pos, cur->d_name.name, cur->d_name.len);
3fd945a7
JL		 2707 spin_unlock(&cur->d_lock);
2708 parent = dget_parent(cur);
2709 } else {
2710 int len, ret;
2711 char buf[NAME_MAX];
2712
2713 /*
2714 * Proactively copy name into buf, in case we need to
2715 * present it as-is.
2716 */
2717 memcpy(buf, cur->d_name.name, cur->d_name.len);
2718 len = cur->d_name.len;
2719 spin_unlock(&cur->d_lock);
2720 parent = dget_parent(cur);
2721
14e034a6 2722 ret = ceph_fscrypt_prepare_readdir(d_inode(parent));
3fd945a7
JL		 2723 if (ret < 0) {
2724 dput(parent);
2725 dput(cur);
2726 return ERR_PTR(ret);
2727 }
2728
2729 if (fscrypt_has_encryption_key(d_inode(parent))) {
2730 len = ceph_encode_encrypted_fname(d_inode(parent),
2731 cur, buf);
2732 if (len < 0) {
2733 dput(parent);
2734 dput(cur);
2735 return ERR_PTR(len);
2736 }
2737 }
2738 pos -= len;
2739 if (pos < 0) {
2740 dput(parent);
2741 break;
2742 }
2743 memcpy(path + pos, buf, len);
2f2dc053 2744 }
3fd945a7
JL		 2745 dput(cur);
2746 cur = parent;
f77f21bb
JL		 2747
2748 /* Are we at the root? */
4c793d4c 2749 if (IS_ROOT(cur))
f77f21bb
JL		 2750 break;
2751
2752 /* Are we out of buffer? */
2753 if (--pos < 0)
2754 break;
2755
2756 path[pos] = '/';
2f2dc053 2757 }
4c793d4c
JL		 2758 inode = d_inode(cur);
2759 base = inode ? ceph_ino(inode) : 0;
2760 dput(cur);
f5946bcc
JL		 2761
2762 if (read_seqretry(&rename_lock, seq))
2763 goto retry;
2764
2765 if (pos < 0) {
2766 /*
2767 * A rename didn't occur, but somehow we didn't end up where
2768 * we thought we would. Throw a warning and try again.
2769 */
38d46409
XL		 2770 pr_warn_client(cl, "did not end path lookup where expected (pos = %d)\n",
2771 pos);
2f2dc053
SW		 2772 goto retry;
2773 }
2774
69a10fb3 2775 *pbase = base;
f77f21bb 2776 *plen = PATH_MAX - 1 - pos;
38d46409
XL		 2777 doutc(cl, "on %p %d built %llx '%.*s'\n", dentry, d_count(dentry),
2778 base, *plen, path + pos);
f77f21bb 2779 return path + pos;
2f2dc053
SW		 2780}
2781
197b7d79
XL		 2782static int build_dentry_path(struct ceph_mds_client *mdsc, struct dentry *dentry,
2783 struct inode *dir, const char **ppath, int *ppathlen,
2784 u64 *pino, bool *pfreepath, bool parent_locked)
2f2dc053
SW		 2785{
2786 char *path;
2787
c6b0b656 2788 rcu_read_lock();
fd36a717
JL		 2789 if (!dir)
2790 dir = d_inode_rcu(dentry->d_parent);
3fd945a7
JL		 2791 if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP &&
2792 !IS_ENCRYPTED(dir)) {
c6b0b656
JL		 2793 *pino = ceph_ino(dir);
2794 rcu_read_unlock();
964fff74
JL		 2795 *ppath = dentry->d_name.name;
2796 *ppathlen = dentry->d_name.len;
2f2dc053
SW		 2797 return 0;
2798 }
c6b0b656 2799 rcu_read_unlock();
197b7d79 2800 path = ceph_mdsc_build_path(mdsc, dentry, ppathlen, pino, 1);
2f2dc053
SW		 2801 if (IS_ERR(path))
2802 return PTR_ERR(path);
2803 *ppath = path;
1bcb3440 2804 *pfreepath = true;
2f2dc053
SW		 2805 return 0;
2806}
2807
2808static int build_inode_path(struct inode *inode,
2809 const char **ppath, int *ppathlen, u64 *pino,
1bcb3440 2810 bool *pfreepath)
2f2dc053 2811{
197b7d79 2812 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
2f2dc053
SW		 2813 struct dentry *dentry;
2814 char *path;
2815
2816 if (ceph_snap(inode) == CEPH_NOSNAP) {
2817 *pino = ceph_ino(inode);
2818 *ppathlen = 0;
2819 return 0;
2820 }
2821 dentry = d_find_alias(inode);
197b7d79 2822 path = ceph_mdsc_build_path(mdsc, dentry, ppathlen, pino, 1);
2f2dc053
SW		 2823 dput(dentry);
2824 if (IS_ERR(path))
2825 return PTR_ERR(path);
2826 *ppath = path;
1bcb3440 2827 *pfreepath = true;
2f2dc053
SW		 2828 return 0;
2829}
2830
2831/*
2832 * request arguments may be specified via an inode *, a dentry *, or
2833 * an explicit ino+path.
2834 */
197b7d79
XL		 2835static int set_request_path_attr(struct ceph_mds_client *mdsc, struct inode *rinode,
2836 struct dentry *rdentry, struct inode *rdiri,
2837 const char *rpath, u64 rino, const char **ppath,
2838 int *pathlen, u64 *ino, bool *freepath,
2839 bool parent_locked)
2f2dc053 2840{
38d46409 2841 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 2842 int r = 0;
2843
2844 if (rinode) {
2845 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
38d46409
XL		 2846 doutc(cl, " inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2847 ceph_snap(rinode));
2f2dc053 2848 } else if (rdentry) {
197b7d79 2849 r = build_dentry_path(mdsc, rdentry, rdiri, ppath, pathlen, ino,
1bcb3440 2850 freepath, parent_locked);
38d46409 2851 doutc(cl, " dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen, *ppath);
795858db 2852 } else if (rpath || rino) {
2f2dc053
SW		 2853 *ino = rino;
2854 *ppath = rpath;
b000056a 2855 *pathlen = rpath ? strlen(rpath) : 0;
38d46409 2856 doutc(cl, " path %.*s\n", *pathlen, rpath);
2f2dc053
SW		 2857 }
2858
2859 return r;
2860}
2861
2d332d5b
JL		 2862static void encode_mclientrequest_tail(void **p,
2863 const struct ceph_mds_request *req)
60267ba3
ID		 2864{
2865 struct ceph_timespec ts;
2866 int i;
2867
2868 ceph_encode_timespec64(&ts, &req->r_stamp);
2869 ceph_encode_copy(p, &ts, sizeof(ts));
2870
24865e75 2871 /* v4: gid_list */
60267ba3
ID		 2872 ceph_encode_32(p, req->r_cred->group_info->ngroups);
2873 for (i = 0; i < req->r_cred->group_info->ngroups; i++)
2874 ceph_encode_64(p, from_kgid(&init_user_ns,
2875 req->r_cred->group_info->gid[i]));
2d332d5b 2876
24865e75
JL		 2877 /* v5: altname */
2878 ceph_encode_32(p, req->r_altname_len);
2879 ceph_encode_copy(p, req->r_altname, req->r_altname_len);
2d332d5b
JL		 2880
2881 /* v6: fscrypt_auth and fscrypt_file */
2882 if (req->r_fscrypt_auth) {
2883 u32 authlen = ceph_fscrypt_auth_len(req->r_fscrypt_auth);
2884
2885 ceph_encode_32(p, authlen);
2886 ceph_encode_copy(p, req->r_fscrypt_auth, authlen);
2887 } else {
2888 ceph_encode_32(p, 0);
2889 }
16be62fc
JL		 2890 if (test_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags)) {
2891 ceph_encode_32(p, sizeof(__le64));
2892 ceph_encode_64(p, req->r_fscrypt_file);
2893 } else {
2894 ceph_encode_32(p, 0);
2895 }
60267ba3
ID		 2896}
2897
5ccd8530
CB		 2898static inline u16 mds_supported_head_version(struct ceph_mds_session *session)
2899{
2900 if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, &session->s_features))
2901 return 1;
2902
2903 if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features))
2904 return 2;
2905
2906 return CEPH_MDS_REQUEST_HEAD_VERSION;
2907}
2908
ce0d5bd3
XL		 2909static struct ceph_mds_request_head_legacy *
2910find_legacy_request_head(void *p, u64 features)
2911{
2912 bool legacy = !(features & CEPH_FEATURE_FS_BTIME);
2913 struct ceph_mds_request_head_old *ohead;
2914
2915 if (legacy)
2916 return (struct ceph_mds_request_head_legacy *)p;
2917 ohead = (struct ceph_mds_request_head_old *)p;
2918 return (struct ceph_mds_request_head_legacy *)&ohead->oldest_client_tid;
2919}
2920
2f2dc053
SW		 2921/*
2922 * called under mdsc->mutex
2923 */
4f1ddb1e 2924static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
2f2dc053 2925 struct ceph_mds_request *req,
4f1ddb1e 2926 bool drop_cap_releases)
2f2dc053 2927{
4f1ddb1e
JL		 2928 int mds = session->s_mds;
2929 struct ceph_mds_client *mdsc = session->s_mdsc;
5ccd8530 2930 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053 2931 struct ceph_msg *msg;
ce0d5bd3 2932 struct ceph_mds_request_head_legacy *lhead;
2f2dc053
SW		 2933 const char *path1 = NULL;
2934 const char *path2 = NULL;
2935 u64 ino1 = 0, ino2 = 0;
2936 int pathlen1 = 0, pathlen2 = 0;
1bcb3440 2937 bool freepath1 = false, freepath2 = false;
a5ffd7b6 2938 struct dentry *old_dentry = NULL;
60267ba3 2939 int len;
2f2dc053
SW		 2940 u16 releases;
2941 void *p, *end;
2942 int ret;
4f1ddb1e 2943 bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
5ccd8530 2944 u16 request_head_version = mds_supported_head_version(session);
673478b6
AM		 2945 kuid_t caller_fsuid = req->r_cred->fsuid;
2946 kgid_t caller_fsgid = req->r_cred->fsgid;
2f2dc053 2947
197b7d79 2948 ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
3dd69aab 2949 req->r_parent, req->r_path1, req->r_ino1.ino,
1bcb3440
JL		 2950 &path1, &pathlen1, &ino1, &freepath1,
2951 test_bit(CEPH_MDS_R_PARENT_LOCKED,
2952 &req->r_req_flags));
2f2dc053
SW		 2953 if (ret < 0) {
2954 msg = ERR_PTR(ret);
2955 goto out;
2956 }
2957
1bcb3440 2958 /* If r_old_dentry is set, then assume that its parent is locked */
a5ffd7b6
XL		 2959 if (req->r_old_dentry &&
2960 !(req->r_old_dentry->d_flags & DCACHE_DISCONNECTED))
2961 old_dentry = req->r_old_dentry;
197b7d79 2962 ret = set_request_path_attr(mdsc, NULL, old_dentry,
fd36a717 2963 req->r_old_dentry_dir,
2f2dc053 2964 req->r_path2, req->r_ino2.ino,
1bcb3440 2965 &path2, &pathlen2, &ino2, &freepath2, true);
2f2dc053
SW		 2966 if (ret < 0) {
2967 msg = ERR_PTR(ret);
2968 goto out_free1;
2969 }
2970
24865e75
JL		 2971 req->r_altname = get_fscrypt_altname(req, &req->r_altname_len);
2972 if (IS_ERR(req->r_altname)) {
2973 msg = ERR_CAST(req->r_altname);
2974 req->r_altname = NULL;
2975 goto out_free2;
2976 }
2977
ce0d5bd3
XL		 2978 /*
2979 * For old cephs without supporting the 32bit retry/fwd feature
2980 * it will copy the raw memories directly when decoding the
2981 * requests. While new cephs will decode the head depending the
2982 * version member, so we need to make sure it will be compatible
2983 * with them both.
2984 */
2985 if (legacy)
2986 len = sizeof(struct ceph_mds_request_head_legacy);
5ccd8530 2987 else if (request_head_version == 1)
ce0d5bd3 2988 len = sizeof(struct ceph_mds_request_head_old);
5ccd8530
CB		 2989 else if (request_head_version == 2)
2990 len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
ce0d5bd3
XL		 2991 else
2992 len = sizeof(struct ceph_mds_request_head);
2f2dc053 2993
2d332d5b
JL		 2994 /* filepaths */
2995 len += 2 * (1 + sizeof(u32) + sizeof(u64));
2996 len += pathlen1 + pathlen2;
2997
2998 /* cap releases */
2f2dc053
SW		 2999 len += sizeof(struct ceph_mds_request_release) *
3000 (!!req->r_inode_drop + !!req->r_dentry_drop +
3001 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
4f1ddb1e 3002
2f2dc053 3003 if (req->r_dentry_drop)
c1dfc277 3004 len += pathlen1;
2f2dc053 3005 if (req->r_old_dentry_drop)
c1dfc277 3006 len += pathlen2;
2f2dc053 3007
2d332d5b
JL		 3008 /* MClientRequest tail */
3009
3010 /* req->r_stamp */
3011 len += sizeof(struct ceph_timespec);
3012
3013 /* gid list */
3014 len += sizeof(u32) + (sizeof(u64) * req->r_cred->group_info->ngroups);
3015
3016 /* alternate name */
24865e75 3017 len += sizeof(u32) + req->r_altname_len;
2d332d5b
JL		 3018
3019 /* fscrypt_auth */
3020 len += sizeof(u32); // fscrypt_auth
3021 if (req->r_fscrypt_auth)
3022 len += ceph_fscrypt_auth_len(req->r_fscrypt_auth);
3023
3024 /* fscrypt_file */
3025 len += sizeof(u32);
16be62fc
JL		 3026 if (test_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags))
3027 len += sizeof(__le64);
2d332d5b 3028
0d9c1ab3 3029 msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
a79832f2
SW		 3030 if (!msg) {
3031 msg = ERR_PTR(-ENOMEM);
2f2dc053 3032 goto out_free2;
a79832f2 3033 }
2f2dc053 3034
6df058c0
SW		 3035 msg->hdr.tid = cpu_to_le64(req->r_tid);
3036
ce0d5bd3
XL		 3037 lhead = find_legacy_request_head(msg->front.iov_base,
3038 session->s_con.peer_features);
3039
5ccd8530
CB		 3040 if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
3041 !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features)) {
3042 WARN_ON_ONCE(!IS_CEPH_MDS_OP_NEWINODE(req->r_op));
3043
673478b6
AM		 3044 if (enable_unsafe_idmap) {
3045 pr_warn_once_client(cl,
3046 "idmapped mount is used and CEPHFS_FEATURE_HAS_OWNER_UIDGID"
3047 " is not supported by MDS. UID/GID-based restrictions may"
3048 " not work properly.\n");
5ccd8530 3049
673478b6
AM		 3050 caller_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
3051 VFSUIDT_INIT(req->r_cred->fsuid));
3052 caller_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
3053 VFSGIDT_INIT(req->r_cred->fsgid));
3054 } else {
3055 pr_err_ratelimited_client(cl,
3056 "idmapped mount is used and CEPHFS_FEATURE_HAS_OWNER_UIDGID"
3057 " is not supported by MDS. Fail request with -EIO.\n");
3058
3059 ret = -EIO;
3060 goto out_err;
3061 }
5ccd8530
CB		 3062 }
3063
4f1ddb1e 3064 /*
ce0d5bd3 3065 * The ceph_mds_request_head_legacy didn't contain a version field, and
4f1ddb1e
JL		 3066 * one was added when we moved the message version from 3->4.
3067 */
3068 if (legacy) {
3069 msg->hdr.version = cpu_to_le16(3);
ce0d5bd3 3070 p = msg->front.iov_base + sizeof(*lhead);
5ccd8530 3071 } else if (request_head_version == 1) {
ce0d5bd3
XL		 3072 struct ceph_mds_request_head_old *ohead = msg->front.iov_base;
3073
3074 msg->hdr.version = cpu_to_le16(4);
3075 ohead->version = cpu_to_le16(1);
3076 p = msg->front.iov_base + sizeof(*ohead);
5ccd8530
CB		 3077 } else if (request_head_version == 2) {
3078 struct ceph_mds_request_head *nhead = msg->front.iov_base;
3079
3080 msg->hdr.version = cpu_to_le16(6);
3081 nhead->version = cpu_to_le16(2);
3082
3083 p = msg->front.iov_base + offsetofend(struct ceph_mds_request_head, ext_num_fwd);
4f1ddb1e 3084 } else {
ce0d5bd3 3085 struct ceph_mds_request_head *nhead = msg->front.iov_base;
5ccd8530
CB		 3086 kuid_t owner_fsuid;
3087 kgid_t owner_fsgid;
4f1ddb1e 3088
2d332d5b 3089 msg->hdr.version = cpu_to_le16(6);
ce0d5bd3 3090 nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
5ccd8530
CB		 3091 nhead->struct_len = cpu_to_le32(sizeof(struct ceph_mds_request_head));
3092
3093 if (IS_CEPH_MDS_OP_NEWINODE(req->r_op)) {
3094 owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
3095 VFSUIDT_INIT(req->r_cred->fsuid));
3096 owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
3097 VFSGIDT_INIT(req->r_cred->fsgid));
3098 nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns, owner_fsuid));
3099 nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns, owner_fsgid));
3100 } else {
3101 nhead->owner_uid = cpu_to_le32(-1);
3102 nhead->owner_gid = cpu_to_le32(-1);
3103 }
3104
ce0d5bd3 3105 p = msg->front.iov_base + sizeof(*nhead);
4f1ddb1e
JL		 3106 }
3107
2f2dc053
SW		 3108 end = msg->front.iov_base + msg->front.iov_len;
3109
ce0d5bd3
XL		 3110 lhead->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
3111 lhead->op = cpu_to_le32(req->r_op);
3112 lhead->caller_uid = cpu_to_le32(from_kuid(&init_user_ns,
673478b6 3113 caller_fsuid));
ce0d5bd3 3114 lhead->caller_gid = cpu_to_le32(from_kgid(&init_user_ns,
673478b6 3115 caller_fsgid));
ce0d5bd3
XL		 3116 lhead->ino = cpu_to_le64(req->r_deleg_ino);
3117 lhead->args = req->r_args;
2f2dc053
SW		 3118
3119 ceph_encode_filepath(&p, end, ino1, path1);
3120 ceph_encode_filepath(&p, end, ino2, path2);
3121
e979cf50
SW		 3122 /* make note of release offset, in case we need to replay */
3123 req->r_request_release_offset = p - msg->front.iov_base;
3124
2f2dc053
SW		 3125 /* cap releases */
3126 releases = 0;
3127 if (req->r_inode_drop)
3128 releases += ceph_encode_inode_release(&p,
2b0143b5 3129 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
719a2514
YZ		 3130 mds, req->r_inode_drop, req->r_inode_unless,
3131 req->r_op == CEPH_MDS_OP_READDIR);
3fd945a7
JL		 3132 if (req->r_dentry_drop) {
3133 ret = ceph_encode_dentry_release(&p, req->r_dentry,
3dd69aab 3134 req->r_parent, mds, req->r_dentry_drop,
ca6c8ae0 3135 req->r_dentry_unless);
3fd945a7
JL		 3136 if (ret < 0)
3137 goto out_err;
3138 releases += ret;
3139 }
3140 if (req->r_old_dentry_drop) {
3141 ret = ceph_encode_dentry_release(&p, req->r_old_dentry,
ca6c8ae0
JL		 3142 req->r_old_dentry_dir, mds,
3143 req->r_old_dentry_drop,
3144 req->r_old_dentry_unless);
3fd945a7
JL		 3145 if (ret < 0)
3146 goto out_err;
3147 releases += ret;
3148 }
2f2dc053
SW		 3149 if (req->r_old_inode_drop)
3150 releases += ceph_encode_inode_release(&p,
2b0143b5 3151 d_inode(req->r_old_dentry),
2f2dc053 3152 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 3153
3154 if (drop_cap_releases) {
3155 releases = 0;
3156 p = msg->front.iov_base + req->r_request_release_offset;
3157 }
3158
ce0d5bd3 3159 lhead->num_releases = cpu_to_le16(releases);
2f2dc053 3160
2d332d5b 3161 encode_mclientrequest_tail(&p, req);
4f1ddb1e 3162
b682c6d4
XL		 3163 if (WARN_ON_ONCE(p > end)) {
3164 ceph_msg_put(msg);
3165 msg = ERR_PTR(-ERANGE);
3166 goto out_free2;
3167 }
3168
2f2dc053
SW		 3169 msg->front.iov_len = p - msg->front.iov_base;
3170 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
3171
25e6bae3
YZ		 3172 if (req->r_pagelist) {
3173 struct ceph_pagelist *pagelist = req->r_pagelist;
25e6bae3
YZ		 3174 ceph_msg_data_add_pagelist(msg, pagelist);
3175 msg->hdr.data_len = cpu_to_le32(pagelist->length);
3176 } else {
3177 msg->hdr.data_len = 0;
ebf18f47 3178 }
02afca6c 3179
2f2dc053
SW		 3180 msg->hdr.data_off = cpu_to_le16(0);
3181
3182out_free2:
3183 if (freepath2)
f77f21bb 3184 ceph_mdsc_free_path((char *)path2, pathlen2);
2f2dc053
SW		 3185out_free1:
3186 if (freepath1)
f77f21bb 3187 ceph_mdsc_free_path((char *)path1, pathlen1);
2f2dc053
SW		 3188out:
3189 return msg;
3fd945a7
JL		 3190out_err:
3191 ceph_msg_put(msg);
3192 msg = ERR_PTR(ret);
3193 goto out_free2;
2f2dc053
SW		 3194}
3195
3196/*
3197 * called under mdsc->mutex if error, under no mutex if
3198 * success.
3199 */
3200static void complete_request(struct ceph_mds_client *mdsc,
3201 struct ceph_mds_request *req)
3202{
70c94820
XL		 3203 req->r_end_latency = ktime_get();
3204
2f2dc053
SW		 3205 if (req->r_callback)
3206 req->r_callback(mdsc, req);
111c7081 3207 complete_all(&req->r_completion);
2f2dc053
SW		 3208}
3209
3210/*
3211 * called under mdsc->mutex
3212 */
396bd62c 3213static int __prepare_send_request(struct ceph_mds_session *session,
2f2dc053 3214 struct ceph_mds_request *req,
396bd62c 3215 bool drop_cap_releases)
2f2dc053 3216{
396bd62c
JL		 3217 int mds = session->s_mds;
3218 struct ceph_mds_client *mdsc = session->s_mdsc;
38d46409 3219 struct ceph_client *cl = mdsc->fsc->client;
ce0d5bd3
XL		 3220 struct ceph_mds_request_head_legacy *lhead;
3221 struct ceph_mds_request_head *nhead;
2f2dc053 3222 struct ceph_msg *msg;
ce0d5bd3
XL		 3223 int flags = 0, old_max_retry;
3224 bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
3225 &session->s_features);
546a5d61
XL		 3226
3227 /*
ce0d5bd3
XL		 3228 * Avoid inifinite retrying after overflow. The client will
3229 * increase the retry count and if the MDS is old version,
3230 * so we limit to retry at most 256 times.
546a5d61 3231 */
ce0d5bd3
XL		 3232 if (req->r_attempts) {
3233 old_max_retry = sizeof_field(struct ceph_mds_request_head_old,
3234 num_retry);
3235 old_max_retry = 1 << (old_max_retry * BITS_PER_BYTE);
3236 if ((old_version && req->r_attempts >= old_max_retry) ||
3237 ((uint32_t)req->r_attempts >= U32_MAX)) {
38d46409
XL		 3238 pr_warn_ratelimited_client(cl, "request tid %llu seq overflow\n",
3239 req->r_tid);
ce0d5bd3
XL		 3240 return -EMULTIHOP;
3241 }
546a5d61 3242 }
2f2dc053 3243
2f2dc053 3244 req->r_attempts++;
e55b71f8
GF		 3245 if (req->r_inode) {
3246 struct ceph_cap *cap =
3247 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
3248
3249 if (cap)
3250 req->r_sent_on_mseq = cap->mseq;
3251 else
3252 req->r_sent_on_mseq = -1;
3253 }
38d46409
XL		 3254 doutc(cl, "%p tid %lld %s (attempt %d)\n", req, req->r_tid,
3255 ceph_mds_op_name(req->r_op), req->r_attempts);
2f2dc053 3256
bc2de10d 3257 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
c5c9a0bf 3258 void *p;
4f1ddb1e 3259
01a92f17
SW		 3260 /*
3261 * Replay. Do not regenerate message (and rebuild
3262 * paths, etc.); just use the original message.
3263 * Rebuilding paths will break for renames because
3264 * d_move mangles the src name.
3265 */
3266 msg = req->r_request;
ce0d5bd3
XL		 3267 lhead = find_legacy_request_head(msg->front.iov_base,
3268 session->s_con.peer_features);
01a92f17 3269
ce0d5bd3 3270 flags = le32_to_cpu(lhead->flags);
01a92f17 3271 flags |= CEPH_MDS_FLAG_REPLAY;
ce0d5bd3 3272 lhead->flags = cpu_to_le32(flags);
01a92f17
SW		 3273
3274 if (req->r_target_inode)
ce0d5bd3 3275 lhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
01a92f17 3276
ce0d5bd3
XL		 3277 lhead->num_retry = req->r_attempts - 1;
3278 if (!old_version) {
3279 nhead = (struct ceph_mds_request_head*)msg->front.iov_base;
3280 nhead->ext_num_retry = cpu_to_le32(req->r_attempts - 1);
3281 }
e979cf50
SW		 3282
3283 /* remove cap/dentry releases from message */
ce0d5bd3 3284 lhead->num_releases = 0;
c5c9a0bf 3285
c5c9a0bf 3286 p = msg->front.iov_base + req->r_request_release_offset;
2d332d5b 3287 encode_mclientrequest_tail(&p, req);
c5c9a0bf
YZ		 3288
3289 msg->front.iov_len = p - msg->front.iov_base;
3290 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 3291 return 0;
3292 }
3293
2f2dc053
SW		 3294 if (req->r_request) {
3295 ceph_msg_put(req->r_request);
3296 req->r_request = NULL;
3297 }
4f1ddb1e 3298 msg = create_request_message(session, req, drop_cap_releases);
2f2dc053 3299 if (IS_ERR(msg)) {
e1518c7c 3300 req->r_err = PTR_ERR(msg);
a79832f2 3301 return PTR_ERR(msg);
2f2dc053
SW		 3302 }
3303 req->r_request = msg;
3304
ce0d5bd3
XL		 3305 lhead = find_legacy_request_head(msg->front.iov_base,
3306 session->s_con.peer_features);
3307 lhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
bc2de10d 3308 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2f2dc053 3309 flags |= CEPH_MDS_FLAG_REPLAY;
3bb48b41
JL		 3310 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags))
3311 flags |= CEPH_MDS_FLAG_ASYNC;
3dd69aab 3312 if (req->r_parent)
2f2dc053 3313 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
ce0d5bd3
XL		 3314 lhead->flags = cpu_to_le32(flags);
3315 lhead->num_fwd = req->r_num_fwd;
3316 lhead->num_retry = req->r_attempts - 1;
3317 if (!old_version) {
3318 nhead = (struct ceph_mds_request_head*)msg->front.iov_base;
3319 nhead->ext_num_fwd = cpu_to_le32(req->r_num_fwd);
3320 nhead->ext_num_retry = cpu_to_le32(req->r_attempts - 1);
3321 }
2f2dc053 3322
38d46409 3323 doutc(cl, " r_parent = %p\n", req->r_parent);
2f2dc053
SW		 3324 return 0;
3325}
3326
9cf54563
XL		 3327/*
3328 * called under mdsc->mutex
3329 */
396bd62c 3330static int __send_request(struct ceph_mds_session *session,
9cf54563
XL		 3331 struct ceph_mds_request *req,
3332 bool drop_cap_releases)
3333{
3334 int err;
3335
396bd62c 3336 err = __prepare_send_request(session, req, drop_cap_releases);
9cf54563
XL		 3337 if (!err) {
3338 ceph_msg_get(req->r_request);
3339 ceph_con_send(&session->s_con, req->r_request);
3340 }
3341
3342 return err;
3343}
3344
2f2dc053
SW		 3345/*
3346 * send request, or put it on the appropriate wait list.
3347 */
d5548492 3348static void __do_request(struct ceph_mds_client *mdsc,
2f2dc053
SW		 3349 struct ceph_mds_request *req)
3350{
38d46409 3351 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 3352 struct ceph_mds_session *session = NULL;
3353 int mds = -1;
48fec5d0 3354 int err = 0;
c4853e97 3355 bool random;
2f2dc053 3356
bc2de10d
JL		 3357 if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
3358 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
eb1b8af3 3359 __unregister_request(mdsc, req);
d5548492 3360 return;
eb1b8af3 3361 }
2f2dc053 3362
a68e564a 3363 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO) {
38d46409 3364 doutc(cl, "metadata corrupted\n");
a68e564a
XL		 3365 err = -EIO;
3366 goto finish;
3367 }
2f2dc053
SW		 3368 if (req->r_timeout &&
3369 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
38d46409 3370 doutc(cl, "timed out\n");
8ccf7fcc 3371 err = -ETIMEDOUT;
2f2dc053
SW		 3372 goto finish;
3373 }
52953d55 3374 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
38d46409 3375 doutc(cl, "forced umount\n");
48fec5d0
YZ		 3376 err = -EIO;
3377 goto finish;
3378 }
52953d55 3379 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
e9e427f0
YZ		 3380 if (mdsc->mdsmap_err) {
3381 err = mdsc->mdsmap_err;
38d46409 3382 doutc(cl, "mdsmap err %d\n", err);
e9e427f0
YZ		 3383 goto finish;
3384 }
cc8e8342 3385 if (mdsc->mdsmap->m_epoch == 0) {
38d46409 3386 doutc(cl, "no mdsmap, waiting for map\n");
cc8e8342 3387 list_add(&req->r_wait, &mdsc->waiting_for_map);
d5548492 3388 return;
cc8e8342 3389 }
e9e427f0
YZ		 3390 if (!(mdsc->fsc->mount_options->flags &
3391 CEPH_MOUNT_OPT_MOUNTWAIT) &&
3392 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
97820058 3393 err = -EHOSTUNREACH;
e9e427f0
YZ		 3394 goto finish;
3395 }
3396 }
2f2dc053 3397
dc69e2e9
SW		 3398 put_request_session(req);
3399
c4853e97 3400 mds = __choose_mds(mdsc, req, &random);
2f2dc053
SW		 3401 if (mds < 0 ||
3402 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
3bb48b41
JL		 3403 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
3404 err = -EJUKEBOX;
3405 goto finish;
3406 }
38d46409 3407 doutc(cl, "no mds or not active, waiting for map\n");
2f2dc053 3408 list_add(&req->r_wait, &mdsc->waiting_for_map);
d5548492 3409 return;
2f2dc053
SW		 3410 }
3411
3412 /* get, open session */
3413 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 3414 if (!session) {
2f2dc053 3415 session = register_session(mdsc, mds);
9c423956
SW		 3416 if (IS_ERR(session)) {
3417 err = PTR_ERR(session);
3418 goto finish;
3419 }
3420 }
5b3248c6 3421 req->r_session = ceph_get_mds_session(session);
dc69e2e9 3422
38d46409
XL		 3423 doutc(cl, "mds%d session %p state %s\n", mds, session,
3424 ceph_session_state_name(session->s_state));
6eb06c46
XL		 3425
3426 /*
3427 * The old ceph will crash the MDSs when see unknown OPs
3428 */
3429 if (req->r_feature_needed > 0 &&
3430 !test_bit(req->r_feature_needed, &session->s_features)) {
3431 err = -EOPNOTSUPP;
3432 goto out_session;
3433 }
3434
2f2dc053
SW		 3435 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
3436 session->s_state != CEPH_MDS_SESSION_HUNG) {
3bb48b41
JL		 3437 /*
3438 * We cannot queue async requests since the caps and delegated
3439 * inodes are bound to the session. Just return -EJUKEBOX and
3440 * let the caller retry a sync request in that case.
3441 */
3442 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
3443 err = -EJUKEBOX;
3444 goto out_session;
3445 }
4ae3713f
JL		 3446
3447 /*
3448 * If the session has been REJECTED, then return a hard error,
3449 * unless it's a CLEANRECOVER mount, in which case we'll queue
3450 * it to the mdsc queue.
3451 */
3452 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
3453 if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
3454 list_add(&req->r_wait, &mdsc->waiting_for_map);
3455 else
3456 err = -EACCES;
3457 goto out_session;
3458 }
3459
2f2dc053 3460 if (session->s_state == CEPH_MDS_SESSION_NEW ||
c4853e97 3461 session->s_state == CEPH_MDS_SESSION_CLOSING) {
b682c6d4
XL		 3462 err = __open_session(mdsc, session);
3463 if (err)
3464 goto out_session;
c4853e97
XL		 3465 /* retry the same mds later */
3466 if (random)
3467 req->r_resend_mds = mds;
3468 }
2f2dc053
SW		 3469 list_add(&req->r_wait, &session->s_waiting);
3470 goto out_session;
3471 }
3472
3473 /* send request */
2f2dc053
SW		 3474 req->r_resend_mds = -1; /* forget any previous mds hint */
3475
3476 if (req->r_request_started == 0) /* note request start time */
3477 req->r_request_started = jiffies;
3478
00061645
XL		 3479 /*
3480 * For async create we will choose the auth MDS of frag in parent
3481 * directory to send the request and ususally this works fine, but
3482 * if the migrated the dirtory to another MDS before it could handle
3483 * it the request will be forwarded.
3484 *
3485 * And then the auth cap will be changed.
3486 */
3487 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && req->r_num_fwd) {
3488 struct ceph_dentry_info *di = ceph_dentry(req->r_dentry);
3489 struct ceph_inode_info *ci;
3490 struct ceph_cap *cap;
3491
3492 /*
3493 * The request maybe handled very fast and the new inode
3494 * hasn't been linked to the dentry yet. We need to wait
3495 * for the ceph_finish_async_create(), which shouldn't be
3496 * stuck too long or fail in thoery, to finish when forwarding
3497 * the request.
3498 */
3499 if (!d_inode(req->r_dentry)) {
3500 err = wait_on_bit(&di->flags, CEPH_DENTRY_ASYNC_CREATE_BIT,
3501 TASK_KILLABLE);
3502 if (err) {
3503 mutex_lock(&req->r_fill_mutex);
3504 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3505 mutex_unlock(&req->r_fill_mutex);
3506 goto out_session;
3507 }
3508 }
3509
3510 ci = ceph_inode(d_inode(req->r_dentry));
3511
3512 spin_lock(&ci->i_ceph_lock);
3513 cap = ci->i_auth_cap;
3514 if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE && mds != cap->mds) {
38d46409
XL		 3515 doutc(cl, "session changed for auth cap %d -> %d\n",
3516 cap->session->s_mds, session->s_mds);
00061645
XL		 3517
3518 /* Remove the auth cap from old session */
3519 spin_lock(&cap->session->s_cap_lock);
3520 cap->session->s_nr_caps--;
3521 list_del_init(&cap->session_caps);
3522 spin_unlock(&cap->session->s_cap_lock);
3523
3524 /* Add the auth cap to the new session */
3525 cap->mds = mds;
3526 cap->session = session;
3527 spin_lock(&session->s_cap_lock);
3528 session->s_nr_caps++;
3529 list_add_tail(&cap->session_caps, &session->s_caps);
3530 spin_unlock(&session->s_cap_lock);
3531
3532 change_auth_cap_ses(ci, session);
3533 }
3534 spin_unlock(&ci->i_ceph_lock);
3535 }
3536
396bd62c 3537 err = __send_request(session, req, false);
2f2dc053
SW		 3538
3539out_session:
3540 ceph_put_mds_session(session);
48fec5d0
YZ		 3541finish:
3542 if (err) {
38d46409 3543 doutc(cl, "early error %d\n", err);
48fec5d0
YZ		 3544 req->r_err = err;
3545 complete_request(mdsc, req);
3546 __unregister_request(mdsc, req);
3547 }
d5548492 3548 return;
2f2dc053
SW		 3549}
3550
3551/*
3552 * called under mdsc->mutex
3553 */
3554static void __wake_requests(struct ceph_mds_client *mdsc,
3555 struct list_head *head)
3556{
38d46409 3557 struct ceph_client *cl = mdsc->fsc->client;
ed75ec2c
YZ		 3558 struct ceph_mds_request *req;
3559 LIST_HEAD(tmp_list);
3560
3561 list_splice_init(head, &tmp_list);
2f2dc053 3562
ed75ec2c
YZ		 3563 while (!list_empty(&tmp_list)) {
3564 req = list_entry(tmp_list.next,
3565 struct ceph_mds_request, r_wait);
2f2dc053 3566 list_del_init(&req->r_wait);
38d46409
XL		 3567 doutc(cl, " wake request %p tid %llu\n", req,
3568 req->r_tid);
2f2dc053
SW		 3569 __do_request(mdsc, req);
3570 }
3571}
3572
3573/*
3574 * Wake up threads with requests pending for @mds, so that they can
29790f26 3575 * resubmit their requests to a possibly different mds.
2f2dc053 3576 */
29790f26 3577static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 3578{
38d46409 3579 struct ceph_client *cl = mdsc->fsc->client;
44ca18f2 3580 struct ceph_mds_request *req;
282c1052 3581 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053 3582
38d46409 3583 doutc(cl, "kick_requests mds%d\n", mds);
282c1052 3584 while (p) {
44ca18f2 3585 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 3586 p = rb_next(p);
bc2de10d 3587 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
44ca18f2 3588 continue;
3de22be6
YZ		 3589 if (req->r_attempts > 0)
3590 continue; /* only new requests */
44ca18f2
SW		 3591 if (req->r_session &&
3592 req->r_session->s_mds == mds) {
38d46409 3593 doutc(cl, " kicking tid %llu\n", req->r_tid);
03974e81 3594 list_del_init(&req->r_wait);
44ca18f2 3595 __do_request(mdsc, req);
2f2dc053
SW		 3596 }
3597 }
3598}
3599
86bda539 3600int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
2f2dc053
SW		 3601 struct ceph_mds_request *req)
3602{
38d46409 3603 struct ceph_client *cl = mdsc->fsc->client;
891f3f5a 3604 int err = 0;
86bda539
JL		 3605
3606 /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
3607 if (req->r_inode)
3608 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
9c1c2b35 3609 if (req->r_parent) {
719a2514
YZ		 3610 struct ceph_inode_info *ci = ceph_inode(req->r_parent);
3611 int fmode = (req->r_op & CEPH_MDS_OP_WRITE) ?
3612 CEPH_FILE_MODE_WR : CEPH_FILE_MODE_RD;
3613 spin_lock(&ci->i_ceph_lock);
3614 ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
3615 __ceph_touch_fmode(ci, mdsc, fmode);
3616 spin_unlock(&ci->i_ceph_lock);
9c1c2b35 3617 }
86bda539
JL		 3618 if (req->r_old_dentry_dir)
3619 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
3620 CEPH_CAP_PIN);
3621
891f3f5a
JL		 3622 if (req->r_inode) {
3623 err = ceph_wait_on_async_create(req->r_inode);
3624 if (err) {
38d46409 3625 doutc(cl, "wait for async create returned: %d\n", err);
891f3f5a
JL		 3626 return err;
3627 }
3628 }
3629
3630 if (!err && req->r_old_inode) {
3631 err = ceph_wait_on_async_create(req->r_old_inode);
3632 if (err) {
38d46409 3633 doutc(cl, "wait for async create returned: %d\n", err);
891f3f5a
JL		 3634 return err;
3635 }
3636 }
3637
38d46409 3638 doutc(cl, "submit_request on %p for inode %p\n", req, dir);
2f2dc053 3639 mutex_lock(&mdsc->mutex);
86bda539 3640 __register_request(mdsc, req, dir);
2f2dc053 3641 __do_request(mdsc, req);
86bda539 3642 err = req->r_err;
2f2dc053 3643 mutex_unlock(&mdsc->mutex);
86bda539 3644 return err;
2f2dc053
SW		 3645}
3646
9eaa7b79
JL		 3647int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
3648 struct ceph_mds_request *req,
3649 ceph_mds_request_wait_callback_t wait_func)
2f2dc053 3650{
38d46409 3651 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 3652 int err;
3653
e1518c7c 3654 /* wait */
38d46409 3655 doutc(cl, "do_request waiting\n");
9eaa7b79
JL		 3656 if (wait_func) {
3657 err = wait_func(mdsc, req);
e1518c7c 3658 } else {
5be73034
ID		 3659 long timeleft = wait_for_completion_killable_timeout(
3660 &req->r_completion,
3661 ceph_timeout_jiffies(req->r_timeout));
3662 if (timeleft > 0)
3663 err = 0;
3664 else if (!timeleft)
8ccf7fcc 3665 err = -ETIMEDOUT; /* timed out */
5be73034
ID		 3666 else
3667 err = timeleft; /* killed */
e1518c7c 3668 }
38d46409 3669 doutc(cl, "do_request waited, got %d\n", err);
e1518c7c 3670 mutex_lock(&mdsc->mutex);
5b1daecd 3671
e1518c7c 3672 /* only abort if we didn't race with a real reply */
bc2de10d 3673 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
e1518c7c
SW		 3674 err = le32_to_cpu(req->r_reply_info.head->result);
3675 } else if (err < 0) {
38d46409 3676 doutc(cl, "aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 3677
3678 /*
3679 * ensure we aren't running concurrently with
3680 * ceph_fill_trace or ceph_readdir_prepopulate, which
3681 * rely on locks (dir mutex) held by our caller.
3682 */
3683 mutex_lock(&req->r_fill_mutex);
e1518c7c 3684 req->r_err = err;
bc2de10d 3685 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
b4556396 3686 mutex_unlock(&req->r_fill_mutex);
5b1daecd 3687
3dd69aab 3688 if (req->r_parent &&
167c9e35
SW		 3689 (req->r_op & CEPH_MDS_OP_WRITE))
3690 ceph_invalidate_dir_request(req);
2f2dc053 3691 } else {
e1518c7c 3692 err = req->r_err;
2f2dc053 3693 }
2f2dc053 3694
e1518c7c 3695 mutex_unlock(&mdsc->mutex);
8340f22c
JL		 3696 return err;
3697}
3698
3699/*
3700 * Synchrously perform an mds request. Take care of all of the
3701 * session setup, forwarding, retry details.
3702 */
3703int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
3704 struct inode *dir,
3705 struct ceph_mds_request *req)
3706{
38d46409 3707 struct ceph_client *cl = mdsc->fsc->client;
8340f22c
JL		 3708 int err;
3709
38d46409 3710 doutc(cl, "do_request on %p\n", req);
8340f22c
JL		 3711
3712 /* issue */
3713 err = ceph_mdsc_submit_request(mdsc, dir, req);
3714 if (!err)
9eaa7b79 3715 err = ceph_mdsc_wait_request(mdsc, req, NULL);
38d46409 3716 doutc(cl, "do_request %p done, result %d\n", req, err);
2f2dc053
SW		 3717 return err;
3718}
3719
167c9e35 3720/*
2f276c51 3721 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 3722 * namespace request.
3723 */
3724void ceph_invalidate_dir_request(struct ceph_mds_request *req)
3725{
8d8f371c
YZ		 3726 struct inode *dir = req->r_parent;
3727 struct inode *old_dir = req->r_old_dentry_dir;
38d46409 3728 struct ceph_client *cl = req->r_mdsc->fsc->client;
167c9e35 3729
38d46409
XL		 3730 doutc(cl, "invalidate_dir_request %p %p (complete, lease(s))\n",
3731 dir, old_dir);
167c9e35 3732
8d8f371c
YZ		 3733 ceph_dir_clear_complete(dir);
3734 if (old_dir)
3735 ceph_dir_clear_complete(old_dir);
167c9e35
SW		 3736 if (req->r_dentry)
3737 ceph_invalidate_dentry_lease(req->r_dentry);
3738 if (req->r_old_dentry)
3739 ceph_invalidate_dentry_lease(req->r_old_dentry);
3740}
3741
2f2dc053
SW		 3742/*
3743 * Handle mds reply.
3744 *
3745 * We take the session mutex and parse and process the reply immediately.
3746 * This preserves the logical ordering of replies, capabilities, etc., sent
3747 * by the MDS as they are applied to our local cache.
3748 */
3749static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
3750{
3751 struct ceph_mds_client *mdsc = session->s_mdsc;
38d46409 3752 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 3753 struct ceph_mds_request *req;
3754 struct ceph_mds_reply_head *head = msg->front.iov_base;
3755 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 3756 struct ceph_snap_realm *realm;
2f2dc053
SW		 3757 u64 tid;
3758 int err, result;
2600d2dd 3759 int mds = session->s_mds;
a68e564a 3760 bool close_sessions = false;
2f2dc053 3761
2f2dc053 3762 if (msg->front.iov_len < sizeof(*head)) {
38d46409 3763 pr_err_client(cl, "got corrupt (short) reply\n");
9ec7cab1 3764 ceph_msg_dump(msg);
2f2dc053
SW		 3765 return;
3766 }
3767
3768 /* get request, session */
6df058c0 3769 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053 3770 mutex_lock(&mdsc->mutex);
fcd00b68 3771 req = lookup_get_request(mdsc, tid);
2f2dc053 3772 if (!req) {
38d46409 3773 doutc(cl, "on unknown tid %llu\n", tid);
2f2dc053
SW		 3774 mutex_unlock(&mdsc->mutex);
3775 return;
3776 }
38d46409 3777 doutc(cl, "handle_reply %p\n", req);
2f2dc053
SW		 3778
3779 /* correct session? */
d96d6049 3780 if (req->r_session != session) {
38d46409
XL		 3781 pr_err_client(cl, "got %llu on session mds%d not mds%d\n",
3782 tid, session->s_mds,
3783 req->r_session ? req->r_session->s_mds : -1);
2f2dc053
SW		 3784 mutex_unlock(&mdsc->mutex);
3785 goto out;
3786 }
3787
3788 /* dup? */
bc2de10d
JL		 3789 if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
3790 (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
38d46409
XL		 3791 pr_warn_client(cl, "got a dup %s reply on %llu from mds%d\n",
3792 head->safe ? "safe" : "unsafe", tid, mds);
2f2dc053
SW		 3793 mutex_unlock(&mdsc->mutex);
3794 goto out;
3795 }
bc2de10d 3796 if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
38d46409
XL		 3797 pr_warn_client(cl, "got unsafe after safe on %llu from mds%d\n",
3798 tid, mds);
85792d0d
SW		 3799 mutex_unlock(&mdsc->mutex);
3800 goto out;
3801 }
2f2dc053
SW		 3802
3803 result = le32_to_cpu(head->result);
3804
2f2dc053 3805 if (head->safe) {
bc2de10d 3806 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
2f2dc053 3807 __unregister_request(mdsc, req);
2f2dc053 3808
07edc057
XL		 3809 /* last request during umount? */
3810 if (mdsc->stopping && !__get_oldest_req(mdsc))
3811 complete_all(&mdsc->safe_umount_waiters);
3812
bc2de10d 3813 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2f2dc053
SW		 3814 /*
3815 * We already handled the unsafe response, now do the
3816 * cleanup. No need to examine the response; the MDS
3817 * doesn't include any result info in the safe
3818 * response. And even if it did, there is nothing
3819 * useful we could do with a revised return value.
3820 */
38d46409 3821 doutc(cl, "got safe reply %llu, mds%d\n", tid, mds);
2f2dc053 3822
2f2dc053
SW		 3823 mutex_unlock(&mdsc->mutex);
3824 goto out;
3825 }
e1518c7c 3826 } else {
bc2de10d 3827 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
2f2dc053
SW		 3828 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
3829 }
3830
38d46409 3831 doutc(cl, "tid %lld result %d\n", tid, result);
b37fe1f9 3832 if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
3859af9e 3833 err = parse_reply_info(session, msg, req, (u64)-1);
b37fe1f9 3834 else
3859af9e
XL		 3835 err = parse_reply_info(session, msg, req,
3836 session->s_con.peer_features);
2f2dc053
SW		 3837 mutex_unlock(&mdsc->mutex);
3838
bca9fc14 3839 /* Must find target inode outside of mutexes to avoid deadlocks */
3859af9e 3840 rinfo = &req->r_reply_info;
bca9fc14 3841 if ((err >= 0) && rinfo->head->is_target) {
ec9595c0 3842 struct inode *in = xchg(&req->r_new_inode, NULL);
bca9fc14
JL		 3843 struct ceph_vino tvino = {
3844 .ino = le64_to_cpu(rinfo->targeti.in->ino),
3845 .snap = le64_to_cpu(rinfo->targeti.in->snapid)
3846 };
3847
ec9595c0
JL		 3848 /*
3849 * If we ended up opening an existing inode, discard
3850 * r_new_inode
3851 */
3852 if (req->r_op == CEPH_MDS_OP_CREATE &&
3853 !req->r_reply_info.has_create_ino) {
3854 /* This should never happen on an async create */
3855 WARN_ON_ONCE(req->r_deleg_ino);
3856 iput(in);
3857 in = NULL;
3858 }
3859
3860 in = ceph_get_inode(mdsc->fsc->sb, tvino, in);
bca9fc14
JL		 3861 if (IS_ERR(in)) {
3862 err = PTR_ERR(in);
3863 mutex_lock(&session->s_mutex);
3864 goto out_err;
3865 }
3866 req->r_target_inode = in;
3867 }
3868
2f2dc053
SW		 3869 mutex_lock(&session->s_mutex);
3870 if (err < 0) {
38d46409
XL		 3871 pr_err_client(cl, "got corrupt reply mds%d(tid:%lld)\n",
3872 mds, tid);
9ec7cab1 3873 ceph_msg_dump(msg);
2f2dc053
SW		 3874 goto out_err;
3875 }
3876
3877 /* snap trace */
982d6011 3878 realm = NULL;
2f2dc053
SW		 3879 if (rinfo->snapblob_len) {
3880 down_write(&mdsc->snap_rwsem);
a68e564a 3881 err = ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 3882 rinfo->snapblob + rinfo->snapblob_len,
3883 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
3884 &realm);
a68e564a
XL		 3885 if (err) {
3886 up_write(&mdsc->snap_rwsem);
3887 close_sessions = true;
3888 if (err == -EIO)
3889 ceph_msg_dump(msg);
3890 goto out_err;
3891 }
2f2dc053
SW		 3892 downgrade_write(&mdsc->snap_rwsem);
3893 } else {
3894 down_read(&mdsc->snap_rwsem);
3895 }
3896
3897 /* insert trace into our cache */
b4556396 3898 mutex_lock(&req->r_fill_mutex);
315f2408 3899 current->journal_info = req;
f5a03b08 3900 err = ceph_fill_trace(mdsc->fsc->sb, req);
2f2dc053 3901 if (err == 0) {
6e8575fa 3902 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 3903 req->r_op == CEPH_MDS_OP_LSSNAP))
af9ffa6d 3904 err = ceph_readdir_prepopulate(req, req->r_session);
2f2dc053 3905 }
315f2408 3906 current->journal_info = NULL;
b4556396 3907 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 3908
3909 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 3910 if (realm)
3911 ceph_put_snap_realm(mdsc, realm);
68cd5b4b 3912
fe33032d
YZ		 3913 if (err == 0) {
3914 if (req->r_target_inode &&
3915 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3916 struct ceph_inode_info *ci =
3917 ceph_inode(req->r_target_inode);
3918 spin_lock(&ci->i_unsafe_lock);
3919 list_add_tail(&req->r_unsafe_target_item,
3920 &ci->i_unsafe_iops);
3921 spin_unlock(&ci->i_unsafe_lock);
3922 }
3923
3924 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
68cd5b4b 3925 }
2f2dc053 3926out_err:
e1518c7c 3927 mutex_lock(&mdsc->mutex);
bc2de10d 3928 if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
e1518c7c
SW		 3929 if (err) {
3930 req->r_err = err;
3931 } else {
5fdb1389 3932 req->r_reply = ceph_msg_get(msg);
bc2de10d 3933 set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
e1518c7c 3934 }
2f2dc053 3935 } else {
38d46409 3936 doutc(cl, "reply arrived after request %lld was aborted\n", tid);
2f2dc053 3937 }
e1518c7c 3938 mutex_unlock(&mdsc->mutex);
2f2dc053 3939
2f2dc053
SW		 3940 mutex_unlock(&session->s_mutex);
3941
3942 /* kick calling process */
3943 complete_request(mdsc, req);
70c94820 3944
8ae99ae2 3945 ceph_update_metadata_metrics(&mdsc->metric, req->r_start_latency,
70c94820 3946 req->r_end_latency, err);
2f2dc053
SW		 3947out:
3948 ceph_mdsc_put_request(req);
a68e564a
XL		 3949
3950 /* Defer closing the sessions after s_mutex lock being released */
3951 if (close_sessions)
3952 ceph_mdsc_close_sessions(mdsc);
2f2dc053
SW		 3953 return;
3954}
3955
3956
3957
3958/*
3959 * handle mds notification that our request has been forwarded.
3960 */
2600d2dd
SW		 3961static void handle_forward(struct ceph_mds_client *mdsc,
3962 struct ceph_mds_session *session,
3963 struct ceph_msg *msg)
2f2dc053 3964{
38d46409 3965 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053 3966 struct ceph_mds_request *req;
a1ea787c 3967 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 3968 u32 next_mds;
3969 u32 fwd_seq;
2f2dc053
SW		 3970 int err = -EINVAL;
3971 void *p = msg->front.iov_base;
3972 void *end = p + msg->front.iov_len;
1980b1bf 3973 bool aborted = false;
2f2dc053 3974
a1ea787c 3975 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 3976 next_mds = ceph_decode_32(&p);
3977 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 3978
3979 mutex_lock(&mdsc->mutex);
fcd00b68 3980 req = lookup_get_request(mdsc, tid);
2f2dc053 3981 if (!req) {
1980b1bf 3982 mutex_unlock(&mdsc->mutex);
38d46409 3983 doutc(cl, "forward tid %llu to mds%d - req dne\n", tid, next_mds);
1980b1bf 3984 return; /* dup reply? */
2f2dc053
SW		 3985 }
3986
bc2de10d 3987 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
38d46409 3988 doutc(cl, "forward tid %llu aborted, unregistering\n", tid);
2a8e5e36 3989 __unregister_request(mdsc, req);
ce0d5bd3 3990 } else if (fwd_seq <= req->r_num_fwd || (uint32_t)fwd_seq >= U32_MAX) {
1980b1bf 3991 /*
ce0d5bd3 3992 * Avoid inifinite retrying after overflow.
1980b1bf 3993 *
ce0d5bd3
XL		 3994 * The MDS will increase the fwd count and in client side
3995 * if the num_fwd is less than the one saved in request
3996 * that means the MDS is an old version and overflowed of
3997 * 8 bits.
1980b1bf 3998 */
ce0d5bd3
XL		 3999 mutex_lock(&req->r_fill_mutex);
4000 req->r_err = -EMULTIHOP;
4001 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
4002 mutex_unlock(&req->r_fill_mutex);
4003 aborted = true;
38d46409
XL		 4004 pr_warn_ratelimited_client(cl, "forward tid %llu seq overflow\n",
4005 tid);
2f2dc053
SW		 4006 } else {
4007 /* resend. forward race not possible; mds would drop */
38d46409 4008 doutc(cl, "forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2a8e5e36 4009 BUG_ON(req->r_err);
bc2de10d 4010 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3de22be6 4011 req->r_attempts = 0;
2f2dc053
SW		 4012 req->r_num_fwd = fwd_seq;
4013 req->r_resend_mds = next_mds;
4014 put_request_session(req);
4015 __do_request(mdsc, req);
4016 }
2f2dc053 4017 mutex_unlock(&mdsc->mutex);
1980b1bf
XL		 4018
4019 /* kick calling process */
4020 if (aborted)
4021 complete_request(mdsc, req);
4022 ceph_mdsc_put_request(req);
2f2dc053
SW		 4023 return;
4024
4025bad:
38d46409 4026 pr_err_client(cl, "decode error err=%d\n", err);
8b0da5c5 4027 ceph_msg_dump(msg);
2f2dc053
SW		 4028}
4029
131d7eb4 4030static int __decode_session_metadata(void **p, void *end,
0b98acd6 4031 bool *blocklisted)
84bf3950
YZ		 4032{
4033 /* map<string,string> */
4034 u32 n;
131d7eb4 4035 bool err_str;
84bf3950
YZ		 4036 ceph_decode_32_safe(p, end, n, bad);
4037 while (n-- > 0) {
4038 u32 len;
4039 ceph_decode_32_safe(p, end, len, bad);
4040 ceph_decode_need(p, end, len, bad);
131d7eb4 4041 err_str = !strncmp(*p, "error_string", len);
84bf3950
YZ		 4042 *p += len;
4043 ceph_decode_32_safe(p, end, len, bad);
4044 ceph_decode_need(p, end, len, bad);
4bb926e8
ID		 4045 /*
4046 * Match "blocklisted (blacklisted)" from newer MDSes,
4047 * or "blacklisted" from older MDSes.
4048 */
131d7eb4 4049 if (err_str && strnstr(*p, "blacklisted", len))
0b98acd6 4050 *blocklisted = true;
84bf3950
YZ		 4051 *p += len;
4052 }
4053 return 0;
4054bad:
4055 return -1;
4056}
4057
2f2dc053
SW		 4058/*
4059 * handle a mds session control message
4060 */
4061static void handle_session(struct ceph_mds_session *session,
4062 struct ceph_msg *msg)
4063{
4064 struct ceph_mds_client *mdsc = session->s_mdsc;
38d46409 4065 struct ceph_client *cl = mdsc->fsc->client;
84bf3950
YZ		 4066 int mds = session->s_mds;
4067 int msg_version = le16_to_cpu(msg->hdr.version);
4068 void *p = msg->front.iov_base;
4069 void *end = p + msg->front.iov_len;
4070 struct ceph_mds_session_head *h;
2f2dc053 4071 u32 op;
0fa82633 4072 u64 seq, features = 0;
2f2dc053 4073 int wake = 0;
0b98acd6 4074 bool blocklisted = false;
2f2dc053 4075
2f2dc053 4076 /* decode */
84bf3950
YZ		 4077 ceph_decode_need(&p, end, sizeof(*h), bad);
4078 h = p;
4079 p += sizeof(*h);
4080
2f2dc053
SW		 4081 op = le32_to_cpu(h->op);
4082 seq = le64_to_cpu(h->seq);
4083
84bf3950
YZ		 4084 if (msg_version >= 3) {
4085 u32 len;
e1c9788c
KH		 4086 /* version >= 2 and < 5, decode metadata, skip otherwise
4087 * as it's handled via flags.
4088 */
4089 if (msg_version >= 5)
4090 ceph_decode_skip_map(&p, end, string, string, bad);
4091 else if (__decode_session_metadata(&p, end, &blocklisted) < 0)
84bf3950 4092 goto bad;
e1c9788c 4093
84bf3950
YZ		 4094 /* version >= 3, feature bits */
4095 ceph_decode_32_safe(&p, end, len, bad);
02e37571
JL		 4096 if (len) {
4097 ceph_decode_64_safe(&p, end, features, bad);
4098 p += len - sizeof(features);
4099 }
84bf3950
YZ		 4100 }
4101
e1c9788c 4102 if (msg_version >= 5) {
ea16567f
LH		 4103 u32 flags, len;
4104
4105 /* version >= 4 */
4106 ceph_decode_skip_16(&p, end, bad); /* struct_v, struct_cv */
4107 ceph_decode_32_safe(&p, end, len, bad); /* len */
4108 ceph_decode_skip_n(&p, end, len, bad); /* metric_spec */
4109
e1c9788c 4110 /* version >= 5, flags */
ea16567f 4111 ceph_decode_32_safe(&p, end, flags, bad);
e1c9788c 4112 if (flags & CEPH_SESSION_BLOCKLISTED) {
38d46409
XL		 4113 pr_warn_client(cl, "mds%d session blocklisted\n",
4114 session->s_mds);
e1c9788c
KH		 4115 blocklisted = true;
4116 }
4117 }
4118
2f2dc053 4119 mutex_lock(&mdsc->mutex);
0a07fc8c 4120 if (op == CEPH_SESSION_CLOSE) {
5b3248c6 4121 ceph_get_mds_session(session);
2600d2dd 4122 __unregister_session(mdsc, session);
0a07fc8c 4123 }
2f2dc053
SW		 4124 /* FIXME: this ttl calculation is generous */
4125 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
4126 mutex_unlock(&mdsc->mutex);
4127
4128 mutex_lock(&session->s_mutex);
4129
38d46409
XL		 4130 doutc(cl, "mds%d %s %p state %s seq %llu\n", mds,
4131 ceph_session_op_name(op), session,
4132 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 4133
4134 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
4135 session->s_state = CEPH_MDS_SESSION_OPEN;
38d46409 4136 pr_info_client(cl, "mds%d came back\n", session->s_mds);
2f2dc053
SW		 4137 }
4138
4139 switch (op) {
4140 case CEPH_SESSION_OPEN:
29790f26 4141 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
38d46409
XL		 4142 pr_info_client(cl, "mds%d reconnect success\n",
4143 session->s_mds);
300e42a2 4144
f48e0342 4145 session->s_features = features;
300e42a2 4146 if (session->s_state == CEPH_MDS_SESSION_OPEN) {
38d46409
XL		 4147 pr_notice_client(cl, "mds%d is already opened\n",
4148 session->s_mds);
300e42a2
XL		 4149 } else {
4150 session->s_state = CEPH_MDS_SESSION_OPEN;
300e42a2
XL		 4151 renewed_caps(mdsc, session, 0);
4152 if (test_bit(CEPHFS_FEATURE_METRIC_COLLECT,
4153 &session->s_features))
4154 metric_schedule_delayed(&mdsc->metric);
4155 }
4156
4157 /*
4158 * The connection maybe broken and the session in client
4159 * side has been reinitialized, need to update the seq
4160 * anyway.
4161 */
4162 if (!session->s_seq && seq)
4163 session->s_seq = seq;
4164
2f2dc053
SW		 4165 wake = 1;
4166 if (mdsc->stopping)
4167 __close_session(mdsc, session);
4168 break;
4169
4170 case CEPH_SESSION_RENEWCAPS:
4171 if (session->s_renew_seq == seq)
4172 renewed_caps(mdsc, session, 1);
4173 break;
4174
4175 case CEPH_SESSION_CLOSE:
29790f26 4176 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
38d46409
XL		 4177 pr_info_client(cl, "mds%d reconnect denied\n",
4178 session->s_mds);
4d681c2f 4179 session->s_state = CEPH_MDS_SESSION_CLOSED;
1c841a96 4180 cleanup_session_requests(mdsc, session);
2f2dc053 4181 remove_session_caps(session);
656e4382 4182 wake = 2; /* for good measure */
f3c60c59 4183 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 4184 break;
4185
4186 case CEPH_SESSION_STALE:
38d46409
XL		 4187 pr_info_client(cl, "mds%d caps went stale, renewing\n",
4188 session->s_mds);
52d60f8e 4189 atomic_inc(&session->s_cap_gen);
1ce208a6 4190 session->s_cap_ttl = jiffies - 1;
2f2dc053
SW		 4191 send_renew_caps(mdsc, session);
4192 break;
4193
4194 case CEPH_SESSION_RECALL_STATE:
e30ee581 4195 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2f2dc053
SW		 4196 break;
4197
186e4f7a 4198 case CEPH_SESSION_FLUSHMSG:
e7d84c6a
XL		 4199 /* flush cap releases */
4200 spin_lock(&session->s_cap_lock);
4201 if (session->s_num_cap_releases)
4202 ceph_flush_cap_releases(mdsc, session);
4203 spin_unlock(&session->s_cap_lock);
4204
186e4f7a
YZ		 4205 send_flushmsg_ack(mdsc, session, seq);
4206 break;
4207
03f4fcb0 4208 case CEPH_SESSION_FORCE_RO:
38d46409 4209 doutc(cl, "force_session_readonly %p\n", session);
03f4fcb0
YZ		 4210 spin_lock(&session->s_cap_lock);
4211 session->s_readonly = true;
4212 spin_unlock(&session->s_cap_lock);
d2f8bb27 4213 wake_up_session_caps(session, FORCE_RO);
03f4fcb0
YZ		 4214 break;
4215
fcff415c
YZ		 4216 case CEPH_SESSION_REJECT:
4217 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
38d46409
XL		 4218 pr_info_client(cl, "mds%d rejected session\n",
4219 session->s_mds);
fcff415c
YZ		 4220 session->s_state = CEPH_MDS_SESSION_REJECTED;
4221 cleanup_session_requests(mdsc, session);
4222 remove_session_caps(session);
0b98acd6
ID		 4223 if (blocklisted)
4224 mdsc->fsc->blocklisted = true;
fcff415c
YZ		 4225 wake = 2; /* for good measure */
4226 break;
4227
2f2dc053 4228 default:
38d46409 4229 pr_err_client(cl, "bad op %d mds%d\n", op, mds);
2f2dc053
SW		 4230 WARN_ON(1);
4231 }
4232
4233 mutex_unlock(&session->s_mutex);
4234 if (wake) {
4235 mutex_lock(&mdsc->mutex);
4236 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 4237 if (wake == 2)
4238 kick_requests(mdsc, mds);
2f2dc053
SW		 4239 mutex_unlock(&mdsc->mutex);
4240 }
0a07fc8c
YZ		 4241 if (op == CEPH_SESSION_CLOSE)
4242 ceph_put_mds_session(session);
2f2dc053
SW		 4243 return;
4244
4245bad:
38d46409
XL		 4246 pr_err_client(cl, "corrupt message mds%d len %d\n", mds,
4247 (int)msg->front.iov_len);
9ec7cab1 4248 ceph_msg_dump(msg);
2f2dc053
SW		 4249 return;
4250}
4251
a25949b9
JL		 4252void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
4253{
38d46409 4254 struct ceph_client *cl = req->r_mdsc->fsc->client;
a25949b9
JL		 4255 int dcaps;
4256
4257 dcaps = xchg(&req->r_dir_caps, 0);
4258 if (dcaps) {
38d46409 4259 doutc(cl, "releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
a25949b9
JL		 4260 ceph_put_cap_refs(ceph_inode(req->r_parent), dcaps);
4261 }
4262}
4263
07045648 4264void ceph_mdsc_release_dir_caps_async(struct ceph_mds_request *req)
e64f44a8 4265{
38d46409 4266 struct ceph_client *cl = req->r_mdsc->fsc->client;
e64f44a8
XL		 4267 int dcaps;
4268
4269 dcaps = xchg(&req->r_dir_caps, 0);
4270 if (dcaps) {
38d46409 4271 doutc(cl, "releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
07045648 4272 ceph_put_cap_refs_async(ceph_inode(req->r_parent), dcaps);
e64f44a8
XL		 4273 }
4274}
4275
2f2dc053
SW		 4276/*
4277 * called under session->mutex.
4278 */
4279static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
4280 struct ceph_mds_session *session)
4281{
4282 struct ceph_mds_request *req, *nreq;
3de22be6 4283 struct rb_node *p;
2f2dc053 4284
38d46409 4285 doutc(mdsc->fsc->client, "mds%d\n", session->s_mds);
2f2dc053
SW		 4286
4287 mutex_lock(&mdsc->mutex);
9cf54563 4288 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item)
396bd62c 4289 __send_request(session, req, true);
3de22be6
YZ		 4290
4291 /*
4292 * also re-send old requests when MDS enters reconnect stage. So that MDS
4293 * can process completed request in clientreplay stage.
4294 */
4295 p = rb_first(&mdsc->request_tree);
4296 while (p) {
4297 req = rb_entry(p, struct ceph_mds_request, r_node);
4298 p = rb_next(p);
bc2de10d 4299 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3de22be6
YZ		 4300 continue;
4301 if (req->r_attempts == 0)
4302 continue; /* only old requests */
a25949b9
JL		 4303 if (!req->r_session)
4304 continue;
4305 if (req->r_session->s_mds != session->s_mds)
4306 continue;
4307
07045648 4308 ceph_mdsc_release_dir_caps_async(req);
a25949b9 4309
396bd62c 4310 __send_request(session, req, true);
3de22be6 4311 }
2f2dc053
SW		 4312 mutex_unlock(&mdsc->mutex);
4313}
4314
81c5a148
YZ		 4315static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
4316{
4317 struct ceph_msg *reply;
4318 struct ceph_pagelist *_pagelist;
4319 struct page *page;
4320 __le32 *addr;
4321 int err = -ENOMEM;
4322
4323 if (!recon_state->allow_multi)
4324 return -ENOSPC;
4325
4326 /* can't handle message that contains both caps and realm */
4327 BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
4328
4329 /* pre-allocate new pagelist */
4330 _pagelist = ceph_pagelist_alloc(GFP_NOFS);
4331 if (!_pagelist)
4332 return -ENOMEM;
4333
4334 reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
4335 if (!reply)
4336 goto fail_msg;
4337
4338 /* placeholder for nr_caps */
4339 err = ceph_pagelist_encode_32(_pagelist, 0);
4340 if (err < 0)
4341 goto fail;
4342
4343 if (recon_state->nr_caps) {
4344 /* currently encoding caps */
4345 err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
4346 if (err)
4347 goto fail;
4348 } else {
4349 /* placeholder for nr_realms (currently encoding relams) */
4350 err = ceph_pagelist_encode_32(_pagelist, 0);
4351 if (err < 0)
4352 goto fail;
4353 }
4354
4355 err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
4356 if (err)
4357 goto fail;
4358
4359 page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
4360 addr = kmap_atomic(page);
4361 if (recon_state->nr_caps) {
4362 /* currently encoding caps */
4363 *addr = cpu_to_le32(recon_state->nr_caps);
4364 } else {
4365 /* currently encoding relams */
4366 *(addr + 1) = cpu_to_le32(recon_state->nr_realms);
4367 }
4368 kunmap_atomic(addr);
4369
4370 reply->hdr.version = cpu_to_le16(5);
4371 reply->hdr.compat_version = cpu_to_le16(4);
4372
4373 reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
4374 ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
4375
4376 ceph_con_send(&recon_state->session->s_con, reply);
4377 ceph_pagelist_release(recon_state->pagelist);
4378
4379 recon_state->pagelist = _pagelist;
4380 recon_state->nr_caps = 0;
4381 recon_state->nr_realms = 0;
4382 recon_state->msg_version = 5;
4383 return 0;
4384fail:
4385 ceph_msg_put(reply);
4386fail_msg:
4387 ceph_pagelist_release(_pagelist);
4388 return err;
4389}
4390
a33f6432
YZ		 4391static struct dentry* d_find_primary(struct inode *inode)
4392{
4393 struct dentry *alias, *dn = NULL;
4394
4395 if (hlist_empty(&inode->i_dentry))
4396 return NULL;
4397
4398 spin_lock(&inode->i_lock);
4399 if (hlist_empty(&inode->i_dentry))
4400 goto out_unlock;
4401
4402 if (S_ISDIR(inode->i_mode)) {
4403 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
4404 if (!IS_ROOT(alias))
4405 dn = dget(alias);
4406 goto out_unlock;
4407 }
4408
4409 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
4410 spin_lock(&alias->d_lock);
4411 if (!d_unhashed(alias) &&
4412 (ceph_dentry(alias)->flags & CEPH_DENTRY_PRIMARY_LINK)) {
4413 dn = dget_dlock(alias);
4414 }
4415 spin_unlock(&alias->d_lock);
4416 if (dn)
4417 break;
4418 }
4419out_unlock:
4420 spin_unlock(&inode->i_lock);
4421 return dn;
4422}
4423
2f2dc053
SW		 4424/*
4425 * Encode information about a cap for a reconnect with the MDS.
4426 */
aaf67de7 4427static int reconnect_caps_cb(struct inode *inode, int mds, void *arg)
2f2dc053 4428{
197b7d79 4429 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
38d46409 4430 struct ceph_client *cl = ceph_inode_to_client(inode);
20cb34ae
SW		 4431 union {
4432 struct ceph_mds_cap_reconnect v2;
4433 struct ceph_mds_cap_reconnect_v1 v1;
4434 } rec;
aaf67de7 4435 struct ceph_inode_info *ci = ceph_inode(inode);
20cb34ae
SW		 4436 struct ceph_reconnect_state *recon_state = arg;
4437 struct ceph_pagelist *pagelist = recon_state->pagelist;
a33f6432 4438 struct dentry *dentry;
aaf67de7 4439 struct ceph_cap *cap;
a33f6432 4440 char *path;
9aaa7eb0 4441 int pathlen = 0, err;
a33f6432 4442 u64 pathbase;
3469ed0d 4443 u64 snap_follows;
2f2dc053 4444
a33f6432
YZ		 4445 dentry = d_find_primary(inode);
4446 if (dentry) {
4447 /* set pathbase to parent dir when msg_version >= 2 */
197b7d79 4448 path = ceph_mdsc_build_path(mdsc, dentry, &pathlen, &pathbase,
a33f6432
YZ		 4449 recon_state->msg_version >= 2);
4450 dput(dentry);
4451 if (IS_ERR(path)) {
4452 err = PTR_ERR(path);
4453 goto out_err;
4454 }
4455 } else {
4456 path = NULL;
a33f6432
YZ		 4457 pathbase = 0;
4458 }
4459
be655596 4460 spin_lock(&ci->i_ceph_lock);
aaf67de7
XL		 4461 cap = __get_cap_for_mds(ci, mds);
4462 if (!cap) {
4463 spin_unlock(&ci->i_ceph_lock);
9aaa7eb0 4464 err = 0;
aaf67de7
XL		 4465 goto out_err;
4466 }
38d46409
XL		 4467 doutc(cl, " adding %p ino %llx.%llx cap %p %lld %s\n", inode,
4468 ceph_vinop(inode), cap, cap->cap_id,
4469 ceph_cap_string(cap->issued));
aaf67de7 4470
2f2dc053
SW		 4471 cap->seq = 0; /* reset cap seq */
4472 cap->issue_seq = 0; /* and issue_seq */
667ca05c 4473 cap->mseq = 0; /* and migrate_seq */
52d60f8e 4474 cap->cap_gen = atomic_read(&cap->session->s_cap_gen);
20cb34ae 4475
a25949b9 4476 /* These are lost when the session goes away */
785892fe
JL		 4477 if (S_ISDIR(inode->i_mode)) {
4478 if (cap->issued & CEPH_CAP_DIR_CREATE) {
4479 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
4480 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
4481 }
a25949b9 4482 cap->issued &= ~CEPH_CAP_ANY_DIR_OPS;
785892fe 4483 }
a25949b9 4484
121f22a1 4485 if (recon_state->msg_version >= 2) {
20cb34ae
SW		 4486 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
4487 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
4488 rec.v2.issued = cpu_to_le32(cap->issued);
4489 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
a33f6432 4490 rec.v2.pathbase = cpu_to_le64(pathbase);
ec1dff25
JL		 4491 rec.v2.flock_len = (__force __le32)
4492 ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
20cb34ae 4493 } else {
c453bdb5
JL		 4494 struct timespec64 ts;
4495
20cb34ae
SW		 4496 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
4497 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
4498 rec.v1.issued = cpu_to_le32(cap->issued);
2d6795fb 4499 rec.v1.size = cpu_to_le64(i_size_read(inode));
c453bdb5
JL		 4500 ts = inode_get_mtime(inode);
4501 ceph_encode_timespec64(&rec.v1.mtime, &ts);
4502 ts = inode_get_atime(inode);
4503 ceph_encode_timespec64(&rec.v1.atime, &ts);
20cb34ae 4504 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
a33f6432 4505 rec.v1.pathbase = cpu_to_le64(pathbase);
20cb34ae 4506 }
3469ed0d
YZ		 4507
4508 if (list_empty(&ci->i_cap_snaps)) {
92776fd2 4509 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3469ed0d
YZ		 4510 } else {
4511 struct ceph_cap_snap *capsnap =
4512 list_first_entry(&ci->i_cap_snaps,
4513 struct ceph_cap_snap, ci_item);
4514 snap_follows = capsnap->follows;
20cb34ae 4515 }
be655596 4516 spin_unlock(&ci->i_ceph_lock);
2f2dc053 4517
121f22a1 4518 if (recon_state->msg_version >= 2) {
40819f6f 4519 int num_fcntl_locks, num_flock_locks;
4deb14a2 4520 struct ceph_filelock *flocks = NULL;
81c5a148 4521 size_t struct_len, total_len = sizeof(u64);
121f22a1 4522 u8 struct_v = 0;
39be95e9
JS		 4523
4524encode_again:
b3f8d68f
YZ		 4525 if (rec.v2.flock_len) {
4526 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
4527 } else {
4528 num_fcntl_locks = 0;
4529 num_flock_locks = 0;
4530 }
4deb14a2 4531 if (num_fcntl_locks + num_flock_locks > 0) {
6da2ec56
KC		 4532 flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
4533 sizeof(struct ceph_filelock),
4534 GFP_NOFS);
4deb14a2
YZ		 4535 if (!flocks) {
4536 err = -ENOMEM;
5ccedf1c 4537 goto out_err;
4deb14a2
YZ		 4538 }
4539 err = ceph_encode_locks_to_buffer(inode, flocks,
4540 num_fcntl_locks,
4541 num_flock_locks);
4542 if (err) {
4543 kfree(flocks);
4544 flocks = NULL;
4545 if (err == -ENOSPC)
4546 goto encode_again;
5ccedf1c 4547 goto out_err;
4deb14a2
YZ		 4548 }
4549 } else {
39be95e9 4550 kfree(flocks);
4deb14a2 4551 flocks = NULL;
39be95e9 4552 }
121f22a1
YZ		 4553
4554 if (recon_state->msg_version >= 3) {
4555 /* version, compat_version and struct_len */
81c5a148 4556 total_len += 2 * sizeof(u8) + sizeof(u32);
3469ed0d 4557 struct_v = 2;
121f22a1 4558 }
39be95e9
JS		 4559 /*
4560 * number of encoded locks is stable, so copy to pagelist
4561 */
121f22a1
YZ		 4562 struct_len = 2 * sizeof(u32) +
4563 (num_fcntl_locks + num_flock_locks) *
4564 sizeof(struct ceph_filelock);
4565 rec.v2.flock_len = cpu_to_le32(struct_len);
4566
a33f6432 4567 struct_len += sizeof(u32) + pathlen + sizeof(rec.v2);
121f22a1 4568
3469ed0d
YZ		 4569 if (struct_v >= 2)
4570 struct_len += sizeof(u64); /* snap_follows */
4571
121f22a1 4572 total_len += struct_len;
81c5a148
YZ		 4573
4574 if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
4575 err = send_reconnect_partial(recon_state);
4576 if (err)
4577 goto out_freeflocks;
4578 pagelist = recon_state->pagelist;
5ccedf1c 4579 }
121f22a1 4580
81c5a148
YZ		 4581 err = ceph_pagelist_reserve(pagelist, total_len);
4582 if (err)
4583 goto out_freeflocks;
4584
4585 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
5ccedf1c
YZ		 4586 if (recon_state->msg_version >= 3) {
4587 ceph_pagelist_encode_8(pagelist, struct_v);
4588 ceph_pagelist_encode_8(pagelist, 1);
4589 ceph_pagelist_encode_32(pagelist, struct_len);
121f22a1 4590 }
a33f6432 4591 ceph_pagelist_encode_string(pagelist, path, pathlen);
5ccedf1c
YZ		 4592 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
4593 ceph_locks_to_pagelist(flocks, pagelist,
4594 num_fcntl_locks, num_flock_locks);
4595 if (struct_v >= 2)
4596 ceph_pagelist_encode_64(pagelist, snap_follows);
81c5a148 4597out_freeflocks:
39be95e9 4598 kfree(flocks);
3612abbd 4599 } else {
5ccedf1c 4600 err = ceph_pagelist_reserve(pagelist,
81c5a148
YZ		 4601 sizeof(u64) + sizeof(u32) +
4602 pathlen + sizeof(rec.v1));
a33f6432
YZ		 4603 if (err)
4604 goto out_err;
5ccedf1c 4605
81c5a148 4606 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
5ccedf1c
YZ		 4607 ceph_pagelist_encode_string(pagelist, path, pathlen);
4608 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
40819f6f 4609 }
44c99757 4610
5ccedf1c 4611out_err:
a33f6432
YZ		 4612 ceph_mdsc_free_path(path, pathlen);
4613 if (!err)
81c5a148
YZ		 4614 recon_state->nr_caps++;
4615 return err;
4616}
4617
4618static int encode_snap_realms(struct ceph_mds_client *mdsc,
4619 struct ceph_reconnect_state *recon_state)
4620{
4621 struct rb_node *p;
4622 struct ceph_pagelist *pagelist = recon_state->pagelist;
38d46409 4623 struct ceph_client *cl = mdsc->fsc->client;
81c5a148
YZ		 4624 int err = 0;
4625
4626 if (recon_state->msg_version >= 4) {
4627 err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
4628 if (err < 0)
4629 goto fail;
4630 }
4631
4632 /*
4633 * snaprealms. we provide mds with the ino, seq (version), and
4634 * parent for all of our realms. If the mds has any newer info,
4635 * it will tell us.
4636 */
4637 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
4638 struct ceph_snap_realm *realm =
4639 rb_entry(p, struct ceph_snap_realm, node);
4640 struct ceph_mds_snaprealm_reconnect sr_rec;
4641
4642 if (recon_state->msg_version >= 4) {
4643 size_t need = sizeof(u8) * 2 + sizeof(u32) +
4644 sizeof(sr_rec);
4645
4646 if (pagelist->length + need > RECONNECT_MAX_SIZE) {
4647 err = send_reconnect_partial(recon_state);
4648 if (err)
4649 goto fail;
4650 pagelist = recon_state->pagelist;
4651 }
4652
4653 err = ceph_pagelist_reserve(pagelist, need);
4654 if (err)
4655 goto fail;
4656
4657 ceph_pagelist_encode_8(pagelist, 1);
4658 ceph_pagelist_encode_8(pagelist, 1);
4659 ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
4660 }
4661
38d46409
XL		 4662 doutc(cl, " adding snap realm %llx seq %lld parent %llx\n",
4663 realm->ino, realm->seq, realm->parent_ino);
81c5a148
YZ		 4664 sr_rec.ino = cpu_to_le64(realm->ino);
4665 sr_rec.seq = cpu_to_le64(realm->seq);
4666 sr_rec.parent = cpu_to_le64(realm->parent_ino);
4667
4668 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
4669 if (err)
4670 goto fail;
4671
4672 recon_state->nr_realms++;
4673 }
4674fail:
93cea5be 4675 return err;
2f2dc053
SW		 4676}
4677
4678
4679/*
4680 * If an MDS fails and recovers, clients need to reconnect in order to
4681 * reestablish shared state. This includes all caps issued through
4682 * this session _and_ the snap_realm hierarchy. Because it's not
4683 * clear which snap realms the mds cares about, we send everything we
4684 * know about.. that ensures we'll then get any new info the
4685 * recovering MDS might have.
4686 *
4687 * This is a relatively heavyweight operation, but it's rare.
2f2dc053 4688 */
34b6c855
SW		 4689static void send_mds_reconnect(struct ceph_mds_client *mdsc,
4690 struct ceph_mds_session *session)
2f2dc053 4691{
38d46409 4692 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053 4693 struct ceph_msg *reply;
34b6c855 4694 int mds = session->s_mds;
9abf82b8 4695 int err = -ENOMEM;
81c5a148
YZ		 4696 struct ceph_reconnect_state recon_state = {
4697 .session = session,
4698 };
c8a96a31 4699 LIST_HEAD(dispose);
2f2dc053 4700
38d46409 4701 pr_info_client(cl, "mds%d reconnect start\n", mds);
2f2dc053 4702
81c5a148
YZ		 4703 recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
4704 if (!recon_state.pagelist)
93cea5be 4705 goto fail_nopagelist;
93cea5be 4706
0d9c1ab3 4707 reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
a79832f2 4708 if (!reply)
93cea5be 4709 goto fail_nomsg;
93cea5be 4710
d4846487
JL		 4711 xa_destroy(&session->s_delegated_inos);
4712
34b6c855
SW		 4713 mutex_lock(&session->s_mutex);
4714 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
4715 session->s_seq = 0;
2f2dc053 4716
38d46409
XL		 4717 doutc(cl, "session %p state %s\n", session,
4718 ceph_session_state_name(session->s_state));
2f2dc053 4719
52d60f8e 4720 atomic_inc(&session->s_cap_gen);
99a9c273
YZ		 4721
4722 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 4723 /* don't know if session is readonly */
4724 session->s_readonly = 0;
99a9c273
YZ		 4725 /*
4726 * notify __ceph_remove_cap() that we are composing cap reconnect.
4727 * If a cap get released before being added to the cap reconnect,
4728 * __ceph_remove_cap() should skip queuing cap release.
4729 */
4730 session->s_cap_reconnect = 1;
e01a5946 4731 /* drop old cap expires; we're about to reestablish that state */
c8a96a31
JL		 4732 detach_cap_releases(session, &dispose);
4733 spin_unlock(&session->s_cap_lock);
4734 dispose_cap_releases(mdsc, &dispose);
e01a5946 4735
5d23371f 4736 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 4737 if (mdsc->fsc->sb->s_root)
4738 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 4739
4740 ceph_con_close(&session->s_con);
4741 ceph_con_open(&session->s_con,
4742 CEPH_ENTITY_TYPE_MDS, mds,
4743 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
4744
4745 /* replay unsafe requests */
4746 replay_unsafe_requests(mdsc, session);
4747
81c5a148
YZ		 4748 ceph_early_kick_flushing_caps(mdsc, session);
4749
5d23371f
YZ		 4750 down_read(&mdsc->snap_rwsem);
4751
81c5a148
YZ		 4752 /* placeholder for nr_caps */
4753 err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
93cea5be
SW		 4754 if (err)
4755 goto fail;
20cb34ae 4756
81c5a148 4757 if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
121f22a1 4758 recon_state.msg_version = 3;
81c5a148
YZ		 4759 recon_state.allow_multi = true;
4760 } else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
4761 recon_state.msg_version = 3;
4762 } else {
23c625ce 4763 recon_state.msg_version = 2;
81c5a148
YZ		 4764 }
4765 /* trsaverse this session's caps */
a25949b9 4766 err = ceph_iterate_session_caps(session, reconnect_caps_cb, &recon_state);
2f2dc053 4767
99a9c273
YZ		 4768 spin_lock(&session->s_cap_lock);
4769 session->s_cap_reconnect = 0;
4770 spin_unlock(&session->s_cap_lock);
4771
81c5a148
YZ		 4772 if (err < 0)
4773 goto fail;
2f2dc053 4774
81c5a148
YZ		 4775 /* check if all realms can be encoded into current message */
4776 if (mdsc->num_snap_realms) {
4777 size_t total_len =
4778 recon_state.pagelist->length +
4779 mdsc->num_snap_realms *
4780 sizeof(struct ceph_mds_snaprealm_reconnect);
4781 if (recon_state.msg_version >= 4) {
4782 /* number of realms */
4783 total_len += sizeof(u32);
4784 /* version, compat_version and struct_len */
4785 total_len += mdsc->num_snap_realms *
4786 (2 * sizeof(u8) + sizeof(u32));
4787 }
4788 if (total_len > RECONNECT_MAX_SIZE) {
4789 if (!recon_state.allow_multi) {
4790 err = -ENOSPC;
4791 goto fail;
4792 }
4793 if (recon_state.nr_caps) {
4794 err = send_reconnect_partial(&recon_state);
4795 if (err)
4796 goto fail;
4797 }
4798 recon_state.msg_version = 5;
4799 }
2f2dc053 4800 }
2f2dc053 4801
81c5a148
YZ		 4802 err = encode_snap_realms(mdsc, &recon_state);
4803 if (err < 0)
4804 goto fail;
4805
4806 if (recon_state.msg_version >= 5) {
4807 err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
4808 if (err < 0)
4809 goto fail;
4810 }
44c99757 4811
81c5a148
YZ		 4812 if (recon_state.nr_caps || recon_state.nr_realms) {
4813 struct page *page =
4814 list_first_entry(&recon_state.pagelist->head,
4815 struct page, lru);
44c99757 4816 __le32 *addr = kmap_atomic(page);
81c5a148
YZ		 4817 if (recon_state.nr_caps) {
4818 WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
4819 *addr = cpu_to_le32(recon_state.nr_caps);
4820 } else if (recon_state.msg_version >= 4) {
4821 *(addr + 1) = cpu_to_le32(recon_state.nr_realms);
4822 }
44c99757 4823 kunmap_atomic(addr);
ebf18f47 4824 }
44c99757 4825
81c5a148
YZ		 4826 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
4827 if (recon_state.msg_version >= 4)
4828 reply->hdr.compat_version = cpu_to_le16(4);
e548e9b9 4829
81c5a148
YZ		 4830 reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
4831 ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
e548e9b9 4832
2f2dc053
SW		 4833 ceph_con_send(&session->s_con, reply);
4834
9abf82b8
SW		 4835 mutex_unlock(&session->s_mutex);
4836
4837 mutex_lock(&mdsc->mutex);
4838 __wake_requests(mdsc, &session->s_waiting);
4839 mutex_unlock(&mdsc->mutex);
4840
2f2dc053 4841 up_read(&mdsc->snap_rwsem);
81c5a148 4842 ceph_pagelist_release(recon_state.pagelist);
2f2dc053
SW		 4843 return;
4844
93cea5be 4845fail:
2f2dc053 4846 ceph_msg_put(reply);
9abf82b8
SW		 4847 up_read(&mdsc->snap_rwsem);
4848 mutex_unlock(&session->s_mutex);
93cea5be 4849fail_nomsg:
81c5a148 4850 ceph_pagelist_release(recon_state.pagelist);
93cea5be 4851fail_nopagelist:
38d46409
XL		 4852 pr_err_client(cl, "error %d preparing reconnect for mds%d\n",
4853 err, mds);
9abf82b8 4854 return;
2f2dc053
SW		 4855}
4856
4857
4858/*
4859 * compare old and new mdsmaps, kicking requests
4860 * and closing out old connections as necessary
4861 *
4862 * called under mdsc->mutex.
4863 */
4864static void check_new_map(struct ceph_mds_client *mdsc,
4865 struct ceph_mdsmap *newmap,
4866 struct ceph_mdsmap *oldmap)
4867{
d517b398 4868 int i, j, err;
2f2dc053
SW		 4869 int oldstate, newstate;
4870 struct ceph_mds_session *s;
d517b398 4871 unsigned long targets[DIV_ROUND_UP(CEPH_MAX_MDS, sizeof(unsigned long))] = {0};
38d46409 4872 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053 4873
38d46409 4874 doutc(cl, "new %u old %u\n", newmap->m_epoch, oldmap->m_epoch);
2f2dc053 4875
d517b398
XL		 4876 if (newmap->m_info) {
4877 for (i = 0; i < newmap->possible_max_rank; i++) {
4878 for (j = 0; j < newmap->m_info[i].num_export_targets; j++)
4879 set_bit(newmap->m_info[i].export_targets[j], targets);
4880 }
4881 }
4882
b38c9eb4 4883 for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
d37b1d99 4884 if (!mdsc->sessions[i])
2f2dc053
SW		 4885 continue;
4886 s = mdsc->sessions[i];
4887 oldstate = ceph_mdsmap_get_state(oldmap, i);
4888 newstate = ceph_mdsmap_get_state(newmap, i);
4889
38d46409
XL		 4890 doutc(cl, "mds%d state %s%s -> %s%s (session %s)\n",
4891 i, ceph_mds_state_name(oldstate),
4892 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
4893 ceph_mds_state_name(newstate),
4894 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
4895 ceph_session_state_name(s->s_state));
2f2dc053 4896
b38c9eb4 4897 if (i >= newmap->possible_max_rank) {
6f0f597b 4898 /* force close session for stopped mds */
5b3248c6 4899 ceph_get_mds_session(s);
6f0f597b
YZ		 4900 __unregister_session(mdsc, s);
4901 __wake_requests(mdsc, &s->s_waiting);
4902 mutex_unlock(&mdsc->mutex);
2827528d 4903
6f0f597b
YZ		 4904 mutex_lock(&s->s_mutex);
4905 cleanup_session_requests(mdsc, s);
4906 remove_session_caps(s);
4907 mutex_unlock(&s->s_mutex);
2827528d 4908
6f0f597b 4909 ceph_put_mds_session(s);
2827528d 4910
6f0f597b
YZ		 4911 mutex_lock(&mdsc->mutex);
4912 kick_requests(mdsc, i);
4913 continue;
4914 }
4915
4916 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
4917 ceph_mdsmap_get_addr(newmap, i),
4918 sizeof(struct ceph_entity_addr))) {
4919 /* just close it */
4920 mutex_unlock(&mdsc->mutex);
4921 mutex_lock(&s->s_mutex);
4922 mutex_lock(&mdsc->mutex);
4923 ceph_con_close(&s->s_con);
4924 mutex_unlock(&s->s_mutex);
4925 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2f2dc053
SW		 4926 } else if (oldstate == newstate) {
4927 continue; /* nothing new with this mds */
4928 }
4929
4930 /*
4931 * send reconnect?
4932 */
4933 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 4934 newstate >= CEPH_MDS_STATE_RECONNECT) {
4935 mutex_unlock(&mdsc->mutex);
d517b398 4936 clear_bit(i, targets);
34b6c855
SW		 4937 send_mds_reconnect(mdsc, s);
4938 mutex_lock(&mdsc->mutex);
4939 }
2f2dc053
SW		 4940
4941 /*
29790f26 4942 * kick request on any mds that has gone active.
2f2dc053
SW		 4943 */
4944 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
4945 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 4946 if (oldstate != CEPH_MDS_STATE_CREATING &&
4947 oldstate != CEPH_MDS_STATE_STARTING)
38d46409
XL		 4948 pr_info_client(cl, "mds%d recovery completed\n",
4949 s->s_mds);
29790f26 4950 kick_requests(mdsc, i);
ea8412b2 4951 mutex_unlock(&mdsc->mutex);
829ad4db 4952 mutex_lock(&s->s_mutex);
ea8412b2 4953 mutex_lock(&mdsc->mutex);
2f2dc053 4954 ceph_kick_flushing_caps(mdsc, s);
829ad4db 4955 mutex_unlock(&s->s_mutex);
d2f8bb27 4956 wake_up_session_caps(s, RECONNECT);
2f2dc053
SW		 4957 }
4958 }
cb170a22 4959
d517b398
XL		 4960 /*
4961 * Only open and reconnect sessions that don't exist yet.
4962 */
4963 for (i = 0; i < newmap->possible_max_rank; i++) {
4964 /*
4965 * In case the import MDS is crashed just after
4966 * the EImportStart journal is flushed, so when
4967 * a standby MDS takes over it and is replaying
4968 * the EImportStart journal the new MDS daemon
4969 * will wait the client to reconnect it, but the
4970 * client may never register/open the session yet.
4971 *
4972 * Will try to reconnect that MDS daemon if the
4973 * rank number is in the export targets array and
4974 * is the up:reconnect state.
4975 */
4976 newstate = ceph_mdsmap_get_state(newmap, i);
4977 if (!test_bit(i, targets) || newstate != CEPH_MDS_STATE_RECONNECT)
4978 continue;
4979
4980 /*
4981 * The session maybe registered and opened by some
4982 * requests which were choosing random MDSes during
4983 * the mdsc->mutex's unlock/lock gap below in rare
4984 * case. But the related MDS daemon will just queue
4985 * that requests and be still waiting for the client's
4986 * reconnection request in up:reconnect state.
4987 */
4988 s = __ceph_lookup_mds_session(mdsc, i);
4989 if (likely(!s)) {
4990 s = __open_export_target_session(mdsc, i);
4991 if (IS_ERR(s)) {
4992 err = PTR_ERR(s);
38d46409
XL		 4993 pr_err_client(cl,
4994 "failed to open export target session, err %d\n",
4995 err);
d517b398
XL		 4996 continue;
4997 }
4998 }
38d46409 4999 doutc(cl, "send reconnect to export target mds.%d\n", i);
d517b398
XL		 5000 mutex_unlock(&mdsc->mutex);
5001 send_mds_reconnect(mdsc, s);
5002 ceph_put_mds_session(s);
5003 mutex_lock(&mdsc->mutex);
5004 }
5005
b38c9eb4 5006 for (i = 0; i < newmap->possible_max_rank && i < mdsc->max_sessions; i++) {
cb170a22
SW		 5007 s = mdsc->sessions[i];
5008 if (!s)
5009 continue;
5010 if (!ceph_mdsmap_is_laggy(newmap, i))
5011 continue;
5012 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
5013 s->s_state == CEPH_MDS_SESSION_HUNG ||
5014 s->s_state == CEPH_MDS_SESSION_CLOSING) {
38d46409 5015 doutc(cl, " connecting to export targets of laggy mds%d\n", i);
cb170a22
SW		 5016 __open_export_target_sessions(mdsc, s);
5017 }
5018 }
2f2dc053
SW		 5019}
5020
5021
5022
5023/*
5024 * leases
5025 */
5026
5027/*
5028 * caller must hold session s_mutex, dentry->d_lock
5029 */
5030void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
5031{
5032 struct ceph_dentry_info *di = ceph_dentry(dentry);
5033
5034 ceph_put_mds_session(di->lease_session);
5035 di->lease_session = NULL;
5036}
5037
2600d2dd
SW		 5038static void handle_lease(struct ceph_mds_client *mdsc,
5039 struct ceph_mds_session *session,
5040 struct ceph_msg *msg)
2f2dc053 5041{
38d46409 5042 struct ceph_client *cl = mdsc->fsc->client;
3d14c5d2 5043 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 5044 struct inode *inode;
2f2dc053
SW		 5045 struct dentry *parent, *dentry;
5046 struct ceph_dentry_info *di;
2600d2dd 5047 int mds = session->s_mds;
2f2dc053 5048 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 5049 u32 seq;
2f2dc053 5050 struct ceph_vino vino;
2f2dc053
SW		 5051 struct qstr dname;
5052 int release = 0;
5053
38d46409 5054 doutc(cl, "from mds%d\n", mds);
2f2dc053 5055
e3dfcab2
XL		 5056 if (!ceph_inc_mds_stopping_blocker(mdsc, session))
5057 return;
5058
2f2dc053
SW		 5059 /* decode */
5060 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
5061 goto bad;
5062 vino.ino = le64_to_cpu(h->ino);
5063 vino.snap = CEPH_NOSNAP;
1e5ea23d 5064 seq = le32_to_cpu(h->seq);
0fcf6c02
YZ		 5065 dname.len = get_unaligned_le32(h + 1);
5066 if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
2f2dc053 5067 goto bad;
0fcf6c02 5068 dname.name = (void *)(h + 1) + sizeof(u32);
2f2dc053 5069
2f2dc053
SW		 5070 /* lookup inode */
5071 inode = ceph_find_inode(sb, vino);
38d46409
XL		 5072 doutc(cl, "%s, ino %llx %p %.*s\n", ceph_lease_op_name(h->action),
5073 vino.ino, inode, dname.len, dname.name);
6cd3bcad
YZ		 5074
5075 mutex_lock(&session->s_mutex);
d37b1d99 5076 if (!inode) {
38d46409 5077 doutc(cl, "no inode %llx\n", vino.ino);
2f2dc053
SW		 5078 goto release;
5079 }
2f2dc053
SW		 5080
5081 /* dentry */
5082 parent = d_find_alias(inode);
5083 if (!parent) {
38d46409 5084 doutc(cl, "no parent dentry on inode %p\n", inode);
2f2dc053
SW		 5085 WARN_ON(1);
5086 goto release; /* hrm... */
5087 }
8387ff25 5088 dname.hash = full_name_hash(parent, dname.name, dname.len);
2f2dc053
SW		 5089 dentry = d_lookup(parent, &dname);
5090 dput(parent);
5091 if (!dentry)
5092 goto release;
5093
5094 spin_lock(&dentry->d_lock);
5095 di = ceph_dentry(dentry);
5096 switch (h->action) {
5097 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 5098 if (di->lease_session == session) {
1e5ea23d
SW		 5099 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
5100 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 5101 __ceph_mdsc_drop_dentry_lease(dentry);
5102 }
5103 release = 1;
5104 break;
5105
5106 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 5107 if (di->lease_session == session &&
52d60f8e 5108 di->lease_gen == atomic_read(&session->s_cap_gen) &&
2f2dc053
SW		 5109 di->lease_renew_from &&
5110 di->lease_renew_after == 0) {
5111 unsigned long duration =
3563dbdd 5112 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 5113
1e5ea23d 5114 di->lease_seq = seq;
9b16f03c 5115 di->time = di->lease_renew_from + duration;
2f2dc053
SW		 5116 di->lease_renew_after = di->lease_renew_from +
5117 (duration >> 1);
5118 di->lease_renew_from = 0;
5119 }
5120 break;
5121 }
5122 spin_unlock(&dentry->d_lock);
5123 dput(dentry);
5124
5125 if (!release)
5126 goto out;
5127
5128release:
5129 /* let's just reuse the same message */
5130 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
5131 ceph_msg_get(msg);
5132 ceph_con_send(&session->s_con, msg);
5133
5134out:
2f2dc053 5135 mutex_unlock(&session->s_mutex);
23c2c76e 5136 iput(inode);
e3dfcab2
XL		 5137
5138 ceph_dec_mds_stopping_blocker(mdsc);
2f2dc053
SW		 5139 return;
5140
5141bad:
e3dfcab2
XL		 5142 ceph_dec_mds_stopping_blocker(mdsc);
5143
38d46409 5144 pr_err_client(cl, "corrupt lease message\n");
9ec7cab1 5145 ceph_msg_dump(msg);
2f2dc053
SW		 5146}
5147
5148void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
2f2dc053
SW		 5149 struct dentry *dentry, char action,
5150 u32 seq)
5151{
38d46409 5152 struct ceph_client *cl = session->s_mdsc->fsc->client;
2f2dc053
SW		 5153 struct ceph_msg *msg;
5154 struct ceph_mds_lease *lease;
8f2a98ef
YZ		 5155 struct inode *dir;
5156 int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
2f2dc053 5157
38d46409
XL		 5158 doutc(cl, "identry %p %s to mds%d\n", dentry, ceph_lease_op_name(action),
5159 session->s_mds);
2f2dc053 5160
b61c2763 5161 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 5162 if (!msg)
2f2dc053
SW		 5163 return;
5164 lease = msg->front.iov_base;
5165 lease->action = action;
2f2dc053 5166 lease->seq = cpu_to_le32(seq);
2f2dc053 5167
8f2a98ef
YZ		 5168 spin_lock(&dentry->d_lock);
5169 dir = d_inode(dentry->d_parent);
5170 lease->ino = cpu_to_le64(ceph_ino(dir));
5171 lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
5172
5173 put_unaligned_le32(dentry->d_name.len, lease + 1);
5174 memcpy((void *)(lease + 1) + 4,
5175 dentry->d_name.name, dentry->d_name.len);
5176 spin_unlock(&dentry->d_lock);
2f2dc053
SW		 5177
5178 ceph_con_send(&session->s_con, msg);
5179}
5180
2f2dc053 5181/*
59b312f3 5182 * lock unlock the session, to wait ongoing session activities
2f2dc053 5183 */
59b312f3 5184static void lock_unlock_session(struct ceph_mds_session *s)
2f2dc053 5185{
59b312f3
XL		 5186 mutex_lock(&s->s_mutex);
5187 mutex_unlock(&s->s_mutex);
2f2dc053
SW		 5188}
5189
131d7eb4
YZ		 5190static void maybe_recover_session(struct ceph_mds_client *mdsc)
5191{
38d46409 5192 struct ceph_client *cl = mdsc->fsc->client;
131d7eb4
YZ		 5193 struct ceph_fs_client *fsc = mdsc->fsc;
5194
5195 if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
5196 return;
2f2dc053 5197
131d7eb4
YZ		 5198 if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
5199 return;
5200
0b98acd6 5201 if (!READ_ONCE(fsc->blocklisted))
131d7eb4
YZ		 5202 return;
5203
38d46409 5204 pr_info_client(cl, "auto reconnect after blocklisted\n");
131d7eb4
YZ		 5205 ceph_force_reconnect(fsc->sb);
5206}
2f2dc053 5207
3e699bd8
XL		 5208bool check_session_state(struct ceph_mds_session *s)
5209{
38d46409
XL		 5210 struct ceph_client *cl = s->s_mdsc->fsc->client;
5211
62575e27
JL		 5212 switch (s->s_state) {
5213 case CEPH_MDS_SESSION_OPEN:
5214 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3e699bd8 5215 s->s_state = CEPH_MDS_SESSION_HUNG;
38d46409 5216 pr_info_client(cl, "mds%d hung\n", s->s_mds);
3e699bd8 5217 }
62575e27
JL		 5218 break;
5219 case CEPH_MDS_SESSION_CLOSING:
62575e27
JL		 5220 case CEPH_MDS_SESSION_NEW:
5221 case CEPH_MDS_SESSION_RESTARTING:
5222 case CEPH_MDS_SESSION_CLOSED:
5223 case CEPH_MDS_SESSION_REJECTED:
3e699bd8 5224 return false;
62575e27 5225 }
3e699bd8
XL		 5226
5227 return true;
5228}
5229
62575e27
JL		 5230/*
5231 * If the sequence is incremented while we're waiting on a REQUEST_CLOSE reply,
5232 * then we need to retransmit that request.
5233 */
5234void inc_session_sequence(struct ceph_mds_session *s)
5235{
38d46409
XL		 5236 struct ceph_client *cl = s->s_mdsc->fsc->client;
5237
62575e27
JL		 5238 lockdep_assert_held(&s->s_mutex);
5239
5240 s->s_seq++;
5241
5242 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
5243 int ret;
5244
38d46409 5245 doutc(cl, "resending session close request for mds%d\n", s->s_mds);
62575e27
JL		 5246 ret = request_close_session(s);
5247 if (ret < 0)
38d46409
XL		 5248 pr_err_client(cl, "unable to close session to mds%d: %d\n",
5249 s->s_mds, ret);
62575e27
JL		 5250 }
5251}
5252
2f2dc053 5253/*
bf2ba432
LH		 5254 * delayed work -- periodically trim expired leases, renew caps with mds. If
5255 * the @delay parameter is set to 0 or if it's more than 5 secs, the default
5256 * workqueue delay value of 5 secs will be used.
2f2dc053 5257 */
bf2ba432 5258static void schedule_delayed(struct ceph_mds_client *mdsc, unsigned long delay)
2f2dc053 5259{
bf2ba432
LH		 5260 unsigned long max_delay = HZ * 5;
5261
5262 /* 5 secs default delay */
5263 if (!delay || (delay > max_delay))
5264 delay = max_delay;
5265 schedule_delayed_work(&mdsc->delayed_work,
5266 round_jiffies_relative(delay));
2f2dc053
SW		 5267}
5268
5269static void delayed_work(struct work_struct *work)
5270{
2f2dc053
SW		 5271 struct ceph_mds_client *mdsc =
5272 container_of(work, struct ceph_mds_client, delayed_work.work);
bf2ba432 5273 unsigned long delay;
2f2dc053
SW		 5274 int renew_interval;
5275 int renew_caps;
bf2ba432 5276 int i;
2f2dc053 5277
38d46409 5278 doutc(mdsc->fsc->client, "mdsc delayed_work\n");
75c9627e 5279
e7e607bd 5280 if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHED)
fa996773
XL		 5281 return;
5282
2f2dc053
SW		 5283 mutex_lock(&mdsc->mutex);
5284 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
5285 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
5286 mdsc->last_renew_caps);
5287 if (renew_caps)
5288 mdsc->last_renew_caps = jiffies;
5289
5290 for (i = 0; i < mdsc->max_sessions; i++) {
5291 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
d37b1d99 5292 if (!s)
2f2dc053 5293 continue;
3e699bd8
XL		 5294
5295 if (!check_session_state(s)) {
2f2dc053
SW		 5296 ceph_put_mds_session(s);
5297 continue;
5298 }
5299 mutex_unlock(&mdsc->mutex);
5300
5301 mutex_lock(&s->s_mutex);
5302 if (renew_caps)
5303 send_renew_caps(mdsc, s);
5304 else
5305 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 5306 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
5307 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 5308 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 5309 mutex_unlock(&s->s_mutex);
5310 ceph_put_mds_session(s);
5311
5312 mutex_lock(&mdsc->mutex);
5313 }
5314 mutex_unlock(&mdsc->mutex);
5315
bf2ba432 5316 delay = ceph_check_delayed_caps(mdsc);
37c4efc1
YZ		 5317
5318 ceph_queue_cap_reclaim_work(mdsc);
5319
5320 ceph_trim_snapid_map(mdsc);
5321
131d7eb4
YZ		 5322 maybe_recover_session(mdsc);
5323
bf2ba432 5324 schedule_delayed(mdsc, delay);
2f2dc053
SW		 5325}
5326
3d14c5d2 5327int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 5328
2f2dc053 5329{
3d14c5d2 5330 struct ceph_mds_client *mdsc;
f9009efa 5331 int err;
3d14c5d2
YS		 5332
5333 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
5334 if (!mdsc)
5335 return -ENOMEM;
5336 mdsc->fsc = fsc;
2f2dc053
SW		 5337 mutex_init(&mdsc->mutex);
5338 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
d37b1d99 5339 if (!mdsc->mdsmap) {
f9009efa
XL		 5340 err = -ENOMEM;
5341 goto err_mdsc;
fb3101b6 5342 }
2d06eeb8 5343
2f2dc053 5344 init_completion(&mdsc->safe_umount_waiters);
e3dfcab2
XL		 5345 spin_lock_init(&mdsc->stopping_lock);
5346 atomic_set(&mdsc->stopping_blockers, 0);
5347 init_completion(&mdsc->stopping_waiter);
f3c60c59 5348 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053 5349 INIT_LIST_HEAD(&mdsc->waiting_for_map);
0c44a8e0
LH		 5350 mdsc->quotarealms_inodes = RB_ROOT;
5351 mutex_init(&mdsc->quotarealms_inodes_mutex);
2f2dc053 5352 init_rwsem(&mdsc->snap_rwsem);
a105f00c 5353 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 5354 INIT_LIST_HEAD(&mdsc->snap_empty);
5355 spin_lock_init(&mdsc->snap_empty_lock);
44ca18f2 5356 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 5357 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
5358 mdsc->last_renew_caps = jiffies;
5359 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3a3430af 5360 INIT_LIST_HEAD(&mdsc->cap_wait_list);
2f2dc053
SW		 5361 spin_lock_init(&mdsc->cap_delay_lock);
5362 INIT_LIST_HEAD(&mdsc->snap_flush_list);
5363 spin_lock_init(&mdsc->snap_flush_lock);
553adfd9 5364 mdsc->last_cap_flush_tid = 1;
e4500b5e 5365 INIT_LIST_HEAD(&mdsc->cap_flush_list);
db354052 5366 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 5367 spin_lock_init(&mdsc->cap_dirty_lock);
5368 init_waitqueue_head(&mdsc->cap_flushing_wq);
37c4efc1 5369 INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
f9009efa
XL		 5370 err = ceph_metric_init(&mdsc->metric);
5371 if (err)
5372 goto err_mdsmap;
37c4efc1
YZ		 5373
5374 spin_lock_init(&mdsc->dentry_list_lock);
5375 INIT_LIST_HEAD(&mdsc->dentry_leases);
5376 INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
2d06eeb8 5377
37151668 5378 ceph_caps_init(mdsc);
fe33032d 5379 ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
37151668 5380
75c9627e
YZ		 5381 spin_lock_init(&mdsc->snapid_map_lock);
5382 mdsc->snapid_map_tree = RB_ROOT;
5383 INIT_LIST_HEAD(&mdsc->snapid_map_lru);
5384
10183a69
YZ		 5385 init_rwsem(&mdsc->pool_perm_rwsem);
5386 mdsc->pool_perm_tree = RB_ROOT;
5387
dfeb84d4
YZ		 5388 strscpy(mdsc->nodename, utsname()->nodename,
5389 sizeof(mdsc->nodename));
a7caa88f
XL		 5390
5391 fsc->mdsc = mdsc;
5f44f142 5392 return 0;
f9009efa
XL		 5393
5394err_mdsmap:
5395 kfree(mdsc->mdsmap);
5396err_mdsc:
5397 kfree(mdsc);
5398 return err;
2f2dc053
SW		 5399}
5400
5401/*
5402 * Wait for safe replies on open mds requests. If we time out, drop
5403 * all requests from the tree to avoid dangling dentry refs.
5404 */
5405static void wait_requests(struct ceph_mds_client *mdsc)
5406{
38d46409 5407 struct ceph_client *cl = mdsc->fsc->client;
a319bf56 5408 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 5409 struct ceph_mds_request *req;
2f2dc053
SW		 5410
5411 mutex_lock(&mdsc->mutex);
44ca18f2 5412 if (__get_oldest_req(mdsc)) {
2f2dc053 5413 mutex_unlock(&mdsc->mutex);
44ca18f2 5414
38d46409 5415 doutc(cl, "waiting for requests\n");
2f2dc053 5416 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 5417 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 5418
5419 /* tear down remaining requests */
44ca18f2
SW		 5420 mutex_lock(&mdsc->mutex);
5421 while ((req = __get_oldest_req(mdsc))) {
38d46409 5422 doutc(cl, "timed out on tid %llu\n", req->r_tid);
428138c9 5423 list_del_init(&req->r_wait);
44ca18f2 5424 __unregister_request(mdsc, req);
2f2dc053
SW		 5425 }
5426 }
5427 mutex_unlock(&mdsc->mutex);
38d46409 5428 doutc(cl, "done\n");
2f2dc053
SW		 5429}
5430
d095559c
XL		 5431void send_flush_mdlog(struct ceph_mds_session *s)
5432{
38d46409 5433 struct ceph_client *cl = s->s_mdsc->fsc->client;
d095559c
XL		 5434 struct ceph_msg *msg;
5435
5436 /*
5437 * Pre-luminous MDS crashes when it sees an unknown session request
5438 */
5439 if (!CEPH_HAVE_FEATURE(s->s_con.peer_features, SERVER_LUMINOUS))
5440 return;
5441
5442 mutex_lock(&s->s_mutex);
38d46409
XL		 5443 doutc(cl, "request mdlog flush to mds%d (%s)s seq %lld\n",
5444 s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
d095559c
XL		 5445 msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_FLUSH_MDLOG,
5446 s->s_seq);
5447 if (!msg) {
38d46409
XL		 5448 pr_err_client(cl, "failed to request mdlog flush to mds%d (%s) seq %lld\n",
5449 s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
d095559c
XL		 5450 } else {
5451 ceph_con_send(&s->s_con, msg);
5452 }
5453 mutex_unlock(&s->s_mutex);
5454}
5455
2f2dc053
SW		 5456/*
5457 * called before mount is ro, and before dentries are torn down.
5458 * (hmm, does this still race with new lookups?)
5459 */
5460void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
5461{
38d46409 5462 doutc(mdsc->fsc->client, "begin\n");
e7e607bd 5463 mdsc->stopping = CEPH_MDSC_STOPPING_BEGIN;
2f2dc053 5464
d095559c 5465 ceph_mdsc_iterate_sessions(mdsc, send_flush_mdlog, true);
59b312f3 5466 ceph_mdsc_iterate_sessions(mdsc, lock_unlock_session, false);
afcdaea3 5467 ceph_flush_dirty_caps(mdsc);
2f2dc053 5468 wait_requests(mdsc);
17c688c3
SW		 5469
5470 /*
5471 * wait for reply handlers to drop their request refs and
5472 * their inode/dcache refs
5473 */
5474 ceph_msgr_flush();
0c44a8e0
LH		 5475
5476 ceph_cleanup_quotarealms_inodes(mdsc);
38d46409 5477 doutc(mdsc->fsc->client, "done\n");
2f2dc053
SW		 5478}
5479
5480/*
1b2ba3c5 5481 * flush the mdlog and wait for all write mds requests to flush.
2f2dc053 5482 */
1b2ba3c5
XL		 5483static void flush_mdlog_and_wait_mdsc_unsafe_requests(struct ceph_mds_client *mdsc,
5484 u64 want_tid)
2f2dc053 5485{
38d46409 5486 struct ceph_client *cl = mdsc->fsc->client;
80fc7314 5487 struct ceph_mds_request *req = NULL, *nextreq;
1b2ba3c5 5488 struct ceph_mds_session *last_session = NULL;
44ca18f2 5489 struct rb_node *n;
2f2dc053
SW		 5490
5491 mutex_lock(&mdsc->mutex);
38d46409 5492 doutc(cl, "want %lld\n", want_tid);
80fc7314 5493restart:
44ca18f2
SW		 5494 req = __get_oldest_req(mdsc);
5495 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 5496 /* find next request */
5497 n = rb_next(&req->r_node);
5498 if (n)
5499 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
5500 else
5501 nextreq = NULL;
e8a7b8b1
YZ		 5502 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
5503 (req->r_op & CEPH_MDS_OP_WRITE)) {
1b2ba3c5
XL		 5504 struct ceph_mds_session *s = req->r_session;
5505
5506 if (!s) {
5507 req = nextreq;
5508 continue;
5509 }
5510
44ca18f2
SW		 5511 /* write op */
5512 ceph_mdsc_get_request(req);
80fc7314
SW		 5513 if (nextreq)
5514 ceph_mdsc_get_request(nextreq);
1b2ba3c5 5515 s = ceph_get_mds_session(s);
44ca18f2 5516 mutex_unlock(&mdsc->mutex);
1b2ba3c5
XL		 5517
5518 /* send flush mdlog request to MDS */
5519 if (last_session != s) {
5520 send_flush_mdlog(s);
5521 ceph_put_mds_session(last_session);
5522 last_session = s;
5523 } else {
5524 ceph_put_mds_session(s);
5525 }
38d46409
XL		 5526 doutc(cl, "wait on %llu (want %llu)\n",
5527 req->r_tid, want_tid);
44ca18f2 5528 wait_for_completion(&req->r_safe_completion);
1b2ba3c5 5529
44ca18f2 5530 mutex_lock(&mdsc->mutex);
44ca18f2 5531 ceph_mdsc_put_request(req);
80fc7314
SW		 5532 if (!nextreq)
5533 break; /* next dne before, so we're done! */
5534 if (RB_EMPTY_NODE(&nextreq->r_node)) {
5535 /* next request was removed from tree */
5536 ceph_mdsc_put_request(nextreq);
5537 goto restart;
5538 }
5539 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 5540 }
80fc7314 5541 req = nextreq;
2f2dc053
SW		 5542 }
5543 mutex_unlock(&mdsc->mutex);
1b2ba3c5 5544 ceph_put_mds_session(last_session);
38d46409 5545 doutc(cl, "done\n");
2f2dc053
SW		 5546}
5547
5548void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
5549{
38d46409 5550 struct ceph_client *cl = mdsc->fsc->client;
0e294387 5551 u64 want_tid, want_flush;
2f2dc053 5552
50c9132d 5553 if (READ_ONCE(mdsc->fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 5554 return;
5555
38d46409 5556 doutc(cl, "sync\n");
2f2dc053
SW		 5557 mutex_lock(&mdsc->mutex);
5558 want_tid = mdsc->last_tid;
2f2dc053 5559 mutex_unlock(&mdsc->mutex);
2f2dc053 5560
afcdaea3 5561 ceph_flush_dirty_caps(mdsc);
d3383a8e 5562 spin_lock(&mdsc->cap_dirty_lock);
8310b089 5563 want_flush = mdsc->last_cap_flush_tid;
c8799fc4
YZ		 5564 if (!list_empty(&mdsc->cap_flush_list)) {
5565 struct ceph_cap_flush *cf =
5566 list_last_entry(&mdsc->cap_flush_list,
5567 struct ceph_cap_flush, g_list);
5568 cf->wake = true;
5569 }
d3383a8e
YZ		 5570 spin_unlock(&mdsc->cap_dirty_lock);
5571
38d46409 5572 doutc(cl, "sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
2f2dc053 5573
1b2ba3c5 5574 flush_mdlog_and_wait_mdsc_unsafe_requests(mdsc, want_tid);
0e294387 5575 wait_caps_flush(mdsc, want_flush);
2f2dc053
SW		 5576}
5577
f3c60c59
SW		 5578/*
5579 * true if all sessions are closed, or we force unmount
5580 */
fcff415c 5581static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
f3c60c59 5582{
52953d55 5583 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
f3c60c59 5584 return true;
fcff415c 5585 return atomic_read(&mdsc->num_sessions) <= skipped;
f3c60c59 5586}
2f2dc053
SW		 5587
5588/*
a68e564a 5589 * called after sb is ro or when metadata corrupted.
2f2dc053
SW		 5590 */
5591void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
5592{
a319bf56 5593 struct ceph_options *opts = mdsc->fsc->client->options;
38d46409 5594 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 5595 struct ceph_mds_session *session;
5596 int i;
fcff415c 5597 int skipped = 0;
2f2dc053 5598
38d46409 5599 doutc(cl, "begin\n");
2f2dc053 5600
2f2dc053 5601 /* close sessions */
f3c60c59
SW		 5602 mutex_lock(&mdsc->mutex);
5603 for (i = 0; i < mdsc->max_sessions; i++) {
5604 session = __ceph_lookup_mds_session(mdsc, i);
5605 if (!session)
5606 continue;
2f2dc053 5607 mutex_unlock(&mdsc->mutex);
f3c60c59 5608 mutex_lock(&session->s_mutex);
fcff415c
YZ		 5609 if (__close_session(mdsc, session) <= 0)
5610 skipped++;
f3c60c59
SW		 5611 mutex_unlock(&session->s_mutex);
5612 ceph_put_mds_session(session);
2f2dc053
SW		 5613 mutex_lock(&mdsc->mutex);
5614 }
f3c60c59
SW		 5615 mutex_unlock(&mdsc->mutex);
5616
38d46409 5617 doutc(cl, "waiting for sessions to close\n");
fcff415c
YZ		 5618 wait_event_timeout(mdsc->session_close_wq,
5619 done_closing_sessions(mdsc, skipped),
a319bf56 5620 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 5621
5622 /* tear down remaining sessions */
f3c60c59 5623 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 5624 for (i = 0; i < mdsc->max_sessions; i++) {
5625 if (mdsc->sessions[i]) {
5b3248c6 5626 session = ceph_get_mds_session(mdsc->sessions[i]);
2600d2dd 5627 __unregister_session(mdsc, session);
2f2dc053
SW		 5628 mutex_unlock(&mdsc->mutex);
5629 mutex_lock(&session->s_mutex);
5630 remove_session_caps(session);
5631 mutex_unlock(&session->s_mutex);
5632 ceph_put_mds_session(session);
5633 mutex_lock(&mdsc->mutex);
5634 }
5635 }
2f2dc053 5636 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 5637 mutex_unlock(&mdsc->mutex);
5638
75c9627e 5639 ceph_cleanup_snapid_map(mdsc);
5ed91587 5640 ceph_cleanup_global_and_empty_realms(mdsc);
2f2dc053 5641
37c4efc1 5642 cancel_work_sync(&mdsc->cap_reclaim_work);
2f2dc053
SW		 5643 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
5644
38d46409 5645 doutc(cl, "done\n");
2f2dc053
SW		 5646}
5647
48fec5d0
YZ		 5648void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
5649{
5650 struct ceph_mds_session *session;
5651 int mds;
5652
38d46409 5653 doutc(mdsc->fsc->client, "force umount\n");
48fec5d0
YZ		 5654
5655 mutex_lock(&mdsc->mutex);
5656 for (mds = 0; mds < mdsc->max_sessions; mds++) {
5657 session = __ceph_lookup_mds_session(mdsc, mds);
5658 if (!session)
5659 continue;
d468e729
YZ		 5660
5661 if (session->s_state == CEPH_MDS_SESSION_REJECTED)
5662 __unregister_session(mdsc, session);
5663 __wake_requests(mdsc, &session->s_waiting);
48fec5d0 5664 mutex_unlock(&mdsc->mutex);
d468e729 5665
48fec5d0
YZ		 5666 mutex_lock(&session->s_mutex);
5667 __close_session(mdsc, session);
5668 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
5669 cleanup_session_requests(mdsc, session);
5670 remove_session_caps(session);
5671 }
5672 mutex_unlock(&session->s_mutex);
5673 ceph_put_mds_session(session);
d468e729 5674
48fec5d0
YZ		 5675 mutex_lock(&mdsc->mutex);
5676 kick_requests(mdsc, mds);
5677 }
5678 __wake_requests(mdsc, &mdsc->waiting_for_map);
5679 mutex_unlock(&mdsc->mutex);
5680}
5681
3d14c5d2 5682static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053 5683{
38d46409 5684 doutc(mdsc->fsc->client, "stop\n");
fa996773
XL		 5685 /*
5686 * Make sure the delayed work stopped before releasing
5687 * the resources.
5688 *
5689 * Because the cancel_delayed_work_sync() will only
5690 * guarantee that the work finishes executing. But the
5691 * delayed work will re-arm itself again after that.
5692 */
5693 flush_delayed_work(&mdsc->delayed_work);
5694
2f2dc053
SW		 5695 if (mdsc->mdsmap)
5696 ceph_mdsmap_destroy(mdsc->mdsmap);
5697 kfree(mdsc->sessions);
37151668 5698 ceph_caps_finalize(mdsc);
10183a69 5699 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 5700}
5701
3d14c5d2
YS		 5702void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
5703{
5704 struct ceph_mds_client *mdsc = fsc->mdsc;
38d46409 5705 doutc(fsc->client, "%p\n", mdsc);
ef550f6f 5706
50c55aec
CX		 5707 if (!mdsc)
5708 return;
5709
ef550f6f
SW		 5710 /* flush out any connection work with references to us */
5711 ceph_msgr_flush();
5712
62a65f36
YZ		 5713 ceph_mdsc_stop(mdsc);
5714
f9009efa
XL		 5715 ceph_metric_destroy(&mdsc->metric);
5716
3d14c5d2
YS		 5717 fsc->mdsc = NULL;
5718 kfree(mdsc);
38d46409 5719 doutc(fsc->client, "%p done\n", mdsc);
3d14c5d2
YS		 5720}
5721
430afbad
YZ		 5722void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5723{
5724 struct ceph_fs_client *fsc = mdsc->fsc;
38d46409 5725 struct ceph_client *cl = fsc->client;
430afbad
YZ		 5726 const char *mds_namespace = fsc->mount_options->mds_namespace;
5727 void *p = msg->front.iov_base;
5728 void *end = p + msg->front.iov_len;
5729 u32 epoch;
430afbad
YZ		 5730 u32 num_fs;
5731 u32 mount_fscid = (u32)-1;
430afbad
YZ		 5732 int err = -EINVAL;
5733
5734 ceph_decode_need(&p, end, sizeof(u32), bad);
5735 epoch = ceph_decode_32(&p);
5736
38d46409 5737 doutc(cl, "epoch %u\n", epoch);
430afbad 5738
06a1ad43
JL		 5739 /* struct_v, struct_cv, map_len, epoch, legacy_client_fscid */
5740 ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 3, bad);
430afbad 5741
06a1ad43 5742 ceph_decode_32_safe(&p, end, num_fs, bad);
430afbad
YZ		 5743 while (num_fs-- > 0) {
5744 void *info_p, *info_end;
5745 u32 info_len;
430afbad
YZ		 5746 u32 fscid, namelen;
5747
5748 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
06a1ad43 5749 p += 2; // info_v, info_cv
430afbad
YZ		 5750 info_len = ceph_decode_32(&p);
5751 ceph_decode_need(&p, end, info_len, bad);
5752 info_p = p;
5753 info_end = p + info_len;
5754 p = info_end;
5755
5756 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
5757 fscid = ceph_decode_32(&info_p);
5758 namelen = ceph_decode_32(&info_p);
5759 ceph_decode_need(&info_p, info_end, namelen, bad);
5760
5761 if (mds_namespace &&
5762 strlen(mds_namespace) == namelen &&
5763 !strncmp(mds_namespace, (char *)info_p, namelen)) {
5764 mount_fscid = fscid;
5765 break;
5766 }
5767 }
5768
5769 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
5770 if (mount_fscid != (u32)-1) {
5771 fsc->client->monc.fs_cluster_id = mount_fscid;
5772 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
5773 0, true);
5774 ceph_monc_renew_subs(&fsc->client->monc);
5775 } else {
5776 err = -ENOENT;
5777 goto err_out;
5778 }
5779 return;
76bd6ec4 5780
430afbad 5781bad:
38d46409
XL		 5782 pr_err_client(cl, "error decoding fsmap %d. Shutting down mount.\n",
5783 err);
631ed4b0 5784 ceph_umount_begin(mdsc->fsc->sb);
8b0da5c5 5785 ceph_msg_dump(msg);
430afbad
YZ		 5786err_out:
5787 mutex_lock(&mdsc->mutex);
76bd6ec4 5788 mdsc->mdsmap_err = err;
430afbad
YZ		 5789 __wake_requests(mdsc, &mdsc->waiting_for_map);
5790 mutex_unlock(&mdsc->mutex);
430afbad 5791}
2f2dc053
SW		 5792
5793/*
5794 * handle mds map update.
5795 */
430afbad 5796void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
2f2dc053 5797{
38d46409 5798 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 5799 u32 epoch;
5800 u32 maplen;
5801 void *p = msg->front.iov_base;
5802 void *end = p + msg->front.iov_len;
5803 struct ceph_mdsmap *newmap, *oldmap;
5804 struct ceph_fsid fsid;
5805 int err = -EINVAL;
5806
5807 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
5808 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 5809 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 5810 return;
c89136ea
SW		 5811 epoch = ceph_decode_32(&p);
5812 maplen = ceph_decode_32(&p);
38d46409 5813 doutc(cl, "epoch %u len %d\n", epoch, (int)maplen);
2f2dc053
SW		 5814
5815 /* do we need it? */
2f2dc053
SW		 5816 mutex_lock(&mdsc->mutex);
5817 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
38d46409 5818 doutc(cl, "epoch %u <= our %u\n", epoch, mdsc->mdsmap->m_epoch);
2f2dc053
SW		 5819 mutex_unlock(&mdsc->mutex);
5820 return;
5821 }
5822
197b7d79 5823 newmap = ceph_mdsmap_decode(mdsc, &p, end, ceph_msgr2(mdsc->fsc->client));
2f2dc053
SW		 5824 if (IS_ERR(newmap)) {
5825 err = PTR_ERR(newmap);
5826 goto bad_unlock;
5827 }
5828
5829 /* swap into place */
5830 if (mdsc->mdsmap) {
5831 oldmap = mdsc->mdsmap;
5832 mdsc->mdsmap = newmap;
5833 check_new_map(mdsc, newmap, oldmap);
5834 ceph_mdsmap_destroy(oldmap);
5835 } else {
5836 mdsc->mdsmap = newmap; /* first mds map */
5837 }
719784ba
CX		 5838 mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
5839 MAX_LFS_FILESIZE);
2f2dc053
SW		 5840
5841 __wake_requests(mdsc, &mdsc->waiting_for_map);
82dcabad
ID		 5842 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
5843 mdsc->mdsmap->m_epoch);
2f2dc053
SW		 5844
5845 mutex_unlock(&mdsc->mutex);
bf2ba432 5846 schedule_delayed(mdsc, 0);
2f2dc053
SW		 5847 return;
5848
5849bad_unlock:
5850 mutex_unlock(&mdsc->mutex);
5851bad:
38d46409
XL		 5852 pr_err_client(cl, "error decoding mdsmap %d. Shutting down mount.\n",
5853 err);
631ed4b0 5854 ceph_umount_begin(mdsc->fsc->sb);
8b0da5c5 5855 ceph_msg_dump(msg);
2f2dc053
SW		 5856 return;
5857}
5858
4972cf60 5859static struct ceph_connection *mds_get_con(struct ceph_connection *con)
2f2dc053
SW		 5860{
5861 struct ceph_mds_session *s = con->private;
5862
5b3248c6 5863 if (ceph_get_mds_session(s))
2f2dc053 5864 return con;
2f2dc053
SW		 5865 return NULL;
5866}
5867
4972cf60 5868static void mds_put_con(struct ceph_connection *con)
2f2dc053
SW		 5869{
5870 struct ceph_mds_session *s = con->private;
5871
2f2dc053
SW		 5872 ceph_put_mds_session(s);
5873}
5874
5875/*
5876 * if the client is unresponsive for long enough, the mds will kill
5877 * the session entirely.
5878 */
4972cf60 5879static void mds_peer_reset(struct ceph_connection *con)
2f2dc053
SW		 5880{
5881 struct ceph_mds_session *s = con->private;
7e70f0ed 5882 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 5883
38d46409
XL		 5884 pr_warn_client(mdsc->fsc->client, "mds%d closed our session\n",
5885 s->s_mds);
cbcb358b
XL		 5886 if (READ_ONCE(mdsc->fsc->mount_state) != CEPH_MOUNT_FENCE_IO &&
5887 ceph_mdsmap_get_state(mdsc->mdsmap, s->s_mds) >= CEPH_MDS_STATE_RECONNECT)
a68e564a 5888 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 5889}
5890
4972cf60 5891static void mds_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
2f2dc053
SW		 5892{
5893 struct ceph_mds_session *s = con->private;
5894 struct ceph_mds_client *mdsc = s->s_mdsc;
38d46409 5895 struct ceph_client *cl = mdsc->fsc->client;
2f2dc053
SW		 5896 int type = le16_to_cpu(msg->hdr.type);
5897
2600d2dd
SW		 5898 mutex_lock(&mdsc->mutex);
5899 if (__verify_registered_session(mdsc, s) < 0) {
5900 mutex_unlock(&mdsc->mutex);
5901 goto out;
5902 }
5903 mutex_unlock(&mdsc->mutex);
5904
2f2dc053
SW		 5905 switch (type) {
5906 case CEPH_MSG_MDS_MAP:
430afbad
YZ		 5907 ceph_mdsc_handle_mdsmap(mdsc, msg);
5908 break;
5909 case CEPH_MSG_FS_MAP_USER:
5910 ceph_mdsc_handle_fsmap(mdsc, msg);
2f2dc053
SW		 5911 break;
5912 case CEPH_MSG_CLIENT_SESSION:
5913 handle_session(s, msg);
5914 break;
5915 case CEPH_MSG_CLIENT_REPLY:
5916 handle_reply(s, msg);
5917 break;
5918 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 5919 handle_forward(mdsc, s, msg);
2f2dc053
SW		 5920 break;
5921 case CEPH_MSG_CLIENT_CAPS:
5922 ceph_handle_caps(s, msg);
5923 break;
5924 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 5925 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 5926 break;
5927 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 5928 handle_lease(mdsc, s, msg);
2f2dc053 5929 break;
fb18a575
LH		 5930 case CEPH_MSG_CLIENT_QUOTA:
5931 ceph_handle_quota(mdsc, s, msg);
5932 break;
2f2dc053
SW		 5933
5934 default:
38d46409
XL		 5935 pr_err_client(cl, "received unknown message type %d %s\n",
5936 type, ceph_msg_type_name(type));
2f2dc053 5937 }
2600d2dd 5938out:
2f2dc053
SW		 5939 ceph_msg_put(msg);
5940}
5941
4e7a5dcd
SW		 5942/*
5943 * authentication
5944 */
a3530df3
AE		 5945
5946/*
5947 * Note: returned pointer is the address of a structure that's
5948 * managed separately. Caller must *not* attempt to free it.
5949 */
4972cf60
ID		 5950static struct ceph_auth_handshake *
5951mds_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
4e7a5dcd
SW		 5952{
5953 struct ceph_mds_session *s = con->private;
5954 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 5955 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 5956 struct ceph_auth_handshake *auth = &s->s_auth;
ce287162 5957 int ret;
4e7a5dcd 5958
ce287162
ID		 5959 ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5960 force_new, proto, NULL, NULL);
5961 if (ret)
5962 return ERR_PTR(ret);
74f1869f 5963
a3530df3 5964 return auth;
4e7a5dcd
SW		 5965}
5966
4972cf60 5967static int mds_add_authorizer_challenge(struct ceph_connection *con,
6daca13d
ID		 5968 void *challenge_buf, int challenge_buf_len)
5969{
5970 struct ceph_mds_session *s = con->private;
5971 struct ceph_mds_client *mdsc = s->s_mdsc;
5972 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5973
5974 return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
5975 challenge_buf, challenge_buf_len);
5976}
4e7a5dcd 5977
4972cf60 5978static int mds_verify_authorizer_reply(struct ceph_connection *con)
4e7a5dcd
SW		 5979{
5980 struct ceph_mds_session *s = con->private;
5981 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 5982 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
285ea34f 5983 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 5984
285ea34f
ID		 5985 return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5986 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5987 NULL, NULL, NULL, NULL);
4e7a5dcd
SW		 5988}
5989
4972cf60 5990static int mds_invalidate_authorizer(struct ceph_connection *con)
9bd2e6f8
SW		 5991{
5992 struct ceph_mds_session *s = con->private;
5993 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 5994 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 5995
27859f97 5996 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 5997
3d14c5d2 5998 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 5999}
6000
cd1a677c
ID		 6001static int mds_get_auth_request(struct ceph_connection *con,
6002 void *buf, int *buf_len,
6003 void **authorizer, int *authorizer_len)
6004{
6005 struct ceph_mds_session *s = con->private;
6006 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
6007 struct ceph_auth_handshake *auth = &s->s_auth;
6008 int ret;
6009
6010 ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
6011 buf, buf_len);
6012 if (ret)
6013 return ret;
6014
6015 *authorizer = auth->authorizer_buf;
6016 *authorizer_len = auth->authorizer_buf_len;
6017 return 0;
6018}
6019
6020static int mds_handle_auth_reply_more(struct ceph_connection *con,
6021 void *reply, int reply_len,
6022 void *buf, int *buf_len,
6023 void **authorizer, int *authorizer_len)
6024{
6025 struct ceph_mds_session *s = con->private;
6026 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
6027 struct ceph_auth_handshake *auth = &s->s_auth;
6028 int ret;
6029
6030 ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
6031 buf, buf_len);
6032 if (ret)
6033 return ret;
6034
6035 *authorizer = auth->authorizer_buf;
6036 *authorizer_len = auth->authorizer_buf_len;
6037 return 0;
6038}
6039
6040static int mds_handle_auth_done(struct ceph_connection *con,
6041 u64 global_id, void *reply, int reply_len,
6042 u8 *session_key, int *session_key_len,
6043 u8 *con_secret, int *con_secret_len)
6044{
6045 struct ceph_mds_session *s = con->private;
6046 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
6047 struct ceph_auth_handshake *auth = &s->s_auth;
6048
6049 return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
6050 session_key, session_key_len,
6051 con_secret, con_secret_len);
6052}
6053
6054static int mds_handle_auth_bad_method(struct ceph_connection *con,
6055 int used_proto, int result,
6056 const int *allowed_protos, int proto_cnt,
6057 const int *allowed_modes, int mode_cnt)
6058{
6059 struct ceph_mds_session *s = con->private;
6060 struct ceph_mon_client *monc = &s->s_mdsc->fsc->client->monc;
6061 int ret;
6062
6063 if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_MDS,
6064 used_proto, result,
6065 allowed_protos, proto_cnt,
6066 allowed_modes, mode_cnt)) {
6067 ret = ceph_monc_validate_auth(monc);
6068 if (ret)
6069 return ret;
6070 }
6071
6072 return -EACCES;
6073}
6074
53ded495
AE		 6075static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
6076 struct ceph_msg_header *hdr, int *skip)
6077{
6078 struct ceph_msg *msg;
6079 int type = (int) le16_to_cpu(hdr->type);
6080 int front_len = (int) le32_to_cpu(hdr->front_len);
6081
6082 if (con->in_msg)
6083 return con->in_msg;
6084
6085 *skip = 0;
6086 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
6087 if (!msg) {
6088 pr_err("unable to allocate msg type %d len %d\n",
6089 type, front_len);
6090 return NULL;
6091 }
53ded495
AE		 6092
6093 return msg;
6094}
6095
79dbd1ba 6096static int mds_sign_message(struct ceph_msg *msg)
33d07337 6097{
79dbd1ba 6098 struct ceph_mds_session *s = msg->con->private;
33d07337 6099 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 6100
33d07337
YZ		 6101 return ceph_auth_sign_message(auth, msg);
6102}
6103
79dbd1ba 6104static int mds_check_message_signature(struct ceph_msg *msg)
33d07337 6105{
79dbd1ba 6106 struct ceph_mds_session *s = msg->con->private;
33d07337 6107 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 6108
33d07337
YZ		 6109 return ceph_auth_check_message_signature(auth, msg);
6110}
6111
9e32789f 6112static const struct ceph_connection_operations mds_con_ops = {
4972cf60
ID		 6113 .get = mds_get_con,
6114 .put = mds_put_con,
53ded495 6115 .alloc_msg = mds_alloc_msg,
4972cf60
ID		 6116 .dispatch = mds_dispatch,
6117 .peer_reset = mds_peer_reset,
6118 .get_authorizer = mds_get_authorizer,
6119 .add_authorizer_challenge = mds_add_authorizer_challenge,
6120 .verify_authorizer_reply = mds_verify_authorizer_reply,
6121 .invalidate_authorizer = mds_invalidate_authorizer,
79dbd1ba
ID		 6122 .sign_message = mds_sign_message,
6123 .check_message_signature = mds_check_message_signature,
cd1a677c
ID		 6124 .get_auth_request = mds_get_auth_request,
6125 .handle_auth_reply_more = mds_handle_auth_reply_more,
6126 .handle_auth_done = mds_handle_auth_done,
6127 .handle_auth_bad_method = mds_handle_auth_bad_method,
2f2dc053
SW		 6128};
6129
2f2dc053 6130/* eof */
Linux 6.x block layer and io_uring tree(s)