]> www.infradead.org Git - users/dwmw2/linux.git/blob
1740047
[users/dwmw2/linux.git] /
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/fs.h>
5 #include <linux/wait.h>
6 #include <linux/slab.h>
7 #include <linux/gfp.h>
8 #include <linux/sched.h>
9 #include <linux/debugfs.h>
10 #include <linux/seq_file.h>
11 #include <linux/ratelimit.h>
12 #include <linux/bits.h>
13 #include <linux/ktime.h>
14 #include <linux/bitmap.h>
15 #include <linux/mnt_idmapping.h>
16
17 #include "super.h"
18 #include "mds_client.h"
19 #include "crypto.h"
20
21 #include <linux/ceph/ceph_features.h>
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>
27
28 #define RECONNECT_MAX_SIZE (INT_MAX - PAGE_SIZE)
29
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52  * are no longer valid.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
55 struct ceph_reconnect_state {
56         struct ceph_mds_session *session;
57         int nr_caps, nr_realms;
58         struct ceph_pagelist *pagelist;
59         unsigned msg_version;
60         bool allow_multi;
61 };
62
63 static void __wake_requests(struct ceph_mds_client *mdsc,
64                             struct list_head *head);
65 static void ceph_cap_release_work(struct work_struct *work);
66 static void ceph_cap_reclaim_work(struct work_struct *work);
67
68 static const struct ceph_connection_operations mds_con_ops;
69
70
71 /*
72  * mds reply parsing
73  */
74
75 static 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;
94 bad:
95         return -EIO;
96 }
97
98 /*
99  * parse individual inode info
100  */
101 static int parse_reply_info_in(void **p, void *end,
102                                struct ceph_mds_reply_info_in *info,
103                                u64 features)
104 {
105         int err = 0;
106         u8 struct_v = 0;
107
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);
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
133         ceph_decode_copy_safe(p, end, &info->dir_layout,
134                               sizeof(info->dir_layout), bad);
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;
139
140         if (features == (u64)-1) {
141                 /* inline data */
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;
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                 }
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 */
164                 ceph_decode_64_safe(p, end, info->change_attr, bad);
165
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
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
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
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                 }
234                 *p = end;
235         } else {
236                 /* legacy (unversioned) struct */
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                 }
265
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));
269                         ceph_decode_64_safe(p, end, info->change_attr, bad);
270                 }
271
272                 info->dir_pin = -ENODATA;
273                 /* info->snap_btime and info->rsnaps remain zero */
274         }
275         return 0;
276 bad:
277         err = -EIO;
278 out_bad:
279         return err;
280 }
281
282 static 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) {
287                 u8 struct_v, struct_compat;
288                 u32 struct_len;
289                 ceph_decode_8_safe(p, end, struct_v, bad);
290                 ceph_decode_8_safe(p, end, struct_compat, bad);
291                 /* struct_v is expected to be >= 1. we only understand
292                  * encoding whose struct_compat == 1. */
293                 if (!struct_v || struct_compat != 1)
294                         goto bad;
295                 ceph_decode_32_safe(p, end, struct_len, bad);
296                 ceph_decode_need(p, end, struct_len, bad);
297                 end = *p + struct_len;
298         }
299
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;
308 bad:
309         return -EIO;
310 }
311
312 static int parse_reply_info_lease(void **p, void *end,
313                                   struct ceph_mds_reply_lease **lease,
314                                   u64 features, u32 *altname_len, u8 **altname)
315 {
316         u8 struct_v;
317         u32 struct_len;
318         void *lend;
319
320         if (features == (u64)-1) {
321                 u8 struct_compat;
322
323                 ceph_decode_8_safe(p, end, struct_v, bad);
324                 ceph_decode_8_safe(p, end, struct_compat, bad);
325
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;
330
331                 ceph_decode_32_safe(p, end, struct_len, bad);
332         } else {
333                 struct_len = sizeof(**lease);
334                 *altname_len = 0;
335                 *altname = NULL;
336         }
337
338         lend = *p + struct_len;
339         ceph_decode_need(p, end, struct_len, bad);
340         *lease = *p;
341         *p += sizeof(**lease);
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;
355         return 0;
356 bad:
357         return -EIO;
358 }
359
360 /*
361  * parse a normal reply, which may contain a (dir+)dentry and/or a
362  * target inode.
363  */
364 static int parse_reply_info_trace(void **p, void *end,
365                                   struct ceph_mds_reply_info_parsed *info,
366                                   u64 features)
367 {
368         int err;
369
370         if (info->head->is_dentry) {
371                 err = parse_reply_info_in(p, end, &info->diri, features);
372                 if (err < 0)
373                         goto out_bad;
374
375                 err = parse_reply_info_dir(p, end, &info->dirfrag, features);
376                 if (err < 0)
377                         goto out_bad;
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;
383
384                 err = parse_reply_info_lease(p, end, &info->dlease, features,
385                                              &info->altname_len, &info->altname);
386                 if (err < 0)
387                         goto out_bad;
388         }
389
390         if (info->head->is_target) {
391                 err = parse_reply_info_in(p, end, &info->targeti, features);
392                 if (err < 0)
393                         goto out_bad;
394         }
395
396         if (unlikely(*p != end))
397                 goto bad;
398         return 0;
399
400 bad:
401         err = -EIO;
402 out_bad:
403         pr_err("problem parsing mds trace %d\n", err);
404         return err;
405 }
406
407 /*
408  * parse readdir results
409  */
410 static int parse_reply_info_readdir(void **p, void *end,
411                                     struct ceph_mds_request *req,
412                                     u64 features)
413 {
414         struct ceph_mds_reply_info_parsed *info = &req->r_reply_info;
415         struct ceph_client *cl = req->r_mdsc->fsc->client;
416         u32 num, i = 0;
417         int err;
418
419         err = parse_reply_info_dir(p, end, &info->dir_dir, features);
420         if (err < 0)
421                 goto out_bad;
422
423         ceph_decode_need(p, end, sizeof(num) + 2, bad);
424         num = ceph_decode_32(p);
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);
429                 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
430                 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
431         }
432         if (num == 0)
433                 goto done;
434
435         BUG_ON(!info->dir_entries);
436         if ((unsigned long)(info->dir_entries + num) >
437             (unsigned long)info->dir_entries + info->dir_buf_size) {
438                 pr_err_client(cl, "dir contents are larger than expected\n");
439                 WARN_ON(1);
440                 goto bad;
441         }
442
443         info->dir_nr = num;
444         while (num) {
445                 struct inode *inode = d_inode(req->r_dentry);
446                 struct ceph_inode_info *ci = ceph_inode(inode);
447                 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
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
454                 /* dentry */
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;
459                 doutc(cl, "parsed dir dname '%.*s'\n", _name_len, _name);
460
461                 if (info->hash_order)
462                         rde->raw_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
463                                                       _name, _name_len);
464
465                 /* dentry lease */
466                 err = parse_reply_info_lease(p, end, &rde->lease, features,
467                                              &altname_len, &altname);
468                 if (err)
469                         goto out_bad;
470
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) {
519                         pr_err_client(cl, "unable to decode %.*s, got %d\n",
520                                       _name_len, _name, err);
521                         goto out_bad;
522                 }
523                 rde->name = oname.name;
524                 rde->name_len = oname.len;
525
526                 /* inode */
527                 err = parse_reply_info_in(p, end, &rde->inode, features);
528                 if (err < 0)
529                         goto out_bad;
530                 /* ceph_readdir_prepopulate() will update it */
531                 rde->offset = 0;
532                 i++;
533                 num--;
534         }
535
536 done:
537         /* Skip over any unrecognized fields */
538         *p = end;
539         return 0;
540
541 bad:
542         err = -EIO;
543 out_bad:
544         pr_err_client(cl, "problem parsing dir contents %d\n", err);
545         return err;
546 }
547
548 /*
549  * parse fcntl F_GETLK results
550  */
551 static int parse_reply_info_filelock(void **p, void *end,
552                                      struct ceph_mds_reply_info_parsed *info,
553                                      u64 features)
554 {
555         if (*p + sizeof(*info->filelock_reply) > end)
556                 goto bad;
557
558         info->filelock_reply = *p;
559
560         /* Skip over any unrecognized fields */
561         *p = end;
562         return 0;
563 bad:
564         return -EIO;
565 }
566
567
568 #if BITS_PER_LONG == 64
569
570 #define DELEGATED_INO_AVAILABLE         xa_mk_value(1)
571
572 static int ceph_parse_deleg_inos(void **p, void *end,
573                                  struct ceph_mds_session *s)
574 {
575         struct ceph_client *cl = s->s_mdsc->fsc->client;
576         u32 sets;
577
578         ceph_decode_32_safe(p, end, sets, bad);
579         doutc(cl, "got %u sets of delegated inodes\n", sets);
580         while (sets--) {
581                 u64 start, len;
582
583                 ceph_decode_64_safe(p, end, start, bad);
584                 ceph_decode_64_safe(p, end, len, bad);
585
586                 /* Don't accept a delegation of system inodes */
587                 if (start < CEPH_INO_SYSTEM_BASE) {
588                         pr_warn_ratelimited_client(cl,
589                                 "ignoring reserved inode range delegation (start=0x%llx len=0x%llx)\n",
590                                 start, len);
591                         continue;
592                 }
593                 while (len--) {
594                         int err = xa_insert(&s->s_delegated_inos, start++,
595                                             DELEGATED_INO_AVAILABLE,
596                                             GFP_KERNEL);
597                         if (!err) {
598                                 doutc(cl, "added delegated inode 0x%llx\n", start - 1);
599                         } else if (err == -EBUSY) {
600                                 pr_warn_client(cl,
601                                         "MDS delegated inode 0x%llx more than once.\n",
602                                         start - 1);
603                         } else {
604                                 return err;
605                         }
606                 }
607         }
608         return 0;
609 bad:
610         return -EIO;
611 }
612
613 u64 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
626 int 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  */
637 static 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;
646 bad:
647         return -EIO;
648 }
649
650 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
651 {
652         return 0;
653 }
654
655 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
656 {
657         return 0;
658 }
659 #endif /* BITS_PER_LONG == 64 */
660
661 /*
662  * parse create results
663  */
664 static int parse_reply_info_create(void **p, void *end,
665                                   struct ceph_mds_reply_info_parsed *info,
666                                   u64 features, struct ceph_mds_session *s)
667 {
668         int ret;
669
670         if (features == (u64)-1 ||
671             (features & CEPH_FEATURE_REPLY_CREATE_INODE)) {
672                 if (*p == end) {
673                         /* Malformed reply? */
674                         info->has_create_ino = false;
675                 } else if (test_bit(CEPHFS_FEATURE_DELEG_INO, &s->s_features)) {
676                         info->has_create_ino = true;
677                         /* struct_v, struct_compat, and len */
678                         ceph_decode_skip_n(p, end, 2 + sizeof(u32), bad);
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 */
685                         ceph_decode_64_safe(p, end, info->ino, bad);
686                         info->has_create_ino = true;
687                 }
688         } else {
689                 if (*p != end)
690                         goto bad;
691         }
692
693         /* Skip over any unrecognized fields */
694         *p = end;
695         return 0;
696 bad:
697         return -EIO;
698 }
699
700 static 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         }
718 bad:
719         return -EIO;
720 }
721
722 /*
723  * parse extra results
724  */
725 static int parse_reply_info_extra(void **p, void *end,
726                                   struct ceph_mds_request *req,
727                                   u64 features, struct ceph_mds_session *s)
728 {
729         struct ceph_mds_reply_info_parsed *info = &req->r_reply_info;
730         u32 op = le32_to_cpu(info->head->op);
731
732         if (op == CEPH_MDS_OP_GETFILELOCK)
733                 return parse_reply_info_filelock(p, end, info, features);
734         else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
735                 return parse_reply_info_readdir(p, end, req, features);
736         else if (op == CEPH_MDS_OP_CREATE)
737                 return parse_reply_info_create(p, end, info, features, s);
738         else if (op == CEPH_MDS_OP_GETVXATTR)
739                 return parse_reply_info_getvxattr(p, end, info, features);
740         else
741                 return -EIO;
742 }
743
744 /*
745  * parse entire mds reply
746  */
747 static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,
748                             struct ceph_mds_request *req, u64 features)
749 {
750         struct ceph_mds_reply_info_parsed *info = &req->r_reply_info;
751         struct ceph_client *cl = s->s_mdsc->fsc->client;
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) {
763                 ceph_decode_need(&p, end, len, bad);
764                 err = parse_reply_info_trace(&p, p+len, info, features);
765                 if (err < 0)
766                         goto out_bad;
767         }
768
769         /* extra */
770         ceph_decode_32_safe(&p, end, len, bad);
771         if (len > 0) {
772                 ceph_decode_need(&p, end, len, bad);
773                 err = parse_reply_info_extra(&p, p+len, req, features, s);
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
788 bad:
789         err = -EIO;
790 out_bad:
791         pr_err_client(cl, "mds parse_reply err %d\n", err);
792         ceph_msg_dump(msg);
793         return err;
794 }
795
796 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
797 {
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);
804         if (!info->dir_entries)
805                 return;
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         }
813         free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
814 }
815
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  * CDentry. That means the just created file is possibly 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  */
836 int ceph_wait_on_conflict_unlink(struct dentry *dentry)
837 {
838         struct ceph_fs_client *fsc = ceph_sb_to_fs_client(dentry->d_sb);
839         struct ceph_client *cl = fsc->client;
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))
864                         pr_warn_client(cl, "dentry %p:%pd async unlink bit is not set\n",
865                                        dentry, dentry);
866
867                 if (!d_same_name(udentry, pdentry, &dname))
868                         goto next;
869
870                 found = dget_dlock(udentry);
871                 spin_unlock(&udentry->d_lock);
872                 break;
873 next:
874                 spin_unlock(&udentry->d_lock);
875         }
876         rcu_read_unlock();
877
878         if (likely(!found))
879                 return 0;
880
881         doutc(cl, "dentry %p:%pd conflict with old %p:%pd\n", dentry, dentry,
882               found, found);
883
884         err = wait_on_bit(&di->flags, CEPH_DENTRY_ASYNC_UNLINK_BIT,
885                           TASK_KILLABLE);
886         dput(found);
887         return err;
888 }
889
890
891 /*
892  * sessions
893  */
894 const char *ceph_session_state_name(int s)
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";
902         case CEPH_MDS_SESSION_CLOSED: return "closed";
903         case CEPH_MDS_SESSION_RESTARTING: return "restarting";
904         case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
905         case CEPH_MDS_SESSION_REJECTED: return "rejected";
906         default: return "???";
907         }
908 }
909
910 struct ceph_mds_session *ceph_get_mds_session(struct ceph_mds_session *s)
911 {
912         if (refcount_inc_not_zero(&s->s_ref))
913                 return s;
914         return NULL;
915 }
916
917 void ceph_put_mds_session(struct ceph_mds_session *s)
918 {
919         if (IS_ERR_OR_NULL(s))
920                 return;
921
922         if (refcount_dec_and_test(&s->s_ref)) {
923                 if (s->s_auth.authorizer)
924                         ceph_auth_destroy_authorizer(s->s_auth.authorizer);
925                 WARN_ON(mutex_is_locked(&s->s_mutex));
926                 xa_destroy(&s->s_delegated_inos);
927                 kfree(s);
928         }
929 }
930
931 /*
932  * called under mdsc->mutex
933  */
934 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
935                                                    int mds)
936 {
937         if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
938                 return NULL;
939         return ceph_get_mds_session(mdsc->sessions[mds]);
940 }
941
942 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
943 {
944         if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
945                 return false;
946         else
947                 return true;
948 }
949
950 static 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
959 /*
960  * create+register a new session for given mds.
961  * called under mdsc->mutex.
962  */
963 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
964                                                  int mds)
965 {
966         struct ceph_client *cl = mdsc->fsc->client;
967         struct ceph_mds_session *s;
968
969         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO)
970                 return ERR_PTR(-EIO);
971
972         if (mds >= mdsc->mdsmap->possible_max_rank)
973                 return ERR_PTR(-EINVAL);
974
975         s = kzalloc(sizeof(*s), GFP_NOFS);
976         if (!s)
977                 return ERR_PTR(-ENOMEM);
978
979         if (mds >= mdsc->max_sessions) {
980                 int newmax = 1 << get_count_order(mds + 1);
981                 struct ceph_mds_session **sa;
982                 size_t ptr_size = sizeof(struct ceph_mds_session *);
983
984                 doutc(cl, "realloc to %d\n", newmax);
985                 sa = kcalloc(newmax, ptr_size, GFP_NOFS);
986                 if (!sa)
987                         goto fail_realloc;
988                 if (mdsc->sessions) {
989                         memcpy(sa, mdsc->sessions,
990                                mdsc->max_sessions * ptr_size);
991                         kfree(mdsc->sessions);
992                 }
993                 mdsc->sessions = sa;
994                 mdsc->max_sessions = newmax;
995         }
996
997         doutc(cl, "mds%d\n", mds);
998         s->s_mdsc = mdsc;
999         s->s_mds = mds;
1000         s->s_state = CEPH_MDS_SESSION_NEW;
1001         mutex_init(&s->s_mutex);
1002
1003         ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
1004
1005         atomic_set(&s->s_cap_gen, 1);
1006         s->s_cap_ttl = jiffies - 1;
1007
1008         spin_lock_init(&s->s_cap_lock);
1009         INIT_LIST_HEAD(&s->s_caps);
1010         refcount_set(&s->s_ref, 1);
1011         INIT_LIST_HEAD(&s->s_waiting);
1012         INIT_LIST_HEAD(&s->s_unsafe);
1013         xa_init(&s->s_delegated_inos);
1014         INIT_LIST_HEAD(&s->s_cap_releases);
1015         INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
1016
1017         INIT_LIST_HEAD(&s->s_cap_dirty);
1018         INIT_LIST_HEAD(&s->s_cap_flushing);
1019
1020         mdsc->sessions[mds] = s;
1021         atomic_inc(&mdsc->num_sessions);
1022         refcount_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
1023
1024         ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
1025                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
1026
1027         return s;
1028
1029 fail_realloc:
1030         kfree(s);
1031         return ERR_PTR(-ENOMEM);
1032 }
1033
1034 /*
1035  * called under mdsc->mutex
1036  */
1037 static void __unregister_session(struct ceph_mds_client *mdsc,
1038                                struct ceph_mds_session *s)
1039 {
1040         doutc(mdsc->fsc->client, "mds%d %p\n", s->s_mds, s);
1041         BUG_ON(mdsc->sessions[s->s_mds] != s);
1042         mdsc->sessions[s->s_mds] = NULL;
1043         ceph_con_close(&s->s_con);
1044         ceph_put_mds_session(s);
1045         atomic_dec(&mdsc->num_sessions);
1046 }
1047
1048 /*
1049  * drop session refs in request.
1050  *
1051  * should be last request ref, or hold mdsc->mutex
1052  */
1053 static void put_request_session(struct ceph_mds_request *req)
1054 {
1055         if (req->r_session) {
1056                 ceph_put_mds_session(req->r_session);
1057                 req->r_session = NULL;
1058         }
1059 }
1060
1061 void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
1062                                 void (*cb)(struct ceph_mds_session *),
1063                                 bool check_state)
1064 {
1065         int mds;
1066
1067         mutex_lock(&mdsc->mutex);
1068         for (mds = 0; mds < mdsc->max_sessions; ++mds) {
1069                 struct ceph_mds_session *s;
1070
1071                 s = __ceph_lookup_mds_session(mdsc, mds);
1072                 if (!s)
1073                         continue;
1074
1075                 if (check_state && !check_session_state(s)) {
1076                         ceph_put_mds_session(s);
1077                         continue;
1078                 }
1079
1080                 mutex_unlock(&mdsc->mutex);
1081                 cb(s);
1082                 ceph_put_mds_session(s);
1083                 mutex_lock(&mdsc->mutex);
1084         }
1085         mutex_unlock(&mdsc->mutex);
1086 }
1087
1088 void ceph_mdsc_release_request(struct kref *kref)
1089 {
1090         struct ceph_mds_request *req = container_of(kref,
1091                                                     struct ceph_mds_request,
1092                                                     r_kref);
1093         ceph_mdsc_release_dir_caps_async(req);
1094         destroy_reply_info(&req->r_reply_info);
1095         if (req->r_request)
1096                 ceph_msg_put(req->r_request);
1097         if (req->r_reply)
1098                 ceph_msg_put(req->r_reply);
1099         if (req->r_inode) {
1100                 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
1101                 iput(req->r_inode);
1102         }
1103         if (req->r_parent) {
1104                 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
1105                 iput(req->r_parent);
1106         }
1107         iput(req->r_target_inode);
1108         iput(req->r_new_inode);
1109         if (req->r_dentry)
1110                 dput(req->r_dentry);
1111         if (req->r_old_dentry)
1112                 dput(req->r_old_dentry);
1113         if (req->r_old_dentry_dir) {
1114                 /*
1115                  * track (and drop pins for) r_old_dentry_dir
1116                  * separately, since r_old_dentry's d_parent may have
1117                  * changed between the dir mutex being dropped and
1118                  * this request being freed.
1119                  */
1120                 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
1121                                   CEPH_CAP_PIN);
1122                 iput(req->r_old_dentry_dir);
1123         }
1124         kfree(req->r_path1);
1125         kfree(req->r_path2);
1126         put_cred(req->r_cred);
1127         if (req->r_mnt_idmap)
1128                 mnt_idmap_put(req->r_mnt_idmap);
1129         if (req->r_pagelist)
1130                 ceph_pagelist_release(req->r_pagelist);
1131         kfree(req->r_fscrypt_auth);
1132         kfree(req->r_altname);
1133         put_request_session(req);
1134         ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
1135         WARN_ON_ONCE(!list_empty(&req->r_wait));
1136         kmem_cache_free(ceph_mds_request_cachep, req);
1137 }
1138
1139 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
1140
1141 /*
1142  * lookup session, bump ref if found.
1143  *
1144  * called under mdsc->mutex.
1145  */
1146 static struct ceph_mds_request *
1147 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
1148 {
1149         struct ceph_mds_request *req;
1150
1151         req = lookup_request(&mdsc->request_tree, tid);
1152         if (req)
1153                 ceph_mdsc_get_request(req);
1154
1155         return req;
1156 }
1157
1158 /*
1159  * Register an in-flight request, and assign a tid.  Link to directory
1160  * are modifying (if any).
1161  *
1162  * Called under mdsc->mutex.
1163  */
1164 static void __register_request(struct ceph_mds_client *mdsc,
1165                                struct ceph_mds_request *req,
1166                                struct inode *dir)
1167 {
1168         struct ceph_client *cl = mdsc->fsc->client;
1169         int ret = 0;
1170
1171         req->r_tid = ++mdsc->last_tid;
1172         if (req->r_num_caps) {
1173                 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
1174                                         req->r_num_caps);
1175                 if (ret < 0) {
1176                         pr_err_client(cl, "%p failed to reserve caps: %d\n",
1177                                       req, ret);
1178                         /* set req->r_err to fail early from __do_request */
1179                         req->r_err = ret;
1180                         return;
1181                 }
1182         }
1183         doutc(cl, "%p tid %lld\n", req, req->r_tid);
1184         ceph_mdsc_get_request(req);
1185         insert_request(&mdsc->request_tree, req);
1186
1187         req->r_cred = get_current_cred();
1188         if (!req->r_mnt_idmap)
1189                 req->r_mnt_idmap = &nop_mnt_idmap;
1190
1191         if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
1192                 mdsc->oldest_tid = req->r_tid;
1193
1194         if (dir) {
1195                 struct ceph_inode_info *ci = ceph_inode(dir);
1196
1197                 ihold(dir);
1198                 req->r_unsafe_dir = dir;
1199                 spin_lock(&ci->i_unsafe_lock);
1200                 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
1201                 spin_unlock(&ci->i_unsafe_lock);
1202         }
1203 }
1204
1205 static void __unregister_request(struct ceph_mds_client *mdsc,
1206                                  struct ceph_mds_request *req)
1207 {
1208         doutc(mdsc->fsc->client, "%p tid %lld\n", req, req->r_tid);
1209
1210         /* Never leave an unregistered request on an unsafe list! */
1211         list_del_init(&req->r_unsafe_item);
1212
1213         if (req->r_tid == mdsc->oldest_tid) {
1214                 struct rb_node *p = rb_next(&req->r_node);
1215                 mdsc->oldest_tid = 0;
1216                 while (p) {
1217                         struct ceph_mds_request *next_req =
1218                                 rb_entry(p, struct ceph_mds_request, r_node);
1219                         if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
1220                                 mdsc->oldest_tid = next_req->r_tid;
1221                                 break;
1222                         }
1223                         p = rb_next(p);
1224                 }
1225         }
1226
1227         erase_request(&mdsc->request_tree, req);
1228
1229         if (req->r_unsafe_dir) {
1230                 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
1231                 spin_lock(&ci->i_unsafe_lock);
1232                 list_del_init(&req->r_unsafe_dir_item);
1233                 spin_unlock(&ci->i_unsafe_lock);
1234         }
1235         if (req->r_target_inode &&
1236             test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
1237                 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
1238                 spin_lock(&ci->i_unsafe_lock);
1239                 list_del_init(&req->r_unsafe_target_item);
1240                 spin_unlock(&ci->i_unsafe_lock);
1241         }
1242
1243         if (req->r_unsafe_dir) {
1244                 iput(req->r_unsafe_dir);
1245                 req->r_unsafe_dir = NULL;
1246         }
1247
1248         complete_all(&req->r_safe_completion);
1249
1250         ceph_mdsc_put_request(req);
1251 }
1252
1253 /*
1254  * Walk back up the dentry tree until we hit a dentry representing a
1255  * non-snapshot inode. We do this using the rcu_read_lock (which must be held
1256  * when calling this) to ensure that the objects won't disappear while we're
1257  * working with them. Once we hit a candidate dentry, we attempt to take a
1258  * reference to it, and return that as the result.
1259  */
1260 static struct inode *get_nonsnap_parent(struct dentry *dentry)
1261 {
1262         struct inode *inode = NULL;
1263
1264         while (dentry && !IS_ROOT(dentry)) {
1265                 inode = d_inode_rcu(dentry);
1266                 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
1267                         break;
1268                 dentry = dentry->d_parent;
1269         }
1270         if (inode)
1271                 inode = igrab(inode);
1272         return inode;
1273 }
1274
1275 /*
1276  * Choose mds to send request to next.  If there is a hint set in the
1277  * request (e.g., due to a prior forward hint from the mds), use that.
1278  * Otherwise, consult frag tree and/or caps to identify the
1279  * appropriate mds.  If all else fails, choose randomly.
1280  *
1281  * Called under mdsc->mutex.
1282  */
1283 static int __choose_mds(struct ceph_mds_client *mdsc,
1284                         struct ceph_mds_request *req,
1285                         bool *random)
1286 {
1287         struct inode *inode;
1288         struct ceph_inode_info *ci;
1289         struct ceph_cap *cap;
1290         int mode = req->r_direct_mode;
1291         int mds = -1;
1292         u32 hash = req->r_direct_hash;
1293         bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
1294         struct ceph_client *cl = mdsc->fsc->client;
1295
1296         if (random)
1297                 *random = false;
1298
1299         /*
1300          * is there a specific mds we should try?  ignore hint if we have
1301          * no session and the mds is not up (active or recovering).
1302          */
1303         if (req->r_resend_mds >= 0 &&
1304             (__have_session(mdsc, req->r_resend_mds) ||
1305              ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
1306                 doutc(cl, "using resend_mds mds%d\n", req->r_resend_mds);
1307                 return req->r_resend_mds;
1308         }
1309
1310         if (mode == USE_RANDOM_MDS)
1311                 goto random;
1312
1313         inode = NULL;
1314         if (req->r_inode) {
1315                 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
1316                         inode = req->r_inode;
1317                         ihold(inode);
1318                 } else {
1319                         /* req->r_dentry is non-null for LSSNAP request */
1320                         rcu_read_lock();
1321                         inode = get_nonsnap_parent(req->r_dentry);
1322                         rcu_read_unlock();
1323                         doutc(cl, "using snapdir's parent %p %llx.%llx\n",
1324                               inode, ceph_vinop(inode));
1325                 }
1326         } else if (req->r_dentry) {
1327                 /* ignore race with rename; old or new d_parent is okay */
1328                 struct dentry *parent;
1329                 struct inode *dir;
1330
1331                 rcu_read_lock();
1332                 parent = READ_ONCE(req->r_dentry->d_parent);
1333                 dir = req->r_parent ? : d_inode_rcu(parent);
1334
1335                 if (!dir || dir->i_sb != mdsc->fsc->sb) {
1336                         /*  not this fs or parent went negative */
1337                         inode = d_inode(req->r_dentry);
1338                         if (inode)
1339                                 ihold(inode);
1340                 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
1341                         /* direct snapped/virtual snapdir requests
1342                          * based on parent dir inode */
1343                         inode = get_nonsnap_parent(parent);
1344                         doutc(cl, "using nonsnap parent %p %llx.%llx\n",
1345                               inode, ceph_vinop(inode));
1346                 } else {
1347                         /* dentry target */
1348                         inode = d_inode(req->r_dentry);
1349                         if (!inode || mode == USE_AUTH_MDS) {
1350                                 /* dir + name */
1351                                 inode = igrab(dir);
1352                                 hash = ceph_dentry_hash(dir, req->r_dentry);
1353                                 is_hash = true;
1354                         } else {
1355                                 ihold(inode);
1356                         }
1357                 }
1358                 rcu_read_unlock();
1359         }
1360
1361         if (!inode)
1362                 goto random;
1363
1364         doutc(cl, "%p %llx.%llx is_hash=%d (0x%x) mode %d\n", inode,
1365               ceph_vinop(inode), (int)is_hash, hash, mode);
1366         ci = ceph_inode(inode);
1367
1368         if (is_hash && S_ISDIR(inode->i_mode)) {
1369                 struct ceph_inode_frag frag;
1370                 int found;
1371
1372                 ceph_choose_frag(ci, hash, &frag, &found);
1373                 if (found) {
1374                         if (mode == USE_ANY_MDS && frag.ndist > 0) {
1375                                 u8 r;
1376
1377                                 /* choose a random replica */
1378                                 get_random_bytes(&r, 1);
1379                                 r %= frag.ndist;
1380                                 mds = frag.dist[r];
1381                                 doutc(cl, "%p %llx.%llx frag %u mds%d (%d/%d)\n",
1382                                       inode, ceph_vinop(inode), frag.frag,
1383                                       mds, (int)r, frag.ndist);
1384                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1385                                     CEPH_MDS_STATE_ACTIVE &&
1386                                     !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
1387                                         goto out;
1388                         }
1389
1390                         /* since this file/dir wasn't known to be
1391                          * replicated, then we want to look for the
1392                          * authoritative mds. */
1393                         if (frag.mds >= 0) {
1394                                 /* choose auth mds */
1395                                 mds = frag.mds;
1396                                 doutc(cl, "%p %llx.%llx frag %u mds%d (auth)\n",
1397                                       inode, ceph_vinop(inode), frag.frag, mds);
1398                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1399                                     CEPH_MDS_STATE_ACTIVE) {
1400                                         if (!ceph_mdsmap_is_laggy(mdsc->mdsmap,
1401                                                                   mds))
1402                                                 goto out;
1403                                 }
1404                         }
1405                         mode = USE_AUTH_MDS;
1406                 }
1407         }
1408
1409         spin_lock(&ci->i_ceph_lock);
1410         cap = NULL;
1411         if (mode == USE_AUTH_MDS)
1412                 cap = ci->i_auth_cap;
1413         if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1414                 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1415         if (!cap) {
1416                 spin_unlock(&ci->i_ceph_lock);
1417                 iput(inode);
1418                 goto random;
1419         }
1420         mds = cap->session->s_mds;
1421         doutc(cl, "%p %llx.%llx mds%d (%scap %p)\n", inode,
1422               ceph_vinop(inode), mds,
1423               cap == ci->i_auth_cap ? "auth " : "", cap);
1424         spin_unlock(&ci->i_ceph_lock);
1425 out:
1426         iput(inode);
1427         return mds;
1428
1429 random:
1430         if (random)
1431                 *random = true;
1432
1433         mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
1434         doutc(cl, "chose random mds%d\n", mds);
1435         return mds;
1436 }
1437
1438
1439 /*
1440  * session messages
1441  */
1442 struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq)
1443 {
1444         struct ceph_msg *msg;
1445         struct ceph_mds_session_head *h;
1446
1447         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1448                            false);
1449         if (!msg) {
1450                 pr_err("ENOMEM creating session %s msg\n",
1451                        ceph_session_op_name(op));
1452                 return NULL;
1453         }
1454         h = msg->front.iov_base;
1455         h->op = cpu_to_le32(op);
1456         h->seq = cpu_to_le64(seq);
1457
1458         return msg;
1459 }
1460
1461 static const unsigned char feature_bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1462 #define FEATURE_BYTES(c) (DIV_ROUND_UP((size_t)feature_bits[c - 1] + 1, 64) * 8)
1463 static int encode_supported_features(void **p, void *end)
1464 {
1465         static const size_t count = ARRAY_SIZE(feature_bits);
1466
1467         if (count > 0) {
1468                 size_t i;
1469                 size_t size = FEATURE_BYTES(count);
1470                 unsigned long bit;
1471
1472                 if (WARN_ON_ONCE(*p + 4 + size > end))
1473                         return -ERANGE;
1474
1475                 ceph_encode_32(p, size);
1476                 memset(*p, 0, size);
1477                 for (i = 0; i < count; i++) {
1478                         bit = feature_bits[i];
1479                         ((unsigned char *)(*p))[bit / 8] |= BIT(bit % 8);
1480                 }
1481                 *p += size;
1482         } else {
1483                 if (WARN_ON_ONCE(*p + 4 > end))
1484                         return -ERANGE;
1485
1486                 ceph_encode_32(p, 0);
1487         }
1488
1489         return 0;
1490 }
1491
1492 static const unsigned char metric_bits[] = CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED;
1493 #define METRIC_BYTES(cnt) (DIV_ROUND_UP((size_t)metric_bits[cnt - 1] + 1, 64) * 8)
1494 static int encode_metric_spec(void **p, void *end)
1495 {
1496         static const size_t count = ARRAY_SIZE(metric_bits);
1497
1498         /* header */
1499         if (WARN_ON_ONCE(*p + 2 > end))
1500                 return -ERANGE;
1501
1502         ceph_encode_8(p, 1); /* version */
1503         ceph_encode_8(p, 1); /* compat */
1504
1505         if (count > 0) {
1506                 size_t i;
1507                 size_t size = METRIC_BYTES(count);
1508
1509                 if (WARN_ON_ONCE(*p + 4 + 4 + size > end))
1510                         return -ERANGE;
1511
1512                 /* metric spec info length */
1513                 ceph_encode_32(p, 4 + size);
1514
1515                 /* metric spec */
1516                 ceph_encode_32(p, size);
1517                 memset(*p, 0, size);
1518                 for (i = 0; i < count; i++)
1519                         ((unsigned char *)(*p))[i / 8] |= BIT(metric_bits[i] % 8);
1520                 *p += size;
1521         } else {
1522                 if (WARN_ON_ONCE(*p + 4 + 4 > end))
1523                         return -ERANGE;
1524
1525                 /* metric spec info length */
1526                 ceph_encode_32(p, 4);
1527                 /* metric spec */
1528                 ceph_encode_32(p, 0);
1529         }
1530
1531         return 0;
1532 }
1533
1534 /*
1535  * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1536  * to include additional client metadata fields.
1537  */
1538 static struct ceph_msg *
1539 create_session_full_msg(struct ceph_mds_client *mdsc, int op, u64 seq)
1540 {
1541         struct ceph_msg *msg;
1542         struct ceph_mds_session_head *h;
1543         int i;
1544         int extra_bytes = 0;
1545         int metadata_key_count = 0;
1546         struct ceph_options *opt = mdsc->fsc->client->options;
1547         struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
1548         struct ceph_client *cl = mdsc->fsc->client;
1549         size_t size, count;
1550         void *p, *end;
1551         int ret;
1552
1553         const char* metadata[][2] = {
1554                 {"hostname", mdsc->nodename},
1555                 {"kernel_version", init_utsname()->release},
1556                 {"entity_id", opt->name ? : ""},
1557                 {"root", fsopt->server_path ? : "/"},
1558                 {NULL, NULL}
1559         };
1560
1561         /* Calculate serialized length of metadata */
1562         extra_bytes = 4;  /* map length */
1563         for (i = 0; metadata[i][0]; ++i) {
1564                 extra_bytes += 8 + strlen(metadata[i][0]) +
1565                         strlen(metadata[i][1]);
1566                 metadata_key_count++;
1567         }
1568
1569         /* supported feature */
1570         size = 0;
1571         count = ARRAY_SIZE(feature_bits);
1572         if (count > 0)
1573                 size = FEATURE_BYTES(count);
1574         extra_bytes += 4 + size;
1575
1576         /* metric spec */
1577         size = 0;
1578         count = ARRAY_SIZE(metric_bits);
1579         if (count > 0)
1580                 size = METRIC_BYTES(count);
1581         extra_bytes += 2 + 4 + 4 + size;
1582
1583         /* flags, mds auth caps and oldest_client_tid */
1584         extra_bytes += 4 + 4 + 8;
1585
1586         /* Allocate the message */
1587         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
1588                            GFP_NOFS, false);
1589         if (!msg) {
1590                 pr_err_client(cl, "ENOMEM creating session open msg\n");
1591                 return ERR_PTR(-ENOMEM);
1592         }
1593         p = msg->front.iov_base;
1594         end = p + msg->front.iov_len;
1595
1596         h = p;
1597         h->op = cpu_to_le32(op);
1598         h->seq = cpu_to_le64(seq);
1599
1600         /*
1601          * Serialize client metadata into waiting buffer space, using
1602          * the format that userspace expects for map<string, string>
1603          *
1604          * ClientSession messages with metadata are v7
1605          */
1606         msg->hdr.version = cpu_to_le16(7);
1607         msg->hdr.compat_version = cpu_to_le16(1);
1608
1609         /* The write pointer, following the session_head structure */
1610         p += sizeof(*h);
1611
1612         /* Number of entries in the map */
1613         ceph_encode_32(&p, metadata_key_count);
1614
1615         /* Two length-prefixed strings for each entry in the map */
1616         for (i = 0; metadata[i][0]; ++i) {
1617                 size_t const key_len = strlen(metadata[i][0]);
1618                 size_t const val_len = strlen(metadata[i][1]);
1619
1620                 ceph_encode_32(&p, key_len);
1621                 memcpy(p, metadata[i][0], key_len);
1622                 p += key_len;
1623                 ceph_encode_32(&p, val_len);
1624                 memcpy(p, metadata[i][1], val_len);
1625                 p += val_len;
1626         }
1627
1628         ret = encode_supported_features(&p, end);
1629         if (ret) {
1630                 pr_err_client(cl, "encode_supported_features failed!\n");
1631                 ceph_msg_put(msg);
1632                 return ERR_PTR(ret);
1633         }
1634
1635         ret = encode_metric_spec(&p, end);
1636         if (ret) {
1637                 pr_err_client(cl, "encode_metric_spec failed!\n");
1638                 ceph_msg_put(msg);
1639                 return ERR_PTR(ret);
1640         }
1641
1642         /* version == 5, flags */
1643         ceph_encode_32(&p, 0);
1644
1645         /* version == 6, mds auth caps */
1646         ceph_encode_32(&p, 0);
1647
1648         /* version == 7, oldest_client_tid */
1649         ceph_encode_64(&p, mdsc->oldest_tid);
1650
1651         msg->front.iov_len = p - msg->front.iov_base;
1652         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1653
1654         return msg;
1655 }
1656
1657 /*
1658  * send session open request.
1659  *
1660  * called under mdsc->mutex
1661  */
1662 static int __open_session(struct ceph_mds_client *mdsc,
1663                           struct ceph_mds_session *session)
1664 {
1665         struct ceph_msg *msg;
1666         int mstate;
1667         int mds = session->s_mds;
1668
1669         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO)
1670                 return -EIO;
1671
1672         /* wait for mds to go active? */
1673         mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1674         doutc(mdsc->fsc->client, "open_session to mds%d (%s)\n", mds,
1675               ceph_mds_state_name(mstate));
1676         session->s_state = CEPH_MDS_SESSION_OPENING;
1677         session->s_renew_requested = jiffies;
1678
1679         /* send connect message */
1680         msg = create_session_full_msg(mdsc, CEPH_SESSION_REQUEST_OPEN,
1681                                       session->s_seq);
1682         if (IS_ERR(msg))
1683                 return PTR_ERR(msg);
1684         ceph_con_send(&session->s_con, msg);
1685         return 0;
1686 }
1687
1688 /*
1689  * open sessions for any export targets for the given mds
1690  *
1691  * called under mdsc->mutex
1692  */
1693 static struct ceph_mds_session *
1694 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1695 {
1696         struct ceph_mds_session *session;
1697         int ret;
1698
1699         session = __ceph_lookup_mds_session(mdsc, target);
1700         if (!session) {
1701                 session = register_session(mdsc, target);
1702                 if (IS_ERR(session))
1703                         return session;
1704         }
1705         if (session->s_state == CEPH_MDS_SESSION_NEW ||
1706             session->s_state == CEPH_MDS_SESSION_CLOSING) {
1707                 ret = __open_session(mdsc, session);
1708                 if (ret)
1709                         return ERR_PTR(ret);
1710         }
1711
1712         return session;
1713 }
1714
1715 struct ceph_mds_session *
1716 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1717 {
1718         struct ceph_mds_session *session;
1719         struct ceph_client *cl = mdsc->fsc->client;
1720
1721         doutc(cl, "to mds%d\n", target);
1722
1723         mutex_lock(&mdsc->mutex);
1724         session = __open_export_target_session(mdsc, target);
1725         mutex_unlock(&mdsc->mutex);
1726
1727         return session;
1728 }
1729
1730 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1731                                           struct ceph_mds_session *session)
1732 {
1733         struct ceph_mds_info *mi;
1734         struct ceph_mds_session *ts;
1735         int i, mds = session->s_mds;
1736         struct ceph_client *cl = mdsc->fsc->client;
1737
1738         if (mds >= mdsc->mdsmap->possible_max_rank)
1739                 return;
1740
1741         mi = &mdsc->mdsmap->m_info[mds];
1742         doutc(cl, "for mds%d (%d targets)\n", session->s_mds,
1743               mi->num_export_targets);
1744
1745         for (i = 0; i < mi->num_export_targets; i++) {
1746                 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1747                 ceph_put_mds_session(ts);
1748         }
1749 }
1750
1751 /*
1752  * session caps
1753  */
1754
1755 static void detach_cap_releases(struct ceph_mds_session *session,
1756                                 struct list_head *target)
1757 {
1758         struct ceph_client *cl = session->s_mdsc->fsc->client;
1759
1760         lockdep_assert_held(&session->s_cap_lock);
1761
1762         list_splice_init(&session->s_cap_releases, target);
1763         session->s_num_cap_releases = 0;
1764         doutc(cl, "mds%d\n", session->s_mds);
1765 }
1766
1767 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1768                                  struct list_head *dispose)
1769 {
1770         while (!list_empty(dispose)) {
1771                 struct ceph_cap *cap;
1772                 /* zero out the in-progress message */
1773                 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1774                 list_del(&cap->session_caps);
1775                 ceph_put_cap(mdsc, cap);
1776         }
1777 }
1778
1779 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1780                                      struct ceph_mds_session *session)
1781 {
1782         struct ceph_client *cl = mdsc->fsc->client;
1783         struct ceph_mds_request *req;
1784         struct rb_node *p;
1785
1786         doutc(cl, "mds%d\n", session->s_mds);
1787         mutex_lock(&mdsc->mutex);
1788         while (!list_empty(&session->s_unsafe)) {
1789                 req = list_first_entry(&session->s_unsafe,
1790                                        struct ceph_mds_request, r_unsafe_item);
1791                 pr_warn_ratelimited_client(cl, " dropping unsafe request %llu\n",
1792                                            req->r_tid);
1793                 if (req->r_target_inode)
1794                         mapping_set_error(req->r_target_inode->i_mapping, -EIO);
1795                 if (req->r_unsafe_dir)
1796                         mapping_set_error(req->r_unsafe_dir->i_mapping, -EIO);
1797                 __unregister_request(mdsc, req);
1798         }
1799         /* zero r_attempts, so kick_requests() will re-send requests */
1800         p = rb_first(&mdsc->request_tree);
1801         while (p) {
1802                 req = rb_entry(p, struct ceph_mds_request, r_node);
1803                 p = rb_next(p);
1804                 if (req->r_session &&
1805                     req->r_session->s_mds == session->s_mds)
1806                         req->r_attempts = 0;
1807         }
1808         mutex_unlock(&mdsc->mutex);
1809 }
1810
1811 /*
1812  * Helper to safely iterate over all caps associated with a session, with
1813  * special care taken to handle a racing __ceph_remove_cap().
1814  *
1815  * Caller must hold session s_mutex.
1816  */
1817 int ceph_iterate_session_caps(struct ceph_mds_session *session,
1818                               int (*cb)(struct inode *, int mds, void *),
1819                               void *arg)
1820 {
1821         struct ceph_client *cl = session->s_mdsc->fsc->client;
1822         struct list_head *p;
1823         struct ceph_cap *cap;
1824         struct inode *inode, *last_inode = NULL;
1825         struct ceph_cap *old_cap = NULL;
1826         int ret;
1827
1828         doutc(cl, "%p mds%d\n", session, session->s_mds);
1829         spin_lock(&session->s_cap_lock);
1830         p = session->s_caps.next;
1831         while (p != &session->s_caps) {
1832                 int mds;
1833
1834                 cap = list_entry(p, struct ceph_cap, session_caps);
1835                 inode = igrab(&cap->ci->netfs.inode);
1836                 if (!inode) {
1837                         p = p->next;
1838                         continue;
1839                 }
1840                 session->s_cap_iterator = cap;
1841                 mds = cap->mds;
1842                 spin_unlock(&session->s_cap_lock);
1843
1844                 if (last_inode) {
1845                         iput(last_inode);
1846                         last_inode = NULL;
1847                 }
1848                 if (old_cap) {
1849                         ceph_put_cap(session->s_mdsc, old_cap);
1850                         old_cap = NULL;
1851                 }
1852
1853                 ret = cb(inode, mds, arg);
1854                 last_inode = inode;
1855
1856                 spin_lock(&session->s_cap_lock);
1857                 p = p->next;
1858                 if (!cap->ci) {
1859                         doutc(cl, "finishing cap %p removal\n", cap);
1860                         BUG_ON(cap->session != session);
1861                         cap->session = NULL;
1862                         list_del_init(&cap->session_caps);
1863                         session->s_nr_caps--;
1864                         atomic64_dec(&session->s_mdsc->metric.total_caps);
1865                         if (cap->queue_release)
1866                                 __ceph_queue_cap_release(session, cap);
1867                         else
1868                                 old_cap = cap;  /* put_cap it w/o locks held */
1869                 }
1870                 if (ret < 0)
1871                         goto out;
1872         }
1873         ret = 0;
1874 out:
1875         session->s_cap_iterator = NULL;
1876         spin_unlock(&session->s_cap_lock);
1877
1878         iput(last_inode);
1879         if (old_cap)
1880                 ceph_put_cap(session->s_mdsc, old_cap);
1881
1882         return ret;
1883 }
1884
1885 static int remove_session_caps_cb(struct inode *inode, int mds, void *arg)
1886 {
1887         struct ceph_inode_info *ci = ceph_inode(inode);
1888         struct ceph_client *cl = ceph_inode_to_client(inode);
1889         bool invalidate = false;
1890         struct ceph_cap *cap;
1891         int iputs = 0;
1892
1893         spin_lock(&ci->i_ceph_lock);
1894         cap = __get_cap_for_mds(ci, mds);
1895         if (cap) {
1896                 doutc(cl, " removing cap %p, ci is %p, inode is %p\n",
1897                       cap, ci, &ci->netfs.inode);
1898
1899                 iputs = ceph_purge_inode_cap(inode, cap, &invalidate);
1900         }
1901         spin_unlock(&ci->i_ceph_lock);
1902
1903         if (cap)
1904                 wake_up_all(&ci->i_cap_wq);
1905         if (invalidate)
1906                 ceph_queue_invalidate(inode);
1907         while (iputs--)
1908                 iput(inode);
1909         return 0;
1910 }
1911
1912 /*
1913  * caller must hold session s_mutex
1914  */
1915 static void remove_session_caps(struct ceph_mds_session *session)
1916 {
1917         struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1918         struct super_block *sb = fsc->sb;
1919         LIST_HEAD(dispose);
1920
1921         doutc(fsc->client, "on %p\n", session);
1922         ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
1923
1924         wake_up_all(&fsc->mdsc->cap_flushing_wq);
1925
1926         spin_lock(&session->s_cap_lock);
1927         if (session->s_nr_caps > 0) {
1928                 struct inode *inode;
1929                 struct ceph_cap *cap, *prev = NULL;
1930                 struct ceph_vino vino;
1931                 /*
1932                  * iterate_session_caps() skips inodes that are being
1933                  * deleted, we need to wait until deletions are complete.
1934                  * __wait_on_freeing_inode() is designed for the job,
1935                  * but it is not exported, so use lookup inode function
1936                  * to access it.
1937                  */
1938                 while (!list_empty(&session->s_caps)) {
1939                         cap = list_entry(session->s_caps.next,
1940                                          struct ceph_cap, session_caps);
1941                         if (cap == prev)
1942                                 break;
1943                         prev = cap;
1944                         vino = cap->ci->i_vino;
1945                         spin_unlock(&session->s_cap_lock);
1946
1947                         inode = ceph_find_inode(sb, vino);
1948                         iput(inode);
1949
1950                         spin_lock(&session->s_cap_lock);
1951                 }
1952         }
1953
1954         // drop cap expires and unlock s_cap_lock
1955         detach_cap_releases(session, &dispose);
1956
1957         BUG_ON(session->s_nr_caps > 0);
1958         BUG_ON(!list_empty(&session->s_cap_flushing));
1959         spin_unlock(&session->s_cap_lock);
1960         dispose_cap_releases(session->s_mdsc, &dispose);
1961 }
1962
1963 enum {
1964         RECONNECT,
1965         RENEWCAPS,
1966         FORCE_RO,
1967 };
1968
1969 /*
1970  * wake up any threads waiting on this session's caps.  if the cap is
1971  * old (didn't get renewed on the client reconnect), remove it now.
1972  *
1973  * caller must hold s_mutex.
1974  */
1975 static int wake_up_session_cb(struct inode *inode, int mds, void *arg)
1976 {
1977         struct ceph_inode_info *ci = ceph_inode(inode);
1978         unsigned long ev = (unsigned long)arg;
1979
1980         if (ev == RECONNECT) {
1981                 spin_lock(&ci->i_ceph_lock);
1982                 ci->i_wanted_max_size = 0;
1983                 ci->i_requested_max_size = 0;
1984                 spin_unlock(&ci->i_ceph_lock);
1985         } else if (ev == RENEWCAPS) {
1986                 struct ceph_cap *cap;
1987
1988                 spin_lock(&ci->i_ceph_lock);
1989                 cap = __get_cap_for_mds(ci, mds);
1990                 /* mds did not re-issue stale cap */
1991                 if (cap && cap->cap_gen < atomic_read(&cap->session->s_cap_gen))
1992                         cap->issued = cap->implemented = CEPH_CAP_PIN;
1993                 spin_unlock(&ci->i_ceph_lock);
1994         } else if (ev == FORCE_RO) {
1995         }
1996         wake_up_all(&ci->i_cap_wq);
1997         return 0;
1998 }
1999
2000 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
2001 {
2002         struct ceph_client *cl = session->s_mdsc->fsc->client;
2003
2004         doutc(cl, "session %p mds%d\n", session, session->s_mds);
2005         ceph_iterate_session_caps(session, wake_up_session_cb,
2006                                   (void *)(unsigned long)ev);
2007 }
2008
2009 /*
2010  * Send periodic message to MDS renewing all currently held caps.  The
2011  * ack will reset the expiration for all caps from this session.
2012  *
2013  * caller holds s_mutex
2014  */
2015 static int send_renew_caps(struct ceph_mds_client *mdsc,
2016                            struct ceph_mds_session *session)
2017 {
2018         struct ceph_client *cl = mdsc->fsc->client;
2019         struct ceph_msg *msg;
2020         int state;
2021
2022         if (time_after_eq(jiffies, session->s_cap_ttl) &&
2023             time_after_eq(session->s_cap_ttl, session->s_renew_requested))
2024                 pr_info_client(cl, "mds%d caps stale\n", session->s_mds);
2025         session->s_renew_requested = jiffies;
2026
2027         /* do not try to renew caps until a recovering mds has reconnected
2028          * with its clients. */
2029         state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
2030         if (state < CEPH_MDS_STATE_RECONNECT) {
2031                 doutc(cl, "ignoring mds%d (%s)\n", session->s_mds,
2032                       ceph_mds_state_name(state));
2033                 return 0;
2034         }
2035
2036         doutc(cl, "to mds%d (%s)\n", session->s_mds,
2037               ceph_mds_state_name(state));
2038         msg = create_session_full_msg(mdsc, CEPH_SESSION_REQUEST_RENEWCAPS,
2039                                       ++session->s_renew_seq);
2040         if (IS_ERR(msg))
2041                 return PTR_ERR(msg);
2042         ceph_con_send(&session->s_con, msg);
2043         return 0;
2044 }
2045
2046 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
2047                              struct ceph_mds_session *session, u64 seq)
2048 {
2049         struct ceph_client *cl = mdsc->fsc->client;
2050         struct ceph_msg *msg;
2051
2052         doutc(cl, "to mds%d (%s)s seq %lld\n", session->s_mds,
2053               ceph_session_state_name(session->s_state), seq);
2054         msg = ceph_create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
2055         if (!msg)
2056                 return -ENOMEM;
2057         ceph_con_send(&session->s_con, msg);
2058         return 0;
2059 }
2060
2061
2062 /*
2063  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
2064  *
2065  * Called under session->s_mutex
2066  */
2067 static void renewed_caps(struct ceph_mds_client *mdsc,
2068                          struct ceph_mds_session *session, int is_renew)
2069 {
2070         struct ceph_client *cl = mdsc->fsc->client;
2071         int was_stale;
2072         int wake = 0;
2073
2074         spin_lock(&session->s_cap_lock);
2075         was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2076
2077         session->s_cap_ttl = session->s_renew_requested +
2078                 mdsc->mdsmap->m_session_timeout*HZ;
2079
2080         if (was_stale) {
2081                 if (time_before(jiffies, session->s_cap_ttl)) {
2082                         pr_info_client(cl, "mds%d caps renewed\n",
2083                                        session->s_mds);
2084                         wake = 1;
2085                 } else {
2086                         pr_info_client(cl, "mds%d caps still stale\n",
2087                                        session->s_mds);
2088                 }
2089         }
2090         doutc(cl, "mds%d ttl now %lu, was %s, now %s\n", session->s_mds,
2091               session->s_cap_ttl, was_stale ? "stale" : "fresh",
2092               time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
2093         spin_unlock(&session->s_cap_lock);
2094
2095         if (wake)
2096                 wake_up_session_caps(session, RENEWCAPS);
2097 }
2098
2099 /*
2100  * send a session close request
2101  */
2102 static int request_close_session(struct ceph_mds_session *session)
2103 {
2104         struct ceph_client *cl = session->s_mdsc->fsc->client;
2105         struct ceph_msg *msg;
2106
2107         doutc(cl, "mds%d state %s seq %lld\n", session->s_mds,
2108               ceph_session_state_name(session->s_state), session->s_seq);
2109         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_CLOSE,
2110                                       session->s_seq);
2111         if (!msg)
2112                 return -ENOMEM;
2113         ceph_con_send(&session->s_con, msg);
2114         return 1;
2115 }
2116
2117 /*
2118  * Called with s_mutex held.
2119  */
2120 static int __close_session(struct ceph_mds_client *mdsc,
2121                          struct ceph_mds_session *session)
2122 {
2123         if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
2124                 return 0;
2125         session->s_state = CEPH_MDS_SESSION_CLOSING;
2126         return request_close_session(session);
2127 }
2128
2129 static bool drop_negative_children(struct dentry *dentry)
2130 {
2131         struct dentry *child;
2132         bool all_negative = true;
2133
2134         if (!d_is_dir(dentry))
2135                 goto out;
2136
2137         spin_lock(&dentry->d_lock);
2138         hlist_for_each_entry(child, &dentry->d_children, d_sib) {
2139                 if (d_really_is_positive(child)) {
2140                         all_negative = false;
2141                         break;
2142                 }
2143         }
2144         spin_unlock(&dentry->d_lock);
2145
2146         if (all_negative)
2147                 shrink_dcache_parent(dentry);
2148 out:
2149         return all_negative;
2150 }
2151
2152 /*
2153  * Trim old(er) caps.
2154  *
2155  * Because we can't cache an inode without one or more caps, we do
2156  * this indirectly: if a cap is unused, we prune its aliases, at which
2157  * point the inode will hopefully get dropped to.
2158  *
2159  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
2160  * memory pressure from the MDS, though, so it needn't be perfect.
2161  */
2162 static int trim_caps_cb(struct inode *inode, int mds, void *arg)
2163 {
2164         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
2165         struct ceph_client *cl = mdsc->fsc->client;
2166         int *remaining = arg;
2167         struct ceph_inode_info *ci = ceph_inode(inode);
2168         int used, wanted, oissued, mine;
2169         struct ceph_cap *cap;
2170
2171         if (*remaining <= 0)
2172                 return -1;
2173
2174         spin_lock(&ci->i_ceph_lock);
2175         cap = __get_cap_for_mds(ci, mds);
2176         if (!cap) {
2177                 spin_unlock(&ci->i_ceph_lock);
2178                 return 0;
2179         }
2180         mine = cap->issued | cap->implemented;
2181         used = __ceph_caps_used(ci);
2182         wanted = __ceph_caps_file_wanted(ci);
2183         oissued = __ceph_caps_issued_other(ci, cap);
2184
2185         doutc(cl, "%p %llx.%llx cap %p mine %s oissued %s used %s wanted %s\n",
2186               inode, ceph_vinop(inode), cap, ceph_cap_string(mine),
2187               ceph_cap_string(oissued), ceph_cap_string(used),
2188               ceph_cap_string(wanted));
2189         if (cap == ci->i_auth_cap) {
2190                 if (ci->i_dirty_caps || ci->i_flushing_caps ||
2191                     !list_empty(&ci->i_cap_snaps))
2192                         goto out;
2193                 if ((used | wanted) & CEPH_CAP_ANY_WR)
2194                         goto out;
2195                 /* Note: it's possible that i_filelock_ref becomes non-zero
2196                  * after dropping auth caps. It doesn't hurt because reply
2197                  * of lock mds request will re-add auth caps. */
2198                 if (atomic_read(&ci->i_filelock_ref) > 0)
2199                         goto out;
2200         }
2201         /* The inode has cached pages, but it's no longer used.
2202          * we can safely drop it */
2203         if (S_ISREG(inode->i_mode) &&
2204             wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
2205             !(oissued & CEPH_CAP_FILE_CACHE)) {
2206           used = 0;
2207           oissued = 0;
2208         }
2209         if ((used | wanted) & ~oissued & mine)
2210                 goto out;   /* we need these caps */
2211
2212         if (oissued) {
2213                 /* we aren't the only cap.. just remove us */
2214                 ceph_remove_cap(mdsc, cap, true);
2215                 (*remaining)--;
2216         } else {
2217                 struct dentry *dentry;
2218                 /* try dropping referring dentries */
2219                 spin_unlock(&ci->i_ceph_lock);
2220                 dentry = d_find_any_alias(inode);
2221                 if (dentry && drop_negative_children(dentry)) {
2222                         int count;
2223                         dput(dentry);
2224                         d_prune_aliases(inode);
2225                         count = icount_read(inode);
2226                         if (count == 1)
2227                                 (*remaining)--;
2228                         doutc(cl, "%p %llx.%llx cap %p pruned, count now %d\n",
2229                               inode, ceph_vinop(inode), cap, count);
2230                 } else {
2231                         dput(dentry);
2232                 }
2233                 return 0;
2234         }
2235
2236 out:
2237         spin_unlock(&ci->i_ceph_lock);
2238         return 0;
2239 }
2240
2241 /*
2242  * Trim session cap count down to some max number.
2243  */
2244 int ceph_trim_caps(struct ceph_mds_client *mdsc,
2245                    struct ceph_mds_session *session,
2246                    int max_caps)
2247 {
2248         struct ceph_client *cl = mdsc->fsc->client;
2249         int trim_caps = session->s_nr_caps - max_caps;
2250
2251         doutc(cl, "mds%d start: %d / %d, trim %d\n", session->s_mds,
2252               session->s_nr_caps, max_caps, trim_caps);
2253         if (trim_caps > 0) {
2254                 int remaining = trim_caps;
2255
2256                 ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
2257                 doutc(cl, "mds%d done: %d / %d, trimmed %d\n",
2258                       session->s_mds, session->s_nr_caps, max_caps,
2259                       trim_caps - remaining);
2260         }
2261
2262         ceph_flush_session_cap_releases(mdsc, session);
2263         return 0;
2264 }
2265
2266 static int check_caps_flush(struct ceph_mds_client *mdsc,
2267                             u64 want_flush_tid)
2268 {
2269         struct ceph_client *cl = mdsc->fsc->client;
2270         int ret = 1;
2271
2272         spin_lock(&mdsc->cap_dirty_lock);
2273         if (!list_empty(&mdsc->cap_flush_list)) {
2274                 struct ceph_cap_flush *cf =
2275                         list_first_entry(&mdsc->cap_flush_list,
2276                                          struct ceph_cap_flush, g_list);
2277                 if (cf->tid <= want_flush_tid) {
2278                         doutc(cl, "still flushing tid %llu <= %llu\n",
2279                               cf->tid, want_flush_tid);
2280                         ret = 0;
2281                 }
2282         }
2283         spin_unlock(&mdsc->cap_dirty_lock);
2284         return ret;
2285 }
2286
2287 /*
2288  * flush all dirty inode data to disk.
2289  *
2290  * returns true if we've flushed through want_flush_tid
2291  */
2292 static void wait_caps_flush(struct ceph_mds_client *mdsc,
2293                             u64 want_flush_tid)
2294 {
2295         struct ceph_client *cl = mdsc->fsc->client;
2296
2297         doutc(cl, "want %llu\n", want_flush_tid);
2298
2299         wait_event(mdsc->cap_flushing_wq,
2300                    check_caps_flush(mdsc, want_flush_tid));
2301
2302         doutc(cl, "ok, flushed thru %llu\n", want_flush_tid);
2303 }
2304
2305 /*
2306  * called under s_mutex
2307  */
2308 static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
2309                                    struct ceph_mds_session *session)
2310 {
2311         struct ceph_client *cl = mdsc->fsc->client;
2312         struct ceph_msg *msg = NULL;
2313         struct ceph_mds_cap_release *head;
2314         struct ceph_mds_cap_item *item;
2315         struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
2316         struct ceph_cap *cap;
2317         LIST_HEAD(tmp_list);
2318         int num_cap_releases;
2319         __le32  barrier, *cap_barrier;
2320
2321         down_read(&osdc->lock);
2322         barrier = cpu_to_le32(osdc->epoch_barrier);
2323         up_read(&osdc->lock);
2324
2325         spin_lock(&session->s_cap_lock);
2326 again:
2327         list_splice_init(&session->s_cap_releases, &tmp_list);
2328         num_cap_releases = session->s_num_cap_releases;
2329         session->s_num_cap_releases = 0;
2330         spin_unlock(&session->s_cap_lock);
2331
2332         while (!list_empty(&tmp_list)) {
2333                 if (!msg) {
2334                         msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
2335                                         PAGE_SIZE, GFP_NOFS, false);
2336                         if (!msg)
2337                                 goto out_err;
2338                         head = msg->front.iov_base;
2339                         head->num = cpu_to_le32(0);
2340                         msg->front.iov_len = sizeof(*head);
2341
2342                         msg->hdr.version = cpu_to_le16(2);
2343                         msg->hdr.compat_version = cpu_to_le16(1);
2344                 }
2345
2346                 cap = list_first_entry(&tmp_list, struct ceph_cap,
2347                                         session_caps);
2348                 list_del(&cap->session_caps);
2349                 num_cap_releases--;
2350
2351                 head = msg->front.iov_base;
2352                 put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
2353                                    &head->num);
2354                 item = msg->front.iov_base + msg->front.iov_len;
2355                 item->ino = cpu_to_le64(cap->cap_ino);
2356                 item->cap_id = cpu_to_le64(cap->cap_id);
2357                 item->migrate_seq = cpu_to_le32(cap->mseq);
2358                 item->issue_seq = cpu_to_le32(cap->issue_seq);
2359                 msg->front.iov_len += sizeof(*item);
2360
2361                 ceph_put_cap(mdsc, cap);
2362
2363                 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
2364                         // Append cap_barrier field
2365                         cap_barrier = msg->front.iov_base + msg->front.iov_len;
2366                         *cap_barrier = barrier;
2367                         msg->front.iov_len += sizeof(*cap_barrier);
2368
2369                         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2370                         doutc(cl, "mds%d %p\n", session->s_mds, msg);
2371                         ceph_con_send(&session->s_con, msg);
2372                         msg = NULL;
2373                 }
2374         }
2375
2376         BUG_ON(num_cap_releases != 0);
2377
2378         spin_lock(&session->s_cap_lock);
2379         if (!list_empty(&session->s_cap_releases))
2380                 goto again;
2381         spin_unlock(&session->s_cap_lock);
2382
2383         if (msg) {
2384                 // Append cap_barrier field
2385                 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2386                 *cap_barrier = barrier;
2387                 msg->front.iov_len += sizeof(*cap_barrier);
2388
2389                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2390                 doutc(cl, "mds%d %p\n", session->s_mds, msg);
2391                 ceph_con_send(&session->s_con, msg);
2392         }
2393         return;
2394 out_err:
2395         pr_err_client(cl, "mds%d, failed to allocate message\n",
2396                       session->s_mds);
2397         spin_lock(&session->s_cap_lock);
2398         list_splice(&tmp_list, &session->s_cap_releases);
2399         session->s_num_cap_releases += num_cap_releases;
2400         spin_unlock(&session->s_cap_lock);
2401 }
2402
2403 static void ceph_cap_release_work(struct work_struct *work)
2404 {
2405         struct ceph_mds_session *session =
2406                 container_of(work, struct ceph_mds_session, s_cap_release_work);
2407
2408         mutex_lock(&session->s_mutex);
2409         if (session->s_state == CEPH_MDS_SESSION_OPEN ||
2410             session->s_state == CEPH_MDS_SESSION_HUNG)
2411                 ceph_send_cap_releases(session->s_mdsc, session);
2412         mutex_unlock(&session->s_mutex);
2413         ceph_put_mds_session(session);
2414 }
2415
2416 void ceph_flush_session_cap_releases(struct ceph_mds_client *mdsc,
2417                              struct ceph_mds_session *session)
2418 {
2419         struct ceph_client *cl = mdsc->fsc->client;
2420         if (mdsc->stopping)
2421                 return;
2422
2423         ceph_get_mds_session(session);
2424         if (queue_work(mdsc->fsc->cap_wq,
2425                        &session->s_cap_release_work)) {
2426                 doutc(cl, "cap release work queued\n");
2427         } else {
2428                 ceph_put_mds_session(session);
2429                 doutc(cl, "failed to queue cap release work\n");
2430         }
2431 }
2432
2433 /*
2434  * caller holds session->s_cap_lock
2435  */
2436 void __ceph_queue_cap_release(struct ceph_mds_session *session,
2437                               struct ceph_cap *cap)
2438 {
2439         list_add_tail(&cap->session_caps, &session->s_cap_releases);
2440         session->s_num_cap_releases++;
2441
2442         if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
2443                 ceph_flush_session_cap_releases(session->s_mdsc, session);
2444 }
2445
2446 static void ceph_cap_reclaim_work(struct work_struct *work)
2447 {
2448         struct ceph_mds_client *mdsc =
2449                 container_of(work, struct ceph_mds_client, cap_reclaim_work);
2450         int ret = ceph_trim_dentries(mdsc);
2451         if (ret == -EAGAIN)
2452                 ceph_queue_cap_reclaim_work(mdsc);
2453 }
2454
2455 void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2456 {
2457         struct ceph_client *cl = mdsc->fsc->client;
2458         if (mdsc->stopping)
2459                 return;
2460
2461         if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
2462                 doutc(cl, "caps reclaim work queued\n");
2463         } else {
2464                 doutc(cl, "failed to queue caps release work\n");
2465         }
2466 }
2467
2468 void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2469 {
2470         int val;
2471         if (!nr)
2472                 return;
2473         val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
2474         if ((val % CEPH_CAPS_PER_RELEASE) < nr) {
2475                 atomic_set(&mdsc->cap_reclaim_pending, 0);
2476                 ceph_queue_cap_reclaim_work(mdsc);
2477         }
2478 }
2479
2480 void ceph_queue_cap_unlink_work(struct ceph_mds_client *mdsc)
2481 {
2482         struct ceph_client *cl = mdsc->fsc->client;
2483         if (mdsc->stopping)
2484                 return;
2485
2486         if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_unlink_work)) {
2487                 doutc(cl, "caps unlink work queued\n");
2488         } else {
2489                 doutc(cl, "failed to queue caps unlink work\n");
2490         }
2491 }
2492
2493 static void ceph_cap_unlink_work(struct work_struct *work)
2494 {
2495         struct ceph_mds_client *mdsc =
2496                 container_of(work, struct ceph_mds_client, cap_unlink_work);
2497         struct ceph_client *cl = mdsc->fsc->client;
2498
2499         doutc(cl, "begin\n");
2500         spin_lock(&mdsc->cap_delay_lock);
2501         while (!list_empty(&mdsc->cap_unlink_delay_list)) {
2502                 struct ceph_inode_info *ci;
2503                 struct inode *inode;
2504
2505                 ci = list_first_entry(&mdsc->cap_unlink_delay_list,
2506                                       struct ceph_inode_info,
2507                                       i_cap_delay_list);
2508                 list_del_init(&ci->i_cap_delay_list);
2509
2510                 inode = igrab(&ci->netfs.inode);
2511                 if (inode) {
2512                         spin_unlock(&mdsc->cap_delay_lock);
2513                         doutc(cl, "on %p %llx.%llx\n", inode,
2514                               ceph_vinop(inode));
2515                         ceph_check_caps(ci, CHECK_CAPS_FLUSH);
2516                         iput(inode);
2517                         spin_lock(&mdsc->cap_delay_lock);
2518                 }
2519         }
2520         spin_unlock(&mdsc->cap_delay_lock);
2521         doutc(cl, "done\n");
2522 }
2523
2524 /*
2525  * requests
2526  */
2527
2528 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2529                                     struct inode *dir)
2530 {
2531         struct ceph_inode_info *ci = ceph_inode(dir);
2532         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2533         struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2534         size_t size = sizeof(struct ceph_mds_reply_dir_entry);
2535         unsigned int num_entries;
2536         u64 bytes_count;
2537         int order;
2538
2539         spin_lock(&ci->i_ceph_lock);
2540         num_entries = ci->i_files + ci->i_subdirs;
2541         spin_unlock(&ci->i_ceph_lock);
2542         num_entries = max(num_entries, 1U);
2543         num_entries = min(num_entries, opt->max_readdir);
2544
2545         bytes_count = (u64)size * num_entries;
2546         if (unlikely(bytes_count > ULONG_MAX))
2547                 bytes_count = ULONG_MAX;
2548
2549         order = get_order((unsigned long)bytes_count);
2550         while (order >= 0) {
2551                 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2552                                                              __GFP_NOWARN |
2553                                                              __GFP_ZERO,
2554                                                              order);
2555                 if (rinfo->dir_entries)
2556                         break;
2557                 order--;
2558         }
2559         if (!rinfo->dir_entries || unlikely(order < 0))
2560                 return -ENOMEM;
2561
2562         num_entries = (PAGE_SIZE << order) / size;
2563         num_entries = min(num_entries, opt->max_readdir);
2564
2565         rinfo->dir_buf_size = PAGE_SIZE << order;
2566         req->r_num_caps = num_entries + 1;
2567         req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2568         req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2569         return 0;
2570 }
2571
2572 /*
2573  * Create an mds request.
2574  */
2575 struct ceph_mds_request *
2576 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2577 {
2578         struct ceph_mds_request *req;
2579
2580         req = kmem_cache_zalloc(ceph_mds_request_cachep, GFP_NOFS);
2581         if (!req)
2582                 return ERR_PTR(-ENOMEM);
2583
2584         mutex_init(&req->r_fill_mutex);
2585         req->r_mdsc = mdsc;
2586         req->r_started = jiffies;
2587         req->r_start_latency = ktime_get();
2588         req->r_resend_mds = -1;
2589         INIT_LIST_HEAD(&req->r_unsafe_dir_item);
2590         INIT_LIST_HEAD(&req->r_unsafe_target_item);
2591         req->r_fmode = -1;
2592         req->r_feature_needed = -1;
2593         kref_init(&req->r_kref);
2594         RB_CLEAR_NODE(&req->r_node);
2595         INIT_LIST_HEAD(&req->r_wait);
2596         init_completion(&req->r_completion);
2597         init_completion(&req->r_safe_completion);
2598         INIT_LIST_HEAD(&req->r_unsafe_item);
2599
2600         ktime_get_coarse_real_ts64(&req->r_stamp);
2601
2602         req->r_op = op;
2603         req->r_direct_mode = mode;
2604         return req;
2605 }
2606
2607 /*
2608  * return oldest (lowest) request, tid in request tree, 0 if none.
2609  *
2610  * called under mdsc->mutex.
2611  */
2612 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2613 {
2614         if (RB_EMPTY_ROOT(&mdsc->request_tree))
2615                 return NULL;
2616         return rb_entry(rb_first(&mdsc->request_tree),
2617                         struct ceph_mds_request, r_node);
2618 }
2619
2620 static inline  u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2621 {
2622         return mdsc->oldest_tid;
2623 }
2624
2625 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
2626 static u8 *get_fscrypt_altname(const struct ceph_mds_request *req, u32 *plen)
2627 {
2628         struct inode *dir = req->r_parent;
2629         struct dentry *dentry = req->r_dentry;
2630         const struct qstr *name = req->r_dname;
2631         u8 *cryptbuf = NULL;
2632         u32 len = 0;
2633         int ret = 0;
2634
2635         /* only encode if we have parent and dentry */
2636         if (!dir || !dentry)
2637                 goto success;
2638
2639         /* No-op unless this is encrypted */
2640         if (!IS_ENCRYPTED(dir))
2641                 goto success;
2642
2643         ret = ceph_fscrypt_prepare_readdir(dir);
2644         if (ret < 0)
2645                 return ERR_PTR(ret);
2646
2647         /* No key? Just ignore it. */
2648         if (!fscrypt_has_encryption_key(dir))
2649                 goto success;
2650
2651         if (!name)
2652                 name = &dentry->d_name;
2653
2654         if (!fscrypt_fname_encrypted_size(dir, name->len, NAME_MAX, &len)) {
2655                 WARN_ON_ONCE(1);
2656                 return ERR_PTR(-ENAMETOOLONG);
2657         }
2658
2659         /* No need to append altname if name is short enough */
2660         if (len <= CEPH_NOHASH_NAME_MAX) {
2661                 len = 0;
2662                 goto success;
2663         }
2664
2665         cryptbuf = kmalloc(len, GFP_KERNEL);
2666         if (!cryptbuf)
2667                 return ERR_PTR(-ENOMEM);
2668
2669         ret = fscrypt_fname_encrypt(dir, name, cryptbuf, len);
2670         if (ret) {
2671                 kfree(cryptbuf);
2672                 return ERR_PTR(ret);
2673         }
2674 success:
2675         *plen = len;
2676         return cryptbuf;
2677 }
2678 #else
2679 static u8 *get_fscrypt_altname(const struct ceph_mds_request *req, u32 *plen)
2680 {
2681         *plen = 0;
2682         return NULL;
2683 }
2684 #endif
2685
2686 /**
2687  * ceph_mdsc_build_path - build a path string to a given dentry
2688  * @mdsc: mds client
2689  * @dentry: dentry to which path should be built
2690  * @path_info: output path, length, base ino+snap, and freepath ownership flag
2691  * @for_wire: is this path going to be sent to the MDS?
2692  *
2693  * Build a string that represents the path to the dentry. This is mostly called
2694  * for two different purposes:
2695  *
2696  * 1) we need to build a path string to send to the MDS (for_wire == true)
2697  * 2) we need a path string for local presentation (e.g. debugfs)
2698  *    (for_wire == false)
2699  *
2700  * The path is built in reverse, starting with the dentry. Walk back up toward
2701  * the root, building the path until the first non-snapped inode is reached
2702  * (for_wire) or the root inode is reached (!for_wire).
2703  *
2704  * Encode hidden .snap dirs as a double /, i.e.
2705  *   foo/.snap/bar -> foo//bar
2706  */
2707 char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc, struct dentry *dentry,
2708                            struct ceph_path_info *path_info, int for_wire)
2709 {
2710         struct ceph_client *cl = mdsc->fsc->client;
2711         struct dentry *cur;
2712         struct inode *inode;
2713         char *path;
2714         int pos;
2715         unsigned seq;
2716         u64 base;
2717
2718         if (!dentry)
2719                 return ERR_PTR(-EINVAL);
2720
2721         path = __getname();
2722         if (!path)
2723                 return ERR_PTR(-ENOMEM);
2724 retry:
2725         pos = PATH_MAX - 1;
2726         path[pos] = '\0';
2727
2728         seq = read_seqbegin(&rename_lock);
2729         cur = dget(dentry);
2730         for (;;) {
2731                 struct dentry *parent;
2732
2733                 spin_lock(&cur->d_lock);
2734                 inode = d_inode(cur);
2735                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
2736                         doutc(cl, "path+%d: %p SNAPDIR\n", pos, cur);
2737                         spin_unlock(&cur->d_lock);
2738                         parent = dget_parent(cur);
2739                 } else if (for_wire && inode && dentry != cur &&
2740                            ceph_snap(inode) == CEPH_NOSNAP) {
2741                         spin_unlock(&cur->d_lock);
2742                         pos++; /* get rid of any prepended '/' */
2743                         break;
2744                 } else if (!for_wire || !IS_ENCRYPTED(d_inode(cur->d_parent))) {
2745                         pos -= cur->d_name.len;
2746                         if (pos < 0) {
2747                                 spin_unlock(&cur->d_lock);
2748                                 break;
2749                         }
2750                         memcpy(path + pos, cur->d_name.name, cur->d_name.len);
2751                         spin_unlock(&cur->d_lock);
2752                         parent = dget_parent(cur);
2753                 } else {
2754                         int len, ret;
2755                         char buf[NAME_MAX];
2756
2757                         /*
2758                          * Proactively copy name into buf, in case we need to
2759                          * present it as-is.
2760                          */
2761                         memcpy(buf, cur->d_name.name, cur->d_name.len);
2762                         len = cur->d_name.len;
2763                         spin_unlock(&cur->d_lock);
2764                         parent = dget_parent(cur);
2765
2766                         ret = ceph_fscrypt_prepare_readdir(d_inode(parent));
2767                         if (ret < 0) {
2768                                 dput(parent);
2769                                 dput(cur);
2770                                 return ERR_PTR(ret);
2771                         }
2772
2773                         if (fscrypt_has_encryption_key(d_inode(parent))) {
2774                                 len = ceph_encode_encrypted_dname(d_inode(parent),
2775                                                                   buf, len);
2776                                 if (len < 0) {
2777                                         dput(parent);
2778                                         dput(cur);
2779                                         return ERR_PTR(len);
2780                                 }
2781                         }
2782                         pos -= len;
2783                         if (pos < 0) {
2784                                 dput(parent);
2785                                 break;
2786                         }
2787                         memcpy(path + pos, buf, len);
2788                 }
2789                 dput(cur);
2790                 cur = parent;
2791
2792                 /* Are we at the root? */
2793                 if (IS_ROOT(cur))
2794                         break;
2795
2796                 /* Are we out of buffer? */
2797                 if (--pos < 0)
2798                         break;
2799
2800                 path[pos] = '/';
2801         }
2802         inode = d_inode(cur);
2803         base = inode ? ceph_ino(inode) : 0;
2804         dput(cur);
2805
2806         if (read_seqretry(&rename_lock, seq))
2807                 goto retry;
2808
2809         if (pos < 0) {
2810                 /*
2811                  * The path is longer than PATH_MAX and this function
2812                  * cannot ever succeed.  Creating paths that long is
2813                  * possible with Ceph, but Linux cannot use them.
2814                  */
2815                 return ERR_PTR(-ENAMETOOLONG);
2816         }
2817
2818         /* Initialize the output structure */
2819         memset(path_info, 0, sizeof(*path_info));
2820
2821         path_info->vino.ino = base;
2822         path_info->pathlen = PATH_MAX - 1 - pos;
2823         path_info->path = path + pos;
2824         path_info->freepath = true;
2825
2826         /* Set snap from dentry if available */
2827         if (d_inode(dentry))
2828                 path_info->vino.snap = ceph_snap(d_inode(dentry));
2829         else
2830                 path_info->vino.snap = CEPH_NOSNAP;
2831
2832         doutc(cl, "on %p %d built %llx '%.*s'\n", dentry, d_count(dentry),
2833               base, PATH_MAX - 1 - pos, path + pos);
2834         return path + pos;
2835 }
2836
2837 static int build_dentry_path(struct ceph_mds_client *mdsc, struct dentry *dentry,
2838                              struct inode *dir, struct ceph_path_info *path_info,
2839                              bool parent_locked)
2840 {
2841         char *path;
2842
2843         rcu_read_lock();
2844         if (!dir)
2845                 dir = d_inode_rcu(dentry->d_parent);
2846         if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP &&
2847             !IS_ENCRYPTED(dir)) {
2848                 path_info->vino.ino = ceph_ino(dir);
2849                 path_info->vino.snap = ceph_snap(dir);
2850                 rcu_read_unlock();
2851                 path_info->path = dentry->d_name.name;
2852                 path_info->pathlen = dentry->d_name.len;
2853                 path_info->freepath = false;
2854                 return 0;
2855         }
2856         rcu_read_unlock();
2857         path = ceph_mdsc_build_path(mdsc, dentry, path_info, 1);
2858         if (IS_ERR(path))
2859                 return PTR_ERR(path);
2860         /*
2861          * ceph_mdsc_build_path already fills path_info, including snap handling.
2862          */
2863         return 0;
2864 }
2865
2866 static int build_inode_path(struct inode *inode, struct ceph_path_info *path_info)
2867 {
2868         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
2869         struct dentry *dentry;
2870         char *path;
2871
2872         if (ceph_snap(inode) == CEPH_NOSNAP) {
2873                 path_info->vino.ino = ceph_ino(inode);
2874                 path_info->vino.snap = ceph_snap(inode);
2875                 path_info->pathlen = 0;
2876                 path_info->freepath = false;
2877                 return 0;
2878         }
2879         dentry = d_find_alias(inode);
2880         path = ceph_mdsc_build_path(mdsc, dentry, path_info, 1);
2881         dput(dentry);
2882         if (IS_ERR(path))
2883                 return PTR_ERR(path);
2884         /*
2885          * ceph_mdsc_build_path already fills path_info, including snap from dentry.
2886          * Override with inode's snap since that's what this function is for.
2887          */
2888         path_info->vino.snap = ceph_snap(inode);
2889         return 0;
2890 }
2891
2892 /*
2893  * request arguments may be specified via an inode *, a dentry *, or
2894  * an explicit ino+path.
2895  */
2896 static int set_request_path_attr(struct ceph_mds_client *mdsc, struct inode *rinode,
2897                                  struct dentry *rdentry, struct inode *rdiri,
2898                                  const char *rpath, u64 rino,
2899                                  struct ceph_path_info *path_info,
2900                                  bool parent_locked)
2901 {
2902         struct ceph_client *cl = mdsc->fsc->client;
2903         int r = 0;
2904
2905         /* Initialize the output structure */
2906         memset(path_info, 0, sizeof(*path_info));
2907
2908         if (rinode) {
2909                 r = build_inode_path(rinode, path_info);
2910                 doutc(cl, " inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2911                       ceph_snap(rinode));
2912         } else if (rdentry) {
2913                 r = build_dentry_path(mdsc, rdentry, rdiri, path_info, parent_locked);
2914                 doutc(cl, " dentry %p %llx/%.*s\n", rdentry, path_info->vino.ino,
2915                       path_info->pathlen, path_info->path);
2916         } else if (rpath || rino) {
2917                 path_info->vino.ino = rino;
2918                 path_info->vino.snap = CEPH_NOSNAP;
2919                 path_info->path = rpath;
2920                 path_info->pathlen = rpath ? strlen(rpath) : 0;
2921                 path_info->freepath = false;
2922
2923                 doutc(cl, " path %.*s\n", path_info->pathlen, rpath);
2924         }
2925
2926         return r;
2927 }
2928
2929 static void encode_mclientrequest_tail(void **p,
2930                                        const struct ceph_mds_request *req)
2931 {
2932         struct ceph_timespec ts;
2933         int i;
2934
2935         ceph_encode_timespec64(&ts, &req->r_stamp);
2936         ceph_encode_copy(p, &ts, sizeof(ts));
2937
2938         /* v4: gid_list */
2939         ceph_encode_32(p, req->r_cred->group_info->ngroups);
2940         for (i = 0; i < req->r_cred->group_info->ngroups; i++)
2941                 ceph_encode_64(p, from_kgid(&init_user_ns,
2942                                             req->r_cred->group_info->gid[i]));
2943
2944         /* v5: altname */
2945         ceph_encode_32(p, req->r_altname_len);
2946         ceph_encode_copy(p, req->r_altname, req->r_altname_len);
2947
2948         /* v6: fscrypt_auth and fscrypt_file */
2949         if (req->r_fscrypt_auth) {
2950                 u32 authlen = ceph_fscrypt_auth_len(req->r_fscrypt_auth);
2951
2952                 ceph_encode_32(p, authlen);
2953                 ceph_encode_copy(p, req->r_fscrypt_auth, authlen);
2954         } else {
2955                 ceph_encode_32(p, 0);
2956         }
2957         if (test_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags)) {
2958                 ceph_encode_32(p, sizeof(__le64));
2959                 ceph_encode_64(p, req->r_fscrypt_file);
2960         } else {
2961                 ceph_encode_32(p, 0);
2962         }
2963 }
2964
2965 static inline u16 mds_supported_head_version(struct ceph_mds_session *session)
2966 {
2967         if (!test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD, &session->s_features))
2968                 return 1;
2969
2970         if (!test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features))
2971                 return 2;
2972
2973         return CEPH_MDS_REQUEST_HEAD_VERSION;
2974 }
2975
2976 static struct ceph_mds_request_head_legacy *
2977 find_legacy_request_head(void *p, u64 features)
2978 {
2979         bool legacy = !(features & CEPH_FEATURE_FS_BTIME);
2980         struct ceph_mds_request_head *head;
2981
2982         if (legacy)
2983                 return (struct ceph_mds_request_head_legacy *)p;
2984         head = (struct ceph_mds_request_head *)p;
2985         return (struct ceph_mds_request_head_legacy *)&head->oldest_client_tid;
2986 }
2987
2988 /*
2989  * called under mdsc->mutex
2990  */
2991 static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
2992                                                struct ceph_mds_request *req,
2993                                                bool drop_cap_releases)
2994 {
2995         int mds = session->s_mds;
2996         struct ceph_mds_client *mdsc = session->s_mdsc;
2997         struct ceph_client *cl = mdsc->fsc->client;
2998         struct ceph_msg *msg;
2999         struct ceph_mds_request_head_legacy *lhead;
3000         struct ceph_path_info path_info1 = {0};
3001         struct ceph_path_info path_info2 = {0};
3002         struct dentry *old_dentry = NULL;
3003         int len;
3004         u16 releases;
3005         void *p, *end;
3006         int ret;
3007         bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
3008         u16 request_head_version = mds_supported_head_version(session);
3009         kuid_t caller_fsuid = req->r_cred->fsuid;
3010         kgid_t caller_fsgid = req->r_cred->fsgid;
3011         bool parent_locked = test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
3012
3013         ret = set_request_path_attr(mdsc, req->r_inode, req->r_dentry,
3014                                     req->r_parent, req->r_path1, req->r_ino1.ino,
3015                                     &path_info1, parent_locked);
3016         if (ret < 0) {
3017                 msg = ERR_PTR(ret);
3018                 goto out;
3019         }
3020
3021         /*
3022          * When the parent directory's i_rwsem is *not* locked, req->r_parent may
3023          * have become stale (e.g. after a concurrent rename) between the time the
3024          * dentry was looked up and now.  If we detect that the stored r_parent
3025          * does not match the inode number we just encoded for the request, switch
3026          * to the correct inode so that the MDS receives a valid parent reference.
3027          */
3028         if (!parent_locked && req->r_parent && path_info1.vino.ino &&
3029             ceph_ino(req->r_parent) != path_info1.vino.ino) {
3030                 struct inode *old_parent = req->r_parent;
3031                 struct inode *correct_dir = ceph_get_inode(mdsc->fsc->sb, path_info1.vino, NULL);
3032                 if (!IS_ERR(correct_dir)) {
3033                         WARN_ONCE(1, "ceph: r_parent mismatch (had %llx wanted %llx) - updating\n",
3034                                   ceph_ino(old_parent), path_info1.vino.ino);
3035                         /*
3036                          * Transfer CEPH_CAP_PIN from the old parent to the new one.
3037                          * The pin was taken earlier in ceph_mdsc_submit_request().
3038                          */
3039                         ceph_put_cap_refs(ceph_inode(old_parent), CEPH_CAP_PIN);
3040                         iput(old_parent);
3041                         req->r_parent = correct_dir;
3042                         ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
3043                 }
3044         }
3045
3046         /* If r_old_dentry is set, then assume that its parent is locked */
3047         if (req->r_old_dentry &&
3048             !(req->r_old_dentry->d_flags & DCACHE_DISCONNECTED))
3049                 old_dentry = req->r_old_dentry;
3050         ret = set_request_path_attr(mdsc, NULL, old_dentry,
3051                                     req->r_old_dentry_dir,
3052                                     req->r_path2, req->r_ino2.ino,
3053                                     &path_info2, true);
3054         if (ret < 0) {
3055                 msg = ERR_PTR(ret);
3056                 goto out_free1;
3057         }
3058
3059         req->r_altname = get_fscrypt_altname(req, &req->r_altname_len);
3060         if (IS_ERR(req->r_altname)) {
3061                 msg = ERR_CAST(req->r_altname);
3062                 req->r_altname = NULL;
3063                 goto out_free2;
3064         }
3065
3066         /*
3067          * For old cephs without supporting the 32bit retry/fwd feature
3068          * it will copy the raw memories directly when decoding the
3069          * requests. While new cephs will decode the head depending the
3070          * version member, so we need to make sure it will be compatible
3071          * with them both.
3072          */
3073         if (legacy)
3074                 len = sizeof(struct ceph_mds_request_head_legacy);
3075         else if (request_head_version == 1)
3076                 len = offsetofend(struct ceph_mds_request_head, args);
3077         else if (request_head_version == 2)
3078                 len = offsetofend(struct ceph_mds_request_head, ext_num_fwd);
3079         else
3080                 len = sizeof(struct ceph_mds_request_head);
3081
3082         /* filepaths */
3083         len += 2 * (1 + sizeof(u32) + sizeof(u64));
3084         len += path_info1.pathlen + path_info2.pathlen;
3085
3086         /* cap releases */
3087         len += sizeof(struct ceph_mds_request_release) *
3088                 (!!req->r_inode_drop + !!req->r_dentry_drop +
3089                  !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
3090
3091         if (req->r_dentry_drop)
3092                 len += path_info1.pathlen;
3093         if (req->r_old_dentry_drop)
3094                 len += path_info2.pathlen;
3095
3096         /* MClientRequest tail */
3097
3098         /* req->r_stamp */
3099         len += sizeof(struct ceph_timespec);
3100
3101         /* gid list */
3102         len += sizeof(u32) + (sizeof(u64) * req->r_cred->group_info->ngroups);
3103
3104         /* alternate name */
3105         len += sizeof(u32) + req->r_altname_len;
3106
3107         /* fscrypt_auth */
3108         len += sizeof(u32); // fscrypt_auth
3109         if (req->r_fscrypt_auth)
3110                 len += ceph_fscrypt_auth_len(req->r_fscrypt_auth);
3111
3112         /* fscrypt_file */
3113         len += sizeof(u32);
3114         if (test_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags))
3115                 len += sizeof(__le64);
3116
3117         msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
3118         if (!msg) {
3119                 msg = ERR_PTR(-ENOMEM);
3120                 goto out_free2;
3121         }
3122
3123         msg->hdr.tid = cpu_to_le64(req->r_tid);
3124
3125         lhead = find_legacy_request_head(msg->front.iov_base,
3126                                          session->s_con.peer_features);
3127
3128         if ((req->r_mnt_idmap != &nop_mnt_idmap) &&
3129             !test_bit(CEPHFS_FEATURE_HAS_OWNER_UIDGID, &session->s_features)) {
3130                 WARN_ON_ONCE(!IS_CEPH_MDS_OP_NEWINODE(req->r_op));
3131
3132                 if (enable_unsafe_idmap) {
3133                         pr_warn_once_client(cl,
3134                                 "idmapped mount is used and CEPHFS_FEATURE_HAS_OWNER_UIDGID"
3135                                 " is not supported by MDS. UID/GID-based restrictions may"
3136                                 " not work properly.\n");
3137
3138                         caller_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
3139                                                    VFSUIDT_INIT(req->r_cred->fsuid));
3140                         caller_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
3141                                                    VFSGIDT_INIT(req->r_cred->fsgid));
3142                 } else {
3143                         pr_err_ratelimited_client(cl,
3144                                 "idmapped mount is used and CEPHFS_FEATURE_HAS_OWNER_UIDGID"
3145                                 " is not supported by MDS. Fail request with -EIO.\n");
3146
3147                         ret = -EIO;
3148                         goto out_err;
3149                 }
3150         }
3151
3152         /*
3153          * The ceph_mds_request_head_legacy didn't contain a version field, and
3154          * one was added when we moved the message version from 3->4.
3155          */
3156         if (legacy) {
3157                 msg->hdr.version = cpu_to_le16(3);
3158                 p = msg->front.iov_base + sizeof(*lhead);
3159         } else if (request_head_version == 1) {
3160                 struct ceph_mds_request_head *nhead = msg->front.iov_base;
3161
3162                 msg->hdr.version = cpu_to_le16(4);
3163                 nhead->version = cpu_to_le16(1);
3164                 p = msg->front.iov_base + offsetofend(struct ceph_mds_request_head, args);
3165         } else if (request_head_version == 2) {
3166                 struct ceph_mds_request_head *nhead = msg->front.iov_base;
3167
3168                 msg->hdr.version = cpu_to_le16(6);
3169                 nhead->version = cpu_to_le16(2);
3170
3171                 p = msg->front.iov_base + offsetofend(struct ceph_mds_request_head, ext_num_fwd);
3172         } else {
3173                 struct ceph_mds_request_head *nhead = msg->front.iov_base;
3174                 kuid_t owner_fsuid;
3175                 kgid_t owner_fsgid;
3176
3177                 msg->hdr.version = cpu_to_le16(6);
3178                 nhead->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
3179                 nhead->struct_len = cpu_to_le32(sizeof(struct ceph_mds_request_head));
3180
3181                 if (IS_CEPH_MDS_OP_NEWINODE(req->r_op)) {
3182                         owner_fsuid = from_vfsuid(req->r_mnt_idmap, &init_user_ns,
3183                                                 VFSUIDT_INIT(req->r_cred->fsuid));
3184                         owner_fsgid = from_vfsgid(req->r_mnt_idmap, &init_user_ns,
3185                                                 VFSGIDT_INIT(req->r_cred->fsgid));
3186                         nhead->owner_uid = cpu_to_le32(from_kuid(&init_user_ns, owner_fsuid));
3187                         nhead->owner_gid = cpu_to_le32(from_kgid(&init_user_ns, owner_fsgid));
3188                 } else {
3189                         nhead->owner_uid = cpu_to_le32(-1);
3190                         nhead->owner_gid = cpu_to_le32(-1);
3191                 }
3192
3193                 p = msg->front.iov_base + sizeof(*nhead);
3194         }
3195
3196         end = msg->front.iov_base + msg->front.iov_len;
3197
3198         lhead->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
3199         lhead->op = cpu_to_le32(req->r_op);
3200         lhead->caller_uid = cpu_to_le32(from_kuid(&init_user_ns,
3201                                                   caller_fsuid));
3202         lhead->caller_gid = cpu_to_le32(from_kgid(&init_user_ns,
3203                                                   caller_fsgid));
3204         lhead->ino = cpu_to_le64(req->r_deleg_ino);
3205         lhead->args = req->r_args;
3206
3207         ceph_encode_filepath(&p, end, path_info1.vino.ino, path_info1.path);
3208         ceph_encode_filepath(&p, end, path_info2.vino.ino, path_info2.path);
3209
3210         /* make note of release offset, in case we need to replay */
3211         req->r_request_release_offset = p - msg->front.iov_base;
3212
3213         /* cap releases */
3214         releases = 0;
3215         if (req->r_inode_drop)
3216                 releases += ceph_encode_inode_release(&p,
3217                       req->r_inode ? req->r_inode : d_inode(req->r_dentry),
3218                       mds, req->r_inode_drop, req->r_inode_unless,
3219                       req->r_op == CEPH_MDS_OP_READDIR);
3220         if (req->r_dentry_drop) {
3221                 ret = ceph_encode_dentry_release(&p, req->r_dentry,
3222                                 req->r_parent, mds, req->r_dentry_drop,
3223                                 req->r_dentry_unless);
3224                 if (ret < 0)
3225                         goto out_err;
3226                 releases += ret;
3227         }
3228         if (req->r_old_dentry_drop) {
3229                 ret = ceph_encode_dentry_release(&p, req->r_old_dentry,
3230                                 req->r_old_dentry_dir, mds,
3231                                 req->r_old_dentry_drop,
3232                                 req->r_old_dentry_unless);
3233                 if (ret < 0)
3234                         goto out_err;
3235                 releases += ret;
3236         }
3237         if (req->r_old_inode_drop)
3238                 releases += ceph_encode_inode_release(&p,
3239                       d_inode(req->r_old_dentry),
3240                       mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
3241
3242         if (drop_cap_releases) {
3243                 releases = 0;
3244                 p = msg->front.iov_base + req->r_request_release_offset;
3245         }
3246
3247         lhead->num_releases = cpu_to_le16(releases);
3248
3249         encode_mclientrequest_tail(&p, req);
3250
3251         if (WARN_ON_ONCE(p > end)) {
3252                 ceph_msg_put(msg);
3253                 msg = ERR_PTR(-ERANGE);
3254                 goto out_free2;
3255         }
3256
3257         msg->front.iov_len = p - msg->front.iov_base;
3258         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
3259
3260         if (req->r_pagelist) {
3261                 struct ceph_pagelist *pagelist = req->r_pagelist;
3262                 ceph_msg_data_add_pagelist(msg, pagelist);
3263                 msg->hdr.data_len = cpu_to_le32(pagelist->length);
3264         } else {
3265                 msg->hdr.data_len = 0;
3266         }
3267
3268         msg->hdr.data_off = cpu_to_le16(0);
3269
3270 out_free2:
3271         ceph_mdsc_free_path_info(&path_info2);
3272 out_free1:
3273         ceph_mdsc_free_path_info(&path_info1);
3274 out:
3275         return msg;
3276 out_err:
3277         ceph_msg_put(msg);
3278         msg = ERR_PTR(ret);
3279         goto out_free2;
3280 }
3281
3282 /*
3283  * called under mdsc->mutex if error, under no mutex if
3284  * success.
3285  */
3286 static void complete_request(struct ceph_mds_client *mdsc,
3287                              struct ceph_mds_request *req)
3288 {
3289         req->r_end_latency = ktime_get();
3290
3291         if (req->r_callback)
3292                 req->r_callback(mdsc, req);
3293         complete_all(&req->r_completion);
3294 }
3295
3296 /*
3297  * called under mdsc->mutex
3298  */
3299 static int __prepare_send_request(struct ceph_mds_session *session,
3300                                   struct ceph_mds_request *req,
3301                                   bool drop_cap_releases)
3302 {
3303         int mds = session->s_mds;
3304         struct ceph_mds_client *mdsc = session->s_mdsc;
3305         struct ceph_client *cl = mdsc->fsc->client;
3306         struct ceph_mds_request_head_legacy *lhead;
3307         struct ceph_mds_request_head *nhead;
3308         struct ceph_msg *msg;
3309         int flags = 0, old_max_retry;
3310         bool old_version = !test_bit(CEPHFS_FEATURE_32BITS_RETRY_FWD,
3311                                      &session->s_features);
3312
3313         /*
3314          * Avoid infinite retrying after overflow. The client will
3315          * increase the retry count and if the MDS is old version,
3316          * so we limit to retry at most 256 times.
3317          */
3318         if (req->r_attempts) {
3319                old_max_retry = sizeof_field(struct ceph_mds_request_head,
3320                                             num_retry);
3321                old_max_retry = 1 << (old_max_retry * BITS_PER_BYTE);
3322                if ((old_version && req->r_attempts >= old_max_retry) ||
3323                    ((uint32_t)req->r_attempts >= U32_MAX)) {
3324                         pr_warn_ratelimited_client(cl, "request tid %llu seq overflow\n",
3325                                                    req->r_tid);
3326                         return -EMULTIHOP;
3327                }
3328         }
3329
3330         req->r_attempts++;
3331         if (req->r_inode) {
3332                 struct ceph_cap *cap =
3333                         ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
3334
3335                 if (cap)
3336                         req->r_sent_on_mseq = cap->mseq;
3337                 else
3338                         req->r_sent_on_mseq = -1;
3339         }
3340         doutc(cl, "%p tid %lld %s (attempt %d)\n", req, req->r_tid,
3341               ceph_mds_op_name(req->r_op), req->r_attempts);
3342
3343         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3344                 void *p;
3345
3346                 /*
3347                  * Replay.  Do not regenerate message (and rebuild
3348                  * paths, etc.); just use the original message.
3349                  * Rebuilding paths will break for renames because
3350                  * d_move mangles the src name.
3351                  */
3352                 msg = req->r_request;
3353                 lhead = find_legacy_request_head(msg->front.iov_base,
3354                                                  session->s_con.peer_features);
3355
3356                 flags = le32_to_cpu(lhead->flags);
3357                 flags |= CEPH_MDS_FLAG_REPLAY;
3358                 lhead->flags = cpu_to_le32(flags);
3359
3360                 if (req->r_target_inode)
3361                         lhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
3362
3363                 lhead->num_retry = req->r_attempts - 1;
3364                 if (!old_version) {
3365                         nhead = (struct ceph_mds_request_head*)msg->front.iov_base;
3366                         nhead->ext_num_retry = cpu_to_le32(req->r_attempts - 1);
3367                 }
3368
3369                 /* remove cap/dentry releases from message */
3370                 lhead->num_releases = 0;
3371
3372                 p = msg->front.iov_base + req->r_request_release_offset;
3373                 encode_mclientrequest_tail(&p, req);
3374
3375                 msg->front.iov_len = p - msg->front.iov_base;
3376                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
3377                 return 0;
3378         }
3379
3380         if (req->r_request) {
3381                 ceph_msg_put(req->r_request);
3382                 req->r_request = NULL;
3383         }
3384         msg = create_request_message(session, req, drop_cap_releases);
3385         if (IS_ERR(msg)) {
3386                 req->r_err = PTR_ERR(msg);
3387                 return PTR_ERR(msg);
3388         }
3389         req->r_request = msg;
3390
3391         lhead = find_legacy_request_head(msg->front.iov_base,
3392                                          session->s_con.peer_features);
3393         lhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
3394         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3395                 flags |= CEPH_MDS_FLAG_REPLAY;
3396         if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags))
3397                 flags |= CEPH_MDS_FLAG_ASYNC;
3398         if (req->r_parent)
3399                 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
3400         lhead->flags = cpu_to_le32(flags);
3401         lhead->num_fwd = req->r_num_fwd;
3402         lhead->num_retry = req->r_attempts - 1;
3403         if (!old_version) {
3404                 nhead = (struct ceph_mds_request_head*)msg->front.iov_base;
3405                 nhead->ext_num_fwd = cpu_to_le32(req->r_num_fwd);
3406                 nhead->ext_num_retry = cpu_to_le32(req->r_attempts - 1);
3407         }
3408
3409         doutc(cl, " r_parent = %p\n", req->r_parent);
3410         return 0;
3411 }
3412
3413 /*
3414  * called under mdsc->mutex
3415  */
3416 static int __send_request(struct ceph_mds_session *session,
3417                           struct ceph_mds_request *req,
3418                           bool drop_cap_releases)
3419 {
3420         int err;
3421
3422         err = __prepare_send_request(session, req, drop_cap_releases);
3423         if (!err) {
3424                 ceph_msg_get(req->r_request);
3425                 ceph_con_send(&session->s_con, req->r_request);
3426         }
3427
3428         return err;
3429 }
3430
3431 /*
3432  * send request, or put it on the appropriate wait list.
3433  */
3434 static void __do_request(struct ceph_mds_client *mdsc,
3435                         struct ceph_mds_request *req)
3436 {
3437         struct ceph_client *cl = mdsc->fsc->client;
3438         struct ceph_mds_session *session = NULL;
3439         int mds = -1;
3440         int err = 0;
3441         bool random;
3442
3443         if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
3444                 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
3445                         __unregister_request(mdsc, req);
3446                 return;
3447         }
3448
3449         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO) {
3450                 doutc(cl, "metadata corrupted\n");
3451                 err = -EIO;
3452                 goto finish;
3453         }
3454         if (req->r_timeout &&
3455             time_after_eq(jiffies, req->r_started + req->r_timeout)) {
3456                 doutc(cl, "timed out\n");
3457                 err = -ETIMEDOUT;
3458                 goto finish;
3459         }
3460         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
3461                 doutc(cl, "forced umount\n");
3462                 err = -EIO;
3463                 goto finish;
3464         }
3465         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
3466                 if (mdsc->mdsmap_err) {
3467                         err = mdsc->mdsmap_err;
3468                         doutc(cl, "mdsmap err %d\n", err);
3469                         goto finish;
3470                 }
3471                 if (mdsc->mdsmap->m_epoch == 0) {
3472                         doutc(cl, "no mdsmap, waiting for map\n");
3473                         list_add(&req->r_wait, &mdsc->waiting_for_map);
3474                         return;
3475                 }
3476                 if (!(mdsc->fsc->mount_options->flags &
3477                       CEPH_MOUNT_OPT_MOUNTWAIT) &&
3478                     !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
3479                         err = -EHOSTUNREACH;
3480                         goto finish;
3481                 }
3482         }
3483
3484         put_request_session(req);
3485
3486         mds = __choose_mds(mdsc, req, &random);
3487         if (mds < 0 ||
3488             ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
3489                 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
3490                         err = -EJUKEBOX;
3491                         goto finish;
3492                 }
3493                 doutc(cl, "no mds or not active, waiting for map\n");
3494                 list_add(&req->r_wait, &mdsc->waiting_for_map);
3495                 return;
3496         }
3497
3498         /* get, open session */
3499         session = __ceph_lookup_mds_session(mdsc, mds);
3500         if (!session) {
3501                 session = register_session(mdsc, mds);
3502                 if (IS_ERR(session)) {
3503                         err = PTR_ERR(session);
3504                         goto finish;
3505                 }
3506         }
3507         req->r_session = ceph_get_mds_session(session);
3508
3509         doutc(cl, "mds%d session %p state %s\n", mds, session,
3510               ceph_session_state_name(session->s_state));
3511
3512         /*
3513          * The old ceph will crash the MDSs when see unknown OPs
3514          */
3515         if (req->r_feature_needed > 0 &&
3516             !test_bit(req->r_feature_needed, &session->s_features)) {
3517                 err = -EOPNOTSUPP;
3518                 goto out_session;
3519         }
3520
3521         if (session->s_state != CEPH_MDS_SESSION_OPEN &&
3522             session->s_state != CEPH_MDS_SESSION_HUNG) {
3523                 /*
3524                  * We cannot queue async requests since the caps and delegated
3525                  * inodes are bound to the session. Just return -EJUKEBOX and
3526                  * let the caller retry a sync request in that case.
3527                  */
3528                 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
3529                         err = -EJUKEBOX;
3530                         goto out_session;
3531                 }
3532
3533                 /*
3534                  * If the session has been REJECTED, then return a hard error,
3535                  * unless it's a CLEANRECOVER mount, in which case we'll queue
3536                  * it to the mdsc queue.
3537                  */
3538                 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
3539                         if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
3540                                 list_add(&req->r_wait, &mdsc->waiting_for_map);
3541                         else
3542                                 err = -EACCES;
3543                         goto out_session;
3544                 }
3545
3546                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
3547                     session->s_state == CEPH_MDS_SESSION_CLOSING) {
3548                         err = __open_session(mdsc, session);
3549                         if (err)
3550                                 goto out_session;
3551                         /* retry the same mds later */
3552                         if (random)
3553                                 req->r_resend_mds = mds;
3554                 }
3555                 list_add(&req->r_wait, &session->s_waiting);
3556                 goto out_session;
3557         }
3558
3559         /* send request */
3560         req->r_resend_mds = -1;   /* forget any previous mds hint */
3561
3562         if (req->r_request_started == 0)   /* note request start time */
3563                 req->r_request_started = jiffies;
3564
3565         /*
3566          * For async create we will choose the auth MDS of frag in parent
3567          * directory to send the request and usually this works fine, but
3568          * if the migrated the dirtory to another MDS before it could handle
3569          * it the request will be forwarded.
3570          *
3571          * And then the auth cap will be changed.
3572          */
3573         if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && req->r_num_fwd) {
3574                 struct ceph_dentry_info *di = ceph_dentry(req->r_dentry);
3575                 struct ceph_inode_info *ci;
3576                 struct ceph_cap *cap;
3577
3578                 /*
3579                  * The request maybe handled very fast and the new inode
3580                  * hasn't been linked to the dentry yet. We need to wait
3581                  * for the ceph_finish_async_create(), which shouldn't be
3582                  * stuck too long or fail in thoery, to finish when forwarding
3583                  * the request.
3584                  */
3585                 if (!d_inode(req->r_dentry)) {
3586                         err = wait_on_bit(&di->flags, CEPH_DENTRY_ASYNC_CREATE_BIT,
3587                                           TASK_KILLABLE);
3588                         if (err) {
3589                                 mutex_lock(&req->r_fill_mutex);
3590                                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3591                                 mutex_unlock(&req->r_fill_mutex);
3592                                 goto out_session;
3593                         }
3594                 }
3595
3596                 ci = ceph_inode(d_inode(req->r_dentry));
3597
3598                 spin_lock(&ci->i_ceph_lock);
3599                 cap = ci->i_auth_cap;
3600                 if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE && mds != cap->mds) {
3601                         doutc(cl, "session changed for auth cap %d -> %d\n",
3602                               cap->session->s_mds, session->s_mds);
3603
3604                         /* Remove the auth cap from old session */
3605                         spin_lock(&cap->session->s_cap_lock);
3606                         cap->session->s_nr_caps--;
3607                         list_del_init(&cap->session_caps);
3608                         spin_unlock(&cap->session->s_cap_lock);
3609
3610                         /* Add the auth cap to the new session */
3611                         cap->mds = mds;
3612                         cap->session = session;
3613                         spin_lock(&session->s_cap_lock);
3614                         session->s_nr_caps++;
3615                         list_add_tail(&cap->session_caps, &session->s_caps);
3616                         spin_unlock(&session->s_cap_lock);
3617
3618                         change_auth_cap_ses(ci, session);
3619                 }
3620                 spin_unlock(&ci->i_ceph_lock);
3621         }
3622
3623         err = __send_request(session, req, false);
3624
3625 out_session:
3626         ceph_put_mds_session(session);
3627 finish:
3628         if (err) {
3629                 doutc(cl, "early error %d\n", err);
3630                 req->r_err = err;
3631                 complete_request(mdsc, req);
3632                 __unregister_request(mdsc, req);
3633         }
3634         return;
3635 }
3636
3637 /*
3638  * called under mdsc->mutex
3639  */
3640 static void __wake_requests(struct ceph_mds_client *mdsc,
3641                             struct list_head *head)
3642 {
3643         struct ceph_client *cl = mdsc->fsc->client;
3644         struct ceph_mds_request *req;
3645         LIST_HEAD(tmp_list);
3646
3647         list_splice_init(head, &tmp_list);
3648
3649         while (!list_empty(&tmp_list)) {
3650                 req = list_entry(tmp_list.next,
3651                                  struct ceph_mds_request, r_wait);
3652                 list_del_init(&req->r_wait);
3653                 doutc(cl, " wake request %p tid %llu\n", req,
3654                       req->r_tid);
3655                 __do_request(mdsc, req);
3656         }
3657 }
3658
3659 /*
3660  * Wake up threads with requests pending for @mds, so that they can
3661  * resubmit their requests to a possibly different mds.
3662  */
3663 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
3664 {
3665         struct ceph_client *cl = mdsc->fsc->client;
3666         struct ceph_mds_request *req;
3667         struct rb_node *p = rb_first(&mdsc->request_tree);
3668
3669         doutc(cl, "kick_requests mds%d\n", mds);
3670         while (p) {
3671                 req = rb_entry(p, struct ceph_mds_request, r_node);
3672                 p = rb_next(p);
3673                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3674                         continue;
3675                 if (req->r_attempts > 0)
3676                         continue; /* only new requests */
3677                 if (req->r_session &&
3678                     req->r_session->s_mds == mds) {
3679                         doutc(cl, " kicking tid %llu\n", req->r_tid);
3680                         list_del_init(&req->r_wait);
3681                         __do_request(mdsc, req);
3682                 }
3683         }
3684 }
3685
3686 int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
3687                               struct ceph_mds_request *req)
3688 {
3689         struct ceph_client *cl = mdsc->fsc->client;
3690         int err = 0;
3691
3692         /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
3693         if (req->r_inode)
3694                 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3695         if (req->r_parent) {
3696                 struct ceph_inode_info *ci = ceph_inode(req->r_parent);
3697                 int fmode = (req->r_op & CEPH_MDS_OP_WRITE) ?
3698                             CEPH_FILE_MODE_WR : CEPH_FILE_MODE_RD;
3699                 spin_lock(&ci->i_ceph_lock);
3700                 ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
3701                 __ceph_touch_fmode(ci, mdsc, fmode);
3702                 spin_unlock(&ci->i_ceph_lock);
3703         }
3704         if (req->r_old_dentry_dir)
3705                 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
3706                                   CEPH_CAP_PIN);
3707
3708         if (req->r_inode) {
3709                 err = ceph_wait_on_async_create(req->r_inode);
3710                 if (err) {
3711                         doutc(cl, "wait for async create returned: %d\n", err);
3712                         return err;
3713                 }
3714         }
3715
3716         if (!err && req->r_old_inode) {
3717                 err = ceph_wait_on_async_create(req->r_old_inode);
3718                 if (err) {
3719                         doutc(cl, "wait for async create returned: %d\n", err);
3720                         return err;
3721                 }
3722         }
3723
3724         doutc(cl, "submit_request on %p for inode %p\n", req, dir);
3725         mutex_lock(&mdsc->mutex);
3726         __register_request(mdsc, req, dir);
3727         __do_request(mdsc, req);
3728         err = req->r_err;
3729         mutex_unlock(&mdsc->mutex);
3730         return err;
3731 }
3732
3733 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
3734                            struct ceph_mds_request *req,
3735                            ceph_mds_request_wait_callback_t wait_func)
3736 {
3737         struct ceph_client *cl = mdsc->fsc->client;
3738         int err;
3739
3740         /* wait */
3741         doutc(cl, "do_request waiting\n");
3742         if (wait_func) {
3743                 err = wait_func(mdsc, req);
3744         } else {
3745                 long timeleft = wait_for_completion_killable_timeout(
3746                                         &req->r_completion,
3747                                         ceph_timeout_jiffies(req->r_timeout));
3748                 if (timeleft > 0)
3749                         err = 0;
3750                 else if (!timeleft)
3751                         err = -ETIMEDOUT;  /* timed out */
3752                 else
3753                         err = timeleft;  /* killed */
3754         }
3755         doutc(cl, "do_request waited, got %d\n", err);
3756         mutex_lock(&mdsc->mutex);
3757
3758         /* only abort if we didn't race with a real reply */
3759         if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
3760                 err = le32_to_cpu(req->r_reply_info.head->result);
3761         } else if (err < 0) {
3762                 doutc(cl, "aborted request %lld with %d\n", req->r_tid, err);
3763
3764                 /*
3765                  * ensure we aren't running concurrently with
3766                  * ceph_fill_trace or ceph_readdir_prepopulate, which
3767                  * rely on locks (dir mutex) held by our caller.
3768                  */
3769                 mutex_lock(&req->r_fill_mutex);
3770                 req->r_err = err;
3771                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3772                 mutex_unlock(&req->r_fill_mutex);
3773
3774                 if (req->r_parent &&
3775                     (req->r_op & CEPH_MDS_OP_WRITE))
3776                         ceph_invalidate_dir_request(req);
3777         } else {
3778                 err = req->r_err;
3779         }
3780
3781         mutex_unlock(&mdsc->mutex);
3782         return err;
3783 }
3784
3785 /*
3786  * Synchrously perform an mds request.  Take care of all of the
3787  * session setup, forwarding, retry details.
3788  */
3789 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
3790                          struct inode *dir,
3791                          struct ceph_mds_request *req)
3792 {
3793         struct ceph_client *cl = mdsc->fsc->client;
3794         int err;
3795
3796         doutc(cl, "do_request on %p\n", req);
3797
3798         /* issue */
3799         err = ceph_mdsc_submit_request(mdsc, dir, req);
3800         if (!err)
3801                 err = ceph_mdsc_wait_request(mdsc, req, NULL);
3802         doutc(cl, "do_request %p done, result %d\n", req, err);
3803         return err;
3804 }
3805
3806 /*
3807  * Invalidate dir's completeness, dentry lease state on an aborted MDS
3808  * namespace request.
3809  */
3810 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
3811 {
3812         struct inode *dir = req->r_parent;
3813         struct inode *old_dir = req->r_old_dentry_dir;
3814         struct ceph_client *cl = req->r_mdsc->fsc->client;
3815
3816         doutc(cl, "invalidate_dir_request %p %p (complete, lease(s))\n",
3817               dir, old_dir);
3818
3819         ceph_dir_clear_complete(dir);
3820         if (old_dir)
3821                 ceph_dir_clear_complete(old_dir);
3822         if (req->r_dentry)
3823                 ceph_invalidate_dentry_lease(req->r_dentry);
3824         if (req->r_old_dentry)
3825                 ceph_invalidate_dentry_lease(req->r_old_dentry);
3826 }
3827
3828 /*
3829  * Handle mds reply.
3830  *
3831  * We take the session mutex and parse and process the reply immediately.
3832  * This preserves the logical ordering of replies, capabilities, etc., sent
3833  * by the MDS as they are applied to our local cache.
3834  */
3835 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
3836 {
3837         struct ceph_mds_client *mdsc = session->s_mdsc;
3838         struct ceph_client *cl = mdsc->fsc->client;
3839         struct ceph_mds_request *req;
3840         struct ceph_mds_reply_head *head = msg->front.iov_base;
3841         struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
3842         struct ceph_snap_realm *realm;
3843         u64 tid;
3844         int err, result;
3845         int mds = session->s_mds;
3846         bool close_sessions = false;
3847
3848         if (msg->front.iov_len < sizeof(*head)) {
3849                 pr_err_client(cl, "got corrupt (short) reply\n");
3850                 ceph_msg_dump(msg);
3851                 return;
3852         }
3853
3854         /* get request, session */
3855         tid = le64_to_cpu(msg->hdr.tid);
3856         mutex_lock(&mdsc->mutex);
3857         req = lookup_get_request(mdsc, tid);
3858         if (!req) {
3859                 doutc(cl, "on unknown tid %llu\n", tid);
3860                 mutex_unlock(&mdsc->mutex);
3861                 return;
3862         }
3863         doutc(cl, "handle_reply %p\n", req);
3864
3865         /* correct session? */
3866         if (req->r_session != session) {
3867                 pr_err_client(cl, "got %llu on session mds%d not mds%d\n",
3868                               tid, session->s_mds,
3869                               req->r_session ? req->r_session->s_mds : -1);
3870                 mutex_unlock(&mdsc->mutex);
3871                 goto out;
3872         }
3873
3874         /* dup? */
3875         if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
3876             (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
3877                 pr_warn_client(cl, "got a dup %s reply on %llu from mds%d\n",
3878                                head->safe ? "safe" : "unsafe", tid, mds);
3879                 mutex_unlock(&mdsc->mutex);
3880                 goto out;
3881         }
3882         if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
3883                 pr_warn_client(cl, "got unsafe after safe on %llu from mds%d\n",
3884                                tid, mds);
3885                 mutex_unlock(&mdsc->mutex);
3886                 goto out;
3887         }
3888
3889         result = le32_to_cpu(head->result);
3890
3891         if (head->safe) {
3892                 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
3893                 __unregister_request(mdsc, req);
3894
3895                 /* last request during umount? */
3896                 if (mdsc->stopping && !__get_oldest_req(mdsc))
3897                         complete_all(&mdsc->safe_umount_waiters);
3898
3899                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3900                         /*
3901                          * We already handled the unsafe response, now do the
3902                          * cleanup.  No need to examine the response; the MDS
3903                          * doesn't include any result info in the safe
3904                          * response.  And even if it did, there is nothing
3905                          * useful we could do with a revised return value.
3906                          */
3907                         doutc(cl, "got safe reply %llu, mds%d\n", tid, mds);
3908
3909                         mutex_unlock(&mdsc->mutex);
3910                         goto out;
3911                 }
3912         } else {
3913                 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
3914                 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
3915         }
3916
3917         doutc(cl, "tid %lld result %d\n", tid, result);
3918         if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
3919                 err = parse_reply_info(session, msg, req, (u64)-1);
3920         else
3921                 err = parse_reply_info(session, msg, req,
3922                                        session->s_con.peer_features);
3923         mutex_unlock(&mdsc->mutex);
3924
3925         /* Must find target inode outside of mutexes to avoid deadlocks */
3926         rinfo = &req->r_reply_info;
3927         if ((err >= 0) && rinfo->head->is_target) {
3928                 struct inode *in = xchg(&req->r_new_inode, NULL);
3929                 struct ceph_vino tvino = {
3930                         .ino  = le64_to_cpu(rinfo->targeti.in->ino),
3931                         .snap = le64_to_cpu(rinfo->targeti.in->snapid)
3932                 };
3933
3934                 /*
3935                  * If we ended up opening an existing inode, discard
3936                  * r_new_inode
3937                  */
3938                 if (req->r_op == CEPH_MDS_OP_CREATE &&
3939                     !req->r_reply_info.has_create_ino) {
3940                         /* This should never happen on an async create */
3941                         WARN_ON_ONCE(req->r_deleg_ino);
3942                         iput(in);
3943                         in = NULL;
3944                 }
3945
3946                 in = ceph_get_inode(mdsc->fsc->sb, tvino, in);
3947                 if (IS_ERR(in)) {
3948                         err = PTR_ERR(in);
3949                         mutex_lock(&session->s_mutex);
3950                         goto out_err;
3951                 }
3952                 req->r_target_inode = in;
3953         }
3954
3955         mutex_lock(&session->s_mutex);
3956         if (err < 0) {
3957                 pr_err_client(cl, "got corrupt reply mds%d(tid:%lld)\n",
3958                               mds, tid);
3959                 ceph_msg_dump(msg);
3960                 goto out_err;
3961         }
3962
3963         /* snap trace */
3964         realm = NULL;
3965         if (rinfo->snapblob_len) {
3966                 down_write(&mdsc->snap_rwsem);
3967                 err = ceph_update_snap_trace(mdsc, rinfo->snapblob,
3968                                 rinfo->snapblob + rinfo->snapblob_len,
3969                                 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
3970                                 &realm);
3971                 if (err) {
3972                         up_write(&mdsc->snap_rwsem);
3973                         close_sessions = true;
3974                         if (err == -EIO)
3975                                 ceph_msg_dump(msg);
3976                         goto out_err;
3977                 }
3978                 downgrade_write(&mdsc->snap_rwsem);
3979         } else {
3980                 down_read(&mdsc->snap_rwsem);
3981         }
3982
3983         /* insert trace into our cache */
3984         mutex_lock(&req->r_fill_mutex);
3985         current->journal_info = req;
3986         err = ceph_fill_trace(mdsc->fsc->sb, req);
3987         if (err == 0) {
3988                 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
3989                                     req->r_op == CEPH_MDS_OP_LSSNAP))
3990                         err = ceph_readdir_prepopulate(req, req->r_session);
3991         }
3992         current->journal_info = NULL;
3993         mutex_unlock(&req->r_fill_mutex);
3994
3995         up_read(&mdsc->snap_rwsem);
3996         if (realm)
3997                 ceph_put_snap_realm(mdsc, realm);
3998
3999         if (err == 0) {
4000                 if (req->r_target_inode &&
4001                     test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
4002                         struct ceph_inode_info *ci =
4003                                 ceph_inode(req->r_target_inode);
4004                         spin_lock(&ci->i_unsafe_lock);
4005                         list_add_tail(&req->r_unsafe_target_item,
4006                                       &ci->i_unsafe_iops);
4007                         spin_unlock(&ci->i_unsafe_lock);
4008                 }
4009
4010                 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
4011         }
4012 out_err:
4013         mutex_lock(&mdsc->mutex);
4014         if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
4015                 if (err) {
4016                         req->r_err = err;
4017                 } else {
4018                         req->r_reply =  ceph_msg_get(msg);
4019                         set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
4020                 }
4021         } else {
4022                 doutc(cl, "reply arrived after request %lld was aborted\n", tid);
4023         }
4024         mutex_unlock(&mdsc->mutex);
4025
4026         mutex_unlock(&session->s_mutex);
4027
4028         /* kick calling process */
4029         complete_request(mdsc, req);
4030
4031         ceph_update_metadata_metrics(&mdsc->metric, req->r_start_latency,
4032                                      req->r_end_latency, err);
4033 out:
4034         ceph_mdsc_put_request(req);
4035
4036         /* Defer closing the sessions after s_mutex lock being released */
4037         if (close_sessions)
4038                 ceph_mdsc_close_sessions(mdsc);
4039         return;
4040 }
4041
4042
4043
4044 /*
4045  * handle mds notification that our request has been forwarded.
4046  */
4047 static void handle_forward(struct ceph_mds_client *mdsc,
4048                            struct ceph_mds_session *session,
4049                            struct ceph_msg *msg)
4050 {
4051         struct ceph_client *cl = mdsc->fsc->client;
4052         struct ceph_mds_request *req;
4053         u64 tid = le64_to_cpu(msg->hdr.tid);
4054         u32 next_mds;
4055         u32 fwd_seq;
4056         int err = -EINVAL;
4057         void *p = msg->front.iov_base;
4058         void *end = p + msg->front.iov_len;
4059         bool aborted = false;
4060
4061         ceph_decode_need(&p, end, 2*sizeof(u32), bad);
4062         next_mds = ceph_decode_32(&p);
4063         fwd_seq = ceph_decode_32(&p);
4064
4065         mutex_lock(&mdsc->mutex);
4066         req = lookup_get_request(mdsc, tid);
4067         if (!req) {
4068                 mutex_unlock(&mdsc->mutex);
4069                 doutc(cl, "forward tid %llu to mds%d - req dne\n", tid, next_mds);
4070                 return;  /* dup reply? */
4071         }
4072
4073         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
4074                 doutc(cl, "forward tid %llu aborted, unregistering\n", tid);
4075                 __unregister_request(mdsc, req);
4076         } else if (fwd_seq <= req->r_num_fwd || (uint32_t)fwd_seq >= U32_MAX) {
4077                 /*
4078                  * Avoid infinite retrying after overflow.
4079                  *
4080                  * The MDS will increase the fwd count and in client side
4081                  * if the num_fwd is less than the one saved in request
4082                  * that means the MDS is an old version and overflowed of
4083                  * 8 bits.
4084                  */
4085                 mutex_lock(&req->r_fill_mutex);
4086                 req->r_err = -EMULTIHOP;
4087                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
4088                 mutex_unlock(&req->r_fill_mutex);
4089                 aborted = true;
4090                 pr_warn_ratelimited_client(cl, "forward tid %llu seq overflow\n",
4091                                            tid);
4092         } else {
4093                 /* resend. forward race not possible; mds would drop */
4094                 doutc(cl, "forward tid %llu to mds%d (we resend)\n", tid, next_mds);
4095                 BUG_ON(req->r_err);
4096                 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
4097                 req->r_attempts = 0;
4098                 req->r_num_fwd = fwd_seq;
4099                 req->r_resend_mds = next_mds;
4100                 put_request_session(req);
4101                 __do_request(mdsc, req);
4102         }
4103         mutex_unlock(&mdsc->mutex);
4104
4105         /* kick calling process */
4106         if (aborted)
4107                 complete_request(mdsc, req);
4108         ceph_mdsc_put_request(req);
4109         return;
4110
4111 bad:
4112         pr_err_client(cl, "decode error err=%d\n", err);
4113         ceph_msg_dump(msg);
4114 }
4115
4116 static int __decode_session_metadata(void **p, void *end,
4117                                      bool *blocklisted)
4118 {
4119         /* map<string,string> */
4120         u32 n;
4121         bool err_str;
4122         ceph_decode_32_safe(p, end, n, bad);
4123         while (n-- > 0) {
4124                 u32 len;
4125                 ceph_decode_32_safe(p, end, len, bad);
4126                 ceph_decode_need(p, end, len, bad);
4127                 err_str = !strncmp(*p, "error_string", len);
4128                 *p += len;
4129                 ceph_decode_32_safe(p, end, len, bad);
4130                 ceph_decode_need(p, end, len, bad);
4131                 /*
4132                  * Match "blocklisted (blacklisted)" from newer MDSes,
4133                  * or "blacklisted" from older MDSes.
4134                  */
4135                 if (err_str && strnstr(*p, "blacklisted", len))
4136                         *blocklisted = true;
4137                 *p += len;
4138         }
4139         return 0;
4140 bad:
4141         return -1;
4142 }
4143
4144 /*
4145  * handle a mds session control message
4146  */
4147 static void handle_session(struct ceph_mds_session *session,
4148                            struct ceph_msg *msg)
4149 {
4150         struct ceph_mds_client *mdsc = session->s_mdsc;
4151         struct ceph_client *cl = mdsc->fsc->client;
4152         int mds = session->s_mds;
4153         int msg_version = le16_to_cpu(msg->hdr.version);
4154         void *p = msg->front.iov_base;
4155         void *end = p + msg->front.iov_len;
4156         struct ceph_mds_session_head *h;
4157         struct ceph_mds_cap_auth *cap_auths = NULL;
4158         u32 op, cap_auths_num = 0;
4159         u64 seq, features = 0;
4160         int wake = 0;
4161         bool blocklisted = false;
4162         u32 i;
4163
4164
4165         /* decode */
4166         ceph_decode_need(&p, end, sizeof(*h), bad);
4167         h = p;
4168         p += sizeof(*h);
4169
4170         op = le32_to_cpu(h->op);
4171         seq = le64_to_cpu(h->seq);
4172
4173         if (msg_version >= 3) {
4174                 u32 len;
4175                 /* version >= 2 and < 5, decode metadata, skip otherwise
4176                  * as it's handled via flags.
4177                  */
4178                 if (msg_version >= 5)
4179                         ceph_decode_skip_map(&p, end, string, string, bad);
4180                 else if (__decode_session_metadata(&p, end, &blocklisted) < 0)
4181                         goto bad;
4182
4183                 /* version >= 3, feature bits */
4184                 ceph_decode_32_safe(&p, end, len, bad);
4185                 if (len) {
4186                         ceph_decode_64_safe(&p, end, features, bad);
4187                         p += len - sizeof(features);
4188                 }
4189         }
4190
4191         if (msg_version >= 5) {
4192                 u32 flags, len;
4193
4194                 /* version >= 4 */
4195                 ceph_decode_skip_16(&p, end, bad); /* struct_v, struct_cv */
4196                 ceph_decode_32_safe(&p, end, len, bad); /* len */
4197                 ceph_decode_skip_n(&p, end, len, bad); /* metric_spec */
4198
4199                 /* version >= 5, flags   */
4200                 ceph_decode_32_safe(&p, end, flags, bad);
4201                 if (flags & CEPH_SESSION_BLOCKLISTED) {
4202                         pr_warn_client(cl, "mds%d session blocklisted\n",
4203                                        session->s_mds);
4204                         blocklisted = true;
4205                 }
4206         }
4207
4208         if (msg_version >= 6) {
4209                 ceph_decode_32_safe(&p, end, cap_auths_num, bad);
4210                 doutc(cl, "cap_auths_num %d\n", cap_auths_num);
4211
4212                 if (cap_auths_num && op != CEPH_SESSION_OPEN) {
4213                         WARN_ON_ONCE(op != CEPH_SESSION_OPEN);
4214                         goto skip_cap_auths;
4215                 }
4216
4217                 cap_auths = kcalloc(cap_auths_num,
4218                                     sizeof(struct ceph_mds_cap_auth),
4219                                     GFP_KERNEL);
4220                 if (!cap_auths) {
4221                         pr_err_client(cl, "No memory for cap_auths\n");
4222                         return;
4223                 }
4224
4225                 for (i = 0; i < cap_auths_num; i++) {
4226                         u32 _len, j;
4227
4228                         /* struct_v, struct_compat, and struct_len in MDSCapAuth */
4229                         ceph_decode_skip_n(&p, end, 2 + sizeof(u32), bad);
4230
4231                         /* struct_v, struct_compat, and struct_len in MDSCapMatch */
4232                         ceph_decode_skip_n(&p, end, 2 + sizeof(u32), bad);
4233                         ceph_decode_64_safe(&p, end, cap_auths[i].match.uid, bad);
4234                         ceph_decode_32_safe(&p, end, _len, bad);
4235                         if (_len) {
4236                                 cap_auths[i].match.gids = kcalloc(_len, sizeof(u32),
4237                                                                   GFP_KERNEL);
4238                                 if (!cap_auths[i].match.gids) {
4239                                         pr_err_client(cl, "No memory for gids\n");
4240                                         goto fail;
4241                                 }
4242
4243                                 cap_auths[i].match.num_gids = _len;
4244                                 for (j = 0; j < _len; j++)
4245                                         ceph_decode_32_safe(&p, end,
4246                                                             cap_auths[i].match.gids[j],
4247                                                             bad);
4248                         }
4249
4250                         ceph_decode_32_safe(&p, end, _len, bad);
4251                         if (_len) {
4252                                 cap_auths[i].match.path = kcalloc(_len + 1, sizeof(char),
4253                                                                   GFP_KERNEL);
4254                                 if (!cap_auths[i].match.path) {
4255                                         pr_err_client(cl, "No memory for path\n");
4256                                         goto fail;
4257                                 }
4258                                 ceph_decode_copy(&p, cap_auths[i].match.path, _len);
4259
4260                                 /* Remove the tailing '/' */
4261                                 while (_len && cap_auths[i].match.path[_len - 1] == '/') {
4262                                         cap_auths[i].match.path[_len - 1] = '\0';
4263                                         _len -= 1;
4264                                 }
4265                         }
4266
4267                         ceph_decode_32_safe(&p, end, _len, bad);
4268                         if (_len) {
4269                                 cap_auths[i].match.fs_name = kcalloc(_len + 1, sizeof(char),
4270                                                                      GFP_KERNEL);
4271                                 if (!cap_auths[i].match.fs_name) {
4272                                         pr_err_client(cl, "No memory for fs_name\n");
4273                                         goto fail;
4274                                 }
4275                                 ceph_decode_copy(&p, cap_auths[i].match.fs_name, _len);
4276                         }
4277
4278                         ceph_decode_8_safe(&p, end, cap_auths[i].match.root_squash, bad);
4279                         ceph_decode_8_safe(&p, end, cap_auths[i].readable, bad);
4280                         ceph_decode_8_safe(&p, end, cap_auths[i].writeable, bad);
4281                         doutc(cl, "uid %lld, num_gids %u, path %s, fs_name %s, root_squash %d, readable %d, writeable %d\n",
4282                               cap_auths[i].match.uid, cap_auths[i].match.num_gids,
4283                               cap_auths[i].match.path, cap_auths[i].match.fs_name,
4284                               cap_auths[i].match.root_squash,
4285                               cap_auths[i].readable, cap_auths[i].writeable);
4286                 }
4287         }
4288
4289 skip_cap_auths:
4290         mutex_lock(&mdsc->mutex);
4291         if (op == CEPH_SESSION_OPEN) {
4292                 if (mdsc->s_cap_auths) {
4293                         for (i = 0; i < mdsc->s_cap_auths_num; i++) {
4294                                 kfree(mdsc->s_cap_auths[i].match.gids);
4295                                 kfree(mdsc->s_cap_auths[i].match.path);
4296                                 kfree(mdsc->s_cap_auths[i].match.fs_name);
4297                         }
4298                         kfree(mdsc->s_cap_auths);
4299                 }
4300                 mdsc->s_cap_auths_num = cap_auths_num;
4301                 mdsc->s_cap_auths = cap_auths;
4302         }
4303         if (op == CEPH_SESSION_CLOSE) {
4304                 ceph_get_mds_session(session);
4305                 __unregister_session(mdsc, session);
4306         }
4307         /* FIXME: this ttl calculation is generous */
4308         session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
4309         mutex_unlock(&mdsc->mutex);
4310
4311         mutex_lock(&session->s_mutex);
4312
4313         doutc(cl, "mds%d %s %p state %s seq %llu\n", mds,
4314               ceph_session_op_name(op), session,
4315               ceph_session_state_name(session->s_state), seq);
4316
4317         if (session->s_state == CEPH_MDS_SESSION_HUNG) {
4318                 session->s_state = CEPH_MDS_SESSION_OPEN;
4319                 pr_info_client(cl, "mds%d came back\n", session->s_mds);
4320         }
4321
4322         switch (op) {
4323         case CEPH_SESSION_OPEN:
4324                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
4325                         pr_info_client(cl, "mds%d reconnect success\n",
4326                                        session->s_mds);
4327
4328                 session->s_features = features;
4329                 if (session->s_state == CEPH_MDS_SESSION_OPEN) {
4330                         pr_notice_client(cl, "mds%d is already opened\n",
4331                                          session->s_mds);
4332                 } else {
4333                         session->s_state = CEPH_MDS_SESSION_OPEN;
4334                         renewed_caps(mdsc, session, 0);
4335                         if (test_bit(CEPHFS_FEATURE_METRIC_COLLECT,
4336                                      &session->s_features))
4337                                 metric_schedule_delayed(&mdsc->metric);
4338                 }
4339
4340                 /*
4341                  * The connection maybe broken and the session in client
4342                  * side has been reinitialized, need to update the seq
4343                  * anyway.
4344                  */
4345                 if (!session->s_seq && seq)
4346                         session->s_seq = seq;
4347
4348                 wake = 1;
4349                 if (mdsc->stopping)
4350                         __close_session(mdsc, session);
4351                 break;
4352
4353         case CEPH_SESSION_RENEWCAPS:
4354                 if (session->s_renew_seq == seq)
4355                         renewed_caps(mdsc, session, 1);
4356                 break;
4357
4358         case CEPH_SESSION_CLOSE:
4359                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
4360                         pr_info_client(cl, "mds%d reconnect denied\n",
4361                                        session->s_mds);
4362                 session->s_state = CEPH_MDS_SESSION_CLOSED;
4363                 cleanup_session_requests(mdsc, session);
4364                 remove_session_caps(session);
4365                 wake = 2; /* for good measure */
4366                 wake_up_all(&mdsc->session_close_wq);
4367                 break;
4368
4369         case CEPH_SESSION_STALE:
4370                 pr_info_client(cl, "mds%d caps went stale, renewing\n",
4371                                session->s_mds);
4372                 atomic_inc(&session->s_cap_gen);
4373                 session->s_cap_ttl = jiffies - 1;
4374                 send_renew_caps(mdsc, session);
4375                 break;
4376
4377         case CEPH_SESSION_RECALL_STATE:
4378                 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
4379                 break;
4380
4381         case CEPH_SESSION_FLUSHMSG:
4382                 /* flush cap releases */
4383                 spin_lock(&session->s_cap_lock);
4384                 if (session->s_num_cap_releases)
4385                         ceph_flush_session_cap_releases(mdsc, session);
4386                 spin_unlock(&session->s_cap_lock);
4387
4388                 send_flushmsg_ack(mdsc, session, seq);
4389                 break;
4390
4391         case CEPH_SESSION_FORCE_RO:
4392                 doutc(cl, "force_session_readonly %p\n", session);
4393                 spin_lock(&session->s_cap_lock);
4394                 session->s_readonly = true;
4395                 spin_unlock(&session->s_cap_lock);
4396                 wake_up_session_caps(session, FORCE_RO);
4397                 break;
4398
4399         case CEPH_SESSION_REJECT:
4400                 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
4401                 pr_info_client(cl, "mds%d rejected session\n",
4402                                session->s_mds);
4403                 session->s_state = CEPH_MDS_SESSION_REJECTED;
4404                 cleanup_session_requests(mdsc, session);
4405                 remove_session_caps(session);
4406                 if (blocklisted)
4407                         mdsc->fsc->blocklisted = true;
4408                 wake = 2; /* for good measure */
4409                 break;
4410
4411         default:
4412                 pr_err_client(cl, "bad op %d mds%d\n", op, mds);
4413                 WARN_ON(1);
4414         }
4415
4416         mutex_unlock(&session->s_mutex);
4417         if (wake) {
4418                 mutex_lock(&mdsc->mutex);
4419                 __wake_requests(mdsc, &session->s_waiting);
4420                 if (wake == 2)
4421                         kick_requests(mdsc, mds);
4422                 mutex_unlock(&mdsc->mutex);
4423         }
4424         if (op == CEPH_SESSION_CLOSE)
4425                 ceph_put_mds_session(session);
4426         return;
4427
4428 bad:
4429         pr_err_client(cl, "corrupt message mds%d len %d\n", mds,
4430                       (int)msg->front.iov_len);
4431         ceph_msg_dump(msg);
4432 fail:
4433         for (i = 0; i < cap_auths_num; i++) {
4434                 kfree(cap_auths[i].match.gids);
4435                 kfree(cap_auths[i].match.path);
4436                 kfree(cap_auths[i].match.fs_name);
4437         }
4438         kfree(cap_auths);
4439         return;
4440 }
4441
4442 void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
4443 {
4444         struct ceph_client *cl = req->r_mdsc->fsc->client;
4445         int dcaps;
4446
4447         dcaps = xchg(&req->r_dir_caps, 0);
4448         if (dcaps) {
4449                 doutc(cl, "releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
4450                 ceph_put_cap_refs(ceph_inode(req->r_parent), dcaps);
4451         }
4452 }
4453
4454 void ceph_mdsc_release_dir_caps_async(struct ceph_mds_request *req)
4455 {
4456         struct ceph_client *cl = req->r_mdsc->fsc->client;
4457         int dcaps;
4458
4459         dcaps = xchg(&req->r_dir_caps, 0);
4460         if (dcaps) {
4461                 doutc(cl, "releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
4462                 ceph_put_cap_refs_async(ceph_inode(req->r_parent), dcaps);
4463         }
4464 }
4465
4466 /*
4467  * called under session->mutex.
4468  */
4469 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
4470                                    struct ceph_mds_session *session)
4471 {
4472         struct ceph_mds_request *req, *nreq;
4473         struct rb_node *p;
4474
4475         doutc(mdsc->fsc->client, "mds%d\n", session->s_mds);
4476
4477         mutex_lock(&mdsc->mutex);
4478         list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item)
4479                 __send_request(session, req, true);
4480
4481         /*
4482          * also re-send old requests when MDS enters reconnect stage. So that MDS
4483          * can process completed request in clientreplay stage.
4484          */
4485         p = rb_first(&mdsc->request_tree);
4486         while (p) {
4487                 req = rb_entry(p, struct ceph_mds_request, r_node);
4488                 p = rb_next(p);
4489                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
4490                         continue;
4491                 if (req->r_attempts == 0)
4492                         continue; /* only old requests */
4493                 if (!req->r_session)
4494                         continue;
4495                 if (req->r_session->s_mds != session->s_mds)
4496                         continue;
4497
4498                 ceph_mdsc_release_dir_caps_async(req);
4499
4500                 __send_request(session, req, true);
4501         }
4502         mutex_unlock(&mdsc->mutex);
4503 }
4504
4505 static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
4506 {
4507         struct ceph_msg *reply;
4508         struct ceph_pagelist *_pagelist;
4509         struct page *page;
4510         __le32 *addr;
4511         int err = -ENOMEM;
4512
4513         if (!recon_state->allow_multi)
4514                 return -ENOSPC;
4515
4516         /* can't handle message that contains both caps and realm */
4517         BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
4518
4519         /* pre-allocate new pagelist */
4520         _pagelist = ceph_pagelist_alloc(GFP_NOFS);
4521         if (!_pagelist)
4522                 return -ENOMEM;
4523
4524         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
4525         if (!reply)
4526                 goto fail_msg;
4527
4528         /* placeholder for nr_caps */
4529         err = ceph_pagelist_encode_32(_pagelist, 0);
4530         if (err < 0)
4531                 goto fail;
4532
4533         if (recon_state->nr_caps) {
4534                 /* currently encoding caps */
4535                 err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
4536                 if (err)
4537                         goto fail;
4538         } else {
4539                 /* placeholder for nr_realms (currently encoding relams) */
4540                 err = ceph_pagelist_encode_32(_pagelist, 0);
4541                 if (err < 0)
4542                         goto fail;
4543         }
4544
4545         err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
4546         if (err)
4547                 goto fail;
4548
4549         page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
4550         addr = kmap_atomic(page);
4551         if (recon_state->nr_caps) {
4552                 /* currently encoding caps */
4553                 *addr = cpu_to_le32(recon_state->nr_caps);
4554         } else {
4555                 /* currently encoding relams */
4556                 *(addr + 1) = cpu_to_le32(recon_state->nr_realms);
4557         }
4558         kunmap_atomic(addr);
4559
4560         reply->hdr.version = cpu_to_le16(5);
4561         reply->hdr.compat_version = cpu_to_le16(4);
4562
4563         reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
4564         ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
4565
4566         ceph_con_send(&recon_state->session->s_con, reply);
4567         ceph_pagelist_release(recon_state->pagelist);
4568
4569         recon_state->pagelist = _pagelist;
4570         recon_state->nr_caps = 0;
4571         recon_state->nr_realms = 0;
4572         recon_state->msg_version = 5;
4573         return 0;
4574 fail:
4575         ceph_msg_put(reply);
4576 fail_msg:
4577         ceph_pagelist_release(_pagelist);
4578         return err;
4579 }
4580
4581 static struct dentry* d_find_primary(struct inode *inode)
4582 {
4583         struct dentry *alias, *dn = NULL;
4584
4585         if (hlist_empty(&inode->i_dentry))
4586                 return NULL;
4587
4588         spin_lock(&inode->i_lock);
4589         if (hlist_empty(&inode->i_dentry))
4590                 goto out_unlock;
4591
4592         if (S_ISDIR(inode->i_mode)) {
4593                 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
4594                 if (!IS_ROOT(alias))
4595                         dn = dget(alias);
4596                 goto out_unlock;
4597         }
4598
4599         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
4600                 spin_lock(&alias->d_lock);
4601                 if (!d_unhashed(alias) &&
4602                     (ceph_dentry(alias)->flags & CEPH_DENTRY_PRIMARY_LINK)) {
4603                         dn = dget_dlock(alias);
4604                 }
4605                 spin_unlock(&alias->d_lock);
4606                 if (dn)
4607                         break;
4608         }
4609 out_unlock:
4610         spin_unlock(&inode->i_lock);
4611         return dn;
4612 }
4613
4614 /*
4615  * Encode information about a cap for a reconnect with the MDS.
4616  */
4617 static int reconnect_caps_cb(struct inode *inode, int mds, void *arg)
4618 {
4619         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
4620         struct ceph_client *cl = ceph_inode_to_client(inode);
4621         union {
4622                 struct ceph_mds_cap_reconnect v2;
4623                 struct ceph_mds_cap_reconnect_v1 v1;
4624         } rec;
4625         struct ceph_inode_info *ci = ceph_inode(inode);
4626         struct ceph_reconnect_state *recon_state = arg;
4627         struct ceph_pagelist *pagelist = recon_state->pagelist;
4628         struct dentry *dentry;
4629         struct ceph_cap *cap;
4630         struct ceph_path_info path_info = {0};
4631         int err;
4632         u64 snap_follows;
4633
4634         dentry = d_find_primary(inode);
4635         if (dentry) {
4636                 /* set pathbase to parent dir when msg_version >= 2 */
4637                 char *path = ceph_mdsc_build_path(mdsc, dentry, &path_info,
4638                                             recon_state->msg_version >= 2);
4639                 dput(dentry);
4640                 if (IS_ERR(path)) {
4641                         err = PTR_ERR(path);
4642                         goto out_err;
4643                 }
4644         }
4645
4646         spin_lock(&ci->i_ceph_lock);
4647         cap = __get_cap_for_mds(ci, mds);
4648         if (!cap) {
4649                 spin_unlock(&ci->i_ceph_lock);
4650                 err = 0;
4651                 goto out_err;
4652         }
4653         doutc(cl, " adding %p ino %llx.%llx cap %p %lld %s\n", inode,
4654               ceph_vinop(inode), cap, cap->cap_id,
4655               ceph_cap_string(cap->issued));
4656
4657         cap->seq = 0;        /* reset cap seq */
4658         cap->issue_seq = 0;  /* and issue_seq */
4659         cap->mseq = 0;       /* and migrate_seq */
4660         cap->cap_gen = atomic_read(&cap->session->s_cap_gen);
4661
4662         /* These are lost when the session goes away */
4663         if (S_ISDIR(inode->i_mode)) {
4664                 if (cap->issued & CEPH_CAP_DIR_CREATE) {
4665                         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
4666                         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
4667                 }
4668                 cap->issued &= ~CEPH_CAP_ANY_DIR_OPS;
4669         }
4670
4671         if (recon_state->msg_version >= 2) {
4672                 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
4673                 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
4674                 rec.v2.issued = cpu_to_le32(cap->issued);
4675                 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
4676                 rec.v2.pathbase = cpu_to_le64(path_info.vino.ino);
4677                 rec.v2.flock_len = (__force __le32)
4678                         ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
4679         } else {
4680                 struct timespec64 ts;
4681
4682                 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
4683                 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
4684                 rec.v1.issued = cpu_to_le32(cap->issued);
4685                 rec.v1.size = cpu_to_le64(i_size_read(inode));
4686                 ts = inode_get_mtime(inode);
4687                 ceph_encode_timespec64(&rec.v1.mtime, &ts);
4688                 ts = inode_get_atime(inode);
4689                 ceph_encode_timespec64(&rec.v1.atime, &ts);
4690                 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
4691                 rec.v1.pathbase = cpu_to_le64(path_info.vino.ino);
4692         }
4693
4694         if (list_empty(&ci->i_cap_snaps)) {
4695                 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
4696         } else {
4697                 struct ceph_cap_snap *capsnap =
4698                         list_first_entry(&ci->i_cap_snaps,
4699                                          struct ceph_cap_snap, ci_item);
4700                 snap_follows = capsnap->follows;
4701         }
4702         spin_unlock(&ci->i_ceph_lock);
4703
4704         if (recon_state->msg_version >= 2) {
4705                 int num_fcntl_locks, num_flock_locks;
4706                 struct ceph_filelock *flocks = NULL;
4707                 size_t struct_len, total_len = sizeof(u64);
4708                 u8 struct_v = 0;
4709
4710 encode_again:
4711                 if (rec.v2.flock_len) {
4712                         ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
4713                 } else {
4714                         num_fcntl_locks = 0;
4715                         num_flock_locks = 0;
4716                 }
4717                 if (num_fcntl_locks + num_flock_locks > 0) {
4718                         flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
4719                                                sizeof(struct ceph_filelock),
4720                                                GFP_NOFS);
4721                         if (!flocks) {
4722                                 err = -ENOMEM;
4723                                 goto out_err;
4724                         }
4725                         err = ceph_encode_locks_to_buffer(inode, flocks,
4726                                                           num_fcntl_locks,
4727                                                           num_flock_locks);
4728                         if (err) {
4729                                 kfree(flocks);
4730                                 flocks = NULL;
4731                                 if (err == -ENOSPC)
4732                                         goto encode_again;
4733                                 goto out_err;
4734                         }
4735                 } else {
4736                         kfree(flocks);
4737                         flocks = NULL;
4738                 }
4739
4740                 if (recon_state->msg_version >= 3) {
4741                         /* version, compat_version and struct_len */
4742                         total_len += 2 * sizeof(u8) + sizeof(u32);
4743                         struct_v = 2;
4744                 }
4745                 /*
4746                  * number of encoded locks is stable, so copy to pagelist
4747                  */
4748                 struct_len = 2 * sizeof(u32) +
4749                             (num_fcntl_locks + num_flock_locks) *
4750                             sizeof(struct ceph_filelock);
4751                 rec.v2.flock_len = cpu_to_le32(struct_len);
4752
4753                 struct_len += sizeof(u32) + path_info.pathlen + sizeof(rec.v2);
4754
4755                 if (struct_v >= 2)
4756                         struct_len += sizeof(u64); /* snap_follows */
4757
4758                 total_len += struct_len;
4759
4760                 if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
4761                         err = send_reconnect_partial(recon_state);
4762                         if (err)
4763                                 goto out_freeflocks;
4764                         pagelist = recon_state->pagelist;
4765                 }
4766
4767                 err = ceph_pagelist_reserve(pagelist, total_len);
4768                 if (err)
4769                         goto out_freeflocks;
4770
4771                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
4772                 if (recon_state->msg_version >= 3) {
4773                         ceph_pagelist_encode_8(pagelist, struct_v);
4774                         ceph_pagelist_encode_8(pagelist, 1);
4775                         ceph_pagelist_encode_32(pagelist, struct_len);
4776                 }
4777                 ceph_pagelist_encode_string(pagelist, (char *)path_info.path, path_info.pathlen);
4778                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
4779                 ceph_locks_to_pagelist(flocks, pagelist,
4780                                        num_fcntl_locks, num_flock_locks);
4781                 if (struct_v >= 2)
4782                         ceph_pagelist_encode_64(pagelist, snap_follows);
4783 out_freeflocks:
4784                 kfree(flocks);
4785         } else {
4786                 err = ceph_pagelist_reserve(pagelist,
4787                                             sizeof(u64) + sizeof(u32) +
4788                                             path_info.pathlen + sizeof(rec.v1));
4789                 if (err)
4790                         goto out_err;
4791
4792                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
4793                 ceph_pagelist_encode_string(pagelist, (char *)path_info.path, path_info.pathlen);
4794                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
4795         }
4796
4797 out_err:
4798         ceph_mdsc_free_path_info(&path_info);
4799         if (!err)
4800                 recon_state->nr_caps++;
4801         return err;
4802 }
4803
4804 static int encode_snap_realms(struct ceph_mds_client *mdsc,
4805                               struct ceph_reconnect_state *recon_state)
4806 {
4807         struct rb_node *p;
4808         struct ceph_pagelist *pagelist = recon_state->pagelist;
4809         struct ceph_client *cl = mdsc->fsc->client;
4810         int err = 0;
4811
4812         if (recon_state->msg_version >= 4) {
4813                 err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
4814                 if (err < 0)
4815                         goto fail;
4816         }
4817
4818         /*
4819          * snaprealms.  we provide mds with the ino, seq (version), and
4820          * parent for all of our realms.  If the mds has any newer info,
4821          * it will tell us.
4822          */
4823         for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
4824                 struct ceph_snap_realm *realm =
4825                        rb_entry(p, struct ceph_snap_realm, node);
4826                 struct ceph_mds_snaprealm_reconnect sr_rec;
4827
4828                 if (recon_state->msg_version >= 4) {
4829                         size_t need = sizeof(u8) * 2 + sizeof(u32) +
4830                                       sizeof(sr_rec);
4831
4832                         if (pagelist->length + need > RECONNECT_MAX_SIZE) {
4833                                 err = send_reconnect_partial(recon_state);
4834                                 if (err)
4835                                         goto fail;
4836                                 pagelist = recon_state->pagelist;
4837                         }
4838
4839                         err = ceph_pagelist_reserve(pagelist, need);
4840                         if (err)
4841                                 goto fail;
4842
4843                         ceph_pagelist_encode_8(pagelist, 1);
4844                         ceph_pagelist_encode_8(pagelist, 1);
4845                         ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
4846                 }
4847
4848                 doutc(cl, " adding snap realm %llx seq %lld parent %llx\n",
4849                       realm->ino, realm->seq, realm->parent_ino);
4850                 sr_rec.ino = cpu_to_le64(realm->ino);
4851                 sr_rec.seq = cpu_to_le64(realm->seq);
4852                 sr_rec.parent = cpu_to_le64(realm->parent_ino);
4853
4854                 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
4855                 if (err)
4856                         goto fail;
4857
4858                 recon_state->nr_realms++;
4859         }
4860 fail:
4861         return err;
4862 }
4863
4864
4865 /*
4866  * If an MDS fails and recovers, clients need to reconnect in order to
4867  * reestablish shared state.  This includes all caps issued through
4868  * this session _and_ the snap_realm hierarchy.  Because it's not
4869  * clear which snap realms the mds cares about, we send everything we
4870  * know about.. that ensures we'll then get any new info the
4871  * recovering MDS might have.
4872  *
4873  * This is a relatively heavyweight operation, but it's rare.
4874  */
4875 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
4876                                struct ceph_mds_session *session)
4877 {
4878         struct ceph_client *cl = mdsc->fsc->client;
4879         struct ceph_msg *reply;
4880         int mds = session->s_mds;
4881         int err = -ENOMEM;
4882         struct ceph_reconnect_state recon_state = {
4883                 .session = session,
4884         };
4885         LIST_HEAD(dispose);
4886
4887         pr_info_client(cl, "mds%d reconnect start\n", mds);
4888
4889         recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
4890         if (!recon_state.pagelist)
4891                 goto fail_nopagelist;
4892
4893         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
4894         if (!reply)
4895                 goto fail_nomsg;
4896
4897         xa_destroy(&session->s_delegated_inos);
4898
4899         mutex_lock(&session->s_mutex);
4900         session->s_state = CEPH_MDS_SESSION_RECONNECTING;
4901         session->s_seq = 0;
4902
4903         doutc(cl, "session %p state %s\n", session,
4904               ceph_session_state_name(session->s_state));
4905
4906         atomic_inc(&session->s_cap_gen);
4907
4908         spin_lock(&session->s_cap_lock);
4909         /* don't know if session is readonly */
4910         session->s_readonly = 0;
4911         /*
4912          * notify __ceph_remove_cap() that we are composing cap reconnect.
4913          * If a cap get released before being added to the cap reconnect,
4914          * __ceph_remove_cap() should skip queuing cap release.
4915          */
4916         session->s_cap_reconnect = 1;
4917         /* drop old cap expires; we're about to reestablish that state */
4918         detach_cap_releases(session, &dispose);
4919         spin_unlock(&session->s_cap_lock);
4920         dispose_cap_releases(mdsc, &dispose);
4921
4922         /* trim unused caps to reduce MDS's cache rejoin time */
4923         if (mdsc->fsc->sb->s_root)
4924                 shrink_dcache_parent(mdsc->fsc->sb->s_root);
4925
4926         ceph_con_close(&session->s_con);
4927         ceph_con_open(&session->s_con,
4928                       CEPH_ENTITY_TYPE_MDS, mds,
4929                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
4930
4931         /* replay unsafe requests */
4932         replay_unsafe_requests(mdsc, session);
4933
4934         ceph_early_kick_flushing_caps(mdsc, session);
4935
4936         down_read(&mdsc->snap_rwsem);
4937
4938         /* placeholder for nr_caps */
4939         err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
4940         if (err)
4941                 goto fail;
4942
4943         if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
4944                 recon_state.msg_version = 3;
4945                 recon_state.allow_multi = true;
4946         } else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
4947                 recon_state.msg_version = 3;
4948         } else {
4949                 recon_state.msg_version = 2;
4950         }
4951         /* traverse this session's caps */
4952         err = ceph_iterate_session_caps(session, reconnect_caps_cb, &recon_state);
4953
4954         spin_lock(&session->s_cap_lock);
4955         session->s_cap_reconnect = 0;
4956         spin_unlock(&session->s_cap_lock);
4957
4958         if (err < 0)
4959                 goto fail;
4960
4961         /* check if all realms can be encoded into current message */
4962         if (mdsc->num_snap_realms) {
4963                 size_t total_len =
4964                         recon_state.pagelist->length +
4965                         mdsc->num_snap_realms *
4966                         sizeof(struct ceph_mds_snaprealm_reconnect);
4967                 if (recon_state.msg_version >= 4) {
4968                         /* number of realms */
4969                         total_len += sizeof(u32);
4970                         /* version, compat_version and struct_len */
4971                         total_len += mdsc->num_snap_realms *
4972                                      (2 * sizeof(u8) + sizeof(u32));
4973                 }
4974                 if (total_len > RECONNECT_MAX_SIZE) {
4975                         if (!recon_state.allow_multi) {
4976                                 err = -ENOSPC;
4977                                 goto fail;
4978                         }
4979                         if (recon_state.nr_caps) {
4980                                 err = send_reconnect_partial(&recon_state);
4981                                 if (err)
4982                                         goto fail;
4983                         }
4984                         recon_state.msg_version = 5;
4985                 }
4986         }
4987
4988         err = encode_snap_realms(mdsc, &recon_state);
4989         if (err < 0)
4990                 goto fail;
4991
4992         if (recon_state.msg_version >= 5) {
4993                 err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
4994                 if (err < 0)
4995                         goto fail;
4996         }
4997
4998         if (recon_state.nr_caps || recon_state.nr_realms) {
4999                 struct page *page =
5000                         list_first_entry(&recon_state.pagelist->head,
5001                                         struct page, lru);
5002                 __le32 *addr = kmap_atomic(page);
5003                 if (recon_state.nr_caps) {
5004                         WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
5005                         *addr = cpu_to_le32(recon_state.nr_caps);
5006                 } else if (recon_state.msg_version >= 4) {
5007                         *(addr + 1) = cpu_to_le32(recon_state.nr_realms);
5008                 }
5009                 kunmap_atomic(addr);
5010         }
5011
5012         reply->hdr.version = cpu_to_le16(recon_state.msg_version);
5013         if (recon_state.msg_version >= 4)
5014                 reply->hdr.compat_version = cpu_to_le16(4);
5015
5016         reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
5017         ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
5018
5019         ceph_con_send(&session->s_con, reply);
5020
5021         mutex_unlock(&session->s_mutex);
5022
5023         mutex_lock(&mdsc->mutex);
5024         __wake_requests(mdsc, &session->s_waiting);
5025         mutex_unlock(&mdsc->mutex);
5026
5027         up_read(&mdsc->snap_rwsem);
5028         ceph_pagelist_release(recon_state.pagelist);
5029         return;
5030
5031 fail:
5032         ceph_msg_put(reply);
5033         up_read(&mdsc->snap_rwsem);
5034         mutex_unlock(&session->s_mutex);
5035 fail_nomsg:
5036         ceph_pagelist_release(recon_state.pagelist);
5037 fail_nopagelist:
5038         pr_err_client(cl, "error %d preparing reconnect for mds%d\n",
5039                       err, mds);
5040         return;
5041 }
5042
5043
5044 /*
5045  * compare old and new mdsmaps, kicking requests
5046  * and closing out old connections as necessary
5047  *
5048  * called under mdsc->mutex.
5049  */
5050 static void check_new_map(struct ceph_mds_client *mdsc,
5051                           struct ceph_mdsmap *newmap,
5052                           struct ceph_mdsmap *oldmap)
5053 {
5054         int i, j, err;
5055         int oldstate, newstate;
5056         struct ceph_mds_session *s;
5057         unsigned long targets[DIV_ROUND_UP(CEPH_MAX_MDS, sizeof(unsigned long))] = {0};
5058         struct ceph_client *cl = mdsc->fsc->client;
5059
5060         doutc(cl, "new %u old %u\n", newmap->m_epoch, oldmap->m_epoch);
5061
5062         if (newmap->m_info) {
5063                 for (i = 0; i < newmap->possible_max_rank; i++) {
5064                         for (j = 0; j < newmap->m_info[i].num_export_targets; j++)
5065                                 set_bit(newmap->m_info[i].export_targets[j], targets);
5066                 }
5067         }
5068
5069         for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
5070                 if (!mdsc->sessions[i])
5071                         continue;
5072                 s = mdsc->sessions[i];
5073                 oldstate = ceph_mdsmap_get_state(oldmap, i);
5074                 newstate = ceph_mdsmap_get_state(newmap, i);
5075
5076                 doutc(cl, "mds%d state %s%s -> %s%s (session %s)\n",
5077                       i, ceph_mds_state_name(oldstate),
5078                       ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
5079                       ceph_mds_state_name(newstate),
5080                       ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
5081                       ceph_session_state_name(s->s_state));
5082
5083                 if (i >= newmap->possible_max_rank) {
5084                         /* force close session for stopped mds */
5085                         ceph_get_mds_session(s);
5086                         __unregister_session(mdsc, s);
5087                         __wake_requests(mdsc, &s->s_waiting);
5088                         mutex_unlock(&mdsc->mutex);
5089
5090                         mutex_lock(&s->s_mutex);
5091                         cleanup_session_requests(mdsc, s);
5092                         remove_session_caps(s);
5093                         mutex_unlock(&s->s_mutex);
5094
5095                         ceph_put_mds_session(s);
5096
5097                         mutex_lock(&mdsc->mutex);
5098                         kick_requests(mdsc, i);
5099                         continue;
5100                 }
5101
5102                 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
5103                            ceph_mdsmap_get_addr(newmap, i),
5104                            sizeof(struct ceph_entity_addr))) {
5105                         /* just close it */
5106                         mutex_unlock(&mdsc->mutex);
5107                         mutex_lock(&s->s_mutex);
5108                         mutex_lock(&mdsc->mutex);
5109                         ceph_con_close(&s->s_con);
5110                         mutex_unlock(&s->s_mutex);
5111                         s->s_state = CEPH_MDS_SESSION_RESTARTING;
5112                 } else if (oldstate == newstate) {
5113                         continue;  /* nothing new with this mds */
5114                 }
5115
5116                 /*
5117                  * send reconnect?
5118                  */
5119                 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
5120                     newstate >= CEPH_MDS_STATE_RECONNECT) {
5121                         mutex_unlock(&mdsc->mutex);
5122                         clear_bit(i, targets);
5123                         send_mds_reconnect(mdsc, s);
5124                         mutex_lock(&mdsc->mutex);
5125                 }
5126
5127                 /*
5128                  * kick request on any mds that has gone active.
5129                  */
5130                 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
5131                     newstate >= CEPH_MDS_STATE_ACTIVE) {
5132                         if (oldstate != CEPH_MDS_STATE_CREATING &&
5133                             oldstate != CEPH_MDS_STATE_STARTING)
5134                                 pr_info_client(cl, "mds%d recovery completed\n",
5135                                                s->s_mds);
5136                         kick_requests(mdsc, i);
5137                         mutex_unlock(&mdsc->mutex);
5138                         mutex_lock(&s->s_mutex);
5139                         mutex_lock(&mdsc->mutex);
5140                         ceph_kick_flushing_caps(mdsc, s);
5141                         mutex_unlock(&s->s_mutex);
5142                         wake_up_session_caps(s, RECONNECT);
5143                 }
5144         }
5145
5146         /*
5147          * Only open and reconnect sessions that don't exist yet.
5148          */
5149         for (i = 0; i < newmap->possible_max_rank; i++) {
5150                 /*
5151                  * In case the import MDS is crashed just after
5152                  * the EImportStart journal is flushed, so when
5153                  * a standby MDS takes over it and is replaying
5154                  * the EImportStart journal the new MDS daemon
5155                  * will wait the client to reconnect it, but the
5156                  * client may never register/open the session yet.
5157                  *
5158                  * Will try to reconnect that MDS daemon if the
5159                  * rank number is in the export targets array and
5160                  * is the up:reconnect state.
5161                  */
5162                 newstate = ceph_mdsmap_get_state(newmap, i);
5163                 if (!test_bit(i, targets) || newstate != CEPH_MDS_STATE_RECONNECT)
5164                         continue;
5165
5166                 /*
5167                  * The session maybe registered and opened by some
5168                  * requests which were choosing random MDSes during
5169                  * the mdsc->mutex's unlock/lock gap below in rare
5170                  * case. But the related MDS daemon will just queue
5171                  * that requests and be still waiting for the client's
5172                  * reconnection request in up:reconnect state.
5173                  */
5174                 s = __ceph_lookup_mds_session(mdsc, i);
5175                 if (likely(!s)) {
5176                         s = __open_export_target_session(mdsc, i);
5177                         if (IS_ERR(s)) {
5178                                 err = PTR_ERR(s);
5179                                 pr_err_client(cl,
5180                                               "failed to open export target session, err %d\n",
5181                                               err);
5182                                 continue;
5183                         }
5184                 }
5185                 doutc(cl, "send reconnect to export target mds.%d\n", i);
5186                 mutex_unlock(&mdsc->mutex);
5187                 send_mds_reconnect(mdsc, s);
5188                 ceph_put_mds_session(s);
5189                 mutex_lock(&mdsc->mutex);
5190         }
5191
5192         for (i = 0; i < newmap->possible_max_rank && i < mdsc->max_sessions; i++) {
5193                 s = mdsc->sessions[i];
5194                 if (!s)
5195                         continue;
5196                 if (!ceph_mdsmap_is_laggy(newmap, i))
5197                         continue;
5198                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
5199                     s->s_state == CEPH_MDS_SESSION_HUNG ||
5200                     s->s_state == CEPH_MDS_SESSION_CLOSING) {
5201                         doutc(cl, " connecting to export targets of laggy mds%d\n", i);
5202                         __open_export_target_sessions(mdsc, s);
5203                 }
5204         }
5205 }
5206
5207
5208
5209 /*
5210  * leases
5211  */
5212
5213 /*
5214  * caller must hold session s_mutex, dentry->d_lock
5215  */
5216 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
5217 {
5218         struct ceph_dentry_info *di = ceph_dentry(dentry);
5219
5220         ceph_put_mds_session(di->lease_session);
5221         di->lease_session = NULL;
5222 }
5223
5224 static void handle_lease(struct ceph_mds_client *mdsc,
5225                          struct ceph_mds_session *session,
5226                          struct ceph_msg *msg)
5227 {
5228         struct ceph_client *cl = mdsc->fsc->client;
5229         struct super_block *sb = mdsc->fsc->sb;
5230         struct inode *inode;
5231         struct dentry *parent, *dentry;
5232         struct ceph_dentry_info *di;
5233         int mds = session->s_mds;
5234         struct ceph_mds_lease *h = msg->front.iov_base;
5235         u32 seq;
5236         struct ceph_vino vino;
5237         struct qstr dname;
5238         int release = 0;
5239
5240         doutc(cl, "from mds%d\n", mds);
5241
5242         if (!ceph_inc_mds_stopping_blocker(mdsc, session))
5243                 return;
5244
5245         /* decode */
5246         if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
5247                 goto bad;
5248         vino.ino = le64_to_cpu(h->ino);
5249         vino.snap = CEPH_NOSNAP;
5250         seq = le32_to_cpu(h->seq);
5251         dname.len = get_unaligned_le32(h + 1);
5252         if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
5253                 goto bad;
5254         dname.name = (void *)(h + 1) + sizeof(u32);
5255
5256         /* lookup inode */
5257         inode = ceph_find_inode(sb, vino);
5258         doutc(cl, "%s, ino %llx %p %.*s\n", ceph_lease_op_name(h->action),
5259               vino.ino, inode, dname.len, dname.name);
5260
5261         mutex_lock(&session->s_mutex);
5262         if (!inode) {
5263                 doutc(cl, "no inode %llx\n", vino.ino);
5264                 goto release;
5265         }
5266
5267         /* dentry */
5268         parent = d_find_alias(inode);
5269         if (!parent) {
5270                 doutc(cl, "no parent dentry on inode %p\n", inode);
5271                 WARN_ON(1);
5272                 goto release;  /* hrm... */
5273         }
5274         dname.hash = full_name_hash(parent, dname.name, dname.len);
5275         dentry = d_lookup(parent, &dname);
5276         dput(parent);
5277         if (!dentry)
5278                 goto release;
5279
5280         spin_lock(&dentry->d_lock);
5281         di = ceph_dentry(dentry);
5282         switch (h->action) {
5283         case CEPH_MDS_LEASE_REVOKE:
5284                 if (di->lease_session == session) {
5285                         if (ceph_seq_cmp(di->lease_seq, seq) > 0)
5286                                 h->seq = cpu_to_le32(di->lease_seq);
5287                         __ceph_mdsc_drop_dentry_lease(dentry);
5288                 }
5289                 release = 1;
5290                 break;
5291
5292         case CEPH_MDS_LEASE_RENEW:
5293                 if (di->lease_session == session &&
5294                     di->lease_gen == atomic_read(&session->s_cap_gen) &&
5295                     di->lease_renew_from &&
5296                     di->lease_renew_after == 0) {
5297                         unsigned long duration =
5298                                 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
5299
5300                         di->lease_seq = seq;
5301                         di->time = di->lease_renew_from + duration;
5302                         di->lease_renew_after = di->lease_renew_from +
5303                                 (duration >> 1);
5304                         di->lease_renew_from = 0;
5305                 }
5306                 break;
5307         }
5308         spin_unlock(&dentry->d_lock);
5309         dput(dentry);
5310
5311         if (!release)
5312                 goto out;
5313
5314 release:
5315         /* let's just reuse the same message */
5316         h->action = CEPH_MDS_LEASE_REVOKE_ACK;
5317         ceph_msg_get(msg);
5318         ceph_con_send(&session->s_con, msg);
5319
5320 out:
5321         mutex_unlock(&session->s_mutex);
5322         iput(inode);
5323
5324         ceph_dec_mds_stopping_blocker(mdsc);
5325         return;
5326
5327 bad:
5328         ceph_dec_mds_stopping_blocker(mdsc);
5329
5330         pr_err_client(cl, "corrupt lease message\n");
5331         ceph_msg_dump(msg);
5332 }
5333
5334 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
5335                               struct dentry *dentry, char action,
5336                               u32 seq)
5337 {
5338         struct ceph_client *cl = session->s_mdsc->fsc->client;
5339         struct ceph_msg *msg;
5340         struct ceph_mds_lease *lease;
5341         struct inode *dir;
5342         int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
5343
5344         doutc(cl, "identry %p %s to mds%d\n", dentry, ceph_lease_op_name(action),
5345               session->s_mds);
5346
5347         msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
5348         if (!msg)
5349                 return;
5350         lease = msg->front.iov_base;
5351         lease->action = action;
5352         lease->seq = cpu_to_le32(seq);
5353
5354         spin_lock(&dentry->d_lock);
5355         dir = d_inode(dentry->d_parent);
5356         lease->ino = cpu_to_le64(ceph_ino(dir));
5357         lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
5358
5359         put_unaligned_le32(dentry->d_name.len, lease + 1);
5360         memcpy((void *)(lease + 1) + 4,
5361                dentry->d_name.name, dentry->d_name.len);
5362         spin_unlock(&dentry->d_lock);
5363
5364         ceph_con_send(&session->s_con, msg);
5365 }
5366
5367 /*
5368  * lock unlock the session, to wait ongoing session activities
5369  */
5370 static void lock_unlock_session(struct ceph_mds_session *s)
5371 {
5372         mutex_lock(&s->s_mutex);
5373         mutex_unlock(&s->s_mutex);
5374 }
5375
5376 static void maybe_recover_session(struct ceph_mds_client *mdsc)
5377 {
5378         struct ceph_client *cl = mdsc->fsc->client;
5379         struct ceph_fs_client *fsc = mdsc->fsc;
5380
5381         if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
5382                 return;
5383
5384         if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
5385                 return;
5386
5387         if (!READ_ONCE(fsc->blocklisted))
5388                 return;
5389
5390         pr_info_client(cl, "auto reconnect after blocklisted\n");
5391         ceph_force_reconnect(fsc->sb);
5392 }
5393
5394 bool check_session_state(struct ceph_mds_session *s)
5395 {
5396         struct ceph_client *cl = s->s_mdsc->fsc->client;
5397
5398         switch (s->s_state) {
5399         case CEPH_MDS_SESSION_OPEN:
5400                 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
5401                         s->s_state = CEPH_MDS_SESSION_HUNG;
5402                         pr_info_client(cl, "mds%d hung\n", s->s_mds);
5403                 }
5404                 break;
5405         case CEPH_MDS_SESSION_CLOSING:
5406         case CEPH_MDS_SESSION_NEW:
5407         case CEPH_MDS_SESSION_RESTARTING:
5408         case CEPH_MDS_SESSION_CLOSED:
5409         case CEPH_MDS_SESSION_REJECTED:
5410                 return false;
5411         }
5412
5413         return true;
5414 }
5415
5416 /*
5417  * If the sequence is incremented while we're waiting on a REQUEST_CLOSE reply,
5418  * then we need to retransmit that request.
5419  */
5420 void inc_session_sequence(struct ceph_mds_session *s)
5421 {
5422         struct ceph_client *cl = s->s_mdsc->fsc->client;
5423
5424         lockdep_assert_held(&s->s_mutex);
5425
5426         s->s_seq++;
5427
5428         if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
5429                 int ret;
5430
5431                 doutc(cl, "resending session close request for mds%d\n", s->s_mds);
5432                 ret = request_close_session(s);
5433                 if (ret < 0)
5434                         pr_err_client(cl, "unable to close session to mds%d: %d\n",
5435                                       s->s_mds, ret);
5436         }
5437 }
5438
5439 /*
5440  * delayed work -- periodically trim expired leases, renew caps with mds.  If
5441  * the @delay parameter is set to 0 or if it's more than 5 secs, the default
5442  * workqueue delay value of 5 secs will be used.
5443  */
5444 static void schedule_delayed(struct ceph_mds_client *mdsc, unsigned long delay)
5445 {
5446         unsigned long max_delay = HZ * 5;
5447
5448         /* 5 secs default delay */
5449         if (!delay || (delay > max_delay))
5450                 delay = max_delay;
5451         schedule_delayed_work(&mdsc->delayed_work,
5452                               round_jiffies_relative(delay));
5453 }
5454
5455 static void delayed_work(struct work_struct *work)
5456 {
5457         struct ceph_mds_client *mdsc =
5458                 container_of(work, struct ceph_mds_client, delayed_work.work);
5459         unsigned long delay;
5460         int renew_interval;
5461         int renew_caps;
5462         int i;
5463
5464         doutc(mdsc->fsc->client, "mdsc delayed_work\n");
5465
5466         if (mdsc->stopping >= CEPH_MDSC_STOPPING_FLUSHED)
5467                 return;
5468
5469         mutex_lock(&mdsc->mutex);
5470         renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
5471         renew_caps = time_after_eq(jiffies, HZ*renew_interval +
5472                                    mdsc->last_renew_caps);
5473         if (renew_caps)
5474                 mdsc->last_renew_caps = jiffies;
5475
5476         for (i = 0; i < mdsc->max_sessions; i++) {
5477                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
5478                 if (!s)
5479                         continue;
5480
5481                 if (!check_session_state(s)) {
5482                         ceph_put_mds_session(s);
5483                         continue;
5484                 }
5485                 mutex_unlock(&mdsc->mutex);
5486
5487                 ceph_flush_session_cap_releases(mdsc, s);
5488
5489                 mutex_lock(&s->s_mutex);
5490                 if (renew_caps)
5491                         send_renew_caps(mdsc, s);
5492                 else
5493                         ceph_con_keepalive(&s->s_con);
5494                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
5495                     s->s_state == CEPH_MDS_SESSION_HUNG)
5496                         ceph_send_cap_releases(mdsc, s);
5497                 mutex_unlock(&s->s_mutex);
5498                 ceph_put_mds_session(s);
5499
5500                 mutex_lock(&mdsc->mutex);
5501         }
5502         mutex_unlock(&mdsc->mutex);
5503
5504         delay = ceph_check_delayed_caps(mdsc);
5505
5506         ceph_queue_cap_reclaim_work(mdsc);
5507
5508         ceph_trim_snapid_map(mdsc);
5509
5510         maybe_recover_session(mdsc);
5511
5512         schedule_delayed(mdsc, delay);
5513 }
5514
5515 int ceph_mdsc_init(struct ceph_fs_client *fsc)
5516
5517 {
5518         struct ceph_mds_client *mdsc;
5519         int err;
5520
5521         mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
5522         if (!mdsc)
5523                 return -ENOMEM;
5524         mdsc->fsc = fsc;
5525         mutex_init(&mdsc->mutex);
5526         mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
5527         if (!mdsc->mdsmap) {
5528                 err = -ENOMEM;
5529                 goto err_mdsc;
5530         }
5531
5532         init_completion(&mdsc->safe_umount_waiters);
5533         spin_lock_init(&mdsc->stopping_lock);
5534         atomic_set(&mdsc->stopping_blockers, 0);
5535         init_completion(&mdsc->stopping_waiter);
5536         atomic64_set(&mdsc->dirty_folios, 0);
5537         init_waitqueue_head(&mdsc->flush_end_wq);
5538         init_waitqueue_head(&mdsc->session_close_wq);
5539         INIT_LIST_HEAD(&mdsc->waiting_for_map);
5540         mdsc->quotarealms_inodes = RB_ROOT;
5541         mutex_init(&mdsc->quotarealms_inodes_mutex);
5542         init_rwsem(&mdsc->snap_rwsem);
5543         mdsc->snap_realms = RB_ROOT;
5544         INIT_LIST_HEAD(&mdsc->snap_empty);
5545         spin_lock_init(&mdsc->snap_empty_lock);
5546         mdsc->request_tree = RB_ROOT;
5547         INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
5548         mdsc->last_renew_caps = jiffies;
5549         INIT_LIST_HEAD(&mdsc->cap_delay_list);
5550 #ifdef CONFIG_DEBUG_FS
5551         INIT_LIST_HEAD(&mdsc->cap_wait_list);
5552 #endif
5553         spin_lock_init(&mdsc->cap_delay_lock);
5554         INIT_LIST_HEAD(&mdsc->cap_unlink_delay_list);
5555         INIT_LIST_HEAD(&mdsc->snap_flush_list);
5556         spin_lock_init(&mdsc->snap_flush_lock);
5557         mdsc->last_cap_flush_tid = 1;
5558         INIT_LIST_HEAD(&mdsc->cap_flush_list);
5559         INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
5560         spin_lock_init(&mdsc->cap_dirty_lock);
5561         init_waitqueue_head(&mdsc->cap_flushing_wq);
5562         INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
5563         INIT_WORK(&mdsc->cap_unlink_work, ceph_cap_unlink_work);
5564         err = ceph_metric_init(&mdsc->metric);
5565         if (err)
5566                 goto err_mdsmap;
5567
5568         spin_lock_init(&mdsc->dentry_list_lock);
5569         INIT_LIST_HEAD(&mdsc->dentry_leases);
5570         INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
5571
5572         ceph_caps_init(mdsc);
5573         ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
5574
5575         spin_lock_init(&mdsc->snapid_map_lock);
5576         mdsc->snapid_map_tree = RB_ROOT;
5577         INIT_LIST_HEAD(&mdsc->snapid_map_lru);
5578
5579         init_rwsem(&mdsc->pool_perm_rwsem);
5580         mdsc->pool_perm_tree = RB_ROOT;
5581
5582         strscpy(mdsc->nodename, utsname()->nodename,
5583                 sizeof(mdsc->nodename));
5584
5585         fsc->mdsc = mdsc;
5586         return 0;
5587
5588 err_mdsmap:
5589         kfree(mdsc->mdsmap);
5590 err_mdsc:
5591         kfree(mdsc);
5592         return err;
5593 }
5594
5595 /*
5596  * Wait for safe replies on open mds requests.  If we time out, drop
5597  * all requests from the tree to avoid dangling dentry refs.
5598  */
5599 static void wait_requests(struct ceph_mds_client *mdsc)
5600 {
5601         struct ceph_client *cl = mdsc->fsc->client;
5602         struct ceph_options *opts = mdsc->fsc->client->options;
5603         struct ceph_mds_request *req;
5604
5605         mutex_lock(&mdsc->mutex);
5606         if (__get_oldest_req(mdsc)) {
5607                 mutex_unlock(&mdsc->mutex);
5608
5609                 doutc(cl, "waiting for requests\n");
5610                 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
5611                                     ceph_timeout_jiffies(opts->mount_timeout));
5612
5613                 /* tear down remaining requests */
5614                 mutex_lock(&mdsc->mutex);
5615                 while ((req = __get_oldest_req(mdsc))) {
5616                         doutc(cl, "timed out on tid %llu\n", req->r_tid);
5617                         list_del_init(&req->r_wait);
5618                         __unregister_request(mdsc, req);
5619                 }
5620         }
5621         mutex_unlock(&mdsc->mutex);
5622         doutc(cl, "done\n");
5623 }
5624
5625 void send_flush_mdlog(struct ceph_mds_session *s)
5626 {
5627         struct ceph_client *cl = s->s_mdsc->fsc->client;
5628         struct ceph_msg *msg;
5629
5630         /*
5631          * Pre-luminous MDS crashes when it sees an unknown session request
5632          */
5633         if (!CEPH_HAVE_FEATURE(s->s_con.peer_features, SERVER_LUMINOUS))
5634                 return;
5635
5636         mutex_lock(&s->s_mutex);
5637         doutc(cl, "request mdlog flush to mds%d (%s)s seq %lld\n",
5638               s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
5639         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_FLUSH_MDLOG,
5640                                       s->s_seq);
5641         if (!msg) {
5642                 pr_err_client(cl, "failed to request mdlog flush to mds%d (%s) seq %lld\n",
5643                               s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
5644         } else {
5645                 ceph_con_send(&s->s_con, msg);
5646         }
5647         mutex_unlock(&s->s_mutex);
5648 }
5649
5650 static int ceph_mds_auth_match(struct ceph_mds_client *mdsc,
5651                                struct ceph_mds_cap_auth *auth,
5652                                const struct cred *cred,
5653                                char *tpath)
5654 {
5655         u32 caller_uid = from_kuid(&init_user_ns, cred->fsuid);
5656         u32 caller_gid = from_kgid(&init_user_ns, cred->fsgid);
5657         struct ceph_client *cl = mdsc->fsc->client;
5658         const char *fs_name = mdsc->fsc->mount_options->mds_namespace;
5659         const char *spath = mdsc->fsc->mount_options->server_path;
5660         bool gid_matched = false;
5661         u32 gid, tlen, len;
5662         int i, j;
5663
5664         doutc(cl, "fsname check fs_name=%s  match.fs_name=%s\n",
5665               fs_name, auth->match.fs_name ? auth->match.fs_name : "");
5666         if (auth->match.fs_name && strcmp(auth->match.fs_name, fs_name)) {
5667                 /* fsname mismatch, try next one */
5668                 return 0;
5669         }
5670
5671         doutc(cl, "match.uid %lld\n", auth->match.uid);
5672         if (auth->match.uid != MDS_AUTH_UID_ANY) {
5673                 if (auth->match.uid != caller_uid)
5674                         return 0;
5675                 if (auth->match.num_gids) {
5676                         for (i = 0; i < auth->match.num_gids; i++) {
5677                                 if (caller_gid == auth->match.gids[i])
5678                                         gid_matched = true;
5679                         }
5680                         if (!gid_matched && cred->group_info->ngroups) {
5681                                 for (i = 0; i < cred->group_info->ngroups; i++) {
5682                                         gid = from_kgid(&init_user_ns,
5683                                                         cred->group_info->gid[i]);
5684                                         for (j = 0; j < auth->match.num_gids; j++) {
5685                                                 if (gid == auth->match.gids[j]) {
5686                                                         gid_matched = true;
5687                                                         break;
5688                                                 }
5689                                         }
5690                                         if (gid_matched)
5691                                                 break;
5692                                 }
5693                         }
5694                         if (!gid_matched)
5695                                 return 0;
5696                 }
5697         }
5698
5699         /* path match */
5700         if (auth->match.path) {
5701                 if (!tpath)
5702                         return 0;
5703
5704                 tlen = strlen(tpath);
5705                 len = strlen(auth->match.path);
5706                 if (len) {
5707                         char *_tpath = tpath;
5708                         bool free_tpath = false;
5709                         int m, n;
5710
5711                         doutc(cl, "server path %s, tpath %s, match.path %s\n",
5712                               spath, tpath, auth->match.path);
5713                         if (spath && (m = strlen(spath)) != 1) {
5714                                 /* mount path + '/' + tpath + an extra space */
5715                                 n = m + 1 + tlen + 1;
5716                                 _tpath = kmalloc(n, GFP_NOFS);
5717                                 if (!_tpath)
5718                                         return -ENOMEM;
5719                                 /* remove the leading '/' */
5720                                 snprintf(_tpath, n, "%s/%s", spath + 1, tpath);
5721                                 free_tpath = true;
5722                                 tlen = strlen(_tpath);
5723                         }
5724
5725                         /*
5726                          * Please note the tailing '/' for match.path has already
5727                          * been removed when parsing.
5728                          *
5729                          * Remove the tailing '/' for the target path.
5730                          */
5731                         while (tlen && _tpath[tlen - 1] == '/') {
5732                                 _tpath[tlen - 1] = '\0';
5733                                 tlen -= 1;
5734                         }
5735                         doutc(cl, "_tpath %s\n", _tpath);
5736
5737                         /*
5738                          * In case first == _tpath && tlen == len:
5739                          *  match.path=/foo  --> /foo _path=/foo     --> match
5740                          *  match.path=/foo/ --> /foo _path=/foo     --> match
5741                          *
5742                          * In case first == _tmatch.path && tlen > len:
5743                          *  match.path=/foo/ --> /foo _path=/foo/    --> match
5744                          *  match.path=/foo  --> /foo _path=/foo/    --> match
5745                          *  match.path=/foo/ --> /foo _path=/foo/d   --> match
5746                          *  match.path=/foo  --> /foo _path=/food    --> mismatch
5747                          *
5748                          * All the other cases                       --> mismatch
5749                          */
5750                         bool path_matched = true;
5751                         char *first = strstr(_tpath, auth->match.path);
5752                         if (first != _tpath ||
5753                             (tlen > len && _tpath[len] != '/')) {
5754                                 path_matched = false;
5755                         }
5756
5757                         if (free_tpath)
5758                                 kfree(_tpath);
5759
5760                         if (!path_matched)
5761                                 return 0;
5762                 }
5763         }
5764
5765         doutc(cl, "matched\n");
5766         return 1;
5767 }
5768
5769 int ceph_mds_check_access(struct ceph_mds_client *mdsc, char *tpath, int mask)
5770 {
5771         const struct cred *cred = get_current_cred();
5772         u32 caller_uid = from_kuid(&init_user_ns, cred->fsuid);
5773         u32 caller_gid = from_kgid(&init_user_ns, cred->fsgid);
5774         struct ceph_mds_cap_auth *rw_perms_s = NULL;
5775         struct ceph_client *cl = mdsc->fsc->client;
5776         bool root_squash_perms = true;
5777         int i, err;
5778
5779         doutc(cl, "tpath '%s', mask %d, caller_uid %d, caller_gid %d\n",
5780               tpath, mask, caller_uid, caller_gid);
5781
5782         for (i = 0; i < mdsc->s_cap_auths_num; i++) {
5783                 struct ceph_mds_cap_auth *s = &mdsc->s_cap_auths[i];
5784
5785                 err = ceph_mds_auth_match(mdsc, s, cred, tpath);
5786                 if (err < 0) {
5787                         put_cred(cred);
5788                         return err;
5789                 } else if (err > 0) {
5790                         /* always follow the last auth caps' permission */
5791                         root_squash_perms = true;
5792                         rw_perms_s = NULL;
5793                         if ((mask & MAY_WRITE) && s->writeable &&
5794                             s->match.root_squash && (!caller_uid || !caller_gid))
5795                                 root_squash_perms = false;
5796
5797                         if (((mask & MAY_WRITE) && !s->writeable) ||
5798                             ((mask & MAY_READ) && !s->readable))
5799                                 rw_perms_s = s;
5800                 }
5801         }
5802
5803         put_cred(cred);
5804
5805         doutc(cl, "root_squash_perms %d, rw_perms_s %p\n", root_squash_perms,
5806               rw_perms_s);
5807         if (root_squash_perms && rw_perms_s == NULL) {
5808                 doutc(cl, "access allowed\n");
5809                 return 0;
5810         }
5811
5812         if (!root_squash_perms) {
5813                 doutc(cl, "root_squash is enabled and user(%d %d) isn't allowed to write",
5814                       caller_uid, caller_gid);
5815         }
5816         if (rw_perms_s) {
5817                 doutc(cl, "mds auth caps readable/writeable %d/%d while request r/w %d/%d",
5818                       rw_perms_s->readable, rw_perms_s->writeable,
5819                       !!(mask & MAY_READ), !!(mask & MAY_WRITE));
5820         }
5821         doutc(cl, "access denied\n");
5822         return -EACCES;
5823 }
5824
5825 /*
5826  * called before mount is ro, and before dentries are torn down.
5827  * (hmm, does this still race with new lookups?)
5828  */
5829 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
5830 {
5831         doutc(mdsc->fsc->client, "begin\n");
5832         mdsc->stopping = CEPH_MDSC_STOPPING_BEGIN;
5833
5834         ceph_mdsc_iterate_sessions(mdsc, send_flush_mdlog, true);
5835         ceph_mdsc_iterate_sessions(mdsc, lock_unlock_session, false);
5836         ceph_flush_dirty_caps(mdsc);
5837         wait_requests(mdsc);
5838
5839         /*
5840          * wait for reply handlers to drop their request refs and
5841          * their inode/dcache refs
5842          */
5843         ceph_msgr_flush();
5844
5845         ceph_cleanup_quotarealms_inodes(mdsc);
5846         doutc(mdsc->fsc->client, "done\n");
5847 }
5848
5849 /*
5850  * flush the mdlog and wait for all write mds requests to flush.
5851  */
5852 static void flush_mdlog_and_wait_mdsc_unsafe_requests(struct ceph_mds_client *mdsc,
5853                                                  u64 want_tid)
5854 {
5855         struct ceph_client *cl = mdsc->fsc->client;
5856         struct ceph_mds_request *req = NULL, *nextreq;
5857         struct ceph_mds_session *last_session = NULL;
5858         struct rb_node *n;
5859
5860         mutex_lock(&mdsc->mutex);
5861         doutc(cl, "want %lld\n", want_tid);
5862 restart:
5863         req = __get_oldest_req(mdsc);
5864         while (req && req->r_tid <= want_tid) {
5865                 /* find next request */
5866                 n = rb_next(&req->r_node);
5867                 if (n)
5868                         nextreq = rb_entry(n, struct ceph_mds_request, r_node);
5869                 else
5870                         nextreq = NULL;
5871                 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
5872                     (req->r_op & CEPH_MDS_OP_WRITE)) {
5873                         struct ceph_mds_session *s = req->r_session;
5874
5875                         if (!s) {
5876                                 req = nextreq;
5877                                 continue;
5878                         }
5879
5880                         /* write op */
5881                         ceph_mdsc_get_request(req);
5882                         if (nextreq)
5883                                 ceph_mdsc_get_request(nextreq);
5884                         s = ceph_get_mds_session(s);
5885                         mutex_unlock(&mdsc->mutex);
5886
5887                         /* send flush mdlog request to MDS */
5888                         if (last_session != s) {
5889                                 send_flush_mdlog(s);
5890                                 ceph_put_mds_session(last_session);
5891                                 last_session = s;
5892                         } else {
5893                                 ceph_put_mds_session(s);
5894                         }
5895                         doutc(cl, "wait on %llu (want %llu)\n",
5896                               req->r_tid, want_tid);
5897                         wait_for_completion(&req->r_safe_completion);
5898
5899                         mutex_lock(&mdsc->mutex);
5900                         ceph_mdsc_put_request(req);
5901                         if (!nextreq)
5902                                 break;  /* next dne before, so we're done! */
5903                         if (RB_EMPTY_NODE(&nextreq->r_node)) {
5904                                 /* next request was removed from tree */
5905                                 ceph_mdsc_put_request(nextreq);
5906                                 goto restart;
5907                         }
5908                         ceph_mdsc_put_request(nextreq);  /* won't go away */
5909                 }
5910                 req = nextreq;
5911         }
5912         mutex_unlock(&mdsc->mutex);
5913         ceph_put_mds_session(last_session);
5914         doutc(cl, "done\n");
5915 }
5916
5917 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
5918 {
5919         struct ceph_client *cl = mdsc->fsc->client;
5920         u64 want_tid, want_flush;
5921
5922         if (READ_ONCE(mdsc->fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN)
5923                 return;
5924
5925         doutc(cl, "sync\n");
5926         mutex_lock(&mdsc->mutex);
5927         want_tid = mdsc->last_tid;
5928         mutex_unlock(&mdsc->mutex);
5929
5930         ceph_flush_dirty_caps(mdsc);
5931         ceph_flush_cap_releases(mdsc);
5932         spin_lock(&mdsc->cap_dirty_lock);
5933         want_flush = mdsc->last_cap_flush_tid;
5934         if (!list_empty(&mdsc->cap_flush_list)) {
5935                 struct ceph_cap_flush *cf =
5936                         list_last_entry(&mdsc->cap_flush_list,
5937                                         struct ceph_cap_flush, g_list);
5938                 cf->wake = true;
5939         }
5940         spin_unlock(&mdsc->cap_dirty_lock);
5941
5942         doutc(cl, "sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
5943
5944         flush_mdlog_and_wait_mdsc_unsafe_requests(mdsc, want_tid);
5945         wait_caps_flush(mdsc, want_flush);
5946 }
5947
5948 /*
5949  * true if all sessions are closed, or we force unmount
5950  */
5951 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
5952 {
5953         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
5954                 return true;
5955         return atomic_read(&mdsc->num_sessions) <= skipped;
5956 }
5957
5958 /*
5959  * called after sb is ro or when metadata corrupted.
5960  */
5961 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
5962 {
5963         struct ceph_options *opts = mdsc->fsc->client->options;
5964         struct ceph_client *cl = mdsc->fsc->client;
5965         struct ceph_mds_session *session;
5966         int i;
5967         int skipped = 0;
5968
5969         doutc(cl, "begin\n");
5970
5971         /* close sessions */
5972         mutex_lock(&mdsc->mutex);
5973         for (i = 0; i < mdsc->max_sessions; i++) {
5974                 session = __ceph_lookup_mds_session(mdsc, i);
5975                 if (!session)
5976                         continue;
5977                 mutex_unlock(&mdsc->mutex);
5978                 mutex_lock(&session->s_mutex);
5979                 if (__close_session(mdsc, session) <= 0)
5980                         skipped++;
5981                 mutex_unlock(&session->s_mutex);
5982                 ceph_put_mds_session(session);
5983                 mutex_lock(&mdsc->mutex);
5984         }
5985         mutex_unlock(&mdsc->mutex);
5986
5987         doutc(cl, "waiting for sessions to close\n");
5988         wait_event_timeout(mdsc->session_close_wq,
5989                            done_closing_sessions(mdsc, skipped),
5990                            ceph_timeout_jiffies(opts->mount_timeout));
5991
5992         /* tear down remaining sessions */
5993         mutex_lock(&mdsc->mutex);
5994         for (i = 0; i < mdsc->max_sessions; i++) {
5995                 if (mdsc->sessions[i]) {
5996                         session = ceph_get_mds_session(mdsc->sessions[i]);
5997                         __unregister_session(mdsc, session);
5998                         mutex_unlock(&mdsc->mutex);
5999                         mutex_lock(&session->s_mutex);
6000                         remove_session_caps(session);
6001                         mutex_unlock(&session->s_mutex);
6002                         ceph_put_mds_session(session);
6003                         mutex_lock(&mdsc->mutex);
6004                 }
6005         }
6006         WARN_ON(!list_empty(&mdsc->cap_delay_list));
6007         mutex_unlock(&mdsc->mutex);
6008
6009         ceph_cleanup_snapid_map(mdsc);
6010         ceph_cleanup_global_and_empty_realms(mdsc);
6011
6012         cancel_work_sync(&mdsc->cap_reclaim_work);
6013         cancel_work_sync(&mdsc->cap_unlink_work);
6014         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
6015
6016         doutc(cl, "done\n");
6017 }
6018
6019 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
6020 {
6021         struct ceph_mds_session *session;
6022         int mds;
6023
6024         doutc(mdsc->fsc->client, "force umount\n");
6025
6026         mutex_lock(&mdsc->mutex);
6027         for (mds = 0; mds < mdsc->max_sessions; mds++) {
6028                 session = __ceph_lookup_mds_session(mdsc, mds);
6029                 if (!session)
6030                         continue;
6031
6032                 if (session->s_state == CEPH_MDS_SESSION_REJECTED)
6033                         __unregister_session(mdsc, session);
6034                 __wake_requests(mdsc, &session->s_waiting);
6035                 mutex_unlock(&mdsc->mutex);
6036
6037                 mutex_lock(&session->s_mutex);
6038                 __close_session(mdsc, session);
6039                 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
6040                         cleanup_session_requests(mdsc, session);
6041                         remove_session_caps(session);
6042                 }
6043                 mutex_unlock(&session->s_mutex);
6044                 ceph_put_mds_session(session);
6045
6046                 mutex_lock(&mdsc->mutex);
6047                 kick_requests(mdsc, mds);
6048         }
6049         __wake_requests(mdsc, &mdsc->waiting_for_map);
6050         mutex_unlock(&mdsc->mutex);
6051 }
6052
6053 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
6054 {
6055         doutc(mdsc->fsc->client, "stop\n");
6056         /*
6057          * Make sure the delayed work stopped before releasing
6058          * the resources.
6059          *
6060          * Because the cancel_delayed_work_sync() will only
6061          * guarantee that the work finishes executing. But the
6062          * delayed work will re-arm itself again after that.
6063          */
6064         flush_delayed_work(&mdsc->delayed_work);
6065
6066         if (mdsc->mdsmap)
6067                 ceph_mdsmap_destroy(mdsc->mdsmap);
6068         kfree(mdsc->sessions);
6069         ceph_caps_finalize(mdsc);
6070
6071         if (mdsc->s_cap_auths) {
6072                 int i;
6073
6074                 for (i = 0; i < mdsc->s_cap_auths_num; i++) {
6075                         kfree(mdsc->s_cap_auths[i].match.gids);
6076                         kfree(mdsc->s_cap_auths[i].match.path);
6077                         kfree(mdsc->s_cap_auths[i].match.fs_name);
6078                 }
6079                 kfree(mdsc->s_cap_auths);
6080         }
6081
6082         ceph_pool_perm_destroy(mdsc);
6083 }
6084
6085 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
6086 {
6087         struct ceph_mds_client *mdsc = fsc->mdsc;
6088         doutc(fsc->client, "%p\n", mdsc);
6089
6090         if (!mdsc)
6091                 return;
6092
6093         /* flush out any connection work with references to us */
6094         ceph_msgr_flush();
6095
6096         ceph_mdsc_stop(mdsc);
6097
6098         ceph_metric_destroy(&mdsc->metric);
6099
6100         fsc->mdsc = NULL;
6101         kfree(mdsc);
6102         doutc(fsc->client, "%p done\n", mdsc);
6103 }
6104
6105 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
6106 {
6107         struct ceph_fs_client *fsc = mdsc->fsc;
6108         struct ceph_client *cl = fsc->client;
6109         const char *mds_namespace = fsc->mount_options->mds_namespace;
6110         void *p = msg->front.iov_base;
6111         void *end = p + msg->front.iov_len;
6112         u32 epoch;
6113         u32 num_fs;
6114         u32 mount_fscid = (u32)-1;
6115         int err = -EINVAL;
6116
6117         ceph_decode_need(&p, end, sizeof(u32), bad);
6118         epoch = ceph_decode_32(&p);
6119
6120         doutc(cl, "epoch %u\n", epoch);
6121
6122         /* struct_v, struct_cv, map_len, epoch, legacy_client_fscid */
6123         ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 3, bad);
6124
6125         ceph_decode_32_safe(&p, end, num_fs, bad);
6126         while (num_fs-- > 0) {
6127                 void *info_p, *info_end;
6128                 u32 info_len;
6129                 u32 fscid, namelen;
6130
6131                 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
6132                 p += 2;         // info_v, info_cv
6133                 info_len = ceph_decode_32(&p);
6134                 ceph_decode_need(&p, end, info_len, bad);
6135                 info_p = p;
6136                 info_end = p + info_len;
6137                 p = info_end;
6138
6139                 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
6140                 fscid = ceph_decode_32(&info_p);
6141                 namelen = ceph_decode_32(&info_p);
6142                 ceph_decode_need(&info_p, info_end, namelen, bad);
6143
6144                 if (mds_namespace &&
6145                     strlen(mds_namespace) == namelen &&
6146                     !strncmp(mds_namespace, (char *)info_p, namelen)) {
6147                         mount_fscid = fscid;
6148                         break;
6149                 }
6150         }
6151
6152         ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
6153         if (mount_fscid != (u32)-1) {
6154                 fsc->client->monc.fs_cluster_id = mount_fscid;
6155                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
6156                                    0, true);
6157                 ceph_monc_renew_subs(&fsc->client->monc);
6158         } else {
6159                 err = -ENOENT;
6160                 goto err_out;
6161         }
6162         return;
6163
6164 bad:
6165         pr_err_client(cl, "error decoding fsmap %d. Shutting down mount.\n",
6166                       err);
6167         ceph_umount_begin(mdsc->fsc->sb);
6168         ceph_msg_dump(msg);
6169 err_out:
6170         mutex_lock(&mdsc->mutex);
6171         mdsc->mdsmap_err = err;
6172         __wake_requests(mdsc, &mdsc->waiting_for_map);
6173         mutex_unlock(&mdsc->mutex);
6174 }
6175
6176 /*
6177  * handle mds map update.
6178  */
6179 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
6180 {
6181         struct ceph_client *cl = mdsc->fsc->client;
6182         u32 epoch;
6183         u32 maplen;
6184         void *p = msg->front.iov_base;
6185         void *end = p + msg->front.iov_len;
6186         struct ceph_mdsmap *newmap, *oldmap;
6187         struct ceph_fsid fsid;
6188         int err = -EINVAL;
6189
6190         ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
6191         ceph_decode_copy(&p, &fsid, sizeof(fsid));
6192         if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
6193                 return;
6194         epoch = ceph_decode_32(&p);
6195         maplen = ceph_decode_32(&p);
6196         doutc(cl, "epoch %u len %d\n", epoch, (int)maplen);
6197
6198         /* do we need it? */
6199         mutex_lock(&mdsc->mutex);
6200         if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
6201                 doutc(cl, "epoch %u <= our %u\n", epoch, mdsc->mdsmap->m_epoch);
6202                 mutex_unlock(&mdsc->mutex);
6203                 return;
6204         }
6205
6206         newmap = ceph_mdsmap_decode(mdsc, &p, end, ceph_msgr2(mdsc->fsc->client));
6207         if (IS_ERR(newmap)) {
6208                 err = PTR_ERR(newmap);
6209                 goto bad_unlock;
6210         }
6211
6212         /* swap into place */
6213         if (mdsc->mdsmap) {
6214                 oldmap = mdsc->mdsmap;
6215                 mdsc->mdsmap = newmap;
6216                 check_new_map(mdsc, newmap, oldmap);
6217                 ceph_mdsmap_destroy(oldmap);
6218         } else {
6219                 mdsc->mdsmap = newmap;  /* first mds map */
6220         }
6221         mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
6222                                         MAX_LFS_FILESIZE);
6223
6224         __wake_requests(mdsc, &mdsc->waiting_for_map);
6225         ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
6226                           mdsc->mdsmap->m_epoch);
6227
6228         mutex_unlock(&mdsc->mutex);
6229         schedule_delayed(mdsc, 0);
6230         return;
6231
6232 bad_unlock:
6233         mutex_unlock(&mdsc->mutex);
6234 bad:
6235         pr_err_client(cl, "error decoding mdsmap %d. Shutting down mount.\n",
6236                       err);
6237         ceph_umount_begin(mdsc->fsc->sb);
6238         ceph_msg_dump(msg);
6239         return;
6240 }
6241
6242 static struct ceph_connection *mds_get_con(struct ceph_connection *con)
6243 {
6244         struct ceph_mds_session *s = con->private;
6245
6246         if (ceph_get_mds_session(s))
6247                 return con;
6248         return NULL;
6249 }
6250
6251 static void mds_put_con(struct ceph_connection *con)
6252 {
6253         struct ceph_mds_session *s = con->private;
6254
6255         ceph_put_mds_session(s);
6256 }
6257
6258 /*
6259  * if the client is unresponsive for long enough, the mds will kill
6260  * the session entirely.
6261  */
6262 static void mds_peer_reset(struct ceph_connection *con)
6263 {
6264         struct ceph_mds_session *s = con->private;
6265         struct ceph_mds_client *mdsc = s->s_mdsc;
6266
6267         pr_warn_client(mdsc->fsc->client, "mds%d closed our session\n",
6268                        s->s_mds);
6269         if (READ_ONCE(mdsc->fsc->mount_state) != CEPH_MOUNT_FENCE_IO &&
6270             ceph_mdsmap_get_state(mdsc->mdsmap, s->s_mds) >= CEPH_MDS_STATE_RECONNECT)
6271                 send_mds_reconnect(mdsc, s);
6272 }
6273
6274 static void mds_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
6275 {
6276         struct ceph_mds_session *s = con->private;
6277         struct ceph_mds_client *mdsc = s->s_mdsc;
6278         struct ceph_client *cl = mdsc->fsc->client;
6279         int type = le16_to_cpu(msg->hdr.type);
6280
6281         mutex_lock(&mdsc->mutex);
6282         if (__verify_registered_session(mdsc, s) < 0) {
6283                 mutex_unlock(&mdsc->mutex);
6284                 goto out;
6285         }
6286         mutex_unlock(&mdsc->mutex);
6287
6288         switch (type) {
6289         case CEPH_MSG_MDS_MAP:
6290                 ceph_mdsc_handle_mdsmap(mdsc, msg);
6291                 break;
6292         case CEPH_MSG_FS_MAP_USER:
6293                 ceph_mdsc_handle_fsmap(mdsc, msg);
6294                 break;
6295         case CEPH_MSG_CLIENT_SESSION:
6296                 handle_session(s, msg);
6297                 break;
6298         case CEPH_MSG_CLIENT_REPLY:
6299                 handle_reply(s, msg);
6300                 break;
6301         case CEPH_MSG_CLIENT_REQUEST_FORWARD:
6302                 handle_forward(mdsc, s, msg);
6303                 break;
6304         case CEPH_MSG_CLIENT_CAPS:
6305                 ceph_handle_caps(s, msg);
6306                 break;
6307         case CEPH_MSG_CLIENT_SNAP:
6308                 ceph_handle_snap(mdsc, s, msg);
6309                 break;
6310         case CEPH_MSG_CLIENT_LEASE:
6311                 handle_lease(mdsc, s, msg);
6312                 break;
6313         case CEPH_MSG_CLIENT_QUOTA:
6314                 ceph_handle_quota(mdsc, s, msg);
6315                 break;
6316
6317         default:
6318                 pr_err_client(cl, "received unknown message type %d %s\n",
6319                               type, ceph_msg_type_name(type));
6320         }
6321 out:
6322         ceph_msg_put(msg);
6323 }
6324
6325 /*
6326  * authentication
6327  */
6328
6329 /*
6330  * Note: returned pointer is the address of a structure that's
6331  * managed separately.  Caller must *not* attempt to free it.
6332  */
6333 static struct ceph_auth_handshake *
6334 mds_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
6335 {
6336         struct ceph_mds_session *s = con->private;
6337         struct ceph_mds_client *mdsc = s->s_mdsc;
6338         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
6339         struct ceph_auth_handshake *auth = &s->s_auth;
6340         int ret;
6341
6342         ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
6343                                          force_new, proto, NULL, NULL);
6344         if (ret)
6345                 return ERR_PTR(ret);
6346
6347         return auth;
6348 }
6349
6350 static int mds_add_authorizer_challenge(struct ceph_connection *con,
6351                                     void *challenge_buf, int challenge_buf_len)
6352 {
6353         struct ceph_mds_session *s = con->private;
6354         struct ceph_mds_client *mdsc = s->s_mdsc;
6355         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
6356
6357         return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
6358                                             challenge_buf, challenge_buf_len);
6359 }
6360
6361 static int mds_verify_authorizer_reply(struct ceph_connection *con)
6362 {
6363         struct ceph_mds_session *s = con->private;
6364         struct ceph_mds_client *mdsc = s->s_mdsc;
6365         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
6366         struct ceph_auth_handshake *auth = &s->s_auth;
6367
6368         return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
6369                 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
6370                 NULL, NULL, NULL, NULL);
6371 }
6372
6373 static int mds_invalidate_authorizer(struct ceph_connection *con)
6374 {
6375         struct ceph_mds_session *s = con->private;
6376         struct ceph_mds_client *mdsc = s->s_mdsc;
6377         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
6378
6379         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
6380
6381         return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
6382 }
6383
6384 static int mds_get_auth_request(struct ceph_connection *con,
6385                                 void *buf, int *buf_len,
6386                                 void **authorizer, int *authorizer_len)
6387 {
6388         struct ceph_mds_session *s = con->private;
6389         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
6390         struct ceph_auth_handshake *auth = &s->s_auth;
6391         int ret;
6392
6393         ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
6394                                        buf, buf_len);
6395         if (ret)
6396                 return ret;
6397
6398         *authorizer = auth->authorizer_buf;
6399         *authorizer_len = auth->authorizer_buf_len;
6400         return 0;
6401 }
6402
6403 static int mds_handle_auth_reply_more(struct ceph_connection *con,
6404                                       void *reply, int reply_len,
6405                                       void *buf, int *buf_len,
6406                                       void **authorizer, int *authorizer_len)
6407 {
6408         struct ceph_mds_session *s = con->private;
6409         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
6410         struct ceph_auth_handshake *auth = &s->s_auth;
6411         int ret;
6412
6413         ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
6414                                               buf, buf_len);
6415         if (ret)
6416                 return ret;
6417
6418         *authorizer = auth->authorizer_buf;
6419         *authorizer_len = auth->authorizer_buf_len;
6420         return 0;
6421 }
6422
6423 static int mds_handle_auth_done(struct ceph_connection *con,
6424                                 u64 global_id, void *reply, int reply_len,
6425                                 u8 *session_key, int *session_key_len,
6426                                 u8 *con_secret, int *con_secret_len)
6427 {
6428         struct ceph_mds_session *s = con->private;
6429         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
6430         struct ceph_auth_handshake *auth = &s->s_auth;
6431
6432         return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
6433                                                session_key, session_key_len,
6434                                                con_secret, con_secret_len);
6435 }
6436
6437 static int mds_handle_auth_bad_method(struct ceph_connection *con,
6438                                       int used_proto, int result,
6439                                       const int *allowed_protos, int proto_cnt,
6440                                       const int *allowed_modes, int mode_cnt)
6441 {
6442         struct ceph_mds_session *s = con->private;
6443         struct ceph_mon_client *monc = &s->s_mdsc->fsc->client->monc;
6444         int ret;
6445
6446         if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_MDS,
6447                                             used_proto, result,
6448                                             allowed_protos, proto_cnt,
6449                                             allowed_modes, mode_cnt)) {
6450                 ret = ceph_monc_validate_auth(monc);
6451                 if (ret)
6452                         return ret;
6453         }
6454
6455         return -EACCES;
6456 }
6457
6458 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
6459                                 struct ceph_msg_header *hdr, int *skip)
6460 {
6461         struct ceph_msg *msg;
6462         int type = (int) le16_to_cpu(hdr->type);
6463         int front_len = (int) le32_to_cpu(hdr->front_len);
6464
6465         if (con->in_msg)
6466                 return con->in_msg;
6467
6468         *skip = 0;
6469         msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
6470         if (!msg) {
6471                 pr_err("unable to allocate msg type %d len %d\n",
6472                        type, front_len);
6473                 return NULL;
6474         }
6475
6476         return msg;
6477 }
6478
6479 static int mds_sign_message(struct ceph_msg *msg)
6480 {
6481        struct ceph_mds_session *s = msg->con->private;
6482        struct ceph_auth_handshake *auth = &s->s_auth;
6483
6484        return ceph_auth_sign_message(auth, msg);
6485 }
6486
6487 static int mds_check_message_signature(struct ceph_msg *msg)
6488 {
6489        struct ceph_mds_session *s = msg->con->private;
6490        struct ceph_auth_handshake *auth = &s->s_auth;
6491
6492        return ceph_auth_check_message_signature(auth, msg);
6493 }
6494
6495 static const struct ceph_connection_operations mds_con_ops = {
6496         .get = mds_get_con,
6497         .put = mds_put_con,
6498         .alloc_msg = mds_alloc_msg,
6499         .dispatch = mds_dispatch,
6500         .peer_reset = mds_peer_reset,
6501         .get_authorizer = mds_get_authorizer,
6502         .add_authorizer_challenge = mds_add_authorizer_challenge,
6503         .verify_authorizer_reply = mds_verify_authorizer_reply,
6504         .invalidate_authorizer = mds_invalidate_authorizer,
6505         .sign_message = mds_sign_message,
6506         .check_message_signature = mds_check_message_signature,
6507         .get_auth_request = mds_get_auth_request,
6508         .handle_auth_reply_more = mds_handle_auth_reply_more,
6509         .handle_auth_done = mds_handle_auth_done,
6510         .handle_auth_bad_method = mds_handle_auth_bad_method,
6511 };
6512
6513 /* eof */