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  50  * are no longer valid.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