root/fs/ceph/mds_client.c

DEFINITIONS

1. parse_reply_info_quota
2. parse_reply_info_in
3. parse_reply_info_dir
4. parse_reply_info_lease
5. parse_reply_info_trace
6. parse_reply_info_readdir
7. parse_reply_info_filelock
8. parse_reply_info_create
9. parse_reply_info_extra
10. parse_reply_info
11. destroy_reply_info
12. ceph_session_state_name
13. get_session
14. ceph_put_mds_session
15. __ceph_lookup_mds_session
16. __have_session
17. __verify_registered_session
18. register_session
19. __unregister_session
20. put_request_session
21. ceph_mdsc_release_request
22. DEFINE_RB_FUNCS
23. __register_request
24. __unregister_request
25. get_nonsnap_parent
26. __choose_mds
27. create_session_msg
28. encode_supported_features
29. create_session_open_msg
30. __open_session
31. __open_export_target_session
32. ceph_mdsc_open_export_target_session
33. __open_export_target_sessions
34. ceph_mdsc_open_export_target_sessions
35. detach_cap_releases
36. dispose_cap_releases
37. cleanup_session_requests
38. ceph_iterate_session_caps
39. remove_session_caps_cb
40. remove_session_caps
41. wake_up_session_cb
42. wake_up_session_caps
43. send_renew_caps
44. send_flushmsg_ack
45. renewed_caps
46. request_close_session
47. __close_session
48. drop_negative_children
49. trim_caps_cb
50. ceph_trim_caps
51. check_caps_flush
52. wait_caps_flush
53. ceph_send_cap_releases
54. ceph_cap_release_work
55. ceph_flush_cap_releases
56. __ceph_queue_cap_release
57. ceph_cap_reclaim_work
58. ceph_queue_cap_reclaim_work
59. ceph_reclaim_caps_nr
60. ceph_alloc_readdir_reply_buffer
61. ceph_mdsc_create_request
62. __get_oldest_req
63. __get_oldest_tid
64. ceph_mdsc_build_path
65. build_dentry_path
66. build_inode_path
67. set_request_path_attr
68. create_request_message
69. complete_request
70. __prepare_send_request
71. __do_request
72. __wake_requests
73. kick_requests
74. ceph_mdsc_submit_request
75. ceph_mdsc_wait_request
76. ceph_mdsc_do_request
77. ceph_invalidate_dir_request
78. handle_reply
79. handle_forward
80. __decode_session_metadata
81. handle_session
82. replay_unsafe_requests
83. send_reconnect_partial
84. encode_caps_cb
85. encode_snap_realms
86. send_mds_reconnect
87. check_new_map
88. __ceph_mdsc_drop_dentry_lease
89. handle_lease
90. ceph_mdsc_lease_send_msg
91. lock_unlock_sessions
92. maybe_recover_session
93. schedule_delayed
94. delayed_work
95. ceph_mdsc_init
96. wait_requests
97. ceph_mdsc_pre_umount
98. wait_unsafe_requests
99. ceph_mdsc_sync
100. done_closing_sessions
101. ceph_mdsc_close_sessions
102. ceph_mdsc_force_umount
103. ceph_mdsc_stop
104. ceph_mdsc_destroy
105. ceph_mdsc_handle_fsmap
106. ceph_mdsc_handle_mdsmap
107. con_get
108. con_put
109. peer_reset
110. dispatch
111. get_authorizer
112. add_authorizer_challenge
113. verify_authorizer_reply
114. invalidate_authorizer
115. mds_alloc_msg
116. mds_sign_message
117. mds_check_message_signature

   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 
  13 #include "super.h"
  14 #include "mds_client.h"
  15 
  16 #include <linux/ceph/ceph_features.h>
  17 #include <linux/ceph/messenger.h>
  18 #include <linux/ceph/decode.h>
  19 #include <linux/ceph/pagelist.h>
  20 #include <linux/ceph/auth.h>
  21 #include <linux/ceph/debugfs.h>
  22 
  23 #define RECONNECT_MAX_SIZE (INT_MAX - PAGE_SIZE)
  24 
  25 /*
  26  * A cluster of MDS (metadata server) daemons is responsible for
  27  * managing the file system namespace (the directory hierarchy and
  28  * inodes) and for coordinating shared access to storage.  Metadata is
  29  * partitioning hierarchically across a number of servers, and that
  30  * partition varies over time as the cluster adjusts the distribution
  31  * in order to balance load.
  32  *
  33  * The MDS client is primarily responsible to managing synchronous
  34  * metadata requests for operations like open, unlink, and so forth.
  35  * If there is a MDS failure, we find out about it when we (possibly
  36  * request and) receive a new MDS map, and can resubmit affected
  37  * requests.
  38  *
  39  * For the most part, though, we take advantage of a lossless
  40  * communications channel to the MDS, and do not need to worry about
  41  * timing out or resubmitting requests.
  42  *
  43  * We maintain a stateful "session" with each MDS we interact with.
  44  * Within each session, we sent periodic heartbeat messages to ensure
  45  * any capabilities or leases we have been issues remain valid.  If
  46  * the session times out and goes stale, our leases and capabilities
  47  * are no longer valid.
  48  */
  49 
  50 struct ceph_reconnect_state {
  51         struct ceph_mds_session *session;
  52         int nr_caps, nr_realms;
  53         struct ceph_pagelist *pagelist;
  54         unsigned msg_version;
  55         bool allow_multi;
  56 };
  57 
  58 static void __wake_requests(struct ceph_mds_client *mdsc,
  59                             struct list_head *head);
  60 static void ceph_cap_release_work(struct work_struct *work);
  61 static void ceph_cap_reclaim_work(struct work_struct *work);
  62 
  63 static const struct ceph_connection_operations mds_con_ops;
  64 
  65 
  66 /*
  67  * mds reply parsing
  68  */
  69 
  70 static int parse_reply_info_quota(void **p, void *end,
  71                                   struct ceph_mds_reply_info_in *info)
  72 {
  73         u8 struct_v, struct_compat;
  74         u32 struct_len;
  75 
  76         ceph_decode_8_safe(p, end, struct_v, bad);
  77         ceph_decode_8_safe(p, end, struct_compat, bad);
  78         /* struct_v is expected to be >= 1. we only
  79          * understand encoding with struct_compat == 1. */
  80         if (!struct_v || struct_compat != 1)
  81                 goto bad;
  82         ceph_decode_32_safe(p, end, struct_len, bad);
  83         ceph_decode_need(p, end, struct_len, bad);
  84         end = *p + struct_len;
  85         ceph_decode_64_safe(p, end, info->max_bytes, bad);
  86         ceph_decode_64_safe(p, end, info->max_files, bad);
  87         *p = end;
  88         return 0;
  89 bad:
  90         return -EIO;
  91 }
  92 
  93 /*
  94  * parse individual inode info
  95  */
  96 static int parse_reply_info_in(void **p, void *end,
  97                                struct ceph_mds_reply_info_in *info,
  98                                u64 features)
  99 {
 100         int err = 0;
 101         u8 struct_v = 0;
 102 
 103         if (features == (u64)-1) {
 104                 u32 struct_len;
 105                 u8 struct_compat;
 106                 ceph_decode_8_safe(p, end, struct_v, bad);
 107                 ceph_decode_8_safe(p, end, struct_compat, bad);
 108                 /* struct_v is expected to be >= 1. we only understand
 109                  * encoding with struct_compat == 1. */
 110                 if (!struct_v || struct_compat != 1)
 111                         goto bad;
 112                 ceph_decode_32_safe(p, end, struct_len, bad);
 113                 ceph_decode_need(p, end, struct_len, bad);
 114                 end = *p + struct_len;
 115         }
 116 
 117         ceph_decode_need(p, end, sizeof(struct ceph_mds_reply_inode), bad);
 118         info->in = *p;
 119         *p += sizeof(struct ceph_mds_reply_inode) +
 120                 sizeof(*info->in->fragtree.splits) *
 121                 le32_to_cpu(info->in->fragtree.nsplits);
 122 
 123         ceph_decode_32_safe(p, end, info->symlink_len, bad);
 124         ceph_decode_need(p, end, info->symlink_len, bad);
 125         info->symlink = *p;
 126         *p += info->symlink_len;
 127 
 128         ceph_decode_copy_safe(p, end, &info->dir_layout,
 129                               sizeof(info->dir_layout), bad);
 130         ceph_decode_32_safe(p, end, info->xattr_len, bad);
 131         ceph_decode_need(p, end, info->xattr_len, bad);
 132         info->xattr_data = *p;
 133         *p += info->xattr_len;
 134 
 135         if (features == (u64)-1) {
 136                 /* inline data */
 137                 ceph_decode_64_safe(p, end, info->inline_version, bad);
 138                 ceph_decode_32_safe(p, end, info->inline_len, bad);
 139                 ceph_decode_need(p, end, info->inline_len, bad);
 140                 info->inline_data = *p;
 141                 *p += info->inline_len;
 142                 /* quota */
 143                 err = parse_reply_info_quota(p, end, info);
 144                 if (err < 0)
 145                         goto out_bad;
 146                 /* pool namespace */
 147                 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
 148                 if (info->pool_ns_len > 0) {
 149                         ceph_decode_need(p, end, info->pool_ns_len, bad);
 150                         info->pool_ns_data = *p;
 151                         *p += info->pool_ns_len;
 152                 }
 153 
 154                 /* btime */
 155                 ceph_decode_need(p, end, sizeof(info->btime), bad);
 156                 ceph_decode_copy(p, &info->btime, sizeof(info->btime));
 157 
 158                 /* change attribute */
 159                 ceph_decode_64_safe(p, end, info->change_attr, bad);
 160 
 161                 /* dir pin */
 162                 if (struct_v >= 2) {
 163                         ceph_decode_32_safe(p, end, info->dir_pin, bad);
 164                 } else {
 165                         info->dir_pin = -ENODATA;
 166                 }
 167 
 168                 /* snapshot birth time, remains zero for v<=2 */
 169                 if (struct_v >= 3) {
 170                         ceph_decode_need(p, end, sizeof(info->snap_btime), bad);
 171                         ceph_decode_copy(p, &info->snap_btime,
 172                                          sizeof(info->snap_btime));
 173                 } else {
 174                         memset(&info->snap_btime, 0, sizeof(info->snap_btime));
 175                 }
 176 
 177                 *p = end;
 178         } else {
 179                 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
 180                         ceph_decode_64_safe(p, end, info->inline_version, bad);
 181                         ceph_decode_32_safe(p, end, info->inline_len, bad);
 182                         ceph_decode_need(p, end, info->inline_len, bad);
 183                         info->inline_data = *p;
 184                         *p += info->inline_len;
 185                 } else
 186                         info->inline_version = CEPH_INLINE_NONE;
 187 
 188                 if (features & CEPH_FEATURE_MDS_QUOTA) {
 189                         err = parse_reply_info_quota(p, end, info);
 190                         if (err < 0)
 191                                 goto out_bad;
 192                 } else {
 193                         info->max_bytes = 0;
 194                         info->max_files = 0;
 195                 }
 196 
 197                 info->pool_ns_len = 0;
 198                 info->pool_ns_data = NULL;
 199                 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
 200                         ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
 201                         if (info->pool_ns_len > 0) {
 202                                 ceph_decode_need(p, end, info->pool_ns_len, bad);
 203                                 info->pool_ns_data = *p;
 204                                 *p += info->pool_ns_len;
 205                         }
 206                 }
 207 
 208                 if (features & CEPH_FEATURE_FS_BTIME) {
 209                         ceph_decode_need(p, end, sizeof(info->btime), bad);
 210                         ceph_decode_copy(p, &info->btime, sizeof(info->btime));
 211                         ceph_decode_64_safe(p, end, info->change_attr, bad);
 212                 }
 213 
 214                 info->dir_pin = -ENODATA;
 215                 /* info->snap_btime remains zero */
 216         }
 217         return 0;
 218 bad:
 219         err = -EIO;
 220 out_bad:
 221         return err;
 222 }
 223 
 224 static int parse_reply_info_dir(void **p, void *end,
 225                                 struct ceph_mds_reply_dirfrag **dirfrag,
 226                                 u64 features)
 227 {
 228         if (features == (u64)-1) {
 229                 u8 struct_v, struct_compat;
 230                 u32 struct_len;
 231                 ceph_decode_8_safe(p, end, struct_v, bad);
 232                 ceph_decode_8_safe(p, end, struct_compat, bad);
 233                 /* struct_v is expected to be >= 1. we only understand
 234                  * encoding whose struct_compat == 1. */
 235                 if (!struct_v || struct_compat != 1)
 236                         goto bad;
 237                 ceph_decode_32_safe(p, end, struct_len, bad);
 238                 ceph_decode_need(p, end, struct_len, bad);
 239                 end = *p + struct_len;
 240         }
 241 
 242         ceph_decode_need(p, end, sizeof(**dirfrag), bad);
 243         *dirfrag = *p;
 244         *p += sizeof(**dirfrag) + sizeof(u32) * le32_to_cpu((*dirfrag)->ndist);
 245         if (unlikely(*p > end))
 246                 goto bad;
 247         if (features == (u64)-1)
 248                 *p = end;
 249         return 0;
 250 bad:
 251         return -EIO;
 252 }
 253 
 254 static int parse_reply_info_lease(void **p, void *end,
 255                                   struct ceph_mds_reply_lease **lease,
 256                                   u64 features)
 257 {
 258         if (features == (u64)-1) {
 259                 u8 struct_v, struct_compat;
 260                 u32 struct_len;
 261                 ceph_decode_8_safe(p, end, struct_v, bad);
 262                 ceph_decode_8_safe(p, end, struct_compat, bad);
 263                 /* struct_v is expected to be >= 1. we only understand
 264                  * encoding whose struct_compat == 1. */
 265                 if (!struct_v || struct_compat != 1)
 266                         goto bad;
 267                 ceph_decode_32_safe(p, end, struct_len, bad);
 268                 ceph_decode_need(p, end, struct_len, bad);
 269                 end = *p + struct_len;
 270         }
 271 
 272         ceph_decode_need(p, end, sizeof(**lease), bad);
 273         *lease = *p;
 274         *p += sizeof(**lease);
 275         if (features == (u64)-1)
 276                 *p = end;
 277         return 0;
 278 bad:
 279         return -EIO;
 280 }
 281 
 282 /*
 283  * parse a normal reply, which may contain a (dir+)dentry and/or a
 284  * target inode.
 285  */
 286 static int parse_reply_info_trace(void **p, void *end,
 287                                   struct ceph_mds_reply_info_parsed *info,
 288                                   u64 features)
 289 {
 290         int err;
 291 
 292         if (info->head->is_dentry) {
 293                 err = parse_reply_info_in(p, end, &info->diri, features);
 294                 if (err < 0)
 295                         goto out_bad;
 296 
 297                 err = parse_reply_info_dir(p, end, &info->dirfrag, features);
 298                 if (err < 0)
 299                         goto out_bad;
 300 
 301                 ceph_decode_32_safe(p, end, info->dname_len, bad);
 302                 ceph_decode_need(p, end, info->dname_len, bad);
 303                 info->dname = *p;
 304                 *p += info->dname_len;
 305 
 306                 err = parse_reply_info_lease(p, end, &info->dlease, features);
 307                 if (err < 0)
 308                         goto out_bad;
 309         }
 310 
 311         if (info->head->is_target) {
 312                 err = parse_reply_info_in(p, end, &info->targeti, features);
 313                 if (err < 0)
 314                         goto out_bad;
 315         }
 316 
 317         if (unlikely(*p != end))
 318                 goto bad;
 319         return 0;
 320 
 321 bad:
 322         err = -EIO;
 323 out_bad:
 324         pr_err("problem parsing mds trace %d\n", err);
 325         return err;
 326 }
 327 
 328 /*
 329  * parse readdir results
 330  */
 331 static int parse_reply_info_readdir(void **p, void *end,
 332                                 struct ceph_mds_reply_info_parsed *info,
 333                                 u64 features)
 334 {
 335         u32 num, i = 0;
 336         int err;
 337 
 338         err = parse_reply_info_dir(p, end, &info->dir_dir, features);
 339         if (err < 0)
 340                 goto out_bad;
 341 
 342         ceph_decode_need(p, end, sizeof(num) + 2, bad);
 343         num = ceph_decode_32(p);
 344         {
 345                 u16 flags = ceph_decode_16(p);
 346                 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
 347                 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
 348                 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
 349                 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
 350         }
 351         if (num == 0)
 352                 goto done;
 353 
 354         BUG_ON(!info->dir_entries);
 355         if ((unsigned long)(info->dir_entries + num) >
 356             (unsigned long)info->dir_entries + info->dir_buf_size) {
 357                 pr_err("dir contents are larger than expected\n");
 358                 WARN_ON(1);
 359                 goto bad;
 360         }
 361 
 362         info->dir_nr = num;
 363         while (num) {
 364                 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
 365                 /* dentry */
 366                 ceph_decode_32_safe(p, end, rde->name_len, bad);
 367                 ceph_decode_need(p, end, rde->name_len, bad);
 368                 rde->name = *p;
 369                 *p += rde->name_len;
 370                 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
 371 
 372                 /* dentry lease */
 373                 err = parse_reply_info_lease(p, end, &rde->lease, features);
 374                 if (err)
 375                         goto out_bad;
 376                 /* inode */
 377                 err = parse_reply_info_in(p, end, &rde->inode, features);
 378                 if (err < 0)
 379                         goto out_bad;
 380                 /* ceph_readdir_prepopulate() will update it */
 381                 rde->offset = 0;
 382                 i++;
 383                 num--;
 384         }
 385 
 386 done:
 387         /* Skip over any unrecognized fields */
 388         *p = end;
 389         return 0;
 390 
 391 bad:
 392         err = -EIO;
 393 out_bad:
 394         pr_err("problem parsing dir contents %d\n", err);
 395         return err;
 396 }
 397 
 398 /*
 399  * parse fcntl F_GETLK results
 400  */
 401 static int parse_reply_info_filelock(void **p, void *end,
 402                                      struct ceph_mds_reply_info_parsed *info,
 403                                      u64 features)
 404 {
 405         if (*p + sizeof(*info->filelock_reply) > end)
 406                 goto bad;
 407 
 408         info->filelock_reply = *p;
 409 
 410         /* Skip over any unrecognized fields */
 411         *p = end;
 412         return 0;
 413 bad:
 414         return -EIO;
 415 }
 416 
 417 /*
 418  * parse create results
 419  */
 420 static int parse_reply_info_create(void **p, void *end,
 421                                   struct ceph_mds_reply_info_parsed *info,
 422                                   u64 features)
 423 {
 424         if (features == (u64)-1 ||
 425             (features & CEPH_FEATURE_REPLY_CREATE_INODE)) {
 426                 /* Malformed reply? */
 427                 if (*p == end) {
 428                         info->has_create_ino = false;
 429                 } else {
 430                         info->has_create_ino = true;
 431                         ceph_decode_64_safe(p, end, info->ino, bad);
 432                 }
 433         } else {
 434                 if (*p != end)
 435                         goto bad;
 436         }
 437 
 438         /* Skip over any unrecognized fields */
 439         *p = end;
 440         return 0;
 441 bad:
 442         return -EIO;
 443 }
 444 
 445 /*
 446  * parse extra results
 447  */
 448 static int parse_reply_info_extra(void **p, void *end,
 449                                   struct ceph_mds_reply_info_parsed *info,
 450                                   u64 features)
 451 {
 452         u32 op = le32_to_cpu(info->head->op);
 453 
 454         if (op == CEPH_MDS_OP_GETFILELOCK)
 455                 return parse_reply_info_filelock(p, end, info, features);
 456         else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
 457                 return parse_reply_info_readdir(p, end, info, features);
 458         else if (op == CEPH_MDS_OP_CREATE)
 459                 return parse_reply_info_create(p, end, info, features);
 460         else
 461                 return -EIO;
 462 }
 463 
 464 /*
 465  * parse entire mds reply
 466  */
 467 static int parse_reply_info(struct ceph_msg *msg,
 468                             struct ceph_mds_reply_info_parsed *info,
 469                             u64 features)
 470 {
 471         void *p, *end;
 472         u32 len;
 473         int err;
 474 
 475         info->head = msg->front.iov_base;
 476         p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
 477         end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
 478 
 479         /* trace */
 480         ceph_decode_32_safe(&p, end, len, bad);
 481         if (len > 0) {
 482                 ceph_decode_need(&p, end, len, bad);
 483                 err = parse_reply_info_trace(&p, p+len, info, features);
 484                 if (err < 0)
 485                         goto out_bad;
 486         }
 487 
 488         /* extra */
 489         ceph_decode_32_safe(&p, end, len, bad);
 490         if (len > 0) {
 491                 ceph_decode_need(&p, end, len, bad);
 492                 err = parse_reply_info_extra(&p, p+len, info, features);
 493                 if (err < 0)
 494                         goto out_bad;
 495         }
 496 
 497         /* snap blob */
 498         ceph_decode_32_safe(&p, end, len, bad);
 499         info->snapblob_len = len;
 500         info->snapblob = p;
 501         p += len;
 502 
 503         if (p != end)
 504                 goto bad;
 505         return 0;
 506 
 507 bad:
 508         err = -EIO;
 509 out_bad:
 510         pr_err("mds parse_reply err %d\n", err);
 511         return err;
 512 }
 513 
 514 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
 515 {
 516         if (!info->dir_entries)
 517                 return;
 518         free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
 519 }
 520 
 521 
 522 /*
 523  * sessions
 524  */
 525 const char *ceph_session_state_name(int s)
 526 {
 527         switch (s) {
 528         case CEPH_MDS_SESSION_NEW: return "new";
 529         case CEPH_MDS_SESSION_OPENING: return "opening";
 530         case CEPH_MDS_SESSION_OPEN: return "open";
 531         case CEPH_MDS_SESSION_HUNG: return "hung";
 532         case CEPH_MDS_SESSION_CLOSING: return "closing";
 533         case CEPH_MDS_SESSION_RESTARTING: return "restarting";
 534         case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
 535         case CEPH_MDS_SESSION_REJECTED: return "rejected";
 536         default: return "???";
 537         }
 538 }
 539 
 540 static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
 541 {
 542         if (refcount_inc_not_zero(&s->s_ref)) {
 543                 dout("mdsc get_session %p %d -> %d\n", s,
 544                      refcount_read(&s->s_ref)-1, refcount_read(&s->s_ref));
 545                 return s;
 546         } else {
 547                 dout("mdsc get_session %p 0 -- FAIL\n", s);
 548                 return NULL;
 549         }
 550 }
 551 
 552 void ceph_put_mds_session(struct ceph_mds_session *s)
 553 {
 554         dout("mdsc put_session %p %d -> %d\n", s,
 555              refcount_read(&s->s_ref), refcount_read(&s->s_ref)-1);
 556         if (refcount_dec_and_test(&s->s_ref)) {
 557                 if (s->s_auth.authorizer)
 558                         ceph_auth_destroy_authorizer(s->s_auth.authorizer);
 559                 kfree(s);
 560         }
 561 }
 562 
 563 /*
 564  * called under mdsc->mutex
 565  */
 566 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
 567                                                    int mds)
 568 {
 569         if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
 570                 return NULL;
 571         return get_session(mdsc->sessions[mds]);
 572 }
 573 
 574 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
 575 {
 576         if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
 577                 return false;
 578         else
 579                 return true;
 580 }
 581 
 582 static int __verify_registered_session(struct ceph_mds_client *mdsc,
 583                                        struct ceph_mds_session *s)
 584 {
 585         if (s->s_mds >= mdsc->max_sessions ||
 586             mdsc->sessions[s->s_mds] != s)
 587                 return -ENOENT;
 588         return 0;
 589 }
 590 
 591 /*
 592  * create+register a new session for given mds.
 593  * called under mdsc->mutex.
 594  */
 595 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
 596                                                  int mds)
 597 {
 598         struct ceph_mds_session *s;
 599 
 600         if (mds >= mdsc->mdsmap->m_num_mds)
 601                 return ERR_PTR(-EINVAL);
 602 
 603         s = kzalloc(sizeof(*s), GFP_NOFS);
 604         if (!s)
 605                 return ERR_PTR(-ENOMEM);
 606 
 607         if (mds >= mdsc->max_sessions) {
 608                 int newmax = 1 << get_count_order(mds + 1);
 609                 struct ceph_mds_session **sa;
 610 
 611                 dout("%s: realloc to %d\n", __func__, newmax);
 612                 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
 613                 if (!sa)
 614                         goto fail_realloc;
 615                 if (mdsc->sessions) {
 616                         memcpy(sa, mdsc->sessions,
 617                                mdsc->max_sessions * sizeof(void *));
 618                         kfree(mdsc->sessions);
 619                 }
 620                 mdsc->sessions = sa;
 621                 mdsc->max_sessions = newmax;
 622         }
 623 
 624         dout("%s: mds%d\n", __func__, mds);
 625         s->s_mdsc = mdsc;
 626         s->s_mds = mds;
 627         s->s_state = CEPH_MDS_SESSION_NEW;
 628         s->s_ttl = 0;
 629         s->s_seq = 0;
 630         mutex_init(&s->s_mutex);
 631 
 632         ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
 633 
 634         spin_lock_init(&s->s_gen_ttl_lock);
 635         s->s_cap_gen = 1;
 636         s->s_cap_ttl = jiffies - 1;
 637 
 638         spin_lock_init(&s->s_cap_lock);
 639         s->s_renew_requested = 0;
 640         s->s_renew_seq = 0;
 641         INIT_LIST_HEAD(&s->s_caps);
 642         s->s_nr_caps = 0;
 643         refcount_set(&s->s_ref, 1);
 644         INIT_LIST_HEAD(&s->s_waiting);
 645         INIT_LIST_HEAD(&s->s_unsafe);
 646         s->s_num_cap_releases = 0;
 647         s->s_cap_reconnect = 0;
 648         s->s_cap_iterator = NULL;
 649         INIT_LIST_HEAD(&s->s_cap_releases);
 650         INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
 651 
 652         INIT_LIST_HEAD(&s->s_cap_flushing);
 653 
 654         mdsc->sessions[mds] = s;
 655         atomic_inc(&mdsc->num_sessions);
 656         refcount_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
 657 
 658         ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
 659                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
 660 
 661         return s;
 662 
 663 fail_realloc:
 664         kfree(s);
 665         return ERR_PTR(-ENOMEM);
 666 }
 667 
 668 /*
 669  * called under mdsc->mutex
 670  */
 671 static void __unregister_session(struct ceph_mds_client *mdsc,
 672                                struct ceph_mds_session *s)
 673 {
 674         dout("__unregister_session mds%d %p\n", s->s_mds, s);
 675         BUG_ON(mdsc->sessions[s->s_mds] != s);
 676         mdsc->sessions[s->s_mds] = NULL;
 677         s->s_state = 0;
 678         ceph_con_close(&s->s_con);
 679         ceph_put_mds_session(s);
 680         atomic_dec(&mdsc->num_sessions);
 681 }
 682 
 683 /*
 684  * drop session refs in request.
 685  *
 686  * should be last request ref, or hold mdsc->mutex
 687  */
 688 static void put_request_session(struct ceph_mds_request *req)
 689 {
 690         if (req->r_session) {
 691                 ceph_put_mds_session(req->r_session);
 692                 req->r_session = NULL;
 693         }
 694 }
 695 
 696 void ceph_mdsc_release_request(struct kref *kref)
 697 {
 698         struct ceph_mds_request *req = container_of(kref,
 699                                                     struct ceph_mds_request,
 700                                                     r_kref);
 701         destroy_reply_info(&req->r_reply_info);
 702         if (req->r_request)
 703                 ceph_msg_put(req->r_request);
 704         if (req->r_reply)
 705                 ceph_msg_put(req->r_reply);
 706         if (req->r_inode) {
 707                 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
 708                 /* avoid calling iput_final() in mds dispatch threads */
 709                 ceph_async_iput(req->r_inode);
 710         }
 711         if (req->r_parent) {
 712                 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
 713                 ceph_async_iput(req->r_parent);
 714         }
 715         ceph_async_iput(req->r_target_inode);
 716         if (req->r_dentry)
 717                 dput(req->r_dentry);
 718         if (req->r_old_dentry)
 719                 dput(req->r_old_dentry);
 720         if (req->r_old_dentry_dir) {
 721                 /*
 722                  * track (and drop pins for) r_old_dentry_dir
 723                  * separately, since r_old_dentry's d_parent may have
 724                  * changed between the dir mutex being dropped and
 725                  * this request being freed.
 726                  */
 727                 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
 728                                   CEPH_CAP_PIN);
 729                 ceph_async_iput(req->r_old_dentry_dir);
 730         }
 731         kfree(req->r_path1);
 732         kfree(req->r_path2);
 733         if (req->r_pagelist)
 734                 ceph_pagelist_release(req->r_pagelist);
 735         put_request_session(req);
 736         ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
 737         WARN_ON_ONCE(!list_empty(&req->r_wait));
 738         kfree(req);
 739 }
 740 
 741 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
 742 
 743 /*
 744  * lookup session, bump ref if found.
 745  *
 746  * called under mdsc->mutex.
 747  */
 748 static struct ceph_mds_request *
 749 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
 750 {
 751         struct ceph_mds_request *req;
 752 
 753         req = lookup_request(&mdsc->request_tree, tid);
 754         if (req)
 755                 ceph_mdsc_get_request(req);
 756 
 757         return req;
 758 }
 759 
 760 /*
 761  * Register an in-flight request, and assign a tid.  Link to directory
 762  * are modifying (if any).
 763  *
 764  * Called under mdsc->mutex.
 765  */
 766 static void __register_request(struct ceph_mds_client *mdsc,
 767                                struct ceph_mds_request *req,
 768                                struct inode *dir)
 769 {
 770         int ret = 0;
 771 
 772         req->r_tid = ++mdsc->last_tid;
 773         if (req->r_num_caps) {
 774                 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
 775                                         req->r_num_caps);
 776                 if (ret < 0) {
 777                         pr_err("__register_request %p "
 778                                "failed to reserve caps: %d\n", req, ret);
 779                         /* set req->r_err to fail early from __do_request */
 780                         req->r_err = ret;
 781                         return;
 782                 }
 783         }
 784         dout("__register_request %p tid %lld\n", req, req->r_tid);
 785         ceph_mdsc_get_request(req);
 786         insert_request(&mdsc->request_tree, req);
 787 
 788         req->r_uid = current_fsuid();
 789         req->r_gid = current_fsgid();
 790 
 791         if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
 792                 mdsc->oldest_tid = req->r_tid;
 793 
 794         if (dir) {
 795                 ihold(dir);
 796                 req->r_unsafe_dir = dir;
 797         }
 798 }
 799 
 800 static void __unregister_request(struct ceph_mds_client *mdsc,
 801                                  struct ceph_mds_request *req)
 802 {
 803         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
 804 
 805         /* Never leave an unregistered request on an unsafe list! */
 806         list_del_init(&req->r_unsafe_item);
 807 
 808         if (req->r_tid == mdsc->oldest_tid) {
 809                 struct rb_node *p = rb_next(&req->r_node);
 810                 mdsc->oldest_tid = 0;
 811                 while (p) {
 812                         struct ceph_mds_request *next_req =
 813                                 rb_entry(p, struct ceph_mds_request, r_node);
 814                         if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
 815                                 mdsc->oldest_tid = next_req->r_tid;
 816                                 break;
 817                         }
 818                         p = rb_next(p);
 819                 }
 820         }
 821 
 822         erase_request(&mdsc->request_tree, req);
 823 
 824         if (req->r_unsafe_dir  &&
 825             test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
 826                 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
 827                 spin_lock(&ci->i_unsafe_lock);
 828                 list_del_init(&req->r_unsafe_dir_item);
 829                 spin_unlock(&ci->i_unsafe_lock);
 830         }
 831         if (req->r_target_inode &&
 832             test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
 833                 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
 834                 spin_lock(&ci->i_unsafe_lock);
 835                 list_del_init(&req->r_unsafe_target_item);
 836                 spin_unlock(&ci->i_unsafe_lock);
 837         }
 838 
 839         if (req->r_unsafe_dir) {
 840                 /* avoid calling iput_final() in mds dispatch threads */
 841                 ceph_async_iput(req->r_unsafe_dir);
 842                 req->r_unsafe_dir = NULL;
 843         }
 844 
 845         complete_all(&req->r_safe_completion);
 846 
 847         ceph_mdsc_put_request(req);
 848 }
 849 
 850 /*
 851  * Walk back up the dentry tree until we hit a dentry representing a
 852  * non-snapshot inode. We do this using the rcu_read_lock (which must be held
 853  * when calling this) to ensure that the objects won't disappear while we're
 854  * working with them. Once we hit a candidate dentry, we attempt to take a
 855  * reference to it, and return that as the result.
 856  */
 857 static struct inode *get_nonsnap_parent(struct dentry *dentry)
 858 {
 859         struct inode *inode = NULL;
 860 
 861         while (dentry && !IS_ROOT(dentry)) {
 862                 inode = d_inode_rcu(dentry);
 863                 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
 864                         break;
 865                 dentry = dentry->d_parent;
 866         }
 867         if (inode)
 868                 inode = igrab(inode);
 869         return inode;
 870 }
 871 
 872 /*
 873  * Choose mds to send request to next.  If there is a hint set in the
 874  * request (e.g., due to a prior forward hint from the mds), use that.
 875  * Otherwise, consult frag tree and/or caps to identify the
 876  * appropriate mds.  If all else fails, choose randomly.
 877  *
 878  * Called under mdsc->mutex.
 879  */
 880 static int __choose_mds(struct ceph_mds_client *mdsc,
 881                         struct ceph_mds_request *req)
 882 {
 883         struct inode *inode;
 884         struct ceph_inode_info *ci;
 885         struct ceph_cap *cap;
 886         int mode = req->r_direct_mode;
 887         int mds = -1;
 888         u32 hash = req->r_direct_hash;
 889         bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
 890 
 891         /*
 892          * is there a specific mds we should try?  ignore hint if we have
 893          * no session and the mds is not up (active or recovering).
 894          */
 895         if (req->r_resend_mds >= 0 &&
 896             (__have_session(mdsc, req->r_resend_mds) ||
 897              ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
 898                 dout("choose_mds using resend_mds mds%d\n",
 899                      req->r_resend_mds);
 900                 return req->r_resend_mds;
 901         }
 902 
 903         if (mode == USE_RANDOM_MDS)
 904                 goto random;
 905 
 906         inode = NULL;
 907         if (req->r_inode) {
 908                 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
 909                         inode = req->r_inode;
 910                         ihold(inode);
 911                 } else {
 912                         /* req->r_dentry is non-null for LSSNAP request */
 913                         rcu_read_lock();
 914                         inode = get_nonsnap_parent(req->r_dentry);
 915                         rcu_read_unlock();
 916                         dout("__choose_mds using snapdir's parent %p\n", inode);
 917                 }
 918         } else if (req->r_dentry) {
 919                 /* ignore race with rename; old or new d_parent is okay */
 920                 struct dentry *parent;
 921                 struct inode *dir;
 922 
 923                 rcu_read_lock();
 924                 parent = READ_ONCE(req->r_dentry->d_parent);
 925                 dir = req->r_parent ? : d_inode_rcu(parent);
 926 
 927                 if (!dir || dir->i_sb != mdsc->fsc->sb) {
 928                         /*  not this fs or parent went negative */
 929                         inode = d_inode(req->r_dentry);
 930                         if (inode)
 931                                 ihold(inode);
 932                 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
 933                         /* direct snapped/virtual snapdir requests
 934                          * based on parent dir inode */
 935                         inode = get_nonsnap_parent(parent);
 936                         dout("__choose_mds using nonsnap parent %p\n", inode);
 937                 } else {
 938                         /* dentry target */
 939                         inode = d_inode(req->r_dentry);
 940                         if (!inode || mode == USE_AUTH_MDS) {
 941                                 /* dir + name */
 942                                 inode = igrab(dir);
 943                                 hash = ceph_dentry_hash(dir, req->r_dentry);
 944                                 is_hash = true;
 945                         } else {
 946                                 ihold(inode);
 947                         }
 948                 }
 949                 rcu_read_unlock();
 950         }
 951 
 952         dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
 953              (int)hash, mode);
 954         if (!inode)
 955                 goto random;
 956         ci = ceph_inode(inode);
 957 
 958         if (is_hash && S_ISDIR(inode->i_mode)) {
 959                 struct ceph_inode_frag frag;
 960                 int found;
 961 
 962                 ceph_choose_frag(ci, hash, &frag, &found);
 963                 if (found) {
 964                         if (mode == USE_ANY_MDS && frag.ndist > 0) {
 965                                 u8 r;
 966 
 967                                 /* choose a random replica */
 968                                 get_random_bytes(&r, 1);
 969                                 r %= frag.ndist;
 970                                 mds = frag.dist[r];
 971                                 dout("choose_mds %p %llx.%llx "
 972                                      "frag %u mds%d (%d/%d)\n",
 973                                      inode, ceph_vinop(inode),
 974                                      frag.frag, mds,
 975                                      (int)r, frag.ndist);
 976                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
 977                                     CEPH_MDS_STATE_ACTIVE)
 978                                         goto out;
 979                         }
 980 
 981                         /* since this file/dir wasn't known to be
 982                          * replicated, then we want to look for the
 983                          * authoritative mds. */
 984                         mode = USE_AUTH_MDS;
 985                         if (frag.mds >= 0) {
 986                                 /* choose auth mds */
 987                                 mds = frag.mds;
 988                                 dout("choose_mds %p %llx.%llx "
 989                                      "frag %u mds%d (auth)\n",
 990                                      inode, ceph_vinop(inode), frag.frag, mds);
 991                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
 992                                     CEPH_MDS_STATE_ACTIVE)
 993                                         goto out;
 994                         }
 995                 }
 996         }
 997 
 998         spin_lock(&ci->i_ceph_lock);
 999         cap = NULL;
1000         if (mode == USE_AUTH_MDS)
1001                 cap = ci->i_auth_cap;
1002         if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1003                 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1004         if (!cap) {
1005                 spin_unlock(&ci->i_ceph_lock);
1006                 ceph_async_iput(inode);
1007                 goto random;
1008         }
1009         mds = cap->session->s_mds;
1010         dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
1011              inode, ceph_vinop(inode), mds,
1012              cap == ci->i_auth_cap ? "auth " : "", cap);
1013         spin_unlock(&ci->i_ceph_lock);
1014 out:
1015         /* avoid calling iput_final() while holding mdsc->mutex or
1016          * in mds dispatch threads */
1017         ceph_async_iput(inode);
1018         return mds;
1019 
1020 random:
1021         mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
1022         dout("choose_mds chose random mds%d\n", mds);
1023         return mds;
1024 }
1025 
1026 
1027 /*
1028  * session messages
1029  */
1030 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
1031 {
1032         struct ceph_msg *msg;
1033         struct ceph_mds_session_head *h;
1034 
1035         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1036                            false);
1037         if (!msg) {
1038                 pr_err("create_session_msg ENOMEM creating msg\n");
1039                 return NULL;
1040         }
1041         h = msg->front.iov_base;
1042         h->op = cpu_to_le32(op);
1043         h->seq = cpu_to_le64(seq);
1044 
1045         return msg;
1046 }
1047 
1048 static void encode_supported_features(void **p, void *end)
1049 {
1050         static const unsigned char bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1051         static const size_t count = ARRAY_SIZE(bits);
1052 
1053         if (count > 0) {
1054                 size_t i;
1055                 size_t size = ((size_t)bits[count - 1] + 64) / 64 * 8;
1056 
1057                 BUG_ON(*p + 4 + size > end);
1058                 ceph_encode_32(p, size);
1059                 memset(*p, 0, size);
1060                 for (i = 0; i < count; i++)
1061                         ((unsigned char*)(*p))[i / 8] |= 1 << (bits[i] % 8);
1062                 *p += size;
1063         } else {
1064                 BUG_ON(*p + 4 > end);
1065                 ceph_encode_32(p, 0);
1066         }
1067 }
1068 
1069 /*
1070  * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1071  * to include additional client metadata fields.
1072  */
1073 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
1074 {
1075         struct ceph_msg *msg;
1076         struct ceph_mds_session_head *h;
1077         int i = -1;
1078         int extra_bytes = 0;
1079         int metadata_key_count = 0;
1080         struct ceph_options *opt = mdsc->fsc->client->options;
1081         struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
1082         void *p, *end;
1083 
1084         const char* metadata[][2] = {
1085                 {"hostname", mdsc->nodename},
1086                 {"kernel_version", init_utsname()->release},
1087                 {"entity_id", opt->name ? : ""},
1088                 {"root", fsopt->server_path ? : "/"},
1089                 {NULL, NULL}
1090         };
1091 
1092         /* Calculate serialized length of metadata */
1093         extra_bytes = 4;  /* map length */
1094         for (i = 0; metadata[i][0]; ++i) {
1095                 extra_bytes += 8 + strlen(metadata[i][0]) +
1096                         strlen(metadata[i][1]);
1097                 metadata_key_count++;
1098         }
1099         /* supported feature */
1100         extra_bytes += 4 + 8;
1101 
1102         /* Allocate the message */
1103         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
1104                            GFP_NOFS, false);
1105         if (!msg) {
1106                 pr_err("create_session_msg ENOMEM creating msg\n");
1107                 return NULL;
1108         }
1109         p = msg->front.iov_base;
1110         end = p + msg->front.iov_len;
1111 
1112         h = p;
1113         h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
1114         h->seq = cpu_to_le64(seq);
1115 
1116         /*
1117          * Serialize client metadata into waiting buffer space, using
1118          * the format that userspace expects for map<string, string>
1119          *
1120          * ClientSession messages with metadata are v2
1121          */
1122         msg->hdr.version = cpu_to_le16(3);
1123         msg->hdr.compat_version = cpu_to_le16(1);
1124 
1125         /* The write pointer, following the session_head structure */
1126         p += sizeof(*h);
1127 
1128         /* Number of entries in the map */
1129         ceph_encode_32(&p, metadata_key_count);
1130 
1131         /* Two length-prefixed strings for each entry in the map */
1132         for (i = 0; metadata[i][0]; ++i) {
1133                 size_t const key_len = strlen(metadata[i][0]);
1134                 size_t const val_len = strlen(metadata[i][1]);
1135 
1136                 ceph_encode_32(&p, key_len);
1137                 memcpy(p, metadata[i][0], key_len);
1138                 p += key_len;
1139                 ceph_encode_32(&p, val_len);
1140                 memcpy(p, metadata[i][1], val_len);
1141                 p += val_len;
1142         }
1143 
1144         encode_supported_features(&p, end);
1145         msg->front.iov_len = p - msg->front.iov_base;
1146         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1147 
1148         return msg;
1149 }
1150 
1151 /*
1152  * send session open request.
1153  *
1154  * called under mdsc->mutex
1155  */
1156 static int __open_session(struct ceph_mds_client *mdsc,
1157                           struct ceph_mds_session *session)
1158 {
1159         struct ceph_msg *msg;
1160         int mstate;
1161         int mds = session->s_mds;
1162 
1163         /* wait for mds to go active? */
1164         mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1165         dout("open_session to mds%d (%s)\n", mds,
1166              ceph_mds_state_name(mstate));
1167         session->s_state = CEPH_MDS_SESSION_OPENING;
1168         session->s_renew_requested = jiffies;
1169 
1170         /* send connect message */
1171         msg = create_session_open_msg(mdsc, session->s_seq);
1172         if (!msg)
1173                 return -ENOMEM;
1174         ceph_con_send(&session->s_con, msg);
1175         return 0;
1176 }
1177 
1178 /*
1179  * open sessions for any export targets for the given mds
1180  *
1181  * called under mdsc->mutex
1182  */
1183 static struct ceph_mds_session *
1184 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1185 {
1186         struct ceph_mds_session *session;
1187 
1188         session = __ceph_lookup_mds_session(mdsc, target);
1189         if (!session) {
1190                 session = register_session(mdsc, target);
1191                 if (IS_ERR(session))
1192                         return session;
1193         }
1194         if (session->s_state == CEPH_MDS_SESSION_NEW ||
1195             session->s_state == CEPH_MDS_SESSION_CLOSING)
1196                 __open_session(mdsc, session);
1197 
1198         return session;
1199 }
1200 
1201 struct ceph_mds_session *
1202 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1203 {
1204         struct ceph_mds_session *session;
1205 
1206         dout("open_export_target_session to mds%d\n", target);
1207 
1208         mutex_lock(&mdsc->mutex);
1209         session = __open_export_target_session(mdsc, target);
1210         mutex_unlock(&mdsc->mutex);
1211 
1212         return session;
1213 }
1214 
1215 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1216                                           struct ceph_mds_session *session)
1217 {
1218         struct ceph_mds_info *mi;
1219         struct ceph_mds_session *ts;
1220         int i, mds = session->s_mds;
1221 
1222         if (mds >= mdsc->mdsmap->m_num_mds)
1223                 return;
1224 
1225         mi = &mdsc->mdsmap->m_info[mds];
1226         dout("open_export_target_sessions for mds%d (%d targets)\n",
1227              session->s_mds, mi->num_export_targets);
1228 
1229         for (i = 0; i < mi->num_export_targets; i++) {
1230                 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1231                 if (!IS_ERR(ts))
1232                         ceph_put_mds_session(ts);
1233         }
1234 }
1235 
1236 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1237                                            struct ceph_mds_session *session)
1238 {
1239         mutex_lock(&mdsc->mutex);
1240         __open_export_target_sessions(mdsc, session);
1241         mutex_unlock(&mdsc->mutex);
1242 }
1243 
1244 /*
1245  * session caps
1246  */
1247 
1248 static void detach_cap_releases(struct ceph_mds_session *session,
1249                                 struct list_head *target)
1250 {
1251         lockdep_assert_held(&session->s_cap_lock);
1252 
1253         list_splice_init(&session->s_cap_releases, target);
1254         session->s_num_cap_releases = 0;
1255         dout("dispose_cap_releases mds%d\n", session->s_mds);
1256 }
1257 
1258 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1259                                  struct list_head *dispose)
1260 {
1261         while (!list_empty(dispose)) {
1262                 struct ceph_cap *cap;
1263                 /* zero out the in-progress message */
1264                 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1265                 list_del(&cap->session_caps);
1266                 ceph_put_cap(mdsc, cap);
1267         }
1268 }
1269 
1270 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1271                                      struct ceph_mds_session *session)
1272 {
1273         struct ceph_mds_request *req;
1274         struct rb_node *p;
1275         struct ceph_inode_info *ci;
1276 
1277         dout("cleanup_session_requests mds%d\n", session->s_mds);
1278         mutex_lock(&mdsc->mutex);
1279         while (!list_empty(&session->s_unsafe)) {
1280                 req = list_first_entry(&session->s_unsafe,
1281                                        struct ceph_mds_request, r_unsafe_item);
1282                 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1283                                     req->r_tid);
1284                 if (req->r_target_inode) {
1285                         /* dropping unsafe change of inode's attributes */
1286                         ci = ceph_inode(req->r_target_inode);
1287                         errseq_set(&ci->i_meta_err, -EIO);
1288                 }
1289                 if (req->r_unsafe_dir) {
1290                         /* dropping unsafe directory operation */
1291                         ci = ceph_inode(req->r_unsafe_dir);
1292                         errseq_set(&ci->i_meta_err, -EIO);
1293                 }
1294                 __unregister_request(mdsc, req);
1295         }
1296         /* zero r_attempts, so kick_requests() will re-send requests */
1297         p = rb_first(&mdsc->request_tree);
1298         while (p) {
1299                 req = rb_entry(p, struct ceph_mds_request, r_node);
1300                 p = rb_next(p);
1301                 if (req->r_session &&
1302                     req->r_session->s_mds == session->s_mds)
1303                         req->r_attempts = 0;
1304         }
1305         mutex_unlock(&mdsc->mutex);
1306 }
1307 
1308 /*
1309  * Helper to safely iterate over all caps associated with a session, with
1310  * special care taken to handle a racing __ceph_remove_cap().
1311  *
1312  * Caller must hold session s_mutex.
1313  */
1314 int ceph_iterate_session_caps(struct ceph_mds_session *session,
1315                               int (*cb)(struct inode *, struct ceph_cap *,
1316                                         void *), void *arg)
1317 {
1318         struct list_head *p;
1319         struct ceph_cap *cap;
1320         struct inode *inode, *last_inode = NULL;
1321         struct ceph_cap *old_cap = NULL;
1322         int ret;
1323 
1324         dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1325         spin_lock(&session->s_cap_lock);
1326         p = session->s_caps.next;
1327         while (p != &session->s_caps) {
1328                 cap = list_entry(p, struct ceph_cap, session_caps);
1329                 inode = igrab(&cap->ci->vfs_inode);
1330                 if (!inode) {
1331                         p = p->next;
1332                         continue;
1333                 }
1334                 session->s_cap_iterator = cap;
1335                 spin_unlock(&session->s_cap_lock);
1336 
1337                 if (last_inode) {
1338                         /* avoid calling iput_final() while holding
1339                          * s_mutex or in mds dispatch threads */
1340                         ceph_async_iput(last_inode);
1341                         last_inode = NULL;
1342                 }
1343                 if (old_cap) {
1344                         ceph_put_cap(session->s_mdsc, old_cap);
1345                         old_cap = NULL;
1346                 }
1347 
1348                 ret = cb(inode, cap, arg);
1349                 last_inode = inode;
1350 
1351                 spin_lock(&session->s_cap_lock);
1352                 p = p->next;
1353                 if (!cap->ci) {
1354                         dout("iterate_session_caps  finishing cap %p removal\n",
1355                              cap);
1356                         BUG_ON(cap->session != session);
1357                         cap->session = NULL;
1358                         list_del_init(&cap->session_caps);
1359                         session->s_nr_caps--;
1360                         if (cap->queue_release)
1361                                 __ceph_queue_cap_release(session, cap);
1362                         else
1363                                 old_cap = cap;  /* put_cap it w/o locks held */
1364                 }
1365                 if (ret < 0)
1366                         goto out;
1367         }
1368         ret = 0;
1369 out:
1370         session->s_cap_iterator = NULL;
1371         spin_unlock(&session->s_cap_lock);
1372 
1373         ceph_async_iput(last_inode);
1374         if (old_cap)
1375                 ceph_put_cap(session->s_mdsc, old_cap);
1376 
1377         return ret;
1378 }
1379 
1380 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1381                                   void *arg)
1382 {
1383         struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
1384         struct ceph_inode_info *ci = ceph_inode(inode);
1385         LIST_HEAD(to_remove);
1386         bool dirty_dropped = false;
1387         bool invalidate = false;
1388 
1389         dout("removing cap %p, ci is %p, inode is %p\n",
1390              cap, ci, &ci->vfs_inode);
1391         spin_lock(&ci->i_ceph_lock);
1392         if (cap->mds_wanted | cap->issued)
1393                 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1394         __ceph_remove_cap(cap, false);
1395         if (!ci->i_auth_cap) {
1396                 struct ceph_cap_flush *cf;
1397                 struct ceph_mds_client *mdsc = fsc->mdsc;
1398 
1399                 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1400                         if (inode->i_data.nrpages > 0)
1401                                 invalidate = true;
1402                         if (ci->i_wrbuffer_ref > 0)
1403                                 mapping_set_error(&inode->i_data, -EIO);
1404                 }
1405 
1406                 while (!list_empty(&ci->i_cap_flush_list)) {
1407                         cf = list_first_entry(&ci->i_cap_flush_list,
1408                                               struct ceph_cap_flush, i_list);
1409                         list_move(&cf->i_list, &to_remove);
1410                 }
1411 
1412                 spin_lock(&mdsc->cap_dirty_lock);
1413 
1414                 list_for_each_entry(cf, &to_remove, i_list)
1415                         list_del(&cf->g_list);
1416 
1417                 if (!list_empty(&ci->i_dirty_item)) {
1418                         pr_warn_ratelimited(
1419                                 " dropping dirty %s state for %p %lld\n",
1420                                 ceph_cap_string(ci->i_dirty_caps),
1421                                 inode, ceph_ino(inode));
1422                         ci->i_dirty_caps = 0;
1423                         list_del_init(&ci->i_dirty_item);
1424                         dirty_dropped = true;
1425                 }
1426                 if (!list_empty(&ci->i_flushing_item)) {
1427                         pr_warn_ratelimited(
1428                                 " dropping dirty+flushing %s state for %p %lld\n",
1429                                 ceph_cap_string(ci->i_flushing_caps),
1430                                 inode, ceph_ino(inode));
1431                         ci->i_flushing_caps = 0;
1432                         list_del_init(&ci->i_flushing_item);
1433                         mdsc->num_cap_flushing--;
1434                         dirty_dropped = true;
1435                 }
1436                 spin_unlock(&mdsc->cap_dirty_lock);
1437 
1438                 if (dirty_dropped) {
1439                         errseq_set(&ci->i_meta_err, -EIO);
1440 
1441                         if (ci->i_wrbuffer_ref_head == 0 &&
1442                             ci->i_wr_ref == 0 &&
1443                             ci->i_dirty_caps == 0 &&
1444                             ci->i_flushing_caps == 0) {
1445                                 ceph_put_snap_context(ci->i_head_snapc);
1446                                 ci->i_head_snapc = NULL;
1447                         }
1448                 }
1449 
1450                 if (atomic_read(&ci->i_filelock_ref) > 0) {
1451                         /* make further file lock syscall return -EIO */
1452                         ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1453                         pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1454                                             inode, ceph_ino(inode));
1455                 }
1456 
1457                 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1458                         list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
1459                         ci->i_prealloc_cap_flush = NULL;
1460                 }
1461         }
1462         spin_unlock(&ci->i_ceph_lock);
1463         while (!list_empty(&to_remove)) {
1464                 struct ceph_cap_flush *cf;
1465                 cf = list_first_entry(&to_remove,
1466                                       struct ceph_cap_flush, i_list);
1467                 list_del(&cf->i_list);
1468                 ceph_free_cap_flush(cf);
1469         }
1470 
1471         wake_up_all(&ci->i_cap_wq);
1472         if (invalidate)
1473                 ceph_queue_invalidate(inode);
1474         if (dirty_dropped)
1475                 iput(inode);
1476         return 0;
1477 }
1478 
1479 /*
1480  * caller must hold session s_mutex
1481  */
1482 static void remove_session_caps(struct ceph_mds_session *session)
1483 {
1484         struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1485         struct super_block *sb = fsc->sb;
1486         LIST_HEAD(dispose);
1487 
1488         dout("remove_session_caps on %p\n", session);
1489         ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
1490 
1491         wake_up_all(&fsc->mdsc->cap_flushing_wq);
1492 
1493         spin_lock(&session->s_cap_lock);
1494         if (session->s_nr_caps > 0) {
1495                 struct inode *inode;
1496                 struct ceph_cap *cap, *prev = NULL;
1497                 struct ceph_vino vino;
1498                 /*
1499                  * iterate_session_caps() skips inodes that are being
1500                  * deleted, we need to wait until deletions are complete.
1501                  * __wait_on_freeing_inode() is designed for the job,
1502                  * but it is not exported, so use lookup inode function
1503                  * to access it.
1504                  */
1505                 while (!list_empty(&session->s_caps)) {
1506                         cap = list_entry(session->s_caps.next,
1507                                          struct ceph_cap, session_caps);
1508                         if (cap == prev)
1509                                 break;
1510                         prev = cap;
1511                         vino = cap->ci->i_vino;
1512                         spin_unlock(&session->s_cap_lock);
1513 
1514                         inode = ceph_find_inode(sb, vino);
1515                          /* avoid calling iput_final() while holding s_mutex */
1516                         ceph_async_iput(inode);
1517 
1518                         spin_lock(&session->s_cap_lock);
1519                 }
1520         }
1521 
1522         // drop cap expires and unlock s_cap_lock
1523         detach_cap_releases(session, &dispose);
1524 
1525         BUG_ON(session->s_nr_caps > 0);
1526         BUG_ON(!list_empty(&session->s_cap_flushing));
1527         spin_unlock(&session->s_cap_lock);
1528         dispose_cap_releases(session->s_mdsc, &dispose);
1529 }
1530 
1531 enum {
1532         RECONNECT,
1533         RENEWCAPS,
1534         FORCE_RO,
1535 };
1536 
1537 /*
1538  * wake up any threads waiting on this session's caps.  if the cap is
1539  * old (didn't get renewed on the client reconnect), remove it now.
1540  *
1541  * caller must hold s_mutex.
1542  */
1543 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1544                               void *arg)
1545 {
1546         struct ceph_inode_info *ci = ceph_inode(inode);
1547         unsigned long ev = (unsigned long)arg;
1548 
1549         if (ev == RECONNECT) {
1550                 spin_lock(&ci->i_ceph_lock);
1551                 ci->i_wanted_max_size = 0;
1552                 ci->i_requested_max_size = 0;
1553                 spin_unlock(&ci->i_ceph_lock);
1554         } else if (ev == RENEWCAPS) {
1555                 if (cap->cap_gen < cap->session->s_cap_gen) {
1556                         /* mds did not re-issue stale cap */
1557                         spin_lock(&ci->i_ceph_lock);
1558                         cap->issued = cap->implemented = CEPH_CAP_PIN;
1559                         /* make sure mds knows what we want */
1560                         if (__ceph_caps_file_wanted(ci) & ~cap->mds_wanted)
1561                                 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1562                         spin_unlock(&ci->i_ceph_lock);
1563                 }
1564         } else if (ev == FORCE_RO) {
1565         }
1566         wake_up_all(&ci->i_cap_wq);
1567         return 0;
1568 }
1569 
1570 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
1571 {
1572         dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1573         ceph_iterate_session_caps(session, wake_up_session_cb,
1574                                   (void *)(unsigned long)ev);
1575 }
1576 
1577 /*
1578  * Send periodic message to MDS renewing all currently held caps.  The
1579  * ack will reset the expiration for all caps from this session.
1580  *
1581  * caller holds s_mutex
1582  */
1583 static int send_renew_caps(struct ceph_mds_client *mdsc,
1584                            struct ceph_mds_session *session)
1585 {
1586         struct ceph_msg *msg;
1587         int state;
1588 
1589         if (time_after_eq(jiffies, session->s_cap_ttl) &&
1590             time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1591                 pr_info("mds%d caps stale\n", session->s_mds);
1592         session->s_renew_requested = jiffies;
1593 
1594         /* do not try to renew caps until a recovering mds has reconnected
1595          * with its clients. */
1596         state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1597         if (state < CEPH_MDS_STATE_RECONNECT) {
1598                 dout("send_renew_caps ignoring mds%d (%s)\n",
1599                      session->s_mds, ceph_mds_state_name(state));
1600                 return 0;
1601         }
1602 
1603         dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1604                 ceph_mds_state_name(state));
1605         msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1606                                  ++session->s_renew_seq);
1607         if (!msg)
1608                 return -ENOMEM;
1609         ceph_con_send(&session->s_con, msg);
1610         return 0;
1611 }
1612 
1613 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1614                              struct ceph_mds_session *session, u64 seq)
1615 {
1616         struct ceph_msg *msg;
1617 
1618         dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1619              session->s_mds, ceph_session_state_name(session->s_state), seq);
1620         msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1621         if (!msg)
1622                 return -ENOMEM;
1623         ceph_con_send(&session->s_con, msg);
1624         return 0;
1625 }
1626 
1627 
1628 /*
1629  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1630  *
1631  * Called under session->s_mutex
1632  */
1633 static void renewed_caps(struct ceph_mds_client *mdsc,
1634                          struct ceph_mds_session *session, int is_renew)
1635 {
1636         int was_stale;
1637         int wake = 0;
1638 
1639         spin_lock(&session->s_cap_lock);
1640         was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1641 
1642         session->s_cap_ttl = session->s_renew_requested +
1643                 mdsc->mdsmap->m_session_timeout*HZ;
1644 
1645         if (was_stale) {
1646                 if (time_before(jiffies, session->s_cap_ttl)) {
1647                         pr_info("mds%d caps renewed\n", session->s_mds);
1648                         wake = 1;
1649                 } else {
1650                         pr_info("mds%d caps still stale\n", session->s_mds);
1651                 }
1652         }
1653         dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1654              session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1655              time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1656         spin_unlock(&session->s_cap_lock);
1657 
1658         if (wake)
1659                 wake_up_session_caps(session, RENEWCAPS);
1660 }
1661 
1662 /*
1663  * send a session close request
1664  */
1665 static int request_close_session(struct ceph_mds_client *mdsc,
1666                                  struct ceph_mds_session *session)
1667 {
1668         struct ceph_msg *msg;
1669 
1670         dout("request_close_session mds%d state %s seq %lld\n",
1671              session->s_mds, ceph_session_state_name(session->s_state),
1672              session->s_seq);
1673         msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1674         if (!msg)
1675                 return -ENOMEM;
1676         ceph_con_send(&session->s_con, msg);
1677         return 1;
1678 }
1679 
1680 /*
1681  * Called with s_mutex held.
1682  */
1683 static int __close_session(struct ceph_mds_client *mdsc,
1684                          struct ceph_mds_session *session)
1685 {
1686         if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1687                 return 0;
1688         session->s_state = CEPH_MDS_SESSION_CLOSING;
1689         return request_close_session(mdsc, session);
1690 }
1691 
1692 static bool drop_negative_children(struct dentry *dentry)
1693 {
1694         struct dentry *child;
1695         bool all_negative = true;
1696 
1697         if (!d_is_dir(dentry))
1698                 goto out;
1699 
1700         spin_lock(&dentry->d_lock);
1701         list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1702                 if (d_really_is_positive(child)) {
1703                         all_negative = false;
1704                         break;
1705                 }
1706         }
1707         spin_unlock(&dentry->d_lock);
1708 
1709         if (all_negative)
1710                 shrink_dcache_parent(dentry);
1711 out:
1712         return all_negative;
1713 }
1714 
1715 /*
1716  * Trim old(er) caps.
1717  *
1718  * Because we can't cache an inode without one or more caps, we do
1719  * this indirectly: if a cap is unused, we prune its aliases, at which
1720  * point the inode will hopefully get dropped to.
1721  *
1722  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1723  * memory pressure from the MDS, though, so it needn't be perfect.
1724  */
1725 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1726 {
1727         int *remaining = arg;
1728         struct ceph_inode_info *ci = ceph_inode(inode);
1729         int used, wanted, oissued, mine;
1730 
1731         if (*remaining <= 0)
1732                 return -1;
1733 
1734         spin_lock(&ci->i_ceph_lock);
1735         mine = cap->issued | cap->implemented;
1736         used = __ceph_caps_used(ci);
1737         wanted = __ceph_caps_file_wanted(ci);
1738         oissued = __ceph_caps_issued_other(ci, cap);
1739 
1740         dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1741              inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1742              ceph_cap_string(used), ceph_cap_string(wanted));
1743         if (cap == ci->i_auth_cap) {
1744                 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1745                     !list_empty(&ci->i_cap_snaps))
1746                         goto out;
1747                 if ((used | wanted) & CEPH_CAP_ANY_WR)
1748                         goto out;
1749                 /* Note: it's possible that i_filelock_ref becomes non-zero
1750                  * after dropping auth caps. It doesn't hurt because reply
1751                  * of lock mds request will re-add auth caps. */
1752                 if (atomic_read(&ci->i_filelock_ref) > 0)
1753                         goto out;
1754         }
1755         /* The inode has cached pages, but it's no longer used.
1756          * we can safely drop it */
1757         if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1758             !(oissued & CEPH_CAP_FILE_CACHE)) {
1759           used = 0;
1760           oissued = 0;
1761         }
1762         if ((used | wanted) & ~oissued & mine)
1763                 goto out;   /* we need these caps */
1764 
1765         if (oissued) {
1766                 /* we aren't the only cap.. just remove us */
1767                 __ceph_remove_cap(cap, true);
1768                 (*remaining)--;
1769         } else {
1770                 struct dentry *dentry;
1771                 /* try dropping referring dentries */
1772                 spin_unlock(&ci->i_ceph_lock);
1773                 dentry = d_find_any_alias(inode);
1774                 if (dentry && drop_negative_children(dentry)) {
1775                         int count;
1776                         dput(dentry);
1777                         d_prune_aliases(inode);
1778                         count = atomic_read(&inode->i_count);
1779                         if (count == 1)
1780                                 (*remaining)--;
1781                         dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1782                              inode, cap, count);
1783                 } else {
1784                         dput(dentry);
1785                 }
1786                 return 0;
1787         }
1788 
1789 out:
1790         spin_unlock(&ci->i_ceph_lock);
1791         return 0;
1792 }
1793 
1794 /*
1795  * Trim session cap count down to some max number.
1796  */
1797 int ceph_trim_caps(struct ceph_mds_client *mdsc,
1798                    struct ceph_mds_session *session,
1799                    int max_caps)
1800 {
1801         int trim_caps = session->s_nr_caps - max_caps;
1802 
1803         dout("trim_caps mds%d start: %d / %d, trim %d\n",
1804              session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1805         if (trim_caps > 0) {
1806                 int remaining = trim_caps;
1807 
1808                 ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
1809                 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1810                      session->s_mds, session->s_nr_caps, max_caps,
1811                         trim_caps - remaining);
1812         }
1813 
1814         ceph_flush_cap_releases(mdsc, session);
1815         return 0;
1816 }
1817 
1818 static int check_caps_flush(struct ceph_mds_client *mdsc,
1819                             u64 want_flush_tid)
1820 {
1821         int ret = 1;
1822 
1823         spin_lock(&mdsc->cap_dirty_lock);
1824         if (!list_empty(&mdsc->cap_flush_list)) {
1825                 struct ceph_cap_flush *cf =
1826                         list_first_entry(&mdsc->cap_flush_list,
1827                                          struct ceph_cap_flush, g_list);
1828                 if (cf->tid <= want_flush_tid) {
1829                         dout("check_caps_flush still flushing tid "
1830                              "%llu <= %llu\n", cf->tid, want_flush_tid);
1831                         ret = 0;
1832                 }
1833         }
1834         spin_unlock(&mdsc->cap_dirty_lock);
1835         return ret;
1836 }
1837 
1838 /*
1839  * flush all dirty inode data to disk.
1840  *
1841  * returns true if we've flushed through want_flush_tid
1842  */
1843 static void wait_caps_flush(struct ceph_mds_client *mdsc,
1844                             u64 want_flush_tid)
1845 {
1846         dout("check_caps_flush want %llu\n", want_flush_tid);
1847 
1848         wait_event(mdsc->cap_flushing_wq,
1849                    check_caps_flush(mdsc, want_flush_tid));
1850 
1851         dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
1852 }
1853 
1854 /*
1855  * called under s_mutex
1856  */
1857 static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1858                                    struct ceph_mds_session *session)
1859 {
1860         struct ceph_msg *msg = NULL;
1861         struct ceph_mds_cap_release *head;
1862         struct ceph_mds_cap_item *item;
1863         struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
1864         struct ceph_cap *cap;
1865         LIST_HEAD(tmp_list);
1866         int num_cap_releases;
1867         __le32  barrier, *cap_barrier;
1868 
1869         down_read(&osdc->lock);
1870         barrier = cpu_to_le32(osdc->epoch_barrier);
1871         up_read(&osdc->lock);
1872 
1873         spin_lock(&session->s_cap_lock);
1874 again:
1875         list_splice_init(&session->s_cap_releases, &tmp_list);
1876         num_cap_releases = session->s_num_cap_releases;
1877         session->s_num_cap_releases = 0;
1878         spin_unlock(&session->s_cap_lock);
1879 
1880         while (!list_empty(&tmp_list)) {
1881                 if (!msg) {
1882                         msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
1883                                         PAGE_SIZE, GFP_NOFS, false);
1884                         if (!msg)
1885                                 goto out_err;
1886                         head = msg->front.iov_base;
1887                         head->num = cpu_to_le32(0);
1888                         msg->front.iov_len = sizeof(*head);
1889 
1890                         msg->hdr.version = cpu_to_le16(2);
1891                         msg->hdr.compat_version = cpu_to_le16(1);
1892                 }
1893 
1894                 cap = list_first_entry(&tmp_list, struct ceph_cap,
1895                                         session_caps);
1896                 list_del(&cap->session_caps);
1897                 num_cap_releases--;
1898 
1899                 head = msg->front.iov_base;
1900                 put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
1901                                    &head->num);
1902                 item = msg->front.iov_base + msg->front.iov_len;
1903                 item->ino = cpu_to_le64(cap->cap_ino);
1904                 item->cap_id = cpu_to_le64(cap->cap_id);
1905                 item->migrate_seq = cpu_to_le32(cap->mseq);
1906                 item->seq = cpu_to_le32(cap->issue_seq);
1907                 msg->front.iov_len += sizeof(*item);
1908 
1909                 ceph_put_cap(mdsc, cap);
1910 
1911                 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1912                         // Append cap_barrier field
1913                         cap_barrier = msg->front.iov_base + msg->front.iov_len;
1914                         *cap_barrier = barrier;
1915                         msg->front.iov_len += sizeof(*cap_barrier);
1916 
1917                         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1918                         dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1919                         ceph_con_send(&session->s_con, msg);
1920                         msg = NULL;
1921                 }
1922         }
1923 
1924         BUG_ON(num_cap_releases != 0);
1925 
1926         spin_lock(&session->s_cap_lock);
1927         if (!list_empty(&session->s_cap_releases))
1928                 goto again;
1929         spin_unlock(&session->s_cap_lock);
1930 
1931         if (msg) {
1932                 // Append cap_barrier field
1933                 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1934                 *cap_barrier = barrier;
1935                 msg->front.iov_len += sizeof(*cap_barrier);
1936 
1937                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1938                 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1939                 ceph_con_send(&session->s_con, msg);
1940         }
1941         return;
1942 out_err:
1943         pr_err("send_cap_releases mds%d, failed to allocate message\n",
1944                 session->s_mds);
1945         spin_lock(&session->s_cap_lock);
1946         list_splice(&tmp_list, &session->s_cap_releases);
1947         session->s_num_cap_releases += num_cap_releases;
1948         spin_unlock(&session->s_cap_lock);
1949 }
1950 
1951 static void ceph_cap_release_work(struct work_struct *work)
1952 {
1953         struct ceph_mds_session *session =
1954                 container_of(work, struct ceph_mds_session, s_cap_release_work);
1955 
1956         mutex_lock(&session->s_mutex);
1957         if (session->s_state == CEPH_MDS_SESSION_OPEN ||
1958             session->s_state == CEPH_MDS_SESSION_HUNG)
1959                 ceph_send_cap_releases(session->s_mdsc, session);
1960         mutex_unlock(&session->s_mutex);
1961         ceph_put_mds_session(session);
1962 }
1963 
1964 void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
1965                              struct ceph_mds_session *session)
1966 {
1967         if (mdsc->stopping)
1968                 return;
1969 
1970         get_session(session);
1971         if (queue_work(mdsc->fsc->cap_wq,
1972                        &session->s_cap_release_work)) {
1973                 dout("cap release work queued\n");
1974         } else {
1975                 ceph_put_mds_session(session);
1976                 dout("failed to queue cap release work\n");
1977         }
1978 }
1979 
1980 /*
1981  * caller holds session->s_cap_lock
1982  */
1983 void __ceph_queue_cap_release(struct ceph_mds_session *session,
1984                               struct ceph_cap *cap)
1985 {
1986         list_add_tail(&cap->session_caps, &session->s_cap_releases);
1987         session->s_num_cap_releases++;
1988 
1989         if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
1990                 ceph_flush_cap_releases(session->s_mdsc, session);
1991 }
1992 
1993 static void ceph_cap_reclaim_work(struct work_struct *work)
1994 {
1995         struct ceph_mds_client *mdsc =
1996                 container_of(work, struct ceph_mds_client, cap_reclaim_work);
1997         int ret = ceph_trim_dentries(mdsc);
1998         if (ret == -EAGAIN)
1999                 ceph_queue_cap_reclaim_work(mdsc);
2000 }
2001 
2002 void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2003 {
2004         if (mdsc->stopping)
2005                 return;
2006 
2007         if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
2008                 dout("caps reclaim work queued\n");
2009         } else {
2010                 dout("failed to queue caps release work\n");
2011         }
2012 }
2013 
2014 void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2015 {
2016         int val;
2017         if (!nr)
2018                 return;
2019         val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
2020         if (!(val % CEPH_CAPS_PER_RELEASE)) {
2021                 atomic_set(&mdsc->cap_reclaim_pending, 0);
2022                 ceph_queue_cap_reclaim_work(mdsc);
2023         }
2024 }
2025 
2026 /*
2027  * requests
2028  */
2029 
2030 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2031                                     struct inode *dir)
2032 {
2033         struct ceph_inode_info *ci = ceph_inode(dir);
2034         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2035         struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2036         size_t size = sizeof(struct ceph_mds_reply_dir_entry);
2037         int order, num_entries;
2038 
2039         spin_lock(&ci->i_ceph_lock);
2040         num_entries = ci->i_files + ci->i_subdirs;
2041         spin_unlock(&ci->i_ceph_lock);
2042         num_entries = max(num_entries, 1);
2043         num_entries = min(num_entries, opt->max_readdir);
2044 
2045         order = get_order(size * num_entries);
2046         while (order >= 0) {
2047                 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2048                                                              __GFP_NOWARN,
2049                                                              order);
2050                 if (rinfo->dir_entries)
2051                         break;
2052                 order--;
2053         }
2054         if (!rinfo->dir_entries)
2055                 return -ENOMEM;
2056 
2057         num_entries = (PAGE_SIZE << order) / size;
2058         num_entries = min(num_entries, opt->max_readdir);
2059 
2060         rinfo->dir_buf_size = PAGE_SIZE << order;
2061         req->r_num_caps = num_entries + 1;
2062         req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2063         req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2064         return 0;
2065 }
2066 
2067 /*
2068  * Create an mds request.
2069  */
2070 struct ceph_mds_request *
2071 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2072 {
2073         struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
2074         struct timespec64 ts;
2075 
2076         if (!req)
2077                 return ERR_PTR(-ENOMEM);
2078 
2079         mutex_init(&req->r_fill_mutex);
2080         req->r_mdsc = mdsc;
2081         req->r_started = jiffies;
2082         req->r_resend_mds = -1;
2083         INIT_LIST_HEAD(&req->r_unsafe_dir_item);
2084         INIT_LIST_HEAD(&req->r_unsafe_target_item);
2085         req->r_fmode = -1;
2086         kref_init(&req->r_kref);
2087         RB_CLEAR_NODE(&req->r_node);
2088         INIT_LIST_HEAD(&req->r_wait);
2089         init_completion(&req->r_completion);
2090         init_completion(&req->r_safe_completion);
2091         INIT_LIST_HEAD(&req->r_unsafe_item);
2092 
2093         ktime_get_coarse_real_ts64(&ts);
2094         req->r_stamp = timespec64_trunc(ts, mdsc->fsc->sb->s_time_gran);
2095 
2096         req->r_op = op;
2097         req->r_direct_mode = mode;
2098         return req;
2099 }
2100 
2101 /*
2102  * return oldest (lowest) request, tid in request tree, 0 if none.
2103  *
2104  * called under mdsc->mutex.
2105  */
2106 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2107 {
2108         if (RB_EMPTY_ROOT(&mdsc->request_tree))
2109                 return NULL;
2110         return rb_entry(rb_first(&mdsc->request_tree),
2111                         struct ceph_mds_request, r_node);
2112 }
2113 
2114 static inline  u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2115 {
2116         return mdsc->oldest_tid;
2117 }
2118 
2119 /*
2120  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
2121  * on build_path_from_dentry in fs/cifs/dir.c.
2122  *
2123  * If @stop_on_nosnap, generate path relative to the first non-snapped
2124  * inode.
2125  *
2126  * Encode hidden .snap dirs as a double /, i.e.
2127  *   foo/.snap/bar -> foo//bar
2128  */
2129 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *pbase,
2130                            int stop_on_nosnap)
2131 {
2132         struct dentry *temp;
2133         char *path;
2134         int pos;
2135         unsigned seq;
2136         u64 base;
2137 
2138         if (!dentry)
2139                 return ERR_PTR(-EINVAL);
2140 
2141         path = __getname();
2142         if (!path)
2143                 return ERR_PTR(-ENOMEM);
2144 retry:
2145         pos = PATH_MAX - 1;
2146         path[pos] = '\0';
2147 
2148         seq = read_seqbegin(&rename_lock);
2149         rcu_read_lock();
2150         temp = dentry;
2151         for (;;) {
2152                 struct inode *inode;
2153 
2154                 spin_lock(&temp->d_lock);
2155                 inode = d_inode(temp);
2156                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
2157                         dout("build_path path+%d: %p SNAPDIR\n",
2158                              pos, temp);
2159                 } else if (stop_on_nosnap && inode && dentry != temp &&
2160                            ceph_snap(inode) == CEPH_NOSNAP) {
2161                         spin_unlock(&temp->d_lock);
2162                         pos++; /* get rid of any prepended '/' */
2163                         break;
2164                 } else {
2165                         pos -= temp->d_name.len;
2166                         if (pos < 0) {
2167                                 spin_unlock(&temp->d_lock);
2168                                 break;
2169                         }
2170                         memcpy(path + pos, temp->d_name.name, temp->d_name.len);
2171                 }
2172                 spin_unlock(&temp->d_lock);
2173                 temp = READ_ONCE(temp->d_parent);
2174 
2175                 /* Are we at the root? */
2176                 if (IS_ROOT(temp))
2177                         break;
2178 
2179                 /* Are we out of buffer? */
2180                 if (--pos < 0)
2181                         break;
2182 
2183                 path[pos] = '/';
2184         }
2185         base = ceph_ino(d_inode(temp));
2186         rcu_read_unlock();
2187         if (pos < 0 || read_seqretry(&rename_lock, seq)) {
2188                 pr_err("build_path did not end path lookup where "
2189                        "expected, pos is %d\n", pos);
2190                 /* presumably this is only possible if racing with a
2191                    rename of one of the parent directories (we can not
2192                    lock the dentries above us to prevent this, but
2193                    retrying should be harmless) */
2194                 goto retry;
2195         }
2196 
2197         *pbase = base;
2198         *plen = PATH_MAX - 1 - pos;
2199         dout("build_path on %p %d built %llx '%.*s'\n",
2200              dentry, d_count(dentry), base, *plen, path + pos);
2201         return path + pos;
2202 }
2203 
2204 static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2205                              const char **ppath, int *ppathlen, u64 *pino,
2206                              bool *pfreepath, bool parent_locked)
2207 {
2208         char *path;
2209 
2210         rcu_read_lock();
2211         if (!dir)
2212                 dir = d_inode_rcu(dentry->d_parent);
2213         if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP) {
2214                 *pino = ceph_ino(dir);
2215                 rcu_read_unlock();
2216                 *ppath = dentry->d_name.name;
2217                 *ppathlen = dentry->d_name.len;
2218                 return 0;
2219         }
2220         rcu_read_unlock();
2221         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2222         if (IS_ERR(path))
2223                 return PTR_ERR(path);
2224         *ppath = path;
2225         *pfreepath = true;
2226         return 0;
2227 }
2228 
2229 static int build_inode_path(struct inode *inode,
2230                             const char **ppath, int *ppathlen, u64 *pino,
2231                             bool *pfreepath)
2232 {
2233         struct dentry *dentry;
2234         char *path;
2235 
2236         if (ceph_snap(inode) == CEPH_NOSNAP) {
2237                 *pino = ceph_ino(inode);
2238                 *ppathlen = 0;
2239                 return 0;
2240         }
2241         dentry = d_find_alias(inode);
2242         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2243         dput(dentry);
2244         if (IS_ERR(path))
2245                 return PTR_ERR(path);
2246         *ppath = path;
2247         *pfreepath = true;
2248         return 0;
2249 }
2250 
2251 /*
2252  * request arguments may be specified via an inode *, a dentry *, or
2253  * an explicit ino+path.
2254  */
2255 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
2256                                   struct inode *rdiri, const char *rpath,
2257                                   u64 rino, const char **ppath, int *pathlen,
2258                                   u64 *ino, bool *freepath, bool parent_locked)
2259 {
2260         int r = 0;
2261 
2262         if (rinode) {
2263                 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
2264                 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2265                      ceph_snap(rinode));
2266         } else if (rdentry) {
2267                 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
2268                                         freepath, parent_locked);
2269                 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
2270                      *ppath);
2271         } else if (rpath || rino) {
2272                 *ino = rino;
2273                 *ppath = rpath;
2274                 *pathlen = rpath ? strlen(rpath) : 0;
2275                 dout(" path %.*s\n", *pathlen, rpath);
2276         }
2277 
2278         return r;
2279 }
2280 
2281 /*
2282  * called under mdsc->mutex
2283  */
2284 static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
2285                                                struct ceph_mds_request *req,
2286                                                int mds, bool drop_cap_releases)
2287 {
2288         struct ceph_msg *msg;
2289         struct ceph_mds_request_head *head;
2290         const char *path1 = NULL;
2291         const char *path2 = NULL;
2292         u64 ino1 = 0, ino2 = 0;
2293         int pathlen1 = 0, pathlen2 = 0;
2294         bool freepath1 = false, freepath2 = false;
2295         int len;
2296         u16 releases;
2297         void *p, *end;
2298         int ret;
2299 
2300         ret = set_request_path_attr(req->r_inode, req->r_dentry,
2301                               req->r_parent, req->r_path1, req->r_ino1.ino,
2302                               &path1, &pathlen1, &ino1, &freepath1,
2303                               test_bit(CEPH_MDS_R_PARENT_LOCKED,
2304                                         &req->r_req_flags));
2305         if (ret < 0) {
2306                 msg = ERR_PTR(ret);
2307                 goto out;
2308         }
2309 
2310         /* If r_old_dentry is set, then assume that its parent is locked */
2311         ret = set_request_path_attr(NULL, req->r_old_dentry,
2312                               req->r_old_dentry_dir,
2313                               req->r_path2, req->r_ino2.ino,
2314                               &path2, &pathlen2, &ino2, &freepath2, true);
2315         if (ret < 0) {
2316                 msg = ERR_PTR(ret);
2317                 goto out_free1;
2318         }
2319 
2320         len = sizeof(*head) +
2321                 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
2322                 sizeof(struct ceph_timespec);
2323 
2324         /* calculate (max) length for cap releases */
2325         len += sizeof(struct ceph_mds_request_release) *
2326                 (!!req->r_inode_drop + !!req->r_dentry_drop +
2327                  !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2328         if (req->r_dentry_drop)
2329                 len += pathlen1;
2330         if (req->r_old_dentry_drop)
2331                 len += pathlen2;
2332 
2333         msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
2334         if (!msg) {
2335                 msg = ERR_PTR(-ENOMEM);
2336                 goto out_free2;
2337         }
2338 
2339         msg->hdr.version = cpu_to_le16(2);
2340         msg->hdr.tid = cpu_to_le64(req->r_tid);
2341 
2342         head = msg->front.iov_base;
2343         p = msg->front.iov_base + sizeof(*head);
2344         end = msg->front.iov_base + msg->front.iov_len;
2345 
2346         head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2347         head->op = cpu_to_le32(req->r_op);
2348         head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
2349         head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2350         head->args = req->r_args;
2351 
2352         ceph_encode_filepath(&p, end, ino1, path1);
2353         ceph_encode_filepath(&p, end, ino2, path2);
2354 
2355         /* make note of release offset, in case we need to replay */
2356         req->r_request_release_offset = p - msg->front.iov_base;
2357 
2358         /* cap releases */
2359         releases = 0;
2360         if (req->r_inode_drop)
2361                 releases += ceph_encode_inode_release(&p,
2362                       req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2363                       mds, req->r_inode_drop, req->r_inode_unless, 0);
2364         if (req->r_dentry_drop)
2365                 releases += ceph_encode_dentry_release(&p, req->r_dentry,
2366                                 req->r_parent, mds, req->r_dentry_drop,
2367                                 req->r_dentry_unless);
2368         if (req->r_old_dentry_drop)
2369                 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2370                                 req->r_old_dentry_dir, mds,
2371                                 req->r_old_dentry_drop,
2372                                 req->r_old_dentry_unless);
2373         if (req->r_old_inode_drop)
2374                 releases += ceph_encode_inode_release(&p,
2375                       d_inode(req->r_old_dentry),
2376                       mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
2377 
2378         if (drop_cap_releases) {
2379                 releases = 0;
2380                 p = msg->front.iov_base + req->r_request_release_offset;
2381         }
2382 
2383         head->num_releases = cpu_to_le16(releases);
2384 
2385         /* time stamp */
2386         {
2387                 struct ceph_timespec ts;
2388                 ceph_encode_timespec64(&ts, &req->r_stamp);
2389                 ceph_encode_copy(&p, &ts, sizeof(ts));
2390         }
2391 
2392         BUG_ON(p > end);
2393         msg->front.iov_len = p - msg->front.iov_base;
2394         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2395 
2396         if (req->r_pagelist) {
2397                 struct ceph_pagelist *pagelist = req->r_pagelist;
2398                 ceph_msg_data_add_pagelist(msg, pagelist);
2399                 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2400         } else {
2401                 msg->hdr.data_len = 0;
2402         }
2403 
2404         msg->hdr.data_off = cpu_to_le16(0);
2405 
2406 out_free2:
2407         if (freepath2)
2408                 ceph_mdsc_free_path((char *)path2, pathlen2);
2409 out_free1:
2410         if (freepath1)
2411                 ceph_mdsc_free_path((char *)path1, pathlen1);
2412 out:
2413         return msg;
2414 }
2415 
2416 /*
2417  * called under mdsc->mutex if error, under no mutex if
2418  * success.
2419  */
2420 static void complete_request(struct ceph_mds_client *mdsc,
2421                              struct ceph_mds_request *req)
2422 {
2423         if (req->r_callback)
2424                 req->r_callback(mdsc, req);
2425         complete_all(&req->r_completion);
2426 }
2427 
2428 /*
2429  * called under mdsc->mutex
2430  */
2431 static int __prepare_send_request(struct ceph_mds_client *mdsc,
2432                                   struct ceph_mds_request *req,
2433                                   int mds, bool drop_cap_releases)
2434 {
2435         struct ceph_mds_request_head *rhead;
2436         struct ceph_msg *msg;
2437         int flags = 0;
2438 
2439         req->r_attempts++;
2440         if (req->r_inode) {
2441                 struct ceph_cap *cap =
2442                         ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2443 
2444                 if (cap)
2445                         req->r_sent_on_mseq = cap->mseq;
2446                 else
2447                         req->r_sent_on_mseq = -1;
2448         }
2449         dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2450              req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2451 
2452         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2453                 void *p;
2454                 /*
2455                  * Replay.  Do not regenerate message (and rebuild
2456                  * paths, etc.); just use the original message.
2457                  * Rebuilding paths will break for renames because
2458                  * d_move mangles the src name.
2459                  */
2460                 msg = req->r_request;
2461                 rhead = msg->front.iov_base;
2462 
2463                 flags = le32_to_cpu(rhead->flags);
2464                 flags |= CEPH_MDS_FLAG_REPLAY;
2465                 rhead->flags = cpu_to_le32(flags);
2466 
2467                 if (req->r_target_inode)
2468                         rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2469 
2470                 rhead->num_retry = req->r_attempts - 1;
2471 
2472                 /* remove cap/dentry releases from message */
2473                 rhead->num_releases = 0;
2474 
2475                 /* time stamp */
2476                 p = msg->front.iov_base + req->r_request_release_offset;
2477                 {
2478                         struct ceph_timespec ts;
2479                         ceph_encode_timespec64(&ts, &req->r_stamp);
2480                         ceph_encode_copy(&p, &ts, sizeof(ts));
2481                 }
2482 
2483                 msg->front.iov_len = p - msg->front.iov_base;
2484                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2485                 return 0;
2486         }
2487 
2488         if (req->r_request) {
2489                 ceph_msg_put(req->r_request);
2490                 req->r_request = NULL;
2491         }
2492         msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2493         if (IS_ERR(msg)) {
2494                 req->r_err = PTR_ERR(msg);
2495                 return PTR_ERR(msg);
2496         }
2497         req->r_request = msg;
2498 
2499         rhead = msg->front.iov_base;
2500         rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2501         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2502                 flags |= CEPH_MDS_FLAG_REPLAY;
2503         if (req->r_parent)
2504                 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2505         rhead->flags = cpu_to_le32(flags);
2506         rhead->num_fwd = req->r_num_fwd;
2507         rhead->num_retry = req->r_attempts - 1;
2508         rhead->ino = 0;
2509 
2510         dout(" r_parent = %p\n", req->r_parent);
2511         return 0;
2512 }
2513 
2514 /*
2515  * send request, or put it on the appropriate wait list.
2516  */
2517 static void __do_request(struct ceph_mds_client *mdsc,
2518                         struct ceph_mds_request *req)
2519 {
2520         struct ceph_mds_session *session = NULL;
2521         int mds = -1;
2522         int err = 0;
2523 
2524         if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2525                 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
2526                         __unregister_request(mdsc, req);
2527                 return;
2528         }
2529 
2530         if (req->r_timeout &&
2531             time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2532                 dout("do_request timed out\n");
2533                 err = -EIO;
2534                 goto finish;
2535         }
2536         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2537                 dout("do_request forced umount\n");
2538                 err = -EIO;
2539                 goto finish;
2540         }
2541         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2542                 if (mdsc->mdsmap_err) {
2543                         err = mdsc->mdsmap_err;
2544                         dout("do_request mdsmap err %d\n", err);
2545                         goto finish;
2546                 }
2547                 if (mdsc->mdsmap->m_epoch == 0) {
2548                         dout("do_request no mdsmap, waiting for map\n");
2549                         list_add(&req->r_wait, &mdsc->waiting_for_map);
2550                         return;
2551                 }
2552                 if (!(mdsc->fsc->mount_options->flags &
2553                       CEPH_MOUNT_OPT_MOUNTWAIT) &&
2554                     !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2555                         err = -EHOSTUNREACH;
2556                         goto finish;
2557                 }
2558         }
2559 
2560         put_request_session(req);
2561 
2562         mds = __choose_mds(mdsc, req);
2563         if (mds < 0 ||
2564             ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2565                 dout("do_request no mds or not active, waiting for map\n");
2566                 list_add(&req->r_wait, &mdsc->waiting_for_map);
2567                 return;
2568         }
2569 
2570         /* get, open session */
2571         session = __ceph_lookup_mds_session(mdsc, mds);
2572         if (!session) {
2573                 session = register_session(mdsc, mds);
2574                 if (IS_ERR(session)) {
2575                         err = PTR_ERR(session);
2576                         goto finish;
2577                 }
2578         }
2579         req->r_session = get_session(session);
2580 
2581         dout("do_request mds%d session %p state %s\n", mds, session,
2582              ceph_session_state_name(session->s_state));
2583         if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2584             session->s_state != CEPH_MDS_SESSION_HUNG) {
2585                 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2586                         err = -EACCES;
2587                         goto out_session;
2588                 }
2589                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2590                     session->s_state == CEPH_MDS_SESSION_CLOSING)
2591                         __open_session(mdsc, session);
2592                 list_add(&req->r_wait, &session->s_waiting);
2593                 goto out_session;
2594         }
2595 
2596         /* send request */
2597         req->r_resend_mds = -1;   /* forget any previous mds hint */
2598 
2599         if (req->r_request_started == 0)   /* note request start time */
2600                 req->r_request_started = jiffies;
2601 
2602         err = __prepare_send_request(mdsc, req, mds, false);
2603         if (!err) {
2604                 ceph_msg_get(req->r_request);
2605                 ceph_con_send(&session->s_con, req->r_request);
2606         }
2607 
2608 out_session:
2609         ceph_put_mds_session(session);
2610 finish:
2611         if (err) {
2612                 dout("__do_request early error %d\n", err);
2613                 req->r_err = err;
2614                 complete_request(mdsc, req);
2615                 __unregister_request(mdsc, req);
2616         }
2617         return;
2618 }
2619 
2620 /*
2621  * called under mdsc->mutex
2622  */
2623 static void __wake_requests(struct ceph_mds_client *mdsc,
2624                             struct list_head *head)
2625 {
2626         struct ceph_mds_request *req;
2627         LIST_HEAD(tmp_list);
2628 
2629         list_splice_init(head, &tmp_list);
2630 
2631         while (!list_empty(&tmp_list)) {
2632                 req = list_entry(tmp_list.next,
2633                                  struct ceph_mds_request, r_wait);
2634                 list_del_init(&req->r_wait);
2635                 dout(" wake request %p tid %llu\n", req, req->r_tid);
2636                 __do_request(mdsc, req);
2637         }
2638 }
2639 
2640 /*
2641  * Wake up threads with requests pending for @mds, so that they can
2642  * resubmit their requests to a possibly different mds.
2643  */
2644 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2645 {
2646         struct ceph_mds_request *req;
2647         struct rb_node *p = rb_first(&mdsc->request_tree);
2648 
2649         dout("kick_requests mds%d\n", mds);
2650         while (p) {
2651                 req = rb_entry(p, struct ceph_mds_request, r_node);
2652                 p = rb_next(p);
2653                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2654                         continue;
2655                 if (req->r_attempts > 0)
2656                         continue; /* only new requests */
2657                 if (req->r_session &&
2658                     req->r_session->s_mds == mds) {
2659                         dout(" kicking tid %llu\n", req->r_tid);
2660                         list_del_init(&req->r_wait);
2661                         __do_request(mdsc, req);
2662                 }
2663         }
2664 }
2665 
2666 int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
2667                               struct ceph_mds_request *req)
2668 {
2669         int err;
2670 
2671         /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
2672         if (req->r_inode)
2673                 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2674         if (req->r_parent) {
2675                 ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
2676                 ihold(req->r_parent);
2677         }
2678         if (req->r_old_dentry_dir)
2679                 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2680                                   CEPH_CAP_PIN);
2681 
2682         dout("submit_request on %p for inode %p\n", req, dir);
2683         mutex_lock(&mdsc->mutex);
2684         __register_request(mdsc, req, dir);
2685         __do_request(mdsc, req);
2686         err = req->r_err;
2687         mutex_unlock(&mdsc->mutex);
2688         return err;
2689 }
2690 
2691 static int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
2692                                   struct ceph_mds_request *req)
2693 {
2694         int err;
2695 
2696         /* wait */
2697         dout("do_request waiting\n");
2698         if (!req->r_timeout && req->r_wait_for_completion) {
2699                 err = req->r_wait_for_completion(mdsc, req);
2700         } else {
2701                 long timeleft = wait_for_completion_killable_timeout(
2702                                         &req->r_completion,
2703                                         ceph_timeout_jiffies(req->r_timeout));
2704                 if (timeleft > 0)
2705                         err = 0;
2706                 else if (!timeleft)
2707                         err = -EIO;  /* timed out */
2708                 else
2709                         err = timeleft;  /* killed */
2710         }
2711         dout("do_request waited, got %d\n", err);
2712         mutex_lock(&mdsc->mutex);
2713 
2714         /* only abort if we didn't race with a real reply */
2715         if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2716                 err = le32_to_cpu(req->r_reply_info.head->result);
2717         } else if (err < 0) {
2718                 dout("aborted request %lld with %d\n", req->r_tid, err);
2719 
2720                 /*
2721                  * ensure we aren't running concurrently with
2722                  * ceph_fill_trace or ceph_readdir_prepopulate, which
2723                  * rely on locks (dir mutex) held by our caller.
2724                  */
2725                 mutex_lock(&req->r_fill_mutex);
2726                 req->r_err = err;
2727                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
2728                 mutex_unlock(&req->r_fill_mutex);
2729 
2730                 if (req->r_parent &&
2731                     (req->r_op & CEPH_MDS_OP_WRITE))
2732                         ceph_invalidate_dir_request(req);
2733         } else {
2734                 err = req->r_err;
2735         }
2736 
2737         mutex_unlock(&mdsc->mutex);
2738         return err;
2739 }
2740 
2741 /*
2742  * Synchrously perform an mds request.  Take care of all of the
2743  * session setup, forwarding, retry details.
2744  */
2745 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2746                          struct inode *dir,
2747                          struct ceph_mds_request *req)
2748 {
2749         int err;
2750 
2751         dout("do_request on %p\n", req);
2752 
2753         /* issue */
2754         err = ceph_mdsc_submit_request(mdsc, dir, req);
2755         if (!err)
2756                 err = ceph_mdsc_wait_request(mdsc, req);
2757         dout("do_request %p done, result %d\n", req, err);
2758         return err;
2759 }
2760 
2761 /*
2762  * Invalidate dir's completeness, dentry lease state on an aborted MDS
2763  * namespace request.
2764  */
2765 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2766 {
2767         struct inode *dir = req->r_parent;
2768         struct inode *old_dir = req->r_old_dentry_dir;
2769 
2770         dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
2771 
2772         ceph_dir_clear_complete(dir);
2773         if (old_dir)
2774                 ceph_dir_clear_complete(old_dir);
2775         if (req->r_dentry)
2776                 ceph_invalidate_dentry_lease(req->r_dentry);
2777         if (req->r_old_dentry)
2778                 ceph_invalidate_dentry_lease(req->r_old_dentry);
2779 }
2780 
2781 /*
2782  * Handle mds reply.
2783  *
2784  * We take the session mutex and parse and process the reply immediately.
2785  * This preserves the logical ordering of replies, capabilities, etc., sent
2786  * by the MDS as they are applied to our local cache.
2787  */
2788 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2789 {
2790         struct ceph_mds_client *mdsc = session->s_mdsc;
2791         struct ceph_mds_request *req;
2792         struct ceph_mds_reply_head *head = msg->front.iov_base;
2793         struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
2794         struct ceph_snap_realm *realm;
2795         u64 tid;
2796         int err, result;
2797         int mds = session->s_mds;
2798 
2799         if (msg->front.iov_len < sizeof(*head)) {
2800                 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2801                 ceph_msg_dump(msg);
2802                 return;
2803         }
2804 
2805         /* get request, session */
2806         tid = le64_to_cpu(msg->hdr.tid);
2807         mutex_lock(&mdsc->mutex);
2808         req = lookup_get_request(mdsc, tid);
2809         if (!req) {
2810                 dout("handle_reply on unknown tid %llu\n", tid);
2811                 mutex_unlock(&mdsc->mutex);
2812                 return;
2813         }
2814         dout("handle_reply %p\n", req);
2815 
2816         /* correct session? */
2817         if (req->r_session != session) {
2818                 pr_err("mdsc_handle_reply got %llu on session mds%d"
2819                        " not mds%d\n", tid, session->s_mds,
2820                        req->r_session ? req->r_session->s_mds : -1);
2821                 mutex_unlock(&mdsc->mutex);
2822                 goto out;
2823         }
2824 
2825         /* dup? */
2826         if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
2827             (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
2828                 pr_warn("got a dup %s reply on %llu from mds%d\n",
2829                            head->safe ? "safe" : "unsafe", tid, mds);
2830                 mutex_unlock(&mdsc->mutex);
2831                 goto out;
2832         }
2833         if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
2834                 pr_warn("got unsafe after safe on %llu from mds%d\n",
2835                            tid, mds);
2836                 mutex_unlock(&mdsc->mutex);
2837                 goto out;
2838         }
2839 
2840         result = le32_to_cpu(head->result);
2841 
2842         /*
2843          * Handle an ESTALE
2844          * if we're not talking to the authority, send to them
2845          * if the authority has changed while we weren't looking,
2846          * send to new authority
2847          * Otherwise we just have to return an ESTALE
2848          */
2849         if (result == -ESTALE) {
2850                 dout("got ESTALE on request %llu\n", req->r_tid);
2851                 req->r_resend_mds = -1;
2852                 if (req->r_direct_mode != USE_AUTH_MDS) {
2853                         dout("not using auth, setting for that now\n");
2854                         req->r_direct_mode = USE_AUTH_MDS;
2855                         __do_request(mdsc, req);
2856                         mutex_unlock(&mdsc->mutex);
2857                         goto out;
2858                 } else  {
2859                         int mds = __choose_mds(mdsc, req);
2860                         if (mds >= 0 && mds != req->r_session->s_mds) {
2861                                 dout("but auth changed, so resending\n");
2862                                 __do_request(mdsc, req);
2863                                 mutex_unlock(&mdsc->mutex);
2864                                 goto out;
2865                         }
2866                 }
2867                 dout("have to return ESTALE on request %llu\n", req->r_tid);
2868         }
2869 
2870 
2871         if (head->safe) {
2872                 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
2873                 __unregister_request(mdsc, req);
2874 
2875                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2876                         /*
2877                          * We already handled the unsafe response, now do the
2878                          * cleanup.  No need to examine the response; the MDS
2879                          * doesn't include any result info in the safe
2880                          * response.  And even if it did, there is nothing
2881                          * useful we could do with a revised return value.
2882                          */
2883                         dout("got safe reply %llu, mds%d\n", tid, mds);
2884 
2885                         /* last unsafe request during umount? */
2886                         if (mdsc->stopping && !__get_oldest_req(mdsc))
2887                                 complete_all(&mdsc->safe_umount_waiters);
2888                         mutex_unlock(&mdsc->mutex);
2889                         goto out;
2890                 }
2891         } else {
2892                 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
2893                 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2894                 if (req->r_unsafe_dir) {
2895                         struct ceph_inode_info *ci =
2896                                         ceph_inode(req->r_unsafe_dir);
2897                         spin_lock(&ci->i_unsafe_lock);
2898                         list_add_tail(&req->r_unsafe_dir_item,
2899                                       &ci->i_unsafe_dirops);
2900                         spin_unlock(&ci->i_unsafe_lock);
2901                 }
2902         }
2903 
2904         dout("handle_reply tid %lld result %d\n", tid, result);
2905         rinfo = &req->r_reply_info;
2906         if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
2907                 err = parse_reply_info(msg, rinfo, (u64)-1);
2908         else
2909                 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2910         mutex_unlock(&mdsc->mutex);
2911 
2912         mutex_lock(&session->s_mutex);
2913         if (err < 0) {
2914                 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
2915                 ceph_msg_dump(msg);
2916                 goto out_err;
2917         }
2918 
2919         /* snap trace */
2920         realm = NULL;
2921         if (rinfo->snapblob_len) {
2922                 down_write(&mdsc->snap_rwsem);
2923                 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2924                                 rinfo->snapblob + rinfo->snapblob_len,
2925                                 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2926                                 &realm);
2927                 downgrade_write(&mdsc->snap_rwsem);
2928         } else {
2929                 down_read(&mdsc->snap_rwsem);
2930         }
2931 
2932         /* insert trace into our cache */
2933         mutex_lock(&req->r_fill_mutex);
2934         current->journal_info = req;
2935         err = ceph_fill_trace(mdsc->fsc->sb, req);
2936         if (err == 0) {
2937                 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
2938                                     req->r_op == CEPH_MDS_OP_LSSNAP))
2939                         ceph_readdir_prepopulate(req, req->r_session);
2940         }
2941         current->journal_info = NULL;
2942         mutex_unlock(&req->r_fill_mutex);
2943 
2944         up_read(&mdsc->snap_rwsem);
2945         if (realm)
2946                 ceph_put_snap_realm(mdsc, realm);
2947 
2948         if (err == 0) {
2949                 if (req->r_target_inode &&
2950                     test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2951                         struct ceph_inode_info *ci =
2952                                 ceph_inode(req->r_target_inode);
2953                         spin_lock(&ci->i_unsafe_lock);
2954                         list_add_tail(&req->r_unsafe_target_item,
2955                                       &ci->i_unsafe_iops);
2956                         spin_unlock(&ci->i_unsafe_lock);
2957                 }
2958 
2959                 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2960         }
2961 out_err:
2962         mutex_lock(&mdsc->mutex);
2963         if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
2964                 if (err) {
2965                         req->r_err = err;
2966                 } else {
2967                         req->r_reply =  ceph_msg_get(msg);
2968                         set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
2969                 }
2970         } else {
2971                 dout("reply arrived after request %lld was aborted\n", tid);
2972         }
2973         mutex_unlock(&mdsc->mutex);
2974 
2975         mutex_unlock(&session->s_mutex);
2976 
2977         /* kick calling process */
2978         complete_request(mdsc, req);
2979 out:
2980         ceph_mdsc_put_request(req);
2981         return;
2982 }
2983 
2984 
2985 
2986 /*
2987  * handle mds notification that our request has been forwarded.
2988  */
2989 static void handle_forward(struct ceph_mds_client *mdsc,
2990                            struct ceph_mds_session *session,
2991                            struct ceph_msg *msg)
2992 {
2993         struct ceph_mds_request *req;
2994         u64 tid = le64_to_cpu(msg->hdr.tid);
2995         u32 next_mds;
2996         u32 fwd_seq;
2997         int err = -EINVAL;
2998         void *p = msg->front.iov_base;
2999         void *end = p + msg->front.iov_len;
3000 
3001         ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3002         next_mds = ceph_decode_32(&p);
3003         fwd_seq = ceph_decode_32(&p);
3004 
3005         mutex_lock(&mdsc->mutex);
3006         req = lookup_get_request(mdsc, tid);
3007         if (!req) {
3008                 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
3009                 goto out;  /* dup reply? */
3010         }
3011 
3012         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3013                 dout("forward tid %llu aborted, unregistering\n", tid);
3014                 __unregister_request(mdsc, req);
3015         } else if (fwd_seq <= req->r_num_fwd) {
3016                 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
3017                      tid, next_mds, req->r_num_fwd, fwd_seq);
3018         } else {
3019                 /* resend. forward race not possible; mds would drop */
3020                 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
3021                 BUG_ON(req->r_err);
3022                 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3023                 req->r_attempts = 0;
3024                 req->r_num_fwd = fwd_seq;
3025                 req->r_resend_mds = next_mds;
3026                 put_request_session(req);
3027                 __do_request(mdsc, req);
3028         }
3029         ceph_mdsc_put_request(req);
3030 out:
3031         mutex_unlock(&mdsc->mutex);
3032         return;
3033 
3034 bad:
3035         pr_err("mdsc_handle_forward decode error err=%d\n", err);
3036 }
3037 
3038 static int __decode_session_metadata(void **p, void *end,
3039                                      bool *blacklisted)
3040 {
3041         /* map<string,string> */
3042         u32 n;
3043         bool err_str;
3044         ceph_decode_32_safe(p, end, n, bad);
3045         while (n-- > 0) {
3046                 u32 len;
3047                 ceph_decode_32_safe(p, end, len, bad);
3048                 ceph_decode_need(p, end, len, bad);
3049                 err_str = !strncmp(*p, "error_string", len);
3050                 *p += len;
3051                 ceph_decode_32_safe(p, end, len, bad);
3052                 ceph_decode_need(p, end, len, bad);
3053                 if (err_str && strnstr(*p, "blacklisted", len))
3054                         *blacklisted = true;
3055                 *p += len;
3056         }
3057         return 0;
3058 bad:
3059         return -1;
3060 }
3061 
3062 /*
3063  * handle a mds session control message
3064  */
3065 static void handle_session(struct ceph_mds_session *session,
3066                            struct ceph_msg *msg)
3067 {
3068         struct ceph_mds_client *mdsc = session->s_mdsc;
3069         int mds = session->s_mds;
3070         int msg_version = le16_to_cpu(msg->hdr.version);
3071         void *p = msg->front.iov_base;
3072         void *end = p + msg->front.iov_len;
3073         struct ceph_mds_session_head *h;
3074         u32 op;
3075         u64 seq, features = 0;
3076         int wake = 0;
3077         bool blacklisted = false;
3078 
3079         /* decode */
3080         ceph_decode_need(&p, end, sizeof(*h), bad);
3081         h = p;
3082         p += sizeof(*h);
3083 
3084         op = le32_to_cpu(h->op);
3085         seq = le64_to_cpu(h->seq);
3086 
3087         if (msg_version >= 3) {
3088                 u32 len;
3089                 /* version >= 2, metadata */
3090                 if (__decode_session_metadata(&p, end, &blacklisted) < 0)
3091                         goto bad;
3092                 /* version >= 3, feature bits */
3093                 ceph_decode_32_safe(&p, end, len, bad);
3094                 ceph_decode_64_safe(&p, end, features, bad);
3095                 p += len - sizeof(features);
3096         }
3097 
3098         mutex_lock(&mdsc->mutex);
3099         if (op == CEPH_SESSION_CLOSE) {
3100                 get_session(session);
3101                 __unregister_session(mdsc, session);
3102         }
3103         /* FIXME: this ttl calculation is generous */
3104         session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
3105         mutex_unlock(&mdsc->mutex);
3106 
3107         mutex_lock(&session->s_mutex);
3108 
3109         dout("handle_session mds%d %s %p state %s seq %llu\n",
3110              mds, ceph_session_op_name(op), session,
3111              ceph_session_state_name(session->s_state), seq);
3112 
3113         if (session->s_state == CEPH_MDS_SESSION_HUNG) {
3114                 session->s_state = CEPH_MDS_SESSION_OPEN;
3115                 pr_info("mds%d came back\n", session->s_mds);
3116         }
3117 
3118         switch (op) {
3119         case CEPH_SESSION_OPEN:
3120                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3121                         pr_info("mds%d reconnect success\n", session->s_mds);
3122                 session->s_state = CEPH_MDS_SESSION_OPEN;
3123                 session->s_features = features;
3124                 renewed_caps(mdsc, session, 0);
3125                 wake = 1;
3126                 if (mdsc->stopping)
3127                         __close_session(mdsc, session);
3128                 break;
3129 
3130         case CEPH_SESSION_RENEWCAPS:
3131                 if (session->s_renew_seq == seq)
3132                         renewed_caps(mdsc, session, 1);
3133                 break;
3134 
3135         case CEPH_SESSION_CLOSE:
3136                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3137                         pr_info("mds%d reconnect denied\n", session->s_mds);
3138                 cleanup_session_requests(mdsc, session);
3139                 remove_session_caps(session);
3140                 wake = 2; /* for good measure */
3141                 wake_up_all(&mdsc->session_close_wq);
3142                 break;
3143 
3144         case CEPH_SESSION_STALE:
3145                 pr_info("mds%d caps went stale, renewing\n",
3146                         session->s_mds);
3147                 spin_lock(&session->s_gen_ttl_lock);
3148                 session->s_cap_gen++;
3149                 session->s_cap_ttl = jiffies - 1;
3150                 spin_unlock(&session->s_gen_ttl_lock);
3151                 send_renew_caps(mdsc, session);
3152                 break;
3153 
3154         case CEPH_SESSION_RECALL_STATE:
3155                 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
3156                 break;
3157 
3158         case CEPH_SESSION_FLUSHMSG:
3159                 send_flushmsg_ack(mdsc, session, seq);
3160                 break;
3161 
3162         case CEPH_SESSION_FORCE_RO:
3163                 dout("force_session_readonly %p\n", session);
3164                 spin_lock(&session->s_cap_lock);
3165                 session->s_readonly = true;
3166                 spin_unlock(&session->s_cap_lock);
3167                 wake_up_session_caps(session, FORCE_RO);
3168                 break;
3169 
3170         case CEPH_SESSION_REJECT:
3171                 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
3172                 pr_info("mds%d rejected session\n", session->s_mds);
3173                 session->s_state = CEPH_MDS_SESSION_REJECTED;
3174                 cleanup_session_requests(mdsc, session);
3175                 remove_session_caps(session);
3176                 if (blacklisted)
3177                         mdsc->fsc->blacklisted = true;
3178                 wake = 2; /* for good measure */
3179                 break;
3180 
3181         default:
3182                 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
3183                 WARN_ON(1);
3184         }
3185 
3186         mutex_unlock(&session->s_mutex);
3187         if (wake) {
3188                 mutex_lock(&mdsc->mutex);
3189                 __wake_requests(mdsc, &session->s_waiting);
3190                 if (wake == 2)
3191                         kick_requests(mdsc, mds);
3192                 mutex_unlock(&mdsc->mutex);
3193         }
3194         if (op == CEPH_SESSION_CLOSE)
3195                 ceph_put_mds_session(session);
3196         return;
3197 
3198 bad:
3199         pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
3200                (int)msg->front.iov_len);
3201         ceph_msg_dump(msg);
3202         return;
3203 }
3204 
3205 
3206 /*
3207  * called under session->mutex.
3208  */
3209 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
3210                                    struct ceph_mds_session *session)
3211 {
3212         struct ceph_mds_request *req, *nreq;
3213         struct rb_node *p;
3214         int err;
3215 
3216         dout("replay_unsafe_requests mds%d\n", session->s_mds);
3217 
3218         mutex_lock(&mdsc->mutex);
3219         list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
3220                 err = __prepare_send_request(mdsc, req, session->s_mds, true);
3221                 if (!err) {
3222                         ceph_msg_get(req->r_request);
3223                         ceph_con_send(&session->s_con, req->r_request);
3224                 }
3225         }
3226 
3227         /*
3228          * also re-send old requests when MDS enters reconnect stage. So that MDS
3229          * can process completed request in clientreplay stage.
3230          */
3231         p = rb_first(&mdsc->request_tree);
3232         while (p) {
3233                 req = rb_entry(p, struct ceph_mds_request, r_node);
3234                 p = rb_next(p);
3235                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3236                         continue;
3237                 if (req->r_attempts == 0)
3238                         continue; /* only old requests */
3239                 if (req->r_session &&
3240                     req->r_session->s_mds == session->s_mds) {
3241                         err = __prepare_send_request(mdsc, req,
3242                                                      session->s_mds, true);
3243                         if (!err) {
3244                                 ceph_msg_get(req->r_request);
3245                                 ceph_con_send(&session->s_con, req->r_request);
3246                         }
3247                 }
3248         }
3249         mutex_unlock(&mdsc->mutex);
3250 }
3251 
3252 static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
3253 {
3254         struct ceph_msg *reply;
3255         struct ceph_pagelist *_pagelist;
3256         struct page *page;
3257         __le32 *addr;
3258         int err = -ENOMEM;
3259 
3260         if (!recon_state->allow_multi)
3261                 return -ENOSPC;
3262 
3263         /* can't handle message that contains both caps and realm */
3264         BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
3265 
3266         /* pre-allocate new pagelist */
3267         _pagelist = ceph_pagelist_alloc(GFP_NOFS);
3268         if (!_pagelist)
3269                 return -ENOMEM;
3270 
3271         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3272         if (!reply)
3273                 goto fail_msg;
3274 
3275         /* placeholder for nr_caps */
3276         err = ceph_pagelist_encode_32(_pagelist, 0);
3277         if (err < 0)
3278                 goto fail;
3279 
3280         if (recon_state->nr_caps) {
3281                 /* currently encoding caps */
3282                 err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
3283                 if (err)
3284                         goto fail;
3285         } else {
3286                 /* placeholder for nr_realms (currently encoding relams) */
3287                 err = ceph_pagelist_encode_32(_pagelist, 0);
3288                 if (err < 0)
3289                         goto fail;
3290         }
3291 
3292         err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
3293         if (err)
3294                 goto fail;
3295 
3296         page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
3297         addr = kmap_atomic(page);
3298         if (recon_state->nr_caps) {
3299                 /* currently encoding caps */
3300                 *addr = cpu_to_le32(recon_state->nr_caps);
3301         } else {
3302                 /* currently encoding relams */
3303                 *(addr + 1) = cpu_to_le32(recon_state->nr_realms);
3304         }
3305         kunmap_atomic(addr);
3306 
3307         reply->hdr.version = cpu_to_le16(5);
3308         reply->hdr.compat_version = cpu_to_le16(4);
3309 
3310         reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
3311         ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
3312 
3313         ceph_con_send(&recon_state->session->s_con, reply);
3314         ceph_pagelist_release(recon_state->pagelist);
3315 
3316         recon_state->pagelist = _pagelist;
3317         recon_state->nr_caps = 0;
3318         recon_state->nr_realms = 0;
3319         recon_state->msg_version = 5;
3320         return 0;
3321 fail:
3322         ceph_msg_put(reply);
3323 fail_msg:
3324         ceph_pagelist_release(_pagelist);
3325         return err;
3326 }
3327 
3328 /*
3329  * Encode information about a cap for a reconnect with the MDS.
3330  */
3331 static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
3332                           void *arg)
3333 {
3334         union {
3335                 struct ceph_mds_cap_reconnect v2;
3336                 struct ceph_mds_cap_reconnect_v1 v1;
3337         } rec;
3338         struct ceph_inode_info *ci = cap->ci;
3339         struct ceph_reconnect_state *recon_state = arg;
3340         struct ceph_pagelist *pagelist = recon_state->pagelist;
3341         int err;
3342         u64 snap_follows;
3343 
3344         dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
3345              inode, ceph_vinop(inode), cap, cap->cap_id,
3346              ceph_cap_string(cap->issued));
3347 
3348         spin_lock(&ci->i_ceph_lock);
3349         cap->seq = 0;        /* reset cap seq */
3350         cap->issue_seq = 0;  /* and issue_seq */
3351         cap->mseq = 0;       /* and migrate_seq */
3352         cap->cap_gen = cap->session->s_cap_gen;
3353 
3354         if (recon_state->msg_version >= 2) {
3355                 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
3356                 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3357                 rec.v2.issued = cpu_to_le32(cap->issued);
3358                 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3359                 rec.v2.pathbase = 0;
3360                 rec.v2.flock_len = (__force __le32)
3361                         ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
3362         } else {
3363                 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
3364                 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3365                 rec.v1.issued = cpu_to_le32(cap->issued);
3366                 rec.v1.size = cpu_to_le64(inode->i_size);
3367                 ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
3368                 ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
3369                 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3370                 rec.v1.pathbase = 0;
3371         }
3372 
3373         if (list_empty(&ci->i_cap_snaps)) {
3374                 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3375         } else {
3376                 struct ceph_cap_snap *capsnap =
3377                         list_first_entry(&ci->i_cap_snaps,
3378                                          struct ceph_cap_snap, ci_item);
3379                 snap_follows = capsnap->follows;
3380         }
3381         spin_unlock(&ci->i_ceph_lock);
3382 
3383         if (recon_state->msg_version >= 2) {
3384                 int num_fcntl_locks, num_flock_locks;
3385                 struct ceph_filelock *flocks = NULL;
3386                 size_t struct_len, total_len = sizeof(u64);
3387                 u8 struct_v = 0;
3388 
3389 encode_again:
3390                 if (rec.v2.flock_len) {
3391                         ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
3392                 } else {
3393                         num_fcntl_locks = 0;
3394                         num_flock_locks = 0;
3395                 }
3396                 if (num_fcntl_locks + num_flock_locks > 0) {
3397                         flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
3398                                                sizeof(struct ceph_filelock),
3399                                                GFP_NOFS);
3400                         if (!flocks) {
3401                                 err = -ENOMEM;
3402                                 goto out_err;
3403                         }
3404                         err = ceph_encode_locks_to_buffer(inode, flocks,
3405                                                           num_fcntl_locks,
3406                                                           num_flock_locks);
3407                         if (err) {
3408                                 kfree(flocks);
3409                                 flocks = NULL;
3410                                 if (err == -ENOSPC)
3411                                         goto encode_again;
3412                                 goto out_err;
3413                         }
3414                 } else {
3415                         kfree(flocks);
3416                         flocks = NULL;
3417                 }
3418 
3419                 if (recon_state->msg_version >= 3) {
3420                         /* version, compat_version and struct_len */
3421                         total_len += 2 * sizeof(u8) + sizeof(u32);
3422                         struct_v = 2;
3423                 }
3424                 /*
3425                  * number of encoded locks is stable, so copy to pagelist
3426                  */
3427                 struct_len = 2 * sizeof(u32) +
3428                             (num_fcntl_locks + num_flock_locks) *
3429                             sizeof(struct ceph_filelock);
3430                 rec.v2.flock_len = cpu_to_le32(struct_len);
3431 
3432                 struct_len += sizeof(u32) + sizeof(rec.v2);
3433 
3434                 if (struct_v >= 2)
3435                         struct_len += sizeof(u64); /* snap_follows */
3436 
3437                 total_len += struct_len;
3438 
3439                 if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
3440                         err = send_reconnect_partial(recon_state);
3441                         if (err)
3442                                 goto out_freeflocks;
3443                         pagelist = recon_state->pagelist;
3444                 }
3445 
3446                 err = ceph_pagelist_reserve(pagelist, total_len);
3447                 if (err)
3448                         goto out_freeflocks;
3449 
3450                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3451                 if (recon_state->msg_version >= 3) {
3452                         ceph_pagelist_encode_8(pagelist, struct_v);
3453                         ceph_pagelist_encode_8(pagelist, 1);
3454                         ceph_pagelist_encode_32(pagelist, struct_len);
3455                 }
3456                 ceph_pagelist_encode_string(pagelist, NULL, 0);
3457                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3458                 ceph_locks_to_pagelist(flocks, pagelist,
3459                                        num_fcntl_locks, num_flock_locks);
3460                 if (struct_v >= 2)
3461                         ceph_pagelist_encode_64(pagelist, snap_follows);
3462 out_freeflocks:
3463                 kfree(flocks);
3464         } else {
3465                 u64 pathbase = 0;
3466                 int pathlen = 0;
3467                 char *path = NULL;
3468                 struct dentry *dentry;
3469 
3470                 dentry = d_find_alias(inode);
3471                 if (dentry) {
3472                         path = ceph_mdsc_build_path(dentry,
3473                                                 &pathlen, &pathbase, 0);
3474                         dput(dentry);
3475                         if (IS_ERR(path)) {
3476                                 err = PTR_ERR(path);
3477                                 goto out_err;
3478                         }
3479                         rec.v1.pathbase = cpu_to_le64(pathbase);
3480                 }
3481 
3482                 err = ceph_pagelist_reserve(pagelist,
3483                                             sizeof(u64) + sizeof(u32) +
3484                                             pathlen + sizeof(rec.v1));
3485                 if (err) {
3486                         goto out_freepath;
3487                 }
3488 
3489                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3490                 ceph_pagelist_encode_string(pagelist, path, pathlen);
3491                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3492 out_freepath:
3493                 ceph_mdsc_free_path(path, pathlen);
3494         }
3495 
3496 out_err:
3497         if (err >= 0)
3498                 recon_state->nr_caps++;
3499         return err;
3500 }
3501 
3502 static int encode_snap_realms(struct ceph_mds_client *mdsc,
3503                               struct ceph_reconnect_state *recon_state)
3504 {
3505         struct rb_node *p;
3506         struct ceph_pagelist *pagelist = recon_state->pagelist;
3507         int err = 0;
3508 
3509         if (recon_state->msg_version >= 4) {
3510                 err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
3511                 if (err < 0)
3512                         goto fail;
3513         }
3514 
3515         /*
3516          * snaprealms.  we provide mds with the ino, seq (version), and
3517          * parent for all of our realms.  If the mds has any newer info,
3518          * it will tell us.
3519          */
3520         for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3521                 struct ceph_snap_realm *realm =
3522                        rb_entry(p, struct ceph_snap_realm, node);
3523                 struct ceph_mds_snaprealm_reconnect sr_rec;
3524 
3525                 if (recon_state->msg_version >= 4) {
3526                         size_t need = sizeof(u8) * 2 + sizeof(u32) +
3527                                       sizeof(sr_rec);
3528 
3529                         if (pagelist->length + need > RECONNECT_MAX_SIZE) {
3530                                 err = send_reconnect_partial(recon_state);
3531                                 if (err)
3532                                         goto fail;
3533                                 pagelist = recon_state->pagelist;
3534                         }
3535 
3536                         err = ceph_pagelist_reserve(pagelist, need);
3537                         if (err)
3538                                 goto fail;
3539 
3540                         ceph_pagelist_encode_8(pagelist, 1);
3541                         ceph_pagelist_encode_8(pagelist, 1);
3542                         ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
3543                 }
3544 
3545                 dout(" adding snap realm %llx seq %lld parent %llx\n",
3546                      realm->ino, realm->seq, realm->parent_ino);
3547                 sr_rec.ino = cpu_to_le64(realm->ino);
3548                 sr_rec.seq = cpu_to_le64(realm->seq);
3549                 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3550 
3551                 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3552                 if (err)
3553                         goto fail;
3554 
3555                 recon_state->nr_realms++;
3556         }
3557 fail:
3558         return err;
3559 }
3560 
3561 
3562 /*
3563  * If an MDS fails and recovers, clients need to reconnect in order to
3564  * reestablish shared state.  This includes all caps issued through
3565  * this session _and_ the snap_realm hierarchy.  Because it's not
3566  * clear which snap realms the mds cares about, we send everything we
3567  * know about.. that ensures we'll then get any new info the
3568  * recovering MDS might have.
3569  *
3570  * This is a relatively heavyweight operation, but it's rare.
3571  *
3572  * called with mdsc->mutex held.
3573  */
3574 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
3575                                struct ceph_mds_session *session)
3576 {
3577         struct ceph_msg *reply;
3578         int mds = session->s_mds;
3579         int err = -ENOMEM;
3580         struct ceph_reconnect_state recon_state = {
3581                 .session = session,
3582         };
3583         LIST_HEAD(dispose);
3584 
3585         pr_info("mds%d reconnect start\n", mds);
3586 
3587         recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
3588         if (!recon_state.pagelist)
3589                 goto fail_nopagelist;
3590 
3591         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3592         if (!reply)
3593                 goto fail_nomsg;
3594 
3595         mutex_lock(&session->s_mutex);
3596         session->s_state = CEPH_MDS_SESSION_RECONNECTING;
3597         session->s_seq = 0;
3598 
3599         dout("session %p state %s\n", session,
3600              ceph_session_state_name(session->s_state));
3601 
3602         spin_lock(&session->s_gen_ttl_lock);
3603         session->s_cap_gen++;
3604         spin_unlock(&session->s_gen_ttl_lock);
3605 
3606         spin_lock(&session->s_cap_lock);
3607         /* don't know if session is readonly */
3608         session->s_readonly = 0;
3609         /*
3610          * notify __ceph_remove_cap() that we are composing cap reconnect.
3611          * If a cap get released before being added to the cap reconnect,
3612          * __ceph_remove_cap() should skip queuing cap release.
3613          */
3614         session->s_cap_reconnect = 1;
3615         /* drop old cap expires; we're about to reestablish that state */
3616         detach_cap_releases(session, &dispose);
3617         spin_unlock(&session->s_cap_lock);
3618         dispose_cap_releases(mdsc, &dispose);
3619 
3620         /* trim unused caps to reduce MDS's cache rejoin time */
3621         if (mdsc->fsc->sb->s_root)
3622                 shrink_dcache_parent(mdsc->fsc->sb->s_root);
3623 
3624         ceph_con_close(&session->s_con);
3625         ceph_con_open(&session->s_con,
3626                       CEPH_ENTITY_TYPE_MDS, mds,
3627                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
3628 
3629         /* replay unsafe requests */
3630         replay_unsafe_requests(mdsc, session);
3631 
3632         ceph_early_kick_flushing_caps(mdsc, session);
3633 
3634         down_read(&mdsc->snap_rwsem);
3635 
3636         /* placeholder for nr_caps */
3637         err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
3638         if (err)
3639                 goto fail;
3640 
3641         if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
3642                 recon_state.msg_version = 3;
3643                 recon_state.allow_multi = true;
3644         } else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
3645                 recon_state.msg_version = 3;
3646         } else {
3647                 recon_state.msg_version = 2;
3648         }
3649         /* trsaverse this session's caps */
3650         err = ceph_iterate_session_caps(session, encode_caps_cb, &recon_state);
3651 
3652         spin_lock(&session->s_cap_lock);
3653         session->s_cap_reconnect = 0;
3654         spin_unlock(&session->s_cap_lock);
3655 
3656         if (err < 0)
3657                 goto fail;
3658 
3659         /* check if all realms can be encoded into current message */
3660         if (mdsc->num_snap_realms) {
3661                 size_t total_len =
3662                         recon_state.pagelist->length +
3663                         mdsc->num_snap_realms *
3664                         sizeof(struct ceph_mds_snaprealm_reconnect);
3665                 if (recon_state.msg_version >= 4) {
3666                         /* number of realms */
3667                         total_len += sizeof(u32);
3668                         /* version, compat_version and struct_len */
3669                         total_len += mdsc->num_snap_realms *
3670                                      (2 * sizeof(u8) + sizeof(u32));
3671                 }
3672                 if (total_len > RECONNECT_MAX_SIZE) {
3673                         if (!recon_state.allow_multi) {
3674                                 err = -ENOSPC;
3675                                 goto fail;
3676                         }
3677                         if (recon_state.nr_caps) {
3678                                 err = send_reconnect_partial(&recon_state);
3679                                 if (err)
3680                                         goto fail;
3681                         }
3682                         recon_state.msg_version = 5;
3683                 }
3684         }
3685 
3686         err = encode_snap_realms(mdsc, &recon_state);
3687         if (err < 0)
3688                 goto fail;
3689 
3690         if (recon_state.msg_version >= 5) {
3691                 err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
3692                 if (err < 0)
3693                         goto fail;
3694         }
3695 
3696         if (recon_state.nr_caps || recon_state.nr_realms) {
3697                 struct page *page =
3698                         list_first_entry(&recon_state.pagelist->head,
3699                                         struct page, lru);
3700                 __le32 *addr = kmap_atomic(page);
3701                 if (recon_state.nr_caps) {
3702                         WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
3703                         *addr = cpu_to_le32(recon_state.nr_caps);
3704                 } else if (recon_state.msg_version >= 4) {
3705                         *(addr + 1) = cpu_to_le32(recon_state.nr_realms);
3706                 }
3707                 kunmap_atomic(addr);
3708         }
3709 
3710         reply->hdr.version = cpu_to_le16(recon_state.msg_version);
3711         if (recon_state.msg_version >= 4)
3712                 reply->hdr.compat_version = cpu_to_le16(4);
3713 
3714         reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
3715         ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
3716 
3717         ceph_con_send(&session->s_con, reply);
3718 
3719         mutex_unlock(&session->s_mutex);
3720 
3721         mutex_lock(&mdsc->mutex);
3722         __wake_requests(mdsc, &session->s_waiting);
3723         mutex_unlock(&mdsc->mutex);
3724 
3725         up_read(&mdsc->snap_rwsem);
3726         ceph_pagelist_release(recon_state.pagelist);
3727         return;
3728 
3729 fail:
3730         ceph_msg_put(reply);
3731         up_read(&mdsc->snap_rwsem);
3732         mutex_unlock(&session->s_mutex);
3733 fail_nomsg:
3734         ceph_pagelist_release(recon_state.pagelist);
3735 fail_nopagelist:
3736         pr_err("error %d preparing reconnect for mds%d\n", err, mds);
3737         return;
3738 }
3739 
3740 
3741 /*
3742  * compare old and new mdsmaps, kicking requests
3743  * and closing out old connections as necessary
3744  *
3745  * called under mdsc->mutex.
3746  */
3747 static void check_new_map(struct ceph_mds_client *mdsc,
3748                           struct ceph_mdsmap *newmap,
3749                           struct ceph_mdsmap *oldmap)
3750 {
3751         int i;
3752         int oldstate, newstate;
3753         struct ceph_mds_session *s;
3754 
3755         dout("check_new_map new %u old %u\n",
3756              newmap->m_epoch, oldmap->m_epoch);
3757 
3758         for (i = 0; i < oldmap->m_num_mds && i < mdsc->max_sessions; i++) {
3759                 if (!mdsc->sessions[i])
3760                         continue;
3761                 s = mdsc->sessions[i];
3762                 oldstate = ceph_mdsmap_get_state(oldmap, i);
3763                 newstate = ceph_mdsmap_get_state(newmap, i);
3764 
3765                 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
3766                      i, ceph_mds_state_name(oldstate),
3767                      ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
3768                      ceph_mds_state_name(newstate),
3769                      ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
3770                      ceph_session_state_name(s->s_state));
3771 
3772                 if (i >= newmap->m_num_mds) {
3773                         /* force close session for stopped mds */
3774                         get_session(s);
3775                         __unregister_session(mdsc, s);
3776                         __wake_requests(mdsc, &s->s_waiting);
3777                         mutex_unlock(&mdsc->mutex);
3778 
3779                         mutex_lock(&s->s_mutex);
3780                         cleanup_session_requests(mdsc, s);
3781                         remove_session_caps(s);
3782                         mutex_unlock(&s->s_mutex);
3783 
3784                         ceph_put_mds_session(s);
3785 
3786                         mutex_lock(&mdsc->mutex);
3787                         kick_requests(mdsc, i);
3788                         continue;
3789                 }
3790 
3791                 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
3792                            ceph_mdsmap_get_addr(newmap, i),
3793                            sizeof(struct ceph_entity_addr))) {
3794                         /* just close it */
3795                         mutex_unlock(&mdsc->mutex);
3796                         mutex_lock(&s->s_mutex);
3797                         mutex_lock(&mdsc->mutex);
3798                         ceph_con_close(&s->s_con);
3799                         mutex_unlock(&s->s_mutex);
3800                         s->s_state = CEPH_MDS_SESSION_RESTARTING;
3801                 } else if (oldstate == newstate) {
3802                         continue;  /* nothing new with this mds */
3803                 }
3804 
3805                 /*
3806                  * send reconnect?
3807                  */
3808                 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
3809                     newstate >= CEPH_MDS_STATE_RECONNECT) {
3810                         mutex_unlock(&mdsc->mutex);
3811                         send_mds_reconnect(mdsc, s);
3812                         mutex_lock(&mdsc->mutex);
3813                 }
3814 
3815                 /*
3816                  * kick request on any mds that has gone active.
3817                  */
3818                 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3819                     newstate >= CEPH_MDS_STATE_ACTIVE) {
3820                         if (oldstate != CEPH_MDS_STATE_CREATING &&
3821                             oldstate != CEPH_MDS_STATE_STARTING)
3822                                 pr_info("mds%d recovery completed\n", s->s_mds);
3823                         kick_requests(mdsc, i);
3824                         ceph_kick_flushing_caps(mdsc, s);
3825                         wake_up_session_caps(s, RECONNECT);
3826                 }
3827         }
3828 
3829         for (i = 0; i < newmap->m_num_mds && i < mdsc->max_sessions; i++) {
3830                 s = mdsc->sessions[i];
3831                 if (!s)
3832                         continue;
3833                 if (!ceph_mdsmap_is_laggy(newmap, i))
3834                         continue;
3835                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3836                     s->s_state == CEPH_MDS_SESSION_HUNG ||
3837                     s->s_state == CEPH_MDS_SESSION_CLOSING) {
3838                         dout(" connecting to export targets of laggy mds%d\n",
3839                              i);
3840                         __open_export_target_sessions(mdsc, s);
3841                 }
3842         }
3843 }
3844 
3845 
3846 
3847 /*
3848  * leases
3849  */
3850 
3851 /*
3852  * caller must hold session s_mutex, dentry->d_lock
3853  */
3854 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3855 {
3856         struct ceph_dentry_info *di = ceph_dentry(dentry);
3857 
3858         ceph_put_mds_session(di->lease_session);
3859         di->lease_session = NULL;
3860 }
3861 
3862 static void handle_lease(struct ceph_mds_client *mdsc,
3863                          struct ceph_mds_session *session,
3864                          struct ceph_msg *msg)
3865 {
3866         struct super_block *sb = mdsc->fsc->sb;
3867         struct inode *inode;
3868         struct dentry *parent, *dentry;
3869         struct ceph_dentry_info *di;
3870         int mds = session->s_mds;
3871         struct ceph_mds_lease *h = msg->front.iov_base;
3872         u32 seq;
3873         struct ceph_vino vino;
3874         struct qstr dname;
3875         int release = 0;
3876 
3877         dout("handle_lease from mds%d\n", mds);
3878 
3879         /* decode */
3880         if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3881                 goto bad;
3882         vino.ino = le64_to_cpu(h->ino);
3883         vino.snap = CEPH_NOSNAP;
3884         seq = le32_to_cpu(h->seq);
3885         dname.len = get_unaligned_le32(h + 1);
3886         if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
3887                 goto bad;
3888         dname.name = (void *)(h + 1) + sizeof(u32);
3889 
3890         /* lookup inode */
3891         inode = ceph_find_inode(sb, vino);
3892         dout("handle_lease %s, ino %llx %p %.*s\n",
3893              ceph_lease_op_name(h->action), vino.ino, inode,
3894              dname.len, dname.name);
3895 
3896         mutex_lock(&session->s_mutex);
3897         session->s_seq++;
3898 
3899         if (!inode) {
3900                 dout("handle_lease no inode %llx\n", vino.ino);
3901                 goto release;
3902         }
3903 
3904         /* dentry */
3905         parent = d_find_alias(inode);
3906         if (!parent) {
3907                 dout("no parent dentry on inode %p\n", inode);
3908                 WARN_ON(1);
3909                 goto release;  /* hrm... */
3910         }
3911         dname.hash = full_name_hash(parent, dname.name, dname.len);
3912         dentry = d_lookup(parent, &dname);
3913         dput(parent);
3914         if (!dentry)
3915                 goto release;
3916 
3917         spin_lock(&dentry->d_lock);
3918         di = ceph_dentry(dentry);
3919         switch (h->action) {
3920         case CEPH_MDS_LEASE_REVOKE:
3921                 if (di->lease_session == session) {
3922                         if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3923                                 h->seq = cpu_to_le32(di->lease_seq);
3924                         __ceph_mdsc_drop_dentry_lease(dentry);
3925                 }
3926                 release = 1;
3927                 break;
3928 
3929         case CEPH_MDS_LEASE_RENEW:
3930                 if (di->lease_session == session &&
3931                     di->lease_gen == session->s_cap_gen &&
3932                     di->lease_renew_from &&
3933                     di->lease_renew_after == 0) {
3934                         unsigned long duration =
3935                                 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
3936 
3937                         di->lease_seq = seq;
3938                         di->time = di->lease_renew_from + duration;
3939                         di->lease_renew_after = di->lease_renew_from +
3940                                 (duration >> 1);
3941                         di->lease_renew_from = 0;
3942                 }
3943                 break;
3944         }
3945         spin_unlock(&dentry->d_lock);
3946         dput(dentry);
3947 
3948         if (!release)
3949                 goto out;
3950 
3951 release:
3952         /* let's just reuse the same message */
3953         h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3954         ceph_msg_get(msg);
3955         ceph_con_send(&session->s_con, msg);
3956 
3957 out:
3958         mutex_unlock(&session->s_mutex);
3959         /* avoid calling iput_final() in mds dispatch threads */
3960         ceph_async_iput(inode);
3961         return;
3962 
3963 bad:
3964         pr_err("corrupt lease message\n");
3965         ceph_msg_dump(msg);
3966 }
3967 
3968 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3969                               struct dentry *dentry, char action,
3970                               u32 seq)
3971 {
3972         struct ceph_msg *msg;
3973         struct ceph_mds_lease *lease;
3974         struct inode *dir;
3975         int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
3976 
3977         dout("lease_send_msg identry %p %s to mds%d\n",
3978              dentry, ceph_lease_op_name(action), session->s_mds);
3979 
3980         msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
3981         if (!msg)
3982                 return;
3983         lease = msg->front.iov_base;
3984         lease->action = action;
3985         lease->seq = cpu_to_le32(seq);
3986 
3987         spin_lock(&dentry->d_lock);
3988         dir = d_inode(dentry->d_parent);
3989         lease->ino = cpu_to_le64(ceph_ino(dir));
3990         lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
3991 
3992         put_unaligned_le32(dentry->d_name.len, lease + 1);
3993         memcpy((void *)(lease + 1) + 4,
3994                dentry->d_name.name, dentry->d_name.len);
3995         spin_unlock(&dentry->d_lock);
3996         /*
3997          * if this is a preemptive lease RELEASE, no need to
3998          * flush request stream, since the actual request will
3999          * soon follow.
4000          */
4001         msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
4002 
4003         ceph_con_send(&session->s_con, msg);
4004 }
4005 
4006 /*
4007  * lock unlock sessions, to wait ongoing session activities
4008  */
4009 static void lock_unlock_sessions(struct ceph_mds_client *mdsc)
4010 {
4011         int i;
4012 
4013         mutex_lock(&mdsc->mutex);
4014         for (i = 0; i < mdsc->max_sessions; i++) {
4015                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4016                 if (!s)
4017                         continue;
4018                 mutex_unlock(&mdsc->mutex);
4019                 mutex_lock(&s->s_mutex);
4020                 mutex_unlock(&s->s_mutex);
4021                 ceph_put_mds_session(s);
4022                 mutex_lock(&mdsc->mutex);
4023         }
4024         mutex_unlock(&mdsc->mutex);
4025 }
4026 
4027 static void maybe_recover_session(struct ceph_mds_client *mdsc)
4028 {
4029         struct ceph_fs_client *fsc = mdsc->fsc;
4030 
4031         if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
4032                 return;
4033 
4034         if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
4035                 return;
4036 
4037         if (!READ_ONCE(fsc->blacklisted))
4038                 return;
4039 
4040         if (fsc->last_auto_reconnect &&
4041             time_before(jiffies, fsc->last_auto_reconnect + HZ * 60 * 30))
4042                 return;
4043 
4044         pr_info("auto reconnect after blacklisted\n");
4045         fsc->last_auto_reconnect = jiffies;
4046         ceph_force_reconnect(fsc->sb);
4047 }
4048 
4049 /*
4050  * delayed work -- periodically trim expired leases, renew caps with mds
4051  */
4052 static void schedule_delayed(struct ceph_mds_client *mdsc)
4053 {
4054         int delay = 5;
4055         unsigned hz = round_jiffies_relative(HZ * delay);
4056         schedule_delayed_work(&mdsc->delayed_work, hz);
4057 }
4058 
4059 static void delayed_work(struct work_struct *work)
4060 {
4061         int i;
4062         struct ceph_mds_client *mdsc =
4063                 container_of(work, struct ceph_mds_client, delayed_work.work);
4064         int renew_interval;
4065         int renew_caps;
4066 
4067         dout("mdsc delayed_work\n");
4068 
4069         mutex_lock(&mdsc->mutex);
4070         renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
4071         renew_caps = time_after_eq(jiffies, HZ*renew_interval +
4072                                    mdsc->last_renew_caps);
4073         if (renew_caps)
4074                 mdsc->last_renew_caps = jiffies;
4075 
4076         for (i = 0; i < mdsc->max_sessions; i++) {
4077                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4078                 if (!s)
4079                         continue;
4080                 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
4081                         dout("resending session close request for mds%d\n",
4082                              s->s_mds);
4083                         request_close_session(mdsc, s);
4084                         ceph_put_mds_session(s);
4085                         continue;
4086                 }
4087                 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
4088                         if (s->s_state == CEPH_MDS_SESSION_OPEN) {
4089                                 s->s_state = CEPH_MDS_SESSION_HUNG;
4090                                 pr_info("mds%d hung\n", s->s_mds);
4091                         }
4092                 }
4093                 if (s->s_state == CEPH_MDS_SESSION_NEW ||
4094                     s->s_state == CEPH_MDS_SESSION_RESTARTING ||
4095                     s->s_state == CEPH_MDS_SESSION_REJECTED) {
4096                         /* this mds is failed or recovering, just wait */
4097                         ceph_put_mds_session(s);
4098                         continue;
4099                 }
4100                 mutex_unlock(&mdsc->mutex);
4101 
4102                 mutex_lock(&s->s_mutex);
4103                 if (renew_caps)
4104                         send_renew_caps(mdsc, s);
4105                 else
4106                         ceph_con_keepalive(&s->s_con);
4107                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4108                     s->s_state == CEPH_MDS_SESSION_HUNG)
4109                         ceph_send_cap_releases(mdsc, s);
4110                 mutex_unlock(&s->s_mutex);
4111                 ceph_put_mds_session(s);
4112 
4113                 mutex_lock(&mdsc->mutex);
4114         }
4115         mutex_unlock(&mdsc->mutex);
4116 
4117         ceph_check_delayed_caps(mdsc);
4118 
4119         ceph_queue_cap_reclaim_work(mdsc);
4120 
4121         ceph_trim_snapid_map(mdsc);
4122 
4123         maybe_recover_session(mdsc);
4124 
4125         schedule_delayed(mdsc);
4126 }
4127 
4128 int ceph_mdsc_init(struct ceph_fs_client *fsc)
4129 
4130 {
4131         struct ceph_mds_client *mdsc;
4132 
4133         mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
4134         if (!mdsc)
4135                 return -ENOMEM;
4136         mdsc->fsc = fsc;
4137         mutex_init(&mdsc->mutex);
4138         mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
4139         if (!mdsc->mdsmap) {
4140                 kfree(mdsc);
4141                 return -ENOMEM;
4142         }
4143 
4144         fsc->mdsc = mdsc;
4145         init_completion(&mdsc->safe_umount_waiters);
4146         init_waitqueue_head(&mdsc->session_close_wq);
4147         INIT_LIST_HEAD(&mdsc->waiting_for_map);
4148         mdsc->sessions = NULL;
4149         atomic_set(&mdsc->num_sessions, 0);
4150         mdsc->max_sessions = 0;
4151         mdsc->stopping = 0;
4152         atomic64_set(&mdsc->quotarealms_count, 0);
4153         mdsc->quotarealms_inodes = RB_ROOT;
4154         mutex_init(&mdsc->quotarealms_inodes_mutex);
4155         mdsc->last_snap_seq = 0;
4156         init_rwsem(&mdsc->snap_rwsem);
4157         mdsc->snap_realms = RB_ROOT;
4158         INIT_LIST_HEAD(&mdsc->snap_empty);
4159         mdsc->num_snap_realms = 0;
4160         spin_lock_init(&mdsc->snap_empty_lock);
4161         mdsc->last_tid = 0;
4162         mdsc->oldest_tid = 0;
4163         mdsc->request_tree = RB_ROOT;
4164         INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
4165         mdsc->last_renew_caps = jiffies;
4166         INIT_LIST_HEAD(&mdsc->cap_delay_list);
4167         spin_lock_init(&mdsc->cap_delay_lock);
4168         INIT_LIST_HEAD(&mdsc->snap_flush_list);
4169         spin_lock_init(&mdsc->snap_flush_lock);
4170         mdsc->last_cap_flush_tid = 1;
4171         INIT_LIST_HEAD(&mdsc->cap_flush_list);
4172         INIT_LIST_HEAD(&mdsc->cap_dirty);
4173         INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
4174         mdsc->num_cap_flushing = 0;
4175         spin_lock_init(&mdsc->cap_dirty_lock);
4176         init_waitqueue_head(&mdsc->cap_flushing_wq);
4177         INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
4178         atomic_set(&mdsc->cap_reclaim_pending, 0);
4179 
4180         spin_lock_init(&mdsc->dentry_list_lock);
4181         INIT_LIST_HEAD(&mdsc->dentry_leases);
4182         INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
4183 
4184         ceph_caps_init(mdsc);
4185         ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
4186 
4187         spin_lock_init(&mdsc->snapid_map_lock);
4188         mdsc->snapid_map_tree = RB_ROOT;
4189         INIT_LIST_HEAD(&mdsc->snapid_map_lru);
4190 
4191         init_rwsem(&mdsc->pool_perm_rwsem);
4192         mdsc->pool_perm_tree = RB_ROOT;
4193 
4194         strscpy(mdsc->nodename, utsname()->nodename,
4195                 sizeof(mdsc->nodename));
4196         return 0;
4197 }
4198 
4199 /*
4200  * Wait for safe replies on open mds requests.  If we time out, drop
4201  * all requests from the tree to avoid dangling dentry refs.
4202  */
4203 static void wait_requests(struct ceph_mds_client *mdsc)
4204 {
4205         struct ceph_options *opts = mdsc->fsc->client->options;
4206         struct ceph_mds_request *req;
4207 
4208         mutex_lock(&mdsc->mutex);
4209         if (__get_oldest_req(mdsc)) {
4210                 mutex_unlock(&mdsc->mutex);
4211 
4212                 dout("wait_requests waiting for requests\n");
4213                 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
4214                                     ceph_timeout_jiffies(opts->mount_timeout));
4215 
4216                 /* tear down remaining requests */
4217                 mutex_lock(&mdsc->mutex);
4218                 while ((req = __get_oldest_req(mdsc))) {
4219                         dout("wait_requests timed out on tid %llu\n",
4220                              req->r_tid);
4221                         list_del_init(&req->r_wait);
4222                         __unregister_request(mdsc, req);
4223                 }
4224         }
4225         mutex_unlock(&mdsc->mutex);
4226         dout("wait_requests done\n");
4227 }
4228 
4229 /*
4230  * called before mount is ro, and before dentries are torn down.
4231  * (hmm, does this still race with new lookups?)
4232  */
4233 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
4234 {
4235         dout("pre_umount\n");
4236         mdsc->stopping = 1;
4237 
4238         lock_unlock_sessions(mdsc);
4239         ceph_flush_dirty_caps(mdsc);
4240         wait_requests(mdsc);
4241 
4242         /*
4243          * wait for reply handlers to drop their request refs and
4244          * their inode/dcache refs
4245          */
4246         ceph_msgr_flush();
4247 
4248         ceph_cleanup_quotarealms_inodes(mdsc);
4249 }
4250 
4251 /*
4252  * wait for all write mds requests to flush.
4253  */
4254 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
4255 {
4256         struct ceph_mds_request *req = NULL, *nextreq;
4257         struct rb_node *n;
4258 
4259         mutex_lock(&mdsc->mutex);
4260         dout("wait_unsafe_requests want %lld\n", want_tid);
4261 restart:
4262         req = __get_oldest_req(mdsc);
4263         while (req && req->r_tid <= want_tid) {
4264                 /* find next request */
4265                 n = rb_next(&req->r_node);
4266                 if (n)
4267                         nextreq = rb_entry(n, struct ceph_mds_request, r_node);
4268                 else
4269                         nextreq = NULL;
4270                 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
4271                     (req->r_op & CEPH_MDS_OP_WRITE)) {
4272                         /* write op */
4273                         ceph_mdsc_get_request(req);
4274                         if (nextreq)
4275                                 ceph_mdsc_get_request(nextreq);
4276                         mutex_unlock(&mdsc->mutex);
4277                         dout("wait_unsafe_requests  wait on %llu (want %llu)\n",
4278                              req->r_tid, want_tid);
4279                         wait_for_completion(&req->r_safe_completion);
4280                         mutex_lock(&mdsc->mutex);
4281                         ceph_mdsc_put_request(req);
4282                         if (!nextreq)
4283                                 break;  /* next dne before, so we're done! */
4284                         if (RB_EMPTY_NODE(&nextreq->r_node)) {
4285                                 /* next request was removed from tree */
4286                                 ceph_mdsc_put_request(nextreq);
4287                                 goto restart;
4288                         }
4289                         ceph_mdsc_put_request(nextreq);  /* won't go away */
4290                 }
4291                 req = nextreq;
4292         }
4293         mutex_unlock(&mdsc->mutex);
4294         dout("wait_unsafe_requests done\n");
4295 }
4296 
4297 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
4298 {
4299         u64 want_tid, want_flush;
4300 
4301         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
4302                 return;
4303 
4304         dout("sync\n");
4305         mutex_lock(&mdsc->mutex);
4306         want_tid = mdsc->last_tid;
4307         mutex_unlock(&mdsc->mutex);
4308 
4309         ceph_flush_dirty_caps(mdsc);
4310         spin_lock(&mdsc->cap_dirty_lock);
4311         want_flush = mdsc->last_cap_flush_tid;
4312         if (!list_empty(&mdsc->cap_flush_list)) {
4313                 struct ceph_cap_flush *cf =
4314                         list_last_entry(&mdsc->cap_flush_list,
4315                                         struct ceph_cap_flush, g_list);
4316                 cf->wake = true;
4317         }
4318         spin_unlock(&mdsc->cap_dirty_lock);
4319 
4320         dout("sync want tid %lld flush_seq %lld\n",
4321              want_tid, want_flush);
4322 
4323         wait_unsafe_requests(mdsc, want_tid);
4324         wait_caps_flush(mdsc, want_flush);
4325 }
4326 
4327 /*
4328  * true if all sessions are closed, or we force unmount
4329  */
4330 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
4331 {
4332         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
4333                 return true;
4334         return atomic_read(&mdsc->num_sessions) <= skipped;
4335 }
4336 
4337 /*
4338  * called after sb is ro.
4339  */
4340 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
4341 {
4342         struct ceph_options *opts = mdsc->fsc->client->options;
4343         struct ceph_mds_session *session;
4344         int i;
4345         int skipped = 0;
4346 
4347         dout("close_sessions\n");
4348 
4349         /* close sessions */
4350         mutex_lock(&mdsc->mutex);
4351         for (i = 0; i < mdsc->max_sessions; i++) {
4352                 session = __ceph_lookup_mds_session(mdsc, i);
4353                 if (!session)
4354                         continue;
4355                 mutex_unlock(&mdsc->mutex);
4356                 mutex_lock(&session->s_mutex);
4357                 if (__close_session(mdsc, session) <= 0)
4358                         skipped++;
4359                 mutex_unlock(&session->s_mutex);
4360                 ceph_put_mds_session(session);
4361                 mutex_lock(&mdsc->mutex);
4362         }
4363         mutex_unlock(&mdsc->mutex);
4364 
4365         dout("waiting for sessions to close\n");
4366         wait_event_timeout(mdsc->session_close_wq,
4367                            done_closing_sessions(mdsc, skipped),
4368                            ceph_timeout_jiffies(opts->mount_timeout));
4369 
4370         /* tear down remaining sessions */
4371         mutex_lock(&mdsc->mutex);
4372         for (i = 0; i < mdsc->max_sessions; i++) {
4373                 if (mdsc->sessions[i]) {
4374                         session = get_session(mdsc->sessions[i]);
4375                         __unregister_session(mdsc, session);
4376                         mutex_unlock(&mdsc->mutex);
4377                         mutex_lock(&session->s_mutex);
4378                         remove_session_caps(session);
4379                         mutex_unlock(&session->s_mutex);
4380                         ceph_put_mds_session(session);
4381                         mutex_lock(&mdsc->mutex);
4382                 }
4383         }
4384         WARN_ON(!list_empty(&mdsc->cap_delay_list));
4385         mutex_unlock(&mdsc->mutex);
4386 
4387         ceph_cleanup_snapid_map(mdsc);
4388         ceph_cleanup_empty_realms(mdsc);
4389 
4390         cancel_work_sync(&mdsc->cap_reclaim_work);
4391         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
4392 
4393         dout("stopped\n");
4394 }
4395 
4396 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
4397 {
4398         struct ceph_mds_session *session;
4399         int mds;
4400 
4401         dout("force umount\n");
4402 
4403         mutex_lock(&mdsc->mutex);
4404         for (mds = 0; mds < mdsc->max_sessions; mds++) {
4405                 session = __ceph_lookup_mds_session(mdsc, mds);
4406                 if (!session)
4407                         continue;
4408 
4409                 if (session->s_state == CEPH_MDS_SESSION_REJECTED)
4410                         __unregister_session(mdsc, session);
4411                 __wake_requests(mdsc, &session->s_waiting);
4412                 mutex_unlock(&mdsc->mutex);
4413 
4414                 mutex_lock(&session->s_mutex);
4415                 __close_session(mdsc, session);
4416                 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
4417                         cleanup_session_requests(mdsc, session);
4418                         remove_session_caps(session);
4419                 }
4420                 mutex_unlock(&session->s_mutex);
4421                 ceph_put_mds_session(session);
4422 
4423                 mutex_lock(&mdsc->mutex);
4424                 kick_requests(mdsc, mds);
4425         }
4426         __wake_requests(mdsc, &mdsc->waiting_for_map);
4427         mutex_unlock(&mdsc->mutex);
4428 }
4429 
4430 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
4431 {
4432         dout("stop\n");
4433         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
4434         if (mdsc->mdsmap)
4435                 ceph_mdsmap_destroy(mdsc->mdsmap);
4436         kfree(mdsc->sessions);
4437         ceph_caps_finalize(mdsc);
4438         ceph_pool_perm_destroy(mdsc);
4439 }
4440 
4441 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
4442 {
4443         struct ceph_mds_client *mdsc = fsc->mdsc;
4444         dout("mdsc_destroy %p\n", mdsc);
4445 
4446         if (!mdsc)
4447                 return;
4448 
4449         /* flush out any connection work with references to us */
4450         ceph_msgr_flush();
4451 
4452         ceph_mdsc_stop(mdsc);
4453 
4454         fsc->mdsc = NULL;
4455         kfree(mdsc);
4456         dout("mdsc_destroy %p done\n", mdsc);
4457 }
4458 
4459 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
4460 {
4461         struct ceph_fs_client *fsc = mdsc->fsc;
4462         const char *mds_namespace = fsc->mount_options->mds_namespace;
4463         void *p = msg->front.iov_base;
4464         void *end = p + msg->front.iov_len;
4465         u32 epoch;
4466         u32 map_len;
4467         u32 num_fs;
4468         u32 mount_fscid = (u32)-1;
4469         u8 struct_v, struct_cv;
4470         int err = -EINVAL;
4471 
4472         ceph_decode_need(&p, end, sizeof(u32), bad);
4473         epoch = ceph_decode_32(&p);
4474 
4475         dout("handle_fsmap epoch %u\n", epoch);
4476 
4477         ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
4478         struct_v = ceph_decode_8(&p);
4479         struct_cv = ceph_decode_8(&p);
4480         map_len = ceph_decode_32(&p);
4481 
4482         ceph_decode_need(&p, end, sizeof(u32) * 3, bad);
4483         p += sizeof(u32) * 2; /* skip epoch and legacy_client_fscid */
4484 
4485         num_fs = ceph_decode_32(&p);
4486         while (num_fs-- > 0) {
4487                 void *info_p, *info_end;
4488                 u32 info_len;
4489                 u8 info_v, info_cv;
4490                 u32 fscid, namelen;
4491 
4492                 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
4493                 info_v = ceph_decode_8(&p);
4494                 info_cv = ceph_decode_8(&p);
4495                 info_len = ceph_decode_32(&p);
4496                 ceph_decode_need(&p, end, info_len, bad);
4497                 info_p = p;
4498                 info_end = p + info_len;
4499                 p = info_end;
4500 
4501                 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
4502                 fscid = ceph_decode_32(&info_p);
4503                 namelen = ceph_decode_32(&info_p);
4504                 ceph_decode_need(&info_p, info_end, namelen, bad);
4505 
4506                 if (mds_namespace &&
4507                     strlen(mds_namespace) == namelen &&
4508                     !strncmp(mds_namespace, (char *)info_p, namelen)) {
4509                         mount_fscid = fscid;
4510                         break;
4511                 }
4512         }
4513 
4514         ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
4515         if (mount_fscid != (u32)-1) {
4516                 fsc->client->monc.fs_cluster_id = mount_fscid;
4517                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
4518                                    0, true);
4519                 ceph_monc_renew_subs(&fsc->client->monc);
4520         } else {
4521                 err = -ENOENT;
4522                 goto err_out;
4523         }
4524         return;
4525 
4526 bad:
4527         pr_err("error decoding fsmap\n");
4528 err_out:
4529         mutex_lock(&mdsc->mutex);
4530         mdsc->mdsmap_err = err;
4531         __wake_requests(mdsc, &mdsc->waiting_for_map);
4532         mutex_unlock(&mdsc->mutex);
4533 }
4534 
4535 /*
4536  * handle mds map update.
4537  */
4538 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
4539 {
4540         u32 epoch;
4541         u32 maplen;
4542         void *p = msg->front.iov_base;
4543         void *end = p + msg->front.iov_len;
4544         struct ceph_mdsmap *newmap, *oldmap;
4545         struct ceph_fsid fsid;
4546         int err = -EINVAL;
4547 
4548         ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
4549         ceph_decode_copy(&p, &fsid, sizeof(fsid));
4550         if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
4551                 return;
4552         epoch = ceph_decode_32(&p);
4553         maplen = ceph_decode_32(&p);
4554         dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
4555 
4556         /* do we need it? */
4557         mutex_lock(&mdsc->mutex);
4558         if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
4559                 dout("handle_map epoch %u <= our %u\n",
4560                      epoch, mdsc->mdsmap->m_epoch);
4561                 mutex_unlock(&mdsc->mutex);
4562                 return;
4563         }
4564 
4565         newmap = ceph_mdsmap_decode(&p, end);
4566         if (IS_ERR(newmap)) {
4567                 err = PTR_ERR(newmap);
4568                 goto bad_unlock;
4569         }
4570 
4571         /* swap into place */
4572         if (mdsc->mdsmap) {
4573                 oldmap = mdsc->mdsmap;
4574                 mdsc->mdsmap = newmap;
4575                 check_new_map(mdsc, newmap, oldmap);
4576                 ceph_mdsmap_destroy(oldmap);
4577         } else {
4578                 mdsc->mdsmap = newmap;  /* first mds map */
4579         }
4580         mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
4581                                         MAX_LFS_FILESIZE);
4582 
4583         __wake_requests(mdsc, &mdsc->waiting_for_map);
4584         ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
4585                           mdsc->mdsmap->m_epoch);
4586 
4587         mutex_unlock(&mdsc->mutex);
4588         schedule_delayed(mdsc);
4589         return;
4590 
4591 bad_unlock:
4592         mutex_unlock(&mdsc->mutex);
4593 bad:
4594         pr_err("error decoding mdsmap %d\n", err);
4595         return;
4596 }
4597 
4598 static struct ceph_connection *con_get(struct ceph_connection *con)
4599 {
4600         struct ceph_mds_session *s = con->private;
4601 
4602         if (get_session(s)) {
4603                 dout("mdsc con_get %p ok (%d)\n", s, refcount_read(&s->s_ref));
4604                 return con;
4605         }
4606         dout("mdsc con_get %p FAIL\n", s);
4607         return NULL;
4608 }
4609 
4610 static void con_put(struct ceph_connection *con)
4611 {
4612         struct ceph_mds_session *s = con->private;
4613 
4614         dout("mdsc con_put %p (%d)\n", s, refcount_read(&s->s_ref) - 1);
4615         ceph_put_mds_session(s);
4616 }
4617 
4618 /*
4619  * if the client is unresponsive for long enough, the mds will kill
4620  * the session entirely.
4621  */
4622 static void peer_reset(struct ceph_connection *con)
4623 {
4624         struct ceph_mds_session *s = con->private;
4625         struct ceph_mds_client *mdsc = s->s_mdsc;
4626 
4627         pr_warn("mds%d closed our session\n", s->s_mds);
4628         send_mds_reconnect(mdsc, s);
4629 }
4630 
4631 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4632 {
4633         struct ceph_mds_session *s = con->private;
4634         struct ceph_mds_client *mdsc = s->s_mdsc;
4635         int type = le16_to_cpu(msg->hdr.type);
4636 
4637         mutex_lock(&mdsc->mutex);
4638         if (__verify_registered_session(mdsc, s) < 0) {
4639                 mutex_unlock(&mdsc->mutex);
4640                 goto out;
4641         }
4642         mutex_unlock(&mdsc->mutex);
4643 
4644         switch (type) {
4645         case CEPH_MSG_MDS_MAP:
4646                 ceph_mdsc_handle_mdsmap(mdsc, msg);
4647                 break;
4648         case CEPH_MSG_FS_MAP_USER:
4649                 ceph_mdsc_handle_fsmap(mdsc, msg);
4650                 break;
4651         case CEPH_MSG_CLIENT_SESSION:
4652                 handle_session(s, msg);
4653                 break;
4654         case CEPH_MSG_CLIENT_REPLY:
4655                 handle_reply(s, msg);
4656                 break;
4657         case CEPH_MSG_CLIENT_REQUEST_FORWARD:
4658                 handle_forward(mdsc, s, msg);
4659                 break;
4660         case CEPH_MSG_CLIENT_CAPS:
4661                 ceph_handle_caps(s, msg);
4662                 break;
4663         case CEPH_MSG_CLIENT_SNAP:
4664                 ceph_handle_snap(mdsc, s, msg);
4665                 break;
4666         case CEPH_MSG_CLIENT_LEASE:
4667                 handle_lease(mdsc, s, msg);
4668                 break;
4669         case CEPH_MSG_CLIENT_QUOTA:
4670                 ceph_handle_quota(mdsc, s, msg);
4671                 break;
4672 
4673         default:
4674                 pr_err("received unknown message type %d %s\n", type,
4675                        ceph_msg_type_name(type));
4676         }
4677 out:
4678         ceph_msg_put(msg);
4679 }
4680 
4681 /*
4682  * authentication
4683  */
4684 
4685 /*
4686  * Note: returned pointer is the address of a structure that's
4687  * managed separately.  Caller must *not* attempt to free it.
4688  */
4689 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
4690                                         int *proto, int force_new)
4691 {
4692         struct ceph_mds_session *s = con->private;
4693         struct ceph_mds_client *mdsc = s->s_mdsc;
4694         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4695         struct ceph_auth_handshake *auth = &s->s_auth;
4696 
4697         if (force_new && auth->authorizer) {
4698                 ceph_auth_destroy_authorizer(auth->authorizer);
4699                 auth->authorizer = NULL;
4700         }
4701         if (!auth->authorizer) {
4702                 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4703                                                       auth);
4704                 if (ret)
4705                         return ERR_PTR(ret);
4706         } else {
4707                 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4708                                                       auth);
4709                 if (ret)
4710                         return ERR_PTR(ret);
4711         }
4712         *proto = ac->protocol;
4713 
4714         return auth;
4715 }
4716 
4717 static int add_authorizer_challenge(struct ceph_connection *con,
4718                                     void *challenge_buf, int challenge_buf_len)
4719 {
4720         struct ceph_mds_session *s = con->private;
4721         struct ceph_mds_client *mdsc = s->s_mdsc;
4722         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4723 
4724         return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
4725                                             challenge_buf, challenge_buf_len);
4726 }
4727 
4728 static int verify_authorizer_reply(struct ceph_connection *con)
4729 {
4730         struct ceph_mds_session *s = con->private;
4731         struct ceph_mds_client *mdsc = s->s_mdsc;
4732         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4733 
4734         return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer);
4735 }
4736 
4737 static int invalidate_authorizer(struct ceph_connection *con)
4738 {
4739         struct ceph_mds_session *s = con->private;
4740         struct ceph_mds_client *mdsc = s->s_mdsc;
4741         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4742 
4743         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
4744 
4745         return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
4746 }
4747 
4748 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
4749                                 struct ceph_msg_header *hdr, int *skip)
4750 {
4751         struct ceph_msg *msg;
4752         int type = (int) le16_to_cpu(hdr->type);
4753         int front_len = (int) le32_to_cpu(hdr->front_len);
4754 
4755         if (con->in_msg)
4756                 return con->in_msg;
4757 
4758         *skip = 0;
4759         msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
4760         if (!msg) {
4761                 pr_err("unable to allocate msg type %d len %d\n",
4762                        type, front_len);
4763                 return NULL;
4764         }
4765 
4766         return msg;
4767 }
4768 
4769 static int mds_sign_message(struct ceph_msg *msg)
4770 {
4771        struct ceph_mds_session *s = msg->con->private;
4772        struct ceph_auth_handshake *auth = &s->s_auth;
4773 
4774        return ceph_auth_sign_message(auth, msg);
4775 }
4776 
4777 static int mds_check_message_signature(struct ceph_msg *msg)
4778 {
4779        struct ceph_mds_session *s = msg->con->private;
4780        struct ceph_auth_handshake *auth = &s->s_auth;
4781 
4782        return ceph_auth_check_message_signature(auth, msg);
4783 }
4784 
4785 static const struct ceph_connection_operations mds_con_ops = {
4786         .get = con_get,
4787         .put = con_put,
4788         .dispatch = dispatch,
4789         .get_authorizer = get_authorizer,
4790         .add_authorizer_challenge = add_authorizer_challenge,
4791         .verify_authorizer_reply = verify_authorizer_reply,
4792         .invalidate_authorizer = invalidate_authorizer,
4793         .peer_reset = peer_reset,
4794         .alloc_msg = mds_alloc_msg,
4795         .sign_message = mds_sign_message,
4796         .check_message_signature = mds_check_message_signature,
4797 };
4798 
4799 /* eof */