fs/ceph/mds_client.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/fs.h>
   5 #include <linux/wait.h>
   6 #include <linux/slab.h>
   7 #include <linux/gfp.h>
   8 #include <linux/sched.h>
   9 #include <linux/debugfs.h>
  10 #include <linux/seq_file.h>
  11 #include <linux/ratelimit.h>
  12 #include <linux/bits.h>
  13 #include <linux/ktime.h>
  14 #include <linux/bitmap.h>
  15
  16 #include "super.h"
  17 #include "mds_client.h"
  18
  19 #include <linux/ceph/ceph_features.h>
  20 #include <linux/ceph/messenger.h>
  21 #include <linux/ceph/decode.h>
  22 #include <linux/ceph/pagelist.h>
  23 #include <linux/ceph/auth.h>
  24 #include <linux/ceph/debugfs.h>
  25
  26 #define RECONNECT_MAX_SIZE (INT_MAX - PAGE_SIZE)
  27
  28 /*
  29  * A cluster of MDS (metadata server) daemons is responsible for
  30  * managing the file system namespace (the directory hierarchy and
  31  * inodes) and for coordinating shared access to storage.  Metadata is
  32  * partitioning hierarchically across a number of servers, and that
  33  * partition varies over time as the cluster adjusts the distribution
  34  * in order to balance load.
  35  *
  36  * The MDS client is primarily responsible to managing synchronous
  37  * metadata requests for operations like open, unlink, and so forth.
  38  * If there is a MDS failure, we find out about it when we (possibly
  39  * request and) receive a new MDS map, and can resubmit affected
  40  * requests.
  41  *
  42  * For the most part, though, we take advantage of a lossless
  43  * communications channel to the MDS, and do not need to worry about
  44  * timing out or resubmitting requests.
  45  *
  46  * We maintain a stateful "session" with each MDS we interact with.
  47  * Within each session, we sent periodic heartbeat messages to ensure
  48  * any capabilities or leases we have been issues remain valid.  If
  49  * the session times out and goes stale, our leases and capabilities
  50  * are no longer valid.
  51  */
  52
  53 struct ceph_reconnect_state {
  54         struct ceph_mds_session *session;
  55         int nr_caps, nr_realms;
  56         struct ceph_pagelist *pagelist;
  57         unsigned msg_version;
  58         bool allow_multi;
  59 };
  60
  61 static void __wake_requests(struct ceph_mds_client *mdsc,
  62                             struct list_head *head);
  63 static void ceph_cap_release_work(struct work_struct *work);
  64 static void ceph_cap_reclaim_work(struct work_struct *work);
  65
  66 static const struct ceph_connection_operations mds_con_ops;
  67
  68
  69 /*
  70  * mds reply parsing
  71  */
  72
  73 static int parse_reply_info_quota(void **p, void *end,
  74                                   struct ceph_mds_reply_info_in *info)
  75 {
  76         u8 struct_v, struct_compat;
  77         u32 struct_len;
  78
  79         ceph_decode_8_safe(p, end, struct_v, bad);
  80         ceph_decode_8_safe(p, end, struct_compat, bad);
  81         /* struct_v is expected to be >= 1. we only
  82          * understand encoding with struct_compat == 1. */
  83         if (!struct_v || struct_compat != 1)
  84                 goto bad;
  85         ceph_decode_32_safe(p, end, struct_len, bad);
  86         ceph_decode_need(p, end, struct_len, bad);
  87         end = *p + struct_len;
  88         ceph_decode_64_safe(p, end, info->max_bytes, bad);
  89         ceph_decode_64_safe(p, end, info->max_files, bad);
  90         *p = end;
  91         return 0;
  92 bad:
  93         return -EIO;
  94 }
  95
  96 /*
  97  * parse individual inode info
  98  */
  99 static int parse_reply_info_in(void **p, void *end,
 100                                struct ceph_mds_reply_info_in *info,
 101                                u64 features)
 102 {
 103         int err = 0;
 104         u8 struct_v = 0;
 105
 106         if (features == (u64)-1) {
 107                 u32 struct_len;
 108                 u8 struct_compat;
 109                 ceph_decode_8_safe(p, end, struct_v, bad);
 110                 ceph_decode_8_safe(p, end, struct_compat, bad);
 111                 /* struct_v is expected to be >= 1. we only understand
 112                  * encoding with struct_compat == 1. */
 113                 if (!struct_v || struct_compat != 1)
 114                         goto bad;
 115                 ceph_decode_32_safe(p, end, struct_len, bad);
 116                 ceph_decode_need(p, end, struct_len, bad);
 117                 end = *p + struct_len;
 118         }
 119
 120         ceph_decode_need(p, end, sizeof(struct ceph_mds_reply_inode), bad);
 121         info->in = *p;
 122         *p += sizeof(struct ceph_mds_reply_inode) +
 123                 sizeof(*info->in->fragtree.splits) *
 124                 le32_to_cpu(info->in->fragtree.nsplits);
 125
 126         ceph_decode_32_safe(p, end, info->symlink_len, bad);
 127         ceph_decode_need(p, end, info->symlink_len, bad);
 128         info->symlink = *p;
 129         *p += info->symlink_len;
 130
 131         ceph_decode_copy_safe(p, end, &info->dir_layout,
 132                               sizeof(info->dir_layout), bad);
 133         ceph_decode_32_safe(p, end, info->xattr_len, bad);
 134         ceph_decode_need(p, end, info->xattr_len, bad);
 135         info->xattr_data = *p;
 136         *p += info->xattr_len;
 137
 138         if (features == (u64)-1) {
 139                 /* inline data */
 140                 ceph_decode_64_safe(p, end, info->inline_version, bad);
 141                 ceph_decode_32_safe(p, end, info->inline_len, bad);
 142                 ceph_decode_need(p, end, info->inline_len, bad);
 143                 info->inline_data = *p;
 144                 *p += info->inline_len;
 145                 /* quota */
 146                 err = parse_reply_info_quota(p, end, info);
 147                 if (err < 0)
 148                         goto out_bad;
 149                 /* pool namespace */
 150                 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
 151                 if (info->pool_ns_len > 0) {
 152                         ceph_decode_need(p, end, info->pool_ns_len, bad);
 153                         info->pool_ns_data = *p;
 154                         *p += info->pool_ns_len;
 155                 }
 156
 157                 /* btime */
 158                 ceph_decode_need(p, end, sizeof(info->btime), bad);
 159                 ceph_decode_copy(p, &info->btime, sizeof(info->btime));
 160
 161                 /* change attribute */
 162                 ceph_decode_64_safe(p, end, info->change_attr, bad);
 163
 164                 /* dir pin */
 165                 if (struct_v >= 2) {
 166                         ceph_decode_32_safe(p, end, info->dir_pin, bad);
 167                 } else {
 168                         info->dir_pin = -ENODATA;
 169                 }
 170
 171                 /* snapshot birth time, remains zero for v<=2 */
 172                 if (struct_v >= 3) {
 173                         ceph_decode_need(p, end, sizeof(info->snap_btime), bad);
 174                         ceph_decode_copy(p, &info->snap_btime,
 175                                          sizeof(info->snap_btime));
 176                 } else {
 177                         memset(&info->snap_btime, 0, sizeof(info->snap_btime));
 178                 }
 179
 180                 /* snapshot count, remains zero for v<=3 */
 181                 if (struct_v >= 4) {
 182                         ceph_decode_64_safe(p, end, info->rsnaps, bad);
 183                 } else {
 184                         info->rsnaps = 0;
 185                 }
 186
 187                 *p = end;
 188         } else {
 189                 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
 190                         ceph_decode_64_safe(p, end, info->inline_version, bad);
 191                         ceph_decode_32_safe(p, end, info->inline_len, bad);
 192                         ceph_decode_need(p, end, info->inline_len, bad);
 193                         info->inline_data = *p;
 194                         *p += info->inline_len;
 195                 } else
 196                         info->inline_version = CEPH_INLINE_NONE;
 197
 198                 if (features & CEPH_FEATURE_MDS_QUOTA) {
 199                         err = parse_reply_info_quota(p, end, info);
 200                         if (err < 0)
 201                                 goto out_bad;
 202                 } else {
 203                         info->max_bytes = 0;
 204                         info->max_files = 0;
 205                 }
 206
 207                 info->pool_ns_len = 0;
 208                 info->pool_ns_data = NULL;
 209                 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
 210                         ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
 211                         if (info->pool_ns_len > 0) {
 212                                 ceph_decode_need(p, end, info->pool_ns_len, bad);
 213                                 info->pool_ns_data = *p;
 214                                 *p += info->pool_ns_len;
 215                         }
 216                 }
 217
 218                 if (features & CEPH_FEATURE_FS_BTIME) {
 219                         ceph_decode_need(p, end, sizeof(info->btime), bad);
 220                         ceph_decode_copy(p, &info->btime, sizeof(info->btime));
 221                         ceph_decode_64_safe(p, end, info->change_attr, bad);
 222                 }
 223
 224                 info->dir_pin = -ENODATA;
 225                 /* info->snap_btime and info->rsnaps remain zero */
 226         }
 227         return 0;
 228 bad:
 229         err = -EIO;
 230 out_bad:
 231         return err;
 232 }
 233
 234 static int parse_reply_info_dir(void **p, void *end,
 235                                 struct ceph_mds_reply_dirfrag **dirfrag,
 236                                 u64 features)
 237 {
 238         if (features == (u64)-1) {
 239                 u8 struct_v, struct_compat;
 240                 u32 struct_len;
 241                 ceph_decode_8_safe(p, end, struct_v, bad);
 242                 ceph_decode_8_safe(p, end, struct_compat, bad);
 243                 /* struct_v is expected to be >= 1. we only understand
 244                  * encoding whose struct_compat == 1. */
 245                 if (!struct_v || struct_compat != 1)
 246                         goto bad;
 247                 ceph_decode_32_safe(p, end, struct_len, bad);
 248                 ceph_decode_need(p, end, struct_len, bad);
 249                 end = *p + struct_len;
 250         }
 251
 252         ceph_decode_need(p, end, sizeof(**dirfrag), bad);
 253         *dirfrag = *p;
 254         *p += sizeof(**dirfrag) + sizeof(u32) * le32_to_cpu((*dirfrag)->ndist);
 255         if (unlikely(*p > end))
 256                 goto bad;
 257         if (features == (u64)-1)
 258                 *p = end;
 259         return 0;
 260 bad:
 261         return -EIO;
 262 }
 263
 264 static int parse_reply_info_lease(void **p, void *end,
 265                                   struct ceph_mds_reply_lease **lease,
 266                                   u64 features)
 267 {
 268         if (features == (u64)-1) {
 269                 u8 struct_v, struct_compat;
 270                 u32 struct_len;
 271                 ceph_decode_8_safe(p, end, struct_v, bad);
 272                 ceph_decode_8_safe(p, end, struct_compat, bad);
 273                 /* struct_v is expected to be >= 1. we only understand
 274                  * encoding whose struct_compat == 1. */
 275                 if (!struct_v || struct_compat != 1)
 276                         goto bad;
 277                 ceph_decode_32_safe(p, end, struct_len, bad);
 278                 ceph_decode_need(p, end, struct_len, bad);
 279                 end = *p + struct_len;
 280         }
 281
 282         ceph_decode_need(p, end, sizeof(**lease), bad);
 283         *lease = *p;
 284         *p += sizeof(**lease);
 285         if (features == (u64)-1)
 286                 *p = end;
 287         return 0;
 288 bad:
 289         return -EIO;
 290 }
 291
 292 /*
 293  * parse a normal reply, which may contain a (dir+)dentry and/or a
 294  * target inode.
 295  */
 296 static int parse_reply_info_trace(void **p, void *end,
 297                                   struct ceph_mds_reply_info_parsed *info,
 298                                   u64 features)
 299 {
 300         int err;
 301
 302         if (info->head->is_dentry) {
 303                 err = parse_reply_info_in(p, end, &info->diri, features);
 304                 if (err < 0)
 305                         goto out_bad;
 306
 307                 err = parse_reply_info_dir(p, end, &info->dirfrag, features);
 308                 if (err < 0)
 309                         goto out_bad;
 310
 311                 ceph_decode_32_safe(p, end, info->dname_len, bad);
 312                 ceph_decode_need(p, end, info->dname_len, bad);
 313                 info->dname = *p;
 314                 *p += info->dname_len;
 315
 316                 err = parse_reply_info_lease(p, end, &info->dlease, features);
 317                 if (err < 0)
 318                         goto out_bad;
 319         }
 320
 321         if (info->head->is_target) {
 322                 err = parse_reply_info_in(p, end, &info->targeti, features);
 323                 if (err < 0)
 324                         goto out_bad;
 325         }
 326
 327         if (unlikely(*p != end))
 328                 goto bad;
 329         return 0;
 330
 331 bad:
 332         err = -EIO;
 333 out_bad:
 334         pr_err("problem parsing mds trace %d\n", err);
 335         return err;
 336 }
 337
 338 /*
 339  * parse readdir results
 340  */
 341 static int parse_reply_info_readdir(void **p, void *end,
 342                                 struct ceph_mds_reply_info_parsed *info,
 343                                 u64 features)
 344 {
 345         u32 num, i = 0;
 346         int err;
 347
 348         err = parse_reply_info_dir(p, end, &info->dir_dir, features);
 349         if (err < 0)
 350                 goto out_bad;
 351
 352         ceph_decode_need(p, end, sizeof(num) + 2, bad);
 353         num = ceph_decode_32(p);
 354         {
 355                 u16 flags = ceph_decode_16(p);
 356                 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
 357                 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
 358                 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
 359                 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
 360         }
 361         if (num == 0)
 362                 goto done;
 363
 364         BUG_ON(!info->dir_entries);
 365         if ((unsigned long)(info->dir_entries + num) >
 366             (unsigned long)info->dir_entries + info->dir_buf_size) {
 367                 pr_err("dir contents are larger than expected\n");
 368                 WARN_ON(1);
 369                 goto bad;
 370         }
 371
 372         info->dir_nr = num;
 373         while (num) {
 374                 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
 375                 /* dentry */
 376                 ceph_decode_32_safe(p, end, rde->name_len, bad);
 377                 ceph_decode_need(p, end, rde->name_len, bad);
 378                 rde->name = *p;
 379                 *p += rde->name_len;
 380                 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
 381
 382                 /* dentry lease */
 383                 err = parse_reply_info_lease(p, end, &rde->lease, features);
 384                 if (err)
 385                         goto out_bad;
 386                 /* inode */
 387                 err = parse_reply_info_in(p, end, &rde->inode, features);
 388                 if (err < 0)
 389                         goto out_bad;
 390                 /* ceph_readdir_prepopulate() will update it */
 391                 rde->offset = 0;
 392                 i++;
 393                 num--;
 394         }
 395
 396 done:
 397         /* Skip over any unrecognized fields */
 398         *p = end;
 399         return 0;
 400
 401 bad:
 402         err = -EIO;
 403 out_bad:
 404         pr_err("problem parsing dir contents %d\n", err);
 405         return err;
 406 }
 407
 408 /*
 409  * parse fcntl F_GETLK results
 410  */
 411 static int parse_reply_info_filelock(void **p, void *end,
 412                                      struct ceph_mds_reply_info_parsed *info,
 413                                      u64 features)
 414 {
 415         if (*p + sizeof(*info->filelock_reply) > end)
 416                 goto bad;
 417
 418         info->filelock_reply = *p;
 419
 420         /* Skip over any unrecognized fields */
 421         *p = end;
 422         return 0;
 423 bad:
 424         return -EIO;
 425 }
 426
 427
 428 #if BITS_PER_LONG == 64
 429
 430 #define DELEGATED_INO_AVAILABLE         xa_mk_value(1)
 431
 432 static int ceph_parse_deleg_inos(void **p, void *end,
 433                                  struct ceph_mds_session *s)
 434 {
 435         u32 sets;
 436
 437         ceph_decode_32_safe(p, end, sets, bad);
 438         dout("got %u sets of delegated inodes\n", sets);
 439         while (sets--) {
 440                 u64 start, len, ino;
 441
 442                 ceph_decode_64_safe(p, end, start, bad);
 443                 ceph_decode_64_safe(p, end, len, bad);
 444
 445                 /* Don't accept a delegation of system inodes */
 446                 if (start < CEPH_INO_SYSTEM_BASE) {
 447                         pr_warn_ratelimited("ceph: ignoring reserved inode range delegation (start=0x%llx len=0x%llx)\n",
 448                                         start, len);
 449                         continue;
 450                 }
 451                 while (len--) {
 452                         int err = xa_insert(&s->s_delegated_inos, ino = start++,
 453                                             DELEGATED_INO_AVAILABLE,
 454                                             GFP_KERNEL);
 455                         if (!err) {
 456                                 dout("added delegated inode 0x%llx\n",
 457                                      start - 1);
 458                         } else if (err == -EBUSY) {
 459                                 pr_warn("ceph: MDS delegated inode 0x%llx more than once.\n",
 460                                         start - 1);
 461                         } else {
 462                                 return err;
 463                         }
 464                 }
 465         }
 466         return 0;
 467 bad:
 468         return -EIO;
 469 }
 470
 471 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
 472 {
 473         unsigned long ino;
 474         void *val;
 475
 476         xa_for_each(&s->s_delegated_inos, ino, val) {
 477                 val = xa_erase(&s->s_delegated_inos, ino);
 478                 if (val == DELEGATED_INO_AVAILABLE)
 479                         return ino;
 480         }
 481         return 0;
 482 }
 483
 484 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
 485 {
 486         return xa_insert(&s->s_delegated_inos, ino, DELEGATED_INO_AVAILABLE,
 487                          GFP_KERNEL);
 488 }
 489 #else /* BITS_PER_LONG == 64 */
 490 /*
 491  * FIXME: xarrays can't handle 64-bit indexes on a 32-bit arch. For now, just
 492  * ignore delegated_inos on 32 bit arch. Maybe eventually add xarrays for top
 493  * and bottom words?
 494  */
 495 static int ceph_parse_deleg_inos(void **p, void *end,
 496                                  struct ceph_mds_session *s)
 497 {
 498         u32 sets;
 499
 500         ceph_decode_32_safe(p, end, sets, bad);
 501         if (sets)
 502                 ceph_decode_skip_n(p, end, sets * 2 * sizeof(__le64), bad);
 503         return 0;
 504 bad:
 505         return -EIO;
 506 }
 507
 508 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
 509 {
 510         return 0;
 511 }
 512
 513 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
 514 {
 515         return 0;
 516 }
 517 #endif /* BITS_PER_LONG == 64 */
 518
 519 /*
 520  * parse create results
 521  */
 522 static int parse_reply_info_create(void **p, void *end,
 523                                   struct ceph_mds_reply_info_parsed *info,
 524                                   u64 features, struct ceph_mds_session *s)
 525 {
 526         int ret;
 527
 528         if (features == (u64)-1 ||
 529             (features & CEPH_FEATURE_REPLY_CREATE_INODE)) {
 530                 if (*p == end) {
 531                         /* Malformed reply? */
 532                         info->has_create_ino = false;
 533                 } else if (test_bit(CEPHFS_FEATURE_DELEG_INO, &s->s_features)) {
 534                         info->has_create_ino = true;
 535                         /* struct_v, struct_compat, and len */
 536                         ceph_decode_skip_n(p, end, 2 + sizeof(u32), bad);
 537                         ceph_decode_64_safe(p, end, info->ino, bad);
 538                         ret = ceph_parse_deleg_inos(p, end, s);
 539                         if (ret)
 540                                 return ret;
 541                 } else {
 542                         /* legacy */
 543                         ceph_decode_64_safe(p, end, info->ino, bad);
 544                         info->has_create_ino = true;
 545                 }
 546         } else {
 547                 if (*p != end)
 548                         goto bad;
 549         }
 550
 551         /* Skip over any unrecognized fields */
 552         *p = end;
 553         return 0;
 554 bad:
 555         return -EIO;
 556 }
 557
 558 /*
 559  * parse extra results
 560  */
 561 static int parse_reply_info_extra(void **p, void *end,
 562                                   struct ceph_mds_reply_info_parsed *info,
 563                                   u64 features, struct ceph_mds_session *s)
 564 {
 565         u32 op = le32_to_cpu(info->head->op);
 566
 567         if (op == CEPH_MDS_OP_GETFILELOCK)
 568                 return parse_reply_info_filelock(p, end, info, features);
 569         else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
 570                 return parse_reply_info_readdir(p, end, info, features);
 571         else if (op == CEPH_MDS_OP_CREATE)
 572                 return parse_reply_info_create(p, end, info, features, s);
 573         else
 574                 return -EIO;
 575 }
 576
 577 /*
 578  * parse entire mds reply
 579  */
 580 static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,
 581                             struct ceph_mds_reply_info_parsed *info,
 582                             u64 features)
 583 {
 584         void *p, *end;
 585         u32 len;
 586         int err;
 587
 588         info->head = msg->front.iov_base;
 589         p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
 590         end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
 591
 592         /* trace */
 593         ceph_decode_32_safe(&p, end, len, bad);
 594         if (len > 0) {
 595                 ceph_decode_need(&p, end, len, bad);
 596                 err = parse_reply_info_trace(&p, p+len, info, features);
 597                 if (err < 0)
 598                         goto out_bad;
 599         }
 600
 601         /* extra */
 602         ceph_decode_32_safe(&p, end, len, bad);
 603         if (len > 0) {
 604                 ceph_decode_need(&p, end, len, bad);
 605                 err = parse_reply_info_extra(&p, p+len, info, features, s);
 606                 if (err < 0)
 607                         goto out_bad;
 608         }
 609
 610         /* snap blob */
 611         ceph_decode_32_safe(&p, end, len, bad);
 612         info->snapblob_len = len;
 613         info->snapblob = p;
 614         p += len;
 615
 616         if (p != end)
 617                 goto bad;
 618         return 0;
 619
 620 bad:
 621         err = -EIO;
 622 out_bad:
 623         pr_err("mds parse_reply err %d\n", err);
 624         return err;
 625 }
 626
 627 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
 628 {
 629         if (!info->dir_entries)
 630                 return;
 631         free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
 632 }
 633
 634
 635 /*
 636  * sessions
 637  */
 638 const char *ceph_session_state_name(int s)
 639 {
 640         switch (s) {
 641         case CEPH_MDS_SESSION_NEW: return "new";
 642         case CEPH_MDS_SESSION_OPENING: return "opening";
 643         case CEPH_MDS_SESSION_OPEN: return "open";
 644         case CEPH_MDS_SESSION_HUNG: return "hung";
 645         case CEPH_MDS_SESSION_CLOSING: return "closing";
 646         case CEPH_MDS_SESSION_CLOSED: return "closed";
 647         case CEPH_MDS_SESSION_RESTARTING: return "restarting";
 648         case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
 649         case CEPH_MDS_SESSION_REJECTED: return "rejected";
 650         default: return "???";
 651         }
 652 }
 653
 654 struct ceph_mds_session *ceph_get_mds_session(struct ceph_mds_session *s)
 655 {
 656         if (refcount_inc_not_zero(&s->s_ref)) {
 657                 dout("mdsc get_session %p %d -> %d\n", s,
 658                      refcount_read(&s->s_ref)-1, refcount_read(&s->s_ref));
 659                 return s;
 660         } else {
 661                 dout("mdsc get_session %p 0 -- FAIL\n", s);
 662                 return NULL;
 663         }
 664 }
 665
 666 void ceph_put_mds_session(struct ceph_mds_session *s)
 667 {
 668         if (IS_ERR_OR_NULL(s))
 669                 return;
 670
 671         dout("mdsc put_session %p %d -> %d\n", s,
 672              refcount_read(&s->s_ref), refcount_read(&s->s_ref)-1);
 673         if (refcount_dec_and_test(&s->s_ref)) {
 674                 if (s->s_auth.authorizer)
 675                         ceph_auth_destroy_authorizer(s->s_auth.authorizer);
 676                 WARN_ON(mutex_is_locked(&s->s_mutex));
 677                 xa_destroy(&s->s_delegated_inos);
 678                 kfree(s);
 679         }
 680 }
 681
 682 /*
 683  * called under mdsc->mutex
 684  */
 685 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
 686                                                    int mds)
 687 {
 688         if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
 689                 return NULL;
 690         return ceph_get_mds_session(mdsc->sessions[mds]);
 691 }
 692
 693 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
 694 {
 695         if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
 696                 return false;
 697         else
 698                 return true;
 699 }
 700
 701 static int __verify_registered_session(struct ceph_mds_client *mdsc,
 702                                        struct ceph_mds_session *s)
 703 {
 704         if (s->s_mds >= mdsc->max_sessions ||
 705             mdsc->sessions[s->s_mds] != s)
 706                 return -ENOENT;
 707         return 0;
 708 }
 709
 710 /*
 711  * create+register a new session for given mds.
 712  * called under mdsc->mutex.
 713  */
 714 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
 715                                                  int mds)
 716 {
 717         struct ceph_mds_session *s;
 718
 719         if (mds >= mdsc->mdsmap->possible_max_rank)
 720                 return ERR_PTR(-EINVAL);
 721
 722         s = kzalloc(sizeof(*s), GFP_NOFS);
 723         if (!s)
 724                 return ERR_PTR(-ENOMEM);
 725
 726         if (mds >= mdsc->max_sessions) {
 727                 int newmax = 1 << get_count_order(mds + 1);
 728                 struct ceph_mds_session **sa;
 729
 730                 dout("%s: realloc to %d\n", __func__, newmax);
 731                 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
 732                 if (!sa)
 733                         goto fail_realloc;
 734                 if (mdsc->sessions) {
 735                         memcpy(sa, mdsc->sessions,
 736                                mdsc->max_sessions * sizeof(void *));
 737                         kfree(mdsc->sessions);
 738                 }
 739                 mdsc->sessions = sa;
 740                 mdsc->max_sessions = newmax;
 741         }
 742
 743         dout("%s: mds%d\n", __func__, mds);
 744         s->s_mdsc = mdsc;
 745         s->s_mds = mds;
 746         s->s_state = CEPH_MDS_SESSION_NEW;
 747         mutex_init(&s->s_mutex);
 748
 749         ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
 750
 751         atomic_set(&s->s_cap_gen, 1);
 752         s->s_cap_ttl = jiffies - 1;
 753
 754         spin_lock_init(&s->s_cap_lock);
 755         INIT_LIST_HEAD(&s->s_caps);
 756         refcount_set(&s->s_ref, 1);
 757         INIT_LIST_HEAD(&s->s_waiting);
 758         INIT_LIST_HEAD(&s->s_unsafe);
 759         xa_init(&s->s_delegated_inos);
 760         INIT_LIST_HEAD(&s->s_cap_releases);
 761         INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
 762
 763         INIT_LIST_HEAD(&s->s_cap_dirty);
 764         INIT_LIST_HEAD(&s->s_cap_flushing);
 765
 766         mdsc->sessions[mds] = s;
 767         atomic_inc(&mdsc->num_sessions);
 768         refcount_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
 769
 770         ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
 771                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
 772
 773         return s;
 774
 775 fail_realloc:
 776         kfree(s);
 777         return ERR_PTR(-ENOMEM);
 778 }
 779
 780 /*
 781  * called under mdsc->mutex
 782  */
 783 static void __unregister_session(struct ceph_mds_client *mdsc,
 784                                struct ceph_mds_session *s)
 785 {
 786         dout("__unregister_session mds%d %p\n", s->s_mds, s);
 787         BUG_ON(mdsc->sessions[s->s_mds] != s);
 788         mdsc->sessions[s->s_mds] = NULL;
 789         ceph_con_close(&s->s_con);
 790         ceph_put_mds_session(s);
 791         atomic_dec(&mdsc->num_sessions);
 792 }
 793
 794 /*
 795  * drop session refs in request.
 796  *
 797  * should be last request ref, or hold mdsc->mutex
 798  */
 799 static void put_request_session(struct ceph_mds_request *req)
 800 {
 801         if (req->r_session) {
 802                 ceph_put_mds_session(req->r_session);
 803                 req->r_session = NULL;
 804         }
 805 }
 806
 807 void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
 808                                 void (*cb)(struct ceph_mds_session *),
 809                                 bool check_state)
 810 {
 811         int mds;
 812
 813         mutex_lock(&mdsc->mutex);
 814         for (mds = 0; mds < mdsc->max_sessions; ++mds) {
 815                 struct ceph_mds_session *s;
 816
 817                 s = __ceph_lookup_mds_session(mdsc, mds);
 818                 if (!s)
 819                         continue;
 820
 821                 if (check_state && !check_session_state(s)) {
 822                         ceph_put_mds_session(s);
 823                         continue;
 824                 }
 825
 826                 mutex_unlock(&mdsc->mutex);
 827                 cb(s);
 828                 ceph_put_mds_session(s);
 829                 mutex_lock(&mdsc->mutex);
 830         }
 831         mutex_unlock(&mdsc->mutex);
 832 }
 833
 834 void ceph_mdsc_release_request(struct kref *kref)
 835 {
 836         struct ceph_mds_request *req = container_of(kref,
 837                                                     struct ceph_mds_request,
 838                                                     r_kref);
 839         ceph_mdsc_release_dir_caps_no_check(req);
 840         destroy_reply_info(&req->r_reply_info);
 841         if (req->r_request)
 842                 ceph_msg_put(req->r_request);
 843         if (req->r_reply)
 844                 ceph_msg_put(req->r_reply);
 845         if (req->r_inode) {
 846                 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
 847                 iput(req->r_inode);
 848         }
 849         if (req->r_parent) {
 850                 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
 851                 iput(req->r_parent);
 852         }
 853         iput(req->r_target_inode);
 854         if (req->r_dentry)
 855                 dput(req->r_dentry);
 856         if (req->r_old_dentry)
 857                 dput(req->r_old_dentry);
 858         if (req->r_old_dentry_dir) {
 859                 /*
 860                  * track (and drop pins for) r_old_dentry_dir
 861                  * separately, since r_old_dentry's d_parent may have
 862                  * changed between the dir mutex being dropped and
 863                  * this request being freed.
 864                  */
 865                 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
 866                                   CEPH_CAP_PIN);
 867                 iput(req->r_old_dentry_dir);
 868         }
 869         kfree(req->r_path1);
 870         kfree(req->r_path2);
 871         put_cred(req->r_cred);
 872         if (req->r_pagelist)
 873                 ceph_pagelist_release(req->r_pagelist);
 874         put_request_session(req);
 875         ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
 876         WARN_ON_ONCE(!list_empty(&req->r_wait));
 877         kmem_cache_free(ceph_mds_request_cachep, req);
 878 }
 879
 880 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
 881
 882 /*
 883  * lookup session, bump ref if found.
 884  *
 885  * called under mdsc->mutex.
 886  */
 887 static struct ceph_mds_request *
 888 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
 889 {
 890         struct ceph_mds_request *req;
 891
 892         req = lookup_request(&mdsc->request_tree, tid);
 893         if (req)
 894                 ceph_mdsc_get_request(req);
 895
 896         return req;
 897 }
 898
 899 /*
 900  * Register an in-flight request, and assign a tid.  Link to directory
 901  * are modifying (if any).
 902  *
 903  * Called under mdsc->mutex.
 904  */
 905 static void __register_request(struct ceph_mds_client *mdsc,
 906                                struct ceph_mds_request *req,
 907                                struct inode *dir)
 908 {
 909         int ret = 0;
 910
 911         req->r_tid = ++mdsc->last_tid;
 912         if (req->r_num_caps) {
 913                 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
 914                                         req->r_num_caps);
 915                 if (ret < 0) {
 916                         pr_err("__register_request %p "
 917                                "failed to reserve caps: %d\n", req, ret);
 918                         /* set req->r_err to fail early from __do_request */
 919                         req->r_err = ret;
 920                         return;
 921                 }
 922         }
 923         dout("__register_request %p tid %lld\n", req, req->r_tid);
 924         ceph_mdsc_get_request(req);
 925         insert_request(&mdsc->request_tree, req);
 926
 927         req->r_cred = get_current_cred();
 928
 929         if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
 930                 mdsc->oldest_tid = req->r_tid;
 931
 932         if (dir) {
 933                 struct ceph_inode_info *ci = ceph_inode(dir);
 934
 935                 ihold(dir);
 936                 req->r_unsafe_dir = dir;
 937                 spin_lock(&ci->i_unsafe_lock);
 938                 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
 939                 spin_unlock(&ci->i_unsafe_lock);
 940         }
 941 }
 942
 943 static void __unregister_request(struct ceph_mds_client *mdsc,
 944                                  struct ceph_mds_request *req)
 945 {
 946         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
 947
 948         /* Never leave an unregistered request on an unsafe list! */
 949         list_del_init(&req->r_unsafe_item);
 950
 951         if (req->r_tid == mdsc->oldest_tid) {
 952                 struct rb_node *p = rb_next(&req->r_node);
 953                 mdsc->oldest_tid = 0;
 954                 while (p) {
 955                         struct ceph_mds_request *next_req =
 956                                 rb_entry(p, struct ceph_mds_request, r_node);
 957                         if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
 958                                 mdsc->oldest_tid = next_req->r_tid;
 959                                 break;
 960                         }
 961                         p = rb_next(p);
 962                 }
 963         }
 964
 965         erase_request(&mdsc->request_tree, req);
 966
 967         if (req->r_unsafe_dir) {
 968                 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
 969                 spin_lock(&ci->i_unsafe_lock);
 970                 list_del_init(&req->r_unsafe_dir_item);
 971                 spin_unlock(&ci->i_unsafe_lock);
 972         }
 973         if (req->r_target_inode &&
 974             test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
 975                 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
 976                 spin_lock(&ci->i_unsafe_lock);
 977                 list_del_init(&req->r_unsafe_target_item);
 978                 spin_unlock(&ci->i_unsafe_lock);
 979         }
 980
 981         if (req->r_unsafe_dir) {
 982                 iput(req->r_unsafe_dir);
 983                 req->r_unsafe_dir = NULL;
 984         }
 985
 986         complete_all(&req->r_safe_completion);
 987
 988         ceph_mdsc_put_request(req);
 989 }
 990
 991 /*
 992  * Walk back up the dentry tree until we hit a dentry representing a
 993  * non-snapshot inode. We do this using the rcu_read_lock (which must be held
 994  * when calling this) to ensure that the objects won't disappear while we're
 995  * working with them. Once we hit a candidate dentry, we attempt to take a
 996  * reference to it, and return that as the result.
 997  */
 998 static struct inode *get_nonsnap_parent(struct dentry *dentry)
 999 {
1000         struct inode *inode = NULL;
1001
1002         while (dentry && !IS_ROOT(dentry)) {
1003                 inode = d_inode_rcu(dentry);
1004                 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
1005                         break;
1006                 dentry = dentry->d_parent;
1007         }
1008         if (inode)
1009                 inode = igrab(inode);
1010         return inode;
1011 }
1012
1013 /*
1014  * Choose mds to send request to next.  If there is a hint set in the
1015  * request (e.g., due to a prior forward hint from the mds), use that.
1016  * Otherwise, consult frag tree and/or caps to identify the
1017  * appropriate mds.  If all else fails, choose randomly.
1018  *
1019  * Called under mdsc->mutex.
1020  */
1021 static int __choose_mds(struct ceph_mds_client *mdsc,
1022                         struct ceph_mds_request *req,
1023                         bool *random)
1024 {
1025         struct inode *inode;
1026         struct ceph_inode_info *ci;
1027         struct ceph_cap *cap;
1028         int mode = req->r_direct_mode;
1029         int mds = -1;
1030         u32 hash = req->r_direct_hash;
1031         bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
1032
1033         if (random)
1034                 *random = false;
1035
1036         /*
1037          * is there a specific mds we should try?  ignore hint if we have
1038          * no session and the mds is not up (active or recovering).
1039          */
1040         if (req->r_resend_mds >= 0 &&
1041             (__have_session(mdsc, req->r_resend_mds) ||
1042              ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
1043                 dout("%s using resend_mds mds%d\n", __func__,
1044                      req->r_resend_mds);
1045                 return req->r_resend_mds;
1046         }
1047
1048         if (mode == USE_RANDOM_MDS)
1049                 goto random;
1050
1051         inode = NULL;
1052         if (req->r_inode) {
1053                 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
1054                         inode = req->r_inode;
1055                         ihold(inode);
1056                 } else {
1057                         /* req->r_dentry is non-null for LSSNAP request */
1058                         rcu_read_lock();
1059                         inode = get_nonsnap_parent(req->r_dentry);
1060                         rcu_read_unlock();
1061                         dout("%s using snapdir's parent %p\n", __func__, inode);
1062                 }
1063         } else if (req->r_dentry) {
1064                 /* ignore race with rename; old or new d_parent is okay */
1065                 struct dentry *parent;
1066                 struct inode *dir;
1067
1068                 rcu_read_lock();
1069                 parent = READ_ONCE(req->r_dentry->d_parent);
1070                 dir = req->r_parent ? : d_inode_rcu(parent);
1071
1072                 if (!dir || dir->i_sb != mdsc->fsc->sb) {
1073                         /*  not this fs or parent went negative */
1074                         inode = d_inode(req->r_dentry);
1075                         if (inode)
1076                                 ihold(inode);
1077                 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
1078                         /* direct snapped/virtual snapdir requests
1079                          * based on parent dir inode */
1080                         inode = get_nonsnap_parent(parent);
1081                         dout("%s using nonsnap parent %p\n", __func__, inode);
1082                 } else {
1083                         /* dentry target */
1084                         inode = d_inode(req->r_dentry);
1085                         if (!inode || mode == USE_AUTH_MDS) {
1086                                 /* dir + name */
1087                                 inode = igrab(dir);
1088                                 hash = ceph_dentry_hash(dir, req->r_dentry);
1089                                 is_hash = true;
1090                         } else {
1091                                 ihold(inode);
1092                         }
1093                 }
1094                 rcu_read_unlock();
1095         }
1096
1097         dout("%s %p is_hash=%d (0x%x) mode %d\n", __func__, inode, (int)is_hash,
1098              hash, mode);
1099         if (!inode)
1100                 goto random;
1101         ci = ceph_inode(inode);
1102
1103         if (is_hash && S_ISDIR(inode->i_mode)) {
1104                 struct ceph_inode_frag frag;
1105                 int found;
1106
1107                 ceph_choose_frag(ci, hash, &frag, &found);
1108                 if (found) {
1109                         if (mode == USE_ANY_MDS && frag.ndist > 0) {
1110                                 u8 r;
1111
1112                                 /* choose a random replica */
1113                                 get_random_bytes(&r, 1);
1114                                 r %= frag.ndist;
1115                                 mds = frag.dist[r];
1116                                 dout("%s %p %llx.%llx frag %u mds%d (%d/%d)\n",
1117                                      __func__, inode, ceph_vinop(inode),
1118                                      frag.frag, mds, (int)r, frag.ndist);
1119                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1120                                     CEPH_MDS_STATE_ACTIVE &&
1121                                     !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
1122                                         goto out;
1123                         }
1124
1125                         /* since this file/dir wasn't known to be
1126                          * replicated, then we want to look for the
1127                          * authoritative mds. */
1128                         if (frag.mds >= 0) {
1129                                 /* choose auth mds */
1130                                 mds = frag.mds;
1131                                 dout("%s %p %llx.%llx frag %u mds%d (auth)\n",
1132                                      __func__, inode, ceph_vinop(inode),
1133                                      frag.frag, mds);
1134                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1135                                     CEPH_MDS_STATE_ACTIVE) {
1136                                         if (!ceph_mdsmap_is_laggy(mdsc->mdsmap,
1137                                                                   mds))
1138                                                 goto out;
1139                                 }
1140                         }
1141                         mode = USE_AUTH_MDS;
1142                 }
1143         }
1144
1145         spin_lock(&ci->i_ceph_lock);
1146         cap = NULL;
1147         if (mode == USE_AUTH_MDS)
1148                 cap = ci->i_auth_cap;
1149         if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1150                 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1151         if (!cap) {
1152                 spin_unlock(&ci->i_ceph_lock);
1153                 iput(inode);
1154                 goto random;
1155         }
1156         mds = cap->session->s_mds;
1157         dout("%s %p %llx.%llx mds%d (%scap %p)\n", __func__,
1158              inode, ceph_vinop(inode), mds,
1159              cap == ci->i_auth_cap ? "auth " : "", cap);
1160         spin_unlock(&ci->i_ceph_lock);
1161 out:
1162         iput(inode);
1163         return mds;
1164
1165 random:
1166         if (random)
1167                 *random = true;
1168
1169         mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
1170         dout("%s chose random mds%d\n", __func__, mds);
1171         return mds;
1172 }
1173
1174
1175 /*
1176  * session messages
1177  */
1178 struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq)
1179 {
1180         struct ceph_msg *msg;
1181         struct ceph_mds_session_head *h;
1182
1183         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1184                            false);
1185         if (!msg) {
1186                 pr_err("ENOMEM creating session %s msg\n",
1187                        ceph_session_op_name(op));
1188                 return NULL;
1189         }
1190         h = msg->front.iov_base;
1191         h->op = cpu_to_le32(op);
1192         h->seq = cpu_to_le64(seq);
1193
1194         return msg;
1195 }
1196
1197 static const unsigned char feature_bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1198 #define FEATURE_BYTES(c) (DIV_ROUND_UP((size_t)feature_bits[c - 1] + 1, 64) * 8)
1199 static int encode_supported_features(void **p, void *end)
1200 {
1201         static const size_t count = ARRAY_SIZE(feature_bits);
1202
1203         if (count > 0) {
1204                 size_t i;
1205                 size_t size = FEATURE_BYTES(count);
1206
1207                 if (WARN_ON_ONCE(*p + 4 + size > end))
1208                         return -ERANGE;
1209
1210                 ceph_encode_32(p, size);
1211                 memset(*p, 0, size);
1212                 for (i = 0; i < count; i++)
1213                         ((unsigned char*)(*p))[i / 8] |= BIT(feature_bits[i] % 8);
1214                 *p += size;
1215         } else {
1216                 if (WARN_ON_ONCE(*p + 4 > end))
1217                         return -ERANGE;
1218
1219                 ceph_encode_32(p, 0);
1220         }
1221
1222         return 0;
1223 }
1224
1225 static const unsigned char metric_bits[] = CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED;
1226 #define METRIC_BYTES(cnt) (DIV_ROUND_UP((size_t)metric_bits[cnt - 1] + 1, 64) * 8)
1227 static int encode_metric_spec(void **p, void *end)
1228 {
1229         static const size_t count = ARRAY_SIZE(metric_bits);
1230
1231         /* header */
1232         if (WARN_ON_ONCE(*p + 2 > end))
1233                 return -ERANGE;
1234
1235         ceph_encode_8(p, 1); /* version */
1236         ceph_encode_8(p, 1); /* compat */
1237
1238         if (count > 0) {
1239                 size_t i;
1240                 size_t size = METRIC_BYTES(count);
1241
1242                 if (WARN_ON_ONCE(*p + 4 + 4 + size > end))
1243                         return -ERANGE;
1244
1245                 /* metric spec info length */
1246                 ceph_encode_32(p, 4 + size);
1247
1248                 /* metric spec */
1249                 ceph_encode_32(p, size);
1250                 memset(*p, 0, size);
1251                 for (i = 0; i < count; i++)
1252                         ((unsigned char *)(*p))[i / 8] |= BIT(metric_bits[i] % 8);
1253                 *p += size;
1254         } else {
1255                 if (WARN_ON_ONCE(*p + 4 + 4 > end))
1256                         return -ERANGE;
1257
1258                 /* metric spec info length */
1259                 ceph_encode_32(p, 4);
1260                 /* metric spec */
1261                 ceph_encode_32(p, 0);
1262         }
1263
1264         return 0;
1265 }
1266
1267 /*
1268  * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1269  * to include additional client metadata fields.
1270  */
1271 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
1272 {
1273         struct ceph_msg *msg;
1274         struct ceph_mds_session_head *h;
1275         int i;
1276         int extra_bytes = 0;
1277         int metadata_key_count = 0;
1278         struct ceph_options *opt = mdsc->fsc->client->options;
1279         struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
1280         size_t size, count;
1281         void *p, *end;
1282         int ret;
1283
1284         const char* metadata[][2] = {
1285                 {"hostname", mdsc->nodename},
1286                 {"kernel_version", init_utsname()->release},
1287                 {"entity_id", opt->name ? : ""},
1288                 {"root", fsopt->server_path ? : "/"},
1289                 {NULL, NULL}
1290         };
1291
1292         /* Calculate serialized length of metadata */
1293         extra_bytes = 4;  /* map length */
1294         for (i = 0; metadata[i][0]; ++i) {
1295                 extra_bytes += 8 + strlen(metadata[i][0]) +
1296                         strlen(metadata[i][1]);
1297                 metadata_key_count++;
1298         }
1299
1300         /* supported feature */
1301         size = 0;
1302         count = ARRAY_SIZE(feature_bits);
1303         if (count > 0)
1304                 size = FEATURE_BYTES(count);
1305         extra_bytes += 4 + size;
1306
1307         /* metric spec */
1308         size = 0;
1309         count = ARRAY_SIZE(metric_bits);
1310         if (count > 0)
1311                 size = METRIC_BYTES(count);
1312         extra_bytes += 2 + 4 + 4 + size;
1313
1314         /* Allocate the message */
1315         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
1316                            GFP_NOFS, false);
1317         if (!msg) {
1318                 pr_err("ENOMEM creating session open msg\n");
1319                 return ERR_PTR(-ENOMEM);
1320         }
1321         p = msg->front.iov_base;
1322         end = p + msg->front.iov_len;
1323
1324         h = p;
1325         h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
1326         h->seq = cpu_to_le64(seq);
1327
1328         /*
1329          * Serialize client metadata into waiting buffer space, using
1330          * the format that userspace expects for map<string, string>
1331          *
1332          * ClientSession messages with metadata are v4
1333          */
1334         msg->hdr.version = cpu_to_le16(4);
1335         msg->hdr.compat_version = cpu_to_le16(1);
1336
1337         /* The write pointer, following the session_head structure */
1338         p += sizeof(*h);
1339
1340         /* Number of entries in the map */
1341         ceph_encode_32(&p, metadata_key_count);
1342
1343         /* Two length-prefixed strings for each entry in the map */
1344         for (i = 0; metadata[i][0]; ++i) {
1345                 size_t const key_len = strlen(metadata[i][0]);
1346                 size_t const val_len = strlen(metadata[i][1]);
1347
1348                 ceph_encode_32(&p, key_len);
1349                 memcpy(p, metadata[i][0], key_len);
1350                 p += key_len;
1351                 ceph_encode_32(&p, val_len);
1352                 memcpy(p, metadata[i][1], val_len);
1353                 p += val_len;
1354         }
1355
1356         ret = encode_supported_features(&p, end);
1357         if (ret) {
1358                 pr_err("encode_supported_features failed!\n");
1359                 ceph_msg_put(msg);
1360                 return ERR_PTR(ret);
1361         }
1362
1363         ret = encode_metric_spec(&p, end);
1364         if (ret) {
1365                 pr_err("encode_metric_spec failed!\n");
1366                 ceph_msg_put(msg);
1367                 return ERR_PTR(ret);
1368         }
1369
1370         msg->front.iov_len = p - msg->front.iov_base;
1371         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1372
1373         return msg;
1374 }
1375
1376 /*
1377  * send session open request.
1378  *
1379  * called under mdsc->mutex
1380  */
1381 static int __open_session(struct ceph_mds_client *mdsc,
1382                           struct ceph_mds_session *session)
1383 {
1384         struct ceph_msg *msg;
1385         int mstate;
1386         int mds = session->s_mds;
1387
1388         /* wait for mds to go active? */
1389         mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1390         dout("open_session to mds%d (%s)\n", mds,
1391              ceph_mds_state_name(mstate));
1392         session->s_state = CEPH_MDS_SESSION_OPENING;
1393         session->s_renew_requested = jiffies;
1394
1395         /* send connect message */
1396         msg = create_session_open_msg(mdsc, session->s_seq);
1397         if (IS_ERR(msg))
1398                 return PTR_ERR(msg);
1399         ceph_con_send(&session->s_con, msg);
1400         return 0;
1401 }
1402
1403 /*
1404  * open sessions for any export targets for the given mds
1405  *
1406  * called under mdsc->mutex
1407  */
1408 static struct ceph_mds_session *
1409 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1410 {
1411         struct ceph_mds_session *session;
1412         int ret;
1413
1414         session = __ceph_lookup_mds_session(mdsc, target);
1415         if (!session) {
1416                 session = register_session(mdsc, target);
1417                 if (IS_ERR(session))
1418                         return session;
1419         }
1420         if (session->s_state == CEPH_MDS_SESSION_NEW ||
1421             session->s_state == CEPH_MDS_SESSION_CLOSING) {
1422                 ret = __open_session(mdsc, session);
1423                 if (ret)
1424                         return ERR_PTR(ret);
1425         }
1426
1427         return session;
1428 }
1429
1430 struct ceph_mds_session *
1431 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1432 {
1433         struct ceph_mds_session *session;
1434
1435         dout("open_export_target_session to mds%d\n", target);
1436
1437         mutex_lock(&mdsc->mutex);
1438         session = __open_export_target_session(mdsc, target);
1439         mutex_unlock(&mdsc->mutex);
1440
1441         return session;
1442 }
1443
1444 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1445                                           struct ceph_mds_session *session)
1446 {
1447         struct ceph_mds_info *mi;
1448         struct ceph_mds_session *ts;
1449         int i, mds = session->s_mds;
1450
1451         if (mds >= mdsc->mdsmap->possible_max_rank)
1452                 return;
1453
1454         mi = &mdsc->mdsmap->m_info[mds];
1455         dout("open_export_target_sessions for mds%d (%d targets)\n",
1456              session->s_mds, mi->num_export_targets);
1457
1458         for (i = 0; i < mi->num_export_targets; i++) {
1459                 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1460                 ceph_put_mds_session(ts);
1461         }
1462 }
1463
1464 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1465                                            struct ceph_mds_session *session)
1466 {
1467         mutex_lock(&mdsc->mutex);
1468         __open_export_target_sessions(mdsc, session);
1469         mutex_unlock(&mdsc->mutex);
1470 }
1471
1472 /*
1473  * session caps
1474  */
1475
1476 static void detach_cap_releases(struct ceph_mds_session *session,
1477                                 struct list_head *target)
1478 {
1479         lockdep_assert_held(&session->s_cap_lock);
1480
1481         list_splice_init(&session->s_cap_releases, target);
1482         session->s_num_cap_releases = 0;
1483         dout("dispose_cap_releases mds%d\n", session->s_mds);
1484 }
1485
1486 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1487                                  struct list_head *dispose)
1488 {
1489         while (!list_empty(dispose)) {
1490                 struct ceph_cap *cap;
1491                 /* zero out the in-progress message */
1492                 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1493                 list_del(&cap->session_caps);
1494                 ceph_put_cap(mdsc, cap);
1495         }
1496 }
1497
1498 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1499                                      struct ceph_mds_session *session)
1500 {
1501         struct ceph_mds_request *req;
1502         struct rb_node *p;
1503         struct ceph_inode_info *ci;
1504
1505         dout("cleanup_session_requests mds%d\n", session->s_mds);
1506         mutex_lock(&mdsc->mutex);
1507         while (!list_empty(&session->s_unsafe)) {
1508                 req = list_first_entry(&session->s_unsafe,
1509                                        struct ceph_mds_request, r_unsafe_item);
1510                 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1511                                     req->r_tid);
1512                 if (req->r_target_inode) {
1513                         /* dropping unsafe change of inode's attributes */
1514                         ci = ceph_inode(req->r_target_inode);
1515                         errseq_set(&ci->i_meta_err, -EIO);
1516                 }
1517                 if (req->r_unsafe_dir) {
1518                         /* dropping unsafe directory operation */
1519                         ci = ceph_inode(req->r_unsafe_dir);
1520                         errseq_set(&ci->i_meta_err, -EIO);
1521                 }
1522                 __unregister_request(mdsc, req);
1523         }
1524         /* zero r_attempts, so kick_requests() will re-send requests */
1525         p = rb_first(&mdsc->request_tree);
1526         while (p) {
1527                 req = rb_entry(p, struct ceph_mds_request, r_node);
1528                 p = rb_next(p);
1529                 if (req->r_session &&
1530                     req->r_session->s_mds == session->s_mds)
1531                         req->r_attempts = 0;
1532         }
1533         mutex_unlock(&mdsc->mutex);
1534 }
1535
1536 /*
1537  * Helper to safely iterate over all caps associated with a session, with
1538  * special care taken to handle a racing __ceph_remove_cap().
1539  *
1540  * Caller must hold session s_mutex.
1541  */
1542 int ceph_iterate_session_caps(struct ceph_mds_session *session,
1543                               int (*cb)(struct inode *, struct ceph_cap *,
1544                                         void *), void *arg)
1545 {
1546         struct list_head *p;
1547         struct ceph_cap *cap;
1548         struct inode *inode, *last_inode = NULL;
1549         struct ceph_cap *old_cap = NULL;
1550         int ret;
1551
1552         dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1553         spin_lock(&session->s_cap_lock);
1554         p = session->s_caps.next;
1555         while (p != &session->s_caps) {
1556                 cap = list_entry(p, struct ceph_cap, session_caps);
1557                 inode = igrab(&cap->ci->vfs_inode);
1558                 if (!inode) {
1559                         p = p->next;
1560                         continue;
1561                 }
1562                 session->s_cap_iterator = cap;
1563                 spin_unlock(&session->s_cap_lock);
1564
1565                 if (last_inode) {
1566                         iput(last_inode);
1567                         last_inode = NULL;
1568                 }
1569                 if (old_cap) {
1570                         ceph_put_cap(session->s_mdsc, old_cap);
1571                         old_cap = NULL;
1572                 }
1573
1574                 ret = cb(inode, cap, arg);
1575                 last_inode = inode;
1576
1577                 spin_lock(&session->s_cap_lock);
1578                 p = p->next;
1579                 if (!cap->ci) {
1580                         dout("iterate_session_caps  finishing cap %p removal\n",
1581                              cap);
1582                         BUG_ON(cap->session != session);
1583                         cap->session = NULL;
1584                         list_del_init(&cap->session_caps);
1585                         session->s_nr_caps--;
1586                         atomic64_dec(&session->s_mdsc->metric.total_caps);
1587                         if (cap->queue_release)
1588                                 __ceph_queue_cap_release(session, cap);
1589                         else
1590                                 old_cap = cap;  /* put_cap it w/o locks held */
1591                 }
1592                 if (ret < 0)
1593                         goto out;
1594         }
1595         ret = 0;
1596 out:
1597         session->s_cap_iterator = NULL;
1598         spin_unlock(&session->s_cap_lock);
1599
1600         iput(last_inode);
1601         if (old_cap)
1602                 ceph_put_cap(session->s_mdsc, old_cap);
1603
1604         return ret;
1605 }
1606
1607 static int remove_capsnaps(struct ceph_mds_client *mdsc, struct inode *inode)
1608 {
1609         struct ceph_inode_info *ci = ceph_inode(inode);
1610         struct ceph_cap_snap *capsnap;
1611         int capsnap_release = 0;
1612
1613         lockdep_assert_held(&ci->i_ceph_lock);
1614
1615         dout("removing capsnaps, ci is %p, inode is %p\n", ci, inode);
1616
1617         while (!list_empty(&ci->i_cap_snaps)) {
1618                 capsnap = list_first_entry(&ci->i_cap_snaps,
1619                                            struct ceph_cap_snap, ci_item);
1620                 __ceph_remove_capsnap(inode, capsnap, NULL, NULL);
1621                 ceph_put_snap_context(capsnap->context);
1622                 ceph_put_cap_snap(capsnap);
1623                 capsnap_release++;
1624         }
1625         wake_up_all(&ci->i_cap_wq);
1626         wake_up_all(&mdsc->cap_flushing_wq);
1627         return capsnap_release;
1628 }
1629
1630 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1631                                   void *arg)
1632 {
1633         struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
1634         struct ceph_mds_client *mdsc = fsc->mdsc;
1635         struct ceph_inode_info *ci = ceph_inode(inode);
1636         LIST_HEAD(to_remove);
1637         bool dirty_dropped = false;
1638         bool invalidate = false;
1639         int capsnap_release = 0;
1640
1641         dout("removing cap %p, ci is %p, inode is %p\n",
1642              cap, ci, &ci->vfs_inode);
1643         spin_lock(&ci->i_ceph_lock);
1644         __ceph_remove_cap(cap, false);
1645         if (!ci->i_auth_cap) {
1646                 struct ceph_cap_flush *cf;
1647
1648                 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1649                         if (inode->i_data.nrpages > 0)
1650                                 invalidate = true;
1651                         if (ci->i_wrbuffer_ref > 0)
1652                                 mapping_set_error(&inode->i_data, -EIO);
1653                 }
1654
1655                 while (!list_empty(&ci->i_cap_flush_list)) {
1656                         cf = list_first_entry(&ci->i_cap_flush_list,
1657                                               struct ceph_cap_flush, i_list);
1658                         list_move(&cf->i_list, &to_remove);
1659                 }
1660
1661                 spin_lock(&mdsc->cap_dirty_lock);
1662
1663                 list_for_each_entry(cf, &to_remove, i_list)
1664                         list_del_init(&cf->g_list);
1665
1666                 if (!list_empty(&ci->i_dirty_item)) {
1667                         pr_warn_ratelimited(
1668                                 " dropping dirty %s state for %p %lld\n",
1669                                 ceph_cap_string(ci->i_dirty_caps),
1670                                 inode, ceph_ino(inode));
1671                         ci->i_dirty_caps = 0;
1672                         list_del_init(&ci->i_dirty_item);
1673                         dirty_dropped = true;
1674                 }
1675                 if (!list_empty(&ci->i_flushing_item)) {
1676                         pr_warn_ratelimited(
1677                                 " dropping dirty+flushing %s state for %p %lld\n",
1678                                 ceph_cap_string(ci->i_flushing_caps),
1679                                 inode, ceph_ino(inode));
1680                         ci->i_flushing_caps = 0;
1681                         list_del_init(&ci->i_flushing_item);
1682                         mdsc->num_cap_flushing--;
1683                         dirty_dropped = true;
1684                 }
1685                 spin_unlock(&mdsc->cap_dirty_lock);
1686
1687                 if (dirty_dropped) {
1688                         errseq_set(&ci->i_meta_err, -EIO);
1689
1690                         if (ci->i_wrbuffer_ref_head == 0 &&
1691                             ci->i_wr_ref == 0 &&
1692                             ci->i_dirty_caps == 0 &&
1693                             ci->i_flushing_caps == 0) {
1694                                 ceph_put_snap_context(ci->i_head_snapc);
1695                                 ci->i_head_snapc = NULL;
1696                         }
1697                 }
1698
1699                 if (atomic_read(&ci->i_filelock_ref) > 0) {
1700                         /* make further file lock syscall return -EIO */
1701                         ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1702                         pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1703                                             inode, ceph_ino(inode));
1704                 }
1705
1706                 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1707                         list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
1708                         ci->i_prealloc_cap_flush = NULL;
1709                 }
1710
1711                 if (!list_empty(&ci->i_cap_snaps))
1712                         capsnap_release = remove_capsnaps(mdsc, inode);
1713         }
1714         spin_unlock(&ci->i_ceph_lock);
1715         while (!list_empty(&to_remove)) {
1716                 struct ceph_cap_flush *cf;
1717                 cf = list_first_entry(&to_remove,
1718                                       struct ceph_cap_flush, i_list);
1719                 list_del_init(&cf->i_list);
1720                 if (!cf->is_capsnap)
1721                         ceph_free_cap_flush(cf);
1722         }
1723
1724         wake_up_all(&ci->i_cap_wq);
1725         if (invalidate)
1726                 ceph_queue_invalidate(inode);
1727         if (dirty_dropped)
1728                 iput(inode);
1729         while (capsnap_release--)
1730                 iput(inode);
1731         return 0;
1732 }
1733
1734 /*
1735  * caller must hold session s_mutex
1736  */
1737 static void remove_session_caps(struct ceph_mds_session *session)
1738 {
1739         struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1740         struct super_block *sb = fsc->sb;
1741         LIST_HEAD(dispose);
1742
1743         dout("remove_session_caps on %p\n", session);
1744         ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
1745
1746         wake_up_all(&fsc->mdsc->cap_flushing_wq);
1747
1748         spin_lock(&session->s_cap_lock);
1749         if (session->s_nr_caps > 0) {
1750                 struct inode *inode;
1751                 struct ceph_cap *cap, *prev = NULL;
1752                 struct ceph_vino vino;
1753                 /*
1754                  * iterate_session_caps() skips inodes that are being
1755                  * deleted, we need to wait until deletions are complete.
1756                  * __wait_on_freeing_inode() is designed for the job,
1757                  * but it is not exported, so use lookup inode function
1758                  * to access it.
1759                  */
1760                 while (!list_empty(&session->s_caps)) {
1761                         cap = list_entry(session->s_caps.next,
1762                                          struct ceph_cap, session_caps);
1763                         if (cap == prev)
1764                                 break;
1765                         prev = cap;
1766                         vino = cap->ci->i_vino;
1767                         spin_unlock(&session->s_cap_lock);
1768
1769                         inode = ceph_find_inode(sb, vino);
1770                         iput(inode);
1771
1772                         spin_lock(&session->s_cap_lock);
1773                 }
1774         }
1775
1776         // drop cap expires and unlock s_cap_lock
1777         detach_cap_releases(session, &dispose);
1778
1779         BUG_ON(session->s_nr_caps > 0);
1780         BUG_ON(!list_empty(&session->s_cap_flushing));
1781         spin_unlock(&session->s_cap_lock);
1782         dispose_cap_releases(session->s_mdsc, &dispose);
1783 }
1784
1785 enum {
1786         RECONNECT,
1787         RENEWCAPS,
1788         FORCE_RO,
1789 };
1790
1791 /*
1792  * wake up any threads waiting on this session's caps.  if the cap is
1793  * old (didn't get renewed on the client reconnect), remove it now.
1794  *
1795  * caller must hold s_mutex.
1796  */
1797 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1798                               void *arg)
1799 {
1800         struct ceph_inode_info *ci = ceph_inode(inode);
1801         unsigned long ev = (unsigned long)arg;
1802
1803         if (ev == RECONNECT) {
1804                 spin_lock(&ci->i_ceph_lock);
1805                 ci->i_wanted_max_size = 0;
1806                 ci->i_requested_max_size = 0;
1807                 spin_unlock(&ci->i_ceph_lock);
1808         } else if (ev == RENEWCAPS) {
1809                 if (cap->cap_gen < atomic_read(&cap->session->s_cap_gen)) {
1810                         /* mds did not re-issue stale cap */
1811                         spin_lock(&ci->i_ceph_lock);
1812                         cap->issued = cap->implemented = CEPH_CAP_PIN;
1813                         spin_unlock(&ci->i_ceph_lock);
1814                 }
1815         } else if (ev == FORCE_RO) {
1816         }
1817         wake_up_all(&ci->i_cap_wq);
1818         return 0;
1819 }
1820
1821 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
1822 {
1823         dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1824         ceph_iterate_session_caps(session, wake_up_session_cb,
1825                                   (void *)(unsigned long)ev);
1826 }
1827
1828 /*
1829  * Send periodic message to MDS renewing all currently held caps.  The
1830  * ack will reset the expiration for all caps from this session.
1831  *
1832  * caller holds s_mutex
1833  */
1834 static int send_renew_caps(struct ceph_mds_client *mdsc,
1835                            struct ceph_mds_session *session)
1836 {
1837         struct ceph_msg *msg;
1838         int state;
1839
1840         if (time_after_eq(jiffies, session->s_cap_ttl) &&
1841             time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1842                 pr_info("mds%d caps stale\n", session->s_mds);
1843         session->s_renew_requested = jiffies;
1844
1845         /* do not try to renew caps until a recovering mds has reconnected
1846          * with its clients. */
1847         state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1848         if (state < CEPH_MDS_STATE_RECONNECT) {
1849                 dout("send_renew_caps ignoring mds%d (%s)\n",
1850                      session->s_mds, ceph_mds_state_name(state));
1851                 return 0;
1852         }
1853
1854         dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1855                 ceph_mds_state_name(state));
1856         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1857                                       ++session->s_renew_seq);
1858         if (!msg)
1859                 return -ENOMEM;
1860         ceph_con_send(&session->s_con, msg);
1861         return 0;
1862 }
1863
1864 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1865                              struct ceph_mds_session *session, u64 seq)
1866 {
1867         struct ceph_msg *msg;
1868
1869         dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1870              session->s_mds, ceph_session_state_name(session->s_state), seq);
1871         msg = ceph_create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1872         if (!msg)
1873                 return -ENOMEM;
1874         ceph_con_send(&session->s_con, msg);
1875         return 0;
1876 }
1877
1878
1879 /*
1880  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1881  *
1882  * Called under session->s_mutex
1883  */
1884 static void renewed_caps(struct ceph_mds_client *mdsc,
1885                          struct ceph_mds_session *session, int is_renew)
1886 {
1887         int was_stale;
1888         int wake = 0;
1889
1890         spin_lock(&session->s_cap_lock);
1891         was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1892
1893         session->s_cap_ttl = session->s_renew_requested +
1894                 mdsc->mdsmap->m_session_timeout*HZ;
1895
1896         if (was_stale) {
1897                 if (time_before(jiffies, session->s_cap_ttl)) {
1898                         pr_info("mds%d caps renewed\n", session->s_mds);
1899                         wake = 1;
1900                 } else {
1901                         pr_info("mds%d caps still stale\n", session->s_mds);
1902                 }
1903         }
1904         dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1905              session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1906              time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1907         spin_unlock(&session->s_cap_lock);
1908
1909         if (wake)
1910                 wake_up_session_caps(session, RENEWCAPS);
1911 }
1912
1913 /*
1914  * send a session close request
1915  */
1916 static int request_close_session(struct ceph_mds_session *session)
1917 {
1918         struct ceph_msg *msg;
1919
1920         dout("request_close_session mds%d state %s seq %lld\n",
1921              session->s_mds, ceph_session_state_name(session->s_state),
1922              session->s_seq);
1923         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_CLOSE,
1924                                       session->s_seq);
1925         if (!msg)
1926                 return -ENOMEM;
1927         ceph_con_send(&session->s_con, msg);
1928         return 1;
1929 }
1930
1931 /*
1932  * Called with s_mutex held.
1933  */
1934 static int __close_session(struct ceph_mds_client *mdsc,
1935                          struct ceph_mds_session *session)
1936 {
1937         if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1938                 return 0;
1939         session->s_state = CEPH_MDS_SESSION_CLOSING;
1940         return request_close_session(session);
1941 }
1942
1943 static bool drop_negative_children(struct dentry *dentry)
1944 {
1945         struct dentry *child;
1946         bool all_negative = true;
1947
1948         if (!d_is_dir(dentry))
1949                 goto out;
1950
1951         spin_lock(&dentry->d_lock);
1952         list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1953                 if (d_really_is_positive(child)) {
1954                         all_negative = false;
1955                         break;
1956                 }
1957         }
1958         spin_unlock(&dentry->d_lock);
1959
1960         if (all_negative)
1961                 shrink_dcache_parent(dentry);
1962 out:
1963         return all_negative;
1964 }
1965
1966 /*
1967  * Trim old(er) caps.
1968  *
1969  * Because we can't cache an inode without one or more caps, we do
1970  * this indirectly: if a cap is unused, we prune its aliases, at which
1971  * point the inode will hopefully get dropped to.
1972  *
1973  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1974  * memory pressure from the MDS, though, so it needn't be perfect.
1975  */
1976 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1977 {
1978         int *remaining = arg;
1979         struct ceph_inode_info *ci = ceph_inode(inode);
1980         int used, wanted, oissued, mine;
1981
1982         if (*remaining <= 0)
1983                 return -1;
1984
1985         spin_lock(&ci->i_ceph_lock);
1986         mine = cap->issued | cap->implemented;
1987         used = __ceph_caps_used(ci);
1988         wanted = __ceph_caps_file_wanted(ci);
1989         oissued = __ceph_caps_issued_other(ci, cap);
1990
1991         dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1992              inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1993              ceph_cap_string(used), ceph_cap_string(wanted));
1994         if (cap == ci->i_auth_cap) {
1995                 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1996                     !list_empty(&ci->i_cap_snaps))
1997                         goto out;
1998                 if ((used | wanted) & CEPH_CAP_ANY_WR)
1999                         goto out;
2000                 /* Note: it's possible that i_filelock_ref becomes non-zero
2001                  * after dropping auth caps. It doesn't hurt because reply
2002                  * of lock mds request will re-add auth caps. */
2003                 if (atomic_read(&ci->i_filelock_ref) > 0)
2004                         goto out;
2005         }
2006         /* The inode has cached pages, but it's no longer used.
2007          * we can safely drop it */
2008         if (S_ISREG(inode->i_mode) &&
2009             wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
2010             !(oissued & CEPH_CAP_FILE_CACHE)) {
2011           used = 0;
2012           oissued = 0;
2013         }
2014         if ((used | wanted) & ~oissued & mine)
2015                 goto out;   /* we need these caps */
2016
2017         if (oissued) {
2018                 /* we aren't the only cap.. just remove us */
2019                 __ceph_remove_cap(cap, true);
2020                 (*remaining)--;
2021         } else {
2022                 struct dentry *dentry;
2023                 /* try dropping referring dentries */
2024                 spin_unlock(&ci->i_ceph_lock);
2025                 dentry = d_find_any_alias(inode);
2026                 if (dentry && drop_negative_children(dentry)) {
2027                         int count;
2028                         dput(dentry);
2029                         d_prune_aliases(inode);
2030                         count = atomic_read(&inode->i_count);
2031                         if (count == 1)
2032                                 (*remaining)--;
2033                         dout("trim_caps_cb %p cap %p pruned, count now %d\n",
2034                              inode, cap, count);
2035                 } else {
2036                         dput(dentry);
2037                 }
2038                 return 0;
2039         }
2040
2041 out:
2042         spin_unlock(&ci->i_ceph_lock);
2043         return 0;
2044 }
2045
2046 /*
2047  * Trim session cap count down to some max number.
2048  */
2049 int ceph_trim_caps(struct ceph_mds_client *mdsc,
2050                    struct ceph_mds_session *session,
2051                    int max_caps)
2052 {
2053         int trim_caps = session->s_nr_caps - max_caps;
2054
2055         dout("trim_caps mds%d start: %d / %d, trim %d\n",
2056              session->s_mds, session->s_nr_caps, max_caps, trim_caps);
2057         if (trim_caps > 0) {
2058                 int remaining = trim_caps;
2059
2060                 ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
2061                 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
2062                      session->s_mds, session->s_nr_caps, max_caps,
2063                         trim_caps - remaining);
2064         }
2065
2066         ceph_flush_cap_releases(mdsc, session);
2067         return 0;
2068 }
2069
2070 static int check_caps_flush(struct ceph_mds_client *mdsc,
2071                             u64 want_flush_tid)
2072 {
2073         int ret = 1;
2074
2075         spin_lock(&mdsc->cap_dirty_lock);
2076         if (!list_empty(&mdsc->cap_flush_list)) {
2077                 struct ceph_cap_flush *cf =
2078                         list_first_entry(&mdsc->cap_flush_list,
2079                                          struct ceph_cap_flush, g_list);
2080                 if (cf->tid <= want_flush_tid) {
2081                         dout("check_caps_flush still flushing tid "
2082                              "%llu <= %llu\n", cf->tid, want_flush_tid);
2083                         ret = 0;
2084                 }
2085         }
2086         spin_unlock(&mdsc->cap_dirty_lock);
2087         return ret;
2088 }
2089
2090 /*
2091  * flush all dirty inode data to disk.
2092  *
2093  * returns true if we've flushed through want_flush_tid
2094  */
2095 static void wait_caps_flush(struct ceph_mds_client *mdsc,
2096                             u64 want_flush_tid)
2097 {
2098         dout("check_caps_flush want %llu\n", want_flush_tid);
2099
2100         wait_event(mdsc->cap_flushing_wq,
2101                    check_caps_flush(mdsc, want_flush_tid));
2102
2103         dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2104 }
2105
2106 /*
2107  * called under s_mutex
2108  */
2109 static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
2110                                    struct ceph_mds_session *session)
2111 {
2112         struct ceph_msg *msg = NULL;
2113         struct ceph_mds_cap_release *head;
2114         struct ceph_mds_cap_item *item;
2115         struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
2116         struct ceph_cap *cap;
2117         LIST_HEAD(tmp_list);
2118         int num_cap_releases;
2119         __le32  barrier, *cap_barrier;
2120
2121         down_read(&osdc->lock);
2122         barrier = cpu_to_le32(osdc->epoch_barrier);
2123         up_read(&osdc->lock);
2124
2125         spin_lock(&session->s_cap_lock);
2126 again:
2127         list_splice_init(&session->s_cap_releases, &tmp_list);
2128         num_cap_releases = session->s_num_cap_releases;
2129         session->s_num_cap_releases = 0;
2130         spin_unlock(&session->s_cap_lock);
2131
2132         while (!list_empty(&tmp_list)) {
2133                 if (!msg) {
2134                         msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
2135                                         PAGE_SIZE, GFP_NOFS, false);
2136                         if (!msg)
2137                                 goto out_err;
2138                         head = msg->front.iov_base;
2139                         head->num = cpu_to_le32(0);
2140                         msg->front.iov_len = sizeof(*head);
2141
2142                         msg->hdr.version = cpu_to_le16(2);
2143                         msg->hdr.compat_version = cpu_to_le16(1);
2144                 }
2145
2146                 cap = list_first_entry(&tmp_list, struct ceph_cap,
2147                                         session_caps);
2148                 list_del(&cap->session_caps);
2149                 num_cap_releases--;
2150
2151                 head = msg->front.iov_base;
2152                 put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
2153                                    &head->num);
2154                 item = msg->front.iov_base + msg->front.iov_len;
2155                 item->ino = cpu_to_le64(cap->cap_ino);
2156                 item->cap_id = cpu_to_le64(cap->cap_id);
2157                 item->migrate_seq = cpu_to_le32(cap->mseq);
2158                 item->seq = cpu_to_le32(cap->issue_seq);
2159                 msg->front.iov_len += sizeof(*item);
2160
2161                 ceph_put_cap(mdsc, cap);
2162
2163                 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
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                         msg = NULL;
2173                 }
2174         }
2175
2176         BUG_ON(num_cap_releases != 0);
2177
2178         spin_lock(&session->s_cap_lock);
2179         if (!list_empty(&session->s_cap_releases))
2180                 goto again;
2181         spin_unlock(&session->s_cap_lock);
2182
2183         if (msg) {
2184                 // Append cap_barrier field
2185                 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2186                 *cap_barrier = barrier;
2187                 msg->front.iov_len += sizeof(*cap_barrier);
2188
2189                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2190                 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2191                 ceph_con_send(&session->s_con, msg);
2192         }
2193         return;
2194 out_err:
2195         pr_err("send_cap_releases mds%d, failed to allocate message\n",
2196                 session->s_mds);
2197         spin_lock(&session->s_cap_lock);
2198         list_splice(&tmp_list, &session->s_cap_releases);
2199         session->s_num_cap_releases += num_cap_releases;
2200         spin_unlock(&session->s_cap_lock);
2201 }
2202
2203 static void ceph_cap_release_work(struct work_struct *work)
2204 {
2205         struct ceph_mds_session *session =
2206                 container_of(work, struct ceph_mds_session, s_cap_release_work);
2207
2208         mutex_lock(&session->s_mutex);
2209         if (session->s_state == CEPH_MDS_SESSION_OPEN ||
2210             session->s_state == CEPH_MDS_SESSION_HUNG)
2211                 ceph_send_cap_releases(session->s_mdsc, session);
2212         mutex_unlock(&session->s_mutex);
2213         ceph_put_mds_session(session);
2214 }
2215
2216 void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
2217                              struct ceph_mds_session *session)
2218 {
2219         if (mdsc->stopping)
2220                 return;
2221
2222         ceph_get_mds_session(session);
2223         if (queue_work(mdsc->fsc->cap_wq,
2224                        &session->s_cap_release_work)) {
2225                 dout("cap release work queued\n");
2226         } else {
2227                 ceph_put_mds_session(session);
2228                 dout("failed to queue cap release work\n");
2229         }
2230 }
2231
2232 /*
2233  * caller holds session->s_cap_lock
2234  */
2235 void __ceph_queue_cap_release(struct ceph_mds_session *session,
2236                               struct ceph_cap *cap)
2237 {
2238         list_add_tail(&cap->session_caps, &session->s_cap_releases);
2239         session->s_num_cap_releases++;
2240
2241         if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
2242                 ceph_flush_cap_releases(session->s_mdsc, session);
2243 }
2244
2245 static void ceph_cap_reclaim_work(struct work_struct *work)
2246 {
2247         struct ceph_mds_client *mdsc =
2248                 container_of(work, struct ceph_mds_client, cap_reclaim_work);
2249         int ret = ceph_trim_dentries(mdsc);
2250         if (ret == -EAGAIN)
2251                 ceph_queue_cap_reclaim_work(mdsc);
2252 }
2253
2254 void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2255 {
2256         if (mdsc->stopping)
2257                 return;
2258
2259         if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
2260                 dout("caps reclaim work queued\n");
2261         } else {
2262                 dout("failed to queue caps release work\n");
2263         }
2264 }
2265
2266 void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2267 {
2268         int val;
2269         if (!nr)
2270                 return;
2271         val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
2272         if ((val % CEPH_CAPS_PER_RELEASE) < nr) {
2273                 atomic_set(&mdsc->cap_reclaim_pending, 0);
2274                 ceph_queue_cap_reclaim_work(mdsc);
2275         }
2276 }
2277
2278 /*
2279  * requests
2280  */
2281
2282 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2283                                     struct inode *dir)
2284 {
2285         struct ceph_inode_info *ci = ceph_inode(dir);
2286         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2287         struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2288         size_t size = sizeof(struct ceph_mds_reply_dir_entry);
2289         unsigned int num_entries;
2290         int order;
2291
2292         spin_lock(&ci->i_ceph_lock);
2293         num_entries = ci->i_files + ci->i_subdirs;
2294         spin_unlock(&ci->i_ceph_lock);
2295         num_entries = max(num_entries, 1U);
2296         num_entries = min(num_entries, opt->max_readdir);
2297
2298         order = get_order(size * num_entries);
2299         while (order >= 0) {
2300                 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2301                                                              __GFP_NOWARN,
2302                                                              order);
2303                 if (rinfo->dir_entries)
2304                         break;
2305                 order--;
2306         }
2307         if (!rinfo->dir_entries)
2308                 return -ENOMEM;
2309
2310         num_entries = (PAGE_SIZE << order) / size;
2311         num_entries = min(num_entries, opt->max_readdir);
2312
2313         rinfo->dir_buf_size = PAGE_SIZE << order;
2314         req->r_num_caps = num_entries + 1;
2315         req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2316         req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2317         return 0;
2318 }
2319
2320 /*
2321  * Create an mds request.
2322  */
2323 struct ceph_mds_request *
2324 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2325 {
2326         struct ceph_mds_request *req;
2327
2328         req = kmem_cache_zalloc(ceph_mds_request_cachep, GFP_NOFS);
2329         if (!req)
2330                 return ERR_PTR(-ENOMEM);
2331
2332         mutex_init(&req->r_fill_mutex);
2333         req->r_mdsc = mdsc;
2334         req->r_started = jiffies;
2335         req->r_start_latency = ktime_get();
2336         req->r_resend_mds = -1;
2337         INIT_LIST_HEAD(&req->r_unsafe_dir_item);
2338         INIT_LIST_HEAD(&req->r_unsafe_target_item);
2339         req->r_fmode = -1;
2340         kref_init(&req->r_kref);
2341         RB_CLEAR_NODE(&req->r_node);
2342         INIT_LIST_HEAD(&req->r_wait);
2343         init_completion(&req->r_completion);
2344         init_completion(&req->r_safe_completion);
2345         INIT_LIST_HEAD(&req->r_unsafe_item);
2346
2347         ktime_get_coarse_real_ts64(&req->r_stamp);
2348
2349         req->r_op = op;
2350         req->r_direct_mode = mode;
2351         return req;
2352 }
2353
2354 /*
2355  * return oldest (lowest) request, tid in request tree, 0 if none.
2356  *
2357  * called under mdsc->mutex.
2358  */
2359 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2360 {
2361         if (RB_EMPTY_ROOT(&mdsc->request_tree))
2362                 return NULL;
2363         return rb_entry(rb_first(&mdsc->request_tree),
2364                         struct ceph_mds_request, r_node);
2365 }
2366
2367 static inline  u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2368 {
2369         return mdsc->oldest_tid;
2370 }
2371
2372 /*
2373  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
2374  * on build_path_from_dentry in fs/cifs/dir.c.
2375  *
2376  * If @stop_on_nosnap, generate path relative to the first non-snapped
2377  * inode.
2378  *
2379  * Encode hidden .snap dirs as a double /, i.e.
2380  *   foo/.snap/bar -> foo//bar
2381  */
2382 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *pbase,
2383                            int stop_on_nosnap)
2384 {
2385         struct dentry *temp;
2386         char *path;
2387         int pos;
2388         unsigned seq;
2389         u64 base;
2390
2391         if (!dentry)
2392                 return ERR_PTR(-EINVAL);
2393
2394         path = __getname();
2395         if (!path)
2396                 return ERR_PTR(-ENOMEM);
2397 retry:
2398         pos = PATH_MAX - 1;
2399         path[pos] = '\0';
2400
2401         seq = read_seqbegin(&rename_lock);
2402         rcu_read_lock();
2403         temp = dentry;
2404         for (;;) {
2405                 struct inode *inode;
2406
2407                 spin_lock(&temp->d_lock);
2408                 inode = d_inode(temp);
2409                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
2410                         dout("build_path path+%d: %p SNAPDIR\n",
2411                              pos, temp);
2412                 } else if (stop_on_nosnap && inode && dentry != temp &&
2413                            ceph_snap(inode) == CEPH_NOSNAP) {
2414                         spin_unlock(&temp->d_lock);
2415                         pos++; /* get rid of any prepended '/' */
2416                         break;
2417                 } else {
2418                         pos -= temp->d_name.len;
2419                         if (pos < 0) {
2420                                 spin_unlock(&temp->d_lock);
2421                                 break;
2422                         }
2423                         memcpy(path + pos, temp->d_name.name, temp->d_name.len);
2424                 }
2425                 spin_unlock(&temp->d_lock);
2426                 temp = READ_ONCE(temp->d_parent);
2427
2428                 /* Are we at the root? */
2429                 if (IS_ROOT(temp))
2430                         break;
2431
2432                 /* Are we out of buffer? */
2433                 if (--pos < 0)
2434                         break;
2435
2436                 path[pos] = '/';
2437         }
2438         base = ceph_ino(d_inode(temp));
2439         rcu_read_unlock();
2440
2441         if (read_seqretry(&rename_lock, seq))
2442                 goto retry;
2443
2444         if (pos < 0) {
2445                 /*
2446                  * A rename didn't occur, but somehow we didn't end up where
2447                  * we thought we would. Throw a warning and try again.
2448                  */
2449                 pr_warn("build_path did not end path lookup where "
2450                         "expected, pos is %d\n", pos);
2451                 goto retry;
2452         }
2453
2454         *pbase = base;
2455         *plen = PATH_MAX - 1 - pos;
2456         dout("build_path on %p %d built %llx '%.*s'\n",
2457              dentry, d_count(dentry), base, *plen, path + pos);
2458         return path + pos;
2459 }
2460
2461 static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2462                              const char **ppath, int *ppathlen, u64 *pino,
2463                              bool *pfreepath, bool parent_locked)
2464 {
2465         char *path;
2466
2467         rcu_read_lock();
2468         if (!dir)
2469                 dir = d_inode_rcu(dentry->d_parent);
2470         if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP) {
2471                 *pino = ceph_ino(dir);
2472                 rcu_read_unlock();
2473                 *ppath = dentry->d_name.name;
2474                 *ppathlen = dentry->d_name.len;
2475                 return 0;
2476         }
2477         rcu_read_unlock();
2478         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2479         if (IS_ERR(path))
2480                 return PTR_ERR(path);
2481         *ppath = path;
2482         *pfreepath = true;
2483         return 0;
2484 }
2485
2486 static int build_inode_path(struct inode *inode,
2487                             const char **ppath, int *ppathlen, u64 *pino,
2488                             bool *pfreepath)
2489 {
2490         struct dentry *dentry;
2491         char *path;
2492
2493         if (ceph_snap(inode) == CEPH_NOSNAP) {
2494                 *pino = ceph_ino(inode);
2495                 *ppathlen = 0;
2496                 return 0;
2497         }
2498         dentry = d_find_alias(inode);
2499         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2500         dput(dentry);
2501         if (IS_ERR(path))
2502                 return PTR_ERR(path);
2503         *ppath = path;
2504         *pfreepath = true;
2505         return 0;
2506 }
2507
2508 /*
2509  * request arguments may be specified via an inode *, a dentry *, or
2510  * an explicit ino+path.
2511  */
2512 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
2513                                   struct inode *rdiri, const char *rpath,
2514                                   u64 rino, const char **ppath, int *pathlen,
2515                                   u64 *ino, bool *freepath, bool parent_locked)
2516 {
2517         int r = 0;
2518
2519         if (rinode) {
2520                 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
2521                 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2522                      ceph_snap(rinode));
2523         } else if (rdentry) {
2524                 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
2525                                         freepath, parent_locked);
2526                 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
2527                      *ppath);
2528         } else if (rpath || rino) {
2529                 *ino = rino;
2530                 *ppath = rpath;
2531                 *pathlen = rpath ? strlen(rpath) : 0;
2532                 dout(" path %.*s\n", *pathlen, rpath);
2533         }
2534
2535         return r;
2536 }
2537
2538 static void encode_timestamp_and_gids(void **p,
2539                                       const struct ceph_mds_request *req)
2540 {
2541         struct ceph_timespec ts;
2542         int i;
2543
2544         ceph_encode_timespec64(&ts, &req->r_stamp);
2545         ceph_encode_copy(p, &ts, sizeof(ts));
2546
2547         /* gid_list */
2548         ceph_encode_32(p, req->r_cred->group_info->ngroups);
2549         for (i = 0; i < req->r_cred->group_info->ngroups; i++)
2550                 ceph_encode_64(p, from_kgid(&init_user_ns,
2551                                             req->r_cred->group_info->gid[i]));
2552 }
2553
2554 /*
2555  * called under mdsc->mutex
2556  */
2557 static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
2558                                                struct ceph_mds_request *req,
2559                                                bool drop_cap_releases)
2560 {
2561         int mds = session->s_mds;
2562         struct ceph_mds_client *mdsc = session->s_mdsc;
2563         struct ceph_msg *msg;
2564         struct ceph_mds_request_head_old *head;
2565         const char *path1 = NULL;
2566         const char *path2 = NULL;
2567         u64 ino1 = 0, ino2 = 0;
2568         int pathlen1 = 0, pathlen2 = 0;
2569         bool freepath1 = false, freepath2 = false;
2570         int len;
2571         u16 releases;
2572         void *p, *end;
2573         int ret;
2574         bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
2575
2576         ret = set_request_path_attr(req->r_inode, req->r_dentry,
2577                               req->r_parent, req->r_path1, req->r_ino1.ino,
2578                               &path1, &pathlen1, &ino1, &freepath1,
2579                               test_bit(CEPH_MDS_R_PARENT_LOCKED,
2580                                         &req->r_req_flags));
2581         if (ret < 0) {
2582                 msg = ERR_PTR(ret);
2583                 goto out;
2584         }
2585
2586         /* If r_old_dentry is set, then assume that its parent is locked */
2587         ret = set_request_path_attr(NULL, req->r_old_dentry,
2588                               req->r_old_dentry_dir,
2589                               req->r_path2, req->r_ino2.ino,
2590                               &path2, &pathlen2, &ino2, &freepath2, true);
2591         if (ret < 0) {
2592                 msg = ERR_PTR(ret);
2593                 goto out_free1;
2594         }
2595
2596         len = legacy ? sizeof(*head) : sizeof(struct ceph_mds_request_head);
2597         len += pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
2598                 sizeof(struct ceph_timespec);
2599         len += sizeof(u32) + (sizeof(u64) * req->r_cred->group_info->ngroups);
2600
2601         /* calculate (max) length for cap releases */
2602         len += sizeof(struct ceph_mds_request_release) *
2603                 (!!req->r_inode_drop + !!req->r_dentry_drop +
2604                  !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2605
2606         if (req->r_dentry_drop)
2607                 len += pathlen1;
2608         if (req->r_old_dentry_drop)
2609                 len += pathlen2;
2610
2611         msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
2612         if (!msg) {
2613                 msg = ERR_PTR(-ENOMEM);
2614                 goto out_free2;
2615         }
2616
2617         msg->hdr.tid = cpu_to_le64(req->r_tid);
2618
2619         /*
2620          * The old ceph_mds_request_head didn't contain a version field, and
2621          * one was added when we moved the message version from 3->4.
2622          */
2623         if (legacy) {
2624                 msg->hdr.version = cpu_to_le16(3);
2625                 head = msg->front.iov_base;
2626                 p = msg->front.iov_base + sizeof(*head);
2627         } else {
2628                 struct ceph_mds_request_head *new_head = msg->front.iov_base;
2629
2630                 msg->hdr.version = cpu_to_le16(4);
2631                 new_head->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
2632                 head = (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2633                 p = msg->front.iov_base + sizeof(*new_head);
2634         }
2635
2636         end = msg->front.iov_base + msg->front.iov_len;
2637
2638         head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2639         head->op = cpu_to_le32(req->r_op);
2640         head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns,
2641                                                  req->r_cred->fsuid));
2642         head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns,
2643                                                  req->r_cred->fsgid));
2644         head->ino = cpu_to_le64(req->r_deleg_ino);
2645         head->args = req->r_args;
2646
2647         ceph_encode_filepath(&p, end, ino1, path1);
2648         ceph_encode_filepath(&p, end, ino2, path2);
2649
2650         /* make note of release offset, in case we need to replay */
2651         req->r_request_release_offset = p - msg->front.iov_base;
2652
2653         /* cap releases */
2654         releases = 0;
2655         if (req->r_inode_drop)
2656                 releases += ceph_encode_inode_release(&p,
2657                       req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2658                       mds, req->r_inode_drop, req->r_inode_unless,
2659                       req->r_op == CEPH_MDS_OP_READDIR);
2660         if (req->r_dentry_drop)
2661                 releases += ceph_encode_dentry_release(&p, req->r_dentry,
2662                                 req->r_parent, mds, req->r_dentry_drop,
2663                                 req->r_dentry_unless);
2664         if (req->r_old_dentry_drop)
2665                 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2666                                 req->r_old_dentry_dir, mds,
2667                                 req->r_old_dentry_drop,
2668                                 req->r_old_dentry_unless);
2669         if (req->r_old_inode_drop)
2670                 releases += ceph_encode_inode_release(&p,
2671                       d_inode(req->r_old_dentry),
2672                       mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
2673
2674         if (drop_cap_releases) {
2675                 releases = 0;
2676                 p = msg->front.iov_base + req->r_request_release_offset;
2677         }
2678
2679         head->num_releases = cpu_to_le16(releases);
2680
2681         encode_timestamp_and_gids(&p, req);
2682
2683         if (WARN_ON_ONCE(p > end)) {
2684                 ceph_msg_put(msg);
2685                 msg = ERR_PTR(-ERANGE);
2686                 goto out_free2;
2687         }
2688
2689         msg->front.iov_len = p - msg->front.iov_base;
2690         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2691
2692         if (req->r_pagelist) {
2693                 struct ceph_pagelist *pagelist = req->r_pagelist;
2694                 ceph_msg_data_add_pagelist(msg, pagelist);
2695                 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2696         } else {
2697                 msg->hdr.data_len = 0;
2698         }
2699
2700         msg->hdr.data_off = cpu_to_le16(0);
2701
2702 out_free2:
2703         if (freepath2)
2704                 ceph_mdsc_free_path((char *)path2, pathlen2);
2705 out_free1:
2706         if (freepath1)
2707                 ceph_mdsc_free_path((char *)path1, pathlen1);
2708 out:
2709         return msg;
2710 }
2711
2712 /*
2713  * called under mdsc->mutex if error, under no mutex if
2714  * success.
2715  */
2716 static void complete_request(struct ceph_mds_client *mdsc,
2717                              struct ceph_mds_request *req)
2718 {
2719         req->r_end_latency = ktime_get();
2720
2721         if (req->r_callback)
2722                 req->r_callback(mdsc, req);
2723         complete_all(&req->r_completion);
2724 }
2725
2726 static struct ceph_mds_request_head_old *
2727 find_old_request_head(void *p, u64 features)
2728 {
2729         bool legacy = !(features & CEPH_FEATURE_FS_BTIME);
2730         struct ceph_mds_request_head *new_head;
2731
2732         if (legacy)
2733                 return (struct ceph_mds_request_head_old *)p;
2734         new_head = (struct ceph_mds_request_head *)p;
2735         return (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2736 }
2737
2738 /*
2739  * called under mdsc->mutex
2740  */
2741 static int __prepare_send_request(struct ceph_mds_session *session,
2742                                   struct ceph_mds_request *req,
2743                                   bool drop_cap_releases)
2744 {
2745         int mds = session->s_mds;
2746         struct ceph_mds_client *mdsc = session->s_mdsc;
2747         struct ceph_mds_request_head_old *rhead;
2748         struct ceph_msg *msg;
2749         int flags = 0;
2750
2751         req->r_attempts++;
2752         if (req->r_inode) {
2753                 struct ceph_cap *cap =
2754                         ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2755
2756                 if (cap)
2757                         req->r_sent_on_mseq = cap->mseq;
2758                 else
2759                         req->r_sent_on_mseq = -1;
2760         }
2761         dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2762              req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2763
2764         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2765                 void *p;
2766
2767                 /*
2768                  * Replay.  Do not regenerate message (and rebuild
2769                  * paths, etc.); just use the original message.
2770                  * Rebuilding paths will break for renames because
2771                  * d_move mangles the src name.
2772                  */
2773                 msg = req->r_request;
2774                 rhead = find_old_request_head(msg->front.iov_base,
2775                                               session->s_con.peer_features);
2776
2777                 flags = le32_to_cpu(rhead->flags);
2778                 flags |= CEPH_MDS_FLAG_REPLAY;
2779                 rhead->flags = cpu_to_le32(flags);
2780
2781                 if (req->r_target_inode)
2782                         rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2783
2784                 rhead->num_retry = req->r_attempts - 1;
2785
2786                 /* remove cap/dentry releases from message */
2787                 rhead->num_releases = 0;
2788
2789                 p = msg->front.iov_base + req->r_request_release_offset;
2790                 encode_timestamp_and_gids(&p, req);
2791
2792                 msg->front.iov_len = p - msg->front.iov_base;
2793                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2794                 return 0;
2795         }
2796
2797         if (req->r_request) {
2798                 ceph_msg_put(req->r_request);
2799                 req->r_request = NULL;
2800         }
2801         msg = create_request_message(session, req, drop_cap_releases);
2802         if (IS_ERR(msg)) {
2803                 req->r_err = PTR_ERR(msg);
2804                 return PTR_ERR(msg);
2805         }
2806         req->r_request = msg;
2807
2808         rhead = find_old_request_head(msg->front.iov_base,
2809                                       session->s_con.peer_features);
2810         rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2811         if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2812                 flags |= CEPH_MDS_FLAG_REPLAY;
2813         if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags))
2814                 flags |= CEPH_MDS_FLAG_ASYNC;
2815         if (req->r_parent)
2816                 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2817         rhead->flags = cpu_to_le32(flags);
2818         rhead->num_fwd = req->r_num_fwd;
2819         rhead->num_retry = req->r_attempts - 1;
2820
2821         dout(" r_parent = %p\n", req->r_parent);
2822         return 0;
2823 }
2824
2825 /*
2826  * called under mdsc->mutex
2827  */
2828 static int __send_request(struct ceph_mds_session *session,
2829                           struct ceph_mds_request *req,
2830                           bool drop_cap_releases)
2831 {
2832         int err;
2833
2834         err = __prepare_send_request(session, req, drop_cap_releases);
2835         if (!err) {
2836                 ceph_msg_get(req->r_request);
2837                 ceph_con_send(&session->s_con, req->r_request);
2838         }
2839
2840         return err;
2841 }
2842
2843 /*
2844  * send request, or put it on the appropriate wait list.
2845  */
2846 static void __do_request(struct ceph_mds_client *mdsc,
2847                         struct ceph_mds_request *req)
2848 {
2849         struct ceph_mds_session *session = NULL;
2850         int mds = -1;
2851         int err = 0;
2852         bool random;
2853
2854         if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2855                 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
2856                         __unregister_request(mdsc, req);
2857                 return;
2858         }
2859
2860         if (req->r_timeout &&
2861             time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2862                 dout("do_request timed out\n");
2863                 err = -ETIMEDOUT;
2864                 goto finish;
2865         }
2866         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2867                 dout("do_request forced umount\n");
2868                 err = -EIO;
2869                 goto finish;
2870         }
2871         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2872                 if (mdsc->mdsmap_err) {
2873                         err = mdsc->mdsmap_err;
2874                         dout("do_request mdsmap err %d\n", err);
2875                         goto finish;
2876                 }
2877                 if (mdsc->mdsmap->m_epoch == 0) {
2878                         dout("do_request no mdsmap, waiting for map\n");
2879                         list_add(&req->r_wait, &mdsc->waiting_for_map);
2880                         return;
2881                 }
2882                 if (!(mdsc->fsc->mount_options->flags &
2883                       CEPH_MOUNT_OPT_MOUNTWAIT) &&
2884                     !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2885                         err = -EHOSTUNREACH;
2886                         goto finish;
2887                 }
2888         }
2889
2890         put_request_session(req);
2891
2892         mds = __choose_mds(mdsc, req, &random);
2893         if (mds < 0 ||
2894             ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2895                 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2896                         err = -EJUKEBOX;
2897                         goto finish;
2898                 }
2899                 dout("do_request no mds or not active, waiting for map\n");
2900                 list_add(&req->r_wait, &mdsc->waiting_for_map);
2901                 return;
2902         }
2903
2904         /* get, open session */
2905         session = __ceph_lookup_mds_session(mdsc, mds);
2906         if (!session) {
2907                 session = register_session(mdsc, mds);
2908                 if (IS_ERR(session)) {
2909                         err = PTR_ERR(session);
2910                         goto finish;
2911                 }
2912         }
2913         req->r_session = ceph_get_mds_session(session);
2914
2915         dout("do_request mds%d session %p state %s\n", mds, session,
2916              ceph_session_state_name(session->s_state));
2917         if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2918             session->s_state != CEPH_MDS_SESSION_HUNG) {
2919                 /*
2920                  * We cannot queue async requests since the caps and delegated
2921                  * inodes are bound to the session. Just return -EJUKEBOX and
2922                  * let the caller retry a sync request in that case.
2923                  */
2924                 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2925                         err = -EJUKEBOX;
2926                         goto out_session;
2927                 }
2928
2929                 /*
2930                  * If the session has been REJECTED, then return a hard error,
2931                  * unless it's a CLEANRECOVER mount, in which case we'll queue
2932                  * it to the mdsc queue.
2933                  */
2934                 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2935                         if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
2936                                 list_add(&req->r_wait, &mdsc->waiting_for_map);
2937                         else
2938                                 err = -EACCES;
2939                         goto out_session;
2940                 }
2941
2942                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2943                     session->s_state == CEPH_MDS_SESSION_CLOSING) {
2944                         err = __open_session(mdsc, session);
2945                         if (err)
2946                                 goto out_session;
2947                         /* retry the same mds later */
2948                         if (random)
2949                                 req->r_resend_mds = mds;
2950                 }
2951                 list_add(&req->r_wait, &session->s_waiting);
2952                 goto out_session;
2953         }
2954
2955         /* send request */
2956         req->r_resend_mds = -1;   /* forget any previous mds hint */
2957
2958         if (req->r_request_started == 0)   /* note request start time */
2959                 req->r_request_started = jiffies;
2960
2961         err = __send_request(session, req, false);
2962
2963 out_session:
2964         ceph_put_mds_session(session);
2965 finish:
2966         if (err) {
2967                 dout("__do_request early error %d\n", err);
2968                 req->r_err = err;
2969                 complete_request(mdsc, req);
2970                 __unregister_request(mdsc, req);
2971         }
2972         return;
2973 }
2974
2975 /*
2976  * called under mdsc->mutex
2977  */
2978 static void __wake_requests(struct ceph_mds_client *mdsc,
2979                             struct list_head *head)
2980 {
2981         struct ceph_mds_request *req;
2982         LIST_HEAD(tmp_list);
2983
2984         list_splice_init(head, &tmp_list);
2985
2986         while (!list_empty(&tmp_list)) {
2987                 req = list_entry(tmp_list.next,
2988                                  struct ceph_mds_request, r_wait);
2989                 list_del_init(&req->r_wait);
2990                 dout(" wake request %p tid %llu\n", req, req->r_tid);
2991                 __do_request(mdsc, req);
2992         }
2993 }
2994
2995 /*
2996  * Wake up threads with requests pending for @mds, so that they can
2997  * resubmit their requests to a possibly different mds.
2998  */
2999 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
3000 {
3001         struct ceph_mds_request *req;
3002         struct rb_node *p = rb_first(&mdsc->request_tree);
3003
3004         dout("kick_requests mds%d\n", mds);
3005         while (p) {
3006                 req = rb_entry(p, struct ceph_mds_request, r_node);
3007                 p = rb_next(p);
3008                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3009                         continue;
3010                 if (req->r_attempts > 0)
3011                         continue; /* only new requests */
3012                 if (req->r_session &&
3013                     req->r_session->s_mds == mds) {
3014                         dout(" kicking tid %llu\n", req->r_tid);
3015                         list_del_init(&req->r_wait);
3016                         __do_request(mdsc, req);
3017                 }
3018         }
3019 }
3020
3021 int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
3022                               struct ceph_mds_request *req)
3023 {
3024         int err = 0;
3025
3026         /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
3027         if (req->r_inode)
3028                 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3029         if (req->r_parent) {
3030                 struct ceph_inode_info *ci = ceph_inode(req->r_parent);
3031                 int fmode = (req->r_op & CEPH_MDS_OP_WRITE) ?
3032                             CEPH_FILE_MODE_WR : CEPH_FILE_MODE_RD;
3033                 spin_lock(&ci->i_ceph_lock);
3034                 ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
3035                 __ceph_touch_fmode(ci, mdsc, fmode);
3036                 spin_unlock(&ci->i_ceph_lock);
3037         }
3038         if (req->r_old_dentry_dir)
3039                 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
3040                                   CEPH_CAP_PIN);
3041
3042         if (req->r_inode) {
3043                 err = ceph_wait_on_async_create(req->r_inode);
3044                 if (err) {
3045                         dout("%s: wait for async create returned: %d\n",
3046                              __func__, err);
3047                         return err;
3048                 }
3049         }
3050
3051         if (!err && req->r_old_inode) {
3052                 err = ceph_wait_on_async_create(req->r_old_inode);
3053                 if (err) {
3054                         dout("%s: wait for async create returned: %d\n",
3055                              __func__, err);
3056                         return err;
3057                 }
3058         }
3059
3060         dout("submit_request on %p for inode %p\n", req, dir);
3061         mutex_lock(&mdsc->mutex);
3062         __register_request(mdsc, req, dir);
3063         __do_request(mdsc, req);
3064         err = req->r_err;
3065         mutex_unlock(&mdsc->mutex);
3066         return err;
3067 }
3068
3069 static int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
3070                                   struct ceph_mds_request *req)
3071 {
3072         int err;
3073
3074         /* wait */
3075         dout("do_request waiting\n");
3076         if (!req->r_timeout && req->r_wait_for_completion) {
3077                 err = req->r_wait_for_completion(mdsc, req);
3078         } else {
3079                 long timeleft = wait_for_completion_killable_timeout(
3080                                         &req->r_completion,
3081                                         ceph_timeout_jiffies(req->r_timeout));
3082                 if (timeleft > 0)
3083                         err = 0;
3084                 else if (!timeleft)
3085                         err = -ETIMEDOUT;  /* timed out */
3086                 else
3087                         err = timeleft;  /* killed */
3088         }
3089         dout("do_request waited, got %d\n", err);
3090         mutex_lock(&mdsc->mutex);
3091
3092         /* only abort if we didn't race with a real reply */
3093         if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
3094                 err = le32_to_cpu(req->r_reply_info.head->result);
3095         } else if (err < 0) {
3096                 dout("aborted request %lld with %d\n", req->r_tid, err);
3097
3098                 /*
3099                  * ensure we aren't running concurrently with
3100                  * ceph_fill_trace or ceph_readdir_prepopulate, which
3101                  * rely on locks (dir mutex) held by our caller.
3102                  */
3103                 mutex_lock(&req->r_fill_mutex);
3104                 req->r_err = err;
3105                 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3106                 mutex_unlock(&req->r_fill_mutex);
3107
3108                 if (req->r_parent &&
3109                     (req->r_op & CEPH_MDS_OP_WRITE))
3110                         ceph_invalidate_dir_request(req);
3111         } else {
3112                 err = req->r_err;
3113         }
3114
3115         mutex_unlock(&mdsc->mutex);
3116         return err;
3117 }
3118
3119 /*
3120  * Synchrously perform an mds request.  Take care of all of the
3121  * session setup, forwarding, retry details.
3122  */
3123 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
3124                          struct inode *dir,
3125                          struct ceph_mds_request *req)
3126 {
3127         int err;
3128
3129         dout("do_request on %p\n", req);
3130
3131         /* issue */
3132         err = ceph_mdsc_submit_request(mdsc, dir, req);
3133         if (!err)
3134                 err = ceph_mdsc_wait_request(mdsc, req);
3135         dout("do_request %p done, result %d\n", req, err);
3136         return err;
3137 }
3138
3139 /*
3140  * Invalidate dir's completeness, dentry lease state on an aborted MDS
3141  * namespace request.
3142  */
3143 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
3144 {
3145         struct inode *dir = req->r_parent;
3146         struct inode *old_dir = req->r_old_dentry_dir;
3147
3148         dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
3149
3150         ceph_dir_clear_complete(dir);
3151         if (old_dir)
3152                 ceph_dir_clear_complete(old_dir);
3153         if (req->r_dentry)
3154                 ceph_invalidate_dentry_lease(req->r_dentry);
3155         if (req->r_old_dentry)
3156                 ceph_invalidate_dentry_lease(req->r_old_dentry);
3157 }
3158
3159 /*
3160  * Handle mds reply.
3161  *
3162  * We take the session mutex and parse and process the reply immediately.
3163  * This preserves the logical ordering of replies, capabilities, etc., sent
3164  * by the MDS as they are applied to our local cache.
3165  */
3166 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
3167 {
3168         struct ceph_mds_client *mdsc = session->s_mdsc;
3169         struct ceph_mds_request *req;
3170         struct ceph_mds_reply_head *head = msg->front.iov_base;
3171         struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
3172         struct ceph_snap_realm *realm;
3173         u64 tid;
3174         int err, result;
3175         int mds = session->s_mds;
3176
3177         if (msg->front.iov_len < sizeof(*head)) {
3178                 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
3179                 ceph_msg_dump(msg);
3180                 return;
3181         }
3182
3183         /* get request, session */
3184         tid = le64_to_cpu(msg->hdr.tid);
3185         mutex_lock(&mdsc->mutex);
3186         req = lookup_get_request(mdsc, tid);
3187         if (!req) {
3188                 dout("handle_reply on unknown tid %llu\n", tid);
3189                 mutex_unlock(&mdsc->mutex);
3190                 return;
3191         }
3192         dout("handle_reply %p\n", req);
3193
3194         /* correct session? */
3195         if (req->r_session != session) {
3196                 pr_err("mdsc_handle_reply got %llu on session mds%d"
3197                        " not mds%d\n", tid, session->s_mds,
3198                        req->r_session ? req->r_session->s_mds : -1);
3199                 mutex_unlock(&mdsc->mutex);
3200                 goto out;
3201         }
3202
3203         /* dup? */
3204         if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
3205             (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
3206                 pr_warn("got a dup %s reply on %llu from mds%d\n",
3207                            head->safe ? "safe" : "unsafe", tid, mds);
3208                 mutex_unlock(&mdsc->mutex);
3209                 goto out;
3210         }
3211         if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
3212                 pr_warn("got unsafe after safe on %llu from mds%d\n",
3213                            tid, mds);
3214                 mutex_unlock(&mdsc->mutex);
3215                 goto out;
3216         }
3217
3218         result = le32_to_cpu(head->result);
3219
3220         /*
3221          * Handle an ESTALE
3222          * if we're not talking to the authority, send to them
3223          * if the authority has changed while we weren't looking,
3224          * send to new authority
3225          * Otherwise we just have to return an ESTALE
3226          */
3227         if (result == -ESTALE) {
3228                 dout("got ESTALE on request %llu\n", req->r_tid);
3229                 req->r_resend_mds = -1;
3230                 if (req->r_direct_mode != USE_AUTH_MDS) {
3231                         dout("not using auth, setting for that now\n");
3232                         req->r_direct_mode = USE_AUTH_MDS;
3233                         __do_request(mdsc, req);
3234                         mutex_unlock(&mdsc->mutex);
3235                         goto out;
3236                 } else  {
3237                         int mds = __choose_mds(mdsc, req, NULL);
3238                         if (mds >= 0 && mds != req->r_session->s_mds) {
3239                                 dout("but auth changed, so resending\n");
3240                                 __do_request(mdsc, req);
3241                                 mutex_unlock(&mdsc->mutex);
3242                                 goto out;
3243                         }
3244                 }
3245                 dout("have to return ESTALE on request %llu\n", req->r_tid);
3246         }
3247
3248
3249         if (head->safe) {
3250                 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
3251                 __unregister_request(mdsc, req);
3252
3253                 /* last request during umount? */
3254                 if (mdsc->stopping && !__get_oldest_req(mdsc))
3255                         complete_all(&mdsc->safe_umount_waiters);
3256
3257                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3258                         /*
3259                          * We already handled the unsafe response, now do the
3260                          * cleanup.  No need to examine the response; the MDS
3261                          * doesn't include any result info in the safe
3262                          * response.  And even if it did, there is nothing
3263                          * useful we could do with a revised return value.
3264                          */
3265                         dout("got safe reply %llu, mds%d\n", tid, mds);
3266
3267                         mutex_unlock(&mdsc->mutex);
3268                         goto out;
3269                 }
3270         } else {
3271                 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
3272                 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
3273         }
3274
3275         dout("handle_reply tid %lld result %d\n", tid, result);
3276         rinfo = &req->r_reply_info;
3277         if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
3278                 err = parse_reply_info(session, msg, rinfo, (u64)-1);
3279         else
3280                 err = parse_reply_info(session, msg, rinfo, session->s_con.peer_features);
3281         mutex_unlock(&mdsc->mutex);
3282
3283         /* Must find target inode outside of mutexes to avoid deadlocks */
3284         if ((err >= 0) && rinfo->head->is_target) {
3285                 struct inode *in;
3286                 struct ceph_vino tvino = {
3287                         .ino  = le64_to_cpu(rinfo->targeti.in->ino),
3288                         .snap = le64_to_cpu(rinfo->targeti.in->snapid)
3289                 };
3290
3291                 in = ceph_get_inode(mdsc->fsc->sb, tvino);
3292                 if (IS_ERR(in)) {
3293                         err = PTR_ERR(in);
3294                         mutex_lock(&session->s_mutex);
3295                         goto out_err;
3296                 }
3297                 req->r_target_inode = in;
3298         }
3299
3300         mutex_lock(&session->s_mutex);
3301         if (err < 0) {
3302                 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
3303                 ceph_msg_dump(msg);
3304                 goto out_err;
3305         }
3306
3307         /* snap trace */
3308         realm = NULL;
3309         if (rinfo->snapblob_len) {
3310                 down_write(&mdsc->snap_rwsem);
3311                 ceph_update_snap_trace(mdsc, rinfo->snapblob,
3312                                 rinfo->snapblob + rinfo->snapblob_len,
3313                                 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
3314                                 &realm);
3315                 downgrade_write(&mdsc->snap_rwsem);
3316         } else {
3317                 down_read(&mdsc->snap_rwsem);
3318         }
3319
3320         /* insert trace into our cache */
3321         mutex_lock(&req->r_fill_mutex);
3322         current->journal_info = req;
3323         err = ceph_fill_trace(mdsc->fsc->sb, req);
3324         if (err == 0) {
3325                 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
3326                                     req->r_op == CEPH_MDS_OP_LSSNAP))
3327                         ceph_readdir_prepopulate(req, req->r_session);
3328         }
3329         current->journal_info = NULL;
3330         mutex_unlock(&req->r_fill_mutex);
3331
3332         up_read(&mdsc->snap_rwsem);
3333         if (realm)
3334                 ceph_put_snap_realm(mdsc, realm);
3335
3336         if (err == 0) {
3337                 if (req->r_target_inode &&
3338                     test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3339                         struct ceph_inode_info *ci =
3340                                 ceph_inode(req->r_target_inode);
3341                         spin_lock(&ci->i_unsafe_lock);
3342                         list_add_tail(&req->r_unsafe_target_item,
3343                                       &ci->i_unsafe_iops);
3344                         spin_unlock(&ci->i_unsafe_lock);
3345                 }
3346
3347                 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
3348         }
3349 out_err:
3350         mutex_lock(&mdsc->mutex);
3351         if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3352                 if (err) {
3353                         req->r_err = err;
3354                 } else {
3355                         req->r_reply =  ceph_msg_get(msg);
3356                         set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
3357                 }
3358         } else {
3359                 dout("reply arrived after request %lld was aborted\n", tid);
3360         }
3361         mutex_unlock(&mdsc->mutex);
3362
3363         mutex_unlock(&session->s_mutex);
3364
3365         /* kick calling process */
3366         complete_request(mdsc, req);
3367
3368         ceph_update_metadata_metrics(&mdsc->metric, req->r_start_latency,
3369                                      req->r_end_latency, err);
3370 out:
3371         ceph_mdsc_put_request(req);
3372         return;
3373 }
3374
3375
3376
3377 /*
3378  * handle mds notification that our request has been forwarded.
3379  */
3380 static void handle_forward(struct ceph_mds_client *mdsc,
3381                            struct ceph_mds_session *session,
3382                            struct ceph_msg *msg)
3383 {
3384         struct ceph_mds_request *req;
3385         u64 tid = le64_to_cpu(msg->hdr.tid);
3386         u32 next_mds;
3387         u32 fwd_seq;
3388         int err = -EINVAL;
3389         void *p = msg->front.iov_base;
3390         void *end = p + msg->front.iov_len;
3391
3392         ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3393         next_mds = ceph_decode_32(&p);
3394         fwd_seq = ceph_decode_32(&p);
3395
3396         mutex_lock(&mdsc->mutex);
3397         req = lookup_get_request(mdsc, tid);
3398         if (!req) {
3399                 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
3400                 goto out;  /* dup reply? */
3401         }
3402
3403         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3404                 dout("forward tid %llu aborted, unregistering\n", tid);
3405                 __unregister_request(mdsc, req);
3406         } else if (fwd_seq <= req->r_num_fwd) {
3407                 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
3408                      tid, next_mds, req->r_num_fwd, fwd_seq);
3409         } else {
3410                 /* resend. forward race not possible; mds would drop */
3411                 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
3412                 BUG_ON(req->r_err);
3413                 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3414                 req->r_attempts = 0;
3415                 req->r_num_fwd = fwd_seq;
3416                 req->r_resend_mds = next_mds;
3417                 put_request_session(req);
3418                 __do_request(mdsc, req);
3419         }
3420         ceph_mdsc_put_request(req);
3421 out:
3422         mutex_unlock(&mdsc->mutex);
3423         return;
3424
3425 bad:
3426         pr_err("mdsc_handle_forward decode error err=%d\n", err);
3427 }
3428
3429 static int __decode_session_metadata(void **p, void *end,
3430                                      bool *blocklisted)
3431 {
3432         /* map<string,string> */
3433         u32 n;
3434         bool err_str;
3435         ceph_decode_32_safe(p, end, n, bad);
3436         while (n-- > 0) {
3437                 u32 len;
3438                 ceph_decode_32_safe(p, end, len, bad);
3439                 ceph_decode_need(p, end, len, bad);
3440                 err_str = !strncmp(*p, "error_string", len);
3441                 *p += len;
3442                 ceph_decode_32_safe(p, end, len, bad);
3443                 ceph_decode_need(p, end, len, bad);
3444                 /*
3445                  * Match "blocklisted (blacklisted)" from newer MDSes,
3446                  * or "blacklisted" from older MDSes.
3447                  */
3448                 if (err_str && strnstr(*p, "blacklisted", len))
3449                         *blocklisted = true;
3450                 *p += len;
3451         }
3452         return 0;
3453 bad:
3454         return -1;
3455 }
3456
3457 /*
3458  * handle a mds session control message
3459  */
3460 static void handle_session(struct ceph_mds_session *session,
3461                            struct ceph_msg *msg)
3462 {
3463         struct ceph_mds_client *mdsc = session->s_mdsc;
3464         int mds = session->s_mds;
3465         int msg_version = le16_to_cpu(msg->hdr.version);
3466         void *p = msg->front.iov_base;
3467         void *end = p + msg->front.iov_len;
3468         struct ceph_mds_session_head *h;
3469         u32 op;
3470         u64 seq, features = 0;
3471         int wake = 0;
3472         bool blocklisted = false;
3473
3474         /* decode */
3475         ceph_decode_need(&p, end, sizeof(*h), bad);
3476         h = p;
3477         p += sizeof(*h);
3478
3479         op = le32_to_cpu(h->op);
3480         seq = le64_to_cpu(h->seq);
3481
3482         if (msg_version >= 3) {
3483                 u32 len;
3484                 /* version >= 2, metadata */
3485                 if (__decode_session_metadata(&p, end, &blocklisted) < 0)
3486                         goto bad;
3487                 /* version >= 3, feature bits */
3488                 ceph_decode_32_safe(&p, end, len, bad);
3489                 if (len) {
3490                         ceph_decode_64_safe(&p, end, features, bad);
3491                         p += len - sizeof(features);
3492                 }
3493         }
3494
3495         mutex_lock(&mdsc->mutex);
3496         if (op == CEPH_SESSION_CLOSE) {
3497                 ceph_get_mds_session(session);
3498                 __unregister_session(mdsc, session);
3499         }
3500         /* FIXME: this ttl calculation is generous */
3501         session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
3502         mutex_unlock(&mdsc->mutex);
3503
3504         mutex_lock(&session->s_mutex);
3505
3506         dout("handle_session mds%d %s %p state %s seq %llu\n",
3507              mds, ceph_session_op_name(op), session,
3508              ceph_session_state_name(session->s_state), seq);
3509
3510         if (session->s_state == CEPH_MDS_SESSION_HUNG) {
3511                 session->s_state = CEPH_MDS_SESSION_OPEN;
3512                 pr_info("mds%d came back\n", session->s_mds);
3513         }
3514
3515         switch (op) {
3516         case CEPH_SESSION_OPEN:
3517                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3518                         pr_info("mds%d reconnect success\n", session->s_mds);
3519                 session->s_state = CEPH_MDS_SESSION_OPEN;
3520                 session->s_features = features;
3521                 renewed_caps(mdsc, session, 0);
3522                 if (test_bit(CEPHFS_FEATURE_METRIC_COLLECT, &session->s_features))
3523                         metric_schedule_delayed(&mdsc->metric);
3524                 wake = 1;
3525                 if (mdsc->stopping)
3526                         __close_session(mdsc, session);
3527                 break;
3528
3529         case CEPH_SESSION_RENEWCAPS:
3530                 if (session->s_renew_seq == seq)
3531                         renewed_caps(mdsc, session, 1);
3532                 break;
3533
3534         case CEPH_SESSION_CLOSE:
3535                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3536                         pr_info("mds%d reconnect denied\n", session->s_mds);
3537                 session->s_state = CEPH_MDS_SESSION_CLOSED;
3538                 cleanup_session_requests(mdsc, session);
3539                 remove_session_caps(session);
3540                 wake = 2; /* for good measure */
3541                 wake_up_all(&mdsc->session_close_wq);
3542                 break;
3543
3544         case CEPH_SESSION_STALE:
3545                 pr_info("mds%d caps went stale, renewing\n",
3546                         session->s_mds);
3547                 atomic_inc(&session->s_cap_gen);
3548                 session->s_cap_ttl = jiffies - 1;
3549                 send_renew_caps(mdsc, session);
3550                 break;
3551
3552         case CEPH_SESSION_RECALL_STATE:
3553                 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
3554                 break;
3555
3556         case CEPH_SESSION_FLUSHMSG:
3557                 send_flushmsg_ack(mdsc, session, seq);
3558                 break;
3559
3560         case CEPH_SESSION_FORCE_RO:
3561                 dout("force_session_readonly %p\n", session);
3562                 spin_lock(&session->s_cap_lock);
3563                 session->s_readonly = true;
3564                 spin_unlock(&session->s_cap_lock);
3565                 wake_up_session_caps(session, FORCE_RO);
3566                 break;
3567
3568         case CEPH_SESSION_REJECT:
3569                 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
3570                 pr_info("mds%d rejected session\n", session->s_mds);
3571                 session->s_state = CEPH_MDS_SESSION_REJECTED;
3572                 cleanup_session_requests(mdsc, session);
3573                 remove_session_caps(session);
3574                 if (blocklisted)
3575                         mdsc->fsc->blocklisted = true;
3576                 wake = 2; /* for good measure */
3577                 break;
3578
3579         default:
3580                 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
3581                 WARN_ON(1);
3582         }
3583
3584         mutex_unlock(&session->s_mutex);
3585         if (wake) {
3586                 mutex_lock(&mdsc->mutex);
3587                 __wake_requests(mdsc, &session->s_waiting);
3588                 if (wake == 2)
3589                         kick_requests(mdsc, mds);
3590                 mutex_unlock(&mdsc->mutex);
3591         }
3592         if (op == CEPH_SESSION_CLOSE)
3593                 ceph_put_mds_session(session);
3594         return;
3595
3596 bad:
3597         pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
3598                (int)msg->front.iov_len);
3599         ceph_msg_dump(msg);
3600         return;
3601 }
3602
3603 void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
3604 {
3605         int dcaps;
3606
3607         dcaps = xchg(&req->r_dir_caps, 0);
3608         if (dcaps) {
3609                 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3610                 ceph_put_cap_refs(ceph_inode(req->r_parent), dcaps);
3611         }
3612 }
3613
3614 void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req)
3615 {
3616         int dcaps;
3617
3618         dcaps = xchg(&req->r_dir_caps, 0);
3619         if (dcaps) {
3620                 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3621                 ceph_put_cap_refs_no_check_caps(ceph_inode(req->r_parent),
3622                                                 dcaps);
3623         }
3624 }
3625
3626 /*
3627  * called under session->mutex.
3628  */
3629 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
3630                                    struct ceph_mds_session *session)
3631 {
3632         struct ceph_mds_request *req, *nreq;
3633         struct rb_node *p;
3634
3635         dout("replay_unsafe_requests mds%d\n", session->s_mds);
3636
3637         mutex_lock(&mdsc->mutex);
3638         list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item)
3639                 __send_request(session, req, true);
3640
3641         /*
3642          * also re-send old requests when MDS enters reconnect stage. So that MDS
3643          * can process completed request in clientreplay stage.
3644          */
3645         p = rb_first(&mdsc->request_tree);
3646         while (p) {
3647                 req = rb_entry(p, struct ceph_mds_request, r_node);
3648                 p = rb_next(p);
3649                 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3650                         continue;
3651                 if (req->r_attempts == 0)
3652                         continue; /* only old requests */
3653                 if (!req->r_session)
3654                         continue;
3655                 if (req->r_session->s_mds != session->s_mds)
3656                         continue;
3657
3658                 ceph_mdsc_release_dir_caps_no_check(req);
3659
3660                 __send_request(session, req, true);
3661         }
3662         mutex_unlock(&mdsc->mutex);
3663 }
3664
3665 static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
3666 {
3667         struct ceph_msg *reply;
3668         struct ceph_pagelist *_pagelist;
3669         struct page *page;
3670         __le32 *addr;
3671         int err = -ENOMEM;
3672
3673         if (!recon_state->allow_multi)
3674                 return -ENOSPC;
3675
3676         /* can't handle message that contains both caps and realm */
3677         BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
3678
3679         /* pre-allocate new pagelist */
3680         _pagelist = ceph_pagelist_alloc(GFP_NOFS);
3681         if (!_pagelist)
3682                 return -ENOMEM;
3683
3684         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3685         if (!reply)
3686                 goto fail_msg;
3687
3688         /* placeholder for nr_caps */
3689         err = ceph_pagelist_encode_32(_pagelist, 0);
3690         if (err < 0)
3691                 goto fail;
3692
3693         if (recon_state->nr_caps) {
3694                 /* currently encoding caps */
3695                 err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
3696                 if (err)
3697                         goto fail;
3698         } else {
3699                 /* placeholder for nr_realms (currently encoding relams) */
3700                 err = ceph_pagelist_encode_32(_pagelist, 0);
3701                 if (err < 0)
3702                         goto fail;
3703         }
3704
3705         err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
3706         if (err)
3707                 goto fail;
3708
3709         page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
3710         addr = kmap_atomic(page);
3711         if (recon_state->nr_caps) {
3712                 /* currently encoding caps */
3713                 *addr = cpu_to_le32(recon_state->nr_caps);
3714         } else {
3715                 /* currently encoding relams */
3716                 *(addr + 1) = cpu_to_le32(recon_state->nr_realms);
3717         }
3718         kunmap_atomic(addr);
3719
3720         reply->hdr.version = cpu_to_le16(5);
3721         reply->hdr.compat_version = cpu_to_le16(4);
3722
3723         reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
3724         ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
3725
3726         ceph_con_send(&recon_state->session->s_con, reply);
3727         ceph_pagelist_release(recon_state->pagelist);
3728
3729         recon_state->pagelist = _pagelist;
3730         recon_state->nr_caps = 0;
3731         recon_state->nr_realms = 0;
3732         recon_state->msg_version = 5;
3733         return 0;
3734 fail:
3735         ceph_msg_put(reply);
3736 fail_msg:
3737         ceph_pagelist_release(_pagelist);
3738         return err;
3739 }
3740
3741 static struct dentry* d_find_primary(struct inode *inode)
3742 {
3743         struct dentry *alias, *dn = NULL;
3744
3745         if (hlist_empty(&inode->i_dentry))
3746                 return NULL;
3747
3748         spin_lock(&inode->i_lock);
3749         if (hlist_empty(&inode->i_dentry))
3750                 goto out_unlock;
3751
3752         if (S_ISDIR(inode->i_mode)) {
3753                 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
3754                 if (!IS_ROOT(alias))
3755                         dn = dget(alias);
3756                 goto out_unlock;
3757         }
3758
3759         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
3760                 spin_lock(&alias->d_lock);
3761                 if (!d_unhashed(alias) &&
3762                     (ceph_dentry(alias)->flags & CEPH_DENTRY_PRIMARY_LINK)) {
3763                         dn = dget_dlock(alias);
3764                 }
3765                 spin_unlock(&alias->d_lock);
3766                 if (dn)
3767                         break;
3768         }
3769 out_unlock:
3770         spin_unlock(&inode->i_lock);
3771         return dn;
3772 }
3773
3774 /*
3775  * Encode information about a cap for a reconnect with the MDS.
3776  */
3777 static int reconnect_caps_cb(struct inode *inode, struct ceph_cap *cap,
3778                           void *arg)
3779 {
3780         union {
3781                 struct ceph_mds_cap_reconnect v2;
3782                 struct ceph_mds_cap_reconnect_v1 v1;
3783         } rec;
3784         struct ceph_inode_info *ci = cap->ci;
3785         struct ceph_reconnect_state *recon_state = arg;
3786         struct ceph_pagelist *pagelist = recon_state->pagelist;
3787         struct dentry *dentry;
3788         char *path;
3789         int pathlen, err;
3790         u64 pathbase;
3791         u64 snap_follows;
3792
3793         dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
3794              inode, ceph_vinop(inode), cap, cap->cap_id,
3795              ceph_cap_string(cap->issued));
3796
3797         dentry = d_find_primary(inode);
3798         if (dentry) {
3799                 /* set pathbase to parent dir when msg_version >= 2 */
3800                 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase,
3801                                             recon_state->msg_version >= 2);
3802                 dput(dentry);
3803                 if (IS_ERR(path)) {
3804                         err = PTR_ERR(path);
3805                         goto out_err;
3806                 }
3807         } else {
3808                 path = NULL;
3809                 pathlen = 0;
3810                 pathbase = 0;
3811         }
3812
3813         spin_lock(&ci->i_ceph_lock);
3814         cap->seq = 0;        /* reset cap seq */
3815         cap->issue_seq = 0;  /* and issue_seq */
3816         cap->mseq = 0;       /* and migrate_seq */
3817         cap->cap_gen = atomic_read(&cap->session->s_cap_gen);
3818
3819         /* These are lost when the session goes away */
3820         if (S_ISDIR(inode->i_mode)) {
3821                 if (cap->issued & CEPH_CAP_DIR_CREATE) {
3822                         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
3823                         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
3824                 }
3825                 cap->issued &= ~CEPH_CAP_ANY_DIR_OPS;
3826         }
3827
3828         if (recon_state->msg_version >= 2) {
3829                 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
3830                 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3831                 rec.v2.issued = cpu_to_le32(cap->issued);
3832                 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3833                 rec.v2.pathbase = cpu_to_le64(pathbase);
3834                 rec.v2.flock_len = (__force __le32)
3835                         ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
3836         } else {
3837                 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
3838                 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3839                 rec.v1.issued = cpu_to_le32(cap->issued);
3840                 rec.v1.size = cpu_to_le64(i_size_read(inode));
3841                 ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
3842                 ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
3843                 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3844                 rec.v1.pathbase = cpu_to_le64(pathbase);
3845         }
3846
3847         if (list_empty(&ci->i_cap_snaps)) {
3848                 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3849         } else {
3850                 struct ceph_cap_snap *capsnap =
3851                         list_first_entry(&ci->i_cap_snaps,
3852                                          struct ceph_cap_snap, ci_item);
3853                 snap_follows = capsnap->follows;
3854         }
3855         spin_unlock(&ci->i_ceph_lock);
3856
3857         if (recon_state->msg_version >= 2) {
3858                 int num_fcntl_locks, num_flock_locks;
3859                 struct ceph_filelock *flocks = NULL;
3860                 size_t struct_len, total_len = sizeof(u64);
3861                 u8 struct_v = 0;
3862
3863 encode_again:
3864                 if (rec.v2.flock_len) {
3865                         ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
3866                 } else {
3867                         num_fcntl_locks = 0;
3868                         num_flock_locks = 0;
3869                 }
3870                 if (num_fcntl_locks + num_flock_locks > 0) {
3871                         flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
3872                                                sizeof(struct ceph_filelock),
3873                                                GFP_NOFS);
3874                         if (!flocks) {
3875                                 err = -ENOMEM;
3876                                 goto out_err;
3877                         }
3878                         err = ceph_encode_locks_to_buffer(inode, flocks,
3879                                                           num_fcntl_locks,
3880                                                           num_flock_locks);
3881                         if (err) {
3882                                 kfree(flocks);
3883                                 flocks = NULL;
3884                                 if (err == -ENOSPC)
3885                                         goto encode_again;
3886                                 goto out_err;
3887                         }
3888                 } else {
3889                         kfree(flocks);
3890                         flocks = NULL;
3891                 }
3892
3893                 if (recon_state->msg_version >= 3) {
3894                         /* version, compat_version and struct_len */
3895                         total_len += 2 * sizeof(u8) + sizeof(u32);
3896                         struct_v = 2;
3897                 }
3898                 /*
3899                  * number of encoded locks is stable, so copy to pagelist
3900                  */
3901                 struct_len = 2 * sizeof(u32) +
3902                             (num_fcntl_locks + num_flock_locks) *
3903                             sizeof(struct ceph_filelock);
3904                 rec.v2.flock_len = cpu_to_le32(struct_len);
3905
3906                 struct_len += sizeof(u32) + pathlen + sizeof(rec.v2);
3907
3908                 if (struct_v >= 2)
3909                         struct_len += sizeof(u64); /* snap_follows */
3910
3911                 total_len += struct_len;
3912
3913                 if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
3914                         err = send_reconnect_partial(recon_state);
3915                         if (err)
3916                                 goto out_freeflocks;
3917                         pagelist = recon_state->pagelist;
3918                 }
3919
3920                 err = ceph_pagelist_reserve(pagelist, total_len);
3921                 if (err)
3922                         goto out_freeflocks;
3923
3924                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3925                 if (recon_state->msg_version >= 3) {
3926                         ceph_pagelist_encode_8(pagelist, struct_v);
3927                         ceph_pagelist_encode_8(pagelist, 1);
3928                         ceph_pagelist_encode_32(pagelist, struct_len);
3929                 }
3930                 ceph_pagelist_encode_string(pagelist, path, pathlen);
3931                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3932                 ceph_locks_to_pagelist(flocks, pagelist,
3933                                        num_fcntl_locks, num_flock_locks);
3934                 if (struct_v >= 2)
3935                         ceph_pagelist_encode_64(pagelist, snap_follows);
3936 out_freeflocks:
3937                 kfree(flocks);
3938         } else {
3939                 err = ceph_pagelist_reserve(pagelist,
3940                                             sizeof(u64) + sizeof(u32) +
3941                                             pathlen + sizeof(rec.v1));
3942                 if (err)
3943                         goto out_err;
3944
3945                 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3946                 ceph_pagelist_encode_string(pagelist, path, pathlen);
3947                 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3948         }
3949
3950 out_err:
3951         ceph_mdsc_free_path(path, pathlen);
3952         if (!err)
3953                 recon_state->nr_caps++;
3954         return err;
3955 }
3956
3957 static int encode_snap_realms(struct ceph_mds_client *mdsc,
3958                               struct ceph_reconnect_state *recon_state)
3959 {
3960         struct rb_node *p;
3961         struct ceph_pagelist *pagelist = recon_state->pagelist;
3962         int err = 0;
3963
3964         if (recon_state->msg_version >= 4) {
3965                 err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
3966                 if (err < 0)
3967                         goto fail;
3968         }
3969
3970         /*
3971          * snaprealms.  we provide mds with the ino, seq (version), and
3972          * parent for all of our realms.  If the mds has any newer info,
3973          * it will tell us.
3974          */
3975         for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3976                 struct ceph_snap_realm *realm =
3977                        rb_entry(p, struct ceph_snap_realm, node);
3978                 struct ceph_mds_snaprealm_reconnect sr_rec;
3979
3980                 if (recon_state->msg_version >= 4) {
3981                         size_t need = sizeof(u8) * 2 + sizeof(u32) +
3982                                       sizeof(sr_rec);
3983
3984                         if (pagelist->length + need > RECONNECT_MAX_SIZE) {
3985                                 err = send_reconnect_partial(recon_state);
3986                                 if (err)
3987                                         goto fail;
3988                                 pagelist = recon_state->pagelist;
3989                         }
3990
3991                         err = ceph_pagelist_reserve(pagelist, need);
3992                         if (err)
3993                                 goto fail;
3994
3995                         ceph_pagelist_encode_8(pagelist, 1);
3996                         ceph_pagelist_encode_8(pagelist, 1);
3997                         ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
3998                 }
3999
4000                 dout(" adding snap realm %llx seq %lld parent %llx\n",
4001                      realm->ino, realm->seq, realm->parent_ino);
4002                 sr_rec.ino = cpu_to_le64(realm->ino);
4003                 sr_rec.seq = cpu_to_le64(realm->seq);
4004                 sr_rec.parent = cpu_to_le64(realm->parent_ino);
4005
4006                 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
4007                 if (err)
4008                         goto fail;
4009
4010                 recon_state->nr_realms++;
4011         }
4012 fail:
4013         return err;
4014 }
4015
4016
4017 /*
4018  * If an MDS fails and recovers, clients need to reconnect in order to
4019  * reestablish shared state.  This includes all caps issued through
4020  * this session _and_ the snap_realm hierarchy.  Because it's not
4021  * clear which snap realms the mds cares about, we send everything we
4022  * know about.. that ensures we'll then get any new info the
4023  * recovering MDS might have.
4024  *
4025  * This is a relatively heavyweight operation, but it's rare.
4026  */
4027 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
4028                                struct ceph_mds_session *session)
4029 {
4030         struct ceph_msg *reply;
4031         int mds = session->s_mds;
4032         int err = -ENOMEM;
4033         struct ceph_reconnect_state recon_state = {
4034                 .session = session,
4035         };
4036         LIST_HEAD(dispose);
4037
4038         pr_info("mds%d reconnect start\n", mds);
4039
4040         recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
4041         if (!recon_state.pagelist)
4042                 goto fail_nopagelist;
4043
4044         reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
4045         if (!reply)
4046                 goto fail_nomsg;
4047
4048         xa_destroy(&session->s_delegated_inos);
4049
4050         mutex_lock(&session->s_mutex);
4051         session->s_state = CEPH_MDS_SESSION_RECONNECTING;
4052         session->s_seq = 0;
4053
4054         dout("session %p state %s\n", session,
4055              ceph_session_state_name(session->s_state));
4056
4057         atomic_inc(&session->s_cap_gen);
4058
4059         spin_lock(&session->s_cap_lock);
4060         /* don't know if session is readonly */
4061         session->s_readonly = 0;
4062         /*
4063          * notify __ceph_remove_cap() that we are composing cap reconnect.
4064          * If a cap get released before being added to the cap reconnect,
4065          * __ceph_remove_cap() should skip queuing cap release.
4066          */
4067         session->s_cap_reconnect = 1;
4068         /* drop old cap expires; we're about to reestablish that state */
4069         detach_cap_releases(session, &dispose);
4070         spin_unlock(&session->s_cap_lock);
4071         dispose_cap_releases(mdsc, &dispose);
4072
4073         /* trim unused caps to reduce MDS's cache rejoin time */
4074         if (mdsc->fsc->sb->s_root)
4075                 shrink_dcache_parent(mdsc->fsc->sb->s_root);
4076
4077         ceph_con_close(&session->s_con);
4078         ceph_con_open(&session->s_con,
4079                       CEPH_ENTITY_TYPE_MDS, mds,
4080                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
4081
4082         /* replay unsafe requests */
4083         replay_unsafe_requests(mdsc, session);
4084
4085         ceph_early_kick_flushing_caps(mdsc, session);
4086
4087         down_read(&mdsc->snap_rwsem);
4088
4089         /* placeholder for nr_caps */
4090         err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
4091         if (err)
4092                 goto fail;
4093
4094         if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
4095                 recon_state.msg_version = 3;
4096                 recon_state.allow_multi = true;
4097         } else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
4098                 recon_state.msg_version = 3;
4099         } else {
4100                 recon_state.msg_version = 2;
4101         }
4102         /* trsaverse this session's caps */
4103         err = ceph_iterate_session_caps(session, reconnect_caps_cb, &recon_state);
4104
4105         spin_lock(&session->s_cap_lock);
4106         session->s_cap_reconnect = 0;
4107         spin_unlock(&session->s_cap_lock);
4108
4109         if (err < 0)
4110                 goto fail;
4111
4112         /* check if all realms can be encoded into current message */
4113         if (mdsc->num_snap_realms) {
4114                 size_t total_len =
4115                         recon_state.pagelist->length +
4116                         mdsc->num_snap_realms *
4117                         sizeof(struct ceph_mds_snaprealm_reconnect);
4118                 if (recon_state.msg_version >= 4) {
4119                         /* number of realms */
4120                         total_len += sizeof(u32);
4121                         /* version, compat_version and struct_len */
4122                         total_len += mdsc->num_snap_realms *
4123                                      (2 * sizeof(u8) + sizeof(u32));
4124                 }
4125                 if (total_len > RECONNECT_MAX_SIZE) {
4126                         if (!recon_state.allow_multi) {
4127                                 err = -ENOSPC;
4128                                 goto fail;
4129                         }
4130                         if (recon_state.nr_caps) {
4131                                 err = send_reconnect_partial(&recon_state);
4132                                 if (err)
4133                                         goto fail;
4134                         }
4135                         recon_state.msg_version = 5;
4136                 }
4137         }
4138
4139         err = encode_snap_realms(mdsc, &recon_state);
4140         if (err < 0)
4141                 goto fail;
4142
4143         if (recon_state.msg_version >= 5) {
4144                 err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
4145                 if (err < 0)
4146                         goto fail;
4147         }
4148
4149         if (recon_state.nr_caps || recon_state.nr_realms) {
4150                 struct page *page =
4151                         list_first_entry(&recon_state.pagelist->head,
4152                                         struct page, lru);
4153                 __le32 *addr = kmap_atomic(page);
4154                 if (recon_state.nr_caps) {
4155                         WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
4156                         *addr = cpu_to_le32(recon_state.nr_caps);
4157                 } else if (recon_state.msg_version >= 4) {
4158                         *(addr + 1) = cpu_to_le32(recon_state.nr_realms);
4159                 }
4160                 kunmap_atomic(addr);
4161         }
4162
4163         reply->hdr.version = cpu_to_le16(recon_state.msg_version);
4164         if (recon_state.msg_version >= 4)
4165                 reply->hdr.compat_version = cpu_to_le16(4);
4166
4167         reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
4168         ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
4169
4170         ceph_con_send(&session->s_con, reply);
4171
4172         mutex_unlock(&session->s_mutex);
4173
4174         mutex_lock(&mdsc->mutex);
4175         __wake_requests(mdsc, &session->s_waiting);
4176         mutex_unlock(&mdsc->mutex);
4177
4178         up_read(&mdsc->snap_rwsem);
4179         ceph_pagelist_release(recon_state.pagelist);
4180         return;
4181
4182 fail:
4183         ceph_msg_put(reply);
4184         up_read(&mdsc->snap_rwsem);
4185         mutex_unlock(&session->s_mutex);
4186 fail_nomsg:
4187         ceph_pagelist_release(recon_state.pagelist);
4188 fail_nopagelist:
4189         pr_err("error %d preparing reconnect for mds%d\n", err, mds);
4190         return;
4191 }
4192
4193
4194 /*
4195  * compare old and new mdsmaps, kicking requests
4196  * and closing out old connections as necessary
4197  *
4198  * called under mdsc->mutex.
4199  */
4200 static void check_new_map(struct ceph_mds_client *mdsc,
4201                           struct ceph_mdsmap *newmap,
4202                           struct ceph_mdsmap *oldmap)
4203 {
4204         int i, j, err;
4205         int oldstate, newstate;
4206         struct ceph_mds_session *s;
4207         unsigned long targets[DIV_ROUND_UP(CEPH_MAX_MDS, sizeof(unsigned long))] = {0};
4208
4209         dout("check_new_map new %u old %u\n",
4210              newmap->m_epoch, oldmap->m_epoch);
4211
4212         if (newmap->m_info) {
4213                 for (i = 0; i < newmap->possible_max_rank; i++) {
4214                         for (j = 0; j < newmap->m_info[i].num_export_targets; j++)
4215                                 set_bit(newmap->m_info[i].export_targets[j], targets);
4216                 }
4217         }
4218
4219         for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4220                 if (!mdsc->sessions[i])
4221                         continue;
4222                 s = mdsc->sessions[i];
4223                 oldstate = ceph_mdsmap_get_state(oldmap, i);
4224                 newstate = ceph_mdsmap_get_state(newmap, i);
4225
4226                 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
4227                      i, ceph_mds_state_name(oldstate),
4228                      ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
4229                      ceph_mds_state_name(newstate),
4230                      ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
4231                      ceph_session_state_name(s->s_state));
4232
4233                 if (i >= newmap->possible_max_rank) {
4234                         /* force close session for stopped mds */
4235                         ceph_get_mds_session(s);
4236                         __unregister_session(mdsc, s);
4237                         __wake_requests(mdsc, &s->s_waiting);
4238                         mutex_unlock(&mdsc->mutex);
4239
4240                         mutex_lock(&s->s_mutex);
4241                         cleanup_session_requests(mdsc, s);
4242                         remove_session_caps(s);
4243                         mutex_unlock(&s->s_mutex);
4244
4245                         ceph_put_mds_session(s);
4246
4247                         mutex_lock(&mdsc->mutex);
4248                         kick_requests(mdsc, i);
4249                         continue;
4250                 }
4251
4252                 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
4253                            ceph_mdsmap_get_addr(newmap, i),
4254                            sizeof(struct ceph_entity_addr))) {
4255                         /* just close it */
4256                         mutex_unlock(&mdsc->mutex);
4257                         mutex_lock(&s->s_mutex);
4258                         mutex_lock(&mdsc->mutex);
4259                         ceph_con_close(&s->s_con);
4260                         mutex_unlock(&s->s_mutex);
4261                         s->s_state = CEPH_MDS_SESSION_RESTARTING;
4262                 } else if (oldstate == newstate) {
4263                         continue;  /* nothing new with this mds */
4264                 }
4265
4266                 /*
4267                  * send reconnect?
4268                  */
4269                 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
4270                     newstate >= CEPH_MDS_STATE_RECONNECT) {
4271                         mutex_unlock(&mdsc->mutex);
4272                         clear_bit(i, targets);
4273                         send_mds_reconnect(mdsc, s);
4274                         mutex_lock(&mdsc->mutex);
4275                 }
4276
4277                 /*
4278                  * kick request on any mds that has gone active.
4279                  */
4280                 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
4281                     newstate >= CEPH_MDS_STATE_ACTIVE) {
4282                         if (oldstate != CEPH_MDS_STATE_CREATING &&
4283                             oldstate != CEPH_MDS_STATE_STARTING)
4284                                 pr_info("mds%d recovery completed\n", s->s_mds);
4285                         kick_requests(mdsc, i);
4286                         mutex_unlock(&mdsc->mutex);
4287                         mutex_lock(&s->s_mutex);
4288                         mutex_lock(&mdsc->mutex);
4289                         ceph_kick_flushing_caps(mdsc, s);
4290                         mutex_unlock(&s->s_mutex);
4291                         wake_up_session_caps(s, RECONNECT);
4292                 }
4293         }
4294
4295         /*
4296          * Only open and reconnect sessions that don't exist yet.
4297          */
4298         for (i = 0; i < newmap->possible_max_rank; i++) {
4299                 /*
4300                  * In case the import MDS is crashed just after
4301                  * the EImportStart journal is flushed, so when
4302                  * a standby MDS takes over it and is replaying
4303                  * the EImportStart journal the new MDS daemon
4304                  * will wait the client to reconnect it, but the
4305                  * client may never register/open the session yet.
4306                  *
4307                  * Will try to reconnect that MDS daemon if the
4308                  * rank number is in the export targets array and
4309                  * is the up:reconnect state.
4310                  */
4311                 newstate = ceph_mdsmap_get_state(newmap, i);
4312                 if (!test_bit(i, targets) || newstate != CEPH_MDS_STATE_RECONNECT)
4313                         continue;
4314
4315                 /*
4316                  * The session maybe registered and opened by some
4317                  * requests which were choosing random MDSes during
4318                  * the mdsc->mutex's unlock/lock gap below in rare
4319                  * case. But the related MDS daemon will just queue
4320                  * that requests and be still waiting for the client's
4321                  * reconnection request in up:reconnect state.
4322                  */
4323                 s = __ceph_lookup_mds_session(mdsc, i);
4324                 if (likely(!s)) {
4325                         s = __open_export_target_session(mdsc, i);
4326                         if (IS_ERR(s)) {
4327                                 err = PTR_ERR(s);
4328                                 pr_err("failed to open export target session, err %d\n",
4329                                        err);
4330                                 continue;
4331                         }
4332                 }
4333                 dout("send reconnect to export target mds.%d\n", i);
4334                 mutex_unlock(&mdsc->mutex);
4335                 send_mds_reconnect(mdsc, s);
4336                 ceph_put_mds_session(s);
4337                 mutex_lock(&mdsc->mutex);
4338         }
4339
4340         for (i = 0; i < newmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4341                 s = mdsc->sessions[i];
4342                 if (!s)
4343                         continue;
4344                 if (!ceph_mdsmap_is_laggy(newmap, i))
4345                         continue;
4346                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4347                     s->s_state == CEPH_MDS_SESSION_HUNG ||
4348                     s->s_state == CEPH_MDS_SESSION_CLOSING) {
4349                         dout(" connecting to export targets of laggy mds%d\n",
4350                              i);
4351                         __open_export_target_sessions(mdsc, s);
4352                 }
4353         }
4354 }
4355
4356
4357
4358 /*
4359  * leases
4360  */
4361
4362 /*
4363  * caller must hold session s_mutex, dentry->d_lock
4364  */
4365 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
4366 {
4367         struct ceph_dentry_info *di = ceph_dentry(dentry);
4368
4369         ceph_put_mds_session(di->lease_session);
4370         di->lease_session = NULL;
4371 }
4372
4373 static void handle_lease(struct ceph_mds_client *mdsc,
4374                          struct ceph_mds_session *session,
4375                          struct ceph_msg *msg)
4376 {
4377         struct super_block *sb = mdsc->fsc->sb;
4378         struct inode *inode;
4379         struct dentry *parent, *dentry;
4380         struct ceph_dentry_info *di;
4381         int mds = session->s_mds;
4382         struct ceph_mds_lease *h = msg->front.iov_base;
4383         u32 seq;
4384         struct ceph_vino vino;
4385         struct qstr dname;
4386         int release = 0;
4387
4388         dout("handle_lease from mds%d\n", mds);
4389
4390         /* decode */
4391         if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
4392                 goto bad;
4393         vino.ino = le64_to_cpu(h->ino);
4394         vino.snap = CEPH_NOSNAP;
4395         seq = le32_to_cpu(h->seq);
4396         dname.len = get_unaligned_le32(h + 1);
4397         if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
4398                 goto bad;
4399         dname.name = (void *)(h + 1) + sizeof(u32);
4400
4401         /* lookup inode */
4402         inode = ceph_find_inode(sb, vino);
4403         dout("handle_lease %s, ino %llx %p %.*s\n",
4404              ceph_lease_op_name(h->action), vino.ino, inode,
4405              dname.len, dname.name);
4406
4407         mutex_lock(&session->s_mutex);
4408         inc_session_sequence(session);
4409
4410         if (!inode) {
4411                 dout("handle_lease no inode %llx\n", vino.ino);
4412                 goto release;
4413         }
4414
4415         /* dentry */
4416         parent = d_find_alias(inode);
4417         if (!parent) {
4418                 dout("no parent dentry on inode %p\n", inode);
4419                 WARN_ON(1);
4420                 goto release;  /* hrm... */
4421         }
4422         dname.hash = full_name_hash(parent, dname.name, dname.len);
4423         dentry = d_lookup(parent, &dname);
4424         dput(parent);
4425         if (!dentry)
4426                 goto release;
4427
4428         spin_lock(&dentry->d_lock);
4429         di = ceph_dentry(dentry);
4430         switch (h->action) {
4431         case CEPH_MDS_LEASE_REVOKE:
4432                 if (di->lease_session == session) {
4433                         if (ceph_seq_cmp(di->lease_seq, seq) > 0)
4434                                 h->seq = cpu_to_le32(di->lease_seq);
4435                         __ceph_mdsc_drop_dentry_lease(dentry);
4436                 }
4437                 release = 1;
4438                 break;
4439
4440         case CEPH_MDS_LEASE_RENEW:
4441                 if (di->lease_session == session &&
4442                     di->lease_gen == atomic_read(&session->s_cap_gen) &&
4443                     di->lease_renew_from &&
4444                     di->lease_renew_after == 0) {
4445                         unsigned long duration =
4446                                 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
4447
4448                         di->lease_seq = seq;
4449                         di->time = di->lease_renew_from + duration;
4450                         di->lease_renew_after = di->lease_renew_from +
4451                                 (duration >> 1);
4452                         di->lease_renew_from = 0;
4453                 }
4454                 break;
4455         }
4456         spin_unlock(&dentry->d_lock);
4457         dput(dentry);
4458
4459         if (!release)
4460                 goto out;
4461
4462 release:
4463         /* let's just reuse the same message */
4464         h->action = CEPH_MDS_LEASE_REVOKE_ACK;
4465         ceph_msg_get(msg);
4466         ceph_con_send(&session->s_con, msg);
4467
4468 out:
4469         mutex_unlock(&session->s_mutex);
4470         iput(inode);
4471         return;
4472
4473 bad:
4474         pr_err("corrupt lease message\n");
4475         ceph_msg_dump(msg);
4476 }
4477
4478 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
4479                               struct dentry *dentry, char action,
4480                               u32 seq)
4481 {
4482         struct ceph_msg *msg;
4483         struct ceph_mds_lease *lease;
4484         struct inode *dir;
4485         int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
4486
4487         dout("lease_send_msg identry %p %s to mds%d\n",
4488              dentry, ceph_lease_op_name(action), session->s_mds);
4489
4490         msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
4491         if (!msg)
4492                 return;
4493         lease = msg->front.iov_base;
4494         lease->action = action;
4495         lease->seq = cpu_to_le32(seq);
4496
4497         spin_lock(&dentry->d_lock);
4498         dir = d_inode(dentry->d_parent);
4499         lease->ino = cpu_to_le64(ceph_ino(dir));
4500         lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
4501
4502         put_unaligned_le32(dentry->d_name.len, lease + 1);
4503         memcpy((void *)(lease + 1) + 4,
4504                dentry->d_name.name, dentry->d_name.len);
4505         spin_unlock(&dentry->d_lock);
4506         /*
4507          * if this is a preemptive lease RELEASE, no need to
4508          * flush request stream, since the actual request will
4509          * soon follow.
4510          */
4511         msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
4512
4513         ceph_con_send(&session->s_con, msg);
4514 }
4515
4516 /*
4517  * lock unlock the session, to wait ongoing session activities
4518  */
4519 static void lock_unlock_session(struct ceph_mds_session *s)
4520 {
4521         mutex_lock(&s->s_mutex);
4522         mutex_unlock(&s->s_mutex);
4523 }
4524
4525 static void maybe_recover_session(struct ceph_mds_client *mdsc)
4526 {
4527         struct ceph_fs_client *fsc = mdsc->fsc;
4528
4529         if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
4530                 return;
4531
4532         if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
4533                 return;
4534
4535         if (!READ_ONCE(fsc->blocklisted))
4536                 return;
4537
4538         pr_info("auto reconnect after blocklisted\n");
4539         ceph_force_reconnect(fsc->sb);
4540 }
4541
4542 bool check_session_state(struct ceph_mds_session *s)
4543 {
4544         struct ceph_fs_client *fsc = s->s_mdsc->fsc;
4545
4546         switch (s->s_state) {
4547         case CEPH_MDS_SESSION_OPEN:
4548                 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
4549                         s->s_state = CEPH_MDS_SESSION_HUNG;
4550                         pr_info("mds%d hung\n", s->s_mds);
4551                 }
4552                 break;
4553         case CEPH_MDS_SESSION_CLOSING:
4554                 /* Should never reach this when not force unmounting */
4555                 WARN_ON_ONCE(s->s_ttl &&
4556                              READ_ONCE(fsc->mount_state) != CEPH_MOUNT_SHUTDOWN);
4557                 fallthrough;
4558         case CEPH_MDS_SESSION_NEW:
4559         case CEPH_MDS_SESSION_RESTARTING:
4560         case CEPH_MDS_SESSION_CLOSED:
4561         case CEPH_MDS_SESSION_REJECTED:
4562                 return false;
4563         }
4564
4565         return true;
4566 }
4567
4568 /*
4569  * If the sequence is incremented while we're waiting on a REQUEST_CLOSE reply,
4570  * then we need to retransmit that request.
4571  */
4572 void inc_session_sequence(struct ceph_mds_session *s)
4573 {
4574         lockdep_assert_held(&s->s_mutex);
4575
4576         s->s_seq++;
4577
4578         if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
4579                 int ret;
4580
4581                 dout("resending session close request for mds%d\n", s->s_mds);
4582                 ret = request_close_session(s);
4583                 if (ret < 0)
4584                         pr_err("unable to close session to mds%d: %d\n",
4585                                s->s_mds, ret);
4586         }
4587 }
4588
4589 /*
4590  * delayed work -- periodically trim expired leases, renew caps with mds.  If
4591  * the @delay parameter is set to 0 or if it's more than 5 secs, the default
4592  * workqueue delay value of 5 secs will be used.
4593  */
4594 static void schedule_delayed(struct ceph_mds_client *mdsc, unsigned long delay)
4595 {
4596         unsigned long max_delay = HZ * 5;
4597
4598         /* 5 secs default delay */
4599         if (!delay || (delay > max_delay))
4600                 delay = max_delay;
4601         schedule_delayed_work(&mdsc->delayed_work,
4602                               round_jiffies_relative(delay));
4603 }
4604
4605 static void delayed_work(struct work_struct *work)
4606 {
4607         struct ceph_mds_client *mdsc =
4608                 container_of(work, struct ceph_mds_client, delayed_work.work);
4609         unsigned long delay;
4610         int renew_interval;
4611         int renew_caps;
4612         int i;
4613
4614         dout("mdsc delayed_work\n");
4615
4616         if (mdsc->stopping)
4617                 return;
4618
4619         mutex_lock(&mdsc->mutex);
4620         renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
4621         renew_caps = time_after_eq(jiffies, HZ*renew_interval +
4622                                    mdsc->last_renew_caps);
4623         if (renew_caps)
4624                 mdsc->last_renew_caps = jiffies;
4625
4626         for (i = 0; i < mdsc->max_sessions; i++) {
4627                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4628                 if (!s)
4629                         continue;
4630
4631                 if (!check_session_state(s)) {
4632                         ceph_put_mds_session(s);
4633                         continue;
4634                 }
4635                 mutex_unlock(&mdsc->mutex);
4636
4637                 mutex_lock(&s->s_mutex);
4638                 if (renew_caps)
4639                         send_renew_caps(mdsc, s);
4640                 else
4641                         ceph_con_keepalive(&s->s_con);
4642                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4643                     s->s_state == CEPH_MDS_SESSION_HUNG)
4644                         ceph_send_cap_releases(mdsc, s);
4645                 mutex_unlock(&s->s_mutex);
4646                 ceph_put_mds_session(s);
4647
4648                 mutex_lock(&mdsc->mutex);
4649         }
4650         mutex_unlock(&mdsc->mutex);
4651
4652         delay = ceph_check_delayed_caps(mdsc);
4653
4654         ceph_queue_cap_reclaim_work(mdsc);
4655
4656         ceph_trim_snapid_map(mdsc);
4657
4658         maybe_recover_session(mdsc);
4659
4660         schedule_delayed(mdsc, delay);
4661 }
4662
4663 int ceph_mdsc_init(struct ceph_fs_client *fsc)
4664
4665 {
4666         struct ceph_mds_client *mdsc;
4667         int err;
4668
4669         mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
4670         if (!mdsc)
4671                 return -ENOMEM;
4672         mdsc->fsc = fsc;
4673         mutex_init(&mdsc->mutex);
4674         mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
4675         if (!mdsc->mdsmap) {
4676                 err = -ENOMEM;
4677                 goto err_mdsc;
4678         }
4679
4680         init_completion(&mdsc->safe_umount_waiters);
4681         init_waitqueue_head(&mdsc->session_close_wq);
4682         INIT_LIST_HEAD(&mdsc->waiting_for_map);
4683         mdsc->quotarealms_inodes = RB_ROOT;
4684         mutex_init(&mdsc->quotarealms_inodes_mutex);
4685         init_rwsem(&mdsc->snap_rwsem);
4686         mdsc->snap_realms = RB_ROOT;
4687         INIT_LIST_HEAD(&mdsc->snap_empty);
4688         spin_lock_init(&mdsc->snap_empty_lock);
4689         mdsc->request_tree = RB_ROOT;
4690         INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
4691         mdsc->last_renew_caps = jiffies;
4692         INIT_LIST_HEAD(&mdsc->cap_delay_list);
4693         INIT_LIST_HEAD(&mdsc->cap_wait_list);
4694         spin_lock_init(&mdsc->cap_delay_lock);
4695         INIT_LIST_HEAD(&mdsc->snap_flush_list);
4696         spin_lock_init(&mdsc->snap_flush_lock);
4697         mdsc->last_cap_flush_tid = 1;
4698         INIT_LIST_HEAD(&mdsc->cap_flush_list);
4699         INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
4700         spin_lock_init(&mdsc->cap_dirty_lock);
4701         init_waitqueue_head(&mdsc->cap_flushing_wq);
4702         INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
4703         err = ceph_metric_init(&mdsc->metric);
4704         if (err)
4705                 goto err_mdsmap;
4706
4707         spin_lock_init(&mdsc->dentry_list_lock);
4708         INIT_LIST_HEAD(&mdsc->dentry_leases);
4709         INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
4710
4711         ceph_caps_init(mdsc);
4712         ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
4713
4714         spin_lock_init(&mdsc->snapid_map_lock);
4715         mdsc->snapid_map_tree = RB_ROOT;
4716         INIT_LIST_HEAD(&mdsc->snapid_map_lru);
4717
4718         init_rwsem(&mdsc->pool_perm_rwsem);
4719         mdsc->pool_perm_tree = RB_ROOT;
4720
4721         strscpy(mdsc->nodename, utsname()->nodename,
4722                 sizeof(mdsc->nodename));
4723
4724         fsc->mdsc = mdsc;
4725         return 0;
4726
4727 err_mdsmap:
4728         kfree(mdsc->mdsmap);
4729 err_mdsc:
4730         kfree(mdsc);
4731         return err;
4732 }
4733
4734 /*
4735  * Wait for safe replies on open mds requests.  If we time out, drop
4736  * all requests from the tree to avoid dangling dentry refs.
4737  */
4738 static void wait_requests(struct ceph_mds_client *mdsc)
4739 {
4740         struct ceph_options *opts = mdsc->fsc->client->options;
4741         struct ceph_mds_request *req;
4742
4743         mutex_lock(&mdsc->mutex);
4744         if (__get_oldest_req(mdsc)) {
4745                 mutex_unlock(&mdsc->mutex);
4746
4747                 dout("wait_requests waiting for requests\n");
4748                 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
4749                                     ceph_timeout_jiffies(opts->mount_timeout));
4750
4751                 /* tear down remaining requests */
4752                 mutex_lock(&mdsc->mutex);
4753                 while ((req = __get_oldest_req(mdsc))) {
4754                         dout("wait_requests timed out on tid %llu\n",
4755                              req->r_tid);
4756                         list_del_init(&req->r_wait);
4757                         __unregister_request(mdsc, req);
4758                 }
4759         }
4760         mutex_unlock(&mdsc->mutex);
4761         dout("wait_requests done\n");
4762 }
4763
4764 void send_flush_mdlog(struct ceph_mds_session *s)
4765 {
4766         struct ceph_msg *msg;
4767
4768         /*
4769          * Pre-luminous MDS crashes when it sees an unknown session request
4770          */
4771         if (!CEPH_HAVE_FEATURE(s->s_con.peer_features, SERVER_LUMINOUS))
4772                 return;
4773
4774         mutex_lock(&s->s_mutex);
4775         dout("request mdlog flush to mds%d (%s)s seq %lld\n", s->s_mds,
4776              ceph_session_state_name(s->s_state), s->s_seq);
4777         msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_FLUSH_MDLOG,
4778                                       s->s_seq);
4779         if (!msg) {
4780                 pr_err("failed to request mdlog flush to mds%d (%s) seq %lld\n",
4781                        s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
4782         } else {
4783                 ceph_con_send(&s->s_con, msg);
4784         }
4785         mutex_unlock(&s->s_mutex);
4786 }
4787
4788 /*
4789  * called before mount is ro, and before dentries are torn down.
4790  * (hmm, does this still race with new lookups?)
4791  */
4792 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
4793 {
4794         dout("pre_umount\n");
4795         mdsc->stopping = 1;
4796
4797         ceph_mdsc_iterate_sessions(mdsc, send_flush_mdlog, true);
4798         ceph_mdsc_iterate_sessions(mdsc, lock_unlock_session, false);
4799         ceph_flush_dirty_caps(mdsc);
4800         wait_requests(mdsc);
4801
4802         /*
4803          * wait for reply handlers to drop their request refs and
4804          * their inode/dcache refs
4805          */
4806         ceph_msgr_flush();
4807
4808         ceph_cleanup_quotarealms_inodes(mdsc);
4809 }
4810
4811 /*
4812  * wait for all write mds requests to flush.
4813  */
4814 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
4815 {
4816         struct ceph_mds_request *req = NULL, *nextreq;
4817         struct rb_node *n;
4818
4819         mutex_lock(&mdsc->mutex);
4820         dout("wait_unsafe_requests want %lld\n", want_tid);
4821 restart:
4822         req = __get_oldest_req(mdsc);
4823         while (req && req->r_tid <= want_tid) {
4824                 /* find next request */
4825                 n = rb_next(&req->r_node);
4826                 if (n)
4827                         nextreq = rb_entry(n, struct ceph_mds_request, r_node);
4828                 else
4829                         nextreq = NULL;
4830                 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
4831                     (req->r_op & CEPH_MDS_OP_WRITE)) {
4832                         /* write op */
4833                         ceph_mdsc_get_request(req);
4834                         if (nextreq)
4835                                 ceph_mdsc_get_request(nextreq);
4836                         mutex_unlock(&mdsc->mutex);
4837                         dout("wait_unsafe_requests  wait on %llu (want %llu)\n",
4838                              req->r_tid, want_tid);
4839                         wait_for_completion(&req->r_safe_completion);
4840                         mutex_lock(&mdsc->mutex);
4841                         ceph_mdsc_put_request(req);
4842                         if (!nextreq)
4843                                 break;  /* next dne before, so we're done! */
4844                         if (RB_EMPTY_NODE(&nextreq->r_node)) {
4845                                 /* next request was removed from tree */
4846                                 ceph_mdsc_put_request(nextreq);
4847                                 goto restart;
4848                         }
4849                         ceph_mdsc_put_request(nextreq);  /* won't go away */
4850                 }
4851                 req = nextreq;
4852         }
4853         mutex_unlock(&mdsc->mutex);
4854         dout("wait_unsafe_requests done\n");
4855 }
4856
4857 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
4858 {
4859         u64 want_tid, want_flush;
4860
4861         if (READ_ONCE(mdsc->fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN)
4862                 return;
4863
4864         dout("sync\n");
4865         mutex_lock(&mdsc->mutex);
4866         want_tid = mdsc->last_tid;
4867         mutex_unlock(&mdsc->mutex);
4868
4869         ceph_flush_dirty_caps(mdsc);
4870         spin_lock(&mdsc->cap_dirty_lock);
4871         want_flush = mdsc->last_cap_flush_tid;
4872         if (!list_empty(&mdsc->cap_flush_list)) {
4873                 struct ceph_cap_flush *cf =
4874                         list_last_entry(&mdsc->cap_flush_list,
4875                                         struct ceph_cap_flush, g_list);
4876                 cf->wake = true;
4877         }
4878         spin_unlock(&mdsc->cap_dirty_lock);
4879
4880         dout("sync want tid %lld flush_seq %lld\n",
4881              want_tid, want_flush);
4882
4883         wait_unsafe_requests(mdsc, want_tid);
4884         wait_caps_flush(mdsc, want_flush);
4885 }
4886
4887 /*
4888  * true if all sessions are closed, or we force unmount
4889  */
4890 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
4891 {
4892         if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
4893                 return true;
4894         return atomic_read(&mdsc->num_sessions) <= skipped;
4895 }
4896
4897 /*
4898  * called after sb is ro.
4899  */
4900 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
4901 {
4902         struct ceph_options *opts = mdsc->fsc->client->options;
4903         struct ceph_mds_session *session;
4904         int i;
4905         int skipped = 0;
4906
4907         dout("close_sessions\n");
4908
4909         /* close sessions */
4910         mutex_lock(&mdsc->mutex);
4911         for (i = 0; i < mdsc->max_sessions; i++) {
4912                 session = __ceph_lookup_mds_session(mdsc, i);
4913                 if (!session)
4914                         continue;
4915                 mutex_unlock(&mdsc->mutex);
4916                 mutex_lock(&session->s_mutex);
4917                 if (__close_session(mdsc, session) <= 0)
4918                         skipped++;
4919                 mutex_unlock(&session->s_mutex);
4920                 ceph_put_mds_session(session);
4921                 mutex_lock(&mdsc->mutex);
4922         }
4923         mutex_unlock(&mdsc->mutex);
4924
4925         dout("waiting for sessions to close\n");
4926         wait_event_timeout(mdsc->session_close_wq,
4927                            done_closing_sessions(mdsc, skipped),
4928                            ceph_timeout_jiffies(opts->mount_timeout));
4929
4930         /* tear down remaining sessions */
4931         mutex_lock(&mdsc->mutex);
4932         for (i = 0; i < mdsc->max_sessions; i++) {
4933                 if (mdsc->sessions[i]) {
4934                         session = ceph_get_mds_session(mdsc->sessions[i]);
4935                         __unregister_session(mdsc, session);
4936                         mutex_unlock(&mdsc->mutex);
4937                         mutex_lock(&session->s_mutex);
4938                         remove_session_caps(session);
4939                         mutex_unlock(&session->s_mutex);
4940                         ceph_put_mds_session(session);
4941                         mutex_lock(&mdsc->mutex);
4942                 }
4943         }
4944         WARN_ON(!list_empty(&mdsc->cap_delay_list));
4945         mutex_unlock(&mdsc->mutex);
4946
4947         ceph_cleanup_snapid_map(mdsc);
4948         ceph_cleanup_empty_realms(mdsc);
4949
4950         cancel_work_sync(&mdsc->cap_reclaim_work);
4951         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
4952
4953         dout("stopped\n");
4954 }
4955
4956 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
4957 {
4958         struct ceph_mds_session *session;
4959         int mds;
4960
4961         dout("force umount\n");
4962
4963         mutex_lock(&mdsc->mutex);
4964         for (mds = 0; mds < mdsc->max_sessions; mds++) {
4965                 session = __ceph_lookup_mds_session(mdsc, mds);
4966                 if (!session)
4967                         continue;
4968
4969                 if (session->s_state == CEPH_MDS_SESSION_REJECTED)
4970                         __unregister_session(mdsc, session);
4971                 __wake_requests(mdsc, &session->s_waiting);
4972                 mutex_unlock(&mdsc->mutex);
4973
4974                 mutex_lock(&session->s_mutex);
4975                 __close_session(mdsc, session);
4976                 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
4977                         cleanup_session_requests(mdsc, session);
4978                         remove_session_caps(session);
4979                 }
4980                 mutex_unlock(&session->s_mutex);
4981                 ceph_put_mds_session(session);
4982
4983                 mutex_lock(&mdsc->mutex);
4984                 kick_requests(mdsc, mds);
4985         }
4986         __wake_requests(mdsc, &mdsc->waiting_for_map);
4987         mutex_unlock(&mdsc->mutex);
4988 }
4989
4990 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
4991 {
4992         dout("stop\n");
4993         /*
4994          * Make sure the delayed work stopped before releasing
4995          * the resources.
4996          *
4997          * Because the cancel_delayed_work_sync() will only
4998          * guarantee that the work finishes executing. But the
4999          * delayed work will re-arm itself again after that.
5000          */
5001         flush_delayed_work(&mdsc->delayed_work);
5002
5003         if (mdsc->mdsmap)
5004                 ceph_mdsmap_destroy(mdsc->mdsmap);
5005         kfree(mdsc->sessions);
5006         ceph_caps_finalize(mdsc);
5007         ceph_pool_perm_destroy(mdsc);
5008 }
5009
5010 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
5011 {
5012         struct ceph_mds_client *mdsc = fsc->mdsc;
5013         dout("mdsc_destroy %p\n", mdsc);
5014
5015         if (!mdsc)
5016                 return;
5017
5018         /* flush out any connection work with references to us */
5019         ceph_msgr_flush();
5020
5021         ceph_mdsc_stop(mdsc);
5022
5023         ceph_metric_destroy(&mdsc->metric);
5024
5025         fsc->mdsc = NULL;
5026         kfree(mdsc);
5027         dout("mdsc_destroy %p done\n", mdsc);
5028 }
5029
5030 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5031 {
5032         struct ceph_fs_client *fsc = mdsc->fsc;
5033         const char *mds_namespace = fsc->mount_options->mds_namespace;
5034         void *p = msg->front.iov_base;
5035         void *end = p + msg->front.iov_len;
5036         u32 epoch;
5037         u32 num_fs;
5038         u32 mount_fscid = (u32)-1;
5039         int err = -EINVAL;
5040
5041         ceph_decode_need(&p, end, sizeof(u32), bad);
5042         epoch = ceph_decode_32(&p);
5043
5044         dout("handle_fsmap epoch %u\n", epoch);
5045
5046         /* struct_v, struct_cv, map_len, epoch, legacy_client_fscid */
5047         ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 3, bad);
5048
5049         ceph_decode_32_safe(&p, end, num_fs, bad);
5050         while (num_fs-- > 0) {
5051                 void *info_p, *info_end;
5052                 u32 info_len;
5053                 u32 fscid, namelen;
5054
5055                 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
5056                 p += 2;         // info_v, info_cv
5057                 info_len = ceph_decode_32(&p);
5058                 ceph_decode_need(&p, end, info_len, bad);
5059                 info_p = p;
5060                 info_end = p + info_len;
5061                 p = info_end;
5062
5063                 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
5064                 fscid = ceph_decode_32(&info_p);
5065                 namelen = ceph_decode_32(&info_p);
5066                 ceph_decode_need(&info_p, info_end, namelen, bad);
5067
5068                 if (mds_namespace &&
5069                     strlen(mds_namespace) == namelen &&
5070                     !strncmp(mds_namespace, (char *)info_p, namelen)) {
5071                         mount_fscid = fscid;
5072                         break;
5073                 }
5074         }
5075
5076         ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
5077         if (mount_fscid != (u32)-1) {
5078                 fsc->client->monc.fs_cluster_id = mount_fscid;
5079                 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
5080                                    0, true);
5081                 ceph_monc_renew_subs(&fsc->client->monc);
5082         } else {
5083                 err = -ENOENT;
5084                 goto err_out;
5085         }
5086         return;
5087
5088 bad:
5089         pr_err("error decoding fsmap\n");
5090 err_out:
5091         mutex_lock(&mdsc->mutex);
5092         mdsc->mdsmap_err = err;
5093         __wake_requests(mdsc, &mdsc->waiting_for_map);
5094         mutex_unlock(&mdsc->mutex);
5095 }
5096
5097 /*
5098  * handle mds map update.
5099  */
5100 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5101 {
5102         u32 epoch;
5103         u32 maplen;
5104         void *p = msg->front.iov_base;
5105         void *end = p + msg->front.iov_len;
5106         struct ceph_mdsmap *newmap, *oldmap;
5107         struct ceph_fsid fsid;
5108         int err = -EINVAL;
5109
5110         ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
5111         ceph_decode_copy(&p, &fsid, sizeof(fsid));
5112         if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
5113                 return;
5114         epoch = ceph_decode_32(&p);
5115         maplen = ceph_decode_32(&p);
5116         dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
5117
5118         /* do we need it? */
5119         mutex_lock(&mdsc->mutex);
5120         if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
5121                 dout("handle_map epoch %u <= our %u\n",
5122                      epoch, mdsc->mdsmap->m_epoch);
5123                 mutex_unlock(&mdsc->mutex);
5124                 return;
5125         }
5126
5127         newmap = ceph_mdsmap_decode(&p, end, ceph_msgr2(mdsc->fsc->client));
5128         if (IS_ERR(newmap)) {
5129                 err = PTR_ERR(newmap);
5130                 goto bad_unlock;
5131         }
5132
5133         /* swap into place */
5134         if (mdsc->mdsmap) {
5135                 oldmap = mdsc->mdsmap;
5136                 mdsc->mdsmap = newmap;
5137                 check_new_map(mdsc, newmap, oldmap);
5138                 ceph_mdsmap_destroy(oldmap);
5139         } else {
5140                 mdsc->mdsmap = newmap;  /* first mds map */
5141         }
5142         mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
5143                                         MAX_LFS_FILESIZE);
5144
5145         __wake_requests(mdsc, &mdsc->waiting_for_map);
5146         ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
5147                           mdsc->mdsmap->m_epoch);
5148
5149         mutex_unlock(&mdsc->mutex);
5150         schedule_delayed(mdsc, 0);
5151         return;
5152
5153 bad_unlock:
5154         mutex_unlock(&mdsc->mutex);
5155 bad:
5156         pr_err("error decoding mdsmap %d\n", err);
5157         return;
5158 }
5159
5160 static struct ceph_connection *mds_get_con(struct ceph_connection *con)
5161 {
5162         struct ceph_mds_session *s = con->private;
5163
5164         if (ceph_get_mds_session(s))
5165                 return con;
5166         return NULL;
5167 }
5168
5169 static void mds_put_con(struct ceph_connection *con)
5170 {
5171         struct ceph_mds_session *s = con->private;
5172
5173         ceph_put_mds_session(s);
5174 }
5175
5176 /*
5177  * if the client is unresponsive for long enough, the mds will kill
5178  * the session entirely.
5179  */
5180 static void mds_peer_reset(struct ceph_connection *con)
5181 {
5182         struct ceph_mds_session *s = con->private;
5183         struct ceph_mds_client *mdsc = s->s_mdsc;
5184
5185         pr_warn("mds%d closed our session\n", s->s_mds);
5186         send_mds_reconnect(mdsc, s);
5187 }
5188
5189 static void mds_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5190 {
5191         struct ceph_mds_session *s = con->private;
5192         struct ceph_mds_client *mdsc = s->s_mdsc;
5193         int type = le16_to_cpu(msg->hdr.type);
5194
5195         mutex_lock(&mdsc->mutex);
5196         if (__verify_registered_session(mdsc, s) < 0) {
5197                 mutex_unlock(&mdsc->mutex);
5198                 goto out;
5199         }
5200         mutex_unlock(&mdsc->mutex);
5201
5202         switch (type) {
5203         case CEPH_MSG_MDS_MAP:
5204                 ceph_mdsc_handle_mdsmap(mdsc, msg);
5205                 break;
5206         case CEPH_MSG_FS_MAP_USER:
5207                 ceph_mdsc_handle_fsmap(mdsc, msg);
5208                 break;
5209         case CEPH_MSG_CLIENT_SESSION:
5210                 handle_session(s, msg);
5211                 break;
5212         case CEPH_MSG_CLIENT_REPLY:
5213                 handle_reply(s, msg);
5214                 break;
5215         case CEPH_MSG_CLIENT_REQUEST_FORWARD:
5216                 handle_forward(mdsc, s, msg);
5217                 break;
5218         case CEPH_MSG_CLIENT_CAPS:
5219                 ceph_handle_caps(s, msg);
5220                 break;
5221         case CEPH_MSG_CLIENT_SNAP:
5222                 ceph_handle_snap(mdsc, s, msg);
5223                 break;
5224         case CEPH_MSG_CLIENT_LEASE:
5225                 handle_lease(mdsc, s, msg);
5226                 break;
5227         case CEPH_MSG_CLIENT_QUOTA:
5228                 ceph_handle_quota(mdsc, s, msg);
5229                 break;
5230
5231         default:
5232                 pr_err("received unknown message type %d %s\n", type,
5233                        ceph_msg_type_name(type));
5234         }
5235 out:
5236         ceph_msg_put(msg);
5237 }
5238
5239 /*
5240  * authentication
5241  */
5242
5243 /*
5244  * Note: returned pointer is the address of a structure that's
5245  * managed separately.  Caller must *not* attempt to free it.
5246  */
5247 static struct ceph_auth_handshake *
5248 mds_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5249 {
5250         struct ceph_mds_session *s = con->private;
5251         struct ceph_mds_client *mdsc = s->s_mdsc;
5252         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5253         struct ceph_auth_handshake *auth = &s->s_auth;
5254         int ret;
5255
5256         ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5257                                          force_new, proto, NULL, NULL);
5258         if (ret)
5259                 return ERR_PTR(ret);
5260
5261         return auth;
5262 }
5263
5264 static int mds_add_authorizer_challenge(struct ceph_connection *con,
5265                                     void *challenge_buf, int challenge_buf_len)
5266 {
5267         struct ceph_mds_session *s = con->private;
5268         struct ceph_mds_client *mdsc = s->s_mdsc;
5269         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5270
5271         return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
5272                                             challenge_buf, challenge_buf_len);
5273 }
5274
5275 static int mds_verify_authorizer_reply(struct ceph_connection *con)
5276 {
5277         struct ceph_mds_session *s = con->private;
5278         struct ceph_mds_client *mdsc = s->s_mdsc;
5279         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5280         struct ceph_auth_handshake *auth = &s->s_auth;
5281
5282         return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5283                 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5284                 NULL, NULL, NULL, NULL);
5285 }
5286
5287 static int mds_invalidate_authorizer(struct ceph_connection *con)
5288 {
5289         struct ceph_mds_session *s = con->private;
5290         struct ceph_mds_client *mdsc = s->s_mdsc;
5291         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5292
5293         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
5294
5295         return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
5296 }
5297
5298 static int mds_get_auth_request(struct ceph_connection *con,
5299                                 void *buf, int *buf_len,
5300                                 void **authorizer, int *authorizer_len)
5301 {
5302         struct ceph_mds_session *s = con->private;
5303         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5304         struct ceph_auth_handshake *auth = &s->s_auth;
5305         int ret;
5306
5307         ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5308                                        buf, buf_len);
5309         if (ret)
5310                 return ret;
5311
5312         *authorizer = auth->authorizer_buf;
5313         *authorizer_len = auth->authorizer_buf_len;
5314         return 0;
5315 }
5316
5317 static int mds_handle_auth_reply_more(struct ceph_connection *con,
5318                                       void *reply, int reply_len,
5319                                       void *buf, int *buf_len,
5320                                       void **authorizer, int *authorizer_len)
5321 {
5322         struct ceph_mds_session *s = con->private;
5323         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5324         struct ceph_auth_handshake *auth = &s->s_auth;
5325         int ret;
5326
5327         ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5328                                               buf, buf_len);
5329         if (ret)
5330                 return ret;
5331
5332         *authorizer = auth->authorizer_buf;
5333         *authorizer_len = auth->authorizer_buf_len;
5334         return 0;
5335 }
5336
5337 static int mds_handle_auth_done(struct ceph_connection *con,
5338                                 u64 global_id, void *reply, int reply_len,
5339                                 u8 *session_key, int *session_key_len,
5340                                 u8 *con_secret, int *con_secret_len)
5341 {
5342         struct ceph_mds_session *s = con->private;
5343         struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5344         struct ceph_auth_handshake *auth = &s->s_auth;
5345
5346         return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5347                                                session_key, session_key_len,
5348                                                con_secret, con_secret_len);
5349 }
5350
5351 static int mds_handle_auth_bad_method(struct ceph_connection *con,
5352                                       int used_proto, int result,
5353                                       const int *allowed_protos, int proto_cnt,
5354                                       const int *allowed_modes, int mode_cnt)
5355 {
5356         struct ceph_mds_session *s = con->private;
5357         struct ceph_mon_client *monc = &s->s_mdsc->fsc->client->monc;
5358         int ret;
5359
5360         if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_MDS,
5361                                             used_proto, result,
5362                                             allowed_protos, proto_cnt,
5363                                             allowed_modes, mode_cnt)) {
5364                 ret = ceph_monc_validate_auth(monc);
5365                 if (ret)
5366                         return ret;
5367         }
5368
5369         return -EACCES;
5370 }
5371
5372 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
5373                                 struct ceph_msg_header *hdr, int *skip)
5374 {
5375         struct ceph_msg *msg;
5376         int type = (int) le16_to_cpu(hdr->type);
5377         int front_len = (int) le32_to_cpu(hdr->front_len);
5378
5379         if (con->in_msg)
5380                 return con->in_msg;
5381
5382         *skip = 0;
5383         msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
5384         if (!msg) {
5385                 pr_err("unable to allocate msg type %d len %d\n",
5386                        type, front_len);
5387                 return NULL;
5388         }
5389
5390         return msg;
5391 }
5392
5393 static int mds_sign_message(struct ceph_msg *msg)
5394 {
5395        struct ceph_mds_session *s = msg->con->private;
5396        struct ceph_auth_handshake *auth = &s->s_auth;
5397
5398        return ceph_auth_sign_message(auth, msg);
5399 }
5400
5401 static int mds_check_message_signature(struct ceph_msg *msg)
5402 {
5403        struct ceph_mds_session *s = msg->con->private;
5404        struct ceph_auth_handshake *auth = &s->s_auth;
5405
5406        return ceph_auth_check_message_signature(auth, msg);
5407 }
5408
5409 static const struct ceph_connection_operations mds_con_ops = {
5410         .get = mds_get_con,
5411         .put = mds_put_con,
5412         .alloc_msg = mds_alloc_msg,
5413         .dispatch = mds_dispatch,
5414         .peer_reset = mds_peer_reset,
5415         .get_authorizer = mds_get_authorizer,
5416         .add_authorizer_challenge = mds_add_authorizer_challenge,
5417         .verify_authorizer_reply = mds_verify_authorizer_reply,
5418         .invalidate_authorizer = mds_invalidate_authorizer,
5419         .sign_message = mds_sign_message,
5420         .check_message_signature = mds_check_message_signature,
5421         .get_auth_request = mds_get_auth_request,
5422         .handle_auth_reply_more = mds_handle_auth_reply_more,
5423         .handle_auth_done = mds_handle_auth_done,
5424         .handle_auth_bad_method = mds_handle_auth_bad_method,
5425 };
5426
5427 /* eof */