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  50  * are no longer valid.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 (mds >= mdsc->mdsmap->possible_max_rank)
 810                 return ERR_PTR(-EINVAL);
 811
 812         s = kzalloc(sizeof(*s), GFP_NOFS);
 813         if (!s)
 814                 return ERR_PTR(-ENOMEM);
 815
 816         if (mds >= mdsc->max_sessions) {
 817                 int newmax = 1 << get_count_order(mds + 1);
 818                 struct ceph_mds_session **sa;
 819
 820                 dout("%s: realloc to %d\n", __func__, newmax);
 821                 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
 822                 if (!sa)
 823                         goto fail_realloc;
 824                 if (mdsc->sessions) {
 825                         memcpy(sa, mdsc->sessions,
 826                                mdsc->max_sessions * sizeof(void *));
 827                         kfree(mdsc->sessions);
 828                 }
 829                 mdsc->sessions = sa;
 830                 mdsc->max_sessions = newmax;
 831         }
 832
 833         dout("%s: mds%d\n", __func__, mds);
 834         s->s_mdsc = mdsc;
 835         s->s_mds = mds;
 836         s->s_state = CEPH_MDS_SESSION_NEW;
 837         mutex_init(&s->s_mutex);
 838
 839         ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
 840
 841         atomic_set(&s->s_cap_gen, 1);
 842         s->s_cap_ttl = jiffies - 1;
 843
 844         spin_lock_init(&s->s_cap_lock);
 845         INIT_LIST_HEAD(&s->s_caps);
 846         refcount_set(&s->s_ref, 1);
 847         INIT_LIST_HEAD(&s->s_waiting);
 848         INIT_LIST_HEAD(&s->s_unsafe);
 849         xa_init(&s->s_delegated_inos);
 850         INIT_LIST_HEAD(&s->s_cap_releases);
 851         INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
 852
 853         INIT_LIST_HEAD(&s->s_cap_dirty);
 854         INIT_LIST_HEAD(&s->s_cap_flushing);
 855
 856         mdsc->sessions[mds] = s;
 857         atomic_inc(&mdsc->num_sessions);
 858         refcount_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
 859
 860         ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
 861                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
 862
 863         return s;
 864
 865 fail_realloc:
 866         kfree(s);
 867         return ERR_PTR(-ENOMEM);
 868 }
 869
 870 /*
 871  * called under mdsc->mutex
 872  */
 873 static void __unregister_session(struct ceph_mds_client *mdsc,
 874                                struct ceph_mds_session *s)
 875 {
 876         dout("__unregister_session mds%d %p\n", s->s_mds, s);
 877         BUG_ON(mdsc->sessions[s->s_mds] != s);
 878         mdsc->sessions[s->s_mds] = NULL;
 879         ceph_con_close(&s->s_con);
 880         ceph_put_mds_session(s);
 881         atomic_dec(&mdsc->num_sessions);
 882 }
 883
 884 /*
 885  * drop session refs in request.
 886  *
 887  * should be last request ref, or hold mdsc->mutex
 888  */
 889 static void put_request_session(struct ceph_mds_request *req)
 890 {
 891         if (req->r_session) {
 892                 ceph_put_mds_session(req->r_session);
 893                 req->r_session = NULL;
 894         }
 895 }
 896
 897 void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
 898                                 void (*cb)(struct ceph_mds_session *),
 899                                 bool check_state)
 900 {
 901         int mds;
 902
 903         mutex_lock(&mdsc->mutex);
 904         for (mds = 0; mds < mdsc->max_sessions; ++mds) {
 905                 struct ceph_mds_session *s;
 906
 907                 s = __ceph_lookup_mds_session(mdsc, mds);
 908                 if (!s)
 909                         continue;
 910
 911                 if (check_state && !check_session_state(s)) {
 912                         ceph_put_mds_session(s);
 913                         continue;
 914                 }
 915
 916                 mutex_unlock(&mdsc->mutex);
 917                 cb(s);
 918                 ceph_put_mds_session(s);
 919                 mutex_lock(&mdsc->mutex);
 920         }
 921         mutex_unlock(&mdsc->mutex);
 922 }
 923
 924 void ceph_mdsc_release_request(struct kref *kref)
 925 {
 926         struct ceph_mds_request *req = container_of(kref,
 927                                                     struct ceph_mds_request,
 928                                                     r_kref);
 929         ceph_mdsc_release_dir_caps_no_check(req);
 930         destroy_reply_info(&req->r_reply_info);
 931         if (req->r_request)
 932                 ceph_msg_put(req->r_request);
 933         if (req->r_reply)
 934                 ceph_msg_put(req->r_reply);
 935         if (req->r_inode) {
 936                 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
 937                 iput(req->r_inode);
 938         }
 939         if (req->r_parent) {
 940                 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
 941                 iput(req->r_parent);
 942         }
 943         iput(req->r_target_inode);
 944         if (req->r_dentry)
 945                 dput(req->r_dentry);
 946         if (req->r_old_dentry)
 947                 dput(req->r_old_dentry);
 948         if (req->r_old_dentry_dir) {
 949                 /*
 950                  * track (and drop pins for) r_old_dentry_dir
 951                  * separately, since r_old_dentry's d_parent may have
 952                  * changed between the dir mutex being dropped and
 953                  * this request being freed.
 954                  */
 955                 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
 956                                   CEPH_CAP_PIN);
 957                 iput(req->r_old_dentry_dir);
 958         }
 959         kfree(req->r_path1);
 960         kfree(req->r_path2);
 961         put_cred(req->r_cred);
 962         if (req->r_pagelist)
 963                 ceph_pagelist_release(req->r_pagelist);
 964         put_request_session(req);
 965         ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
 966         WARN_ON_ONCE(!list_empty(&req->r_wait));
 967         kmem_cache_free(ceph_mds_request_cachep, req);
 968 }
 969
 970 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
 971
 972 /*
 973  * lookup session, bump ref if found.
 974  *
 975  * called under mdsc->mutex.
 976  */
 977 static struct ceph_mds_request *
 978 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
 979 {
 980         struct ceph_mds_request *req;
 981
 982         req = lookup_request(&mdsc->request_tree, tid);
 983         if (req)
 984                 ceph_mdsc_get_request(req);
 985
 986         return req;
 987 }
 988
 989 /*
 990  * Register an in-flight request, and assign a tid.  Link to directory
 991  * are modifying (if any).
 992  *
 993  * Called under mdsc->mutex.
 994  */
 995 static void __register_request(struct ceph_mds_client *mdsc,
 996                                struct ceph_mds_request *req,
 997                                struct inode *dir)
 998 {
 999         int ret = 0;
1000
1001         req->r_tid = ++mdsc->last_tid;
1002         if (req->r_num_caps) {
1003                 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
1004                                         req->r_num_caps);
1005                 if (ret < 0) {
1006                         pr_err("__register_request %p "
1007                                "failed to reserve caps: %d\n", req, ret);
1008                         /* set req->r_err to fail early from __do_request */
1009                         req->r_err = ret;
1010                         return;
1011                 }
1012         }
1013         dout("__register_request %p tid %lld\n", req, req->r_tid);
1014         ceph_mdsc_get_request(req);
1015         insert_request(&mdsc->request_tree, req);
1016
1017         req->r_cred = get_current_cred();
1018
1019         if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
1020                 mdsc->oldest_tid = req->r_tid;
1021
1022         if (dir) {
1023                 struct ceph_inode_info *ci = ceph_inode(dir);
1024
1025                 ihold(dir);
1026                 req->r_unsafe_dir = dir;
1027                 spin_lock(&ci->i_unsafe_lock);
1028                 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
1029                 spin_unlock(&ci->i_unsafe_lock);
1030         }
1031 }
1032
1033 static void __unregister_request(struct ceph_mds_client *mdsc,
1034                                  struct ceph_mds_request *req)
1035 {
1036         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
1037
1038         /* Never leave an unregistered request on an unsafe list! */
1039         list_del_init(&req->r_unsafe_item);
1040
1041         if (req->r_tid == mdsc->oldest_tid) {
1042                 struct rb_node *p = rb_next(&req->r_node);
1043                 mdsc->oldest_tid = 0;
1044                 while (p) {
1045                         struct ceph_mds_request *next_req =
1046                                 rb_entry(p, struct ceph_mds_request, r_node);
1047                         if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
1048                                 mdsc->oldest_tid = next_req->r_tid;
1049                                 break;
1050                         }
1051                         p = rb_next(p);
1052                 }
1053         }
1054
1055         erase_request(&mdsc->request_tree, req);
1056
1057         if (req->r_unsafe_dir) {
1058                 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
1059                 spin_lock(&ci->i_unsafe_lock);
1060                 list_del_init(&req->r_unsafe_dir_item);
1061                 spin_unlock(&ci->i_unsafe_lock);
1062         }
1063         if (req->r_target_inode &&
1064             test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
1065                 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
1066                 spin_lock(&ci->i_unsafe_lock);
1067                 list_del_init(&req->r_unsafe_target_item);
1068                 spin_unlock(&ci->i_unsafe_lock);
1069         }
1070
1071         if (req->r_unsafe_dir) {
1072                 iput(req->r_unsafe_dir);
1073                 req->r_unsafe_dir = NULL;
1074         }
1075
1076         complete_all(&req->r_safe_completion);
1077
1078         ceph_mdsc_put_request(req);
1079 }
1080
1081 /*
1082  * Walk back up the dentry tree until we hit a dentry representing a
1083  * non-snapshot inode. We do this using the rcu_read_lock (which must be held
1084  * when calling this) to ensure that the objects won't disappear while we're
1085  * working with them. Once we hit a candidate dentry, we attempt to take a
1086  * reference to it, and return that as the result.
1087  */
1088 static struct inode *get_nonsnap_parent(struct dentry *dentry)
1089 {
1090         struct inode *inode = NULL;
1091
1092         while (dentry && !IS_ROOT(dentry)) {
1093                 inode = d_inode_rcu(dentry);
1094                 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
1095                         break;
1096                 dentry = dentry->d_parent;
1097         }
1098         if (inode)
1099                 inode = igrab(inode);
1100         return inode;
1101 }
1102
1103 /*
1104  * Choose mds to send request to next.  If there is a hint set in the
1105  * request (e.g., due to a prior forward hint from the mds), use that.
1106  * Otherwise, consult frag tree and/or caps to identify the
1107  * appropriate mds.  If all else fails, choose randomly.
1108  *
1109  * Called under mdsc->mutex.
1110  */
1111 static int __choose_mds(struct ceph_mds_client *mdsc,
1112                         struct ceph_mds_request *req,
1113                         bool *random)
1114 {
1115         struct inode *inode;
1116         struct ceph_inode_info *ci;
1117         struct ceph_cap *cap;
1118         int mode = req->r_direct_mode;
1119         int mds = -1;
1120         u32 hash = req->r_direct_hash;
1121         bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
1122
1123         if (random)
1124                 *random = false;
1125
1126         /*
1127          * is there a specific mds we should try?  ignore hint if we have
1128          * no session and the mds is not up (active or recovering).
1129          */
1130         if (req->r_resend_mds >= 0 &&
1131             (__have_session(mdsc, req->r_resend_mds) ||
1132              ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
1133                 dout("%s using resend_mds mds%d\n", __func__,
1134                      req->r_resend_mds);
1135                 return req->r_resend_mds;
1136         }
1137
1138         if (mode == USE_RANDOM_MDS)
1139                 goto random;
1140
1141         inode = NULL;
1142         if (req->r_inode) {
1143                 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
1144                         inode = req->r_inode;
1145                         ihold(inode);
1146                 } else {
1147                         /* req->r_dentry is non-null for LSSNAP request */
1148                         rcu_read_lock();
1149                         inode = get_nonsnap_parent(req->r_dentry);
1150                         rcu_read_unlock();
1151                         dout("%s using snapdir's parent %p\n", __func__, inode);
1152                 }
1153         } else if (req->r_dentry) {
1154                 /* ignore race with rename; old or new d_parent is okay */
1155                 struct dentry *parent;
1156                 struct inode *dir;
1157
1158                 rcu_read_lock();
1159                 parent = READ_ONCE(req->r_dentry->d_parent);
1160                 dir = req->r_parent ? : d_inode_rcu(parent);
1161
1162                 if (!dir || dir->i_sb != mdsc->fsc->sb) {
1163                         /*  not this fs or parent went negative */
1164                         inode = d_inode(req->r_dentry);
1165                         if (inode)
1166                                 ihold(inode);
1167                 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
1168                         /* direct snapped/virtual snapdir requests
1169                          * based on parent dir inode */
1170                         inode = get_nonsnap_parent(parent);
1171                         dout("%s using nonsnap parent %p\n", __func__, inode);
1172                 } else {
1173                         /* dentry target */
1174                         inode = d_inode(req->r_dentry);
1175                         if (!inode || mode == USE_AUTH_MDS) {
1176                                 /* dir + name */
1177                                 inode = igrab(dir);
1178                                 hash = ceph_dentry_hash(dir, req->r_dentry);
1179                                 is_hash = true;
1180                         } else {
1181                                 ihold(inode);
1182                         }
1183                 }
1184                 rcu_read_unlock();
1185         }
1186
1187         dout("%s %p is_hash=%d (0x%x) mode %d\n", __func__, inode, (int)is_hash,
1188              hash, mode);
1189         if (!inode)
1190                 goto random;
1191         ci = ceph_inode(inode);
1192
1193         if (is_hash && S_ISDIR(inode->i_mode)) {
1194                 struct ceph_inode_frag frag;
1195                 int found;
1196
1197                 ceph_choose_frag(ci, hash, &frag, &found);
1198                 if (found) {
1199                         if (mode == USE_ANY_MDS && frag.ndist > 0) {
1200                                 u8 r;
1201
1202                                 /* choose a random replica */
1203                                 get_random_bytes(&r, 1);
1204                                 r %= frag.ndist;
1205                                 mds = frag.dist[r];
1206                                 dout("%s %p %llx.%llx frag %u mds%d (%d/%d)\n",
1207                                      __func__, inode, ceph_vinop(inode),
1208                                      frag.frag, mds, (int)r, frag.ndist);
1209                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1210                                     CEPH_MDS_STATE_ACTIVE &&
1211                                     !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
1212                                         goto out;
1213                         }
1214
1215                         /* since this file/dir wasn't known to be
1216                          * replicated, then we want to look for the
1217                          * authoritative mds. */
1218                         if (frag.mds >= 0) {
1219                                 /* choose auth mds */
1220                                 mds = frag.mds;
1221                                 dout("%s %p %llx.%llx frag %u mds%d (auth)\n",
1222                                      __func__, inode, ceph_vinop(inode),
1223                                      frag.frag, mds);
1224                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1225                                     CEPH_MDS_STATE_ACTIVE) {
1226                                         if (!ceph_mdsmap_is_laggy(mdsc->mdsmap,
1227                                                                   mds))
1228                                                 goto out;
1229                                 }
1230                         }
1231                         mode = USE_AUTH_MDS;
1232                 }
1233         }
1234
1235         spin_lock(&ci->i_ceph_lock);
1236         cap = NULL;
1237         if (mode == USE_AUTH_MDS)
1238                 cap = ci->i_auth_cap;
1239         if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1240                 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1241         if (!cap) {
1242                 spin_unlock(&ci->i_ceph_lock);
1243                 iput(inode);
1244                 goto random;
1245         }
1246         mds = cap->session->s_mds;
1247         dout("%s %p %llx.%llx mds%d (%scap %p)\n", __func__,
1248              inode, ceph_vinop(inode), mds,
1249              cap == ci->i_auth_cap ? "auth " : "", cap);
1250         spin_unlock(&ci->i_ceph_lock);
1251 out:
1252         iput(inode);
1253         return mds;
1254
1255 random:
1256         if (random)
1257                 *random = true;
1258
1259         mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
1260         dout("%s chose random mds%d\n", __func__, mds);
1261         return mds;
1262 }
1263
1264
1265 /*
1266  * session messages
1267  */
1268 struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq)
1269 {
1270         struct ceph_msg *msg;
1271         struct ceph_mds_session_head *h;
1272
1273         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1274                            false);
1275         if (!msg) {
1276                 pr_err("ENOMEM creating session %s msg\n",
1277                        ceph_session_op_name(op));
1278                 return NULL;
1279         }
1280         h = msg->front.iov_base;
1281         h->op = cpu_to_le32(op);
1282         h->seq = cpu_to_le64(seq);
1283
1284         return msg;
1285 }
1286
1287 static const unsigned char feature_bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1288 #define FEATURE_BYTES(c) (DIV_ROUND_UP((size_t)feature_bits[c - 1] + 1, 64) * 8)
1289 static int encode_supported_features(void **p, void *end)
1290 {
1291         static const size_t count = ARRAY_SIZE(feature_bits);
1292
1293         if (count > 0) {
1294                 size_t i;
1295                 size_t size = FEATURE_BYTES(count);
1296                 unsigned long bit;
1297
1298                 if (WARN_ON_ONCE(*p + 4 + size > end))
1299                         return -ERANGE;
1300
1301                 ceph_encode_32(p, size);
1302                 memset(*p, 0, size);
1303                 for (i = 0; i < count; i++) {
1304                         bit = feature_bits[i];
1305                         ((unsigned char *)(*p))[bit / 8] |= BIT(bit % 8);
1306                 }
1307                 *p += size;
1308         } else {
1309                 if (WARN_ON_ONCE(*p + 4 > end))
1310                         return -ERANGE;
1311
1312                 ceph_encode_32(p, 0);
1313         }
1314
1315         return 0;
1316 }
1317
1318 static const unsigned char metric_bits[] = CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED;
1319 #define METRIC_BYTES(cnt) (DIV_ROUND_UP((size_t)metric_bits[cnt - 1] + 1, 64) * 8)
1320 static int encode_metric_spec(void **p, void *end)
1321 {
1322         static const size_t count = ARRAY_SIZE(metric_bits);
1323
1324         /* header */
1325         if (WARN_ON_ONCE(*p + 2 > end))
1326                 return -ERANGE;
1327
1328         ceph_encode_8(p, 1); /* version */
1329         ceph_encode_8(p, 1); /* compat */
1330
1331         if (count > 0) {
1332                 size_t i;
1333                 size_t size = METRIC_BYTES(count);
1334
1335                 if (WARN_ON_ONCE(*p + 4 + 4 + size > end))
1336                         return -ERANGE;
1337
1338                 /* metric spec info length */
1339                 ceph_encode_32(p, 4 + size);
1340
1341                 /* metric spec */
1342                 ceph_encode_32(p, size);
1343                 memset(*p, 0, size);
1344                 for (i = 0; i < count; i++)
1345                         ((unsigned char *)(*p))[i / 8] |= BIT(metric_bits[i] % 8);
1346                 *p += size;
1347         } else {
1348                 if (WARN_ON_ONCE(*p + 4 + 4 > end))
1349                         return -ERANGE;
1350
1351                 /* metric spec info length */
1352                 ceph_encode_32(p, 4);
1353                 /* metric spec */
1354                 ceph_encode_32(p, 0);
1355         }
1356
1357         return 0;
1358 }
1359
1360 /*
1361  * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1362  * to include additional client metadata fields.
1363  */
1364 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
1365 {
1366         struct ceph_msg *msg;
1367         struct ceph_mds_session_head *h;
1368         int i;
1369         int extra_bytes = 0;
1370         int metadata_key_count = 0;
1371         struct ceph_options *opt = mdsc->fsc->client->options;
1372         struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
1373         size_t size, count;
1374         void *p, *end;
1375         int ret;
1376
1377         const char* metadata[][2] = {
1378                 {"hostname", mdsc->nodename},
1379                 {"kernel_version", init_utsname()->release},
1380                 {"entity_id", opt->name ? : ""},
1381                 {"root", fsopt->server_path ? : "/"},
1382                 {NULL, NULL}
1383         };
1384
1385         /* Calculate serialized length of metadata */
1386         extra_bytes = 4;  /* map length */
1387         for (i = 0; metadata[i][0]; ++i) {
1388                 extra_bytes += 8 + strlen(metadata[i][0]) +
1389                         strlen(metadata[i][1]);
1390                 metadata_key_count++;
1391         }
1392
1393         /* supported feature */
1394         size = 0;
1395         count = ARRAY_SIZE(feature_bits);
1396         if (count > 0)
1397                 size = FEATURE_BYTES(count);
1398         extra_bytes += 4 + size;
1399
1400         /* metric spec */
1401         size = 0;
1402         count = ARRAY_SIZE(metric_bits);
1403         if (count > 0)
1404                 size = METRIC_BYTES(count);
1405         extra_bytes += 2 + 4 + 4 + size;
1406
1407         /* Allocate the message */
1408         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
1409                            GFP_NOFS, false);
1410         if (!msg) {
1411                 pr_err("ENOMEM creating session open msg\n");
1412                 return ERR_PTR(-ENOMEM);
1413         }
1414         p = msg->front.iov_base;
1415         end = p + msg->front.iov_len;
1416
1417         h = p;
1418         h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
1419         h->seq = cpu_to_le64(seq);
1420
1421         /*
1422          * Serialize client metadata into waiting buffer space, using
1423          * the format that userspace expects for map<string, string>
1424          *
1425          * ClientSession messages with metadata are v4
1426          */
1427         msg->hdr.version = cpu_to_le16(4);
1428         msg->hdr.compat_version = cpu_to_le16(1);
1429
1430         /* The write pointer, following the session_head structure */
1431         p += sizeof(*h);
1432
1433         /* Number of entries in the map */
1434         ceph_encode_32(&p, metadata_key_count);
1435
1436         /* Two length-prefixed strings for each entry in the map */
1437         for (i = 0; metadata[i][0]; ++i) {
1438                 size_t const key_len = strlen(metadata[i][0]);
1439                 size_t const val_len = strlen(metadata[i][1]);
1440
1441                 ceph_encode_32(&p, key_len);
1442                 memcpy(p, metadata[i][0], key_len);
1443                 p += key_len;
1444                 ceph_encode_32(&p, val_len);
1445                 memcpy(p, metadata[i][1], val_len);
1446                 p += val_len;
1447         }
1448
1449         ret = encode_supported_features(&p, end);
1450         if (ret) {
1451                 pr_err("encode_supported_features failed!\n");
1452                 ceph_msg_put(msg);
1453                 return ERR_PTR(ret);
1454         }
1455
1456         ret = encode_metric_spec(&p, end);
1457         if (ret) {
1458                 pr_err("encode_metric_spec failed!\n");
1459                 ceph_msg_put(msg);
1460                 return ERR_PTR(ret);
1461         }
1462
1463         msg->front.iov_len = p - msg->front.iov_base;
1464         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1465
1466         return msg;
1467 }
1468
1469 /*
1470  * send session open request.
1471  *
1472  * called under mdsc->mutex
1473  */
1474 static int __open_session(struct ceph_mds_client *mdsc,
1475                           struct ceph_mds_session *session)
1476 {
1477         struct ceph_msg *msg;
1478         int mstate;
1479         int mds = session->s_mds;
1480
1481         /* wait for mds to go active? */
1482         mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1483         dout("open_session to mds%d (%s)\n", mds,
1484              ceph_mds_state_name(mstate));
1485         session->s_state = CEPH_MDS_SESSION_OPENING;
1486         session->s_renew_requested = jiffies;
1487
1488         /* send connect message */
1489         msg = create_session_open_msg(mdsc, session->s_seq);
1490         if (IS_ERR(msg))
1491                 return PTR_ERR(msg);
1492         ceph_con_send(&session->s_con, msg);
1493         return 0;
1494 }
1495
1496 /*
1497  * open sessions for any export targets for the given mds
1498  *
1499  * called under mdsc->mutex
1500  */
1501 static struct ceph_mds_session *
1502 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1503 {
1504         struct ceph_mds_session *session;
1505         int ret;
1506
1507         session = __ceph_lookup_mds_session(mdsc, target);
1508         if (!session) {
1509                 session = register_session(mdsc, target);
1510                 if (IS_ERR(session))
1511                         return session;
1512         }
1513         if (session->s_state == CEPH_MDS_SESSION_NEW ||
1514             session->s_state == CEPH_MDS_SESSION_CLOSING) {
1515                 ret = __open_session(mdsc, session);
1516                 if (ret)
1517                         return ERR_PTR(ret);
1518         }
1519
1520         return session;
1521 }
1522
1523 struct ceph_mds_session *
1524 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1525 {
1526         struct ceph_mds_session *session;
1527
1528         dout("open_export_target_session to mds%d\n", target);
1529
1530         mutex_lock(&mdsc->mutex);
1531         session = __open_export_target_session(mdsc, target);
1532         mutex_unlock(&mdsc->mutex);
1533
1534         return session;
1535 }
1536
1537 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1538                                           struct ceph_mds_session *session)
1539 {
1540         struct ceph_mds_info *mi;
1541         struct ceph_mds_session *ts;
1542         int i, mds = session->s_mds;
1543
1544         if (mds >= mdsc->mdsmap->possible_max_rank)
1545                 return;
1546
1547         mi = &mdsc->mdsmap->m_info[mds];
1548         dout("open_export_target_sessions for mds%d (%d targets)\n",
1549              session->s_mds, mi->num_export_targets);
1550
1551         for (i = 0; i < mi->num_export_targets; i++) {
1552                 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1553                 ceph_put_mds_session(ts);
1554         }
1555 }
1556
1557 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1558                                            struct ceph_mds_session *session)
1559 {
1560         mutex_lock(&mdsc->mutex);
1561         __open_export_target_sessions(mdsc, session);
1562         mutex_unlock(&mdsc->mutex);
1563 }
1564
1565 /*
1566  * session caps
1567  */
1568
1569 static void detach_cap_releases(struct ceph_mds_session *session,
1570                                 struct list_head *target)
1571 {
1572         lockdep_assert_held(&session->s_cap_lock);
1573
1574         list_splice_init(&session->s_cap_releases, target);
1575         session->s_num_cap_releases = 0;
1576         dout("dispose_cap_releases mds%d\n", session->s_mds);
1577 }
1578
1579 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1580                                  struct list_head *dispose)
1581 {
1582         while (!list_empty(dispose)) {
1583                 struct ceph_cap *cap;
1584                 /* zero out the in-progress message */
1585                 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1586                 list_del(&cap->session_caps);
1587                 ceph_put_cap(mdsc, cap);
1588         }
1589 }
1590
1591 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1592                                      struct ceph_mds_session *session)
1593 {
1594         struct ceph_mds_request *req;
1595         struct rb_node *p;
1596
1597         dout("cleanup_session_requests mds%d\n", session->s_mds);
1598         mutex_lock(&mdsc->mutex);
1599         while (!list_empty(&session->s_unsafe)) {
1600                 req = list_first_entry(&session->s_unsafe,
1601                                        struct ceph_mds_request, r_unsafe_item);
1602                 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1603                                     req->r_tid);
1604                 if (req->r_target_inode)
1605                         mapping_set_error(req->r_target_inode->i_mapping, -EIO);
1606                 if (req->r_unsafe_dir)
1607                         mapping_set_error(req->r_unsafe_dir->i_mapping, -EIO);
1608                 __unregister_request(mdsc, req);
1609         }
1610         /* zero r_attempts, so kick_requests() will re-send requests */
1611         p = rb_first(&mdsc->request_tree);
1612         while (p) {
1613                 req = rb_entry(p, struct ceph_mds_request, r_node);
1614                 p = rb_next(p);
1615                 if (req->r_session &&
1616                     req->r_session->s_mds == session->s_mds)
1617                         req->r_attempts = 0;
1618         }
1619         mutex_unlock(&mdsc->mutex);
1620 }
1621
1622 /*
1623  * Helper to safely iterate over all caps associated with a session, with
1624  * special care taken to handle a racing __ceph_remove_cap().
1625  *
1626  * Caller must hold session s_mutex.
1627  */
1628 int ceph_iterate_session_caps(struct ceph_mds_session *session,
1629                               int (*cb)(struct inode *, struct ceph_cap *,
1630                                         void *), void *arg)
1631 {
1632         struct list_head *p;
1633         struct ceph_cap *cap;
1634         struct inode *inode, *last_inode = NULL;
1635         struct ceph_cap *old_cap = NULL;
1636         int ret;
1637
1638         dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1639         spin_lock(&session->s_cap_lock);
1640         p = session->s_caps.next;
1641         while (p != &session->s_caps) {
1642                 cap = list_entry(p, struct ceph_cap, session_caps);
1643                 inode = igrab(&cap->ci->netfs.inode);
1644                 if (!inode) {
1645                         p = p->next;
1646                         continue;
1647                 }
1648                 session->s_cap_iterator = cap;
1649                 spin_unlock(&session->s_cap_lock);
1650
1651                 if (last_inode) {
1652                         iput(last_inode);
1653                         last_inode = NULL;
1654                 }
1655                 if (old_cap) {
1656                         ceph_put_cap(session->s_mdsc, old_cap);
1657                         old_cap = NULL;
1658                 }
1659
1660                 ret = cb(inode, cap, arg);
1661                 last_inode = inode;
1662
1663                 spin_lock(&session->s_cap_lock);
1664                 p = p->next;
1665                 if (!cap->ci) {
1666                         dout("iterate_session_caps  finishing cap %p removal\n",
1667                              cap);
1668                         BUG_ON(cap->session != session);
1669                         cap->session = NULL;
1670                         list_del_init(&cap->session_caps);
1671                         session->s_nr_caps--;
1672                         atomic64_dec(&session->s_mdsc->metric.total_caps);
1673                         if (cap->queue_release)
1674                                 __ceph_queue_cap_release(session, cap);
1675                         else
1676                                 old_cap = cap;  /* put_cap it w/o locks held */
1677                 }
1678                 if (ret < 0)
1679                         goto out;
1680         }
1681         ret = 0;
1682 out:
1683         session->s_cap_iterator = NULL;
1684         spin_unlock(&session->s_cap_lock);
1685
1686         iput(last_inode);
1687         if (old_cap)
1688                 ceph_put_cap(session->s_mdsc, old_cap);
1689
1690         return ret;
1691 }
1692
1693 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1694                                   void *arg)
1695 {
1696         struct ceph_inode_info *ci = ceph_inode(inode);
1697         bool invalidate = false;
1698         int iputs;
1699
1700         dout("removing cap %p, ci is %p, inode is %p\n",
1701              cap, ci, &ci->netfs.inode);
1702         spin_lock(&ci->i_ceph_lock);
1703         iputs = ceph_purge_inode_cap(inode, cap, &invalidate);
1704         spin_unlock(&ci->i_ceph_lock);
1705
1706         wake_up_all(&ci->i_cap_wq);
1707         if (invalidate)
1708                 ceph_queue_invalidate(inode);
1709         while (iputs--)
1710                 iput(inode);
1711         return 0;
1712 }
1713
1714 /*
1715  * caller must hold session s_mutex
1716  */
1717 static void remove_session_caps(struct ceph_mds_session *session)
1718 {
1719         struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1720         struct super_block *sb = fsc->sb;
1721         LIST_HEAD(dispose);
1722
1723         dout("remove_session_caps on %p\n", session);
1724         ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
1725
1726         wake_up_all(&fsc->mdsc->cap_flushing_wq);
1727
1728         spin_lock(&session->s_cap_lock);
1729         if (session->s_nr_caps > 0) {
1730                 struct inode *inode;
1731                 struct ceph_cap *cap, *prev = NULL;
1732                 struct ceph_vino vino;
1733                 /*
1734                  * iterate_session_caps() skips inodes that are being
1735                  * deleted, we need to wait until deletions are complete.
1736                  * __wait_on_freeing_inode() is designed for the job,
1737                  * but it is not exported, so use lookup inode function
1738                  * to access it.
1739                  */
1740                 while (!list_empty(&session->s_caps)) {
1741                         cap = list_entry(session->s_caps.next,
1742                                          struct ceph_cap, session_caps);
1743                         if (cap == prev)
1744                                 break;
1745                         prev = cap;
1746                         vino = cap->ci->i_vino;
1747                         spin_unlock(&session->s_cap_lock);
1748
1749                         inode = ceph_find_inode(sb, vino);
1750                         iput(inode);
1751
1752                         spin_lock(&session->s_cap_lock);
1753                 }
1754         }
1755
1756         // drop cap expires and unlock s_cap_lock
1757         detach_cap_releases(session, &dispose);
1758
1759         BUG_ON(session->s_nr_caps > 0);
1760         BUG_ON(!list_empty(&session->s_cap_flushing));
1761         spin_unlock(&session->s_cap_lock);
1762         dispose_cap_releases(session->s_mdsc, &dispose);
1763 }
1764
1765 enum {
1766         RECONNECT,
1767         RENEWCAPS,
1768         FORCE_RO,
1769 };
1770
1771 /*
1772  * wake up any threads waiting on this session's caps.  if the cap is
1773  * old (didn't get renewed on the client reconnect), remove it now.
1774  *
1775  * caller must hold s_mutex.
1776  */
1777 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1778                               void *arg)
1779 {
1780         struct ceph_inode_info *ci = ceph_inode(inode);
1781         unsigned long ev = (unsigned long)arg;
1782
1783         if (ev == RECONNECT) {
1784                 spin_lock(&ci->i_ceph_lock);
1785                 ci->i_wanted_max_size = 0;
1786                 ci->i_requested_max_size = 0;
1787                 spin_unlock(&ci->i_ceph_lock);
1788         } else if (ev == RENEWCAPS) {
1789                 if (cap->cap_gen < atomic_read(&cap->session->s_cap_gen)) {
1790                         /* mds did not re-issue stale cap */
1791                         spin_lock(&ci->i_ceph_lock);
1792                         cap->issued = cap->implemented = CEPH_CAP_PIN;
1793                         spin_unlock(&ci->i_ceph_lock);
1794                 }
1795         } else if (ev == FORCE_RO) {
1796         }
1797         wake_up_all(&ci->i_cap_wq);
1798         return 0;
1799 }
1800
1801 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
1802 {
1803         dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1804         ceph_iterate_session_caps(session, wake_up_session_cb,
1805                                   (void *)(unsigned long)ev);
1806 }
1807
1808 /*
1809  * Send periodic message to MDS renewing all currently held caps.  The
1810  * ack will reset the expiration for all caps from this session.
1811  *
1812  * caller holds s_mutex
1813  */
1814 static int send_renew_caps(struct ceph_mds_client *mdsc,
1815                            struct ceph_mds_session *session)
1816 {
1817         struct ceph_msg *msg;
1818         int state;
1819
1820         if (time_after_eq(jiffies, session->s_cap_ttl) &&
1821             time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1822                 pr_info("mds%d caps stale\n", session->s_mds);
1823         session->s_renew_requested = jiffies;
1824
1825         /* do not try to renew caps until a recovering mds has reconnected
1826          * with its clients. */
1827         state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1828         if (state < CEPH_MDS_STATE_RECONNECT) {
1829                 dout("send_renew_caps ignoring mds%d (%s)\n",
1830                      session->s_mds, ceph_mds_state_name(state));
1831                 return 0;
1832         }
1833
1834         dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1835                 ceph_mds_state_name(state));
1836         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1837                                       ++session->s_renew_seq);
1838         if (!msg)
1839                 return -ENOMEM;
1840         ceph_con_send(&session->s_con, msg);
1841         return 0;
1842 }
1843
1844 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1845                              struct ceph_mds_session *session, u64 seq)
1846 {
1847         struct ceph_msg *msg;
1848
1849         dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1850              session->s_mds, ceph_session_state_name(session->s_state), seq);
1851         msg = ceph_create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1852         if (!msg)
1853                 return -ENOMEM;
1854         ceph_con_send(&session->s_con, msg);
1855         return 0;
1856 }
1857
1858
1859 /*
1860  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1861  *
1862  * Called under session->s_mutex
1863  */
1864 static void renewed_caps(struct ceph_mds_client *mdsc,
1865                          struct ceph_mds_session *session, int is_renew)
1866 {
1867         int was_stale;
1868         int wake = 0;
1869
1870         spin_lock(&session->s_cap_lock);
1871         was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1872
1873         session->s_cap_ttl = session->s_renew_requested +
1874                 mdsc->mdsmap->m_session_timeout*HZ;
1875
1876         if (was_stale) {
1877                 if (time_before(jiffies, session->s_cap_ttl)) {
1878                         pr_info("mds%d caps renewed\n", session->s_mds);
1879                         wake = 1;
1880                 } else {
1881                         pr_info("mds%d caps still stale\n", session->s_mds);
1882                 }
1883         }
1884         dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1885              session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1886              time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1887         spin_unlock(&session->s_cap_lock);
1888
1889         if (wake)
1890                 wake_up_session_caps(session, RENEWCAPS);
1891 }
1892
1893 /*
1894  * send a session close request
1895  */
1896 static int request_close_session(struct ceph_mds_session *session)
1897 {
1898         struct ceph_msg *msg;
1899
1900         dout("request_close_session mds%d state %s seq %lld\n",
1901              session->s_mds, ceph_session_state_name(session->s_state),
1902              session->s_seq);
1903         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_CLOSE,
1904                                       session->s_seq);
1905         if (!msg)
1906                 return -ENOMEM;
1907         ceph_con_send(&session->s_con, msg);
1908         return 1;
1909 }
1910
1911 /*
1912  * Called with s_mutex held.
1913  */
1914 static int __close_session(struct ceph_mds_client *mdsc,
1915                          struct ceph_mds_session *session)
1916 {
1917         if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1918                 return 0;
1919         session->s_state = CEPH_MDS_SESSION_CLOSING;
1920         return request_close_session(session);
1921 }
1922
1923 static bool drop_negative_children(struct dentry *dentry)
1924 {
1925         struct dentry *child;
1926         bool all_negative = true;
1927
1928         if (!d_is_dir(dentry))
1929                 goto out;
1930
1931         spin_lock(&dentry->d_lock);
1932         list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1933                 if (d_really_is_positive(child)) {
1934                         all_negative = false;
1935                         break;
1936                 }
1937         }
1938         spin_unlock(&dentry->d_lock);
1939
1940         if (all_negative)
1941                 shrink_dcache_parent(dentry);
1942 out:
1943         return all_negative;
1944 }
1945
1946 /*
1947  * Trim old(er) caps.
1948  *
1949  * Because we can't cache an inode without one or more caps, we do
1950  * this indirectly: if a cap is unused, we prune its aliases, at which
1951  * point the inode will hopefully get dropped to.
1952  *
1953  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1954  * memory pressure from the MDS, though, so it needn't be perfect.
1955  */
1956 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1957 {
1958         int *remaining = arg;
1959         struct ceph_inode_info *ci = ceph_inode(inode);
1960         int used, wanted, oissued, mine;
1961
1962         if (*remaining <= 0)
1963                 return -1;
1964
1965         spin_lock(&ci->i_ceph_lock);
1966         mine = cap->issued | cap->implemented;
1967         used = __ceph_caps_used(ci);
1968         wanted = __ceph_caps_file_wanted(ci);
1969         oissued = __ceph_caps_issued_other(ci, cap);
1970
1971         dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1972              inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1973              ceph_cap_string(used), ceph_cap_string(wanted));
1974         if (cap == ci->i_auth_cap) {
1975                 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1976                     !list_empty(&ci->i_cap_snaps))
1977                         goto out;
1978                 if ((used | wanted) & CEPH_CAP_ANY_WR)
1979                         goto out;
1980                 /* Note: it's possible that i_filelock_ref becomes non-zero
1981                  * after dropping auth caps. It doesn't hurt because reply
1982                  * of lock mds request will re-add auth caps. */
1983                 if (atomic_read(&ci->i_filelock_ref) > 0)
1984                         goto out;
1985         }
1986         /* The inode has cached pages, but it's no longer used.
1987          * we can safely drop it */
1988         if (S_ISREG(inode->i_mode) &&
1989             wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1990             !(oissued & CEPH_CAP_FILE_CACHE)) {
1991           used = 0;
1992           oissued = 0;
1993         }
1994         if ((used | wanted) & ~oissued & mine)
1995                 goto out;   /* we need these caps */
1996
1997         if (oissued) {
1998                 /* we aren't the only cap.. just remove us */
1999                 ceph_remove_cap(cap, true);
2000                 (*remaining)--;
2001         } else {
2002                 struct dentry *dentry;
2003                 /* try dropping referring dentries */
2004                 spin_unlock(&ci->i_ceph_lock);
2005                 dentry = d_find_any_alias(inode);
2006                 if (dentry && drop_negative_children(dentry)) {
2007                         int count;
2008                         dput(dentry);
2009                         d_prune_aliases(inode);
2010                         count = atomic_read(&inode->i_count);
2011                         if (count == 1)
2012                                 (*remaining)--;
2013                         dout("trim_caps_cb %p cap %p pruned, count now %d\n",
2014                              inode, cap, count);
2015                 } else {
2016                         dput(dentry);
2017                 }
2018                 return 0;
2019         }
2020
2021 out:
2022         spin_unlock(&ci->i_ceph_lock);
2023         return 0;
2024 }
2025
2026 /*
2027  * Trim session cap count down to some max number.
2028  */
2029 int ceph_trim_caps(struct ceph_mds_client *mdsc,
2030                    struct ceph_mds_session *session,
2031                    int max_caps)
2032 {
2033         int trim_caps = session->s_nr_caps - max_caps;
2034
2035         dout("trim_caps mds%d start: %d / %d, trim %d\n",
2036              session->s_mds, session->s_nr_caps, max_caps, trim_caps);
2037         if (trim_caps > 0) {
2038                 int remaining = trim_caps;
2039
2040                 ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
2041                 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
2042                      session->s_mds, session->s_nr_caps, max_caps,
2043                         trim_caps - remaining);
2044         }
2045
2046         ceph_flush_cap_releases(mdsc, session);
2047         return 0;
2048 }
2049
2050 static int check_caps_flush(struct ceph_mds_client *mdsc,
2051                             u64 want_flush_tid)
2052 {
2053         int ret = 1;
2054
2055         spin_lock(&mdsc->cap_dirty_lock);
2056         if (!list_empty(&mdsc->cap_flush_list)) {
2057                 struct ceph_cap_flush *cf =
2058                         list_first_entry(&mdsc->cap_flush_list,
2059                                          struct ceph_cap_flush, g_list);
2060                 if (cf->tid <= want_flush_tid) {
2061                         dout("check_caps_flush still flushing tid "
2062                              "%llu <= %llu\n", cf->tid, want_flush_tid);
2063                         ret = 0;
2064                 }
2065         }
2066         spin_unlock(&mdsc->cap_dirty_lock);
2067         return ret;
2068 }
2069
2070 /*
2071  * flush all dirty inode data to disk.
2072  *
2073  * returns true if we've flushed through want_flush_tid
2074  */
2075 static void wait_caps_flush(struct ceph_mds_client *mdsc,
2076                             u64 want_flush_tid)
2077 {
2078         dout("check_caps_flush want %llu\n", want_flush_tid);
2079
2080         wait_event(mdsc->cap_flushing_wq,
2081                    check_caps_flush(mdsc, want_flush_tid));
2082
2083         dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2084 }
2085
2086 /*
2087  * called under s_mutex
2088  */
2089 static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
2090                                    struct ceph_mds_session *session)
2091 {
2092         struct ceph_msg *msg = NULL;
2093         struct ceph_mds_cap_release *head;
2094         struct ceph_mds_cap_item *item;
2095         struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
2096         struct ceph_cap *cap;
2097         LIST_HEAD(tmp_list);
2098         int num_cap_releases;
2099         __le32  barrier, *cap_barrier;
2100
2101         down_read(&osdc->lock);
2102         barrier = cpu_to_le32(osdc->epoch_barrier);
2103         up_read(&osdc->lock);
2104
2105         spin_lock(&session->s_cap_lock);
2106 again:
2107         list_splice_init(&session->s_cap_releases, &tmp_list);
2108         num_cap_releases = session->s_num_cap_releases;
2109         session->s_num_cap_releases = 0;
2110         spin_unlock(&session->s_cap_lock);
2111
2112         while (!list_empty(&tmp_list)) {
2113                 if (!msg) {
2114                         msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
2115                                         PAGE_SIZE, GFP_NOFS, false);
2116                         if (!msg)
2117                                 goto out_err;
2118                         head = msg->front.iov_base;
2119                         head->num = cpu_to_le32(0);
2120                         msg->front.iov_len = sizeof(*head);
2121
2122                         msg->hdr.version = cpu_to_le16(2);
2123                         msg->hdr.compat_version = cpu_to_le16(1);
2124                 }
2125
2126                 cap = list_first_entry(&tmp_list, struct ceph_cap,
2127                                         session_caps);
2128                 list_del(&cap->session_caps);
2129                 num_cap_releases--;
2130
2131                 head = msg->front.iov_base;
2132                 put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
2133                                    &head->num);
2134                 item = msg->front.iov_base + msg->front.iov_len;
2135                 item->ino = cpu_to_le64(cap->cap_ino);
2136                 item->cap_id = cpu_to_le64(cap->cap_id);
2137                 item->migrate_seq = cpu_to_le32(cap->mseq);
2138                 item->seq = cpu_to_le32(cap->issue_seq);
2139                 msg->front.iov_len += sizeof(*item);
2140
2141                 ceph_put_cap(mdsc, cap);
2142
2143                 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
2144                         // Append cap_barrier field
2145                         cap_barrier = msg->front.iov_base + msg->front.iov_len;
2146                         *cap_barrier = barrier;
2147                         msg->front.iov_len += sizeof(*cap_barrier);
2148
2149                         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2150                         dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2151                         ceph_con_send(&session->s_con, msg);
2152                         msg = NULL;
2153                 }
2154         }
2155
2156         BUG_ON(num_cap_releases != 0);
2157
2158         spin_lock(&session->s_cap_lock);
2159         if (!list_empty(&session->s_cap_releases))
2160                 goto again;
2161         spin_unlock(&session->s_cap_lock);
2162
2163         if (msg) {
2164                 // Append cap_barrier field
2165                 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2166                 *cap_barrier = barrier;
2167                 msg->front.iov_len += sizeof(*cap_barrier);
2168
2169                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2170                 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2171                 ceph_con_send(&session->s_con, msg);
2172         }
2173         return;
2174 out_err:
2175         pr_err("send_cap_releases mds%d, failed to allocate message\n",
2176                 session->s_mds);
2177         spin_lock(&session->s_cap_lock);
2178         list_splice(&tmp_list, &session->s_cap_releases);
2179         session->s_num_cap_releases += num_cap_releases;
2180         spin_unlock(&session->s_cap_lock);
2181 }
2182
2183 static void ceph_cap_release_work(struct work_struct *work)
2184 {
2185         struct ceph_mds_session *session =
2186                 container_of(work, struct ceph_mds_session, s_cap_release_work);
2187
2188         mutex_lock(&session->s_mutex);
2189         if (session->s_state == CEPH_MDS_SESSION_OPEN ||
2190             session->s_state == CEPH_MDS_SESSION_HUNG)
2191                 ceph_send_cap_releases(session->s_mdsc, session);
2192         mutex_unlock(&session->s_mutex);
2193         ceph_put_mds_session(session);
2194 }
2195
2196 void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
2197                              struct ceph_mds_session *session)
2198 {
2199         if (mdsc->stopping)
2200                 return;
2201
2202         ceph_get_mds_session(session);
2203         if (queue_work(mdsc->fsc->cap_wq,
2204                        &session->s_cap_release_work)) {
2205                 dout("cap release work queued\n");
2206         } else {
2207                 ceph_put_mds_session(session);
2208                 dout("failed to queue cap release work\n");
2209         }
2210 }
2211
2212 /*
2213  * caller holds session->s_cap_lock
2214  */
2215 void __ceph_queue_cap_release(struct ceph_mds_session *session,
2216                               struct ceph_cap *cap)
2217 {
2218         list_add_tail(&cap->session_caps, &session->s_cap_releases);
2219         session->s_num_cap_releases++;
2220
2221         if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
2222                 ceph_flush_cap_releases(session->s_mdsc, session);
2223 }
2224
2225 static void ceph_cap_reclaim_work(struct work_struct *work)
2226 {
2227         struct ceph_mds_client *mdsc =
2228                 container_of(work, struct ceph_mds_client, cap_reclaim_work);
2229         int ret = ceph_trim_dentries(mdsc);
2230         if (ret == -EAGAIN)
2231                 ceph_queue_cap_reclaim_work(mdsc);
2232 }
2233
2234 void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2235 {
2236         if (mdsc->stopping)
2237                 return;
2238
2239         if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
2240                 dout("caps reclaim work queued\n");
2241         } else {
2242                 dout("failed to queue caps release work\n");
2243         }
2244 }
2245
2246 void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2247 {
2248         int val;
2249         if (!nr)
2250                 return;
2251         val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
2252         if ((val % CEPH_CAPS_PER_RELEASE) < nr) {
2253                 atomic_set(&mdsc->cap_reclaim_pending, 0);
2254                 ceph_queue_cap_reclaim_work(mdsc);
2255         }
2256 }
2257
2258 /*
2259  * requests
2260  */
2261
2262 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2263                                     struct inode *dir)
2264 {
2265         struct ceph_inode_info *ci = ceph_inode(dir);
2266         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2267         struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2268         size_t size = sizeof(struct ceph_mds_reply_dir_entry);
2269         unsigned int num_entries;
2270         int order;
2271
2272         spin_lock(&ci->i_ceph_lock);
2273         num_entries = ci->i_files + ci->i_subdirs;
2274         spin_unlock(&ci->i_ceph_lock);
2275         num_entries = max(num_entries, 1U);
2276         num_entries = min(num_entries, opt->max_readdir);
2277
2278         order = get_order(size * num_entries);
2279         while (order >= 0) {
2280                 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2281                                                              __GFP_NOWARN |
2282                                                              __GFP_ZERO,
2283                                                              order);
2284                 if (rinfo->dir_entries)
2285                         break;
2286                 order--;
2287         }
2288         if (!rinfo->dir_entries)
2289                 return -ENOMEM;
2290
2291         num_entries = (PAGE_SIZE << order) / size;
2292         num_entries = min(num_entries, opt->max_readdir);
2293
2294         rinfo->dir_buf_size = PAGE_SIZE << order;
2295         req->r_num_caps = num_entries + 1;
2296         req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2297         req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2298         return 0;
2299 }
2300
2301 /*
2302  * Create an mds request.
2303  */
2304 struct ceph_mds_request *
2305 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2306 {
2307         struct ceph_mds_request *req;
2308
2309         req = kmem_cache_zalloc(ceph_mds_request_cachep, GFP_NOFS);
2310         if (!req)
2311                 return ERR_PTR(-ENOMEM);
2312
2313         mutex_init(&req->r_fill_mutex);
2314         req->r_mdsc = mdsc;
2315         req->r_started = jiffies;
2316         req->r_start_latency = ktime_get();
2317         req->r_resend_mds = -1;
2318         INIT_LIST_HEAD(&req->r_unsafe_dir_item);
2319         INIT_LIST_HEAD(&req->r_unsafe_target_item);
2320         req->r_fmode = -1;
2321         req->r_feature_needed = -1;
2322         kref_init(&req->r_kref);
2323         RB_CLEAR_NODE(&req->r_node);
2324         INIT_LIST_HEAD(&req->r_wait);
2325         init_completion(&req->r_completion);
2326         init_completion(&req->r_safe_completion);
2327         INIT_LIST_HEAD(&req->r_unsafe_item);
2328
2329         ktime_get_coarse_real_ts64(&req->r_stamp);
2330
2331         req->r_op = op;
2332         req->r_direct_mode = mode;
2333         return req;
2334 }
2335
2336 /*
2337  * return oldest (lowest) request, tid in request tree, 0 if none.
2338  *
2339  * called under mdsc->mutex.
2340  */
2341 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2342 {
2343         if (RB_EMPTY_ROOT(&mdsc->request_tree))
2344                 return NULL;
2345         return rb_entry(rb_first(&mdsc->request_tree),
2346                         struct ceph_mds_request, r_node);
2347 }
2348
2349 static inline  u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2350 {
2351         return mdsc->oldest_tid;
2352 }
2353
2354 /*
2355  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
2356  * on build_path_from_dentry in fs/cifs/dir.c.
2357  *
2358  * If @stop_on_nosnap, generate path relative to the first non-snapped
2359  * inode.
2360  *
2361  * Encode hidden .snap dirs as a double /, i.e.
2362  *   foo/.snap/bar -> foo//bar
2363  */
2364 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *pbase,
2365                            int stop_on_nosnap)
2366 {
2367         struct dentry *temp;
2368         char *path;
2369         int pos;
2370         unsigned seq;
2371         u64 base;
2372
2373         if (!dentry)
2374                 return ERR_PTR(-EINVAL);
2375
2376         path = __getname();
2377         if (!path)
2378                 return ERR_PTR(-ENOMEM);
2379 retry:
2380         pos = PATH_MAX - 1;
2381         path[pos] = '\0';
2382
2383         seq = read_seqbegin(&rename_lock);
2384         rcu_read_lock();
2385         temp = dentry;
2386         for (;;) {
2387                 struct inode *inode;
2388
2389                 spin_lock(&temp->d_lock);
2390                 inode = d_inode(temp);
2391                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
2392                         dout("build_path path+%d: %p SNAPDIR\n",
2393                              pos, temp);
2394                 } else if (stop_on_nosnap && inode && dentry != temp &&
2395                            ceph_snap(inode) == CEPH_NOSNAP) {
2396                         spin_unlock(&temp->d_lock);
2397                         pos++; /* get rid of any prepended '/' */
2398                         break;
2399                 } else {
2400                         pos -= temp->d_name.len;
2401                         if (pos < 0) {
2402                                 spin_unlock(&temp->d_lock);
2403                                 break;
2404                         }
2405                         memcpy(path + pos, temp->d_name.name, temp->d_name.len);
2406                 }
2407                 spin_unlock(&temp->d_lock);
2408                 temp = READ_ONCE(temp->d_parent);
2409
2410                 /* Are we at the root? */
2411                 if (IS_ROOT(temp))
2412                         break;
2413
2414                 /* Are we out of buffer? */
2415                 if (--pos < 0)
2416                         break;
2417
2418                 path[pos] = '/';
2419         }
2420         base = ceph_ino(d_inode(temp));
2421         rcu_read_unlock();
2422
2423         if (read_seqretry(&rename_lock, seq))
2424                 goto retry;
2425
2426         if (pos < 0) {
2427                 /*
2428                  * A rename didn't occur, but somehow we didn't end up where
2429                  * we thought we would. Throw a warning and try again.
2430                  */
2431                 pr_warn("build_path did not end path lookup where "
2432                         "expected, pos is %d\n", pos);
2433                 goto retry;
2434         }
2435
2436         *pbase = base;
2437         *plen = PATH_MAX - 1 - pos;
2438         dout("build_path on %p %d built %llx '%.*s'\n",
2439              dentry, d_count(dentry), base, *plen, path + pos);
2440         return path + pos;
2441 }
2442
2443 static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2444                              const char **ppath, int *ppathlen, u64 *pino,
2445                              bool *pfreepath, bool parent_locked)
2446 {
2447         char *path;
2448
2449         rcu_read_lock();
2450         if (!dir)
2451                 dir = d_inode_rcu(dentry->d_parent);
2452         if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP) {
2453                 *pino = ceph_ino(dir);
2454                 rcu_read_unlock();
2455                 *ppath = dentry->d_name.name;
2456                 *ppathlen = dentry->d_name.len;
2457                 return 0;
2458         }
2459         rcu_read_unlock();
2460         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2461         if (IS_ERR(path))
2462                 return PTR_ERR(path);
2463         *ppath = path;
2464         *pfreepath = true;
2465         return 0;
2466 }
2467
2468 static int build_inode_path(struct inode *inode,
2469                             const char **ppath, int *ppathlen, u64 *pino,
2470                             bool *pfreepath)
2471 {
2472         struct dentry *dentry;
2473         char *path;
2474
2475         if (ceph_snap(inode) == CEPH_NOSNAP) {
2476                 *pino = ceph_ino(inode);
2477                 *ppathlen = 0;
2478                 return 0;
2479         }
2480         dentry = d_find_alias(inode);
2481         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2482         dput(dentry);
2483         if (IS_ERR(path))
2484                 return PTR_ERR(path);
2485         *ppath = path;
2486         *pfreepath = true;
2487         return 0;
2488 }
2489
2490 /*
2491  * request arguments may be specified via an inode *, a dentry *, or
2492  * an explicit ino+path.
2493  */
2494 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
2495                                   struct inode *rdiri, const char *rpath,
2496                                   u64 rino, const char **ppath, int *pathlen,
2497                                   u64 *ino, bool *freepath, bool parent_locked)
2498 {
2499         int r = 0;
2500
2501         if (rinode) {
2502                 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
2503                 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2504                      ceph_snap(rinode));
2505         } else if (rdentry) {
2506                 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
2507                                         freepath, parent_locked);
2508                 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
2509                      *ppath);
2510         } else if (rpath || rino) {
2511                 *ino = rino;
2512                 *ppath = rpath;
2513                 *pathlen = rpath ? strlen(rpath) : 0;
2514                 dout(" path %.*s\n", *pathlen, rpath);
2515         }
2516
2517         return r;
2518 }
2519
2520 static void encode_timestamp_and_gids(void **p,
2521                                       const struct ceph_mds_request *req)
2522 {
2523         struct ceph_timespec ts;
2524         int i;
2525
2526         ceph_encode_timespec64(&ts, &req->r_stamp);
2527         ceph_encode_copy(p, &ts, sizeof(ts));
2528
2529         /* gid_list */
2530         ceph_encode_32(p, req->r_cred->group_info->ngroups);
2531         for (i = 0; i < req->r_cred->group_info->ngroups; i++)
2532                 ceph_encode_64(p, from_kgid(&init_user_ns,
2533                                             req->r_cred->group_info->gid[i]));
2534 }
2535
2536 /*
2537  * called under mdsc->mutex
2538  */
2539 static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
2540                                                struct ceph_mds_request *req,
2541                                                bool drop_cap_releases)
2542 {
2543         int mds = session->s_mds;
2544         struct ceph_mds_client *mdsc = session->s_mdsc;
2545         struct ceph_msg *msg;
2546         struct ceph_mds_request_head_old *head;
2547         const char *path1 = NULL;
2548         const char *path2 = NULL;
2549         u64 ino1 = 0, ino2 = 0;
2550         int pathlen1 = 0, pathlen2 = 0;
2551         bool freepath1 = false, freepath2 = false;
2552         int len;
2553         u16 releases;
2554         void *p, *end;
2555         int ret;
2556         bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
2557
2558         ret = set_request_path_attr(req->r_inode, req->r_dentry,
2559                               req->r_parent, req->r_path1, req->r_ino1.ino,
2560                               &path1, &pathlen1, &ino1, &freepath1,
2561                               test_bit(CEPH_MDS_R_PARENT_LOCKED,
2562                                         &req->r_req_flags));
2563         if (ret < 0) {
2564                 msg = ERR_PTR(ret);
2565                 goto out;
2566         }
2567
2568         /* If r_old_dentry is set, then assume that its parent is locked */
2569         ret = set_request_path_attr(NULL, req->r_old_dentry,
2570                               req->r_old_dentry_dir,
2571                               req->r_path2, req->r_ino2.ino,
2572                               &path2, &pathlen2, &ino2, &freepath2, true);
2573         if (ret < 0) {
2574                 msg = ERR_PTR(ret);
2575                 goto out_free1;
2576         }
2577
2578         len = legacy ? sizeof(*head) : sizeof(struct ceph_mds_request_head);
2579         len += pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
2580                 sizeof(struct ceph_timespec);
2581         len += sizeof(u32) + (sizeof(u64) * req->r_cred->group_info->ngroups);
2582
2583         /* calculate (max) length for cap releases */
2584         len += sizeof(struct ceph_mds_request_release) *
2585                 (!!req->r_inode_drop + !!req->r_dentry_drop +
2586                  !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2587
2588         if (req->r_dentry_drop)
2589                 len += pathlen1;
2590         if (req->r_old_dentry_drop)
2591                 len += pathlen2;
2592
2593         msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
2594         if (!msg) {
2595                 msg = ERR_PTR(-ENOMEM);
2596                 goto out_free2;
2597         }
2598
2599         msg->hdr.tid = cpu_to_le64(req->r_tid);
2600
2601         /*
2602          * The old ceph_mds_request_head didn't contain a version field, and
2603          * one was added when we moved the message version from 3->4.
2604          */
2605         if (legacy) {
2606                 msg->hdr.version = cpu_to_le16(3);
2607                 head = msg->front.iov_base;
2608                 p = msg->front.iov_base + sizeof(*head);
2609         } else {
2610                 struct ceph_mds_request_head *new_head = msg->front.iov_base;
2611
2612                 msg->hdr.version = cpu_to_le16(4);
2613                 new_head->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
2614                 head = (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2615                 p = msg->front.iov_base + sizeof(*new_head);
2616         }
2617
2618         end = msg->front.iov_base + msg->front.iov_len;
2619
2620         head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2621         head->op = cpu_to_le32(req->r_op);
2622         head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns,
2623                                                  req->r_cred->fsuid));
2624         head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns,
2625                                                  req->r_cred->fsgid));
2626         head->ino = cpu_to_le64(req->r_deleg_ino);
2627         head->args = req->r_args;
2628
2629         ceph_encode_filepath(&p, end, ino1, path1);
2630         ceph_encode_filepath(&p, end, ino2, path2);
2631
2632         /* make note of release offset, in case we need to replay */
2633         req->r_request_release_offset = p - msg->front.iov_base;
2634
2635         /* cap releases */
2636         releases = 0;
2637         if (req->r_inode_drop)
2638                 releases += ceph_encode_inode_release(&p,
2639                       req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2640                       mds, req->r_inode_drop, req->r_inode_unless,
2641                       req->r_op == CEPH_MDS_OP_READDIR);
2642         if (req->r_dentry_drop)
2643                 releases += ceph_encode_dentry_release(&p, req->r_dentry,
2644                                 req->r_parent, mds, req->r_dentry_drop,
2645                                 req->r_dentry_unless);
2646         if (req->r_old_dentry_drop)
2647                 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2648                                 req->r_old_dentry_dir, mds,
2649                                 req->r_old_dentry_drop,
2650                                 req->r_old_dentry_unless);
2651         if (req->r_old_inode_drop)
2652                 releases += ceph_encode_inode_release(&p,
2653                       d_inode(req->r_old_dentry),
2654                       mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
2655
2656         if (drop_cap_releases) {
2657                 releases = 0;
2658                 p = msg->front.iov_base + req->r_request_release_offset;
2659         }
2660
2661         head->num_releases = cpu_to_le16(releases);
2662
2663         encode_timestamp_and_gids(&p, req);
2664
2665         if (WARN_ON_ONCE(p > end)) {
2666                 ceph_msg_put(msg);
2667                 msg = ERR_PTR(-ERANGE);
2668                 goto out_free2;
2669         }
2670
2671         msg->front.iov_len = p - msg->front.iov_base;
2672         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2673
2674         if (req->r_pagelist) {
2675                 struct ceph_pagelist *pagelist = req->r_pagelist;
2676                 ceph_msg_data_add_pagelist(msg, pagelist);
2677                 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2678         } else {
2679                 msg->hdr.data_len = 0;
2680         }
2681
2682         msg->hdr.data_off = cpu_to_le16(0);
2683
2684 out_free2:
2685         if (freepath2)
2686                 ceph_mdsc_free_path((char *)path2, pathlen2);
2687 out_free1:
2688         if (freepath1)
2689                 ceph_mdsc_free_path((char *)path1, pathlen1);
2690 out:
2691         return msg;
2692 }
2693
2694 /*
2695  * called under mdsc->mutex if error, under no mutex if
2696  * success.
2697  */
2698 static void complete_request(struct ceph_mds_client *mdsc,
2699                              struct ceph_mds_request *req)
2700 {
2701         req->r_end_latency = ktime_get();
2702
2703         if (req->r_callback)
2704                 req->r_callback(mdsc, req);
2705         complete_all(&req->r_completion);
2706 }
2707
2708 static struct ceph_mds_request_head_old *
2709 find_old_request_head(void *p, u64 features)
2710 {
2711         bool legacy = !(features & CEPH_FEATURE_FS_BTIME);
2712         struct ceph_mds_request_head *new_head;
2713
2714         if (legacy)
2715                 return (struct ceph_mds_request_head_old *)p;
2716         new_head = (struct ceph_mds_request_head *)p;
2717         return (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2718 }
2719
2720 /*
2721  * called under mdsc->mutex
2722  */
2723 static int __prepare_send_request(struct ceph_mds_session *session,
2724                                   struct ceph_mds_request *req,
2725                                   bool drop_cap_releases)
2726 {
2727         int mds = session->s_mds;
2728         struct ceph_mds_client *mdsc = session->s_mdsc;
2729         struct ceph_mds_request_head_old *rhead;
2730         struct ceph_msg *msg;
2731         int flags = 0, max_retry;
2732
2733         /*
2734          * The type of 'r_attempts' in kernel 'ceph_mds_request'
2735          * is 'int', while in 'ceph_mds_request_head' the type of
2736          * 'num_retry' is '__u8'. So in case the request retries
2737          *  exceeding 256 times, the MDS will receive a incorrect
2738          *  retry seq.
2739          *
2740          * In this case it's ususally a bug in MDS and continue
2741          * retrying the request makes no sense.
2742          *
2743          * In future this could be fixed in ceph code, so avoid
2744          * using the hardcode here.
2745          */
2746         max_retry = sizeof_field(struct ceph_mds_request_head, num_retry);
2747         max_retry = 1 << (max_retry * BITS_PER_BYTE);
2748         if (req->r_attempts >= max_retry) {
2749                 pr_warn_ratelimited("%s request tid %llu seq overflow\n",
2750                                     __func__, req->r_tid);
2751                 return -EMULTIHOP;
2752         }
2753
2754         req->r_attempts++;
2755         if (req->r_inode) {
2756                 struct ceph_cap *cap =
2757                         ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2758
2759                 if (cap)
2760                         req->r_sent_on_mseq = cap->mseq;
2761                 else
2762                         req->r_sent_on_mseq = -1;
2763         }
2764         dout("%s %p tid %lld %s (attempt %d)\n", __func__, req,
2765              req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2766
2767         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2768                 void *p;
2769
2770                 /*
2771                  * Replay.  Do not regenerate message (and rebuild
2772                  * paths, etc.); just use the original message.
2773                  * Rebuilding paths will break for renames because
2774                  * d_move mangles the src name.
2775                  */
2776                 msg = req->r_request;
2777                 rhead = find_old_request_head(msg->front.iov_base,
2778                                               session->s_con.peer_features);
2779
2780                 flags = le32_to_cpu(rhead->flags);
2781                 flags |= CEPH_MDS_FLAG_REPLAY;
2782                 rhead->flags = cpu_to_le32(flags);
2783
2784                 if (req->r_target_inode)
2785                         rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2786
2787                 rhead->num_retry = req->r_attempts - 1;
2788
2789                 /* remove cap/dentry releases from message */
2790                 rhead->num_releases = 0;
2791
2792                 p = msg->front.iov_base + req->r_request_release_offset;
2793                 encode_timestamp_and_gids(&p, req);
2794
2795                 msg->front.iov_len = p - msg->front.iov_base;
2796                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2797                 return 0;
2798         }
2799
2800         if (req->r_request) {
2801                 ceph_msg_put(req->r_request);
2802                 req->r_request = NULL;
2803         }
2804         msg = create_request_message(session, req, drop_cap_releases);
2805         if (IS_ERR(msg)) {
2806                 req->r_err = PTR_ERR(msg);
2807                 return PTR_ERR(msg);
2808         }
2809         req->r_request = msg;
2810
2811         rhead = find_old_request_head(msg->front.iov_base,
2812                                       session->s_con.peer_features);
2813         rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2814         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2815                 flags |= CEPH_MDS_FLAG_REPLAY;
2816         if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags))
2817                 flags |= CEPH_MDS_FLAG_ASYNC;
2818         if (req->r_parent)
2819                 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2820         rhead->flags = cpu_to_le32(flags);
2821         rhead->num_fwd = req->r_num_fwd;
2822         rhead->num_retry = req->r_attempts - 1;
2823
2824         dout(" r_parent = %p\n", req->r_parent);
2825         return 0;
2826 }
2827
2828 /*
2829  * called under mdsc->mutex
2830  */
2831 static int __send_request(struct ceph_mds_session *session,
2832                           struct ceph_mds_request *req,
2833                           bool drop_cap_releases)
2834 {
2835         int err;
2836
2837         err = __prepare_send_request(session, req, drop_cap_releases);
2838         if (!err) {
2839                 ceph_msg_get(req->r_request);
2840                 ceph_con_send(&session->s_con, req->r_request);
2841         }
2842
2843         return err;
2844 }
2845
2846 /*
2847  * send request, or put it on the appropriate wait list.
2848  */
2849 static void __do_request(struct ceph_mds_client *mdsc,
2850                         struct ceph_mds_request *req)
2851 {
2852         struct ceph_mds_session *session = NULL;
2853         int mds = -1;
2854         int err = 0;
2855         bool random;
2856
2857         if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2858                 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
2859                         __unregister_request(mdsc, req);
2860                 return;
2861         }
2862
2863         if (req->r_timeout &&
2864             time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2865                 dout("do_request timed out\n");
2866                 err = -ETIMEDOUT;
2867                 goto finish;
2868         }
2869         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2870                 dout("do_request forced umount\n");
2871                 err = -EIO;
2872                 goto finish;
2873         }
2874         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2875                 if (mdsc->mdsmap_err) {
2876                         err = mdsc->mdsmap_err;
2877                         dout("do_request mdsmap err %d\n", err);
2878                         goto finish;
2879                 }
2880                 if (mdsc->mdsmap->m_epoch == 0) {
2881                         dout("do_request no mdsmap, waiting for map\n");
2882                         list_add(&req->r_wait, &mdsc->waiting_for_map);
2883                         return;
2884                 }
2885                 if (!(mdsc->fsc->mount_options->flags &
2886                       CEPH_MOUNT_OPT_MOUNTWAIT) &&
2887                     !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2888                         err = -EHOSTUNREACH;
2889                         goto finish;
2890                 }
2891         }
2892
2893         put_request_session(req);
2894
2895         mds = __choose_mds(mdsc, req, &random);
2896         if (mds < 0 ||
2897             ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2898                 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2899                         err = -EJUKEBOX;
2900                         goto finish;
2901                 }
2902                 dout("do_request no mds or not active, waiting for map\n");
2903                 list_add(&req->r_wait, &mdsc->waiting_for_map);
2904                 return;
2905         }
2906
2907         /* get, open session */
2908         session = __ceph_lookup_mds_session(mdsc, mds);
2909         if (!session) {
2910                 session = register_session(mdsc, mds);
2911                 if (IS_ERR(session)) {
2912                         err = PTR_ERR(session);
2913                         goto finish;
2914                 }
2915         }
2916         req->r_session = ceph_get_mds_session(session);
2917
2918         dout("do_request mds%d session %p state %s\n", mds, session,
2919              ceph_session_state_name(session->s_state));
2920
2921         /*
2922          * The old ceph will crash the MDSs when see unknown OPs
2923          */
2924         if (req->r_feature_needed > 0 &&
2925             !test_bit(req->r_feature_needed, &session->s_features)) {
2926                 err = -EOPNOTSUPP;
2927                 goto out_session;
2928         }
2929
2930         if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2931             session->s_state != CEPH_MDS_SESSION_HUNG) {
2932                 /*
2933                  * We cannot queue async requests since the caps and delegated
2934                  * inodes are bound to the session. Just return -EJUKEBOX and
2935                  * let the caller retry a sync request in that case.
2936                  */
2937                 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2938                         err = -EJUKEBOX;
2939                         goto out_session;
2940                 }
2941
2942                 /*
2943                  * If the session has been REJECTED, then return a hard error,
2944                  * unless it's a CLEANRECOVER mount, in which case we'll queue
2945                  * it to the mdsc queue.
2946                  */
2947                 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2948                         if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
2949                                 list_add(&req->r_wait, &mdsc->waiting_for_map);
2950                         else
2951                                 err = -EACCES;
2952                         goto out_session;
2953                 }
2954
2955                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2956                     session->s_state == CEPH_MDS_SESSION_CLOSING) {
2957                         err = __open_session(mdsc, session);
2958                         if (err)
2959                                 goto out_session;
2960                         /* retry the same mds later */
2961                         if (random)
2962                                 req->r_resend_mds = mds;
2963                 }
2964                 list_add(&req->r_wait, &session->s_waiting);
2965                 goto out_session;
2966         }
2967
2968         /* send request */
2969         req->r_resend_mds = -1;   /* forget any previous mds hint */
2970
2971         if (req->r_request_started == 0)   /* note request start time */
2972                 req->r_request_started = jiffies;
2973
2974         /*
2975          * For async create we will choose the auth MDS of frag in parent
2976          * directory to send the request and ususally this works fine, but
2977          * if the migrated the dirtory to another MDS before it could handle
2978          * it the request will be forwarded.
2979          *
2980          * And then the auth cap will be changed.
2981          */
2982         if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && req->r_num_fwd) {
2983                 struct ceph_dentry_info *di = ceph_dentry(req->r_dentry);
2984                 struct ceph_inode_info *ci;
2985                 struct ceph_cap *cap;
2986
2987                 /*
2988                  * The request maybe handled very fast and the new inode
2989                  * hasn't been linked to the dentry yet. We need to wait
2990                  * for the ceph_finish_async_create(), which shouldn't be
2991                  * stuck too long or fail in thoery, to finish when forwarding
2992                  * the request.
2993                  */
2994                 if (!d_inode(req->r_dentry)) {
2995                         err = wait_on_bit(&di->flags, CEPH_DENTRY_ASYNC_CREATE_BIT,
2996                                           TASK_KILLABLE);
2997                         if (err) {
2998                                 mutex_lock(&req->r_fill_mutex);
2999                                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3000                                 mutex_unlock(&req->r_fill_mutex);
3001                                 goto out_session;
3002                         }
3003                 }
3004
3005                 ci = ceph_inode(d_inode(req->r_dentry));
3006
3007                 spin_lock(&ci->i_ceph_lock);
3008                 cap = ci->i_auth_cap;
3009                 if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE && mds != cap->mds) {
3010                         dout("do_request session changed for auth cap %d -> %d\n",
3011                              cap->session->s_mds, session->s_mds);
3012
3013                         /* Remove the auth cap from old session */
3014                         spin_lock(&cap->session->s_cap_lock);
3015                         cap->session->s_nr_caps--;
3016                         list_del_init(&cap->session_caps);
3017                         spin_unlock(&cap->session->s_cap_lock);
3018
3019                         /* Add the auth cap to the new session */
3020                         cap->mds = mds;
3021                         cap->session = session;
3022                         spin_lock(&session->s_cap_lock);
3023                         session->s_nr_caps++;
3024                         list_add_tail(&cap->session_caps, &session->s_caps);
3025                         spin_unlock(&session->s_cap_lock);
3026
3027                         change_auth_cap_ses(ci, session);
3028                 }
3029                 spin_unlock(&ci->i_ceph_lock);
3030         }
3031
3032         err = __send_request(session, req, false);
3033
3034 out_session:
3035         ceph_put_mds_session(session);
3036 finish:
3037         if (err) {
3038                 dout("__do_request early error %d\n", err);
3039                 req->r_err = err;
3040                 complete_request(mdsc, req);
3041                 __unregister_request(mdsc, req);
3042         }
3043         return;
3044 }
3045
3046 /*
3047  * called under mdsc->mutex
3048  */
3049 static void __wake_requests(struct ceph_mds_client *mdsc,
3050                             struct list_head *head)
3051 {
3052         struct ceph_mds_request *req;
3053         LIST_HEAD(tmp_list);
3054
3055         list_splice_init(head, &tmp_list);
3056
3057         while (!list_empty(&tmp_list)) {
3058                 req = list_entry(tmp_list.next,
3059                                  struct ceph_mds_request, r_wait);
3060                 list_del_init(&req->r_wait);
3061                 dout(" wake request %p tid %llu\n", req, req->r_tid);
3062                 __do_request(mdsc, req);
3063         }
3064 }
3065
3066 /*
3067  * Wake up threads with requests pending for @mds, so that they can
3068  * resubmit their requests to a possibly different mds.
3069  */
3070 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
3071 {
3072         struct ceph_mds_request *req;
3073         struct rb_node *p = rb_first(&mdsc->request_tree);
3074
3075         dout("kick_requests mds%d\n", mds);
3076         while (p) {
3077                 req = rb_entry(p, struct ceph_mds_request, r_node);
3078                 p = rb_next(p);
3079                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3080                         continue;
3081                 if (req->r_attempts > 0)
3082                         continue; /* only new requests */
3083                 if (req->r_session &&
3084                     req->r_session->s_mds == mds) {
3085                         dout(" kicking tid %llu\n", req->r_tid);
3086                         list_del_init(&req->r_wait);
3087                         __do_request(mdsc, req);
3088                 }
3089         }
3090 }
3091
3092 int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
3093                               struct ceph_mds_request *req)
3094 {
3095         int err = 0;
3096
3097         /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
3098         if (req->r_inode)
3099                 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3100         if (req->r_parent) {
3101                 struct ceph_inode_info *ci = ceph_inode(req->r_parent);
3102                 int fmode = (req->r_op & CEPH_MDS_OP_WRITE) ?
3103                             CEPH_FILE_MODE_WR : CEPH_FILE_MODE_RD;
3104                 spin_lock(&ci->i_ceph_lock);
3105                 ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
3106                 __ceph_touch_fmode(ci, mdsc, fmode);
3107                 spin_unlock(&ci->i_ceph_lock);
3108         }
3109         if (req->r_old_dentry_dir)
3110                 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
3111                                   CEPH_CAP_PIN);
3112
3113         if (req->r_inode) {
3114                 err = ceph_wait_on_async_create(req->r_inode);
3115                 if (err) {
3116                         dout("%s: wait for async create returned: %d\n",
3117                              __func__, err);
3118                         return err;
3119                 }
3120         }
3121
3122         if (!err && req->r_old_inode) {
3123                 err = ceph_wait_on_async_create(req->r_old_inode);
3124                 if (err) {
3125                         dout("%s: wait for async create returned: %d\n",
3126                              __func__, err);
3127                         return err;
3128                 }
3129         }
3130
3131         dout("submit_request on %p for inode %p\n", req, dir);
3132         mutex_lock(&mdsc->mutex);
3133         __register_request(mdsc, req, dir);
3134         __do_request(mdsc, req);
3135         err = req->r_err;
3136         mutex_unlock(&mdsc->mutex);
3137         return err;
3138 }
3139
3140 int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
3141                            struct ceph_mds_request *req,
3142                            ceph_mds_request_wait_callback_t wait_func)
3143 {
3144         int err;
3145
3146         /* wait */
3147         dout("do_request waiting\n");
3148         if (wait_func) {
3149                 err = wait_func(mdsc, req);
3150         } else {
3151                 long timeleft = wait_for_completion_killable_timeout(
3152                                         &req->r_completion,
3153                                         ceph_timeout_jiffies(req->r_timeout));
3154                 if (timeleft > 0)
3155                         err = 0;
3156                 else if (!timeleft)
3157                         err = -ETIMEDOUT;  /* timed out */
3158                 else
3159                         err = timeleft;  /* killed */
3160         }
3161         dout("do_request waited, got %d\n", err);
3162         mutex_lock(&mdsc->mutex);
3163
3164         /* only abort if we didn't race with a real reply */
3165         if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
3166                 err = le32_to_cpu(req->r_reply_info.head->result);
3167         } else if (err < 0) {
3168                 dout("aborted request %lld with %d\n", req->r_tid, err);
3169
3170                 /*
3171                  * ensure we aren't running concurrently with
3172                  * ceph_fill_trace or ceph_readdir_prepopulate, which
3173                  * rely on locks (dir mutex) held by our caller.
3174                  */
3175                 mutex_lock(&req->r_fill_mutex);
3176                 req->r_err = err;
3177                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3178                 mutex_unlock(&req->r_fill_mutex);
3179
3180                 if (req->r_parent &&
3181                     (req->r_op & CEPH_MDS_OP_WRITE))
3182                         ceph_invalidate_dir_request(req);
3183         } else {
3184                 err = req->r_err;
3185         }
3186
3187         mutex_unlock(&mdsc->mutex);
3188         return err;
3189 }
3190
3191 /*
3192  * Synchrously perform an mds request.  Take care of all of the
3193  * session setup, forwarding, retry details.
3194  */
3195 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
3196                          struct inode *dir,
3197                          struct ceph_mds_request *req)
3198 {
3199         int err;
3200
3201         dout("do_request on %p\n", req);
3202
3203         /* issue */
3204         err = ceph_mdsc_submit_request(mdsc, dir, req);
3205         if (!err)
3206                 err = ceph_mdsc_wait_request(mdsc, req, NULL);
3207         dout("do_request %p done, result %d\n", req, err);
3208         return err;
3209 }
3210
3211 /*
3212  * Invalidate dir's completeness, dentry lease state on an aborted MDS
3213  * namespace request.
3214  */
3215 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
3216 {
3217         struct inode *dir = req->r_parent;
3218         struct inode *old_dir = req->r_old_dentry_dir;
3219
3220         dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
3221
3222         ceph_dir_clear_complete(dir);
3223         if (old_dir)
3224                 ceph_dir_clear_complete(old_dir);
3225         if (req->r_dentry)
3226                 ceph_invalidate_dentry_lease(req->r_dentry);
3227         if (req->r_old_dentry)
3228                 ceph_invalidate_dentry_lease(req->r_old_dentry);
3229 }
3230
3231 /*
3232  * Handle mds reply.
3233  *
3234  * We take the session mutex and parse and process the reply immediately.
3235  * This preserves the logical ordering of replies, capabilities, etc., sent
3236  * by the MDS as they are applied to our local cache.
3237  */
3238 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
3239 {
3240         struct ceph_mds_client *mdsc = session->s_mdsc;
3241         struct ceph_mds_request *req;
3242         struct ceph_mds_reply_head *head = msg->front.iov_base;
3243         struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
3244         struct ceph_snap_realm *realm;
3245         u64 tid;
3246         int err, result;
3247         int mds = session->s_mds;
3248
3249         if (msg->front.iov_len < sizeof(*head)) {
3250                 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
3251                 ceph_msg_dump(msg);
3252                 return;
3253         }
3254
3255         /* get request, session */
3256         tid = le64_to_cpu(msg->hdr.tid);
3257         mutex_lock(&mdsc->mutex);
3258         req = lookup_get_request(mdsc, tid);
3259         if (!req) {
3260                 dout("handle_reply on unknown tid %llu\n", tid);
3261                 mutex_unlock(&mdsc->mutex);
3262                 return;
3263         }
3264         dout("handle_reply %p\n", req);
3265
3266         /* correct session? */
3267         if (req->r_session != session) {
3268                 pr_err("mdsc_handle_reply got %llu on session mds%d"
3269                        " not mds%d\n", tid, session->s_mds,
3270                        req->r_session ? req->r_session->s_mds : -1);
3271                 mutex_unlock(&mdsc->mutex);
3272                 goto out;
3273         }
3274
3275         /* dup? */
3276         if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
3277             (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
3278                 pr_warn("got a dup %s reply on %llu from mds%d\n",
3279                            head->safe ? "safe" : "unsafe", tid, mds);
3280                 mutex_unlock(&mdsc->mutex);
3281                 goto out;
3282         }
3283         if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
3284                 pr_warn("got unsafe after safe on %llu from mds%d\n",
3285                            tid, mds);
3286                 mutex_unlock(&mdsc->mutex);
3287                 goto out;
3288         }
3289
3290         result = le32_to_cpu(head->result);
3291
3292         if (head->safe) {
3293                 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
3294                 __unregister_request(mdsc, req);
3295
3296                 /* last request during umount? */
3297                 if (mdsc->stopping && !__get_oldest_req(mdsc))
3298                         complete_all(&mdsc->safe_umount_waiters);
3299
3300                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3301                         /*
3302                          * We already handled the unsafe response, now do the
3303                          * cleanup.  No need to examine the response; the MDS
3304                          * doesn't include any result info in the safe
3305                          * response.  And even if it did, there is nothing
3306                          * useful we could do with a revised return value.
3307                          */
3308                         dout("got safe reply %llu, mds%d\n", tid, mds);
3309
3310                         mutex_unlock(&mdsc->mutex);
3311                         goto out;
3312                 }
3313         } else {
3314                 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
3315                 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
3316         }
3317
3318         dout("handle_reply tid %lld result %d\n", tid, result);
3319         rinfo = &req->r_reply_info;
3320         if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
3321                 err = parse_reply_info(session, msg, rinfo, (u64)-1);
3322         else
3323                 err = parse_reply_info(session, msg, rinfo, session->s_con.peer_features);
3324         mutex_unlock(&mdsc->mutex);
3325
3326         /* Must find target inode outside of mutexes to avoid deadlocks */
3327         if ((err >= 0) && rinfo->head->is_target) {
3328                 struct inode *in;
3329                 struct ceph_vino tvino = {
3330                         .ino  = le64_to_cpu(rinfo->targeti.in->ino),
3331                         .snap = le64_to_cpu(rinfo->targeti.in->snapid)
3332                 };
3333
3334                 in = ceph_get_inode(mdsc->fsc->sb, tvino);
3335                 if (IS_ERR(in)) {
3336                         err = PTR_ERR(in);
3337                         mutex_lock(&session->s_mutex);
3338                         goto out_err;
3339                 }
3340                 req->r_target_inode = in;
3341         }
3342
3343         mutex_lock(&session->s_mutex);
3344         if (err < 0) {
3345                 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
3346                 ceph_msg_dump(msg);
3347                 goto out_err;
3348         }
3349
3350         /* snap trace */
3351         realm = NULL;
3352         if (rinfo->snapblob_len) {
3353                 down_write(&mdsc->snap_rwsem);
3354                 ceph_update_snap_trace(mdsc, rinfo->snapblob,
3355                                 rinfo->snapblob + rinfo->snapblob_len,
3356                                 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
3357                                 &realm);
3358                 downgrade_write(&mdsc->snap_rwsem);
3359         } else {
3360                 down_read(&mdsc->snap_rwsem);
3361         }
3362
3363         /* insert trace into our cache */
3364         mutex_lock(&req->r_fill_mutex);
3365         current->journal_info = req;
3366         err = ceph_fill_trace(mdsc->fsc->sb, req);
3367         if (err == 0) {
3368                 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
3369                                     req->r_op == CEPH_MDS_OP_LSSNAP))
3370                         ceph_readdir_prepopulate(req, req->r_session);
3371         }
3372         current->journal_info = NULL;
3373         mutex_unlock(&req->r_fill_mutex);
3374
3375         up_read(&mdsc->snap_rwsem);
3376         if (realm)
3377                 ceph_put_snap_realm(mdsc, realm);
3378
3379         if (err == 0) {
3380                 if (req->r_target_inode &&
3381                     test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3382                         struct ceph_inode_info *ci =
3383                                 ceph_inode(req->r_target_inode);
3384                         spin_lock(&ci->i_unsafe_lock);
3385                         list_add_tail(&req->r_unsafe_target_item,
3386                                       &ci->i_unsafe_iops);
3387                         spin_unlock(&ci->i_unsafe_lock);
3388                 }
3389
3390                 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
3391         }
3392 out_err:
3393         mutex_lock(&mdsc->mutex);
3394         if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3395                 if (err) {
3396                         req->r_err = err;
3397                 } else {
3398                         req->r_reply =  ceph_msg_get(msg);
3399                         set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
3400                 }
3401         } else {
3402                 dout("reply arrived after request %lld was aborted\n", tid);
3403         }
3404         mutex_unlock(&mdsc->mutex);
3405
3406         mutex_unlock(&session->s_mutex);
3407
3408         /* kick calling process */
3409         complete_request(mdsc, req);
3410
3411         ceph_update_metadata_metrics(&mdsc->metric, req->r_start_latency,
3412                                      req->r_end_latency, err);
3413 out:
3414         ceph_mdsc_put_request(req);
3415         return;
3416 }
3417
3418
3419
3420 /*
3421  * handle mds notification that our request has been forwarded.
3422  */
3423 static void handle_forward(struct ceph_mds_client *mdsc,
3424                            struct ceph_mds_session *session,
3425                            struct ceph_msg *msg)
3426 {
3427         struct ceph_mds_request *req;
3428         u64 tid = le64_to_cpu(msg->hdr.tid);
3429         u32 next_mds;
3430         u32 fwd_seq;
3431         int err = -EINVAL;
3432         void *p = msg->front.iov_base;
3433         void *end = p + msg->front.iov_len;
3434         bool aborted = false;
3435
3436         ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3437         next_mds = ceph_decode_32(&p);
3438         fwd_seq = ceph_decode_32(&p);
3439
3440         mutex_lock(&mdsc->mutex);
3441         req = lookup_get_request(mdsc, tid);
3442         if (!req) {
3443                 mutex_unlock(&mdsc->mutex);
3444                 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
3445                 return;  /* dup reply? */
3446         }
3447
3448         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3449                 dout("forward tid %llu aborted, unregistering\n", tid);
3450                 __unregister_request(mdsc, req);
3451         } else if (fwd_seq <= req->r_num_fwd) {
3452                 /*
3453                  * The type of 'num_fwd' in ceph 'MClientRequestForward'
3454                  * is 'int32_t', while in 'ceph_mds_request_head' the
3455                  * type is '__u8'. So in case the request bounces between
3456                  * MDSes exceeding 256 times, the client will get stuck.
3457                  *
3458                  * In this case it's ususally a bug in MDS and continue
3459                  * bouncing the request makes no sense.
3460                  *
3461                  * In future this could be fixed in ceph code, so avoid
3462                  * using the hardcode here.
3463                  */
3464                 int max = sizeof_field(struct ceph_mds_request_head, num_fwd);
3465                 max = 1 << (max * BITS_PER_BYTE);
3466                 if (req->r_num_fwd >= max) {
3467                         mutex_lock(&req->r_fill_mutex);
3468                         req->r_err = -EMULTIHOP;
3469                         set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3470                         mutex_unlock(&req->r_fill_mutex);
3471                         aborted = true;
3472                         pr_warn_ratelimited("forward tid %llu seq overflow\n",
3473                                             tid);
3474                 } else {
3475                         dout("forward tid %llu to mds%d - old seq %d <= %d\n",
3476                              tid, next_mds, req->r_num_fwd, fwd_seq);
3477                 }
3478         } else {
3479                 /* resend. forward race not possible; mds would drop */
3480                 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
3481                 BUG_ON(req->r_err);
3482                 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3483                 req->r_attempts = 0;
3484                 req->r_num_fwd = fwd_seq;
3485                 req->r_resend_mds = next_mds;
3486                 put_request_session(req);
3487                 __do_request(mdsc, req);
3488         }
3489         mutex_unlock(&mdsc->mutex);
3490
3491         /* kick calling process */
3492         if (aborted)
3493                 complete_request(mdsc, req);
3494         ceph_mdsc_put_request(req);
3495         return;
3496
3497 bad:
3498         pr_err("mdsc_handle_forward decode error err=%d\n", err);
3499 }
3500
3501 static int __decode_session_metadata(void **p, void *end,
3502                                      bool *blocklisted)
3503 {
3504         /* map<string,string> */
3505         u32 n;
3506         bool err_str;
3507         ceph_decode_32_safe(p, end, n, bad);
3508         while (n-- > 0) {
3509                 u32 len;
3510                 ceph_decode_32_safe(p, end, len, bad);
3511                 ceph_decode_need(p, end, len, bad);
3512                 err_str = !strncmp(*p, "error_string", len);
3513                 *p += len;
3514                 ceph_decode_32_safe(p, end, len, bad);
3515                 ceph_decode_need(p, end, len, bad);
3516                 /*
3517                  * Match "blocklisted (blacklisted)" from newer MDSes,
3518                  * or "blacklisted" from older MDSes.
3519                  */
3520                 if (err_str && strnstr(*p, "blacklisted", len))
3521                         *blocklisted = true;
3522                 *p += len;
3523         }
3524         return 0;
3525 bad:
3526         return -1;
3527 }
3528
3529 /*
3530  * handle a mds session control message
3531  */
3532 static void handle_session(struct ceph_mds_session *session,
3533                            struct ceph_msg *msg)
3534 {
3535         struct ceph_mds_client *mdsc = session->s_mdsc;
3536         int mds = session->s_mds;
3537         int msg_version = le16_to_cpu(msg->hdr.version);
3538         void *p = msg->front.iov_base;
3539         void *end = p + msg->front.iov_len;
3540         struct ceph_mds_session_head *h;
3541         u32 op;
3542         u64 seq, features = 0;
3543         int wake = 0;
3544         bool blocklisted = false;
3545
3546         /* decode */
3547         ceph_decode_need(&p, end, sizeof(*h), bad);
3548         h = p;
3549         p += sizeof(*h);
3550
3551         op = le32_to_cpu(h->op);
3552         seq = le64_to_cpu(h->seq);
3553
3554         if (msg_version >= 3) {
3555                 u32 len;
3556                 /* version >= 2 and < 5, decode metadata, skip otherwise
3557                  * as it's handled via flags.
3558                  */
3559                 if (msg_version >= 5)
3560                         ceph_decode_skip_map(&p, end, string, string, bad);
3561                 else if (__decode_session_metadata(&p, end, &blocklisted) < 0)
3562                         goto bad;
3563
3564                 /* version >= 3, feature bits */
3565                 ceph_decode_32_safe(&p, end, len, bad);
3566                 if (len) {
3567                         ceph_decode_64_safe(&p, end, features, bad);
3568                         p += len - sizeof(features);
3569                 }
3570         }
3571
3572         if (msg_version >= 5) {
3573                 u32 flags, len;
3574
3575                 /* version >= 4 */
3576                 ceph_decode_skip_16(&p, end, bad); /* struct_v, struct_cv */
3577                 ceph_decode_32_safe(&p, end, len, bad); /* len */
3578                 ceph_decode_skip_n(&p, end, len, bad); /* metric_spec */
3579
3580                 /* version >= 5, flags   */
3581                 ceph_decode_32_safe(&p, end, flags, bad);
3582                 if (flags & CEPH_SESSION_BLOCKLISTED) {
3583                         pr_warn("mds%d session blocklisted\n", session->s_mds);
3584                         blocklisted = true;
3585                 }
3586         }
3587
3588         mutex_lock(&mdsc->mutex);
3589         if (op == CEPH_SESSION_CLOSE) {
3590                 ceph_get_mds_session(session);
3591                 __unregister_session(mdsc, session);
3592         }
3593         /* FIXME: this ttl calculation is generous */
3594         session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
3595         mutex_unlock(&mdsc->mutex);
3596
3597         mutex_lock(&session->s_mutex);
3598
3599         dout("handle_session mds%d %s %p state %s seq %llu\n",
3600              mds, ceph_session_op_name(op), session,
3601              ceph_session_state_name(session->s_state), seq);
3602
3603         if (session->s_state == CEPH_MDS_SESSION_HUNG) {
3604                 session->s_state = CEPH_MDS_SESSION_OPEN;
3605                 pr_info("mds%d came back\n", session->s_mds);
3606         }
3607
3608         switch (op) {
3609         case CEPH_SESSION_OPEN:
3610                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3611                         pr_info("mds%d reconnect success\n", session->s_mds);
3612
3613                 if (session->s_state == CEPH_MDS_SESSION_OPEN) {
3614                         pr_notice("mds%d is already opened\n", session->s_mds);
3615                 } else {
3616                         session->s_state = CEPH_MDS_SESSION_OPEN;
3617                         session->s_features = features;
3618                         renewed_caps(mdsc, session, 0);
3619                         if (test_bit(CEPHFS_FEATURE_METRIC_COLLECT,
3620                                      &session->s_features))
3621                                 metric_schedule_delayed(&mdsc->metric);
3622                 }
3623
3624                 /*
3625                  * The connection maybe broken and the session in client
3626                  * side has been reinitialized, need to update the seq
3627                  * anyway.
3628                  */
3629                 if (!session->s_seq && seq)
3630                         session->s_seq = seq;
3631
3632                 wake = 1;
3633                 if (mdsc->stopping)
3634                         __close_session(mdsc, session);
3635                 break;
3636
3637         case CEPH_SESSION_RENEWCAPS:
3638                 if (session->s_renew_seq == seq)
3639                         renewed_caps(mdsc, session, 1);
3640                 break;
3641
3642         case CEPH_SESSION_CLOSE:
3643                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3644                         pr_info("mds%d reconnect denied\n", session->s_mds);
3645                 session->s_state = CEPH_MDS_SESSION_CLOSED;
3646                 cleanup_session_requests(mdsc, session);
3647                 remove_session_caps(session);
3648                 wake = 2; /* for good measure */
3649                 wake_up_all(&mdsc->session_close_wq);
3650                 break;
3651
3652         case CEPH_SESSION_STALE:
3653                 pr_info("mds%d caps went stale, renewing\n",
3654                         session->s_mds);
3655                 atomic_inc(&session->s_cap_gen);
3656                 session->s_cap_ttl = jiffies - 1;
3657                 send_renew_caps(mdsc, session);
3658                 break;
3659
3660         case CEPH_SESSION_RECALL_STATE:
3661                 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
3662                 break;
3663
3664         case CEPH_SESSION_FLUSHMSG:
3665                 send_flushmsg_ack(mdsc, session, seq);
3666                 break;
3667
3668         case CEPH_SESSION_FORCE_RO:
3669                 dout("force_session_readonly %p\n", session);
3670                 spin_lock(&session->s_cap_lock);
3671                 session->s_readonly = true;
3672                 spin_unlock(&session->s_cap_lock);
3673                 wake_up_session_caps(session, FORCE_RO);
3674                 break;
3675
3676         case CEPH_SESSION_REJECT:
3677                 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
3678                 pr_info("mds%d rejected session\n", session->s_mds);
3679                 session->s_state = CEPH_MDS_SESSION_REJECTED;
3680                 cleanup_session_requests(mdsc, session);
3681                 remove_session_caps(session);
3682                 if (blocklisted)
3683                         mdsc->fsc->blocklisted = true;
3684                 wake = 2; /* for good measure */
3685                 break;
3686
3687         default:
3688                 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
3689                 WARN_ON(1);
3690         }
3691
3692         mutex_unlock(&session->s_mutex);
3693         if (wake) {
3694                 mutex_lock(&mdsc->mutex);
3695                 __wake_requests(mdsc, &session->s_waiting);
3696                 if (wake == 2)
3697                         kick_requests(mdsc, mds);
3698                 mutex_unlock(&mdsc->mutex);
3699         }
3700         if (op == CEPH_SESSION_CLOSE)
3701                 ceph_put_mds_session(session);
3702         return;
3703
3704 bad:
3705         pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
3706                (int)msg->front.iov_len);
3707         ceph_msg_dump(msg);
3708         return;
3709 }
3710
3711 void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
3712 {
3713         int dcaps;
3714
3715         dcaps = xchg(&req->r_dir_caps, 0);
3716         if (dcaps) {
3717                 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3718                 ceph_put_cap_refs(ceph_inode(req->r_parent), dcaps);
3719         }
3720 }
3721
3722 void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req)
3723 {
3724         int dcaps;
3725
3726         dcaps = xchg(&req->r_dir_caps, 0);
3727         if (dcaps) {
3728                 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3729                 ceph_put_cap_refs_no_check_caps(ceph_inode(req->r_parent),
3730                                                 dcaps);
3731         }
3732 }
3733
3734 /*
3735  * called under session->mutex.
3736  */
3737 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
3738                                    struct ceph_mds_session *session)
3739 {
3740         struct ceph_mds_request *req, *nreq;
3741         struct rb_node *p;
3742
3743         dout("replay_unsafe_requests mds%d\n", session->s_mds);
3744
3745         mutex_lock(&mdsc->mutex);
3746         list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item)
3747                 __send_request(session, req, true);
3748
3749         /*
3750          * also re-send old requests when MDS enters reconnect stage. So that MDS
3751          * can process completed request in clientreplay stage.
3752          */
3753         p = rb_first(&mdsc->request_tree);
3754         while (p) {
3755                 req = rb_entry(p, struct ceph_mds_request, r_node);
3756                 p = rb_next(p);
3757                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3758                         continue;
3759                 if (req->r_attempts == 0)
3760                         continue; /* only old requests */
3761                 if (!req->r_session)
3762                         continue;
3763                 if (req->r_session->s_mds != session->s_mds)
3764                         continue;
3765
3766                 ceph_mdsc_release_dir_caps_no_check(req);
3767
3768                 __send_request(session, req, true);
3769         }
3770         mutex_unlock(&mdsc->mutex);
3771 }
3772
3773 static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
3774 {
3775         struct ceph_msg *reply;
3776         struct ceph_pagelist *_pagelist;
3777         struct page *page;
3778         __le32 *addr;
3779         int err = -ENOMEM;
3780
3781         if (!recon_state->allow_multi)
3782                 return -ENOSPC;
3783
3784         /* can't handle message that contains both caps and realm */
3785         BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
3786
3787         /* pre-allocate new pagelist */
3788         _pagelist = ceph_pagelist_alloc(GFP_NOFS);
3789         if (!_pagelist)
3790                 return -ENOMEM;
3791
3792         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3793         if (!reply)
3794                 goto fail_msg;
3795
3796         /* placeholder for nr_caps */
3797         err = ceph_pagelist_encode_32(_pagelist, 0);
3798         if (err < 0)
3799                 goto fail;
3800
3801         if (recon_state->nr_caps) {
3802                 /* currently encoding caps */
3803                 err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
3804                 if (err)
3805                         goto fail;
3806         } else {
3807                 /* placeholder for nr_realms (currently encoding relams) */
3808                 err = ceph_pagelist_encode_32(_pagelist, 0);
3809                 if (err < 0)
3810                         goto fail;
3811         }
3812
3813         err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
3814         if (err)
3815                 goto fail;
3816
3817         page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
3818         addr = kmap_atomic(page);
3819         if (recon_state->nr_caps) {
3820                 /* currently encoding caps */
3821                 *addr = cpu_to_le32(recon_state->nr_caps);
3822         } else {
3823                 /* currently encoding relams */
3824                 *(addr + 1) = cpu_to_le32(recon_state->nr_realms);
3825         }
3826         kunmap_atomic(addr);
3827
3828         reply->hdr.version = cpu_to_le16(5);
3829         reply->hdr.compat_version = cpu_to_le16(4);
3830
3831         reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
3832         ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
3833
3834         ceph_con_send(&recon_state->session->s_con, reply);
3835         ceph_pagelist_release(recon_state->pagelist);
3836
3837         recon_state->pagelist = _pagelist;
3838         recon_state->nr_caps = 0;
3839         recon_state->nr_realms = 0;
3840         recon_state->msg_version = 5;
3841         return 0;
3842 fail:
3843         ceph_msg_put(reply);
3844 fail_msg:
3845         ceph_pagelist_release(_pagelist);
3846         return err;
3847 }
3848
3849 static struct dentry* d_find_primary(struct inode *inode)
3850 {
3851         struct dentry *alias, *dn = NULL;
3852
3853         if (hlist_empty(&inode->i_dentry))
3854                 return NULL;
3855
3856         spin_lock(&inode->i_lock);
3857         if (hlist_empty(&inode->i_dentry))
3858                 goto out_unlock;
3859
3860         if (S_ISDIR(inode->i_mode)) {
3861                 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
3862                 if (!IS_ROOT(alias))
3863                         dn = dget(alias);
3864                 goto out_unlock;
3865         }
3866
3867         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
3868                 spin_lock(&alias->d_lock);
3869                 if (!d_unhashed(alias) &&
3870                     (ceph_dentry(alias)->flags & CEPH_DENTRY_PRIMARY_LINK)) {
3871                         dn = dget_dlock(alias);
3872                 }
3873                 spin_unlock(&alias->d_lock);
3874                 if (dn)
3875                         break;
3876         }
3877 out_unlock:
3878         spin_unlock(&inode->i_lock);
3879         return dn;
3880 }
3881
3882 /*
3883  * Encode information about a cap for a reconnect with the MDS.
3884  */
3885 static int reconnect_caps_cb(struct inode *inode, struct ceph_cap *cap,
3886                           void *arg)
3887 {
3888         union {
3889                 struct ceph_mds_cap_reconnect v2;
3890                 struct ceph_mds_cap_reconnect_v1 v1;
3891         } rec;
3892         struct ceph_inode_info *ci = cap->ci;
3893         struct ceph_reconnect_state *recon_state = arg;
3894         struct ceph_pagelist *pagelist = recon_state->pagelist;
3895         struct dentry *dentry;
3896         char *path;
3897         int pathlen = 0, err;
3898         u64 pathbase;
3899         u64 snap_follows;
3900
3901         dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
3902              inode, ceph_vinop(inode), cap, cap->cap_id,
3903              ceph_cap_string(cap->issued));
3904
3905         dentry = d_find_primary(inode);
3906         if (dentry) {
3907                 /* set pathbase to parent dir when msg_version >= 2 */
3908                 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase,
3909                                             recon_state->msg_version >= 2);
3910                 dput(dentry);
3911                 if (IS_ERR(path)) {
3912                         err = PTR_ERR(path);
3913                         goto out_err;
3914                 }
3915         } else {
3916                 path = NULL;
3917                 pathbase = 0;
3918         }
3919
3920         spin_lock(&ci->i_ceph_lock);
3921         cap->seq = 0;        /* reset cap seq */
3922         cap->issue_seq = 0;  /* and issue_seq */
3923         cap->mseq = 0;       /* and migrate_seq */
3924         cap->cap_gen = atomic_read(&cap->session->s_cap_gen);
3925
3926         /* These are lost when the session goes away */
3927         if (S_ISDIR(inode->i_mode)) {
3928                 if (cap->issued & CEPH_CAP_DIR_CREATE) {
3929                         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
3930                         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
3931                 }
3932                 cap->issued &= ~CEPH_CAP_ANY_DIR_OPS;
3933         }
3934
3935         if (recon_state->msg_version >= 2) {
3936                 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
3937                 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3938                 rec.v2.issued = cpu_to_le32(cap->issued);
3939                 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3940                 rec.v2.pathbase = cpu_to_le64(pathbase);
3941                 rec.v2.flock_len = (__force __le32)
3942                         ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
3943         } else {
3944                 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
3945                 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3946                 rec.v1.issued = cpu_to_le32(cap->issued);
3947                 rec.v1.size = cpu_to_le64(i_size_read(inode));
3948                 ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
3949                 ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
3950                 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3951                 rec.v1.pathbase = cpu_to_le64(pathbase);
3952         }
3953
3954         if (list_empty(&ci->i_cap_snaps)) {
3955                 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3956         } else {
3957                 struct ceph_cap_snap *capsnap =
3958                         list_first_entry(&ci->i_cap_snaps,
3959                                          struct ceph_cap_snap, ci_item);
3960                 snap_follows = capsnap->follows;
3961         }
3962         spin_unlock(&ci->i_ceph_lock);
3963
3964         if (recon_state->msg_version >= 2) {
3965                 int num_fcntl_locks, num_flock_locks;
3966                 struct ceph_filelock *flocks = NULL;
3967                 size_t struct_len, total_len = sizeof(u64);
3968                 u8 struct_v = 0;
3969
3970 encode_again:
3971                 if (rec.v2.flock_len) {
3972                         ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
3973                 } else {
3974                         num_fcntl_locks = 0;
3975                         num_flock_locks = 0;
3976                 }
3977                 if (num_fcntl_locks + num_flock_locks > 0) {
3978                         flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
3979                                                sizeof(struct ceph_filelock),
3980                                                GFP_NOFS);
3981                         if (!flocks) {
3982                                 err = -ENOMEM;
3983                                 goto out_err;
3984                         }
3985                         err = ceph_encode_locks_to_buffer(inode, flocks,
3986                                                           num_fcntl_locks,
3987                                                           num_flock_locks);
3988                         if (err) {
3989                                 kfree(flocks);
3990                                 flocks = NULL;
3991                                 if (err == -ENOSPC)
3992                                         goto encode_again;
3993                                 goto out_err;
3994                         }
3995                 } else {
3996                         kfree(flocks);
3997                         flocks = NULL;
3998                 }
3999
4000                 if (recon_state->msg_version >= 3) {
4001                         /* version, compat_version and struct_len */
4002                         total_len += 2 * sizeof(u8) + sizeof(u32);
4003                         struct_v = 2;
4004                 }
4005                 /*
4006                  * number of encoded locks is stable, so copy to pagelist
4007                  */
4008                 struct_len = 2 * sizeof(u32) +
4009                             (num_fcntl_locks + num_flock_locks) *
4010                             sizeof(struct ceph_filelock);
4011                 rec.v2.flock_len = cpu_to_le32(struct_len);
4012
4013                 struct_len += sizeof(u32) + pathlen + sizeof(rec.v2);
4014
4015                 if (struct_v >= 2)
4016                         struct_len += sizeof(u64); /* snap_follows */
4017
4018                 total_len += struct_len;
4019
4020                 if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
4021                         err = send_reconnect_partial(recon_state);
4022                         if (err)
4023                                 goto out_freeflocks;
4024                         pagelist = recon_state->pagelist;
4025                 }
4026
4027                 err = ceph_pagelist_reserve(pagelist, total_len);
4028                 if (err)
4029                         goto out_freeflocks;
4030
4031                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
4032                 if (recon_state->msg_version >= 3) {
4033                         ceph_pagelist_encode_8(pagelist, struct_v);
4034                         ceph_pagelist_encode_8(pagelist, 1);
4035                         ceph_pagelist_encode_32(pagelist, struct_len);
4036                 }
4037                 ceph_pagelist_encode_string(pagelist, path, pathlen);
4038                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
4039                 ceph_locks_to_pagelist(flocks, pagelist,
4040                                        num_fcntl_locks, num_flock_locks);
4041                 if (struct_v >= 2)
4042                         ceph_pagelist_encode_64(pagelist, snap_follows);
4043 out_freeflocks:
4044                 kfree(flocks);
4045         } else {
4046                 err = ceph_pagelist_reserve(pagelist,
4047                                             sizeof(u64) + sizeof(u32) +
4048                                             pathlen + sizeof(rec.v1));
4049                 if (err)
4050                         goto out_err;
4051
4052                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
4053                 ceph_pagelist_encode_string(pagelist, path, pathlen);
4054                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
4055         }
4056
4057 out_err:
4058         ceph_mdsc_free_path(path, pathlen);
4059         if (!err)
4060                 recon_state->nr_caps++;
4061         return err;
4062 }
4063
4064 static int encode_snap_realms(struct ceph_mds_client *mdsc,
4065                               struct ceph_reconnect_state *recon_state)
4066 {
4067         struct rb_node *p;
4068         struct ceph_pagelist *pagelist = recon_state->pagelist;
4069         int err = 0;
4070
4071         if (recon_state->msg_version >= 4) {
4072                 err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
4073                 if (err < 0)
4074                         goto fail;
4075         }
4076
4077         /*
4078          * snaprealms.  we provide mds with the ino, seq (version), and
4079          * parent for all of our realms.  If the mds has any newer info,
4080          * it will tell us.
4081          */
4082         for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
4083                 struct ceph_snap_realm *realm =
4084                        rb_entry(p, struct ceph_snap_realm, node);
4085                 struct ceph_mds_snaprealm_reconnect sr_rec;
4086
4087                 if (recon_state->msg_version >= 4) {
4088                         size_t need = sizeof(u8) * 2 + sizeof(u32) +
4089                                       sizeof(sr_rec);
4090
4091                         if (pagelist->length + need > RECONNECT_MAX_SIZE) {
4092                                 err = send_reconnect_partial(recon_state);
4093                                 if (err)
4094                                         goto fail;
4095                                 pagelist = recon_state->pagelist;
4096                         }
4097
4098                         err = ceph_pagelist_reserve(pagelist, need);
4099                         if (err)
4100                                 goto fail;
4101
4102                         ceph_pagelist_encode_8(pagelist, 1);
4103                         ceph_pagelist_encode_8(pagelist, 1);
4104                         ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
4105                 }
4106
4107                 dout(" adding snap realm %llx seq %lld parent %llx\n",
4108                      realm->ino, realm->seq, realm->parent_ino);
4109                 sr_rec.ino = cpu_to_le64(realm->ino);
4110                 sr_rec.seq = cpu_to_le64(realm->seq);
4111                 sr_rec.parent = cpu_to_le64(realm->parent_ino);
4112
4113                 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
4114                 if (err)
4115                         goto fail;
4116
4117                 recon_state->nr_realms++;
4118         }
4119 fail:
4120         return err;
4121 }
4122
4123
4124 /*
4125  * If an MDS fails and recovers, clients need to reconnect in order to
4126  * reestablish shared state.  This includes all caps issued through
4127  * this session _and_ the snap_realm hierarchy.  Because it's not
4128  * clear which snap realms the mds cares about, we send everything we
4129  * know about.. that ensures we'll then get any new info the
4130  * recovering MDS might have.
4131  *
4132  * This is a relatively heavyweight operation, but it's rare.
4133  */
4134 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
4135                                struct ceph_mds_session *session)
4136 {
4137         struct ceph_msg *reply;
4138         int mds = session->s_mds;
4139         int err = -ENOMEM;
4140         struct ceph_reconnect_state recon_state = {
4141                 .session = session,
4142         };
4143         LIST_HEAD(dispose);
4144
4145         pr_info("mds%d reconnect start\n", mds);
4146
4147         recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
4148         if (!recon_state.pagelist)
4149                 goto fail_nopagelist;
4150
4151         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
4152         if (!reply)
4153                 goto fail_nomsg;
4154
4155         xa_destroy(&session->s_delegated_inos);
4156
4157         mutex_lock(&session->s_mutex);
4158         session->s_state = CEPH_MDS_SESSION_RECONNECTING;
4159         session->s_seq = 0;
4160
4161         dout("session %p state %s\n", session,
4162              ceph_session_state_name(session->s_state));
4163
4164         atomic_inc(&session->s_cap_gen);
4165
4166         spin_lock(&session->s_cap_lock);
4167         /* don't know if session is readonly */
4168         session->s_readonly = 0;
4169         /*
4170          * notify __ceph_remove_cap() that we are composing cap reconnect.
4171          * If a cap get released before being added to the cap reconnect,
4172          * __ceph_remove_cap() should skip queuing cap release.
4173          */
4174         session->s_cap_reconnect = 1;
4175         /* drop old cap expires; we're about to reestablish that state */
4176         detach_cap_releases(session, &dispose);
4177         spin_unlock(&session->s_cap_lock);
4178         dispose_cap_releases(mdsc, &dispose);
4179
4180         /* trim unused caps to reduce MDS's cache rejoin time */
4181         if (mdsc->fsc->sb->s_root)
4182                 shrink_dcache_parent(mdsc->fsc->sb->s_root);
4183
4184         ceph_con_close(&session->s_con);
4185         ceph_con_open(&session->s_con,
4186                       CEPH_ENTITY_TYPE_MDS, mds,
4187                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
4188
4189         /* replay unsafe requests */
4190         replay_unsafe_requests(mdsc, session);
4191
4192         ceph_early_kick_flushing_caps(mdsc, session);
4193
4194         down_read(&mdsc->snap_rwsem);
4195
4196         /* placeholder for nr_caps */
4197         err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
4198         if (err)
4199                 goto fail;
4200
4201         if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
4202                 recon_state.msg_version = 3;
4203                 recon_state.allow_multi = true;
4204         } else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
4205                 recon_state.msg_version = 3;
4206         } else {
4207                 recon_state.msg_version = 2;
4208         }
4209         /* trsaverse this session's caps */
4210         err = ceph_iterate_session_caps(session, reconnect_caps_cb, &recon_state);
4211
4212         spin_lock(&session->s_cap_lock);
4213         session->s_cap_reconnect = 0;
4214         spin_unlock(&session->s_cap_lock);
4215
4216         if (err < 0)
4217                 goto fail;
4218
4219         /* check if all realms can be encoded into current message */
4220         if (mdsc->num_snap_realms) {
4221                 size_t total_len =
4222                         recon_state.pagelist->length +
4223                         mdsc->num_snap_realms *
4224                         sizeof(struct ceph_mds_snaprealm_reconnect);
4225                 if (recon_state.msg_version >= 4) {
4226                         /* number of realms */
4227                         total_len += sizeof(u32);
4228                         /* version, compat_version and struct_len */
4229                         total_len += mdsc->num_snap_realms *
4230                                      (2 * sizeof(u8) + sizeof(u32));
4231                 }
4232                 if (total_len > RECONNECT_MAX_SIZE) {
4233                         if (!recon_state.allow_multi) {
4234                                 err = -ENOSPC;
4235                                 goto fail;
4236                         }
4237                         if (recon_state.nr_caps) {
4238                                 err = send_reconnect_partial(&recon_state);
4239                                 if (err)
4240                                         goto fail;
4241                         }
4242                         recon_state.msg_version = 5;
4243                 }
4244         }
4245
4246         err = encode_snap_realms(mdsc, &recon_state);
4247         if (err < 0)
4248                 goto fail;
4249
4250         if (recon_state.msg_version >= 5) {
4251                 err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
4252                 if (err < 0)
4253                         goto fail;
4254         }
4255
4256         if (recon_state.nr_caps || recon_state.nr_realms) {
4257                 struct page *page =
4258                         list_first_entry(&recon_state.pagelist->head,
4259                                         struct page, lru);
4260                 __le32 *addr = kmap_atomic(page);
4261                 if (recon_state.nr_caps) {
4262                         WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
4263                         *addr = cpu_to_le32(recon_state.nr_caps);
4264                 } else if (recon_state.msg_version >= 4) {
4265                         *(addr + 1) = cpu_to_le32(recon_state.nr_realms);
4266                 }
4267                 kunmap_atomic(addr);
4268         }
4269
4270         reply->hdr.version = cpu_to_le16(recon_state.msg_version);
4271         if (recon_state.msg_version >= 4)
4272                 reply->hdr.compat_version = cpu_to_le16(4);
4273
4274         reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
4275         ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
4276
4277         ceph_con_send(&session->s_con, reply);
4278
4279         mutex_unlock(&session->s_mutex);
4280
4281         mutex_lock(&mdsc->mutex);
4282         __wake_requests(mdsc, &session->s_waiting);
4283         mutex_unlock(&mdsc->mutex);
4284
4285         up_read(&mdsc->snap_rwsem);
4286         ceph_pagelist_release(recon_state.pagelist);
4287         return;
4288
4289 fail:
4290         ceph_msg_put(reply);
4291         up_read(&mdsc->snap_rwsem);
4292         mutex_unlock(&session->s_mutex);
4293 fail_nomsg:
4294         ceph_pagelist_release(recon_state.pagelist);
4295 fail_nopagelist:
4296         pr_err("error %d preparing reconnect for mds%d\n", err, mds);
4297         return;
4298 }
4299
4300
4301 /*
4302  * compare old and new mdsmaps, kicking requests
4303  * and closing out old connections as necessary
4304  *
4305  * called under mdsc->mutex.
4306  */
4307 static void check_new_map(struct ceph_mds_client *mdsc,
4308                           struct ceph_mdsmap *newmap,
4309                           struct ceph_mdsmap *oldmap)
4310 {
4311         int i, j, err;
4312         int oldstate, newstate;
4313         struct ceph_mds_session *s;
4314         unsigned long targets[DIV_ROUND_UP(CEPH_MAX_MDS, sizeof(unsigned long))] = {0};
4315
4316         dout("check_new_map new %u old %u\n",
4317              newmap->m_epoch, oldmap->m_epoch);
4318
4319         if (newmap->m_info) {
4320                 for (i = 0; i < newmap->possible_max_rank; i++) {
4321                         for (j = 0; j < newmap->m_info[i].num_export_targets; j++)
4322                                 set_bit(newmap->m_info[i].export_targets[j], targets);
4323                 }
4324         }
4325
4326         for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4327                 if (!mdsc->sessions[i])
4328                         continue;
4329                 s = mdsc->sessions[i];
4330                 oldstate = ceph_mdsmap_get_state(oldmap, i);
4331                 newstate = ceph_mdsmap_get_state(newmap, i);
4332
4333                 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
4334                      i, ceph_mds_state_name(oldstate),
4335                      ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
4336                      ceph_mds_state_name(newstate),
4337                      ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
4338                      ceph_session_state_name(s->s_state));
4339
4340                 if (i >= newmap->possible_max_rank) {
4341                         /* force close session for stopped mds */
4342                         ceph_get_mds_session(s);
4343                         __unregister_session(mdsc, s);
4344                         __wake_requests(mdsc, &s->s_waiting);
4345                         mutex_unlock(&mdsc->mutex);
4346
4347                         mutex_lock(&s->s_mutex);
4348                         cleanup_session_requests(mdsc, s);
4349                         remove_session_caps(s);
4350                         mutex_unlock(&s->s_mutex);
4351
4352                         ceph_put_mds_session(s);
4353
4354                         mutex_lock(&mdsc->mutex);
4355                         kick_requests(mdsc, i);
4356                         continue;
4357                 }
4358
4359                 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
4360                            ceph_mdsmap_get_addr(newmap, i),
4361                            sizeof(struct ceph_entity_addr))) {
4362                         /* just close it */
4363                         mutex_unlock(&mdsc->mutex);
4364                         mutex_lock(&s->s_mutex);
4365                         mutex_lock(&mdsc->mutex);
4366                         ceph_con_close(&s->s_con);
4367                         mutex_unlock(&s->s_mutex);
4368                         s->s_state = CEPH_MDS_SESSION_RESTARTING;
4369                 } else if (oldstate == newstate) {
4370                         continue;  /* nothing new with this mds */
4371                 }
4372
4373                 /*
4374                  * send reconnect?
4375                  */
4376                 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
4377                     newstate >= CEPH_MDS_STATE_RECONNECT) {
4378                         mutex_unlock(&mdsc->mutex);
4379                         clear_bit(i, targets);
4380                         send_mds_reconnect(mdsc, s);
4381                         mutex_lock(&mdsc->mutex);
4382                 }
4383
4384                 /*
4385                  * kick request on any mds that has gone active.
4386                  */
4387                 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
4388                     newstate >= CEPH_MDS_STATE_ACTIVE) {
4389                         if (oldstate != CEPH_MDS_STATE_CREATING &&
4390                             oldstate != CEPH_MDS_STATE_STARTING)
4391                                 pr_info("mds%d recovery completed\n", s->s_mds);
4392                         kick_requests(mdsc, i);
4393                         mutex_unlock(&mdsc->mutex);
4394                         mutex_lock(&s->s_mutex);
4395                         mutex_lock(&mdsc->mutex);
4396                         ceph_kick_flushing_caps(mdsc, s);
4397                         mutex_unlock(&s->s_mutex);
4398                         wake_up_session_caps(s, RECONNECT);
4399                 }
4400         }
4401
4402         /*
4403          * Only open and reconnect sessions that don't exist yet.
4404          */
4405         for (i = 0; i < newmap->possible_max_rank; i++) {
4406                 /*
4407                  * In case the import MDS is crashed just after
4408                  * the EImportStart journal is flushed, so when
4409                  * a standby MDS takes over it and is replaying
4410                  * the EImportStart journal the new MDS daemon
4411                  * will wait the client to reconnect it, but the
4412                  * client may never register/open the session yet.
4413                  *
4414                  * Will try to reconnect that MDS daemon if the
4415                  * rank number is in the export targets array and
4416                  * is the up:reconnect state.
4417                  */
4418                 newstate = ceph_mdsmap_get_state(newmap, i);
4419                 if (!test_bit(i, targets) || newstate != CEPH_MDS_STATE_RECONNECT)
4420                         continue;
4421
4422                 /*
4423                  * The session maybe registered and opened by some
4424                  * requests which were choosing random MDSes during
4425                  * the mdsc->mutex's unlock/lock gap below in rare
4426                  * case. But the related MDS daemon will just queue
4427                  * that requests and be still waiting for the client's
4428                  * reconnection request in up:reconnect state.
4429                  */
4430                 s = __ceph_lookup_mds_session(mdsc, i);
4431                 if (likely(!s)) {
4432                         s = __open_export_target_session(mdsc, i);
4433                         if (IS_ERR(s)) {
4434                                 err = PTR_ERR(s);
4435                                 pr_err("failed to open export target session, err %d\n",
4436                                        err);
4437                                 continue;
4438                         }
4439                 }
4440                 dout("send reconnect to export target mds.%d\n", i);
4441                 mutex_unlock(&mdsc->mutex);
4442                 send_mds_reconnect(mdsc, s);
4443                 ceph_put_mds_session(s);
4444                 mutex_lock(&mdsc->mutex);
4445         }
4446
4447         for (i = 0; i < newmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4448                 s = mdsc->sessions[i];
4449                 if (!s)
4450                         continue;
4451                 if (!ceph_mdsmap_is_laggy(newmap, i))
4452                         continue;
4453                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4454                     s->s_state == CEPH_MDS_SESSION_HUNG ||
4455                     s->s_state == CEPH_MDS_SESSION_CLOSING) {
4456                         dout(" connecting to export targets of laggy mds%d\n",
4457                              i);
4458                         __open_export_target_sessions(mdsc, s);
4459                 }
4460         }
4461 }
4462
4463
4464
4465 /*
4466  * leases
4467  */
4468
4469 /*
4470  * caller must hold session s_mutex, dentry->d_lock
4471  */
4472 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
4473 {
4474         struct ceph_dentry_info *di = ceph_dentry(dentry);
4475
4476         ceph_put_mds_session(di->lease_session);
4477         di->lease_session = NULL;
4478 }
4479
4480 static void handle_lease(struct ceph_mds_client *mdsc,
4481                          struct ceph_mds_session *session,
4482                          struct ceph_msg *msg)
4483 {
4484         struct super_block *sb = mdsc->fsc->sb;
4485         struct inode *inode;
4486         struct dentry *parent, *dentry;
4487         struct ceph_dentry_info *di;
4488         int mds = session->s_mds;
4489         struct ceph_mds_lease *h = msg->front.iov_base;
4490         u32 seq;
4491         struct ceph_vino vino;
4492         struct qstr dname;
4493         int release = 0;
4494
4495         dout("handle_lease from mds%d\n", mds);
4496
4497         /* decode */
4498         if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
4499                 goto bad;
4500         vino.ino = le64_to_cpu(h->ino);
4501         vino.snap = CEPH_NOSNAP;
4502         seq = le32_to_cpu(h->seq);
4503         dname.len = get_unaligned_le32(h + 1);
4504         if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
4505                 goto bad;
4506         dname.name = (void *)(h + 1) + sizeof(u32);
4507
4508         /* lookup inode */
4509         inode = ceph_find_inode(sb, vino);
4510         dout("handle_lease %s, ino %llx %p %.*s\n",
4511              ceph_lease_op_name(h->action), vino.ino, inode,
4512              dname.len, dname.name);
4513
4514         mutex_lock(&session->s_mutex);
4515         inc_session_sequence(session);
4516
4517         if (!inode) {
4518                 dout("handle_lease no inode %llx\n", vino.ino);
4519                 goto release;
4520         }
4521
4522         /* dentry */
4523         parent = d_find_alias(inode);
4524         if (!parent) {
4525                 dout("no parent dentry on inode %p\n", inode);
4526                 WARN_ON(1);
4527                 goto release;  /* hrm... */
4528         }
4529         dname.hash = full_name_hash(parent, dname.name, dname.len);
4530         dentry = d_lookup(parent, &dname);
4531         dput(parent);
4532         if (!dentry)
4533                 goto release;
4534
4535         spin_lock(&dentry->d_lock);
4536         di = ceph_dentry(dentry);
4537         switch (h->action) {
4538         case CEPH_MDS_LEASE_REVOKE:
4539                 if (di->lease_session == session) {
4540                         if (ceph_seq_cmp(di->lease_seq, seq) > 0)
4541                                 h->seq = cpu_to_le32(di->lease_seq);
4542                         __ceph_mdsc_drop_dentry_lease(dentry);
4543                 }
4544                 release = 1;
4545                 break;
4546
4547         case CEPH_MDS_LEASE_RENEW:
4548                 if (di->lease_session == session &&
4549                     di->lease_gen == atomic_read(&session->s_cap_gen) &&
4550                     di->lease_renew_from &&
4551                     di->lease_renew_after == 0) {
4552                         unsigned long duration =
4553                                 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
4554
4555                         di->lease_seq = seq;
4556                         di->time = di->lease_renew_from + duration;
4557                         di->lease_renew_after = di->lease_renew_from +
4558                                 (duration >> 1);
4559                         di->lease_renew_from = 0;
4560                 }
4561                 break;
4562         }
4563         spin_unlock(&dentry->d_lock);
4564         dput(dentry);
4565
4566         if (!release)
4567                 goto out;
4568
4569 release:
4570         /* let's just reuse the same message */
4571         h->action = CEPH_MDS_LEASE_REVOKE_ACK;
4572         ceph_msg_get(msg);
4573         ceph_con_send(&session->s_con, msg);
4574
4575 out:
4576         mutex_unlock(&session->s_mutex);
4577         iput(inode);
4578         return;
4579
4580 bad:
4581         pr_err("corrupt lease message\n");
4582         ceph_msg_dump(msg);
4583 }
4584
4585 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
4586                               struct dentry *dentry, char action,
4587                               u32 seq)
4588 {
4589         struct ceph_msg *msg;
4590         struct ceph_mds_lease *lease;
4591         struct inode *dir;
4592         int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
4593
4594         dout("lease_send_msg identry %p %s to mds%d\n",
4595              dentry, ceph_lease_op_name(action), session->s_mds);
4596
4597         msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
4598         if (!msg)
4599                 return;
4600         lease = msg->front.iov_base;
4601         lease->action = action;
4602         lease->seq = cpu_to_le32(seq);
4603
4604         spin_lock(&dentry->d_lock);
4605         dir = d_inode(dentry->d_parent);
4606         lease->ino = cpu_to_le64(ceph_ino(dir));
4607         lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
4608
4609         put_unaligned_le32(dentry->d_name.len, lease + 1);
4610         memcpy((void *)(lease + 1) + 4,
4611                dentry->d_name.name, dentry->d_name.len);
4612         spin_unlock(&dentry->d_lock);
4613
4614         ceph_con_send(&session->s_con, msg);
4615 }
4616
4617 /*
4618  * lock unlock the session, to wait ongoing session activities
4619  */
4620 static void lock_unlock_session(struct ceph_mds_session *s)
4621 {
4622         mutex_lock(&s->s_mutex);
4623         mutex_unlock(&s->s_mutex);
4624 }
4625
4626 static void maybe_recover_session(struct ceph_mds_client *mdsc)
4627 {
4628         struct ceph_fs_client *fsc = mdsc->fsc;
4629
4630         if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
4631                 return;
4632
4633         if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
4634                 return;
4635
4636         if (!READ_ONCE(fsc->blocklisted))
4637                 return;
4638
4639         pr_info("auto reconnect after blocklisted\n");
4640         ceph_force_reconnect(fsc->sb);
4641 }
4642
4643 bool check_session_state(struct ceph_mds_session *s)
4644 {
4645         switch (s->s_state) {
4646         case CEPH_MDS_SESSION_OPEN:
4647                 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
4648                         s->s_state = CEPH_MDS_SESSION_HUNG;
4649                         pr_info("mds%d hung\n", s->s_mds);
4650                 }
4651                 break;
4652         case CEPH_MDS_SESSION_CLOSING:
4653         case CEPH_MDS_SESSION_NEW:
4654         case CEPH_MDS_SESSION_RESTARTING:
4655         case CEPH_MDS_SESSION_CLOSED:
4656         case CEPH_MDS_SESSION_REJECTED:
4657                 return false;
4658         }
4659
4660         return true;
4661 }
4662
4663 /*
4664  * If the sequence is incremented while we're waiting on a REQUEST_CLOSE reply,
4665  * then we need to retransmit that request.
4666  */
4667 void inc_session_sequence(struct ceph_mds_session *s)
4668 {
4669         lockdep_assert_held(&s->s_mutex);
4670
4671         s->s_seq++;
4672
4673         if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
4674                 int ret;
4675
4676                 dout("resending session close request for mds%d\n", s->s_mds);
4677                 ret = request_close_session(s);
4678                 if (ret < 0)
4679                         pr_err("unable to close session to mds%d: %d\n",
4680                                s->s_mds, ret);
4681         }
4682 }
4683
4684 /*
4685  * delayed work -- periodically trim expired leases, renew caps with mds.  If
4686  * the @delay parameter is set to 0 or if it's more than 5 secs, the default
4687  * workqueue delay value of 5 secs will be used.
4688  */
4689 static void schedule_delayed(struct ceph_mds_client *mdsc, unsigned long delay)
4690 {
4691         unsigned long max_delay = HZ * 5;
4692
4693         /* 5 secs default delay */
4694         if (!delay || (delay > max_delay))
4695                 delay = max_delay;
4696         schedule_delayed_work(&mdsc->delayed_work,
4697                               round_jiffies_relative(delay));
4698 }
4699
4700 static void delayed_work(struct work_struct *work)
4701 {
4702         struct ceph_mds_client *mdsc =
4703                 container_of(work, struct ceph_mds_client, delayed_work.work);
4704         unsigned long delay;
4705         int renew_interval;
4706         int renew_caps;
4707         int i;
4708
4709         dout("mdsc delayed_work\n");
4710
4711         if (mdsc->stopping)
4712                 return;
4713
4714         mutex_lock(&mdsc->mutex);
4715         renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
4716         renew_caps = time_after_eq(jiffies, HZ*renew_interval +
4717                                    mdsc->last_renew_caps);
4718         if (renew_caps)
4719                 mdsc->last_renew_caps = jiffies;
4720
4721         for (i = 0; i < mdsc->max_sessions; i++) {
4722                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4723                 if (!s)
4724                         continue;
4725
4726                 if (!check_session_state(s)) {
4727                         ceph_put_mds_session(s);
4728                         continue;
4729                 }
4730                 mutex_unlock(&mdsc->mutex);
4731
4732                 mutex_lock(&s->s_mutex);
4733                 if (renew_caps)
4734                         send_renew_caps(mdsc, s);
4735                 else
4736                         ceph_con_keepalive(&s->s_con);
4737                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4738                     s->s_state == CEPH_MDS_SESSION_HUNG)
4739                         ceph_send_cap_releases(mdsc, s);
4740                 mutex_unlock(&s->s_mutex);
4741                 ceph_put_mds_session(s);
4742
4743                 mutex_lock(&mdsc->mutex);
4744         }
4745         mutex_unlock(&mdsc->mutex);
4746
4747         delay = ceph_check_delayed_caps(mdsc);
4748
4749         ceph_queue_cap_reclaim_work(mdsc);
4750
4751         ceph_trim_snapid_map(mdsc);
4752
4753         maybe_recover_session(mdsc);
4754
4755         schedule_delayed(mdsc, delay);
4756 }
4757
4758 int ceph_mdsc_init(struct ceph_fs_client *fsc)
4759
4760 {
4761         struct ceph_mds_client *mdsc;
4762         int err;
4763
4764         mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
4765         if (!mdsc)
4766                 return -ENOMEM;
4767         mdsc->fsc = fsc;
4768         mutex_init(&mdsc->mutex);
4769         mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
4770         if (!mdsc->mdsmap) {
4771                 err = -ENOMEM;
4772                 goto err_mdsc;
4773         }
4774
4775         init_completion(&mdsc->safe_umount_waiters);
4776         init_waitqueue_head(&mdsc->session_close_wq);
4777         INIT_LIST_HEAD(&mdsc->waiting_for_map);
4778         mdsc->quotarealms_inodes = RB_ROOT;
4779         mutex_init(&mdsc->quotarealms_inodes_mutex);
4780         init_rwsem(&mdsc->snap_rwsem);
4781         mdsc->snap_realms = RB_ROOT;
4782         INIT_LIST_HEAD(&mdsc->snap_empty);
4783         spin_lock_init(&mdsc->snap_empty_lock);
4784         mdsc->request_tree = RB_ROOT;
4785         INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
4786         mdsc->last_renew_caps = jiffies;
4787         INIT_LIST_HEAD(&mdsc->cap_delay_list);
4788         INIT_LIST_HEAD(&mdsc->cap_wait_list);
4789         spin_lock_init(&mdsc->cap_delay_lock);
4790         INIT_LIST_HEAD(&mdsc->snap_flush_list);
4791         spin_lock_init(&mdsc->snap_flush_lock);
4792         mdsc->last_cap_flush_tid = 1;
4793         INIT_LIST_HEAD(&mdsc->cap_flush_list);
4794         INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
4795         spin_lock_init(&mdsc->cap_dirty_lock);
4796         init_waitqueue_head(&mdsc->cap_flushing_wq);
4797         INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
4798         err = ceph_metric_init(&mdsc->metric);
4799         if (err)
4800                 goto err_mdsmap;
4801
4802         spin_lock_init(&mdsc->dentry_list_lock);
4803         INIT_LIST_HEAD(&mdsc->dentry_leases);
4804         INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
4805
4806         ceph_caps_init(mdsc);
4807         ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
4808
4809         spin_lock_init(&mdsc->snapid_map_lock);
4810         mdsc->snapid_map_tree = RB_ROOT;
4811         INIT_LIST_HEAD(&mdsc->snapid_map_lru);
4812
4813         init_rwsem(&mdsc->pool_perm_rwsem);
4814         mdsc->pool_perm_tree = RB_ROOT;
4815
4816         strscpy(mdsc->nodename, utsname()->nodename,
4817                 sizeof(mdsc->nodename));
4818
4819         fsc->mdsc = mdsc;
4820         return 0;
4821
4822 err_mdsmap:
4823         kfree(mdsc->mdsmap);
4824 err_mdsc:
4825         kfree(mdsc);
4826         return err;
4827 }
4828
4829 /*
4830  * Wait for safe replies on open mds requests.  If we time out, drop
4831  * all requests from the tree to avoid dangling dentry refs.
4832  */
4833 static void wait_requests(struct ceph_mds_client *mdsc)
4834 {
4835         struct ceph_options *opts = mdsc->fsc->client->options;
4836         struct ceph_mds_request *req;
4837
4838         mutex_lock(&mdsc->mutex);
4839         if (__get_oldest_req(mdsc)) {
4840                 mutex_unlock(&mdsc->mutex);
4841
4842                 dout("wait_requests waiting for requests\n");
4843                 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
4844                                     ceph_timeout_jiffies(opts->mount_timeout));
4845
4846                 /* tear down remaining requests */
4847                 mutex_lock(&mdsc->mutex);
4848                 while ((req = __get_oldest_req(mdsc))) {
4849                         dout("wait_requests timed out on tid %llu\n",
4850                              req->r_tid);
4851                         list_del_init(&req->r_wait);
4852                         __unregister_request(mdsc, req);
4853                 }
4854         }
4855         mutex_unlock(&mdsc->mutex);
4856         dout("wait_requests done\n");
4857 }
4858
4859 void send_flush_mdlog(struct ceph_mds_session *s)
4860 {
4861         struct ceph_msg *msg;
4862
4863         /*
4864          * Pre-luminous MDS crashes when it sees an unknown session request
4865          */
4866         if (!CEPH_HAVE_FEATURE(s->s_con.peer_features, SERVER_LUMINOUS))
4867                 return;
4868
4869         mutex_lock(&s->s_mutex);
4870         dout("request mdlog flush to mds%d (%s)s seq %lld\n", s->s_mds,
4871              ceph_session_state_name(s->s_state), s->s_seq);
4872         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_FLUSH_MDLOG,
4873                                       s->s_seq);
4874         if (!msg) {
4875                 pr_err("failed to request mdlog flush to mds%d (%s) seq %lld\n",
4876                        s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
4877         } else {
4878                 ceph_con_send(&s->s_con, msg);
4879         }
4880         mutex_unlock(&s->s_mutex);
4881 }
4882
4883 /*
4884  * called before mount is ro, and before dentries are torn down.
4885  * (hmm, does this still race with new lookups?)
4886  */
4887 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
4888 {
4889         dout("pre_umount\n");
4890         mdsc->stopping = 1;
4891
4892         ceph_mdsc_iterate_sessions(mdsc, send_flush_mdlog, true);
4893         ceph_mdsc_iterate_sessions(mdsc, lock_unlock_session, false);
4894         ceph_flush_dirty_caps(mdsc);
4895         wait_requests(mdsc);
4896
4897         /*
4898          * wait for reply handlers to drop their request refs and
4899          * their inode/dcache refs
4900          */
4901         ceph_msgr_flush();
4902
4903         ceph_cleanup_quotarealms_inodes(mdsc);
4904 }
4905
4906 /*
4907  * flush the mdlog and wait for all write mds requests to flush.
4908  */
4909 static void flush_mdlog_and_wait_mdsc_unsafe_requests(struct ceph_mds_client *mdsc,
4910                                                  u64 want_tid)
4911 {
4912         struct ceph_mds_request *req = NULL, *nextreq;
4913         struct ceph_mds_session *last_session = NULL;
4914         struct rb_node *n;
4915
4916         mutex_lock(&mdsc->mutex);
4917         dout("%s want %lld\n", __func__, want_tid);
4918 restart:
4919         req = __get_oldest_req(mdsc);
4920         while (req && req->r_tid <= want_tid) {
4921                 /* find next request */
4922                 n = rb_next(&req->r_node);
4923                 if (n)
4924                         nextreq = rb_entry(n, struct ceph_mds_request, r_node);
4925                 else
4926                         nextreq = NULL;
4927                 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
4928                     (req->r_op & CEPH_MDS_OP_WRITE)) {
4929                         struct ceph_mds_session *s = req->r_session;
4930
4931                         if (!s) {
4932                                 req = nextreq;
4933                                 continue;
4934                         }
4935
4936                         /* write op */
4937                         ceph_mdsc_get_request(req);
4938                         if (nextreq)
4939                                 ceph_mdsc_get_request(nextreq);
4940                         s = ceph_get_mds_session(s);
4941                         mutex_unlock(&mdsc->mutex);
4942
4943                         /* send flush mdlog request to MDS */
4944                         if (last_session != s) {
4945                                 send_flush_mdlog(s);
4946                                 ceph_put_mds_session(last_session);
4947                                 last_session = s;
4948                         } else {
4949                                 ceph_put_mds_session(s);
4950                         }
4951                         dout("%s wait on %llu (want %llu)\n", __func__,
4952                              req->r_tid, want_tid);
4953                         wait_for_completion(&req->r_safe_completion);
4954
4955                         mutex_lock(&mdsc->mutex);
4956                         ceph_mdsc_put_request(req);
4957                         if (!nextreq)
4958                                 break;  /* next dne before, so we're done! */
4959                         if (RB_EMPTY_NODE(&nextreq->r_node)) {
4960                                 /* next request was removed from tree */
4961                                 ceph_mdsc_put_request(nextreq);
4962                                 goto restart;
4963                         }
4964                         ceph_mdsc_put_request(nextreq);  /* won't go away */
4965                 }
4966                 req = nextreq;
4967         }
4968         mutex_unlock(&mdsc->mutex);
4969         ceph_put_mds_session(last_session);
4970         dout("%s done\n", __func__);
4971 }
4972
4973 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
4974 {
4975         u64 want_tid, want_flush;
4976
4977         if (READ_ONCE(mdsc->fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN)
4978                 return;
4979
4980         dout("sync\n");
4981         mutex_lock(&mdsc->mutex);
4982         want_tid = mdsc->last_tid;
4983         mutex_unlock(&mdsc->mutex);
4984
4985         ceph_flush_dirty_caps(mdsc);
4986         spin_lock(&mdsc->cap_dirty_lock);
4987         want_flush = mdsc->last_cap_flush_tid;
4988         if (!list_empty(&mdsc->cap_flush_list)) {
4989                 struct ceph_cap_flush *cf =
4990                         list_last_entry(&mdsc->cap_flush_list,
4991                                         struct ceph_cap_flush, g_list);
4992                 cf->wake = true;
4993         }
4994         spin_unlock(&mdsc->cap_dirty_lock);
4995
4996         dout("sync want tid %lld flush_seq %lld\n",
4997              want_tid, want_flush);
4998
4999         flush_mdlog_and_wait_mdsc_unsafe_requests(mdsc, want_tid);
5000         wait_caps_flush(mdsc, want_flush);
5001 }
5002
5003 /*
5004  * true if all sessions are closed, or we force unmount
5005  */
5006 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
5007 {
5008         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
5009                 return true;
5010         return atomic_read(&mdsc->num_sessions) <= skipped;
5011 }
5012
5013 /*
5014  * called after sb is ro.
5015  */
5016 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
5017 {
5018         struct ceph_options *opts = mdsc->fsc->client->options;
5019         struct ceph_mds_session *session;
5020         int i;
5021         int skipped = 0;
5022
5023         dout("close_sessions\n");
5024
5025         /* close sessions */
5026         mutex_lock(&mdsc->mutex);
5027         for (i = 0; i < mdsc->max_sessions; i++) {
5028                 session = __ceph_lookup_mds_session(mdsc, i);
5029                 if (!session)
5030                         continue;
5031                 mutex_unlock(&mdsc->mutex);
5032                 mutex_lock(&session->s_mutex);
5033                 if (__close_session(mdsc, session) <= 0)
5034                         skipped++;
5035                 mutex_unlock(&session->s_mutex);
5036                 ceph_put_mds_session(session);
5037                 mutex_lock(&mdsc->mutex);
5038         }
5039         mutex_unlock(&mdsc->mutex);
5040
5041         dout("waiting for sessions to close\n");
5042         wait_event_timeout(mdsc->session_close_wq,
5043                            done_closing_sessions(mdsc, skipped),
5044                            ceph_timeout_jiffies(opts->mount_timeout));
5045
5046         /* tear down remaining sessions */
5047         mutex_lock(&mdsc->mutex);
5048         for (i = 0; i < mdsc->max_sessions; i++) {
5049                 if (mdsc->sessions[i]) {
5050                         session = ceph_get_mds_session(mdsc->sessions[i]);
5051                         __unregister_session(mdsc, session);
5052                         mutex_unlock(&mdsc->mutex);
5053                         mutex_lock(&session->s_mutex);
5054                         remove_session_caps(session);
5055                         mutex_unlock(&session->s_mutex);
5056                         ceph_put_mds_session(session);
5057                         mutex_lock(&mdsc->mutex);
5058                 }
5059         }
5060         WARN_ON(!list_empty(&mdsc->cap_delay_list));
5061         mutex_unlock(&mdsc->mutex);
5062
5063         ceph_cleanup_snapid_map(mdsc);
5064         ceph_cleanup_global_and_empty_realms(mdsc);
5065
5066         cancel_work_sync(&mdsc->cap_reclaim_work);
5067         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
5068
5069         dout("stopped\n");
5070 }
5071
5072 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
5073 {
5074         struct ceph_mds_session *session;
5075         int mds;
5076
5077         dout("force umount\n");
5078
5079         mutex_lock(&mdsc->mutex);
5080         for (mds = 0; mds < mdsc->max_sessions; mds++) {
5081                 session = __ceph_lookup_mds_session(mdsc, mds);
5082                 if (!session)
5083                         continue;
5084
5085                 if (session->s_state == CEPH_MDS_SESSION_REJECTED)
5086                         __unregister_session(mdsc, session);
5087                 __wake_requests(mdsc, &session->s_waiting);
5088                 mutex_unlock(&mdsc->mutex);
5089
5090                 mutex_lock(&session->s_mutex);
5091                 __close_session(mdsc, session);
5092                 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
5093                         cleanup_session_requests(mdsc, session);
5094                         remove_session_caps(session);
5095                 }
5096                 mutex_unlock(&session->s_mutex);
5097                 ceph_put_mds_session(session);
5098
5099                 mutex_lock(&mdsc->mutex);
5100                 kick_requests(mdsc, mds);
5101         }
5102         __wake_requests(mdsc, &mdsc->waiting_for_map);
5103         mutex_unlock(&mdsc->mutex);
5104 }
5105
5106 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
5107 {
5108         dout("stop\n");
5109         /*
5110          * Make sure the delayed work stopped before releasing
5111          * the resources.
5112          *
5113          * Because the cancel_delayed_work_sync() will only
5114          * guarantee that the work finishes executing. But the
5115          * delayed work will re-arm itself again after that.
5116          */
5117         flush_delayed_work(&mdsc->delayed_work);
5118
5119         if (mdsc->mdsmap)
5120                 ceph_mdsmap_destroy(mdsc->mdsmap);
5121         kfree(mdsc->sessions);
5122         ceph_caps_finalize(mdsc);
5123         ceph_pool_perm_destroy(mdsc);
5124 }
5125
5126 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
5127 {
5128         struct ceph_mds_client *mdsc = fsc->mdsc;
5129         dout("mdsc_destroy %p\n", mdsc);
5130
5131         if (!mdsc)
5132                 return;
5133
5134         /* flush out any connection work with references to us */
5135         ceph_msgr_flush();
5136
5137         ceph_mdsc_stop(mdsc);
5138
5139         ceph_metric_destroy(&mdsc->metric);
5140
5141         fsc->mdsc = NULL;
5142         kfree(mdsc);
5143         dout("mdsc_destroy %p done\n", mdsc);
5144 }
5145
5146 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5147 {
5148         struct ceph_fs_client *fsc = mdsc->fsc;
5149         const char *mds_namespace = fsc->mount_options->mds_namespace;
5150         void *p = msg->front.iov_base;
5151         void *end = p + msg->front.iov_len;
5152         u32 epoch;
5153         u32 num_fs;
5154         u32 mount_fscid = (u32)-1;
5155         int err = -EINVAL;
5156
5157         ceph_decode_need(&p, end, sizeof(u32), bad);
5158         epoch = ceph_decode_32(&p);
5159
5160         dout("handle_fsmap epoch %u\n", epoch);
5161
5162         /* struct_v, struct_cv, map_len, epoch, legacy_client_fscid */
5163         ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 3, bad);
5164
5165         ceph_decode_32_safe(&p, end, num_fs, bad);
5166         while (num_fs-- > 0) {
5167                 void *info_p, *info_end;
5168                 u32 info_len;
5169                 u32 fscid, namelen;
5170
5171                 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
5172                 p += 2;         // info_v, info_cv
5173                 info_len = ceph_decode_32(&p);
5174                 ceph_decode_need(&p, end, info_len, bad);
5175                 info_p = p;
5176                 info_end = p + info_len;
5177                 p = info_end;
5178
5179                 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
5180                 fscid = ceph_decode_32(&info_p);
5181                 namelen = ceph_decode_32(&info_p);
5182                 ceph_decode_need(&info_p, info_end, namelen, bad);
5183
5184                 if (mds_namespace &&
5185                     strlen(mds_namespace) == namelen &&
5186                     !strncmp(mds_namespace, (char *)info_p, namelen)) {
5187                         mount_fscid = fscid;
5188                         break;
5189                 }
5190         }
5191
5192         ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
5193         if (mount_fscid != (u32)-1) {
5194                 fsc->client->monc.fs_cluster_id = mount_fscid;
5195                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
5196                                    0, true);
5197                 ceph_monc_renew_subs(&fsc->client->monc);
5198         } else {
5199                 err = -ENOENT;
5200                 goto err_out;
5201         }
5202         return;
5203
5204 bad:
5205         pr_err("error decoding fsmap %d. Shutting down mount.\n", err);
5206         ceph_umount_begin(mdsc->fsc->sb);
5207 err_out:
5208         mutex_lock(&mdsc->mutex);
5209         mdsc->mdsmap_err = err;
5210         __wake_requests(mdsc, &mdsc->waiting_for_map);
5211         mutex_unlock(&mdsc->mutex);
5212 }
5213
5214 /*
5215  * handle mds map update.
5216  */
5217 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5218 {
5219         u32 epoch;
5220         u32 maplen;
5221         void *p = msg->front.iov_base;
5222         void *end = p + msg->front.iov_len;
5223         struct ceph_mdsmap *newmap, *oldmap;
5224         struct ceph_fsid fsid;
5225         int err = -EINVAL;
5226
5227         ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
5228         ceph_decode_copy(&p, &fsid, sizeof(fsid));
5229         if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
5230                 return;
5231         epoch = ceph_decode_32(&p);
5232         maplen = ceph_decode_32(&p);
5233         dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
5234
5235         /* do we need it? */
5236         mutex_lock(&mdsc->mutex);
5237         if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
5238                 dout("handle_map epoch %u <= our %u\n",
5239                      epoch, mdsc->mdsmap->m_epoch);
5240                 mutex_unlock(&mdsc->mutex);
5241                 return;
5242         }
5243
5244         newmap = ceph_mdsmap_decode(&p, end, ceph_msgr2(mdsc->fsc->client));
5245         if (IS_ERR(newmap)) {
5246                 err = PTR_ERR(newmap);
5247                 goto bad_unlock;
5248         }
5249
5250         /* swap into place */
5251         if (mdsc->mdsmap) {
5252                 oldmap = mdsc->mdsmap;
5253                 mdsc->mdsmap = newmap;
5254                 check_new_map(mdsc, newmap, oldmap);
5255                 ceph_mdsmap_destroy(oldmap);
5256         } else {
5257                 mdsc->mdsmap = newmap;  /* first mds map */
5258         }
5259         mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
5260                                         MAX_LFS_FILESIZE);
5261
5262         __wake_requests(mdsc, &mdsc->waiting_for_map);
5263         ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
5264                           mdsc->mdsmap->m_epoch);
5265
5266         mutex_unlock(&mdsc->mutex);
5267         schedule_delayed(mdsc, 0);
5268         return;
5269
5270 bad_unlock:
5271         mutex_unlock(&mdsc->mutex);
5272 bad:
5273         pr_err("error decoding mdsmap %d. Shutting down mount.\n", err);
5274         ceph_umount_begin(mdsc->fsc->sb);
5275         return;
5276 }
5277
5278 static struct ceph_connection *mds_get_con(struct ceph_connection *con)
5279 {
5280         struct ceph_mds_session *s = con->private;
5281
5282         if (ceph_get_mds_session(s))
5283                 return con;
5284         return NULL;
5285 }
5286
5287 static void mds_put_con(struct ceph_connection *con)
5288 {
5289         struct ceph_mds_session *s = con->private;
5290
5291         ceph_put_mds_session(s);
5292 }
5293
5294 /*
5295  * if the client is unresponsive for long enough, the mds will kill
5296  * the session entirely.
5297  */
5298 static void mds_peer_reset(struct ceph_connection *con)
5299 {
5300         struct ceph_mds_session *s = con->private;
5301         struct ceph_mds_client *mdsc = s->s_mdsc;
5302
5303         pr_warn("mds%d closed our session\n", s->s_mds);
5304         send_mds_reconnect(mdsc, s);
5305 }
5306
5307 static void mds_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5308 {
5309         struct ceph_mds_session *s = con->private;
5310         struct ceph_mds_client *mdsc = s->s_mdsc;
5311         int type = le16_to_cpu(msg->hdr.type);
5312
5313         mutex_lock(&mdsc->mutex);
5314         if (__verify_registered_session(mdsc, s) < 0) {
5315                 mutex_unlock(&mdsc->mutex);
5316                 goto out;
5317         }
5318         mutex_unlock(&mdsc->mutex);
5319
5320         switch (type) {
5321         case CEPH_MSG_MDS_MAP:
5322                 ceph_mdsc_handle_mdsmap(mdsc, msg);
5323                 break;
5324         case CEPH_MSG_FS_MAP_USER:
5325                 ceph_mdsc_handle_fsmap(mdsc, msg);
5326                 break;
5327         case CEPH_MSG_CLIENT_SESSION:
5328                 handle_session(s, msg);
5329                 break;
5330         case CEPH_MSG_CLIENT_REPLY:
5331                 handle_reply(s, msg);
5332                 break;
5333         case CEPH_MSG_CLIENT_REQUEST_FORWARD:
5334                 handle_forward(mdsc, s, msg);
5335                 break;
5336         case CEPH_MSG_CLIENT_CAPS:
5337                 ceph_handle_caps(s, msg);
5338                 break;
5339         case CEPH_MSG_CLIENT_SNAP:
5340                 ceph_handle_snap(mdsc, s, msg);
5341                 break;
5342         case CEPH_MSG_CLIENT_LEASE:
5343                 handle_lease(mdsc, s, msg);
5344                 break;
5345         case CEPH_MSG_CLIENT_QUOTA:
5346                 ceph_handle_quota(mdsc, s, msg);
5347                 break;
5348
5349         default:
5350                 pr_err("received unknown message type %d %s\n", type,
5351                        ceph_msg_type_name(type));
5352         }
5353 out:
5354         ceph_msg_put(msg);
5355 }
5356
5357 /*
5358  * authentication
5359  */
5360
5361 /*
5362  * Note: returned pointer is the address of a structure that's
5363  * managed separately.  Caller must *not* attempt to free it.
5364  */
5365 static struct ceph_auth_handshake *
5366 mds_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5367 {
5368         struct ceph_mds_session *s = con->private;
5369         struct ceph_mds_client *mdsc = s->s_mdsc;
5370         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5371         struct ceph_auth_handshake *auth = &s->s_auth;
5372         int ret;
5373
5374         ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5375                                          force_new, proto, NULL, NULL);
5376         if (ret)
5377                 return ERR_PTR(ret);
5378
5379         return auth;
5380 }
5381
5382 static int mds_add_authorizer_challenge(struct ceph_connection *con,
5383                                     void *challenge_buf, int challenge_buf_len)
5384 {
5385         struct ceph_mds_session *s = con->private;
5386         struct ceph_mds_client *mdsc = s->s_mdsc;
5387         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5388
5389         return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
5390                                             challenge_buf, challenge_buf_len);
5391 }
5392
5393 static int mds_verify_authorizer_reply(struct ceph_connection *con)
5394 {
5395         struct ceph_mds_session *s = con->private;
5396         struct ceph_mds_client *mdsc = s->s_mdsc;
5397         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5398         struct ceph_auth_handshake *auth = &s->s_auth;
5399
5400         return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5401                 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5402                 NULL, NULL, NULL, NULL);
5403 }
5404
5405 static int mds_invalidate_authorizer(struct ceph_connection *con)
5406 {
5407         struct ceph_mds_session *s = con->private;
5408         struct ceph_mds_client *mdsc = s->s_mdsc;
5409         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5410
5411         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
5412
5413         return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
5414 }
5415
5416 static int mds_get_auth_request(struct ceph_connection *con,
5417                                 void *buf, int *buf_len,
5418                                 void **authorizer, int *authorizer_len)
5419 {
5420         struct ceph_mds_session *s = con->private;
5421         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5422         struct ceph_auth_handshake *auth = &s->s_auth;
5423         int ret;
5424
5425         ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5426                                        buf, buf_len);
5427         if (ret)
5428                 return ret;
5429
5430         *authorizer = auth->authorizer_buf;
5431         *authorizer_len = auth->authorizer_buf_len;
5432         return 0;
5433 }
5434
5435 static int mds_handle_auth_reply_more(struct ceph_connection *con,
5436                                       void *reply, int reply_len,
5437                                       void *buf, int *buf_len,
5438                                       void **authorizer, int *authorizer_len)
5439 {
5440         struct ceph_mds_session *s = con->private;
5441         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5442         struct ceph_auth_handshake *auth = &s->s_auth;
5443         int ret;
5444
5445         ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5446                                               buf, buf_len);
5447         if (ret)
5448                 return ret;
5449
5450         *authorizer = auth->authorizer_buf;
5451         *authorizer_len = auth->authorizer_buf_len;
5452         return 0;
5453 }
5454
5455 static int mds_handle_auth_done(struct ceph_connection *con,
5456                                 u64 global_id, void *reply, int reply_len,
5457                                 u8 *session_key, int *session_key_len,
5458                                 u8 *con_secret, int *con_secret_len)
5459 {
5460         struct ceph_mds_session *s = con->private;
5461         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5462         struct ceph_auth_handshake *auth = &s->s_auth;
5463
5464         return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5465                                                session_key, session_key_len,
5466                                                con_secret, con_secret_len);
5467 }
5468
5469 static int mds_handle_auth_bad_method(struct ceph_connection *con,
5470                                       int used_proto, int result,
5471                                       const int *allowed_protos, int proto_cnt,
5472                                       const int *allowed_modes, int mode_cnt)
5473 {
5474         struct ceph_mds_session *s = con->private;
5475         struct ceph_mon_client *monc = &s->s_mdsc->fsc->client->monc;
5476         int ret;
5477
5478         if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_MDS,
5479                                             used_proto, result,
5480                                             allowed_protos, proto_cnt,
5481                                             allowed_modes, mode_cnt)) {
5482                 ret = ceph_monc_validate_auth(monc);
5483                 if (ret)
5484                         return ret;
5485         }
5486
5487         return -EACCES;
5488 }
5489
5490 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
5491                                 struct ceph_msg_header *hdr, int *skip)
5492 {
5493         struct ceph_msg *msg;
5494         int type = (int) le16_to_cpu(hdr->type);
5495         int front_len = (int) le32_to_cpu(hdr->front_len);
5496
5497         if (con->in_msg)
5498                 return con->in_msg;
5499
5500         *skip = 0;
5501         msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
5502         if (!msg) {
5503                 pr_err("unable to allocate msg type %d len %d\n",
5504                        type, front_len);
5505                 return NULL;
5506         }
5507
5508         return msg;
5509 }
5510
5511 static int mds_sign_message(struct ceph_msg *msg)
5512 {
5513        struct ceph_mds_session *s = msg->con->private;
5514        struct ceph_auth_handshake *auth = &s->s_auth;
5515
5516        return ceph_auth_sign_message(auth, msg);
5517 }
5518
5519 static int mds_check_message_signature(struct ceph_msg *msg)
5520 {
5521        struct ceph_mds_session *s = msg->con->private;
5522        struct ceph_auth_handshake *auth = &s->s_auth;
5523
5524        return ceph_auth_check_message_signature(auth, msg);
5525 }
5526
5527 static const struct ceph_connection_operations mds_con_ops = {
5528         .get = mds_get_con,
5529         .put = mds_put_con,
5530         .alloc_msg = mds_alloc_msg,
5531         .dispatch = mds_dispatch,
5532         .peer_reset = mds_peer_reset,
5533         .get_authorizer = mds_get_authorizer,
5534         .add_authorizer_challenge = mds_add_authorizer_challenge,
5535         .verify_authorizer_reply = mds_verify_authorizer_reply,
5536         .invalidate_authorizer = mds_invalidate_authorizer,
5537         .sign_message = mds_sign_message,
5538         .check_message_signature = mds_check_message_signature,
5539         .get_auth_request = mds_get_auth_request,
5540         .handle_auth_reply_more = mds_handle_auth_reply_more,
5541         .handle_auth_done = mds_handle_auth_done,
5542         .handle_auth_bad_method = mds_handle_auth_bad_method,
5543 };
5544
5545 /* eof */