fs/ceph/mds_client.c

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  48  * are no longer valid.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