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  50  * are no longer valid.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;
 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, 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("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 static int parse_reply_info_getvxattr(void **p, void *end,
 559                                       struct ceph_mds_reply_info_parsed *info,
 560                                       u64 features)
 561 {
 562         u32 value_len;
 563
 564         ceph_decode_skip_8(p, end, bad); /* skip current version: 1 */
 565         ceph_decode_skip_8(p, end, bad); /* skip first version: 1 */
 566         ceph_decode_skip_32(p, end, bad); /* skip payload length */
 567
 568         ceph_decode_32_safe(p, end, value_len, bad);
 569
 570         if (value_len == end - *p) {
 571           info->xattr_info.xattr_value = *p;
 572           info->xattr_info.xattr_value_len = value_len;
 573           *p = end;
 574           return value_len;
 575         }
 576 bad:
 577         return -EIO;
 578 }
 579
 580 /*
 581  * parse extra results
 582  */
 583 static int parse_reply_info_extra(void **p, void *end,
 584                                   struct ceph_mds_reply_info_parsed *info,
 585                                   u64 features, struct ceph_mds_session *s)
 586 {
 587         u32 op = le32_to_cpu(info->head->op);
 588
 589         if (op == CEPH_MDS_OP_GETFILELOCK)
 590                 return parse_reply_info_filelock(p, end, info, features);
 591         else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
 592                 return parse_reply_info_readdir(p, end, info, features);
 593         else if (op == CEPH_MDS_OP_CREATE)
 594                 return parse_reply_info_create(p, end, info, features, s);
 595         else if (op == CEPH_MDS_OP_GETVXATTR)
 596                 return parse_reply_info_getvxattr(p, end, info, features);
 597         else
 598                 return -EIO;
 599 }
 600
 601 /*
 602  * parse entire mds reply
 603  */
 604 static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,
 605                             struct ceph_mds_reply_info_parsed *info,
 606                             u64 features)
 607 {
 608         void *p, *end;
 609         u32 len;
 610         int err;
 611
 612         info->head = msg->front.iov_base;
 613         p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
 614         end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
 615
 616         /* trace */
 617         ceph_decode_32_safe(&p, end, len, bad);
 618         if (len > 0) {
 619                 ceph_decode_need(&p, end, len, bad);
 620                 err = parse_reply_info_trace(&p, p+len, info, features);
 621                 if (err < 0)
 622                         goto out_bad;
 623         }
 624
 625         /* extra */
 626         ceph_decode_32_safe(&p, end, len, bad);
 627         if (len > 0) {
 628                 ceph_decode_need(&p, end, len, bad);
 629                 err = parse_reply_info_extra(&p, p+len, info, features, s);
 630                 if (err < 0)
 631                         goto out_bad;
 632         }
 633
 634         /* snap blob */
 635         ceph_decode_32_safe(&p, end, len, bad);
 636         info->snapblob_len = len;
 637         info->snapblob = p;
 638         p += len;
 639
 640         if (p != end)
 641                 goto bad;
 642         return 0;
 643
 644 bad:
 645         err = -EIO;
 646 out_bad:
 647         pr_err("mds parse_reply err %d\n", err);
 648         return err;
 649 }
 650
 651 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
 652 {
 653         if (!info->dir_entries)
 654                 return;
 655         free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
 656 }
 657
 658 /*
 659  * In async unlink case the kclient won't wait for the first reply
 660  * from MDS and just drop all the links and unhash the dentry and then
 661  * succeeds immediately.
 662  *
 663  * For any new create/link/rename,etc requests followed by using the
 664  * same file names we must wait for the first reply of the inflight
 665  * unlink request, or the MDS possibly will fail these following
 666  * requests with -EEXIST if the inflight async unlink request was
 667  * delayed for some reasons.
 668  *
 669  * And the worst case is that for the none async openc request it will
 670  * successfully open the file if the CDentry hasn't been unlinked yet,
 671  * but later the previous delayed async unlink request will remove the
 672  * CDenty. That means the just created file is possiblly deleted later
 673  * by accident.
 674  *
 675  * We need to wait for the inflight async unlink requests to finish
 676  * when creating new files/directories by using the same file names.
 677  */
 678 int ceph_wait_on_conflict_unlink(struct dentry *dentry)
 679 {
 680         struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
 681         struct dentry *pdentry = dentry->d_parent;
 682         struct dentry *udentry, *found = NULL;
 683         struct ceph_dentry_info *di;
 684         struct qstr dname;
 685         u32 hash = dentry->d_name.hash;
 686         int err;
 687
 688         dname.name = dentry->d_name.name;
 689         dname.len = dentry->d_name.len;
 690
 691         rcu_read_lock();
 692         hash_for_each_possible_rcu(fsc->async_unlink_conflict, di,
 693                                    hnode, hash) {
 694                 udentry = di->dentry;
 695
 696                 spin_lock(&udentry->d_lock);
 697                 if (udentry->d_name.hash != hash)
 698                         goto next;
 699                 if (unlikely(udentry->d_parent != pdentry))
 700                         goto next;
 701                 if (!hash_hashed(&di->hnode))
 702                         goto next;
 703
 704                 if (!test_bit(CEPH_DENTRY_ASYNC_UNLINK_BIT, &di->flags))
 705                         pr_warn("%s dentry %p:%pd async unlink bit is not set\n",
 706                                 __func__, dentry, dentry);
 707
 708                 if (!d_same_name(udentry, pdentry, &dname))
 709                         goto next;
 710
 711                 spin_unlock(&udentry->d_lock);
 712                 found = dget(udentry);
 713                 break;
 714 next:
 715                 spin_unlock(&udentry->d_lock);
 716         }
 717         rcu_read_unlock();
 718
 719         if (likely(!found))
 720                 return 0;
 721
 722         dout("%s dentry %p:%pd conflict with old %p:%pd\n", __func__,
 723              dentry, dentry, found, found);
 724
 725         err = wait_on_bit(&di->flags, CEPH_DENTRY_ASYNC_UNLINK_BIT,
 726                           TASK_KILLABLE);
 727         dput(found);
 728         return err;
 729 }
 730
 731
 732 /*
 733  * sessions
 734  */
 735 const char *ceph_session_state_name(int s)
 736 {
 737         switch (s) {
 738         case CEPH_MDS_SESSION_NEW: return "new";
 739         case CEPH_MDS_SESSION_OPENING: return "opening";
 740         case CEPH_MDS_SESSION_OPEN: return "open";
 741         case CEPH_MDS_SESSION_HUNG: return "hung";
 742         case CEPH_MDS_SESSION_CLOSING: return "closing";
 743         case CEPH_MDS_SESSION_CLOSED: return "closed";
 744         case CEPH_MDS_SESSION_RESTARTING: return "restarting";
 745         case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
 746         case CEPH_MDS_SESSION_REJECTED: return "rejected";
 747         default: return "???";
 748         }
 749 }
 750
 751 struct ceph_mds_session *ceph_get_mds_session(struct ceph_mds_session *s)
 752 {
 753         if (refcount_inc_not_zero(&s->s_ref))
 754                 return s;
 755         return NULL;
 756 }
 757
 758 void ceph_put_mds_session(struct ceph_mds_session *s)
 759 {
 760         if (IS_ERR_OR_NULL(s))
 761                 return;
 762
 763         if (refcount_dec_and_test(&s->s_ref)) {
 764                 if (s->s_auth.authorizer)
 765                         ceph_auth_destroy_authorizer(s->s_auth.authorizer);
 766                 WARN_ON(mutex_is_locked(&s->s_mutex));
 767                 xa_destroy(&s->s_delegated_inos);
 768                 kfree(s);
 769         }
 770 }
 771
 772 /*
 773  * called under mdsc->mutex
 774  */
 775 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
 776                                                    int mds)
 777 {
 778         if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
 779                 return NULL;
 780         return ceph_get_mds_session(mdsc->sessions[mds]);
 781 }
 782
 783 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
 784 {
 785         if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
 786                 return false;
 787         else
 788                 return true;
 789 }
 790
 791 static int __verify_registered_session(struct ceph_mds_client *mdsc,
 792                                        struct ceph_mds_session *s)
 793 {
 794         if (s->s_mds >= mdsc->max_sessions ||
 795             mdsc->sessions[s->s_mds] != s)
 796                 return -ENOENT;
 797         return 0;
 798 }
 799
 800 /*
 801  * create+register a new session for given mds.
 802  * called under mdsc->mutex.
 803  */
 804 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
 805                                                  int mds)
 806 {
 807         struct ceph_mds_session *s;
 808
 809         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO)
 810                 return ERR_PTR(-EIO);
 811
 812         if (mds >= mdsc->mdsmap->possible_max_rank)
 813                 return ERR_PTR(-EINVAL);
 814
 815         s = kzalloc(sizeof(*s), GFP_NOFS);
 816         if (!s)
 817                 return ERR_PTR(-ENOMEM);
 818
 819         if (mds >= mdsc->max_sessions) {
 820                 int newmax = 1 << get_count_order(mds + 1);
 821                 struct ceph_mds_session **sa;
 822
 823                 dout("%s: realloc to %d\n", __func__, newmax);
 824                 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
 825                 if (!sa)
 826                         goto fail_realloc;
 827                 if (mdsc->sessions) {
 828                         memcpy(sa, mdsc->sessions,
 829                                mdsc->max_sessions * sizeof(void *));
 830                         kfree(mdsc->sessions);
 831                 }
 832                 mdsc->sessions = sa;
 833                 mdsc->max_sessions = newmax;
 834         }
 835
 836         dout("%s: mds%d\n", __func__, mds);
 837         s->s_mdsc = mdsc;
 838         s->s_mds = mds;
 839         s->s_state = CEPH_MDS_SESSION_NEW;
 840         mutex_init(&s->s_mutex);
 841
 842         ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
 843
 844         atomic_set(&s->s_cap_gen, 1);
 845         s->s_cap_ttl = jiffies - 1;
 846
 847         spin_lock_init(&s->s_cap_lock);
 848         INIT_LIST_HEAD(&s->s_caps);
 849         refcount_set(&s->s_ref, 1);
 850         INIT_LIST_HEAD(&s->s_waiting);
 851         INIT_LIST_HEAD(&s->s_unsafe);
 852         xa_init(&s->s_delegated_inos);
 853         INIT_LIST_HEAD(&s->s_cap_releases);
 854         INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
 855
 856         INIT_LIST_HEAD(&s->s_cap_dirty);
 857         INIT_LIST_HEAD(&s->s_cap_flushing);
 858
 859         mdsc->sessions[mds] = s;
 860         atomic_inc(&mdsc->num_sessions);
 861         refcount_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
 862
 863         ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
 864                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
 865
 866         return s;
 867
 868 fail_realloc:
 869         kfree(s);
 870         return ERR_PTR(-ENOMEM);
 871 }
 872
 873 /*
 874  * called under mdsc->mutex
 875  */
 876 static void __unregister_session(struct ceph_mds_client *mdsc,
 877                                struct ceph_mds_session *s)
 878 {
 879         dout("__unregister_session mds%d %p\n", s->s_mds, s);
 880         BUG_ON(mdsc->sessions[s->s_mds] != s);
 881         mdsc->sessions[s->s_mds] = NULL;
 882         ceph_con_close(&s->s_con);
 883         ceph_put_mds_session(s);
 884         atomic_dec(&mdsc->num_sessions);
 885 }
 886
 887 /*
 888  * drop session refs in request.
 889  *
 890  * should be last request ref, or hold mdsc->mutex
 891  */
 892 static void put_request_session(struct ceph_mds_request *req)
 893 {
 894         if (req->r_session) {
 895                 ceph_put_mds_session(req->r_session);
 896                 req->r_session = NULL;
 897         }
 898 }
 899
 900 void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
 901                                 void (*cb)(struct ceph_mds_session *),
 902                                 bool check_state)
 903 {
 904         int mds;
 905
 906         mutex_lock(&mdsc->mutex);
 907         for (mds = 0; mds < mdsc->max_sessions; ++mds) {
 908                 struct ceph_mds_session *s;
 909
 910                 s = __ceph_lookup_mds_session(mdsc, mds);
 911                 if (!s)
 912                         continue;
 913
 914                 if (check_state && !check_session_state(s)) {
 915                         ceph_put_mds_session(s);
 916                         continue;
 917                 }
 918
 919                 mutex_unlock(&mdsc->mutex);
 920                 cb(s);
 921                 ceph_put_mds_session(s);
 922                 mutex_lock(&mdsc->mutex);
 923         }
 924         mutex_unlock(&mdsc->mutex);
 925 }
 926
 927 void ceph_mdsc_release_request(struct kref *kref)
 928 {
 929         struct ceph_mds_request *req = container_of(kref,
 930                                                     struct ceph_mds_request,
 931                                                     r_kref);
 932         ceph_mdsc_release_dir_caps_no_check(req);
 933         destroy_reply_info(&req->r_reply_info);
 934         if (req->r_request)
 935                 ceph_msg_put(req->r_request);
 936         if (req->r_reply)
 937                 ceph_msg_put(req->r_reply);
 938         if (req->r_inode) {
 939                 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
 940                 iput(req->r_inode);
 941         }
 942         if (req->r_parent) {
 943                 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
 944                 iput(req->r_parent);
 945         }
 946         iput(req->r_target_inode);
 947         if (req->r_dentry)
 948                 dput(req->r_dentry);
 949         if (req->r_old_dentry)
 950                 dput(req->r_old_dentry);
 951         if (req->r_old_dentry_dir) {
 952                 /*
 953                  * track (and drop pins for) r_old_dentry_dir
 954                  * separately, since r_old_dentry's d_parent may have
 955                  * changed between the dir mutex being dropped and
 956                  * this request being freed.
 957                  */
 958                 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
 959                                   CEPH_CAP_PIN);
 960                 iput(req->r_old_dentry_dir);
 961         }
 962         kfree(req->r_path1);
 963         kfree(req->r_path2);
 964         put_cred(req->r_cred);
 965         if (req->r_pagelist)
 966                 ceph_pagelist_release(req->r_pagelist);
 967         put_request_session(req);
 968         ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
 969         WARN_ON_ONCE(!list_empty(&req->r_wait));
 970         kmem_cache_free(ceph_mds_request_cachep, req);
 971 }
 972
 973 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
 974
 975 /*
 976  * lookup session, bump ref if found.
 977  *
 978  * called under mdsc->mutex.
 979  */
 980 static struct ceph_mds_request *
 981 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
 982 {
 983         struct ceph_mds_request *req;
 984
 985         req = lookup_request(&mdsc->request_tree, tid);
 986         if (req)
 987                 ceph_mdsc_get_request(req);
 988
 989         return req;
 990 }
 991
 992 /*
 993  * Register an in-flight request, and assign a tid.  Link to directory
 994  * are modifying (if any).
 995  *
 996  * Called under mdsc->mutex.
 997  */
 998 static void __register_request(struct ceph_mds_client *mdsc,
 999                                struct ceph_mds_request *req,
1000                                struct inode *dir)
1001 {
1002         int ret = 0;
1003
1004         req->r_tid = ++mdsc->last_tid;
1005         if (req->r_num_caps) {
1006                 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
1007                                         req->r_num_caps);
1008                 if (ret < 0) {
1009                         pr_err("__register_request %p "
1010                                "failed to reserve caps: %d\n", req, ret);
1011                         /* set req->r_err to fail early from __do_request */
1012                         req->r_err = ret;
1013                         return;
1014                 }
1015         }
1016         dout("__register_request %p tid %lld\n", req, req->r_tid);
1017         ceph_mdsc_get_request(req);
1018         insert_request(&mdsc->request_tree, req);
1019
1020         req->r_cred = get_current_cred();
1021
1022         if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
1023                 mdsc->oldest_tid = req->r_tid;
1024
1025         if (dir) {
1026                 struct ceph_inode_info *ci = ceph_inode(dir);
1027
1028                 ihold(dir);
1029                 req->r_unsafe_dir = dir;
1030                 spin_lock(&ci->i_unsafe_lock);
1031                 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
1032                 spin_unlock(&ci->i_unsafe_lock);
1033         }
1034 }
1035
1036 static void __unregister_request(struct ceph_mds_client *mdsc,
1037                                  struct ceph_mds_request *req)
1038 {
1039         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1040
1041         /* Never leave an unregistered request on an unsafe list! */
1042         list_del_init(&req->r_unsafe_item);
1043
1044         if (req->r_tid == mdsc->oldest_tid) {
1045                 struct rb_node *p = rb_next(&req->r_node);
1046                 mdsc->oldest_tid = 0;
1047                 while (p) {
1048                         struct ceph_mds_request *next_req =
1049                                 rb_entry(p, struct ceph_mds_request, r_node);
1050                         if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
1051                                 mdsc->oldest_tid = next_req->r_tid;
1052                                 break;
1053                         }
1054                         p = rb_next(p);
1055                 }
1056         }
1057
1058         erase_request(&mdsc->request_tree, req);
1059
1060         if (req->r_unsafe_dir) {
1061                 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
1062                 spin_lock(&ci->i_unsafe_lock);
1063                 list_del_init(&req->r_unsafe_dir_item);
1064                 spin_unlock(&ci->i_unsafe_lock);
1065         }
1066         if (req->r_target_inode &&
1067             test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
1068                 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
1069                 spin_lock(&ci->i_unsafe_lock);
1070                 list_del_init(&req->r_unsafe_target_item);
1071                 spin_unlock(&ci->i_unsafe_lock);
1072         }
1073
1074         if (req->r_unsafe_dir) {
1075                 iput(req->r_unsafe_dir);
1076                 req->r_unsafe_dir = NULL;
1077         }
1078
1079         complete_all(&req->r_safe_completion);
1080
1081         ceph_mdsc_put_request(req);
1082 }
1083
1084 /*
1085  * Walk back up the dentry tree until we hit a dentry representing a
1086  * non-snapshot inode. We do this using the rcu_read_lock (which must be held
1087  * when calling this) to ensure that the objects won't disappear while we're
1088  * working with them. Once we hit a candidate dentry, we attempt to take a
1089  * reference to it, and return that as the result.
1090  */
1091 static struct inode *get_nonsnap_parent(struct dentry *dentry)
1092 {
1093         struct inode *inode = NULL;
1094
1095         while (dentry && !IS_ROOT(dentry)) {
1096                 inode = d_inode_rcu(dentry);
1097                 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
1098                         break;
1099                 dentry = dentry->d_parent;
1100         }
1101         if (inode)
1102                 inode = igrab(inode);
1103         return inode;
1104 }
1105
1106 /*
1107  * Choose mds to send request to next.  If there is a hint set in the
1108  * request (e.g., due to a prior forward hint from the mds), use that.
1109  * Otherwise, consult frag tree and/or caps to identify the
1110  * appropriate mds.  If all else fails, choose randomly.
1111  *
1112  * Called under mdsc->mutex.
1113  */
1114 static int __choose_mds(struct ceph_mds_client *mdsc,
1115                         struct ceph_mds_request *req,
1116                         bool *random)
1117 {
1118         struct inode *inode;
1119         struct ceph_inode_info *ci;
1120         struct ceph_cap *cap;
1121         int mode = req->r_direct_mode;
1122         int mds = -1;
1123         u32 hash = req->r_direct_hash;
1124         bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
1125
1126         if (random)
1127                 *random = false;
1128
1129         /*
1130          * is there a specific mds we should try?  ignore hint if we have
1131          * no session and the mds is not up (active or recovering).
1132          */
1133         if (req->r_resend_mds >= 0 &&
1134             (__have_session(mdsc, req->r_resend_mds) ||
1135              ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
1136                 dout("%s using resend_mds mds%d\n", __func__,
1137                      req->r_resend_mds);
1138                 return req->r_resend_mds;
1139         }
1140
1141         if (mode == USE_RANDOM_MDS)
1142                 goto random;
1143
1144         inode = NULL;
1145         if (req->r_inode) {
1146                 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
1147                         inode = req->r_inode;
1148                         ihold(inode);
1149                 } else {
1150                         /* req->r_dentry is non-null for LSSNAP request */
1151                         rcu_read_lock();
1152                         inode = get_nonsnap_parent(req->r_dentry);
1153                         rcu_read_unlock();
1154                         dout("%s using snapdir's parent %p\n", __func__, inode);
1155                 }
1156         } else if (req->r_dentry) {
1157                 /* ignore race with rename; old or new d_parent is okay */
1158                 struct dentry *parent;
1159                 struct inode *dir;
1160
1161                 rcu_read_lock();
1162                 parent = READ_ONCE(req->r_dentry->d_parent);
1163                 dir = req->r_parent ? : d_inode_rcu(parent);
1164
1165                 if (!dir || dir->i_sb != mdsc->fsc->sb) {
1166                         /*  not this fs or parent went negative */
1167                         inode = d_inode(req->r_dentry);
1168                         if (inode)
1169                                 ihold(inode);
1170                 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
1171                         /* direct snapped/virtual snapdir requests
1172                          * based on parent dir inode */
1173                         inode = get_nonsnap_parent(parent);
1174                         dout("%s using nonsnap parent %p\n", __func__, inode);
1175                 } else {
1176                         /* dentry target */
1177                         inode = d_inode(req->r_dentry);
1178                         if (!inode || mode == USE_AUTH_MDS) {
1179                                 /* dir + name */
1180                                 inode = igrab(dir);
1181                                 hash = ceph_dentry_hash(dir, req->r_dentry);
1182                                 is_hash = true;
1183                         } else {
1184                                 ihold(inode);
1185                         }
1186                 }
1187                 rcu_read_unlock();
1188         }
1189
1190         dout("%s %p is_hash=%d (0x%x) mode %d\n", __func__, inode, (int)is_hash,
1191              hash, mode);
1192         if (!inode)
1193                 goto random;
1194         ci = ceph_inode(inode);
1195
1196         if (is_hash && S_ISDIR(inode->i_mode)) {
1197                 struct ceph_inode_frag frag;
1198                 int found;
1199
1200                 ceph_choose_frag(ci, hash, &frag, &found);
1201                 if (found) {
1202                         if (mode == USE_ANY_MDS && frag.ndist > 0) {
1203                                 u8 r;
1204
1205                                 /* choose a random replica */
1206                                 get_random_bytes(&r, 1);
1207                                 r %= frag.ndist;
1208                                 mds = frag.dist[r];
1209                                 dout("%s %p %llx.%llx frag %u mds%d (%d/%d)\n",
1210                                      __func__, inode, ceph_vinop(inode),
1211                                      frag.frag, mds, (int)r, frag.ndist);
1212                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1213                                     CEPH_MDS_STATE_ACTIVE &&
1214                                     !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
1215                                         goto out;
1216                         }
1217
1218                         /* since this file/dir wasn't known to be
1219                          * replicated, then we want to look for the
1220                          * authoritative mds. */
1221                         if (frag.mds >= 0) {
1222                                 /* choose auth mds */
1223                                 mds = frag.mds;
1224                                 dout("%s %p %llx.%llx frag %u mds%d (auth)\n",
1225                                      __func__, inode, ceph_vinop(inode),
1226                                      frag.frag, mds);
1227                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1228                                     CEPH_MDS_STATE_ACTIVE) {
1229                                         if (!ceph_mdsmap_is_laggy(mdsc->mdsmap,
1230                                                                   mds))
1231                                                 goto out;
1232                                 }
1233                         }
1234                         mode = USE_AUTH_MDS;
1235                 }
1236         }
1237
1238         spin_lock(&ci->i_ceph_lock);
1239         cap = NULL;
1240         if (mode == USE_AUTH_MDS)
1241                 cap = ci->i_auth_cap;
1242         if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1243                 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1244         if (!cap) {
1245                 spin_unlock(&ci->i_ceph_lock);
1246                 iput(inode);
1247                 goto random;
1248         }
1249         mds = cap->session->s_mds;
1250         dout("%s %p %llx.%llx mds%d (%scap %p)\n", __func__,
1251              inode, ceph_vinop(inode), mds,
1252              cap == ci->i_auth_cap ? "auth " : "", cap);
1253         spin_unlock(&ci->i_ceph_lock);
1254 out:
1255         iput(inode);
1256         return mds;
1257
1258 random:
1259         if (random)
1260                 *random = true;
1261
1262         mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
1263         dout("%s chose random mds%d\n", __func__, mds);
1264         return mds;
1265 }
1266
1267
1268 /*
1269  * session messages
1270  */
1271 struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq)
1272 {
1273         struct ceph_msg *msg;
1274         struct ceph_mds_session_head *h;
1275
1276         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1277                            false);
1278         if (!msg) {
1279                 pr_err("ENOMEM creating session %s msg\n",
1280                        ceph_session_op_name(op));
1281                 return NULL;
1282         }
1283         h = msg->front.iov_base;
1284         h->op = cpu_to_le32(op);
1285         h->seq = cpu_to_le64(seq);
1286
1287         return msg;
1288 }
1289
1290 static const unsigned char feature_bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1291 #define FEATURE_BYTES(c) (DIV_ROUND_UP((size_t)feature_bits[c - 1] + 1, 64) * 8)
1292 static int encode_supported_features(void **p, void *end)
1293 {
1294         static const size_t count = ARRAY_SIZE(feature_bits);
1295
1296         if (count > 0) {
1297                 size_t i;
1298                 size_t size = FEATURE_BYTES(count);
1299                 unsigned long bit;
1300
1301                 if (WARN_ON_ONCE(*p + 4 + size > end))
1302                         return -ERANGE;
1303
1304                 ceph_encode_32(p, size);
1305                 memset(*p, 0, size);
1306                 for (i = 0; i < count; i++) {
1307                         bit = feature_bits[i];
1308                         ((unsigned char *)(*p))[bit / 8] |= BIT(bit % 8);
1309                 }
1310                 *p += size;
1311         } else {
1312                 if (WARN_ON_ONCE(*p + 4 > end))
1313                         return -ERANGE;
1314
1315                 ceph_encode_32(p, 0);
1316         }
1317
1318         return 0;
1319 }
1320
1321 static const unsigned char metric_bits[] = CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED;
1322 #define METRIC_BYTES(cnt) (DIV_ROUND_UP((size_t)metric_bits[cnt - 1] + 1, 64) * 8)
1323 static int encode_metric_spec(void **p, void *end)
1324 {
1325         static const size_t count = ARRAY_SIZE(metric_bits);
1326
1327         /* header */
1328         if (WARN_ON_ONCE(*p + 2 > end))
1329                 return -ERANGE;
1330
1331         ceph_encode_8(p, 1); /* version */
1332         ceph_encode_8(p, 1); /* compat */
1333
1334         if (count > 0) {
1335                 size_t i;
1336                 size_t size = METRIC_BYTES(count);
1337
1338                 if (WARN_ON_ONCE(*p + 4 + 4 + size > end))
1339                         return -ERANGE;
1340
1341                 /* metric spec info length */
1342                 ceph_encode_32(p, 4 + size);
1343
1344                 /* metric spec */
1345                 ceph_encode_32(p, size);
1346                 memset(*p, 0, size);
1347                 for (i = 0; i < count; i++)
1348                         ((unsigned char *)(*p))[i / 8] |= BIT(metric_bits[i] % 8);
1349                 *p += size;
1350         } else {
1351                 if (WARN_ON_ONCE(*p + 4 + 4 > end))
1352                         return -ERANGE;
1353
1354                 /* metric spec info length */
1355                 ceph_encode_32(p, 4);
1356                 /* metric spec */
1357                 ceph_encode_32(p, 0);
1358         }
1359
1360         return 0;
1361 }
1362
1363 /*
1364  * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1365  * to include additional client metadata fields.
1366  */
1367 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
1368 {
1369         struct ceph_msg *msg;
1370         struct ceph_mds_session_head *h;
1371         int i;
1372         int extra_bytes = 0;
1373         int metadata_key_count = 0;
1374         struct ceph_options *opt = mdsc->fsc->client->options;
1375         struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
1376         size_t size, count;
1377         void *p, *end;
1378         int ret;
1379
1380         const char* metadata[][2] = {
1381                 {"hostname", mdsc->nodename},
1382                 {"kernel_version", init_utsname()->release},
1383                 {"entity_id", opt->name ? : ""},
1384                 {"root", fsopt->server_path ? : "/"},
1385                 {NULL, NULL}
1386         };
1387
1388         /* Calculate serialized length of metadata */
1389         extra_bytes = 4;  /* map length */
1390         for (i = 0; metadata[i][0]; ++i) {
1391                 extra_bytes += 8 + strlen(metadata[i][0]) +
1392                         strlen(metadata[i][1]);
1393                 metadata_key_count++;
1394         }
1395
1396         /* supported feature */
1397         size = 0;
1398         count = ARRAY_SIZE(feature_bits);
1399         if (count > 0)
1400                 size = FEATURE_BYTES(count);
1401         extra_bytes += 4 + size;
1402
1403         /* metric spec */
1404         size = 0;
1405         count = ARRAY_SIZE(metric_bits);
1406         if (count > 0)
1407                 size = METRIC_BYTES(count);
1408         extra_bytes += 2 + 4 + 4 + size;
1409
1410         /* Allocate the message */
1411         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
1412                            GFP_NOFS, false);
1413         if (!msg) {
1414                 pr_err("ENOMEM creating session open msg\n");
1415                 return ERR_PTR(-ENOMEM);
1416         }
1417         p = msg->front.iov_base;
1418         end = p + msg->front.iov_len;
1419
1420         h = p;
1421         h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
1422         h->seq = cpu_to_le64(seq);
1423
1424         /*
1425          * Serialize client metadata into waiting buffer space, using
1426          * the format that userspace expects for map<string, string>
1427          *
1428          * ClientSession messages with metadata are v4
1429          */
1430         msg->hdr.version = cpu_to_le16(4);
1431         msg->hdr.compat_version = cpu_to_le16(1);
1432
1433         /* The write pointer, following the session_head structure */
1434         p += sizeof(*h);
1435
1436         /* Number of entries in the map */
1437         ceph_encode_32(&p, metadata_key_count);
1438
1439         /* Two length-prefixed strings for each entry in the map */
1440         for (i = 0; metadata[i][0]; ++i) {
1441                 size_t const key_len = strlen(metadata[i][0]);
1442                 size_t const val_len = strlen(metadata[i][1]);
1443
1444                 ceph_encode_32(&p, key_len);
1445                 memcpy(p, metadata[i][0], key_len);
1446                 p += key_len;
1447                 ceph_encode_32(&p, val_len);
1448                 memcpy(p, metadata[i][1], val_len);
1449                 p += val_len;
1450         }
1451
1452         ret = encode_supported_features(&p, end);
1453         if (ret) {
1454                 pr_err("encode_supported_features failed!\n");
1455                 ceph_msg_put(msg);
1456                 return ERR_PTR(ret);
1457         }
1458
1459         ret = encode_metric_spec(&p, end);
1460         if (ret) {
1461                 pr_err("encode_metric_spec failed!\n");
1462                 ceph_msg_put(msg);
1463                 return ERR_PTR(ret);
1464         }
1465
1466         msg->front.iov_len = p - msg->front.iov_base;
1467         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1468
1469         return msg;
1470 }
1471
1472 /*
1473  * send session open request.
1474  *
1475  * called under mdsc->mutex
1476  */
1477 static int __open_session(struct ceph_mds_client *mdsc,
1478                           struct ceph_mds_session *session)
1479 {
1480         struct ceph_msg *msg;
1481         int mstate;
1482         int mds = session->s_mds;
1483
1484         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO)
1485                 return -EIO;
1486
1487         /* wait for mds to go active? */
1488         mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1489         dout("open_session to mds%d (%s)\n", mds,
1490              ceph_mds_state_name(mstate));
1491         session->s_state = CEPH_MDS_SESSION_OPENING;
1492         session->s_renew_requested = jiffies;
1493
1494         /* send connect message */
1495         msg = create_session_open_msg(mdsc, session->s_seq);
1496         if (IS_ERR(msg))
1497                 return PTR_ERR(msg);
1498         ceph_con_send(&session->s_con, msg);
1499         return 0;
1500 }
1501
1502 /*
1503  * open sessions for any export targets for the given mds
1504  *
1505  * called under mdsc->mutex
1506  */
1507 static struct ceph_mds_session *
1508 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1509 {
1510         struct ceph_mds_session *session;
1511         int ret;
1512
1513         session = __ceph_lookup_mds_session(mdsc, target);
1514         if (!session) {
1515                 session = register_session(mdsc, target);
1516                 if (IS_ERR(session))
1517                         return session;
1518         }
1519         if (session->s_state == CEPH_MDS_SESSION_NEW ||
1520             session->s_state == CEPH_MDS_SESSION_CLOSING) {
1521                 ret = __open_session(mdsc, session);
1522                 if (ret)
1523                         return ERR_PTR(ret);
1524         }
1525
1526         return session;
1527 }
1528
1529 struct ceph_mds_session *
1530 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1531 {
1532         struct ceph_mds_session *session;
1533
1534         dout("open_export_target_session to mds%d\n", target);
1535
1536         mutex_lock(&mdsc->mutex);
1537         session = __open_export_target_session(mdsc, target);
1538         mutex_unlock(&mdsc->mutex);
1539
1540         return session;
1541 }
1542
1543 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1544                                           struct ceph_mds_session *session)
1545 {
1546         struct ceph_mds_info *mi;
1547         struct ceph_mds_session *ts;
1548         int i, mds = session->s_mds;
1549
1550         if (mds >= mdsc->mdsmap->possible_max_rank)
1551                 return;
1552
1553         mi = &mdsc->mdsmap->m_info[mds];
1554         dout("open_export_target_sessions for mds%d (%d targets)\n",
1555              session->s_mds, mi->num_export_targets);
1556
1557         for (i = 0; i < mi->num_export_targets; i++) {
1558                 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1559                 ceph_put_mds_session(ts);
1560         }
1561 }
1562
1563 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1564                                            struct ceph_mds_session *session)
1565 {
1566         mutex_lock(&mdsc->mutex);
1567         __open_export_target_sessions(mdsc, session);
1568         mutex_unlock(&mdsc->mutex);
1569 }
1570
1571 /*
1572  * session caps
1573  */
1574
1575 static void detach_cap_releases(struct ceph_mds_session *session,
1576                                 struct list_head *target)
1577 {
1578         lockdep_assert_held(&session->s_cap_lock);
1579
1580         list_splice_init(&session->s_cap_releases, target);
1581         session->s_num_cap_releases = 0;
1582         dout("dispose_cap_releases mds%d\n", session->s_mds);
1583 }
1584
1585 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1586                                  struct list_head *dispose)
1587 {
1588         while (!list_empty(dispose)) {
1589                 struct ceph_cap *cap;
1590                 /* zero out the in-progress message */
1591                 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1592                 list_del(&cap->session_caps);
1593                 ceph_put_cap(mdsc, cap);
1594         }
1595 }
1596
1597 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1598                                      struct ceph_mds_session *session)
1599 {
1600         struct ceph_mds_request *req;
1601         struct rb_node *p;
1602
1603         dout("cleanup_session_requests mds%d\n", session->s_mds);
1604         mutex_lock(&mdsc->mutex);
1605         while (!list_empty(&session->s_unsafe)) {
1606                 req = list_first_entry(&session->s_unsafe,
1607                                        struct ceph_mds_request, r_unsafe_item);
1608                 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1609                                     req->r_tid);
1610                 if (req->r_target_inode)
1611                         mapping_set_error(req->r_target_inode->i_mapping, -EIO);
1612                 if (req->r_unsafe_dir)
1613                         mapping_set_error(req->r_unsafe_dir->i_mapping, -EIO);
1614                 __unregister_request(mdsc, req);
1615         }
1616         /* zero r_attempts, so kick_requests() will re-send requests */
1617         p = rb_first(&mdsc->request_tree);
1618         while (p) {
1619                 req = rb_entry(p, struct ceph_mds_request, r_node);
1620                 p = rb_next(p);
1621                 if (req->r_session &&
1622                     req->r_session->s_mds == session->s_mds)
1623                         req->r_attempts = 0;
1624         }
1625         mutex_unlock(&mdsc->mutex);
1626 }
1627
1628 /*
1629  * Helper to safely iterate over all caps associated with a session, with
1630  * special care taken to handle a racing __ceph_remove_cap().
1631  *
1632  * Caller must hold session s_mutex.
1633  */
1634 int ceph_iterate_session_caps(struct ceph_mds_session *session,
1635                               int (*cb)(struct inode *, int mds, void *),
1636                               void *arg)
1637 {
1638         struct list_head *p;
1639         struct ceph_cap *cap;
1640         struct inode *inode, *last_inode = NULL;
1641         struct ceph_cap *old_cap = NULL;
1642         int ret;
1643
1644         dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1645         spin_lock(&session->s_cap_lock);
1646         p = session->s_caps.next;
1647         while (p != &session->s_caps) {
1648                 int mds;
1649
1650                 cap = list_entry(p, struct ceph_cap, session_caps);
1651                 inode = igrab(&cap->ci->netfs.inode);
1652                 if (!inode) {
1653                         p = p->next;
1654                         continue;
1655                 }
1656                 session->s_cap_iterator = cap;
1657                 mds = cap->mds;
1658                 spin_unlock(&session->s_cap_lock);
1659
1660                 if (last_inode) {
1661                         iput(last_inode);
1662                         last_inode = NULL;
1663                 }
1664                 if (old_cap) {
1665                         ceph_put_cap(session->s_mdsc, old_cap);
1666                         old_cap = NULL;
1667                 }
1668
1669                 ret = cb(inode, mds, arg);
1670                 last_inode = inode;
1671
1672                 spin_lock(&session->s_cap_lock);
1673                 p = p->next;
1674                 if (!cap->ci) {
1675                         dout("iterate_session_caps  finishing cap %p removal\n",
1676                              cap);
1677                         BUG_ON(cap->session != session);
1678                         cap->session = NULL;
1679                         list_del_init(&cap->session_caps);
1680                         session->s_nr_caps--;
1681                         atomic64_dec(&session->s_mdsc->metric.total_caps);
1682                         if (cap->queue_release)
1683                                 __ceph_queue_cap_release(session, cap);
1684                         else
1685                                 old_cap = cap;  /* put_cap it w/o locks held */
1686                 }
1687                 if (ret < 0)
1688                         goto out;
1689         }
1690         ret = 0;
1691 out:
1692         session->s_cap_iterator = NULL;
1693         spin_unlock(&session->s_cap_lock);
1694
1695         iput(last_inode);
1696         if (old_cap)
1697                 ceph_put_cap(session->s_mdsc, old_cap);
1698
1699         return ret;
1700 }
1701
1702 static int remove_session_caps_cb(struct inode *inode, int mds, void *arg)
1703 {
1704         struct ceph_inode_info *ci = ceph_inode(inode);
1705         bool invalidate = false;
1706         struct ceph_cap *cap;
1707         int iputs = 0;
1708
1709         spin_lock(&ci->i_ceph_lock);
1710         cap = __get_cap_for_mds(ci, mds);
1711         if (cap) {
1712                 dout(" removing cap %p, ci is %p, inode is %p\n",
1713                      cap, ci, &ci->netfs.inode);
1714
1715                 iputs = ceph_purge_inode_cap(inode, cap, &invalidate);
1716         }
1717         spin_unlock(&ci->i_ceph_lock);
1718
1719         if (cap)
1720                 wake_up_all(&ci->i_cap_wq);
1721         if (invalidate)
1722                 ceph_queue_invalidate(inode);
1723         while (iputs--)
1724                 iput(inode);
1725         return 0;
1726 }
1727
1728 /*
1729  * caller must hold session s_mutex
1730  */
1731 static void remove_session_caps(struct ceph_mds_session *session)
1732 {
1733         struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1734         struct super_block *sb = fsc->sb;
1735         LIST_HEAD(dispose);
1736
1737         dout("remove_session_caps on %p\n", session);
1738         ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
1739
1740         wake_up_all(&fsc->mdsc->cap_flushing_wq);
1741
1742         spin_lock(&session->s_cap_lock);
1743         if (session->s_nr_caps > 0) {
1744                 struct inode *inode;
1745                 struct ceph_cap *cap, *prev = NULL;
1746                 struct ceph_vino vino;
1747                 /*
1748                  * iterate_session_caps() skips inodes that are being
1749                  * deleted, we need to wait until deletions are complete.
1750                  * __wait_on_freeing_inode() is designed for the job,
1751                  * but it is not exported, so use lookup inode function
1752                  * to access it.
1753                  */
1754                 while (!list_empty(&session->s_caps)) {
1755                         cap = list_entry(session->s_caps.next,
1756                                          struct ceph_cap, session_caps);
1757                         if (cap == prev)
1758                                 break;
1759                         prev = cap;
1760                         vino = cap->ci->i_vino;
1761                         spin_unlock(&session->s_cap_lock);
1762
1763                         inode = ceph_find_inode(sb, vino);
1764                         iput(inode);
1765
1766                         spin_lock(&session->s_cap_lock);
1767                 }
1768         }
1769
1770         // drop cap expires and unlock s_cap_lock
1771         detach_cap_releases(session, &dispose);
1772
1773         BUG_ON(session->s_nr_caps > 0);
1774         BUG_ON(!list_empty(&session->s_cap_flushing));
1775         spin_unlock(&session->s_cap_lock);
1776         dispose_cap_releases(session->s_mdsc, &dispose);
1777 }
1778
1779 enum {
1780         RECONNECT,
1781         RENEWCAPS,
1782         FORCE_RO,
1783 };
1784
1785 /*
1786  * wake up any threads waiting on this session's caps.  if the cap is
1787  * old (didn't get renewed on the client reconnect), remove it now.
1788  *
1789  * caller must hold s_mutex.
1790  */
1791 static int wake_up_session_cb(struct inode *inode, int mds, 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                 struct ceph_cap *cap;
1803
1804                 spin_lock(&ci->i_ceph_lock);
1805                 cap = __get_cap_for_mds(ci, mds);
1806                 /* mds did not re-issue stale cap */
1807                 if (cap && cap->cap_gen < atomic_read(&cap->session->s_cap_gen))
1808                         cap->issued = cap->implemented = CEPH_CAP_PIN;
1809                 spin_unlock(&ci->i_ceph_lock);
1810         } else if (ev == FORCE_RO) {
1811         }
1812         wake_up_all(&ci->i_cap_wq);
1813         return 0;
1814 }
1815
1816 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
1817 {
1818         dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1819         ceph_iterate_session_caps(session, wake_up_session_cb,
1820                                   (void *)(unsigned long)ev);
1821 }
1822
1823 /*
1824  * Send periodic message to MDS renewing all currently held caps.  The
1825  * ack will reset the expiration for all caps from this session.
1826  *
1827  * caller holds s_mutex
1828  */
1829 static int send_renew_caps(struct ceph_mds_client *mdsc,
1830                            struct ceph_mds_session *session)
1831 {
1832         struct ceph_msg *msg;
1833         int state;
1834
1835         if (time_after_eq(jiffies, session->s_cap_ttl) &&
1836             time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1837                 pr_info("mds%d caps stale\n", session->s_mds);
1838         session->s_renew_requested = jiffies;
1839
1840         /* do not try to renew caps until a recovering mds has reconnected
1841          * with its clients. */
1842         state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1843         if (state < CEPH_MDS_STATE_RECONNECT) {
1844                 dout("send_renew_caps ignoring mds%d (%s)\n",
1845                      session->s_mds, ceph_mds_state_name(state));
1846                 return 0;
1847         }
1848
1849         dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1850                 ceph_mds_state_name(state));
1851         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1852                                       ++session->s_renew_seq);
1853         if (!msg)
1854                 return -ENOMEM;
1855         ceph_con_send(&session->s_con, msg);
1856         return 0;
1857 }
1858
1859 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1860                              struct ceph_mds_session *session, u64 seq)
1861 {
1862         struct ceph_msg *msg;
1863
1864         dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1865              session->s_mds, ceph_session_state_name(session->s_state), seq);
1866         msg = ceph_create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1867         if (!msg)
1868                 return -ENOMEM;
1869         ceph_con_send(&session->s_con, msg);
1870         return 0;
1871 }
1872
1873
1874 /*
1875  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1876  *
1877  * Called under session->s_mutex
1878  */
1879 static void renewed_caps(struct ceph_mds_client *mdsc,
1880                          struct ceph_mds_session *session, int is_renew)
1881 {
1882         int was_stale;
1883         int wake = 0;
1884
1885         spin_lock(&session->s_cap_lock);
1886         was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1887
1888         session->s_cap_ttl = session->s_renew_requested +
1889                 mdsc->mdsmap->m_session_timeout*HZ;
1890
1891         if (was_stale) {
1892                 if (time_before(jiffies, session->s_cap_ttl)) {
1893                         pr_info("mds%d caps renewed\n", session->s_mds);
1894                         wake = 1;
1895                 } else {
1896                         pr_info("mds%d caps still stale\n", session->s_mds);
1897                 }
1898         }
1899         dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1900              session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1901              time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1902         spin_unlock(&session->s_cap_lock);
1903
1904         if (wake)
1905                 wake_up_session_caps(session, RENEWCAPS);
1906 }
1907
1908 /*
1909  * send a session close request
1910  */
1911 static int request_close_session(struct ceph_mds_session *session)
1912 {
1913         struct ceph_msg *msg;
1914
1915         dout("request_close_session mds%d state %s seq %lld\n",
1916              session->s_mds, ceph_session_state_name(session->s_state),
1917              session->s_seq);
1918         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_CLOSE,
1919                                       session->s_seq);
1920         if (!msg)
1921                 return -ENOMEM;
1922         ceph_con_send(&session->s_con, msg);
1923         return 1;
1924 }
1925
1926 /*
1927  * Called with s_mutex held.
1928  */
1929 static int __close_session(struct ceph_mds_client *mdsc,
1930                          struct ceph_mds_session *session)
1931 {
1932         if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1933                 return 0;
1934         session->s_state = CEPH_MDS_SESSION_CLOSING;
1935         return request_close_session(session);
1936 }
1937
1938 static bool drop_negative_children(struct dentry *dentry)
1939 {
1940         struct dentry *child;
1941         bool all_negative = true;
1942
1943         if (!d_is_dir(dentry))
1944                 goto out;
1945
1946         spin_lock(&dentry->d_lock);
1947         list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1948                 if (d_really_is_positive(child)) {
1949                         all_negative = false;
1950                         break;
1951                 }
1952         }
1953         spin_unlock(&dentry->d_lock);
1954
1955         if (all_negative)
1956                 shrink_dcache_parent(dentry);
1957 out:
1958         return all_negative;
1959 }
1960
1961 /*
1962  * Trim old(er) caps.
1963  *
1964  * Because we can't cache an inode without one or more caps, we do
1965  * this indirectly: if a cap is unused, we prune its aliases, at which
1966  * point the inode will hopefully get dropped to.
1967  *
1968  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1969  * memory pressure from the MDS, though, so it needn't be perfect.
1970  */
1971 static int trim_caps_cb(struct inode *inode, int mds, void *arg)
1972 {
1973         int *remaining = arg;
1974         struct ceph_inode_info *ci = ceph_inode(inode);
1975         int used, wanted, oissued, mine;
1976         struct ceph_cap *cap;
1977
1978         if (*remaining <= 0)
1979                 return -1;
1980
1981         spin_lock(&ci->i_ceph_lock);
1982         cap = __get_cap_for_mds(ci, mds);
1983         if (!cap) {
1984                 spin_unlock(&ci->i_ceph_lock);
1985                 return 0;
1986         }
1987         mine = cap->issued | cap->implemented;
1988         used = __ceph_caps_used(ci);
1989         wanted = __ceph_caps_file_wanted(ci);
1990         oissued = __ceph_caps_issued_other(ci, cap);
1991
1992         dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1993              inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1994              ceph_cap_string(used), ceph_cap_string(wanted));
1995         if (cap == ci->i_auth_cap) {
1996                 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1997                     !list_empty(&ci->i_cap_snaps))
1998                         goto out;
1999                 if ((used | wanted) & CEPH_CAP_ANY_WR)
2000                         goto out;
2001                 /* Note: it's possible that i_filelock_ref becomes non-zero
2002                  * after dropping auth caps. It doesn't hurt because reply
2003                  * of lock mds request will re-add auth caps. */
2004                 if (atomic_read(&ci->i_filelock_ref) > 0)
2005                         goto out;
2006         }
2007         /* The inode has cached pages, but it's no longer used.
2008          * we can safely drop it */
2009         if (S_ISREG(inode->i_mode) &&
2010             wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
2011             !(oissued & CEPH_CAP_FILE_CACHE)) {
2012           used = 0;
2013           oissued = 0;
2014         }
2015         if ((used | wanted) & ~oissued & mine)
2016                 goto out;   /* we need these caps */
2017
2018         if (oissued) {
2019                 /* we aren't the only cap.. just remove us */
2020                 ceph_remove_cap(cap, true);
2021                 (*remaining)--;
2022         } else {
2023                 struct dentry *dentry;
2024                 /* try dropping referring dentries */
2025                 spin_unlock(&ci->i_ceph_lock);
2026                 dentry = d_find_any_alias(inode);
2027                 if (dentry && drop_negative_children(dentry)) {
2028                         int count;
2029                         dput(dentry);
2030                         d_prune_aliases(inode);
2031                         count = atomic_read(&inode->i_count);
2032                         if (count == 1)
2033                                 (*remaining)--;
2034                         dout("trim_caps_cb %p cap %p pruned, count now %d\n",
2035                              inode, cap, count);
2036                 } else {
2037                         dput(dentry);
2038                 }
2039                 return 0;
2040         }
2041
2042 out:
2043         spin_unlock(&ci->i_ceph_lock);
2044         return 0;
2045 }
2046
2047 /*
2048  * Trim session cap count down to some max number.
2049  */
2050 int ceph_trim_caps(struct ceph_mds_client *mdsc,
2051                    struct ceph_mds_session *session,
2052                    int max_caps)
2053 {
2054         int trim_caps = session->s_nr_caps - max_caps;
2055
2056         dout("trim_caps mds%d start: %d / %d, trim %d\n",
2057              session->s_mds, session->s_nr_caps, max_caps, trim_caps);
2058         if (trim_caps > 0) {
2059                 int remaining = trim_caps;
2060
2061                 ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
2062                 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
2063                      session->s_mds, session->s_nr_caps, max_caps,
2064                         trim_caps - remaining);
2065         }
2066
2067         ceph_flush_cap_releases(mdsc, session);
2068         return 0;
2069 }
2070
2071 static int check_caps_flush(struct ceph_mds_client *mdsc,
2072                             u64 want_flush_tid)
2073 {
2074         int ret = 1;
2075
2076         spin_lock(&mdsc->cap_dirty_lock);
2077         if (!list_empty(&mdsc->cap_flush_list)) {
2078                 struct ceph_cap_flush *cf =
2079                         list_first_entry(&mdsc->cap_flush_list,
2080                                          struct ceph_cap_flush, g_list);
2081                 if (cf->tid <= want_flush_tid) {
2082                         dout("check_caps_flush still flushing tid "
2083                              "%llu <= %llu\n", cf->tid, want_flush_tid);
2084                         ret = 0;
2085                 }
2086         }
2087         spin_unlock(&mdsc->cap_dirty_lock);
2088         return ret;
2089 }
2090
2091 /*
2092  * flush all dirty inode data to disk.
2093  *
2094  * returns true if we've flushed through want_flush_tid
2095  */
2096 static void wait_caps_flush(struct ceph_mds_client *mdsc,
2097                             u64 want_flush_tid)
2098 {
2099         dout("check_caps_flush want %llu\n", want_flush_tid);
2100
2101         wait_event(mdsc->cap_flushing_wq,
2102                    check_caps_flush(mdsc, want_flush_tid));
2103
2104         dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2105 }
2106
2107 /*
2108  * called under s_mutex
2109  */
2110 static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
2111                                    struct ceph_mds_session *session)
2112 {
2113         struct ceph_msg *msg = NULL;
2114         struct ceph_mds_cap_release *head;
2115         struct ceph_mds_cap_item *item;
2116         struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
2117         struct ceph_cap *cap;
2118         LIST_HEAD(tmp_list);
2119         int num_cap_releases;
2120         __le32  barrier, *cap_barrier;
2121
2122         down_read(&osdc->lock);
2123         barrier = cpu_to_le32(osdc->epoch_barrier);
2124         up_read(&osdc->lock);
2125
2126         spin_lock(&session->s_cap_lock);
2127 again:
2128         list_splice_init(&session->s_cap_releases, &tmp_list);
2129         num_cap_releases = session->s_num_cap_releases;
2130         session->s_num_cap_releases = 0;
2131         spin_unlock(&session->s_cap_lock);
2132
2133         while (!list_empty(&tmp_list)) {
2134                 if (!msg) {
2135                         msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
2136                                         PAGE_SIZE, GFP_NOFS, false);
2137                         if (!msg)
2138                                 goto out_err;
2139                         head = msg->front.iov_base;
2140                         head->num = cpu_to_le32(0);
2141                         msg->front.iov_len = sizeof(*head);
2142
2143                         msg->hdr.version = cpu_to_le16(2);
2144                         msg->hdr.compat_version = cpu_to_le16(1);
2145                 }
2146
2147                 cap = list_first_entry(&tmp_list, struct ceph_cap,
2148                                         session_caps);
2149                 list_del(&cap->session_caps);
2150                 num_cap_releases--;
2151
2152                 head = msg->front.iov_base;
2153                 put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
2154                                    &head->num);
2155                 item = msg->front.iov_base + msg->front.iov_len;
2156                 item->ino = cpu_to_le64(cap->cap_ino);
2157                 item->cap_id = cpu_to_le64(cap->cap_id);
2158                 item->migrate_seq = cpu_to_le32(cap->mseq);
2159                 item->seq = cpu_to_le32(cap->issue_seq);
2160                 msg->front.iov_len += sizeof(*item);
2161
2162                 ceph_put_cap(mdsc, cap);
2163
2164                 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
2165                         // Append cap_barrier field
2166                         cap_barrier = msg->front.iov_base + msg->front.iov_len;
2167                         *cap_barrier = barrier;
2168                         msg->front.iov_len += sizeof(*cap_barrier);
2169
2170                         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2171                         dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2172                         ceph_con_send(&session->s_con, msg);
2173                         msg = NULL;
2174                 }
2175         }
2176
2177         BUG_ON(num_cap_releases != 0);
2178
2179         spin_lock(&session->s_cap_lock);
2180         if (!list_empty(&session->s_cap_releases))
2181                 goto again;
2182         spin_unlock(&session->s_cap_lock);
2183
2184         if (msg) {
2185                 // Append cap_barrier field
2186                 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2187                 *cap_barrier = barrier;
2188                 msg->front.iov_len += sizeof(*cap_barrier);
2189
2190                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2191                 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2192                 ceph_con_send(&session->s_con, msg);
2193         }
2194         return;
2195 out_err:
2196         pr_err("send_cap_releases mds%d, failed to allocate message\n",
2197                 session->s_mds);
2198         spin_lock(&session->s_cap_lock);
2199         list_splice(&tmp_list, &session->s_cap_releases);
2200         session->s_num_cap_releases += num_cap_releases;
2201         spin_unlock(&session->s_cap_lock);
2202 }
2203
2204 static void ceph_cap_release_work(struct work_struct *work)
2205 {
2206         struct ceph_mds_session *session =
2207                 container_of(work, struct ceph_mds_session, s_cap_release_work);
2208
2209         mutex_lock(&session->s_mutex);
2210         if (session->s_state == CEPH_MDS_SESSION_OPEN ||
2211             session->s_state == CEPH_MDS_SESSION_HUNG)
2212                 ceph_send_cap_releases(session->s_mdsc, session);
2213         mutex_unlock(&session->s_mutex);
2214         ceph_put_mds_session(session);
2215 }
2216
2217 void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
2218                              struct ceph_mds_session *session)
2219 {
2220         if (mdsc->stopping)
2221                 return;
2222
2223         ceph_get_mds_session(session);
2224         if (queue_work(mdsc->fsc->cap_wq,
2225                        &session->s_cap_release_work)) {
2226                 dout("cap release work queued\n");
2227         } else {
2228                 ceph_put_mds_session(session);
2229                 dout("failed to queue cap release work\n");
2230         }
2231 }
2232
2233 /*
2234  * caller holds session->s_cap_lock
2235  */
2236 void __ceph_queue_cap_release(struct ceph_mds_session *session,
2237                               struct ceph_cap *cap)
2238 {
2239         list_add_tail(&cap->session_caps, &session->s_cap_releases);
2240         session->s_num_cap_releases++;
2241
2242         if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
2243                 ceph_flush_cap_releases(session->s_mdsc, session);
2244 }
2245
2246 static void ceph_cap_reclaim_work(struct work_struct *work)
2247 {
2248         struct ceph_mds_client *mdsc =
2249                 container_of(work, struct ceph_mds_client, cap_reclaim_work);
2250         int ret = ceph_trim_dentries(mdsc);
2251         if (ret == -EAGAIN)
2252                 ceph_queue_cap_reclaim_work(mdsc);
2253 }
2254
2255 void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2256 {
2257         if (mdsc->stopping)
2258                 return;
2259
2260         if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
2261                 dout("caps reclaim work queued\n");
2262         } else {
2263                 dout("failed to queue caps release work\n");
2264         }
2265 }
2266
2267 void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2268 {
2269         int val;
2270         if (!nr)
2271                 return;
2272         val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
2273         if ((val % CEPH_CAPS_PER_RELEASE) < nr) {
2274                 atomic_set(&mdsc->cap_reclaim_pending, 0);
2275                 ceph_queue_cap_reclaim_work(mdsc);
2276         }
2277 }
2278
2279 /*
2280  * requests
2281  */
2282
2283 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2284                                     struct inode *dir)
2285 {
2286         struct ceph_inode_info *ci = ceph_inode(dir);
2287         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2288         struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2289         size_t size = sizeof(struct ceph_mds_reply_dir_entry);
2290         unsigned int num_entries;
2291         int order;
2292
2293         spin_lock(&ci->i_ceph_lock);
2294         num_entries = ci->i_files + ci->i_subdirs;
2295         spin_unlock(&ci->i_ceph_lock);
2296         num_entries = max(num_entries, 1U);
2297         num_entries = min(num_entries, opt->max_readdir);
2298
2299         order = get_order(size * num_entries);
2300         while (order >= 0) {
2301                 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2302                                                              __GFP_NOWARN |
2303                                                              __GFP_ZERO,
2304                                                              order);
2305                 if (rinfo->dir_entries)
2306                         break;
2307                 order--;
2308         }
2309         if (!rinfo->dir_entries)
2310                 return -ENOMEM;
2311
2312         num_entries = (PAGE_SIZE << order) / size;
2313         num_entries = min(num_entries, opt->max_readdir);
2314
2315         rinfo->dir_buf_size = PAGE_SIZE << order;
2316         req->r_num_caps = num_entries + 1;
2317         req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2318         req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2319         return 0;
2320 }
2321
2322 /*
2323  * Create an mds request.
2324  */
2325 struct ceph_mds_request *
2326 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2327 {
2328         struct ceph_mds_request *req;
2329
2330         req = kmem_cache_zalloc(ceph_mds_request_cachep, GFP_NOFS);
2331         if (!req)
2332                 return ERR_PTR(-ENOMEM);
2333
2334         mutex_init(&req->r_fill_mutex);
2335         req->r_mdsc = mdsc;
2336         req->r_started = jiffies;
2337         req->r_start_latency = ktime_get();
2338         req->r_resend_mds = -1;
2339         INIT_LIST_HEAD(&req->r_unsafe_dir_item);
2340         INIT_LIST_HEAD(&req->r_unsafe_target_item);
2341         req->r_fmode = -1;
2342         req->r_feature_needed = -1;
2343         kref_init(&req->r_kref);
2344         RB_CLEAR_NODE(&req->r_node);
2345         INIT_LIST_HEAD(&req->r_wait);
2346         init_completion(&req->r_completion);
2347         init_completion(&req->r_safe_completion);
2348         INIT_LIST_HEAD(&req->r_unsafe_item);
2349
2350         ktime_get_coarse_real_ts64(&req->r_stamp);
2351
2352         req->r_op = op;
2353         req->r_direct_mode = mode;
2354         return req;
2355 }
2356
2357 /*
2358  * return oldest (lowest) request, tid in request tree, 0 if none.
2359  *
2360  * called under mdsc->mutex.
2361  */
2362 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2363 {
2364         if (RB_EMPTY_ROOT(&mdsc->request_tree))
2365                 return NULL;
2366         return rb_entry(rb_first(&mdsc->request_tree),
2367                         struct ceph_mds_request, r_node);
2368 }
2369
2370 static inline  u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2371 {
2372         return mdsc->oldest_tid;
2373 }
2374
2375 /*
2376  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
2377  * on build_path_from_dentry in fs/cifs/dir.c.
2378  *
2379  * If @stop_on_nosnap, generate path relative to the first non-snapped
2380  * inode.
2381  *
2382  * Encode hidden .snap dirs as a double /, i.e.
2383  *   foo/.snap/bar -> foo//bar
2384  */
2385 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *pbase,
2386                            int stop_on_nosnap)
2387 {
2388         struct dentry *temp;
2389         char *path;
2390         int pos;
2391         unsigned seq;
2392         u64 base;
2393
2394         if (!dentry)
2395                 return ERR_PTR(-EINVAL);
2396
2397         path = __getname();
2398         if (!path)
2399                 return ERR_PTR(-ENOMEM);
2400 retry:
2401         pos = PATH_MAX - 1;
2402         path[pos] = '\0';
2403
2404         seq = read_seqbegin(&rename_lock);
2405         rcu_read_lock();
2406         temp = dentry;
2407         for (;;) {
2408                 struct inode *inode;
2409
2410                 spin_lock(&temp->d_lock);
2411                 inode = d_inode(temp);
2412                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
2413                         dout("build_path path+%d: %p SNAPDIR\n",
2414                              pos, temp);
2415                 } else if (stop_on_nosnap && inode && dentry != temp &&
2416                            ceph_snap(inode) == CEPH_NOSNAP) {
2417                         spin_unlock(&temp->d_lock);
2418                         pos++; /* get rid of any prepended '/' */
2419                         break;
2420                 } else {
2421                         pos -= temp->d_name.len;
2422                         if (pos < 0) {
2423                                 spin_unlock(&temp->d_lock);
2424                                 break;
2425                         }
2426                         memcpy(path + pos, temp->d_name.name, temp->d_name.len);
2427                 }
2428                 spin_unlock(&temp->d_lock);
2429                 temp = READ_ONCE(temp->d_parent);
2430
2431                 /* Are we at the root? */
2432                 if (IS_ROOT(temp))
2433                         break;
2434
2435                 /* Are we out of buffer? */
2436                 if (--pos < 0)
2437                         break;
2438
2439                 path[pos] = '/';
2440         }
2441         base = ceph_ino(d_inode(temp));
2442         rcu_read_unlock();
2443
2444         if (read_seqretry(&rename_lock, seq))
2445                 goto retry;
2446
2447         if (pos < 0) {
2448                 /*
2449                  * A rename didn't occur, but somehow we didn't end up where
2450                  * we thought we would. Throw a warning and try again.
2451                  */
2452                 pr_warn("build_path did not end path lookup where "
2453                         "expected, pos is %d\n", pos);
2454                 goto retry;
2455         }
2456
2457         *pbase = base;
2458         *plen = PATH_MAX - 1 - pos;
2459         dout("build_path on %p %d built %llx '%.*s'\n",
2460              dentry, d_count(dentry), base, *plen, path + pos);
2461         return path + pos;
2462 }
2463
2464 static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2465                              const char **ppath, int *ppathlen, u64 *pino,
2466                              bool *pfreepath, bool parent_locked)
2467 {
2468         char *path;
2469
2470         rcu_read_lock();
2471         if (!dir)
2472                 dir = d_inode_rcu(dentry->d_parent);
2473         if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP) {
2474                 *pino = ceph_ino(dir);
2475                 rcu_read_unlock();
2476                 *ppath = dentry->d_name.name;
2477                 *ppathlen = dentry->d_name.len;
2478                 return 0;
2479         }
2480         rcu_read_unlock();
2481         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2482         if (IS_ERR(path))
2483                 return PTR_ERR(path);
2484         *ppath = path;
2485         *pfreepath = true;
2486         return 0;
2487 }
2488
2489 static int build_inode_path(struct inode *inode,
2490                             const char **ppath, int *ppathlen, u64 *pino,
2491                             bool *pfreepath)
2492 {
2493         struct dentry *dentry;
2494         char *path;
2495
2496         if (ceph_snap(inode) == CEPH_NOSNAP) {
2497                 *pino = ceph_ino(inode);
2498                 *ppathlen = 0;
2499                 return 0;
2500         }
2501         dentry = d_find_alias(inode);
2502         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2503         dput(dentry);
2504         if (IS_ERR(path))
2505                 return PTR_ERR(path);
2506         *ppath = path;
2507         *pfreepath = true;
2508         return 0;
2509 }
2510
2511 /*
2512  * request arguments may be specified via an inode *, a dentry *, or
2513  * an explicit ino+path.
2514  */
2515 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
2516                                   struct inode *rdiri, const char *rpath,
2517                                   u64 rino, const char **ppath, int *pathlen,
2518                                   u64 *ino, bool *freepath, bool parent_locked)
2519 {
2520         int r = 0;
2521
2522         if (rinode) {
2523                 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
2524                 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2525                      ceph_snap(rinode));
2526         } else if (rdentry) {
2527                 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
2528                                         freepath, parent_locked);
2529                 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
2530                      *ppath);
2531         } else if (rpath || rino) {
2532                 *ino = rino;
2533                 *ppath = rpath;
2534                 *pathlen = rpath ? strlen(rpath) : 0;
2535                 dout(" path %.*s\n", *pathlen, rpath);
2536         }
2537
2538         return r;
2539 }
2540
2541 static void encode_timestamp_and_gids(void **p,
2542                                       const struct ceph_mds_request *req)
2543 {
2544         struct ceph_timespec ts;
2545         int i;
2546
2547         ceph_encode_timespec64(&ts, &req->r_stamp);
2548         ceph_encode_copy(p, &ts, sizeof(ts));
2549
2550         /* gid_list */
2551         ceph_encode_32(p, req->r_cred->group_info->ngroups);
2552         for (i = 0; i < req->r_cred->group_info->ngroups; i++)
2553                 ceph_encode_64(p, from_kgid(&init_user_ns,
2554                                             req->r_cred->group_info->gid[i]));
2555 }
2556
2557 /*
2558  * called under mdsc->mutex
2559  */
2560 static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
2561                                                struct ceph_mds_request *req,
2562                                                bool drop_cap_releases)
2563 {
2564         int mds = session->s_mds;
2565         struct ceph_mds_client *mdsc = session->s_mdsc;
2566         struct ceph_msg *msg;
2567         struct ceph_mds_request_head_old *head;
2568         const char *path1 = NULL;
2569         const char *path2 = NULL;
2570         u64 ino1 = 0, ino2 = 0;
2571         int pathlen1 = 0, pathlen2 = 0;
2572         bool freepath1 = false, freepath2 = false;
2573         struct dentry *old_dentry = NULL;
2574         int len;
2575         u16 releases;
2576         void *p, *end;
2577         int ret;
2578         bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
2579
2580         ret = set_request_path_attr(req->r_inode, req->r_dentry,
2581                               req->r_parent, req->r_path1, req->r_ino1.ino,
2582                               &path1, &pathlen1, &ino1, &freepath1,
2583                               test_bit(CEPH_MDS_R_PARENT_LOCKED,
2584                                         &req->r_req_flags));
2585         if (ret < 0) {
2586                 msg = ERR_PTR(ret);
2587                 goto out;
2588         }
2589
2590         /* If r_old_dentry is set, then assume that its parent is locked */
2591         if (req->r_old_dentry &&
2592             !(req->r_old_dentry->d_flags & DCACHE_DISCONNECTED))
2593                 old_dentry = req->r_old_dentry;
2594         ret = set_request_path_attr(NULL, old_dentry,
2595                               req->r_old_dentry_dir,
2596                               req->r_path2, req->r_ino2.ino,
2597                               &path2, &pathlen2, &ino2, &freepath2, true);
2598         if (ret < 0) {
2599                 msg = ERR_PTR(ret);
2600                 goto out_free1;
2601         }
2602
2603         len = legacy ? sizeof(*head) : sizeof(struct ceph_mds_request_head);
2604         len += pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
2605                 sizeof(struct ceph_timespec);
2606         len += sizeof(u32) + (sizeof(u64) * req->r_cred->group_info->ngroups);
2607
2608         /* calculate (max) length for cap releases */
2609         len += sizeof(struct ceph_mds_request_release) *
2610                 (!!req->r_inode_drop + !!req->r_dentry_drop +
2611                  !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2612
2613         if (req->r_dentry_drop)
2614                 len += pathlen1;
2615         if (req->r_old_dentry_drop)
2616                 len += pathlen2;
2617
2618         msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
2619         if (!msg) {
2620                 msg = ERR_PTR(-ENOMEM);
2621                 goto out_free2;
2622         }
2623
2624         msg->hdr.tid = cpu_to_le64(req->r_tid);
2625
2626         /*
2627          * The old ceph_mds_request_head didn't contain a version field, and
2628          * one was added when we moved the message version from 3->4.
2629          */
2630         if (legacy) {
2631                 msg->hdr.version = cpu_to_le16(3);
2632                 head = msg->front.iov_base;
2633                 p = msg->front.iov_base + sizeof(*head);
2634         } else {
2635                 struct ceph_mds_request_head *new_head = msg->front.iov_base;
2636
2637                 msg->hdr.version = cpu_to_le16(4);
2638                 new_head->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
2639                 head = (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2640                 p = msg->front.iov_base + sizeof(*new_head);
2641         }
2642
2643         end = msg->front.iov_base + msg->front.iov_len;
2644
2645         head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2646         head->op = cpu_to_le32(req->r_op);
2647         head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns,
2648                                                  req->r_cred->fsuid));
2649         head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns,
2650                                                  req->r_cred->fsgid));
2651         head->ino = cpu_to_le64(req->r_deleg_ino);
2652         head->args = req->r_args;
2653
2654         ceph_encode_filepath(&p, end, ino1, path1);
2655         ceph_encode_filepath(&p, end, ino2, path2);
2656
2657         /* make note of release offset, in case we need to replay */
2658         req->r_request_release_offset = p - msg->front.iov_base;
2659
2660         /* cap releases */
2661         releases = 0;
2662         if (req->r_inode_drop)
2663                 releases += ceph_encode_inode_release(&p,
2664                       req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2665                       mds, req->r_inode_drop, req->r_inode_unless,
2666                       req->r_op == CEPH_MDS_OP_READDIR);
2667         if (req->r_dentry_drop)
2668                 releases += ceph_encode_dentry_release(&p, req->r_dentry,
2669                                 req->r_parent, mds, req->r_dentry_drop,
2670                                 req->r_dentry_unless);
2671         if (req->r_old_dentry_drop)
2672                 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2673                                 req->r_old_dentry_dir, mds,
2674                                 req->r_old_dentry_drop,
2675                                 req->r_old_dentry_unless);
2676         if (req->r_old_inode_drop)
2677                 releases += ceph_encode_inode_release(&p,
2678                       d_inode(req->r_old_dentry),
2679                       mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
2680
2681         if (drop_cap_releases) {
2682                 releases = 0;
2683                 p = msg->front.iov_base + req->r_request_release_offset;
2684         }
2685
2686         head->num_releases = cpu_to_le16(releases);
2687
2688         encode_timestamp_and_gids(&p, req);
2689
2690         if (WARN_ON_ONCE(p > end)) {
2691                 ceph_msg_put(msg);
2692                 msg = ERR_PTR(-ERANGE);
2693                 goto out_free2;
2694         }
2695
2696         msg->front.iov_len = p - msg->front.iov_base;
2697         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2698
2699         if (req->r_pagelist) {
2700                 struct ceph_pagelist *pagelist = req->r_pagelist;
2701                 ceph_msg_data_add_pagelist(msg, pagelist);
2702                 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2703         } else {
2704                 msg->hdr.data_len = 0;
2705         }
2706
2707         msg->hdr.data_off = cpu_to_le16(0);
2708
2709 out_free2:
2710         if (freepath2)
2711                 ceph_mdsc_free_path((char *)path2, pathlen2);
2712 out_free1:
2713         if (freepath1)
2714                 ceph_mdsc_free_path((char *)path1, pathlen1);
2715 out:
2716         return msg;
2717 }
2718
2719 /*
2720  * called under mdsc->mutex if error, under no mutex if
2721  * success.
2722  */
2723 static void complete_request(struct ceph_mds_client *mdsc,
2724                              struct ceph_mds_request *req)
2725 {
2726         req->r_end_latency = ktime_get();
2727
2728         if (req->r_callback)
2729                 req->r_callback(mdsc, req);
2730         complete_all(&req->r_completion);
2731 }
2732
2733 static struct ceph_mds_request_head_old *
2734 find_old_request_head(void *p, u64 features)
2735 {
2736         bool legacy = !(features & CEPH_FEATURE_FS_BTIME);
2737         struct ceph_mds_request_head *new_head;
2738
2739         if (legacy)
2740                 return (struct ceph_mds_request_head_old *)p;
2741         new_head = (struct ceph_mds_request_head *)p;
2742         return (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2743 }
2744
2745 /*
2746  * called under mdsc->mutex
2747  */
2748 static int __prepare_send_request(struct ceph_mds_session *session,
2749                                   struct ceph_mds_request *req,
2750                                   bool drop_cap_releases)
2751 {
2752         int mds = session->s_mds;
2753         struct ceph_mds_client *mdsc = session->s_mdsc;
2754         struct ceph_mds_request_head_old *rhead;
2755         struct ceph_msg *msg;
2756         int flags = 0, max_retry;
2757
2758         /*
2759          * The type of 'r_attempts' in kernel 'ceph_mds_request'
2760          * is 'int', while in 'ceph_mds_request_head' the type of
2761          * 'num_retry' is '__u8'. So in case the request retries
2762          *  exceeding 256 times, the MDS will receive a incorrect
2763          *  retry seq.
2764          *
2765          * In this case it's ususally a bug in MDS and continue
2766          * retrying the request makes no sense.
2767          *
2768          * In future this could be fixed in ceph code, so avoid
2769          * using the hardcode here.
2770          */
2771         max_retry = sizeof_field(struct ceph_mds_request_head, num_retry);
2772         max_retry = 1 << (max_retry * BITS_PER_BYTE);
2773         if (req->r_attempts >= max_retry) {
2774                 pr_warn_ratelimited("%s request tid %llu seq overflow\n",
2775                                     __func__, req->r_tid);
2776                 return -EMULTIHOP;
2777         }
2778
2779         req->r_attempts++;
2780         if (req->r_inode) {
2781                 struct ceph_cap *cap =
2782                         ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2783
2784                 if (cap)
2785                         req->r_sent_on_mseq = cap->mseq;
2786                 else
2787                         req->r_sent_on_mseq = -1;
2788         }
2789         dout("%s %p tid %lld %s (attempt %d)\n", __func__, req,
2790              req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2791
2792         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2793                 void *p;
2794
2795                 /*
2796                  * Replay.  Do not regenerate message (and rebuild
2797                  * paths, etc.); just use the original message.
2798                  * Rebuilding paths will break for renames because
2799                  * d_move mangles the src name.
2800                  */
2801                 msg = req->r_request;
2802                 rhead = find_old_request_head(msg->front.iov_base,
2803                                               session->s_con.peer_features);
2804
2805                 flags = le32_to_cpu(rhead->flags);
2806                 flags |= CEPH_MDS_FLAG_REPLAY;
2807                 rhead->flags = cpu_to_le32(flags);
2808
2809                 if (req->r_target_inode)
2810                         rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2811
2812                 rhead->num_retry = req->r_attempts - 1;
2813
2814                 /* remove cap/dentry releases from message */
2815                 rhead->num_releases = 0;
2816
2817                 p = msg->front.iov_base + req->r_request_release_offset;
2818                 encode_timestamp_and_gids(&p, req);
2819
2820                 msg->front.iov_len = p - msg->front.iov_base;
2821                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2822                 return 0;
2823         }
2824
2825         if (req->r_request) {
2826                 ceph_msg_put(req->r_request);
2827                 req->r_request = NULL;
2828         }
2829         msg = create_request_message(session, req, drop_cap_releases);
2830         if (IS_ERR(msg)) {
2831                 req->r_err = PTR_ERR(msg);
2832                 return PTR_ERR(msg);
2833         }
2834         req->r_request = msg;
2835
2836         rhead = find_old_request_head(msg->front.iov_base,
2837                                       session->s_con.peer_features);
2838         rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2839         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2840                 flags |= CEPH_MDS_FLAG_REPLAY;
2841         if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags))
2842                 flags |= CEPH_MDS_FLAG_ASYNC;
2843         if (req->r_parent)
2844                 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2845         rhead->flags = cpu_to_le32(flags);
2846         rhead->num_fwd = req->r_num_fwd;
2847         rhead->num_retry = req->r_attempts - 1;
2848
2849         dout(" r_parent = %p\n", req->r_parent);
2850         return 0;
2851 }
2852
2853 /*
2854  * called under mdsc->mutex
2855  */
2856 static int __send_request(struct ceph_mds_session *session,
2857                           struct ceph_mds_request *req,
2858                           bool drop_cap_releases)
2859 {
2860         int err;
2861
2862         err = __prepare_send_request(session, req, drop_cap_releases);
2863         if (!err) {
2864                 ceph_msg_get(req->r_request);
2865                 ceph_con_send(&session->s_con, req->r_request);
2866         }
2867
2868         return err;
2869 }
2870
2871 /*
2872  * send request, or put it on the appropriate wait list.
2873  */
2874 static void __do_request(struct ceph_mds_client *mdsc,
2875                         struct ceph_mds_request *req)
2876 {
2877         struct ceph_mds_session *session = NULL;
2878         int mds = -1;
2879         int err = 0;
2880         bool random;
2881
2882         if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2883                 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
2884                         __unregister_request(mdsc, req);
2885                 return;
2886         }
2887
2888         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_FENCE_IO) {
2889                 dout("do_request metadata corrupted\n");
2890                 err = -EIO;
2891                 goto finish;
2892         }
2893         if (req->r_timeout &&
2894             time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2895                 dout("do_request timed out\n");
2896                 err = -ETIMEDOUT;
2897                 goto finish;
2898         }
2899         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2900                 dout("do_request forced umount\n");
2901                 err = -EIO;
2902                 goto finish;
2903         }
2904         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2905                 if (mdsc->mdsmap_err) {
2906                         err = mdsc->mdsmap_err;
2907                         dout("do_request mdsmap err %d\n", err);
2908                         goto finish;
2909                 }
2910                 if (mdsc->mdsmap->m_epoch == 0) {
2911                         dout("do_request no mdsmap, waiting for map\n");
2912                         list_add(&req->r_wait, &mdsc->waiting_for_map);
2913                         return;
2914                 }
2915                 if (!(mdsc->fsc->mount_options->flags &
2916                       CEPH_MOUNT_OPT_MOUNTWAIT) &&
2917                     !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2918                         err = -EHOSTUNREACH;
2919                         goto finish;
2920                 }
2921         }
2922
2923         put_request_session(req);
2924
2925         mds = __choose_mds(mdsc, req, &random);
2926         if (mds < 0 ||
2927             ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2928                 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2929                         err = -EJUKEBOX;
2930                         goto finish;
2931                 }
2932                 dout("do_request no mds or not active, waiting for map\n");
2933                 list_add(&req->r_wait, &mdsc->waiting_for_map);
2934                 return;
2935         }
2936
2937         /* get, open session */
2938         session = __ceph_lookup_mds_session(mdsc, mds);
2939         if (!session) {
2940                 session = register_session(mdsc, mds);
2941                 if (IS_ERR(session)) {
2942                         err = PTR_ERR(session);
2943                         goto finish;
2944                 }
2945         }
2946         req->r_session = ceph_get_mds_session(session);
2947
2948         dout("do_request mds%d session %p state %s\n", mds, session,
2949              ceph_session_state_name(session->s_state));
2950
2951         /*
2952          * The old ceph will crash the MDSs when see unknown OPs
2953          */
2954         if (req->r_feature_needed > 0 &&
2955             !test_bit(req->r_feature_needed, &session->s_features)) {
2956                 err = -EOPNOTSUPP;
2957                 goto out_session;
2958         }
2959
2960         if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2961             session->s_state != CEPH_MDS_SESSION_HUNG) {
2962                 /*
2963                  * We cannot queue async requests since the caps and delegated
2964                  * inodes are bound to the session. Just return -EJUKEBOX and
2965                  * let the caller retry a sync request in that case.
2966                  */
2967                 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2968                         err = -EJUKEBOX;
2969                         goto out_session;
2970                 }
2971
2972                 /*
2973                  * If the session has been REJECTED, then return a hard error,
2974                  * unless it's a CLEANRECOVER mount, in which case we'll queue
2975                  * it to the mdsc queue.
2976                  */
2977                 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2978                         if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
2979                                 list_add(&req->r_wait, &mdsc->waiting_for_map);
2980                         else
2981                                 err = -EACCES;
2982                         goto out_session;
2983                 }
2984
2985                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2986                     session->s_state == CEPH_MDS_SESSION_CLOSING) {
2987                         err = __open_session(mdsc, session);
2988                         if (err)
2989                                 goto out_session;
2990                         /* retry the same mds later */
2991                         if (random)
2992                                 req->r_resend_mds = mds;
2993                 }
2994                 list_add(&req->r_wait, &session->s_waiting);
2995                 goto out_session;
2996         }
2997
2998         /* send request */
2999         req->r_resend_mds = -1;   /* forget any previous mds hint */
3000
3001         if (req->r_request_started == 0)   /* note request start time */
3002                 req->r_request_started = jiffies;
3003
3004         /*
3005          * For async create we will choose the auth MDS of frag in parent
3006          * directory to send the request and ususally this works fine, but
3007          * if the migrated the dirtory to another MDS before it could handle
3008          * it the request will be forwarded.
3009          *
3010          * And then the auth cap will be changed.
3011          */
3012         if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && req->r_num_fwd) {
3013                 struct ceph_dentry_info *di = ceph_dentry(req->r_dentry);
3014                 struct ceph_inode_info *ci;
3015                 struct ceph_cap *cap;
3016
3017                 /*
3018                  * The request maybe handled very fast and the new inode
3019                  * hasn't been linked to the dentry yet. We need to wait
3020                  * for the ceph_finish_async_create(), which shouldn't be
3021                  * stuck too long or fail in thoery, to finish when forwarding
3022                  * the request.
3023                  */
3024                 if (!d_inode(req->r_dentry)) {
3025                         err = wait_on_bit(&di->flags, CEPH_DENTRY_ASYNC_CREATE_BIT,
3026                                           TASK_KILLABLE);
3027                         if (err) {
3028                                 mutex_lock(&req->r_fill_mutex);
3029                                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3030                                 mutex_unlock(&req->r_fill_mutex);
3031                                 goto out_session;
3032                         }
3033                 }
3034
3035                 ci = ceph_inode(d_inode(req->r_dentry));
3036
3037                 spin_lock(&ci->i_ceph_lock);
3038                 cap = ci->i_auth_cap;
3039                 if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE && mds != cap->mds) {
3040                         dout("do_request session changed for auth cap %d -> %d\n",
3041                              cap->session->s_mds, session->s_mds);
3042
3043                         /* Remove the auth cap from old session */
3044                         spin_lock(&cap->session->s_cap_lock);
3045                         cap->session->s_nr_caps--;
3046                         list_del_init(&cap->session_caps);
3047                         spin_unlock(&cap->session->s_cap_lock);
3048
3049                         /* Add the auth cap to the new session */
3050                         cap->mds = mds;
3051                         cap->session = session;
3052                         spin_lock(&session->s_cap_lock);
3053                         session->s_nr_caps++;
3054                         list_add_tail(&cap->session_caps, &session->s_caps);
3055                         spin_unlock(&session->s_cap_lock);
3056
3057                         change_auth_cap_ses(ci, session);
3058                 }
3059                 spin_unlock(&ci->i_ceph_lock);
3060         }
3061
3062         err = __send_request(session, req, false);
3063
3064 out_session:
3065         ceph_put_mds_session(session);
3066 finish:
3067         if (err) {
3068                 dout("__do_request early error %d\n", err);
3069                 req->r_err = err;
3070                 complete_request(mdsc, req);
3071                 __unregister_request(mdsc, req);
3072         }
3073         return;
3074 }
3075
3076 /*
3077  * called under mdsc->mutex
3078  */
3079 static void __wake_requests(struct ceph_mds_client *mdsc,
3080                             struct list_head *head)
3081 {
3082         struct ceph_mds_request *req;
3083         LIST_HEAD(tmp_list);
3084
3085         list_splice_init(head, &tmp_list);
3086
3087         while (!list_empty(&tmp_list)) {
3088                 req = list_entry(tmp_list.next,
3089                                  struct ceph_mds_request, r_wait);
3090                 list_del_init(&req->r_wait);
3091                 dout(" wake request %p tid %llu\n", req, req->r_tid);
3092                 __do_request(mdsc, req);
3093         }
3094 }
3095
3096 /*
3097  * Wake up threads with requests pending for @mds, so that they can
3098  * resubmit their requests to a possibly different mds.
3099  */
3100 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
3101 {
3102         struct ceph_mds_request *req;
3103         struct rb_node *p = rb_first(&mdsc->request_tree);
3104
3105         dout("kick_requests mds%d\n", mds);
3106         while (p) {
3107                 req = rb_entry(p, struct ceph_mds_request, r_node);
3108                 p = rb_next(p);
3109                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3110                         continue;
3111                 if (req->r_attempts > 0)
3112                         continue; /* only new requests */
3113                 if (req->r_session &&
3114                     req->r_session->s_mds == mds) {
3115                         dout(" kicking tid %llu\n", req->r_tid);
3116                         list_del_init(&req->r_wait);
3117                         __do_request(mdsc, req);
3118                 }
3119         }
3120 }
3121
3122 int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
3123                               struct ceph_mds_request *req)
3124 {
3125         int err = 0;
3126
3127         /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
3128         if (req->r_inode)
3129                 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3130         if (req->r_parent) {
3131                 struct ceph_inode_info *ci = ceph_inode(req->r_parent);
3132                 int fmode = (req->r_op & CEPH_MDS_OP_WRITE) ?
3133                             CEPH_FILE_MODE_WR : CEPH_FILE_MODE_RD;
3134                 spin_lock(&ci->i_ceph_lock);
3135                 ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
3136                 __ceph_touch_fmode(ci, mdsc, fmode);
3137                 spin_unlock(&ci->i_ceph_lock);
3138         }
3139         if (req->r_old_dentry_dir)
3140                 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
3141                                   CEPH_CAP_PIN);
3142
3143         if (req->r_inode) {
3144                 err = ceph_wait_on_async_create(req->r_inode);
3145                 if (err) {
3146                         dout("%s: wait for async create returned: %d\n",
3147                              __func__, err);
3148                         return err;
3149                 }
3150         }
3151
3152         if (!err && req->r_old_inode) {
3153                 err = ceph_wait_on_async_create(req->r_old_inode);
3154                 if (err) {
3155                         dout("%s: wait for async create returned: %d\n",
3156                              __func__, err);
3157                         return err;
3158                 }
3159         }
3160
3161         dout("submit_request on %p for inode %p\n", req, dir);
3162         mutex_lock(&mdsc->mutex);
3163         __register_request(mdsc, req, dir);
3164         __do_request(mdsc, req);
3165         err = req->r_err;
3166         mutex_unlock(&mdsc->mutex);
3167         return err;
3168 }
3169
3170 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
3171                            struct ceph_mds_request *req,
3172                            ceph_mds_request_wait_callback_t wait_func)
3173 {
3174         int err;
3175
3176         /* wait */
3177         dout("do_request waiting\n");
3178         if (wait_func) {
3179                 err = wait_func(mdsc, req);
3180         } else {
3181                 long timeleft = wait_for_completion_killable_timeout(
3182                                         &req->r_completion,
3183                                         ceph_timeout_jiffies(req->r_timeout));
3184                 if (timeleft > 0)
3185                         err = 0;
3186                 else if (!timeleft)
3187                         err = -ETIMEDOUT;  /* timed out */
3188                 else
3189                         err = timeleft;  /* killed */
3190         }
3191         dout("do_request waited, got %d\n", err);
3192         mutex_lock(&mdsc->mutex);
3193
3194         /* only abort if we didn't race with a real reply */
3195         if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
3196                 err = le32_to_cpu(req->r_reply_info.head->result);
3197         } else if (err < 0) {
3198                 dout("aborted request %lld with %d\n", req->r_tid, err);
3199
3200                 /*
3201                  * ensure we aren't running concurrently with
3202                  * ceph_fill_trace or ceph_readdir_prepopulate, which
3203                  * rely on locks (dir mutex) held by our caller.
3204                  */
3205                 mutex_lock(&req->r_fill_mutex);
3206                 req->r_err = err;
3207                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3208                 mutex_unlock(&req->r_fill_mutex);
3209
3210                 if (req->r_parent &&
3211                     (req->r_op & CEPH_MDS_OP_WRITE))
3212                         ceph_invalidate_dir_request(req);
3213         } else {
3214                 err = req->r_err;
3215         }
3216
3217         mutex_unlock(&mdsc->mutex);
3218         return err;
3219 }
3220
3221 /*
3222  * Synchrously perform an mds request.  Take care of all of the
3223  * session setup, forwarding, retry details.
3224  */
3225 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
3226                          struct inode *dir,
3227                          struct ceph_mds_request *req)
3228 {
3229         int err;
3230
3231         dout("do_request on %p\n", req);
3232
3233         /* issue */
3234         err = ceph_mdsc_submit_request(mdsc, dir, req);
3235         if (!err)
3236                 err = ceph_mdsc_wait_request(mdsc, req, NULL);
3237         dout("do_request %p done, result %d\n", req, err);
3238         return err;
3239 }
3240
3241 /*
3242  * Invalidate dir's completeness, dentry lease state on an aborted MDS
3243  * namespace request.
3244  */
3245 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
3246 {
3247         struct inode *dir = req->r_parent;
3248         struct inode *old_dir = req->r_old_dentry_dir;
3249
3250         dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
3251
3252         ceph_dir_clear_complete(dir);
3253         if (old_dir)
3254                 ceph_dir_clear_complete(old_dir);
3255         if (req->r_dentry)
3256                 ceph_invalidate_dentry_lease(req->r_dentry);
3257         if (req->r_old_dentry)
3258                 ceph_invalidate_dentry_lease(req->r_old_dentry);
3259 }
3260
3261 /*
3262  * Handle mds reply.
3263  *
3264  * We take the session mutex and parse and process the reply immediately.
3265  * This preserves the logical ordering of replies, capabilities, etc., sent
3266  * by the MDS as they are applied to our local cache.
3267  */
3268 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
3269 {
3270         struct ceph_mds_client *mdsc = session->s_mdsc;
3271         struct ceph_mds_request *req;
3272         struct ceph_mds_reply_head *head = msg->front.iov_base;
3273         struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
3274         struct ceph_snap_realm *realm;
3275         u64 tid;
3276         int err, result;
3277         int mds = session->s_mds;
3278         bool close_sessions = false;
3279
3280         if (msg->front.iov_len < sizeof(*head)) {
3281                 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
3282                 ceph_msg_dump(msg);
3283                 return;
3284         }
3285
3286         /* get request, session */
3287         tid = le64_to_cpu(msg->hdr.tid);
3288         mutex_lock(&mdsc->mutex);
3289         req = lookup_get_request(mdsc, tid);
3290         if (!req) {
3291                 dout("handle_reply on unknown tid %llu\n", tid);
3292                 mutex_unlock(&mdsc->mutex);
3293                 return;
3294         }
3295         dout("handle_reply %p\n", req);
3296
3297         /* correct session? */
3298         if (req->r_session != session) {
3299                 pr_err("mdsc_handle_reply got %llu on session mds%d"
3300                        " not mds%d\n", tid, session->s_mds,
3301                        req->r_session ? req->r_session->s_mds : -1);
3302                 mutex_unlock(&mdsc->mutex);
3303                 goto out;
3304         }
3305
3306         /* dup? */
3307         if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
3308             (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
3309                 pr_warn("got a dup %s reply on %llu from mds%d\n",
3310                            head->safe ? "safe" : "unsafe", tid, mds);
3311                 mutex_unlock(&mdsc->mutex);
3312                 goto out;
3313         }
3314         if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
3315                 pr_warn("got unsafe after safe on %llu from mds%d\n",
3316                            tid, mds);
3317                 mutex_unlock(&mdsc->mutex);
3318                 goto out;
3319         }
3320
3321         result = le32_to_cpu(head->result);
3322
3323         if (head->safe) {
3324                 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
3325                 __unregister_request(mdsc, req);
3326
3327                 /* last request during umount? */
3328                 if (mdsc->stopping && !__get_oldest_req(mdsc))
3329                         complete_all(&mdsc->safe_umount_waiters);
3330
3331                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3332                         /*
3333                          * We already handled the unsafe response, now do the
3334                          * cleanup.  No need to examine the response; the MDS
3335                          * doesn't include any result info in the safe
3336                          * response.  And even if it did, there is nothing
3337                          * useful we could do with a revised return value.
3338                          */
3339                         dout("got safe reply %llu, mds%d\n", tid, mds);
3340
3341                         mutex_unlock(&mdsc->mutex);
3342                         goto out;
3343                 }
3344         } else {
3345                 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
3346                 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
3347         }
3348
3349         dout("handle_reply tid %lld result %d\n", tid, result);
3350         rinfo = &req->r_reply_info;
3351         if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
3352                 err = parse_reply_info(session, msg, rinfo, (u64)-1);
3353         else
3354                 err = parse_reply_info(session, msg, rinfo, session->s_con.peer_features);
3355         mutex_unlock(&mdsc->mutex);
3356
3357         /* Must find target inode outside of mutexes to avoid deadlocks */
3358         if ((err >= 0) && rinfo->head->is_target) {
3359                 struct inode *in;
3360                 struct ceph_vino tvino = {
3361                         .ino  = le64_to_cpu(rinfo->targeti.in->ino),
3362                         .snap = le64_to_cpu(rinfo->targeti.in->snapid)
3363                 };
3364
3365                 in = ceph_get_inode(mdsc->fsc->sb, tvino);
3366                 if (IS_ERR(in)) {
3367                         err = PTR_ERR(in);
3368                         mutex_lock(&session->s_mutex);
3369                         goto out_err;
3370                 }
3371                 req->r_target_inode = in;
3372         }
3373
3374         mutex_lock(&session->s_mutex);
3375         if (err < 0) {
3376                 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
3377                 ceph_msg_dump(msg);
3378                 goto out_err;
3379         }
3380
3381         /* snap trace */
3382         realm = NULL;
3383         if (rinfo->snapblob_len) {
3384                 down_write(&mdsc->snap_rwsem);
3385                 err = ceph_update_snap_trace(mdsc, rinfo->snapblob,
3386                                 rinfo->snapblob + rinfo->snapblob_len,
3387                                 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
3388                                 &realm);
3389                 if (err) {
3390                         up_write(&mdsc->snap_rwsem);
3391                         close_sessions = true;
3392                         if (err == -EIO)
3393                                 ceph_msg_dump(msg);
3394                         goto out_err;
3395                 }
3396                 downgrade_write(&mdsc->snap_rwsem);
3397         } else {
3398                 down_read(&mdsc->snap_rwsem);
3399         }
3400
3401         /* insert trace into our cache */
3402         mutex_lock(&req->r_fill_mutex);
3403         current->journal_info = req;
3404         err = ceph_fill_trace(mdsc->fsc->sb, req);
3405         if (err == 0) {
3406                 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
3407                                     req->r_op == CEPH_MDS_OP_LSSNAP))
3408                         ceph_readdir_prepopulate(req, req->r_session);
3409         }
3410         current->journal_info = NULL;
3411         mutex_unlock(&req->r_fill_mutex);
3412
3413         up_read(&mdsc->snap_rwsem);
3414         if (realm)
3415                 ceph_put_snap_realm(mdsc, realm);
3416
3417         if (err == 0) {
3418                 if (req->r_target_inode &&
3419                     test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3420                         struct ceph_inode_info *ci =
3421                                 ceph_inode(req->r_target_inode);
3422                         spin_lock(&ci->i_unsafe_lock);
3423                         list_add_tail(&req->r_unsafe_target_item,
3424                                       &ci->i_unsafe_iops);
3425                         spin_unlock(&ci->i_unsafe_lock);
3426                 }
3427
3428                 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
3429         }
3430 out_err:
3431         mutex_lock(&mdsc->mutex);
3432         if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3433                 if (err) {
3434                         req->r_err = err;
3435                 } else {
3436                         req->r_reply =  ceph_msg_get(msg);
3437                         set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
3438                 }
3439         } else {
3440                 dout("reply arrived after request %lld was aborted\n", tid);
3441         }
3442         mutex_unlock(&mdsc->mutex);
3443
3444         mutex_unlock(&session->s_mutex);
3445
3446         /* kick calling process */
3447         complete_request(mdsc, req);
3448
3449         ceph_update_metadata_metrics(&mdsc->metric, req->r_start_latency,
3450                                      req->r_end_latency, err);
3451 out:
3452         ceph_mdsc_put_request(req);
3453
3454         /* Defer closing the sessions after s_mutex lock being released */
3455         if (close_sessions)
3456                 ceph_mdsc_close_sessions(mdsc);
3457         return;
3458 }
3459
3460
3461
3462 /*
3463  * handle mds notification that our request has been forwarded.
3464  */
3465 static void handle_forward(struct ceph_mds_client *mdsc,
3466                            struct ceph_mds_session *session,
3467                            struct ceph_msg *msg)
3468 {
3469         struct ceph_mds_request *req;
3470         u64 tid = le64_to_cpu(msg->hdr.tid);
3471         u32 next_mds;
3472         u32 fwd_seq;
3473         int err = -EINVAL;
3474         void *p = msg->front.iov_base;
3475         void *end = p + msg->front.iov_len;
3476         bool aborted = false;
3477
3478         ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3479         next_mds = ceph_decode_32(&p);
3480         fwd_seq = ceph_decode_32(&p);
3481
3482         mutex_lock(&mdsc->mutex);
3483         req = lookup_get_request(mdsc, tid);
3484         if (!req) {
3485                 mutex_unlock(&mdsc->mutex);
3486                 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
3487                 return;  /* dup reply? */
3488         }
3489
3490         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3491                 dout("forward tid %llu aborted, unregistering\n", tid);
3492                 __unregister_request(mdsc, req);
3493         } else if (fwd_seq <= req->r_num_fwd) {
3494                 /*
3495                  * The type of 'num_fwd' in ceph 'MClientRequestForward'
3496                  * is 'int32_t', while in 'ceph_mds_request_head' the
3497                  * type is '__u8'. So in case the request bounces between
3498                  * MDSes exceeding 256 times, the client will get stuck.
3499                  *
3500                  * In this case it's ususally a bug in MDS and continue
3501                  * bouncing the request makes no sense.
3502                  *
3503                  * In future this could be fixed in ceph code, so avoid
3504                  * using the hardcode here.
3505                  */
3506                 int max = sizeof_field(struct ceph_mds_request_head, num_fwd);
3507                 max = 1 << (max * BITS_PER_BYTE);
3508                 if (req->r_num_fwd >= max) {
3509                         mutex_lock(&req->r_fill_mutex);
3510                         req->r_err = -EMULTIHOP;
3511                         set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3512                         mutex_unlock(&req->r_fill_mutex);
3513                         aborted = true;
3514                         pr_warn_ratelimited("forward tid %llu seq overflow\n",
3515                                             tid);
3516                 } else {
3517                         dout("forward tid %llu to mds%d - old seq %d <= %d\n",
3518                              tid, next_mds, req->r_num_fwd, fwd_seq);
3519                 }
3520         } else {
3521                 /* resend. forward race not possible; mds would drop */
3522                 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
3523                 BUG_ON(req->r_err);
3524                 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3525                 req->r_attempts = 0;
3526                 req->r_num_fwd = fwd_seq;
3527                 req->r_resend_mds = next_mds;
3528                 put_request_session(req);
3529                 __do_request(mdsc, req);
3530         }
3531         mutex_unlock(&mdsc->mutex);
3532
3533         /* kick calling process */
3534         if (aborted)
3535                 complete_request(mdsc, req);
3536         ceph_mdsc_put_request(req);
3537         return;
3538
3539 bad:
3540         pr_err("mdsc_handle_forward decode error err=%d\n", err);
3541 }
3542
3543 static int __decode_session_metadata(void **p, void *end,
3544                                      bool *blocklisted)
3545 {
3546         /* map<string,string> */
3547         u32 n;
3548         bool err_str;
3549         ceph_decode_32_safe(p, end, n, bad);
3550         while (n-- > 0) {
3551                 u32 len;
3552                 ceph_decode_32_safe(p, end, len, bad);
3553                 ceph_decode_need(p, end, len, bad);
3554                 err_str = !strncmp(*p, "error_string", len);
3555                 *p += len;
3556                 ceph_decode_32_safe(p, end, len, bad);
3557                 ceph_decode_need(p, end, len, bad);
3558                 /*
3559                  * Match "blocklisted (blacklisted)" from newer MDSes,
3560                  * or "blacklisted" from older MDSes.
3561                  */
3562                 if (err_str && strnstr(*p, "blacklisted", len))
3563                         *blocklisted = true;
3564                 *p += len;
3565         }
3566         return 0;
3567 bad:
3568         return -1;
3569 }
3570
3571 /*
3572  * handle a mds session control message
3573  */
3574 static void handle_session(struct ceph_mds_session *session,
3575                            struct ceph_msg *msg)
3576 {
3577         struct ceph_mds_client *mdsc = session->s_mdsc;
3578         int mds = session->s_mds;
3579         int msg_version = le16_to_cpu(msg->hdr.version);
3580         void *p = msg->front.iov_base;
3581         void *end = p + msg->front.iov_len;
3582         struct ceph_mds_session_head *h;
3583         u32 op;
3584         u64 seq, features = 0;
3585         int wake = 0;
3586         bool blocklisted = false;
3587
3588         /* decode */
3589         ceph_decode_need(&p, end, sizeof(*h), bad);
3590         h = p;
3591         p += sizeof(*h);
3592
3593         op = le32_to_cpu(h->op);
3594         seq = le64_to_cpu(h->seq);
3595
3596         if (msg_version >= 3) {
3597                 u32 len;
3598                 /* version >= 2 and < 5, decode metadata, skip otherwise
3599                  * as it's handled via flags.
3600                  */
3601                 if (msg_version >= 5)
3602                         ceph_decode_skip_map(&p, end, string, string, bad);
3603                 else if (__decode_session_metadata(&p, end, &blocklisted) < 0)
3604                         goto bad;
3605
3606                 /* version >= 3, feature bits */
3607                 ceph_decode_32_safe(&p, end, len, bad);
3608                 if (len) {
3609                         ceph_decode_64_safe(&p, end, features, bad);
3610                         p += len - sizeof(features);
3611                 }
3612         }
3613
3614         if (msg_version >= 5) {
3615                 u32 flags, len;
3616
3617                 /* version >= 4 */
3618                 ceph_decode_skip_16(&p, end, bad); /* struct_v, struct_cv */
3619                 ceph_decode_32_safe(&p, end, len, bad); /* len */
3620                 ceph_decode_skip_n(&p, end, len, bad); /* metric_spec */
3621
3622                 /* version >= 5, flags   */
3623                 ceph_decode_32_safe(&p, end, flags, bad);
3624                 if (flags & CEPH_SESSION_BLOCKLISTED) {
3625                         pr_warn("mds%d session blocklisted\n", session->s_mds);
3626                         blocklisted = true;
3627                 }
3628         }
3629
3630         mutex_lock(&mdsc->mutex);
3631         if (op == CEPH_SESSION_CLOSE) {
3632                 ceph_get_mds_session(session);
3633                 __unregister_session(mdsc, session);
3634         }
3635         /* FIXME: this ttl calculation is generous */
3636         session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
3637         mutex_unlock(&mdsc->mutex);
3638
3639         mutex_lock(&session->s_mutex);
3640
3641         dout("handle_session mds%d %s %p state %s seq %llu\n",
3642              mds, ceph_session_op_name(op), session,
3643              ceph_session_state_name(session->s_state), seq);
3644
3645         if (session->s_state == CEPH_MDS_SESSION_HUNG) {
3646                 session->s_state = CEPH_MDS_SESSION_OPEN;
3647                 pr_info("mds%d came back\n", session->s_mds);
3648         }
3649
3650         switch (op) {
3651         case CEPH_SESSION_OPEN:
3652                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3653                         pr_info("mds%d reconnect success\n", session->s_mds);
3654
3655                 if (session->s_state == CEPH_MDS_SESSION_OPEN) {
3656                         pr_notice("mds%d is already opened\n", session->s_mds);
3657                 } else {
3658                         session->s_state = CEPH_MDS_SESSION_OPEN;
3659                         session->s_features = features;
3660                         renewed_caps(mdsc, session, 0);
3661                         if (test_bit(CEPHFS_FEATURE_METRIC_COLLECT,
3662                                      &session->s_features))
3663                                 metric_schedule_delayed(&mdsc->metric);
3664                 }
3665
3666                 /*
3667                  * The connection maybe broken and the session in client
3668                  * side has been reinitialized, need to update the seq
3669                  * anyway.
3670                  */
3671                 if (!session->s_seq && seq)
3672                         session->s_seq = seq;
3673
3674                 wake = 1;
3675                 if (mdsc->stopping)
3676                         __close_session(mdsc, session);
3677                 break;
3678
3679         case CEPH_SESSION_RENEWCAPS:
3680                 if (session->s_renew_seq == seq)
3681                         renewed_caps(mdsc, session, 1);
3682                 break;
3683
3684         case CEPH_SESSION_CLOSE:
3685                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3686                         pr_info("mds%d reconnect denied\n", session->s_mds);
3687                 session->s_state = CEPH_MDS_SESSION_CLOSED;
3688                 cleanup_session_requests(mdsc, session);
3689                 remove_session_caps(session);
3690                 wake = 2; /* for good measure */
3691                 wake_up_all(&mdsc->session_close_wq);
3692                 break;
3693
3694         case CEPH_SESSION_STALE:
3695                 pr_info("mds%d caps went stale, renewing\n",
3696                         session->s_mds);
3697                 atomic_inc(&session->s_cap_gen);
3698                 session->s_cap_ttl = jiffies - 1;
3699                 send_renew_caps(mdsc, session);
3700                 break;
3701
3702         case CEPH_SESSION_RECALL_STATE:
3703                 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
3704                 break;
3705
3706         case CEPH_SESSION_FLUSHMSG:
3707                 /* flush cap releases */
3708                 spin_lock(&session->s_cap_lock);
3709                 if (session->s_num_cap_releases)
3710                         ceph_flush_cap_releases(mdsc, session);
3711                 spin_unlock(&session->s_cap_lock);
3712
3713                 send_flushmsg_ack(mdsc, session, seq);
3714                 break;
3715
3716         case CEPH_SESSION_FORCE_RO:
3717                 dout("force_session_readonly %p\n", session);
3718                 spin_lock(&session->s_cap_lock);
3719                 session->s_readonly = true;
3720                 spin_unlock(&session->s_cap_lock);
3721                 wake_up_session_caps(session, FORCE_RO);
3722                 break;
3723
3724         case CEPH_SESSION_REJECT:
3725                 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
3726                 pr_info("mds%d rejected session\n", session->s_mds);
3727                 session->s_state = CEPH_MDS_SESSION_REJECTED;
3728                 cleanup_session_requests(mdsc, session);
3729                 remove_session_caps(session);
3730                 if (blocklisted)
3731                         mdsc->fsc->blocklisted = true;
3732                 wake = 2; /* for good measure */
3733                 break;
3734
3735         default:
3736                 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
3737                 WARN_ON(1);
3738         }
3739
3740         mutex_unlock(&session->s_mutex);
3741         if (wake) {
3742                 mutex_lock(&mdsc->mutex);
3743                 __wake_requests(mdsc, &session->s_waiting);
3744                 if (wake == 2)
3745                         kick_requests(mdsc, mds);
3746                 mutex_unlock(&mdsc->mutex);
3747         }
3748         if (op == CEPH_SESSION_CLOSE)
3749                 ceph_put_mds_session(session);
3750         return;
3751
3752 bad:
3753         pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
3754                (int)msg->front.iov_len);
3755         ceph_msg_dump(msg);
3756         return;
3757 }
3758
3759 void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
3760 {
3761         int dcaps;
3762
3763         dcaps = xchg(&req->r_dir_caps, 0);
3764         if (dcaps) {
3765                 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3766                 ceph_put_cap_refs(ceph_inode(req->r_parent), dcaps);
3767         }
3768 }
3769
3770 void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req)
3771 {
3772         int dcaps;
3773
3774         dcaps = xchg(&req->r_dir_caps, 0);
3775         if (dcaps) {
3776                 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3777                 ceph_put_cap_refs_no_check_caps(ceph_inode(req->r_parent),
3778                                                 dcaps);
3779         }
3780 }
3781
3782 /*
3783  * called under session->mutex.
3784  */
3785 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
3786                                    struct ceph_mds_session *session)
3787 {
3788         struct ceph_mds_request *req, *nreq;
3789         struct rb_node *p;
3790
3791         dout("replay_unsafe_requests mds%d\n", session->s_mds);
3792
3793         mutex_lock(&mdsc->mutex);
3794         list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item)
3795                 __send_request(session, req, true);
3796
3797         /*
3798          * also re-send old requests when MDS enters reconnect stage. So that MDS
3799          * can process completed request in clientreplay stage.
3800          */
3801         p = rb_first(&mdsc->request_tree);
3802         while (p) {
3803                 req = rb_entry(p, struct ceph_mds_request, r_node);
3804                 p = rb_next(p);
3805                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3806                         continue;
3807                 if (req->r_attempts == 0)
3808                         continue; /* only old requests */
3809                 if (!req->r_session)
3810                         continue;
3811                 if (req->r_session->s_mds != session->s_mds)
3812                         continue;
3813
3814                 ceph_mdsc_release_dir_caps_no_check(req);
3815
3816                 __send_request(session, req, true);
3817         }
3818         mutex_unlock(&mdsc->mutex);
3819 }
3820
3821 static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
3822 {
3823         struct ceph_msg *reply;
3824         struct ceph_pagelist *_pagelist;
3825         struct page *page;
3826         __le32 *addr;
3827         int err = -ENOMEM;
3828
3829         if (!recon_state->allow_multi)
3830                 return -ENOSPC;
3831
3832         /* can't handle message that contains both caps and realm */
3833         BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
3834
3835         /* pre-allocate new pagelist */
3836         _pagelist = ceph_pagelist_alloc(GFP_NOFS);
3837         if (!_pagelist)
3838                 return -ENOMEM;
3839
3840         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3841         if (!reply)
3842                 goto fail_msg;
3843
3844         /* placeholder for nr_caps */
3845         err = ceph_pagelist_encode_32(_pagelist, 0);
3846         if (err < 0)
3847                 goto fail;
3848
3849         if (recon_state->nr_caps) {
3850                 /* currently encoding caps */
3851                 err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
3852                 if (err)
3853                         goto fail;
3854         } else {
3855                 /* placeholder for nr_realms (currently encoding relams) */
3856                 err = ceph_pagelist_encode_32(_pagelist, 0);
3857                 if (err < 0)
3858                         goto fail;
3859         }
3860
3861         err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
3862         if (err)
3863                 goto fail;
3864
3865         page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
3866         addr = kmap_atomic(page);
3867         if (recon_state->nr_caps) {
3868                 /* currently encoding caps */
3869                 *addr = cpu_to_le32(recon_state->nr_caps);
3870         } else {
3871                 /* currently encoding relams */
3872                 *(addr + 1) = cpu_to_le32(recon_state->nr_realms);
3873         }
3874         kunmap_atomic(addr);
3875
3876         reply->hdr.version = cpu_to_le16(5);
3877         reply->hdr.compat_version = cpu_to_le16(4);
3878
3879         reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
3880         ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
3881
3882         ceph_con_send(&recon_state->session->s_con, reply);
3883         ceph_pagelist_release(recon_state->pagelist);
3884
3885         recon_state->pagelist = _pagelist;
3886         recon_state->nr_caps = 0;
3887         recon_state->nr_realms = 0;
3888         recon_state->msg_version = 5;
3889         return 0;
3890 fail:
3891         ceph_msg_put(reply);
3892 fail_msg:
3893         ceph_pagelist_release(_pagelist);
3894         return err;
3895 }
3896
3897 static struct dentry* d_find_primary(struct inode *inode)
3898 {
3899         struct dentry *alias, *dn = NULL;
3900
3901         if (hlist_empty(&inode->i_dentry))
3902                 return NULL;
3903
3904         spin_lock(&inode->i_lock);
3905         if (hlist_empty(&inode->i_dentry))
3906                 goto out_unlock;
3907
3908         if (S_ISDIR(inode->i_mode)) {
3909                 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
3910                 if (!IS_ROOT(alias))
3911                         dn = dget(alias);
3912                 goto out_unlock;
3913         }
3914
3915         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
3916                 spin_lock(&alias->d_lock);
3917                 if (!d_unhashed(alias) &&
3918                     (ceph_dentry(alias)->flags & CEPH_DENTRY_PRIMARY_LINK)) {
3919                         dn = dget_dlock(alias);
3920                 }
3921                 spin_unlock(&alias->d_lock);
3922                 if (dn)
3923                         break;
3924         }
3925 out_unlock:
3926         spin_unlock(&inode->i_lock);
3927         return dn;
3928 }
3929
3930 /*
3931  * Encode information about a cap for a reconnect with the MDS.
3932  */
3933 static int reconnect_caps_cb(struct inode *inode, int mds, void *arg)
3934 {
3935         union {
3936                 struct ceph_mds_cap_reconnect v2;
3937                 struct ceph_mds_cap_reconnect_v1 v1;
3938         } rec;
3939         struct ceph_inode_info *ci = ceph_inode(inode);
3940         struct ceph_reconnect_state *recon_state = arg;
3941         struct ceph_pagelist *pagelist = recon_state->pagelist;
3942         struct dentry *dentry;
3943         struct ceph_cap *cap;
3944         char *path;
3945         int pathlen = 0, err;
3946         u64 pathbase;
3947         u64 snap_follows;
3948
3949         dentry = d_find_primary(inode);
3950         if (dentry) {
3951                 /* set pathbase to parent dir when msg_version >= 2 */
3952                 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase,
3953                                             recon_state->msg_version >= 2);
3954                 dput(dentry);
3955                 if (IS_ERR(path)) {
3956                         err = PTR_ERR(path);
3957                         goto out_err;
3958                 }
3959         } else {
3960                 path = NULL;
3961                 pathbase = 0;
3962         }
3963
3964         spin_lock(&ci->i_ceph_lock);
3965         cap = __get_cap_for_mds(ci, mds);
3966         if (!cap) {
3967                 spin_unlock(&ci->i_ceph_lock);
3968                 err = 0;
3969                 goto out_err;
3970         }
3971         dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
3972              inode, ceph_vinop(inode), cap, cap->cap_id,
3973              ceph_cap_string(cap->issued));
3974
3975         cap->seq = 0;        /* reset cap seq */
3976         cap->issue_seq = 0;  /* and issue_seq */
3977         cap->mseq = 0;       /* and migrate_seq */
3978         cap->cap_gen = atomic_read(&cap->session->s_cap_gen);
3979
3980         /* These are lost when the session goes away */
3981         if (S_ISDIR(inode->i_mode)) {
3982                 if (cap->issued & CEPH_CAP_DIR_CREATE) {
3983                         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
3984                         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
3985                 }
3986                 cap->issued &= ~CEPH_CAP_ANY_DIR_OPS;
3987         }
3988
3989         if (recon_state->msg_version >= 2) {
3990                 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
3991                 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3992                 rec.v2.issued = cpu_to_le32(cap->issued);
3993                 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3994                 rec.v2.pathbase = cpu_to_le64(pathbase);
3995                 rec.v2.flock_len = (__force __le32)
3996                         ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
3997         } else {
3998                 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
3999                 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
4000                 rec.v1.issued = cpu_to_le32(cap->issued);
4001                 rec.v1.size = cpu_to_le64(i_size_read(inode));
4002                 ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
4003                 ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
4004                 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
4005                 rec.v1.pathbase = cpu_to_le64(pathbase);
4006         }
4007
4008         if (list_empty(&ci->i_cap_snaps)) {
4009                 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
4010         } else {
4011                 struct ceph_cap_snap *capsnap =
4012                         list_first_entry(&ci->i_cap_snaps,
4013                                          struct ceph_cap_snap, ci_item);
4014                 snap_follows = capsnap->follows;
4015         }
4016         spin_unlock(&ci->i_ceph_lock);
4017
4018         if (recon_state->msg_version >= 2) {
4019                 int num_fcntl_locks, num_flock_locks;
4020                 struct ceph_filelock *flocks = NULL;
4021                 size_t struct_len, total_len = sizeof(u64);
4022                 u8 struct_v = 0;
4023
4024 encode_again:
4025                 if (rec.v2.flock_len) {
4026                         ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
4027                 } else {
4028                         num_fcntl_locks = 0;
4029                         num_flock_locks = 0;
4030                 }
4031                 if (num_fcntl_locks + num_flock_locks > 0) {
4032                         flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
4033                                                sizeof(struct ceph_filelock),
4034                                                GFP_NOFS);
4035                         if (!flocks) {
4036                                 err = -ENOMEM;
4037                                 goto out_err;
4038                         }
4039                         err = ceph_encode_locks_to_buffer(inode, flocks,
4040                                                           num_fcntl_locks,
4041                                                           num_flock_locks);
4042                         if (err) {
4043                                 kfree(flocks);
4044                                 flocks = NULL;
4045                                 if (err == -ENOSPC)
4046                                         goto encode_again;
4047                                 goto out_err;
4048                         }
4049                 } else {
4050                         kfree(flocks);
4051                         flocks = NULL;
4052                 }
4053
4054                 if (recon_state->msg_version >= 3) {
4055                         /* version, compat_version and struct_len */
4056                         total_len += 2 * sizeof(u8) + sizeof(u32);
4057                         struct_v = 2;
4058                 }
4059                 /*
4060                  * number of encoded locks is stable, so copy to pagelist
4061                  */
4062                 struct_len = 2 * sizeof(u32) +
4063                             (num_fcntl_locks + num_flock_locks) *
4064                             sizeof(struct ceph_filelock);
4065                 rec.v2.flock_len = cpu_to_le32(struct_len);
4066
4067                 struct_len += sizeof(u32) + pathlen + sizeof(rec.v2);
4068
4069                 if (struct_v >= 2)
4070                         struct_len += sizeof(u64); /* snap_follows */
4071
4072                 total_len += struct_len;
4073
4074                 if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
4075                         err = send_reconnect_partial(recon_state);
4076                         if (err)
4077                                 goto out_freeflocks;
4078                         pagelist = recon_state->pagelist;
4079                 }
4080
4081                 err = ceph_pagelist_reserve(pagelist, total_len);
4082                 if (err)
4083                         goto out_freeflocks;
4084
4085                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
4086                 if (recon_state->msg_version >= 3) {
4087                         ceph_pagelist_encode_8(pagelist, struct_v);
4088                         ceph_pagelist_encode_8(pagelist, 1);
4089                         ceph_pagelist_encode_32(pagelist, struct_len);
4090                 }
4091                 ceph_pagelist_encode_string(pagelist, path, pathlen);
4092                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
4093                 ceph_locks_to_pagelist(flocks, pagelist,
4094                                        num_fcntl_locks, num_flock_locks);
4095                 if (struct_v >= 2)
4096                         ceph_pagelist_encode_64(pagelist, snap_follows);
4097 out_freeflocks:
4098                 kfree(flocks);
4099         } else {
4100                 err = ceph_pagelist_reserve(pagelist,
4101                                             sizeof(u64) + sizeof(u32) +
4102                                             pathlen + sizeof(rec.v1));
4103                 if (err)
4104                         goto out_err;
4105
4106                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
4107                 ceph_pagelist_encode_string(pagelist, path, pathlen);
4108                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
4109         }
4110
4111 out_err:
4112         ceph_mdsc_free_path(path, pathlen);
4113         if (!err)
4114                 recon_state->nr_caps++;
4115         return err;
4116 }
4117
4118 static int encode_snap_realms(struct ceph_mds_client *mdsc,
4119                               struct ceph_reconnect_state *recon_state)
4120 {
4121         struct rb_node *p;
4122         struct ceph_pagelist *pagelist = recon_state->pagelist;
4123         int err = 0;
4124
4125         if (recon_state->msg_version >= 4) {
4126                 err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
4127                 if (err < 0)
4128                         goto fail;
4129         }
4130
4131         /*
4132          * snaprealms.  we provide mds with the ino, seq (version), and
4133          * parent for all of our realms.  If the mds has any newer info,
4134          * it will tell us.
4135          */
4136         for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
4137                 struct ceph_snap_realm *realm =
4138                        rb_entry(p, struct ceph_snap_realm, node);
4139                 struct ceph_mds_snaprealm_reconnect sr_rec;
4140
4141                 if (recon_state->msg_version >= 4) {
4142                         size_t need = sizeof(u8) * 2 + sizeof(u32) +
4143                                       sizeof(sr_rec);
4144
4145                         if (pagelist->length + need > RECONNECT_MAX_SIZE) {
4146                                 err = send_reconnect_partial(recon_state);
4147                                 if (err)
4148                                         goto fail;
4149                                 pagelist = recon_state->pagelist;
4150                         }
4151
4152                         err = ceph_pagelist_reserve(pagelist, need);
4153                         if (err)
4154                                 goto fail;
4155
4156                         ceph_pagelist_encode_8(pagelist, 1);
4157                         ceph_pagelist_encode_8(pagelist, 1);
4158                         ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
4159                 }
4160
4161                 dout(" adding snap realm %llx seq %lld parent %llx\n",
4162                      realm->ino, realm->seq, realm->parent_ino);
4163                 sr_rec.ino = cpu_to_le64(realm->ino);
4164                 sr_rec.seq = cpu_to_le64(realm->seq);
4165                 sr_rec.parent = cpu_to_le64(realm->parent_ino);
4166
4167                 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
4168                 if (err)
4169                         goto fail;
4170
4171                 recon_state->nr_realms++;
4172         }
4173 fail:
4174         return err;
4175 }
4176
4177
4178 /*
4179  * If an MDS fails and recovers, clients need to reconnect in order to
4180  * reestablish shared state.  This includes all caps issued through
4181  * this session _and_ the snap_realm hierarchy.  Because it's not
4182  * clear which snap realms the mds cares about, we send everything we
4183  * know about.. that ensures we'll then get any new info the
4184  * recovering MDS might have.
4185  *
4186  * This is a relatively heavyweight operation, but it's rare.
4187  */
4188 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
4189                                struct ceph_mds_session *session)
4190 {
4191         struct ceph_msg *reply;
4192         int mds = session->s_mds;
4193         int err = -ENOMEM;
4194         struct ceph_reconnect_state recon_state = {
4195                 .session = session,
4196         };
4197         LIST_HEAD(dispose);
4198
4199         pr_info("mds%d reconnect start\n", mds);
4200
4201         recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
4202         if (!recon_state.pagelist)
4203                 goto fail_nopagelist;
4204
4205         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
4206         if (!reply)
4207                 goto fail_nomsg;
4208
4209         xa_destroy(&session->s_delegated_inos);
4210
4211         mutex_lock(&session->s_mutex);
4212         session->s_state = CEPH_MDS_SESSION_RECONNECTING;
4213         session->s_seq = 0;
4214
4215         dout("session %p state %s\n", session,
4216              ceph_session_state_name(session->s_state));
4217
4218         atomic_inc(&session->s_cap_gen);
4219
4220         spin_lock(&session->s_cap_lock);
4221         /* don't know if session is readonly */
4222         session->s_readonly = 0;
4223         /*
4224          * notify __ceph_remove_cap() that we are composing cap reconnect.
4225          * If a cap get released before being added to the cap reconnect,
4226          * __ceph_remove_cap() should skip queuing cap release.
4227          */
4228         session->s_cap_reconnect = 1;
4229         /* drop old cap expires; we're about to reestablish that state */
4230         detach_cap_releases(session, &dispose);
4231         spin_unlock(&session->s_cap_lock);
4232         dispose_cap_releases(mdsc, &dispose);
4233
4234         /* trim unused caps to reduce MDS's cache rejoin time */
4235         if (mdsc->fsc->sb->s_root)
4236                 shrink_dcache_parent(mdsc->fsc->sb->s_root);
4237
4238         ceph_con_close(&session->s_con);
4239         ceph_con_open(&session->s_con,
4240                       CEPH_ENTITY_TYPE_MDS, mds,
4241                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
4242
4243         /* replay unsafe requests */
4244         replay_unsafe_requests(mdsc, session);
4245
4246         ceph_early_kick_flushing_caps(mdsc, session);
4247
4248         down_read(&mdsc->snap_rwsem);
4249
4250         /* placeholder for nr_caps */
4251         err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
4252         if (err)
4253                 goto fail;
4254
4255         if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
4256                 recon_state.msg_version = 3;
4257                 recon_state.allow_multi = true;
4258         } else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
4259                 recon_state.msg_version = 3;
4260         } else {
4261                 recon_state.msg_version = 2;
4262         }
4263         /* trsaverse this session's caps */
4264         err = ceph_iterate_session_caps(session, reconnect_caps_cb, &recon_state);
4265
4266         spin_lock(&session->s_cap_lock);
4267         session->s_cap_reconnect = 0;
4268         spin_unlock(&session->s_cap_lock);
4269
4270         if (err < 0)
4271                 goto fail;
4272
4273         /* check if all realms can be encoded into current message */
4274         if (mdsc->num_snap_realms) {
4275                 size_t total_len =
4276                         recon_state.pagelist->length +
4277                         mdsc->num_snap_realms *
4278                         sizeof(struct ceph_mds_snaprealm_reconnect);
4279                 if (recon_state.msg_version >= 4) {
4280                         /* number of realms */
4281                         total_len += sizeof(u32);
4282                         /* version, compat_version and struct_len */
4283                         total_len += mdsc->num_snap_realms *
4284                                      (2 * sizeof(u8) + sizeof(u32));
4285                 }
4286                 if (total_len > RECONNECT_MAX_SIZE) {
4287                         if (!recon_state.allow_multi) {
4288                                 err = -ENOSPC;
4289                                 goto fail;
4290                         }
4291                         if (recon_state.nr_caps) {
4292                                 err = send_reconnect_partial(&recon_state);
4293                                 if (err)
4294                                         goto fail;
4295                         }
4296                         recon_state.msg_version = 5;
4297                 }
4298         }
4299
4300         err = encode_snap_realms(mdsc, &recon_state);
4301         if (err < 0)
4302                 goto fail;
4303
4304         if (recon_state.msg_version >= 5) {
4305                 err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
4306                 if (err < 0)
4307                         goto fail;
4308         }
4309
4310         if (recon_state.nr_caps || recon_state.nr_realms) {
4311                 struct page *page =
4312                         list_first_entry(&recon_state.pagelist->head,
4313                                         struct page, lru);
4314                 __le32 *addr = kmap_atomic(page);
4315                 if (recon_state.nr_caps) {
4316                         WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
4317                         *addr = cpu_to_le32(recon_state.nr_caps);
4318                 } else if (recon_state.msg_version >= 4) {
4319                         *(addr + 1) = cpu_to_le32(recon_state.nr_realms);
4320                 }
4321                 kunmap_atomic(addr);
4322         }
4323
4324         reply->hdr.version = cpu_to_le16(recon_state.msg_version);
4325         if (recon_state.msg_version >= 4)
4326                 reply->hdr.compat_version = cpu_to_le16(4);
4327
4328         reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
4329         ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
4330
4331         ceph_con_send(&session->s_con, reply);
4332
4333         mutex_unlock(&session->s_mutex);
4334
4335         mutex_lock(&mdsc->mutex);
4336         __wake_requests(mdsc, &session->s_waiting);
4337         mutex_unlock(&mdsc->mutex);
4338
4339         up_read(&mdsc->snap_rwsem);
4340         ceph_pagelist_release(recon_state.pagelist);
4341         return;
4342
4343 fail:
4344         ceph_msg_put(reply);
4345         up_read(&mdsc->snap_rwsem);
4346         mutex_unlock(&session->s_mutex);
4347 fail_nomsg:
4348         ceph_pagelist_release(recon_state.pagelist);
4349 fail_nopagelist:
4350         pr_err("error %d preparing reconnect for mds%d\n", err, mds);
4351         return;
4352 }
4353
4354
4355 /*
4356  * compare old and new mdsmaps, kicking requests
4357  * and closing out old connections as necessary
4358  *
4359  * called under mdsc->mutex.
4360  */
4361 static void check_new_map(struct ceph_mds_client *mdsc,
4362                           struct ceph_mdsmap *newmap,
4363                           struct ceph_mdsmap *oldmap)
4364 {
4365         int i, j, err;
4366         int oldstate, newstate;
4367         struct ceph_mds_session *s;
4368         unsigned long targets[DIV_ROUND_UP(CEPH_MAX_MDS, sizeof(unsigned long))] = {0};
4369
4370         dout("check_new_map new %u old %u\n",
4371              newmap->m_epoch, oldmap->m_epoch);
4372
4373         if (newmap->m_info) {
4374                 for (i = 0; i < newmap->possible_max_rank; i++) {
4375                         for (j = 0; j < newmap->m_info[i].num_export_targets; j++)
4376                                 set_bit(newmap->m_info[i].export_targets[j], targets);
4377                 }
4378         }
4379
4380         for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4381                 if (!mdsc->sessions[i])
4382                         continue;
4383                 s = mdsc->sessions[i];
4384                 oldstate = ceph_mdsmap_get_state(oldmap, i);
4385                 newstate = ceph_mdsmap_get_state(newmap, i);
4386
4387                 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
4388                      i, ceph_mds_state_name(oldstate),
4389                      ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
4390                      ceph_mds_state_name(newstate),
4391                      ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
4392                      ceph_session_state_name(s->s_state));
4393
4394                 if (i >= newmap->possible_max_rank) {
4395                         /* force close session for stopped mds */
4396                         ceph_get_mds_session(s);
4397                         __unregister_session(mdsc, s);
4398                         __wake_requests(mdsc, &s->s_waiting);
4399                         mutex_unlock(&mdsc->mutex);
4400
4401                         mutex_lock(&s->s_mutex);
4402                         cleanup_session_requests(mdsc, s);
4403                         remove_session_caps(s);
4404                         mutex_unlock(&s->s_mutex);
4405
4406                         ceph_put_mds_session(s);
4407
4408                         mutex_lock(&mdsc->mutex);
4409                         kick_requests(mdsc, i);
4410                         continue;
4411                 }
4412
4413                 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
4414                            ceph_mdsmap_get_addr(newmap, i),
4415                            sizeof(struct ceph_entity_addr))) {
4416                         /* just close it */
4417                         mutex_unlock(&mdsc->mutex);
4418                         mutex_lock(&s->s_mutex);
4419                         mutex_lock(&mdsc->mutex);
4420                         ceph_con_close(&s->s_con);
4421                         mutex_unlock(&s->s_mutex);
4422                         s->s_state = CEPH_MDS_SESSION_RESTARTING;
4423                 } else if (oldstate == newstate) {
4424                         continue;  /* nothing new with this mds */
4425                 }
4426
4427                 /*
4428                  * send reconnect?
4429                  */
4430                 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
4431                     newstate >= CEPH_MDS_STATE_RECONNECT) {
4432                         mutex_unlock(&mdsc->mutex);
4433                         clear_bit(i, targets);
4434                         send_mds_reconnect(mdsc, s);
4435                         mutex_lock(&mdsc->mutex);
4436                 }
4437
4438                 /*
4439                  * kick request on any mds that has gone active.
4440                  */
4441                 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
4442                     newstate >= CEPH_MDS_STATE_ACTIVE) {
4443                         if (oldstate != CEPH_MDS_STATE_CREATING &&
4444                             oldstate != CEPH_MDS_STATE_STARTING)
4445                                 pr_info("mds%d recovery completed\n", s->s_mds);
4446                         kick_requests(mdsc, i);
4447                         mutex_unlock(&mdsc->mutex);
4448                         mutex_lock(&s->s_mutex);
4449                         mutex_lock(&mdsc->mutex);
4450                         ceph_kick_flushing_caps(mdsc, s);
4451                         mutex_unlock(&s->s_mutex);
4452                         wake_up_session_caps(s, RECONNECT);
4453                 }
4454         }
4455
4456         /*
4457          * Only open and reconnect sessions that don't exist yet.
4458          */
4459         for (i = 0; i < newmap->possible_max_rank; i++) {
4460                 /*
4461                  * In case the import MDS is crashed just after
4462                  * the EImportStart journal is flushed, so when
4463                  * a standby MDS takes over it and is replaying
4464                  * the EImportStart journal the new MDS daemon
4465                  * will wait the client to reconnect it, but the
4466                  * client may never register/open the session yet.
4467                  *
4468                  * Will try to reconnect that MDS daemon if the
4469                  * rank number is in the export targets array and
4470                  * is the up:reconnect state.
4471                  */
4472                 newstate = ceph_mdsmap_get_state(newmap, i);
4473                 if (!test_bit(i, targets) || newstate != CEPH_MDS_STATE_RECONNECT)
4474                         continue;
4475
4476                 /*
4477                  * The session maybe registered and opened by some
4478                  * requests which were choosing random MDSes during
4479                  * the mdsc->mutex's unlock/lock gap below in rare
4480                  * case. But the related MDS daemon will just queue
4481                  * that requests and be still waiting for the client's
4482                  * reconnection request in up:reconnect state.
4483                  */
4484                 s = __ceph_lookup_mds_session(mdsc, i);
4485                 if (likely(!s)) {
4486                         s = __open_export_target_session(mdsc, i);
4487                         if (IS_ERR(s)) {
4488                                 err = PTR_ERR(s);
4489                                 pr_err("failed to open export target session, err %d\n",
4490                                        err);
4491                                 continue;
4492                         }
4493                 }
4494                 dout("send reconnect to export target mds.%d\n", i);
4495                 mutex_unlock(&mdsc->mutex);
4496                 send_mds_reconnect(mdsc, s);
4497                 ceph_put_mds_session(s);
4498                 mutex_lock(&mdsc->mutex);
4499         }
4500
4501         for (i = 0; i < newmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4502                 s = mdsc->sessions[i];
4503                 if (!s)
4504                         continue;
4505                 if (!ceph_mdsmap_is_laggy(newmap, i))
4506                         continue;
4507                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4508                     s->s_state == CEPH_MDS_SESSION_HUNG ||
4509                     s->s_state == CEPH_MDS_SESSION_CLOSING) {
4510                         dout(" connecting to export targets of laggy mds%d\n",
4511                              i);
4512                         __open_export_target_sessions(mdsc, s);
4513                 }
4514         }
4515 }
4516
4517
4518
4519 /*
4520  * leases
4521  */
4522
4523 /*
4524  * caller must hold session s_mutex, dentry->d_lock
4525  */
4526 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
4527 {
4528         struct ceph_dentry_info *di = ceph_dentry(dentry);
4529
4530         ceph_put_mds_session(di->lease_session);
4531         di->lease_session = NULL;
4532 }
4533
4534 static void handle_lease(struct ceph_mds_client *mdsc,
4535                          struct ceph_mds_session *session,
4536                          struct ceph_msg *msg)
4537 {
4538         struct super_block *sb = mdsc->fsc->sb;
4539         struct inode *inode;
4540         struct dentry *parent, *dentry;
4541         struct ceph_dentry_info *di;
4542         int mds = session->s_mds;
4543         struct ceph_mds_lease *h = msg->front.iov_base;
4544         u32 seq;
4545         struct ceph_vino vino;
4546         struct qstr dname;
4547         int release = 0;
4548
4549         dout("handle_lease from mds%d\n", mds);
4550
4551         /* decode */
4552         if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
4553                 goto bad;
4554         vino.ino = le64_to_cpu(h->ino);
4555         vino.snap = CEPH_NOSNAP;
4556         seq = le32_to_cpu(h->seq);
4557         dname.len = get_unaligned_le32(h + 1);
4558         if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
4559                 goto bad;
4560         dname.name = (void *)(h + 1) + sizeof(u32);
4561
4562         /* lookup inode */
4563         inode = ceph_find_inode(sb, vino);
4564         dout("handle_lease %s, ino %llx %p %.*s\n",
4565              ceph_lease_op_name(h->action), vino.ino, inode,
4566              dname.len, dname.name);
4567
4568         mutex_lock(&session->s_mutex);
4569         inc_session_sequence(session);
4570
4571         if (!inode) {
4572                 dout("handle_lease no inode %llx\n", vino.ino);
4573                 goto release;
4574         }
4575
4576         /* dentry */
4577         parent = d_find_alias(inode);
4578         if (!parent) {
4579                 dout("no parent dentry on inode %p\n", inode);
4580                 WARN_ON(1);
4581                 goto release;  /* hrm... */
4582         }
4583         dname.hash = full_name_hash(parent, dname.name, dname.len);
4584         dentry = d_lookup(parent, &dname);
4585         dput(parent);
4586         if (!dentry)
4587                 goto release;
4588
4589         spin_lock(&dentry->d_lock);
4590         di = ceph_dentry(dentry);
4591         switch (h->action) {
4592         case CEPH_MDS_LEASE_REVOKE:
4593                 if (di->lease_session == session) {
4594                         if (ceph_seq_cmp(di->lease_seq, seq) > 0)
4595                                 h->seq = cpu_to_le32(di->lease_seq);
4596                         __ceph_mdsc_drop_dentry_lease(dentry);
4597                 }
4598                 release = 1;
4599                 break;
4600
4601         case CEPH_MDS_LEASE_RENEW:
4602                 if (di->lease_session == session &&
4603                     di->lease_gen == atomic_read(&session->s_cap_gen) &&
4604                     di->lease_renew_from &&
4605                     di->lease_renew_after == 0) {
4606                         unsigned long duration =
4607                                 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
4608
4609                         di->lease_seq = seq;
4610                         di->time = di->lease_renew_from + duration;
4611                         di->lease_renew_after = di->lease_renew_from +
4612                                 (duration >> 1);
4613                         di->lease_renew_from = 0;
4614                 }
4615                 break;
4616         }
4617         spin_unlock(&dentry->d_lock);
4618         dput(dentry);
4619
4620         if (!release)
4621                 goto out;
4622
4623 release:
4624         /* let's just reuse the same message */
4625         h->action = CEPH_MDS_LEASE_REVOKE_ACK;
4626         ceph_msg_get(msg);
4627         ceph_con_send(&session->s_con, msg);
4628
4629 out:
4630         mutex_unlock(&session->s_mutex);
4631         iput(inode);
4632         return;
4633
4634 bad:
4635         pr_err("corrupt lease message\n");
4636         ceph_msg_dump(msg);
4637 }
4638
4639 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
4640                               struct dentry *dentry, char action,
4641                               u32 seq)
4642 {
4643         struct ceph_msg *msg;
4644         struct ceph_mds_lease *lease;
4645         struct inode *dir;
4646         int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
4647
4648         dout("lease_send_msg identry %p %s to mds%d\n",
4649              dentry, ceph_lease_op_name(action), session->s_mds);
4650
4651         msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
4652         if (!msg)
4653                 return;
4654         lease = msg->front.iov_base;
4655         lease->action = action;
4656         lease->seq = cpu_to_le32(seq);
4657
4658         spin_lock(&dentry->d_lock);
4659         dir = d_inode(dentry->d_parent);
4660         lease->ino = cpu_to_le64(ceph_ino(dir));
4661         lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
4662
4663         put_unaligned_le32(dentry->d_name.len, lease + 1);
4664         memcpy((void *)(lease + 1) + 4,
4665                dentry->d_name.name, dentry->d_name.len);
4666         spin_unlock(&dentry->d_lock);
4667
4668         ceph_con_send(&session->s_con, msg);
4669 }
4670
4671 /*
4672  * lock unlock the session, to wait ongoing session activities
4673  */
4674 static void lock_unlock_session(struct ceph_mds_session *s)
4675 {
4676         mutex_lock(&s->s_mutex);
4677         mutex_unlock(&s->s_mutex);
4678 }
4679
4680 static void maybe_recover_session(struct ceph_mds_client *mdsc)
4681 {
4682         struct ceph_fs_client *fsc = mdsc->fsc;
4683
4684         if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
4685                 return;
4686
4687         if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
4688                 return;
4689
4690         if (!READ_ONCE(fsc->blocklisted))
4691                 return;
4692
4693         pr_info("auto reconnect after blocklisted\n");
4694         ceph_force_reconnect(fsc->sb);
4695 }
4696
4697 bool check_session_state(struct ceph_mds_session *s)
4698 {
4699         switch (s->s_state) {
4700         case CEPH_MDS_SESSION_OPEN:
4701                 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
4702                         s->s_state = CEPH_MDS_SESSION_HUNG;
4703                         pr_info("mds%d hung\n", s->s_mds);
4704                 }
4705                 break;
4706         case CEPH_MDS_SESSION_CLOSING:
4707         case CEPH_MDS_SESSION_NEW:
4708         case CEPH_MDS_SESSION_RESTARTING:
4709         case CEPH_MDS_SESSION_CLOSED:
4710         case CEPH_MDS_SESSION_REJECTED:
4711                 return false;
4712         }
4713
4714         return true;
4715 }
4716
4717 /*
4718  * If the sequence is incremented while we're waiting on a REQUEST_CLOSE reply,
4719  * then we need to retransmit that request.
4720  */
4721 void inc_session_sequence(struct ceph_mds_session *s)
4722 {
4723         lockdep_assert_held(&s->s_mutex);
4724
4725         s->s_seq++;
4726
4727         if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
4728                 int ret;
4729
4730                 dout("resending session close request for mds%d\n", s->s_mds);
4731                 ret = request_close_session(s);
4732                 if (ret < 0)
4733                         pr_err("unable to close session to mds%d: %d\n",
4734                                s->s_mds, ret);
4735         }
4736 }
4737
4738 /*
4739  * delayed work -- periodically trim expired leases, renew caps with mds.  If
4740  * the @delay parameter is set to 0 or if it's more than 5 secs, the default
4741  * workqueue delay value of 5 secs will be used.
4742  */
4743 static void schedule_delayed(struct ceph_mds_client *mdsc, unsigned long delay)
4744 {
4745         unsigned long max_delay = HZ * 5;
4746
4747         /* 5 secs default delay */
4748         if (!delay || (delay > max_delay))
4749                 delay = max_delay;
4750         schedule_delayed_work(&mdsc->delayed_work,
4751                               round_jiffies_relative(delay));
4752 }
4753
4754 static void delayed_work(struct work_struct *work)
4755 {
4756         struct ceph_mds_client *mdsc =
4757                 container_of(work, struct ceph_mds_client, delayed_work.work);
4758         unsigned long delay;
4759         int renew_interval;
4760         int renew_caps;
4761         int i;
4762
4763         dout("mdsc delayed_work\n");
4764
4765         if (mdsc->stopping)
4766                 return;
4767
4768         mutex_lock(&mdsc->mutex);
4769         renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
4770         renew_caps = time_after_eq(jiffies, HZ*renew_interval +
4771                                    mdsc->last_renew_caps);
4772         if (renew_caps)
4773                 mdsc->last_renew_caps = jiffies;
4774
4775         for (i = 0; i < mdsc->max_sessions; i++) {
4776                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4777                 if (!s)
4778                         continue;
4779
4780                 if (!check_session_state(s)) {
4781                         ceph_put_mds_session(s);
4782                         continue;
4783                 }
4784                 mutex_unlock(&mdsc->mutex);
4785
4786                 mutex_lock(&s->s_mutex);
4787                 if (renew_caps)
4788                         send_renew_caps(mdsc, s);
4789                 else
4790                         ceph_con_keepalive(&s->s_con);
4791                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4792                     s->s_state == CEPH_MDS_SESSION_HUNG)
4793                         ceph_send_cap_releases(mdsc, s);
4794                 mutex_unlock(&s->s_mutex);
4795                 ceph_put_mds_session(s);
4796
4797                 mutex_lock(&mdsc->mutex);
4798         }
4799         mutex_unlock(&mdsc->mutex);
4800
4801         delay = ceph_check_delayed_caps(mdsc);
4802
4803         ceph_queue_cap_reclaim_work(mdsc);
4804
4805         ceph_trim_snapid_map(mdsc);
4806
4807         maybe_recover_session(mdsc);
4808
4809         schedule_delayed(mdsc, delay);
4810 }
4811
4812 int ceph_mdsc_init(struct ceph_fs_client *fsc)
4813
4814 {
4815         struct ceph_mds_client *mdsc;
4816         int err;
4817
4818         mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
4819         if (!mdsc)
4820                 return -ENOMEM;
4821         mdsc->fsc = fsc;
4822         mutex_init(&mdsc->mutex);
4823         mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
4824         if (!mdsc->mdsmap) {
4825                 err = -ENOMEM;
4826                 goto err_mdsc;
4827         }
4828
4829         init_completion(&mdsc->safe_umount_waiters);
4830         init_waitqueue_head(&mdsc->session_close_wq);
4831         INIT_LIST_HEAD(&mdsc->waiting_for_map);
4832         mdsc->quotarealms_inodes = RB_ROOT;
4833         mutex_init(&mdsc->quotarealms_inodes_mutex);
4834         init_rwsem(&mdsc->snap_rwsem);
4835         mdsc->snap_realms = RB_ROOT;
4836         INIT_LIST_HEAD(&mdsc->snap_empty);
4837         spin_lock_init(&mdsc->snap_empty_lock);
4838         mdsc->request_tree = RB_ROOT;
4839         INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
4840         mdsc->last_renew_caps = jiffies;
4841         INIT_LIST_HEAD(&mdsc->cap_delay_list);
4842         INIT_LIST_HEAD(&mdsc->cap_wait_list);
4843         spin_lock_init(&mdsc->cap_delay_lock);
4844         INIT_LIST_HEAD(&mdsc->snap_flush_list);
4845         spin_lock_init(&mdsc->snap_flush_lock);
4846         mdsc->last_cap_flush_tid = 1;
4847         INIT_LIST_HEAD(&mdsc->cap_flush_list);
4848         INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
4849         spin_lock_init(&mdsc->cap_dirty_lock);
4850         init_waitqueue_head(&mdsc->cap_flushing_wq);
4851         INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
4852         err = ceph_metric_init(&mdsc->metric);
4853         if (err)
4854                 goto err_mdsmap;
4855
4856         spin_lock_init(&mdsc->dentry_list_lock);
4857         INIT_LIST_HEAD(&mdsc->dentry_leases);
4858         INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
4859
4860         ceph_caps_init(mdsc);
4861         ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
4862
4863         spin_lock_init(&mdsc->snapid_map_lock);
4864         mdsc->snapid_map_tree = RB_ROOT;
4865         INIT_LIST_HEAD(&mdsc->snapid_map_lru);
4866
4867         init_rwsem(&mdsc->pool_perm_rwsem);
4868         mdsc->pool_perm_tree = RB_ROOT;
4869
4870         strscpy(mdsc->nodename, utsname()->nodename,
4871                 sizeof(mdsc->nodename));
4872
4873         fsc->mdsc = mdsc;
4874         return 0;
4875
4876 err_mdsmap:
4877         kfree(mdsc->mdsmap);
4878 err_mdsc:
4879         kfree(mdsc);
4880         return err;
4881 }
4882
4883 /*
4884  * Wait for safe replies on open mds requests.  If we time out, drop
4885  * all requests from the tree to avoid dangling dentry refs.
4886  */
4887 static void wait_requests(struct ceph_mds_client *mdsc)
4888 {
4889         struct ceph_options *opts = mdsc->fsc->client->options;
4890         struct ceph_mds_request *req;
4891
4892         mutex_lock(&mdsc->mutex);
4893         if (__get_oldest_req(mdsc)) {
4894                 mutex_unlock(&mdsc->mutex);
4895
4896                 dout("wait_requests waiting for requests\n");
4897                 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
4898                                     ceph_timeout_jiffies(opts->mount_timeout));
4899
4900                 /* tear down remaining requests */
4901                 mutex_lock(&mdsc->mutex);
4902                 while ((req = __get_oldest_req(mdsc))) {
4903                         dout("wait_requests timed out on tid %llu\n",
4904                              req->r_tid);
4905                         list_del_init(&req->r_wait);
4906                         __unregister_request(mdsc, req);
4907                 }
4908         }
4909         mutex_unlock(&mdsc->mutex);
4910         dout("wait_requests done\n");
4911 }
4912
4913 void send_flush_mdlog(struct ceph_mds_session *s)
4914 {
4915         struct ceph_msg *msg;
4916
4917         /*
4918          * Pre-luminous MDS crashes when it sees an unknown session request
4919          */
4920         if (!CEPH_HAVE_FEATURE(s->s_con.peer_features, SERVER_LUMINOUS))
4921                 return;
4922
4923         mutex_lock(&s->s_mutex);
4924         dout("request mdlog flush to mds%d (%s)s seq %lld\n", s->s_mds,
4925              ceph_session_state_name(s->s_state), s->s_seq);
4926         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_FLUSH_MDLOG,
4927                                       s->s_seq);
4928         if (!msg) {
4929                 pr_err("failed to request mdlog flush to mds%d (%s) seq %lld\n",
4930                        s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
4931         } else {
4932                 ceph_con_send(&s->s_con, msg);
4933         }
4934         mutex_unlock(&s->s_mutex);
4935 }
4936
4937 /*
4938  * called before mount is ro, and before dentries are torn down.
4939  * (hmm, does this still race with new lookups?)
4940  */
4941 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
4942 {
4943         dout("pre_umount\n");
4944         mdsc->stopping = 1;
4945
4946         ceph_mdsc_iterate_sessions(mdsc, send_flush_mdlog, true);
4947         ceph_mdsc_iterate_sessions(mdsc, lock_unlock_session, false);
4948         ceph_flush_dirty_caps(mdsc);
4949         wait_requests(mdsc);
4950
4951         /*
4952          * wait for reply handlers to drop their request refs and
4953          * their inode/dcache refs
4954          */
4955         ceph_msgr_flush();
4956
4957         ceph_cleanup_quotarealms_inodes(mdsc);
4958 }
4959
4960 /*
4961  * flush the mdlog and wait for all write mds requests to flush.
4962  */
4963 static void flush_mdlog_and_wait_mdsc_unsafe_requests(struct ceph_mds_client *mdsc,
4964                                                  u64 want_tid)
4965 {
4966         struct ceph_mds_request *req = NULL, *nextreq;
4967         struct ceph_mds_session *last_session = NULL;
4968         struct rb_node *n;
4969
4970         mutex_lock(&mdsc->mutex);
4971         dout("%s want %lld\n", __func__, want_tid);
4972 restart:
4973         req = __get_oldest_req(mdsc);
4974         while (req && req->r_tid <= want_tid) {
4975                 /* find next request */
4976                 n = rb_next(&req->r_node);
4977                 if (n)
4978                         nextreq = rb_entry(n, struct ceph_mds_request, r_node);
4979                 else
4980                         nextreq = NULL;
4981                 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
4982                     (req->r_op & CEPH_MDS_OP_WRITE)) {
4983                         struct ceph_mds_session *s = req->r_session;
4984
4985                         if (!s) {
4986                                 req = nextreq;
4987                                 continue;
4988                         }
4989
4990                         /* write op */
4991                         ceph_mdsc_get_request(req);
4992                         if (nextreq)
4993                                 ceph_mdsc_get_request(nextreq);
4994                         s = ceph_get_mds_session(s);
4995                         mutex_unlock(&mdsc->mutex);
4996
4997                         /* send flush mdlog request to MDS */
4998                         if (last_session != s) {
4999                                 send_flush_mdlog(s);
5000                                 ceph_put_mds_session(last_session);
5001                                 last_session = s;
5002                         } else {
5003                                 ceph_put_mds_session(s);
5004                         }
5005                         dout("%s wait on %llu (want %llu)\n", __func__,
5006                              req->r_tid, want_tid);
5007                         wait_for_completion(&req->r_safe_completion);
5008
5009                         mutex_lock(&mdsc->mutex);
5010                         ceph_mdsc_put_request(req);
5011                         if (!nextreq)
5012                                 break;  /* next dne before, so we're done! */
5013                         if (RB_EMPTY_NODE(&nextreq->r_node)) {
5014                                 /* next request was removed from tree */
5015                                 ceph_mdsc_put_request(nextreq);
5016                                 goto restart;
5017                         }
5018                         ceph_mdsc_put_request(nextreq);  /* won't go away */
5019                 }
5020                 req = nextreq;
5021         }
5022         mutex_unlock(&mdsc->mutex);
5023         ceph_put_mds_session(last_session);
5024         dout("%s done\n", __func__);
5025 }
5026
5027 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
5028 {
5029         u64 want_tid, want_flush;
5030
5031         if (READ_ONCE(mdsc->fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN)
5032                 return;
5033
5034         dout("sync\n");
5035         mutex_lock(&mdsc->mutex);
5036         want_tid = mdsc->last_tid;
5037         mutex_unlock(&mdsc->mutex);
5038
5039         ceph_flush_dirty_caps(mdsc);
5040         spin_lock(&mdsc->cap_dirty_lock);
5041         want_flush = mdsc->last_cap_flush_tid;
5042         if (!list_empty(&mdsc->cap_flush_list)) {
5043                 struct ceph_cap_flush *cf =
5044                         list_last_entry(&mdsc->cap_flush_list,
5045                                         struct ceph_cap_flush, g_list);
5046                 cf->wake = true;
5047         }
5048         spin_unlock(&mdsc->cap_dirty_lock);
5049
5050         dout("sync want tid %lld flush_seq %lld\n",
5051              want_tid, want_flush);
5052
5053         flush_mdlog_and_wait_mdsc_unsafe_requests(mdsc, want_tid);
5054         wait_caps_flush(mdsc, want_flush);
5055 }
5056
5057 /*
5058  * true if all sessions are closed, or we force unmount
5059  */
5060 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
5061 {
5062         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
5063                 return true;
5064         return atomic_read(&mdsc->num_sessions) <= skipped;
5065 }
5066
5067 /*
5068  * called after sb is ro or when metadata corrupted.
5069  */
5070 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
5071 {
5072         struct ceph_options *opts = mdsc->fsc->client->options;
5073         struct ceph_mds_session *session;
5074         int i;
5075         int skipped = 0;
5076
5077         dout("close_sessions\n");
5078
5079         /* close sessions */
5080         mutex_lock(&mdsc->mutex);
5081         for (i = 0; i < mdsc->max_sessions; i++) {
5082                 session = __ceph_lookup_mds_session(mdsc, i);
5083                 if (!session)
5084                         continue;
5085                 mutex_unlock(&mdsc->mutex);
5086                 mutex_lock(&session->s_mutex);
5087                 if (__close_session(mdsc, session) <= 0)
5088                         skipped++;
5089                 mutex_unlock(&session->s_mutex);
5090                 ceph_put_mds_session(session);
5091                 mutex_lock(&mdsc->mutex);
5092         }
5093         mutex_unlock(&mdsc->mutex);
5094
5095         dout("waiting for sessions to close\n");
5096         wait_event_timeout(mdsc->session_close_wq,
5097                            done_closing_sessions(mdsc, skipped),
5098                            ceph_timeout_jiffies(opts->mount_timeout));
5099
5100         /* tear down remaining sessions */
5101         mutex_lock(&mdsc->mutex);
5102         for (i = 0; i < mdsc->max_sessions; i++) {
5103                 if (mdsc->sessions[i]) {
5104                         session = ceph_get_mds_session(mdsc->sessions[i]);
5105                         __unregister_session(mdsc, session);
5106                         mutex_unlock(&mdsc->mutex);
5107                         mutex_lock(&session->s_mutex);
5108                         remove_session_caps(session);
5109                         mutex_unlock(&session->s_mutex);
5110                         ceph_put_mds_session(session);
5111                         mutex_lock(&mdsc->mutex);
5112                 }
5113         }
5114         WARN_ON(!list_empty(&mdsc->cap_delay_list));
5115         mutex_unlock(&mdsc->mutex);
5116
5117         ceph_cleanup_snapid_map(mdsc);
5118         ceph_cleanup_global_and_empty_realms(mdsc);
5119
5120         cancel_work_sync(&mdsc->cap_reclaim_work);
5121         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
5122
5123         dout("stopped\n");
5124 }
5125
5126 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
5127 {
5128         struct ceph_mds_session *session;
5129         int mds;
5130
5131         dout("force umount\n");
5132
5133         mutex_lock(&mdsc->mutex);
5134         for (mds = 0; mds < mdsc->max_sessions; mds++) {
5135                 session = __ceph_lookup_mds_session(mdsc, mds);
5136                 if (!session)
5137                         continue;
5138
5139                 if (session->s_state == CEPH_MDS_SESSION_REJECTED)
5140                         __unregister_session(mdsc, session);
5141                 __wake_requests(mdsc, &session->s_waiting);
5142                 mutex_unlock(&mdsc->mutex);
5143
5144                 mutex_lock(&session->s_mutex);
5145                 __close_session(mdsc, session);
5146                 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
5147                         cleanup_session_requests(mdsc, session);
5148                         remove_session_caps(session);
5149                 }
5150                 mutex_unlock(&session->s_mutex);
5151                 ceph_put_mds_session(session);
5152
5153                 mutex_lock(&mdsc->mutex);
5154                 kick_requests(mdsc, mds);
5155         }
5156         __wake_requests(mdsc, &mdsc->waiting_for_map);
5157         mutex_unlock(&mdsc->mutex);
5158 }
5159
5160 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
5161 {
5162         dout("stop\n");
5163         /*
5164          * Make sure the delayed work stopped before releasing
5165          * the resources.
5166          *
5167          * Because the cancel_delayed_work_sync() will only
5168          * guarantee that the work finishes executing. But the
5169          * delayed work will re-arm itself again after that.
5170          */
5171         flush_delayed_work(&mdsc->delayed_work);
5172
5173         if (mdsc->mdsmap)
5174                 ceph_mdsmap_destroy(mdsc->mdsmap);
5175         kfree(mdsc->sessions);
5176         ceph_caps_finalize(mdsc);
5177         ceph_pool_perm_destroy(mdsc);
5178 }
5179
5180 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
5181 {
5182         struct ceph_mds_client *mdsc = fsc->mdsc;
5183         dout("mdsc_destroy %p\n", mdsc);
5184
5185         if (!mdsc)
5186                 return;
5187
5188         /* flush out any connection work with references to us */
5189         ceph_msgr_flush();
5190
5191         ceph_mdsc_stop(mdsc);
5192
5193         ceph_metric_destroy(&mdsc->metric);
5194
5195         fsc->mdsc = NULL;
5196         kfree(mdsc);
5197         dout("mdsc_destroy %p done\n", mdsc);
5198 }
5199
5200 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5201 {
5202         struct ceph_fs_client *fsc = mdsc->fsc;
5203         const char *mds_namespace = fsc->mount_options->mds_namespace;
5204         void *p = msg->front.iov_base;
5205         void *end = p + msg->front.iov_len;
5206         u32 epoch;
5207         u32 num_fs;
5208         u32 mount_fscid = (u32)-1;
5209         int err = -EINVAL;
5210
5211         ceph_decode_need(&p, end, sizeof(u32), bad);
5212         epoch = ceph_decode_32(&p);
5213
5214         dout("handle_fsmap epoch %u\n", epoch);
5215
5216         /* struct_v, struct_cv, map_len, epoch, legacy_client_fscid */
5217         ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 3, bad);
5218
5219         ceph_decode_32_safe(&p, end, num_fs, bad);
5220         while (num_fs-- > 0) {
5221                 void *info_p, *info_end;
5222                 u32 info_len;
5223                 u32 fscid, namelen;
5224
5225                 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
5226                 p += 2;         // info_v, info_cv
5227                 info_len = ceph_decode_32(&p);
5228                 ceph_decode_need(&p, end, info_len, bad);
5229                 info_p = p;
5230                 info_end = p + info_len;
5231                 p = info_end;
5232
5233                 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
5234                 fscid = ceph_decode_32(&info_p);
5235                 namelen = ceph_decode_32(&info_p);
5236                 ceph_decode_need(&info_p, info_end, namelen, bad);
5237
5238                 if (mds_namespace &&
5239                     strlen(mds_namespace) == namelen &&
5240                     !strncmp(mds_namespace, (char *)info_p, namelen)) {
5241                         mount_fscid = fscid;
5242                         break;
5243                 }
5244         }
5245
5246         ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
5247         if (mount_fscid != (u32)-1) {
5248                 fsc->client->monc.fs_cluster_id = mount_fscid;
5249                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
5250                                    0, true);
5251                 ceph_monc_renew_subs(&fsc->client->monc);
5252         } else {
5253                 err = -ENOENT;
5254                 goto err_out;
5255         }
5256         return;
5257
5258 bad:
5259         pr_err("error decoding fsmap %d. Shutting down mount.\n", err);
5260         ceph_umount_begin(mdsc->fsc->sb);
5261 err_out:
5262         mutex_lock(&mdsc->mutex);
5263         mdsc->mdsmap_err = err;
5264         __wake_requests(mdsc, &mdsc->waiting_for_map);
5265         mutex_unlock(&mdsc->mutex);
5266 }
5267
5268 /*
5269  * handle mds map update.
5270  */
5271 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5272 {
5273         u32 epoch;
5274         u32 maplen;
5275         void *p = msg->front.iov_base;
5276         void *end = p + msg->front.iov_len;
5277         struct ceph_mdsmap *newmap, *oldmap;
5278         struct ceph_fsid fsid;
5279         int err = -EINVAL;
5280
5281         ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
5282         ceph_decode_copy(&p, &fsid, sizeof(fsid));
5283         if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
5284                 return;
5285         epoch = ceph_decode_32(&p);
5286         maplen = ceph_decode_32(&p);
5287         dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
5288
5289         /* do we need it? */
5290         mutex_lock(&mdsc->mutex);
5291         if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
5292                 dout("handle_map epoch %u <= our %u\n",
5293                      epoch, mdsc->mdsmap->m_epoch);
5294                 mutex_unlock(&mdsc->mutex);
5295                 return;
5296         }
5297
5298         newmap = ceph_mdsmap_decode(&p, end, ceph_msgr2(mdsc->fsc->client));
5299         if (IS_ERR(newmap)) {
5300                 err = PTR_ERR(newmap);
5301                 goto bad_unlock;
5302         }
5303
5304         /* swap into place */
5305         if (mdsc->mdsmap) {
5306                 oldmap = mdsc->mdsmap;
5307                 mdsc->mdsmap = newmap;
5308                 check_new_map(mdsc, newmap, oldmap);
5309                 ceph_mdsmap_destroy(oldmap);
5310         } else {
5311                 mdsc->mdsmap = newmap;  /* first mds map */
5312         }
5313         mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
5314                                         MAX_LFS_FILESIZE);
5315
5316         __wake_requests(mdsc, &mdsc->waiting_for_map);
5317         ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
5318                           mdsc->mdsmap->m_epoch);
5319
5320         mutex_unlock(&mdsc->mutex);
5321         schedule_delayed(mdsc, 0);
5322         return;
5323
5324 bad_unlock:
5325         mutex_unlock(&mdsc->mutex);
5326 bad:
5327         pr_err("error decoding mdsmap %d. Shutting down mount.\n", err);
5328         ceph_umount_begin(mdsc->fsc->sb);
5329         return;
5330 }
5331
5332 static struct ceph_connection *mds_get_con(struct ceph_connection *con)
5333 {
5334         struct ceph_mds_session *s = con->private;
5335
5336         if (ceph_get_mds_session(s))
5337                 return con;
5338         return NULL;
5339 }
5340
5341 static void mds_put_con(struct ceph_connection *con)
5342 {
5343         struct ceph_mds_session *s = con->private;
5344
5345         ceph_put_mds_session(s);
5346 }
5347
5348 /*
5349  * if the client is unresponsive for long enough, the mds will kill
5350  * the session entirely.
5351  */
5352 static void mds_peer_reset(struct ceph_connection *con)
5353 {
5354         struct ceph_mds_session *s = con->private;
5355         struct ceph_mds_client *mdsc = s->s_mdsc;
5356
5357         pr_warn("mds%d closed our session\n", s->s_mds);
5358         if (READ_ONCE(mdsc->fsc->mount_state) != CEPH_MOUNT_FENCE_IO)
5359                 send_mds_reconnect(mdsc, s);
5360 }
5361
5362 static void mds_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5363 {
5364         struct ceph_mds_session *s = con->private;
5365         struct ceph_mds_client *mdsc = s->s_mdsc;
5366         int type = le16_to_cpu(msg->hdr.type);
5367
5368         mutex_lock(&mdsc->mutex);
5369         if (__verify_registered_session(mdsc, s) < 0) {
5370                 mutex_unlock(&mdsc->mutex);
5371                 goto out;
5372         }
5373         mutex_unlock(&mdsc->mutex);
5374
5375         switch (type) {
5376         case CEPH_MSG_MDS_MAP:
5377                 ceph_mdsc_handle_mdsmap(mdsc, msg);
5378                 break;
5379         case CEPH_MSG_FS_MAP_USER:
5380                 ceph_mdsc_handle_fsmap(mdsc, msg);
5381                 break;
5382         case CEPH_MSG_CLIENT_SESSION:
5383                 handle_session(s, msg);
5384                 break;
5385         case CEPH_MSG_CLIENT_REPLY:
5386                 handle_reply(s, msg);
5387                 break;
5388         case CEPH_MSG_CLIENT_REQUEST_FORWARD:
5389                 handle_forward(mdsc, s, msg);
5390                 break;
5391         case CEPH_MSG_CLIENT_CAPS:
5392                 ceph_handle_caps(s, msg);
5393                 break;
5394         case CEPH_MSG_CLIENT_SNAP:
5395                 ceph_handle_snap(mdsc, s, msg);
5396                 break;
5397         case CEPH_MSG_CLIENT_LEASE:
5398                 handle_lease(mdsc, s, msg);
5399                 break;
5400         case CEPH_MSG_CLIENT_QUOTA:
5401                 ceph_handle_quota(mdsc, s, msg);
5402                 break;
5403
5404         default:
5405                 pr_err("received unknown message type %d %s\n", type,
5406                        ceph_msg_type_name(type));
5407         }
5408 out:
5409         ceph_msg_put(msg);
5410 }
5411
5412 /*
5413  * authentication
5414  */
5415
5416 /*
5417  * Note: returned pointer is the address of a structure that's
5418  * managed separately.  Caller must *not* attempt to free it.
5419  */
5420 static struct ceph_auth_handshake *
5421 mds_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5422 {
5423         struct ceph_mds_session *s = con->private;
5424         struct ceph_mds_client *mdsc = s->s_mdsc;
5425         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5426         struct ceph_auth_handshake *auth = &s->s_auth;
5427         int ret;
5428
5429         ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5430                                          force_new, proto, NULL, NULL);
5431         if (ret)
5432                 return ERR_PTR(ret);
5433
5434         return auth;
5435 }
5436
5437 static int mds_add_authorizer_challenge(struct ceph_connection *con,
5438                                     void *challenge_buf, int challenge_buf_len)
5439 {
5440         struct ceph_mds_session *s = con->private;
5441         struct ceph_mds_client *mdsc = s->s_mdsc;
5442         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5443
5444         return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
5445                                             challenge_buf, challenge_buf_len);
5446 }
5447
5448 static int mds_verify_authorizer_reply(struct ceph_connection *con)
5449 {
5450         struct ceph_mds_session *s = con->private;
5451         struct ceph_mds_client *mdsc = s->s_mdsc;
5452         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5453         struct ceph_auth_handshake *auth = &s->s_auth;
5454
5455         return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5456                 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5457                 NULL, NULL, NULL, NULL);
5458 }
5459
5460 static int mds_invalidate_authorizer(struct ceph_connection *con)
5461 {
5462         struct ceph_mds_session *s = con->private;
5463         struct ceph_mds_client *mdsc = s->s_mdsc;
5464         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5465
5466         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
5467
5468         return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
5469 }
5470
5471 static int mds_get_auth_request(struct ceph_connection *con,
5472                                 void *buf, int *buf_len,
5473                                 void **authorizer, int *authorizer_len)
5474 {
5475         struct ceph_mds_session *s = con->private;
5476         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5477         struct ceph_auth_handshake *auth = &s->s_auth;
5478         int ret;
5479
5480         ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5481                                        buf, buf_len);
5482         if (ret)
5483                 return ret;
5484
5485         *authorizer = auth->authorizer_buf;
5486         *authorizer_len = auth->authorizer_buf_len;
5487         return 0;
5488 }
5489
5490 static int mds_handle_auth_reply_more(struct ceph_connection *con,
5491                                       void *reply, int reply_len,
5492                                       void *buf, int *buf_len,
5493                                       void **authorizer, int *authorizer_len)
5494 {
5495         struct ceph_mds_session *s = con->private;
5496         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5497         struct ceph_auth_handshake *auth = &s->s_auth;
5498         int ret;
5499
5500         ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5501                                               buf, buf_len);
5502         if (ret)
5503                 return ret;
5504
5505         *authorizer = auth->authorizer_buf;
5506         *authorizer_len = auth->authorizer_buf_len;
5507         return 0;
5508 }
5509
5510 static int mds_handle_auth_done(struct ceph_connection *con,
5511                                 u64 global_id, void *reply, int reply_len,
5512                                 u8 *session_key, int *session_key_len,
5513                                 u8 *con_secret, int *con_secret_len)
5514 {
5515         struct ceph_mds_session *s = con->private;
5516         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5517         struct ceph_auth_handshake *auth = &s->s_auth;
5518
5519         return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5520                                                session_key, session_key_len,
5521                                                con_secret, con_secret_len);
5522 }
5523
5524 static int mds_handle_auth_bad_method(struct ceph_connection *con,
5525                                       int used_proto, int result,
5526                                       const int *allowed_protos, int proto_cnt,
5527                                       const int *allowed_modes, int mode_cnt)
5528 {
5529         struct ceph_mds_session *s = con->private;
5530         struct ceph_mon_client *monc = &s->s_mdsc->fsc->client->monc;
5531         int ret;
5532
5533         if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_MDS,
5534                                             used_proto, result,
5535                                             allowed_protos, proto_cnt,
5536                                             allowed_modes, mode_cnt)) {
5537                 ret = ceph_monc_validate_auth(monc);
5538                 if (ret)
5539                         return ret;
5540         }
5541
5542         return -EACCES;
5543 }
5544
5545 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
5546                                 struct ceph_msg_header *hdr, int *skip)
5547 {
5548         struct ceph_msg *msg;
5549         int type = (int) le16_to_cpu(hdr->type);
5550         int front_len = (int) le32_to_cpu(hdr->front_len);
5551
5552         if (con->in_msg)
5553                 return con->in_msg;
5554
5555         *skip = 0;
5556         msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
5557         if (!msg) {
5558                 pr_err("unable to allocate msg type %d len %d\n",
5559                        type, front_len);
5560                 return NULL;
5561         }
5562
5563         return msg;
5564 }
5565
5566 static int mds_sign_message(struct ceph_msg *msg)
5567 {
5568        struct ceph_mds_session *s = msg->con->private;
5569        struct ceph_auth_handshake *auth = &s->s_auth;
5570
5571        return ceph_auth_sign_message(auth, msg);
5572 }
5573
5574 static int mds_check_message_signature(struct ceph_msg *msg)
5575 {
5576        struct ceph_mds_session *s = msg->con->private;
5577        struct ceph_auth_handshake *auth = &s->s_auth;
5578
5579        return ceph_auth_check_message_signature(auth, msg);
5580 }
5581
5582 static const struct ceph_connection_operations mds_con_ops = {
5583         .get = mds_get_con,
5584         .put = mds_put_con,
5585         .alloc_msg = mds_alloc_msg,
5586         .dispatch = mds_dispatch,
5587         .peer_reset = mds_peer_reset,
5588         .get_authorizer = mds_get_authorizer,
5589         .add_authorizer_challenge = mds_add_authorizer_challenge,
5590         .verify_authorizer_reply = mds_verify_authorizer_reply,
5591         .invalidate_authorizer = mds_invalidate_authorizer,
5592         .sign_message = mds_sign_message,
5593         .check_message_signature = mds_check_message_signature,
5594         .get_auth_request = mds_get_auth_request,
5595         .handle_auth_reply_more = mds_handle_auth_reply_more,
5596         .handle_auth_done = mds_handle_auth_done,
5597         .handle_auth_bad_method = mds_handle_auth_bad_method,
5598 };
5599
5600 /* eof */