fs/ceph/mds_client.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/fs.h>
   4 #include <linux/wait.h>
   5 #include <linux/slab.h>
   6 #include <linux/gfp.h>
   7 #include <linux/sched.h>
   8 #include <linux/debugfs.h>
   9 #include <linux/seq_file.h>
  10 #include <linux/utsname.h>
  11 #include <linux/ratelimit.h>
  12
  13 #include "super.h"
  14 #include "mds_client.h"
  15
  16 #include <linux/ceph/ceph_features.h>
  17 #include <linux/ceph/messenger.h>
  18 #include <linux/ceph/decode.h>
  19 #include <linux/ceph/pagelist.h>
  20 #include <linux/ceph/auth.h>
  21 #include <linux/ceph/debugfs.h>
  22
  23 /*
  24  * A cluster of MDS (metadata server) daemons is responsible for
  25  * managing the file system namespace (the directory hierarchy and
  26  * inodes) and for coordinating shared access to storage.  Metadata is
  27  * partitioning hierarchically across a number of servers, and that
  28  * partition varies over time as the cluster adjusts the distribution
  29  * in order to balance load.
  30  *
  31  * The MDS client is primarily responsible to managing synchronous
  32  * metadata requests for operations like open, unlink, and so forth.
  33  * If there is a MDS failure, we find out about it when we (possibly
  34  * request and) receive a new MDS map, and can resubmit affected
  35  * requests.
  36  *
  37  * For the most part, though, we take advantage of a lossless
  38  * communications channel to the MDS, and do not need to worry about
  39  * timing out or resubmitting requests.
  40  *
  41  * We maintain a stateful "session" with each MDS we interact with.
  42  * Within each session, we sent periodic heartbeat messages to ensure
  43  * any capabilities or leases we have been issues remain valid.  If
  44  * the session times out and goes stale, our leases and capabilities
  45  * are no longer valid.
  46  */
  47
  48 struct ceph_reconnect_state {
  49         int nr_caps;
  50         struct ceph_pagelist *pagelist;
  51         bool flock;
  52 };
  53
  54 static void __wake_requests(struct ceph_mds_client *mdsc,
  55                             struct list_head *head);
  56
  57 static const struct ceph_connection_operations mds_con_ops;
  58
  59
  60 /*
  61  * mds reply parsing
  62  */
  63
  64 /*
  65  * parse individual inode info
  66  */
  67 static int parse_reply_info_in(void **p, void *end,
  68                                struct ceph_mds_reply_info_in *info,
  69                                u64 features)
  70 {
  71         int err = -EIO;
  72
  73         info->in = *p;
  74         *p += sizeof(struct ceph_mds_reply_inode) +
  75                 sizeof(*info->in->fragtree.splits) *
  76                 le32_to_cpu(info->in->fragtree.nsplits);
  77
  78         ceph_decode_32_safe(p, end, info->symlink_len, bad);
  79         ceph_decode_need(p, end, info->symlink_len, bad);
  80         info->symlink = *p;
  81         *p += info->symlink_len;
  82
  83         if (features & CEPH_FEATURE_DIRLAYOUTHASH)
  84                 ceph_decode_copy_safe(p, end, &info->dir_layout,
  85                                       sizeof(info->dir_layout), bad);
  86         else
  87                 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
  88
  89         ceph_decode_32_safe(p, end, info->xattr_len, bad);
  90         ceph_decode_need(p, end, info->xattr_len, bad);
  91         info->xattr_data = *p;
  92         *p += info->xattr_len;
  93
  94         if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
  95                 ceph_decode_64_safe(p, end, info->inline_version, bad);
  96                 ceph_decode_32_safe(p, end, info->inline_len, bad);
  97                 ceph_decode_need(p, end, info->inline_len, bad);
  98                 info->inline_data = *p;
  99                 *p += info->inline_len;
 100         } else
 101                 info->inline_version = CEPH_INLINE_NONE;
 102
 103         if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
 104                 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
 105                 ceph_decode_need(p, end, info->pool_ns_len, bad);
 106                 *p += info->pool_ns_len;
 107         } else {
 108                 info->pool_ns_len = 0;
 109         }
 110
 111         return 0;
 112 bad:
 113         return err;
 114 }
 115
 116 /*
 117  * parse a normal reply, which may contain a (dir+)dentry and/or a
 118  * target inode.
 119  */
 120 static int parse_reply_info_trace(void **p, void *end,
 121                                   struct ceph_mds_reply_info_parsed *info,
 122                                   u64 features)
 123 {
 124         int err;
 125
 126         if (info->head->is_dentry) {
 127                 err = parse_reply_info_in(p, end, &info->diri, features);
 128                 if (err < 0)
 129                         goto out_bad;
 130
 131                 if (unlikely(*p + sizeof(*info->dirfrag) > end))
 132                         goto bad;
 133                 info->dirfrag = *p;
 134                 *p += sizeof(*info->dirfrag) +
 135                         sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
 136                 if (unlikely(*p > end))
 137                         goto bad;
 138
 139                 ceph_decode_32_safe(p, end, info->dname_len, bad);
 140                 ceph_decode_need(p, end, info->dname_len, bad);
 141                 info->dname = *p;
 142                 *p += info->dname_len;
 143                 info->dlease = *p;
 144                 *p += sizeof(*info->dlease);
 145         }
 146
 147         if (info->head->is_target) {
 148                 err = parse_reply_info_in(p, end, &info->targeti, features);
 149                 if (err < 0)
 150                         goto out_bad;
 151         }
 152
 153         if (unlikely(*p != end))
 154                 goto bad;
 155         return 0;
 156
 157 bad:
 158         err = -EIO;
 159 out_bad:
 160         pr_err("problem parsing mds trace %d\n", err);
 161         return err;
 162 }
 163
 164 /*
 165  * parse readdir results
 166  */
 167 static int parse_reply_info_dir(void **p, void *end,
 168                                 struct ceph_mds_reply_info_parsed *info,
 169                                 u64 features)
 170 {
 171         u32 num, i = 0;
 172         int err;
 173
 174         info->dir_dir = *p;
 175         if (*p + sizeof(*info->dir_dir) > end)
 176                 goto bad;
 177         *p += sizeof(*info->dir_dir) +
 178                 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
 179         if (*p > end)
 180                 goto bad;
 181
 182         ceph_decode_need(p, end, sizeof(num) + 2, bad);
 183         num = ceph_decode_32(p);
 184         {
 185                 u16 flags = ceph_decode_16(p);
 186                 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
 187                 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
 188                 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
 189         }
 190         if (num == 0)
 191                 goto done;
 192
 193         BUG_ON(!info->dir_entries);
 194         if ((unsigned long)(info->dir_entries + num) >
 195             (unsigned long)info->dir_entries + info->dir_buf_size) {
 196                 pr_err("dir contents are larger than expected\n");
 197                 WARN_ON(1);
 198                 goto bad;
 199         }
 200
 201         info->dir_nr = num;
 202         while (num) {
 203                 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
 204                 /* dentry */
 205                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 206                 rde->name_len = ceph_decode_32(p);
 207                 ceph_decode_need(p, end, rde->name_len, bad);
 208                 rde->name = *p;
 209                 *p += rde->name_len;
 210                 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
 211                 rde->lease = *p;
 212                 *p += sizeof(struct ceph_mds_reply_lease);
 213
 214                 /* inode */
 215                 err = parse_reply_info_in(p, end, &rde->inode, features);
 216                 if (err < 0)
 217                         goto out_bad;
 218                 /* ceph_readdir_prepopulate() will update it */
 219                 rde->offset = 0;
 220                 i++;
 221                 num--;
 222         }
 223
 224 done:
 225         if (*p != end)
 226                 goto bad;
 227         return 0;
 228
 229 bad:
 230         err = -EIO;
 231 out_bad:
 232         pr_err("problem parsing dir contents %d\n", err);
 233         return err;
 234 }
 235
 236 /*
 237  * parse fcntl F_GETLK results
 238  */
 239 static int parse_reply_info_filelock(void **p, void *end,
 240                                      struct ceph_mds_reply_info_parsed *info,
 241                                      u64 features)
 242 {
 243         if (*p + sizeof(*info->filelock_reply) > end)
 244                 goto bad;
 245
 246         info->filelock_reply = *p;
 247         *p += sizeof(*info->filelock_reply);
 248
 249         if (unlikely(*p != end))
 250                 goto bad;
 251         return 0;
 252
 253 bad:
 254         return -EIO;
 255 }
 256
 257 /*
 258  * parse create results
 259  */
 260 static int parse_reply_info_create(void **p, void *end,
 261                                   struct ceph_mds_reply_info_parsed *info,
 262                                   u64 features)
 263 {
 264         if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
 265                 if (*p == end) {
 266                         info->has_create_ino = false;
 267                 } else {
 268                         info->has_create_ino = true;
 269                         info->ino = ceph_decode_64(p);
 270                 }
 271         }
 272
 273         if (unlikely(*p != end))
 274                 goto bad;
 275         return 0;
 276
 277 bad:
 278         return -EIO;
 279 }
 280
 281 /*
 282  * parse extra results
 283  */
 284 static int parse_reply_info_extra(void **p, void *end,
 285                                   struct ceph_mds_reply_info_parsed *info,
 286                                   u64 features)
 287 {
 288         if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
 289                 return parse_reply_info_filelock(p, end, info, features);
 290         else if (info->head->op == CEPH_MDS_OP_READDIR ||
 291                  info->head->op == CEPH_MDS_OP_LSSNAP)
 292                 return parse_reply_info_dir(p, end, info, features);
 293         else if (info->head->op == CEPH_MDS_OP_CREATE)
 294                 return parse_reply_info_create(p, end, info, features);
 295         else
 296                 return -EIO;
 297 }
 298
 299 /*
 300  * parse entire mds reply
 301  */
 302 static int parse_reply_info(struct ceph_msg *msg,
 303                             struct ceph_mds_reply_info_parsed *info,
 304                             u64 features)
 305 {
 306         void *p, *end;
 307         u32 len;
 308         int err;
 309
 310         info->head = msg->front.iov_base;
 311         p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
 312         end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
 313
 314         /* trace */
 315         ceph_decode_32_safe(&p, end, len, bad);
 316         if (len > 0) {
 317                 ceph_decode_need(&p, end, len, bad);
 318                 err = parse_reply_info_trace(&p, p+len, info, features);
 319                 if (err < 0)
 320                         goto out_bad;
 321         }
 322
 323         /* extra */
 324         ceph_decode_32_safe(&p, end, len, bad);
 325         if (len > 0) {
 326                 ceph_decode_need(&p, end, len, bad);
 327                 err = parse_reply_info_extra(&p, p+len, info, features);
 328                 if (err < 0)
 329                         goto out_bad;
 330         }
 331
 332         /* snap blob */
 333         ceph_decode_32_safe(&p, end, len, bad);
 334         info->snapblob_len = len;
 335         info->snapblob = p;
 336         p += len;
 337
 338         if (p != end)
 339                 goto bad;
 340         return 0;
 341
 342 bad:
 343         err = -EIO;
 344 out_bad:
 345         pr_err("mds parse_reply err %d\n", err);
 346         return err;
 347 }
 348
 349 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
 350 {
 351         if (!info->dir_entries)
 352                 return;
 353         free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
 354 }
 355
 356
 357 /*
 358  * sessions
 359  */
 360 const char *ceph_session_state_name(int s)
 361 {
 362         switch (s) {
 363         case CEPH_MDS_SESSION_NEW: return "new";
 364         case CEPH_MDS_SESSION_OPENING: return "opening";
 365         case CEPH_MDS_SESSION_OPEN: return "open";
 366         case CEPH_MDS_SESSION_HUNG: return "hung";
 367         case CEPH_MDS_SESSION_CLOSING: return "closing";
 368         case CEPH_MDS_SESSION_RESTARTING: return "restarting";
 369         case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
 370         default: return "???";
 371         }
 372 }
 373
 374 static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
 375 {
 376         if (atomic_inc_not_zero(&s->s_ref)) {
 377                 dout("mdsc get_session %p %d -> %d\n", s,
 378                      atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
 379                 return s;
 380         } else {
 381                 dout("mdsc get_session %p 0 -- FAIL", s);
 382                 return NULL;
 383         }
 384 }
 385
 386 void ceph_put_mds_session(struct ceph_mds_session *s)
 387 {
 388         dout("mdsc put_session %p %d -> %d\n", s,
 389              atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
 390         if (atomic_dec_and_test(&s->s_ref)) {
 391                 if (s->s_auth.authorizer)
 392                         ceph_auth_destroy_authorizer(s->s_auth.authorizer);
 393                 kfree(s);
 394         }
 395 }
 396
 397 /*
 398  * called under mdsc->mutex
 399  */
 400 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
 401                                                    int mds)
 402 {
 403         struct ceph_mds_session *session;
 404
 405         if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
 406                 return NULL;
 407         session = mdsc->sessions[mds];
 408         dout("lookup_mds_session %p %d\n", session,
 409              atomic_read(&session->s_ref));
 410         get_session(session);
 411         return session;
 412 }
 413
 414 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
 415 {
 416         if (mds >= mdsc->max_sessions)
 417                 return false;
 418         return mdsc->sessions[mds];
 419 }
 420
 421 static int __verify_registered_session(struct ceph_mds_client *mdsc,
 422                                        struct ceph_mds_session *s)
 423 {
 424         if (s->s_mds >= mdsc->max_sessions ||
 425             mdsc->sessions[s->s_mds] != s)
 426                 return -ENOENT;
 427         return 0;
 428 }
 429
 430 /*
 431  * create+register a new session for given mds.
 432  * called under mdsc->mutex.
 433  */
 434 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
 435                                                  int mds)
 436 {
 437         struct ceph_mds_session *s;
 438
 439         if (mds >= mdsc->mdsmap->m_max_mds)
 440                 return ERR_PTR(-EINVAL);
 441
 442         s = kzalloc(sizeof(*s), GFP_NOFS);
 443         if (!s)
 444                 return ERR_PTR(-ENOMEM);
 445         s->s_mdsc = mdsc;
 446         s->s_mds = mds;
 447         s->s_state = CEPH_MDS_SESSION_NEW;
 448         s->s_ttl = 0;
 449         s->s_seq = 0;
 450         mutex_init(&s->s_mutex);
 451
 452         ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
 453
 454         spin_lock_init(&s->s_gen_ttl_lock);
 455         s->s_cap_gen = 0;
 456         s->s_cap_ttl = jiffies - 1;
 457
 458         spin_lock_init(&s->s_cap_lock);
 459         s->s_renew_requested = 0;
 460         s->s_renew_seq = 0;
 461         INIT_LIST_HEAD(&s->s_caps);
 462         s->s_nr_caps = 0;
 463         s->s_trim_caps = 0;
 464         atomic_set(&s->s_ref, 1);
 465         INIT_LIST_HEAD(&s->s_waiting);
 466         INIT_LIST_HEAD(&s->s_unsafe);
 467         s->s_num_cap_releases = 0;
 468         s->s_cap_reconnect = 0;
 469         s->s_cap_iterator = NULL;
 470         INIT_LIST_HEAD(&s->s_cap_releases);
 471         INIT_LIST_HEAD(&s->s_cap_flushing);
 472         INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
 473
 474         dout("register_session mds%d\n", mds);
 475         if (mds >= mdsc->max_sessions) {
 476                 int newmax = 1 << get_count_order(mds+1);
 477                 struct ceph_mds_session **sa;
 478
 479                 dout("register_session realloc to %d\n", newmax);
 480                 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
 481                 if (sa == NULL)
 482                         goto fail_realloc;
 483                 if (mdsc->sessions) {
 484                         memcpy(sa, mdsc->sessions,
 485                                mdsc->max_sessions * sizeof(void *));
 486                         kfree(mdsc->sessions);
 487                 }
 488                 mdsc->sessions = sa;
 489                 mdsc->max_sessions = newmax;
 490         }
 491         mdsc->sessions[mds] = s;
 492         atomic_inc(&mdsc->num_sessions);
 493         atomic_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
 494
 495         ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
 496                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
 497
 498         return s;
 499
 500 fail_realloc:
 501         kfree(s);
 502         return ERR_PTR(-ENOMEM);
 503 }
 504
 505 /*
 506  * called under mdsc->mutex
 507  */
 508 static void __unregister_session(struct ceph_mds_client *mdsc,
 509                                struct ceph_mds_session *s)
 510 {
 511         dout("__unregister_session mds%d %p\n", s->s_mds, s);
 512         BUG_ON(mdsc->sessions[s->s_mds] != s);
 513         mdsc->sessions[s->s_mds] = NULL;
 514         ceph_con_close(&s->s_con);
 515         ceph_put_mds_session(s);
 516         atomic_dec(&mdsc->num_sessions);
 517 }
 518
 519 /*
 520  * drop session refs in request.
 521  *
 522  * should be last request ref, or hold mdsc->mutex
 523  */
 524 static void put_request_session(struct ceph_mds_request *req)
 525 {
 526         if (req->r_session) {
 527                 ceph_put_mds_session(req->r_session);
 528                 req->r_session = NULL;
 529         }
 530 }
 531
 532 void ceph_mdsc_release_request(struct kref *kref)
 533 {
 534         struct ceph_mds_request *req = container_of(kref,
 535                                                     struct ceph_mds_request,
 536                                                     r_kref);
 537         destroy_reply_info(&req->r_reply_info);
 538         if (req->r_request)
 539                 ceph_msg_put(req->r_request);
 540         if (req->r_reply)
 541                 ceph_msg_put(req->r_reply);
 542         if (req->r_inode) {
 543                 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
 544                 iput(req->r_inode);
 545         }
 546         if (req->r_locked_dir)
 547                 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
 548         iput(req->r_target_inode);
 549         if (req->r_dentry)
 550                 dput(req->r_dentry);
 551         if (req->r_old_dentry)
 552                 dput(req->r_old_dentry);
 553         if (req->r_old_dentry_dir) {
 554                 /*
 555                  * track (and drop pins for) r_old_dentry_dir
 556                  * separately, since r_old_dentry's d_parent may have
 557                  * changed between the dir mutex being dropped and
 558                  * this request being freed.
 559                  */
 560                 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
 561                                   CEPH_CAP_PIN);
 562                 iput(req->r_old_dentry_dir);
 563         }
 564         kfree(req->r_path1);
 565         kfree(req->r_path2);
 566         if (req->r_pagelist)
 567                 ceph_pagelist_release(req->r_pagelist);
 568         put_request_session(req);
 569         ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
 570         kfree(req);
 571 }
 572
 573 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
 574
 575 /*
 576  * lookup session, bump ref if found.
 577  *
 578  * called under mdsc->mutex.
 579  */
 580 static struct ceph_mds_request *
 581 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
 582 {
 583         struct ceph_mds_request *req;
 584
 585         req = lookup_request(&mdsc->request_tree, tid);
 586         if (req)
 587                 ceph_mdsc_get_request(req);
 588
 589         return req;
 590 }
 591
 592 /*
 593  * Register an in-flight request, and assign a tid.  Link to directory
 594  * are modifying (if any).
 595  *
 596  * Called under mdsc->mutex.
 597  */
 598 static void __register_request(struct ceph_mds_client *mdsc,
 599                                struct ceph_mds_request *req,
 600                                struct inode *dir)
 601 {
 602         req->r_tid = ++mdsc->last_tid;
 603         if (req->r_num_caps)
 604                 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
 605                                   req->r_num_caps);
 606         dout("__register_request %p tid %lld\n", req, req->r_tid);
 607         ceph_mdsc_get_request(req);
 608         insert_request(&mdsc->request_tree, req);
 609
 610         req->r_uid = current_fsuid();
 611         req->r_gid = current_fsgid();
 612
 613         if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
 614                 mdsc->oldest_tid = req->r_tid;
 615
 616         if (dir) {
 617                 ihold(dir);
 618                 req->r_unsafe_dir = dir;
 619         }
 620 }
 621
 622 static void __unregister_request(struct ceph_mds_client *mdsc,
 623                                  struct ceph_mds_request *req)
 624 {
 625         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
 626
 627         if (req->r_tid == mdsc->oldest_tid) {
 628                 struct rb_node *p = rb_next(&req->r_node);
 629                 mdsc->oldest_tid = 0;
 630                 while (p) {
 631                         struct ceph_mds_request *next_req =
 632                                 rb_entry(p, struct ceph_mds_request, r_node);
 633                         if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
 634                                 mdsc->oldest_tid = next_req->r_tid;
 635                                 break;
 636                         }
 637                         p = rb_next(p);
 638                 }
 639         }
 640
 641         erase_request(&mdsc->request_tree, req);
 642
 643         if (req->r_unsafe_dir && req->r_got_unsafe) {
 644                 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
 645                 spin_lock(&ci->i_unsafe_lock);
 646                 list_del_init(&req->r_unsafe_dir_item);
 647                 spin_unlock(&ci->i_unsafe_lock);
 648         }
 649         if (req->r_target_inode && req->r_got_unsafe) {
 650                 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
 651                 spin_lock(&ci->i_unsafe_lock);
 652                 list_del_init(&req->r_unsafe_target_item);
 653                 spin_unlock(&ci->i_unsafe_lock);
 654         }
 655
 656         if (req->r_unsafe_dir) {
 657                 iput(req->r_unsafe_dir);
 658                 req->r_unsafe_dir = NULL;
 659         }
 660
 661         complete_all(&req->r_safe_completion);
 662
 663         ceph_mdsc_put_request(req);
 664 }
 665
 666 /*
 667  * Choose mds to send request to next.  If there is a hint set in the
 668  * request (e.g., due to a prior forward hint from the mds), use that.
 669  * Otherwise, consult frag tree and/or caps to identify the
 670  * appropriate mds.  If all else fails, choose randomly.
 671  *
 672  * Called under mdsc->mutex.
 673  */
 674 static struct dentry *get_nonsnap_parent(struct dentry *dentry)
 675 {
 676         /*
 677          * we don't need to worry about protecting the d_parent access
 678          * here because we never renaming inside the snapped namespace
 679          * except to resplice to another snapdir, and either the old or new
 680          * result is a valid result.
 681          */
 682         while (!IS_ROOT(dentry) && ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
 683                 dentry = dentry->d_parent;
 684         return dentry;
 685 }
 686
 687 static int __choose_mds(struct ceph_mds_client *mdsc,
 688                         struct ceph_mds_request *req)
 689 {
 690         struct inode *inode;
 691         struct ceph_inode_info *ci;
 692         struct ceph_cap *cap;
 693         int mode = req->r_direct_mode;
 694         int mds = -1;
 695         u32 hash = req->r_direct_hash;
 696         bool is_hash = req->r_direct_is_hash;
 697
 698         /*
 699          * is there a specific mds we should try?  ignore hint if we have
 700          * no session and the mds is not up (active or recovering).
 701          */
 702         if (req->r_resend_mds >= 0 &&
 703             (__have_session(mdsc, req->r_resend_mds) ||
 704              ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
 705                 dout("choose_mds using resend_mds mds%d\n",
 706                      req->r_resend_mds);
 707                 return req->r_resend_mds;
 708         }
 709
 710         if (mode == USE_RANDOM_MDS)
 711                 goto random;
 712
 713         inode = NULL;
 714         if (req->r_inode) {
 715                 inode = req->r_inode;
 716         } else if (req->r_dentry) {
 717                 /* ignore race with rename; old or new d_parent is okay */
 718                 struct dentry *parent = req->r_dentry->d_parent;
 719                 struct inode *dir = d_inode(parent);
 720
 721                 if (dir->i_sb != mdsc->fsc->sb) {
 722                         /* not this fs! */
 723                         inode = d_inode(req->r_dentry);
 724                 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
 725                         /* direct snapped/virtual snapdir requests
 726                          * based on parent dir inode */
 727                         struct dentry *dn = get_nonsnap_parent(parent);
 728                         inode = d_inode(dn);
 729                         dout("__choose_mds using nonsnap parent %p\n", inode);
 730                 } else {
 731                         /* dentry target */
 732                         inode = d_inode(req->r_dentry);
 733                         if (!inode || mode == USE_AUTH_MDS) {
 734                                 /* dir + name */
 735                                 inode = dir;
 736                                 hash = ceph_dentry_hash(dir, req->r_dentry);
 737                                 is_hash = true;
 738                         }
 739                 }
 740         }
 741
 742         dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
 743              (int)hash, mode);
 744         if (!inode)
 745                 goto random;
 746         ci = ceph_inode(inode);
 747
 748         if (is_hash && S_ISDIR(inode->i_mode)) {
 749                 struct ceph_inode_frag frag;
 750                 int found;
 751
 752                 ceph_choose_frag(ci, hash, &frag, &found);
 753                 if (found) {
 754                         if (mode == USE_ANY_MDS && frag.ndist > 0) {
 755                                 u8 r;
 756
 757                                 /* choose a random replica */
 758                                 get_random_bytes(&r, 1);
 759                                 r %= frag.ndist;
 760                                 mds = frag.dist[r];
 761                                 dout("choose_mds %p %llx.%llx "
 762                                      "frag %u mds%d (%d/%d)\n",
 763                                      inode, ceph_vinop(inode),
 764                                      frag.frag, mds,
 765                                      (int)r, frag.ndist);
 766                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
 767                                     CEPH_MDS_STATE_ACTIVE)
 768                                         return mds;
 769                         }
 770
 771                         /* since this file/dir wasn't known to be
 772                          * replicated, then we want to look for the
 773                          * authoritative mds. */
 774                         mode = USE_AUTH_MDS;
 775                         if (frag.mds >= 0) {
 776                                 /* choose auth mds */
 777                                 mds = frag.mds;
 778                                 dout("choose_mds %p %llx.%llx "
 779                                      "frag %u mds%d (auth)\n",
 780                                      inode, ceph_vinop(inode), frag.frag, mds);
 781                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
 782                                     CEPH_MDS_STATE_ACTIVE)
 783                                         return mds;
 784                         }
 785                 }
 786         }
 787
 788         spin_lock(&ci->i_ceph_lock);
 789         cap = NULL;
 790         if (mode == USE_AUTH_MDS)
 791                 cap = ci->i_auth_cap;
 792         if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
 793                 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
 794         if (!cap) {
 795                 spin_unlock(&ci->i_ceph_lock);
 796                 goto random;
 797         }
 798         mds = cap->session->s_mds;
 799         dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
 800              inode, ceph_vinop(inode), mds,
 801              cap == ci->i_auth_cap ? "auth " : "", cap);
 802         spin_unlock(&ci->i_ceph_lock);
 803         return mds;
 804
 805 random:
 806         mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
 807         dout("choose_mds chose random mds%d\n", mds);
 808         return mds;
 809 }
 810
 811
 812 /*
 813  * session messages
 814  */
 815 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
 816 {
 817         struct ceph_msg *msg;
 818         struct ceph_mds_session_head *h;
 819
 820         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
 821                            false);
 822         if (!msg) {
 823                 pr_err("create_session_msg ENOMEM creating msg\n");
 824                 return NULL;
 825         }
 826         h = msg->front.iov_base;
 827         h->op = cpu_to_le32(op);
 828         h->seq = cpu_to_le64(seq);
 829
 830         return msg;
 831 }
 832
 833 /*
 834  * session message, specialization for CEPH_SESSION_REQUEST_OPEN
 835  * to include additional client metadata fields.
 836  */
 837 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
 838 {
 839         struct ceph_msg *msg;
 840         struct ceph_mds_session_head *h;
 841         int i = -1;
 842         int metadata_bytes = 0;
 843         int metadata_key_count = 0;
 844         struct ceph_options *opt = mdsc->fsc->client->options;
 845         struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
 846         void *p;
 847
 848         const char* metadata[][2] = {
 849                 {"hostname", utsname()->nodename},
 850                 {"kernel_version", utsname()->release},
 851                 {"entity_id", opt->name ? : ""},
 852                 {"root", fsopt->server_path ? : "/"},
 853                 {NULL, NULL}
 854         };
 855
 856         /* Calculate serialized length of metadata */
 857         metadata_bytes = 4;  /* map length */
 858         for (i = 0; metadata[i][0] != NULL; ++i) {
 859                 metadata_bytes += 8 + strlen(metadata[i][0]) +
 860                         strlen(metadata[i][1]);
 861                 metadata_key_count++;
 862         }
 863
 864         /* Allocate the message */
 865         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
 866                            GFP_NOFS, false);
 867         if (!msg) {
 868                 pr_err("create_session_msg ENOMEM creating msg\n");
 869                 return NULL;
 870         }
 871         h = msg->front.iov_base;
 872         h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
 873         h->seq = cpu_to_le64(seq);
 874
 875         /*
 876          * Serialize client metadata into waiting buffer space, using
 877          * the format that userspace expects for map<string, string>
 878          *
 879          * ClientSession messages with metadata are v2
 880          */
 881         msg->hdr.version = cpu_to_le16(2);
 882         msg->hdr.compat_version = cpu_to_le16(1);
 883
 884         /* The write pointer, following the session_head structure */
 885         p = msg->front.iov_base + sizeof(*h);
 886
 887         /* Number of entries in the map */
 888         ceph_encode_32(&p, metadata_key_count);
 889
 890         /* Two length-prefixed strings for each entry in the map */
 891         for (i = 0; metadata[i][0] != NULL; ++i) {
 892                 size_t const key_len = strlen(metadata[i][0]);
 893                 size_t const val_len = strlen(metadata[i][1]);
 894
 895                 ceph_encode_32(&p, key_len);
 896                 memcpy(p, metadata[i][0], key_len);
 897                 p += key_len;
 898                 ceph_encode_32(&p, val_len);
 899                 memcpy(p, metadata[i][1], val_len);
 900                 p += val_len;
 901         }
 902
 903         return msg;
 904 }
 905
 906 /*
 907  * send session open request.
 908  *
 909  * called under mdsc->mutex
 910  */
 911 static int __open_session(struct ceph_mds_client *mdsc,
 912                           struct ceph_mds_session *session)
 913 {
 914         struct ceph_msg *msg;
 915         int mstate;
 916         int mds = session->s_mds;
 917
 918         /* wait for mds to go active? */
 919         mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
 920         dout("open_session to mds%d (%s)\n", mds,
 921              ceph_mds_state_name(mstate));
 922         session->s_state = CEPH_MDS_SESSION_OPENING;
 923         session->s_renew_requested = jiffies;
 924
 925         /* send connect message */
 926         msg = create_session_open_msg(mdsc, session->s_seq);
 927         if (!msg)
 928                 return -ENOMEM;
 929         ceph_con_send(&session->s_con, msg);
 930         return 0;
 931 }
 932
 933 /*
 934  * open sessions for any export targets for the given mds
 935  *
 936  * called under mdsc->mutex
 937  */
 938 static struct ceph_mds_session *
 939 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
 940 {
 941         struct ceph_mds_session *session;
 942
 943         session = __ceph_lookup_mds_session(mdsc, target);
 944         if (!session) {
 945                 session = register_session(mdsc, target);
 946                 if (IS_ERR(session))
 947                         return session;
 948         }
 949         if (session->s_state == CEPH_MDS_SESSION_NEW ||
 950             session->s_state == CEPH_MDS_SESSION_CLOSING)
 951                 __open_session(mdsc, session);
 952
 953         return session;
 954 }
 955
 956 struct ceph_mds_session *
 957 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
 958 {
 959         struct ceph_mds_session *session;
 960
 961         dout("open_export_target_session to mds%d\n", target);
 962
 963         mutex_lock(&mdsc->mutex);
 964         session = __open_export_target_session(mdsc, target);
 965         mutex_unlock(&mdsc->mutex);
 966
 967         return session;
 968 }
 969
 970 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
 971                                           struct ceph_mds_session *session)
 972 {
 973         struct ceph_mds_info *mi;
 974         struct ceph_mds_session *ts;
 975         int i, mds = session->s_mds;
 976
 977         if (mds >= mdsc->mdsmap->m_max_mds)
 978                 return;
 979
 980         mi = &mdsc->mdsmap->m_info[mds];
 981         dout("open_export_target_sessions for mds%d (%d targets)\n",
 982              session->s_mds, mi->num_export_targets);
 983
 984         for (i = 0; i < mi->num_export_targets; i++) {
 985                 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
 986                 if (!IS_ERR(ts))
 987                         ceph_put_mds_session(ts);
 988         }
 989 }
 990
 991 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 992                                            struct ceph_mds_session *session)
 993 {
 994         mutex_lock(&mdsc->mutex);
 995         __open_export_target_sessions(mdsc, session);
 996         mutex_unlock(&mdsc->mutex);
 997 }
 998
 999 /*
1000  * session caps
1001  */
1002
1003 /* caller holds s_cap_lock, we drop it */
1004 static void cleanup_cap_releases(struct ceph_mds_client *mdsc,
1005                                  struct ceph_mds_session *session)
1006         __releases(session->s_cap_lock)
1007 {
1008         LIST_HEAD(tmp_list);
1009         list_splice_init(&session->s_cap_releases, &tmp_list);
1010         session->s_num_cap_releases = 0;
1011         spin_unlock(&session->s_cap_lock);
1012
1013         dout("cleanup_cap_releases mds%d\n", session->s_mds);
1014         while (!list_empty(&tmp_list)) {
1015                 struct ceph_cap *cap;
1016                 /* zero out the in-progress message */
1017                 cap = list_first_entry(&tmp_list,
1018                                         struct ceph_cap, session_caps);
1019                 list_del(&cap->session_caps);
1020                 ceph_put_cap(mdsc, cap);
1021         }
1022 }
1023
1024 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1025                                      struct ceph_mds_session *session)
1026 {
1027         struct ceph_mds_request *req;
1028         struct rb_node *p;
1029
1030         dout("cleanup_session_requests mds%d\n", session->s_mds);
1031         mutex_lock(&mdsc->mutex);
1032         while (!list_empty(&session->s_unsafe)) {
1033                 req = list_first_entry(&session->s_unsafe,
1034                                        struct ceph_mds_request, r_unsafe_item);
1035                 list_del_init(&req->r_unsafe_item);
1036                 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1037                                     req->r_tid);
1038                 __unregister_request(mdsc, req);
1039         }
1040         /* zero r_attempts, so kick_requests() will re-send requests */
1041         p = rb_first(&mdsc->request_tree);
1042         while (p) {
1043                 req = rb_entry(p, struct ceph_mds_request, r_node);
1044                 p = rb_next(p);
1045                 if (req->r_session &&
1046                     req->r_session->s_mds == session->s_mds)
1047                         req->r_attempts = 0;
1048         }
1049         mutex_unlock(&mdsc->mutex);
1050 }
1051
1052 /*
1053  * Helper to safely iterate over all caps associated with a session, with
1054  * special care taken to handle a racing __ceph_remove_cap().
1055  *
1056  * Caller must hold session s_mutex.
1057  */
1058 static int iterate_session_caps(struct ceph_mds_session *session,
1059                                  int (*cb)(struct inode *, struct ceph_cap *,
1060                                             void *), void *arg)
1061 {
1062         struct list_head *p;
1063         struct ceph_cap *cap;
1064         struct inode *inode, *last_inode = NULL;
1065         struct ceph_cap *old_cap = NULL;
1066         int ret;
1067
1068         dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1069         spin_lock(&session->s_cap_lock);
1070         p = session->s_caps.next;
1071         while (p != &session->s_caps) {
1072                 cap = list_entry(p, struct ceph_cap, session_caps);
1073                 inode = igrab(&cap->ci->vfs_inode);
1074                 if (!inode) {
1075                         p = p->next;
1076                         continue;
1077                 }
1078                 session->s_cap_iterator = cap;
1079                 spin_unlock(&session->s_cap_lock);
1080
1081                 if (last_inode) {
1082                         iput(last_inode);
1083                         last_inode = NULL;
1084                 }
1085                 if (old_cap) {
1086                         ceph_put_cap(session->s_mdsc, old_cap);
1087                         old_cap = NULL;
1088                 }
1089
1090                 ret = cb(inode, cap, arg);
1091                 last_inode = inode;
1092
1093                 spin_lock(&session->s_cap_lock);
1094                 p = p->next;
1095                 if (cap->ci == NULL) {
1096                         dout("iterate_session_caps  finishing cap %p removal\n",
1097                              cap);
1098                         BUG_ON(cap->session != session);
1099                         cap->session = NULL;
1100                         list_del_init(&cap->session_caps);
1101                         session->s_nr_caps--;
1102                         if (cap->queue_release) {
1103                                 list_add_tail(&cap->session_caps,
1104                                               &session->s_cap_releases);
1105                                 session->s_num_cap_releases++;
1106                         } else {
1107                                 old_cap = cap;  /* put_cap it w/o locks held */
1108                         }
1109                 }
1110                 if (ret < 0)
1111                         goto out;
1112         }
1113         ret = 0;
1114 out:
1115         session->s_cap_iterator = NULL;
1116         spin_unlock(&session->s_cap_lock);
1117
1118         iput(last_inode);
1119         if (old_cap)
1120                 ceph_put_cap(session->s_mdsc, old_cap);
1121
1122         return ret;
1123 }
1124
1125 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1126                                   void *arg)
1127 {
1128         struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
1129         struct ceph_inode_info *ci = ceph_inode(inode);
1130         LIST_HEAD(to_remove);
1131         bool drop = false;
1132         bool invalidate = false;
1133
1134         dout("removing cap %p, ci is %p, inode is %p\n",
1135              cap, ci, &ci->vfs_inode);
1136         spin_lock(&ci->i_ceph_lock);
1137         __ceph_remove_cap(cap, false);
1138         if (!ci->i_auth_cap) {
1139                 struct ceph_cap_flush *cf;
1140                 struct ceph_mds_client *mdsc = fsc->mdsc;
1141
1142                 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1143
1144                 if (ci->i_wrbuffer_ref > 0 &&
1145                     ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
1146                         invalidate = true;
1147
1148                 while (true) {
1149                         struct rb_node *n = rb_first(&ci->i_cap_flush_tree);
1150                         if (!n)
1151                                 break;
1152                         cf = rb_entry(n, struct ceph_cap_flush, i_node);
1153                         rb_erase(&cf->i_node, &ci->i_cap_flush_tree);
1154                         list_add(&cf->list, &to_remove);
1155                 }
1156
1157                 spin_lock(&mdsc->cap_dirty_lock);
1158
1159                 list_for_each_entry(cf, &to_remove, list)
1160                         rb_erase(&cf->g_node, &mdsc->cap_flush_tree);
1161
1162                 if (!list_empty(&ci->i_dirty_item)) {
1163                         pr_warn_ratelimited(
1164                                 " dropping dirty %s state for %p %lld\n",
1165                                 ceph_cap_string(ci->i_dirty_caps),
1166                                 inode, ceph_ino(inode));
1167                         ci->i_dirty_caps = 0;
1168                         list_del_init(&ci->i_dirty_item);
1169                         drop = true;
1170                 }
1171                 if (!list_empty(&ci->i_flushing_item)) {
1172                         pr_warn_ratelimited(
1173                                 " dropping dirty+flushing %s state for %p %lld\n",
1174                                 ceph_cap_string(ci->i_flushing_caps),
1175                                 inode, ceph_ino(inode));
1176                         ci->i_flushing_caps = 0;
1177                         list_del_init(&ci->i_flushing_item);
1178                         mdsc->num_cap_flushing--;
1179                         drop = true;
1180                 }
1181                 spin_unlock(&mdsc->cap_dirty_lock);
1182
1183                 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1184                         list_add(&ci->i_prealloc_cap_flush->list, &to_remove);
1185                         ci->i_prealloc_cap_flush = NULL;
1186                 }
1187         }
1188         spin_unlock(&ci->i_ceph_lock);
1189         while (!list_empty(&to_remove)) {
1190                 struct ceph_cap_flush *cf;
1191                 cf = list_first_entry(&to_remove,
1192                                       struct ceph_cap_flush, list);
1193                 list_del(&cf->list);
1194                 ceph_free_cap_flush(cf);
1195         }
1196
1197         wake_up_all(&ci->i_cap_wq);
1198         if (invalidate)
1199                 ceph_queue_invalidate(inode);
1200         if (drop)
1201                 iput(inode);
1202         return 0;
1203 }
1204
1205 /*
1206  * caller must hold session s_mutex
1207  */
1208 static void remove_session_caps(struct ceph_mds_session *session)
1209 {
1210         struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1211         struct super_block *sb = fsc->sb;
1212         dout("remove_session_caps on %p\n", session);
1213         iterate_session_caps(session, remove_session_caps_cb, fsc);
1214
1215         spin_lock(&session->s_cap_lock);
1216         if (session->s_nr_caps > 0) {
1217                 struct inode *inode;
1218                 struct ceph_cap *cap, *prev = NULL;
1219                 struct ceph_vino vino;
1220                 /*
1221                  * iterate_session_caps() skips inodes that are being
1222                  * deleted, we need to wait until deletions are complete.
1223                  * __wait_on_freeing_inode() is designed for the job,
1224                  * but it is not exported, so use lookup inode function
1225                  * to access it.
1226                  */
1227                 while (!list_empty(&session->s_caps)) {
1228                         cap = list_entry(session->s_caps.next,
1229                                          struct ceph_cap, session_caps);
1230                         if (cap == prev)
1231                                 break;
1232                         prev = cap;
1233                         vino = cap->ci->i_vino;
1234                         spin_unlock(&session->s_cap_lock);
1235
1236                         inode = ceph_find_inode(sb, vino);
1237                         iput(inode);
1238
1239                         spin_lock(&session->s_cap_lock);
1240                 }
1241         }
1242
1243         // drop cap expires and unlock s_cap_lock
1244         cleanup_cap_releases(session->s_mdsc, session);
1245
1246         BUG_ON(session->s_nr_caps > 0);
1247         BUG_ON(!list_empty(&session->s_cap_flushing));
1248 }
1249
1250 /*
1251  * wake up any threads waiting on this session's caps.  if the cap is
1252  * old (didn't get renewed on the client reconnect), remove it now.
1253  *
1254  * caller must hold s_mutex.
1255  */
1256 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1257                               void *arg)
1258 {
1259         struct ceph_inode_info *ci = ceph_inode(inode);
1260
1261         if (arg) {
1262                 spin_lock(&ci->i_ceph_lock);
1263                 ci->i_wanted_max_size = 0;
1264                 ci->i_requested_max_size = 0;
1265                 spin_unlock(&ci->i_ceph_lock);
1266         }
1267         wake_up_all(&ci->i_cap_wq);
1268         return 0;
1269 }
1270
1271 static void wake_up_session_caps(struct ceph_mds_session *session,
1272                                  int reconnect)
1273 {
1274         dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1275         iterate_session_caps(session, wake_up_session_cb,
1276                              (void *)(unsigned long)reconnect);
1277 }
1278
1279 /*
1280  * Send periodic message to MDS renewing all currently held caps.  The
1281  * ack will reset the expiration for all caps from this session.
1282  *
1283  * caller holds s_mutex
1284  */
1285 static int send_renew_caps(struct ceph_mds_client *mdsc,
1286                            struct ceph_mds_session *session)
1287 {
1288         struct ceph_msg *msg;
1289         int state;
1290
1291         if (time_after_eq(jiffies, session->s_cap_ttl) &&
1292             time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1293                 pr_info("mds%d caps stale\n", session->s_mds);
1294         session->s_renew_requested = jiffies;
1295
1296         /* do not try to renew caps until a recovering mds has reconnected
1297          * with its clients. */
1298         state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1299         if (state < CEPH_MDS_STATE_RECONNECT) {
1300                 dout("send_renew_caps ignoring mds%d (%s)\n",
1301                      session->s_mds, ceph_mds_state_name(state));
1302                 return 0;
1303         }
1304
1305         dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1306                 ceph_mds_state_name(state));
1307         msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1308                                  ++session->s_renew_seq);
1309         if (!msg)
1310                 return -ENOMEM;
1311         ceph_con_send(&session->s_con, msg);
1312         return 0;
1313 }
1314
1315 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1316                              struct ceph_mds_session *session, u64 seq)
1317 {
1318         struct ceph_msg *msg;
1319
1320         dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1321              session->s_mds, ceph_session_state_name(session->s_state), seq);
1322         msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1323         if (!msg)
1324                 return -ENOMEM;
1325         ceph_con_send(&session->s_con, msg);
1326         return 0;
1327 }
1328
1329
1330 /*
1331  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1332  *
1333  * Called under session->s_mutex
1334  */
1335 static void renewed_caps(struct ceph_mds_client *mdsc,
1336                          struct ceph_mds_session *session, int is_renew)
1337 {
1338         int was_stale;
1339         int wake = 0;
1340
1341         spin_lock(&session->s_cap_lock);
1342         was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1343
1344         session->s_cap_ttl = session->s_renew_requested +
1345                 mdsc->mdsmap->m_session_timeout*HZ;
1346
1347         if (was_stale) {
1348                 if (time_before(jiffies, session->s_cap_ttl)) {
1349                         pr_info("mds%d caps renewed\n", session->s_mds);
1350                         wake = 1;
1351                 } else {
1352                         pr_info("mds%d caps still stale\n", session->s_mds);
1353                 }
1354         }
1355         dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1356              session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1357              time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1358         spin_unlock(&session->s_cap_lock);
1359
1360         if (wake)
1361                 wake_up_session_caps(session, 0);
1362 }
1363
1364 /*
1365  * send a session close request
1366  */
1367 static int request_close_session(struct ceph_mds_client *mdsc,
1368                                  struct ceph_mds_session *session)
1369 {
1370         struct ceph_msg *msg;
1371
1372         dout("request_close_session mds%d state %s seq %lld\n",
1373              session->s_mds, ceph_session_state_name(session->s_state),
1374              session->s_seq);
1375         msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1376         if (!msg)
1377                 return -ENOMEM;
1378         ceph_con_send(&session->s_con, msg);
1379         return 0;
1380 }
1381
1382 /*
1383  * Called with s_mutex held.
1384  */
1385 static int __close_session(struct ceph_mds_client *mdsc,
1386                          struct ceph_mds_session *session)
1387 {
1388         if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1389                 return 0;
1390         session->s_state = CEPH_MDS_SESSION_CLOSING;
1391         return request_close_session(mdsc, session);
1392 }
1393
1394 /*
1395  * Trim old(er) caps.
1396  *
1397  * Because we can't cache an inode without one or more caps, we do
1398  * this indirectly: if a cap is unused, we prune its aliases, at which
1399  * point the inode will hopefully get dropped to.
1400  *
1401  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1402  * memory pressure from the MDS, though, so it needn't be perfect.
1403  */
1404 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1405 {
1406         struct ceph_mds_session *session = arg;
1407         struct ceph_inode_info *ci = ceph_inode(inode);
1408         int used, wanted, oissued, mine;
1409
1410         if (session->s_trim_caps <= 0)
1411                 return -1;
1412
1413         spin_lock(&ci->i_ceph_lock);
1414         mine = cap->issued | cap->implemented;
1415         used = __ceph_caps_used(ci);
1416         wanted = __ceph_caps_file_wanted(ci);
1417         oissued = __ceph_caps_issued_other(ci, cap);
1418
1419         dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1420              inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1421              ceph_cap_string(used), ceph_cap_string(wanted));
1422         if (cap == ci->i_auth_cap) {
1423                 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1424                     !list_empty(&ci->i_cap_snaps))
1425                         goto out;
1426                 if ((used | wanted) & CEPH_CAP_ANY_WR)
1427                         goto out;
1428         }
1429         /* The inode has cached pages, but it's no longer used.
1430          * we can safely drop it */
1431         if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1432             !(oissued & CEPH_CAP_FILE_CACHE)) {
1433           used = 0;
1434           oissued = 0;
1435         }
1436         if ((used | wanted) & ~oissued & mine)
1437                 goto out;   /* we need these caps */
1438
1439         session->s_trim_caps--;
1440         if (oissued) {
1441                 /* we aren't the only cap.. just remove us */
1442                 __ceph_remove_cap(cap, true);
1443         } else {
1444                 /* try dropping referring dentries */
1445                 spin_unlock(&ci->i_ceph_lock);
1446                 d_prune_aliases(inode);
1447                 dout("trim_caps_cb %p cap %p  pruned, count now %d\n",
1448                      inode, cap, atomic_read(&inode->i_count));
1449                 return 0;
1450         }
1451
1452 out:
1453         spin_unlock(&ci->i_ceph_lock);
1454         return 0;
1455 }
1456
1457 /*
1458  * Trim session cap count down to some max number.
1459  */
1460 static int trim_caps(struct ceph_mds_client *mdsc,
1461                      struct ceph_mds_session *session,
1462                      int max_caps)
1463 {
1464         int trim_caps = session->s_nr_caps - max_caps;
1465
1466         dout("trim_caps mds%d start: %d / %d, trim %d\n",
1467              session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1468         if (trim_caps > 0) {
1469                 session->s_trim_caps = trim_caps;
1470                 iterate_session_caps(session, trim_caps_cb, session);
1471                 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1472                      session->s_mds, session->s_nr_caps, max_caps,
1473                         trim_caps - session->s_trim_caps);
1474                 session->s_trim_caps = 0;
1475         }
1476
1477         ceph_send_cap_releases(mdsc, session);
1478         return 0;
1479 }
1480
1481 static int check_capsnap_flush(struct ceph_inode_info *ci,
1482                                u64 want_snap_seq)
1483 {
1484         int ret = 1;
1485         spin_lock(&ci->i_ceph_lock);
1486         if (want_snap_seq > 0 && !list_empty(&ci->i_cap_snaps)) {
1487                 struct ceph_cap_snap *capsnap =
1488                         list_first_entry(&ci->i_cap_snaps,
1489                                          struct ceph_cap_snap, ci_item);
1490                 ret = capsnap->follows >= want_snap_seq;
1491         }
1492         spin_unlock(&ci->i_ceph_lock);
1493         return ret;
1494 }
1495
1496 static int check_caps_flush(struct ceph_mds_client *mdsc,
1497                             u64 want_flush_tid)
1498 {
1499         struct rb_node *n;
1500         struct ceph_cap_flush *cf;
1501         int ret = 1;
1502
1503         spin_lock(&mdsc->cap_dirty_lock);
1504         n = rb_first(&mdsc->cap_flush_tree);
1505         cf = n ? rb_entry(n, struct ceph_cap_flush, g_node) : NULL;
1506         if (cf && cf->tid <= want_flush_tid) {
1507                 dout("check_caps_flush still flushing tid %llu <= %llu\n",
1508                      cf->tid, want_flush_tid);
1509                 ret = 0;
1510         }
1511         spin_unlock(&mdsc->cap_dirty_lock);
1512         return ret;
1513 }
1514
1515 /*
1516  * flush all dirty inode data to disk.
1517  *
1518  * returns true if we've flushed through want_flush_tid
1519  */
1520 static void wait_caps_flush(struct ceph_mds_client *mdsc,
1521                             u64 want_flush_tid, u64 want_snap_seq)
1522 {
1523         int mds;
1524
1525         dout("check_caps_flush want %llu snap want %llu\n",
1526              want_flush_tid, want_snap_seq);
1527         mutex_lock(&mdsc->mutex);
1528         for (mds = 0; mds < mdsc->max_sessions; ) {
1529                 struct ceph_mds_session *session = mdsc->sessions[mds];
1530                 struct inode *inode = NULL;
1531
1532                 if (!session) {
1533                         mds++;
1534                         continue;
1535                 }
1536                 get_session(session);
1537                 mutex_unlock(&mdsc->mutex);
1538
1539                 mutex_lock(&session->s_mutex);
1540                 if (!list_empty(&session->s_cap_snaps_flushing)) {
1541                         struct ceph_cap_snap *capsnap =
1542                                 list_first_entry(&session->s_cap_snaps_flushing,
1543                                                  struct ceph_cap_snap,
1544                                                  flushing_item);
1545                         struct ceph_inode_info *ci = capsnap->ci;
1546                         if (!check_capsnap_flush(ci, want_snap_seq)) {
1547                                 dout("check_cap_flush still flushing snap %p "
1548                                      "follows %lld <= %lld to mds%d\n",
1549                                      &ci->vfs_inode, capsnap->follows,
1550                                      want_snap_seq, mds);
1551                                 inode = igrab(&ci->vfs_inode);
1552                         }
1553                 }
1554                 mutex_unlock(&session->s_mutex);
1555                 ceph_put_mds_session(session);
1556
1557                 if (inode) {
1558                         wait_event(mdsc->cap_flushing_wq,
1559                                    check_capsnap_flush(ceph_inode(inode),
1560                                                        want_snap_seq));
1561                         iput(inode);
1562                 } else {
1563                         mds++;
1564                 }
1565
1566                 mutex_lock(&mdsc->mutex);
1567         }
1568         mutex_unlock(&mdsc->mutex);
1569
1570         wait_event(mdsc->cap_flushing_wq,
1571                    check_caps_flush(mdsc, want_flush_tid));
1572
1573         dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
1574 }
1575
1576 /*
1577  * called under s_mutex
1578  */
1579 void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1580                             struct ceph_mds_session *session)
1581 {
1582         struct ceph_msg *msg = NULL;
1583         struct ceph_mds_cap_release *head;
1584         struct ceph_mds_cap_item *item;
1585         struct ceph_cap *cap;
1586         LIST_HEAD(tmp_list);
1587         int num_cap_releases;
1588
1589         spin_lock(&session->s_cap_lock);
1590 again:
1591         list_splice_init(&session->s_cap_releases, &tmp_list);
1592         num_cap_releases = session->s_num_cap_releases;
1593         session->s_num_cap_releases = 0;
1594         spin_unlock(&session->s_cap_lock);
1595
1596         while (!list_empty(&tmp_list)) {
1597                 if (!msg) {
1598                         msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
1599                                         PAGE_SIZE, GFP_NOFS, false);
1600                         if (!msg)
1601                                 goto out_err;
1602                         head = msg->front.iov_base;
1603                         head->num = cpu_to_le32(0);
1604                         msg->front.iov_len = sizeof(*head);
1605                 }
1606                 cap = list_first_entry(&tmp_list, struct ceph_cap,
1607                                         session_caps);
1608                 list_del(&cap->session_caps);
1609                 num_cap_releases--;
1610
1611                 head = msg->front.iov_base;
1612                 le32_add_cpu(&head->num, 1);
1613                 item = msg->front.iov_base + msg->front.iov_len;
1614                 item->ino = cpu_to_le64(cap->cap_ino);
1615                 item->cap_id = cpu_to_le64(cap->cap_id);
1616                 item->migrate_seq = cpu_to_le32(cap->mseq);
1617                 item->seq = cpu_to_le32(cap->issue_seq);
1618                 msg->front.iov_len += sizeof(*item);
1619
1620                 ceph_put_cap(mdsc, cap);
1621
1622                 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1623                         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1624                         dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1625                         ceph_con_send(&session->s_con, msg);
1626                         msg = NULL;
1627                 }
1628         }
1629
1630         BUG_ON(num_cap_releases != 0);
1631
1632         spin_lock(&session->s_cap_lock);
1633         if (!list_empty(&session->s_cap_releases))
1634                 goto again;
1635         spin_unlock(&session->s_cap_lock);
1636
1637         if (msg) {
1638                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1639                 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1640                 ceph_con_send(&session->s_con, msg);
1641         }
1642         return;
1643 out_err:
1644         pr_err("send_cap_releases mds%d, failed to allocate message\n",
1645                 session->s_mds);
1646         spin_lock(&session->s_cap_lock);
1647         list_splice(&tmp_list, &session->s_cap_releases);
1648         session->s_num_cap_releases += num_cap_releases;
1649         spin_unlock(&session->s_cap_lock);
1650 }
1651
1652 /*
1653  * requests
1654  */
1655
1656 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1657                                     struct inode *dir)
1658 {
1659         struct ceph_inode_info *ci = ceph_inode(dir);
1660         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1661         struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
1662         size_t size = sizeof(struct ceph_mds_reply_dir_entry);
1663         int order, num_entries;
1664
1665         spin_lock(&ci->i_ceph_lock);
1666         num_entries = ci->i_files + ci->i_subdirs;
1667         spin_unlock(&ci->i_ceph_lock);
1668         num_entries = max(num_entries, 1);
1669         num_entries = min(num_entries, opt->max_readdir);
1670
1671         order = get_order(size * num_entries);
1672         while (order >= 0) {
1673                 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
1674                                                              __GFP_NOWARN,
1675                                                              order);
1676                 if (rinfo->dir_entries)
1677                         break;
1678                 order--;
1679         }
1680         if (!rinfo->dir_entries)
1681                 return -ENOMEM;
1682
1683         num_entries = (PAGE_SIZE << order) / size;
1684         num_entries = min(num_entries, opt->max_readdir);
1685
1686         rinfo->dir_buf_size = PAGE_SIZE << order;
1687         req->r_num_caps = num_entries + 1;
1688         req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1689         req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1690         return 0;
1691 }
1692
1693 /*
1694  * Create an mds request.
1695  */
1696 struct ceph_mds_request *
1697 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1698 {
1699         struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1700
1701         if (!req)
1702                 return ERR_PTR(-ENOMEM);
1703
1704         mutex_init(&req->r_fill_mutex);
1705         req->r_mdsc = mdsc;
1706         req->r_started = jiffies;
1707         req->r_resend_mds = -1;
1708         INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1709         INIT_LIST_HEAD(&req->r_unsafe_target_item);
1710         req->r_fmode = -1;
1711         kref_init(&req->r_kref);
1712         RB_CLEAR_NODE(&req->r_node);
1713         INIT_LIST_HEAD(&req->r_wait);
1714         init_completion(&req->r_completion);
1715         init_completion(&req->r_safe_completion);
1716         INIT_LIST_HEAD(&req->r_unsafe_item);
1717
1718         req->r_stamp = current_fs_time(mdsc->fsc->sb);
1719
1720         req->r_op = op;
1721         req->r_direct_mode = mode;
1722         return req;
1723 }
1724
1725 /*
1726  * return oldest (lowest) request, tid in request tree, 0 if none.
1727  *
1728  * called under mdsc->mutex.
1729  */
1730 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1731 {
1732         if (RB_EMPTY_ROOT(&mdsc->request_tree))
1733                 return NULL;
1734         return rb_entry(rb_first(&mdsc->request_tree),
1735                         struct ceph_mds_request, r_node);
1736 }
1737
1738 static inline  u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1739 {
1740         return mdsc->oldest_tid;
1741 }
1742
1743 /*
1744  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
1745  * on build_path_from_dentry in fs/cifs/dir.c.
1746  *
1747  * If @stop_on_nosnap, generate path relative to the first non-snapped
1748  * inode.
1749  *
1750  * Encode hidden .snap dirs as a double /, i.e.
1751  *   foo/.snap/bar -> foo//bar
1752  */
1753 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1754                            int stop_on_nosnap)
1755 {
1756         struct dentry *temp;
1757         char *path;
1758         int len, pos;
1759         unsigned seq;
1760
1761         if (dentry == NULL)
1762                 return ERR_PTR(-EINVAL);
1763
1764 retry:
1765         len = 0;
1766         seq = read_seqbegin(&rename_lock);
1767         rcu_read_lock();
1768         for (temp = dentry; !IS_ROOT(temp);) {
1769                 struct inode *inode = d_inode(temp);
1770                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1771                         len++;  /* slash only */
1772                 else if (stop_on_nosnap && inode &&
1773                          ceph_snap(inode) == CEPH_NOSNAP)
1774                         break;
1775                 else
1776                         len += 1 + temp->d_name.len;
1777                 temp = temp->d_parent;
1778         }
1779         rcu_read_unlock();
1780         if (len)
1781                 len--;  /* no leading '/' */
1782
1783         path = kmalloc(len+1, GFP_NOFS);
1784         if (path == NULL)
1785                 return ERR_PTR(-ENOMEM);
1786         pos = len;
1787         path[pos] = 0;  /* trailing null */
1788         rcu_read_lock();
1789         for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1790                 struct inode *inode;
1791
1792                 spin_lock(&temp->d_lock);
1793                 inode = d_inode(temp);
1794                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1795                         dout("build_path path+%d: %p SNAPDIR\n",
1796                              pos, temp);
1797                 } else if (stop_on_nosnap && inode &&
1798                            ceph_snap(inode) == CEPH_NOSNAP) {
1799                         spin_unlock(&temp->d_lock);
1800                         break;
1801                 } else {
1802                         pos -= temp->d_name.len;
1803                         if (pos < 0) {
1804                                 spin_unlock(&temp->d_lock);
1805                                 break;
1806                         }
1807                         strncpy(path + pos, temp->d_name.name,
1808                                 temp->d_name.len);
1809                 }
1810                 spin_unlock(&temp->d_lock);
1811                 if (pos)
1812                         path[--pos] = '/';
1813                 temp = temp->d_parent;
1814         }
1815         rcu_read_unlock();
1816         if (pos != 0 || read_seqretry(&rename_lock, seq)) {
1817                 pr_err("build_path did not end path lookup where "
1818                        "expected, namelen is %d, pos is %d\n", len, pos);
1819                 /* presumably this is only possible if racing with a
1820                    rename of one of the parent directories (we can not
1821                    lock the dentries above us to prevent this, but
1822                    retrying should be harmless) */
1823                 kfree(path);
1824                 goto retry;
1825         }
1826
1827         *base = ceph_ino(d_inode(temp));
1828         *plen = len;
1829         dout("build_path on %p %d built %llx '%.*s'\n",
1830              dentry, d_count(dentry), *base, len, path);
1831         return path;
1832 }
1833
1834 static int build_dentry_path(struct dentry *dentry,
1835                              const char **ppath, int *ppathlen, u64 *pino,
1836                              int *pfreepath)
1837 {
1838         char *path;
1839
1840         if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP) {
1841                 *pino = ceph_ino(d_inode(dentry->d_parent));
1842                 *ppath = dentry->d_name.name;
1843                 *ppathlen = dentry->d_name.len;
1844                 return 0;
1845         }
1846         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1847         if (IS_ERR(path))
1848                 return PTR_ERR(path);
1849         *ppath = path;
1850         *pfreepath = 1;
1851         return 0;
1852 }
1853
1854 static int build_inode_path(struct inode *inode,
1855                             const char **ppath, int *ppathlen, u64 *pino,
1856                             int *pfreepath)
1857 {
1858         struct dentry *dentry;
1859         char *path;
1860
1861         if (ceph_snap(inode) == CEPH_NOSNAP) {
1862                 *pino = ceph_ino(inode);
1863                 *ppathlen = 0;
1864                 return 0;
1865         }
1866         dentry = d_find_alias(inode);
1867         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1868         dput(dentry);
1869         if (IS_ERR(path))
1870                 return PTR_ERR(path);
1871         *ppath = path;
1872         *pfreepath = 1;
1873         return 0;
1874 }
1875
1876 /*
1877  * request arguments may be specified via an inode *, a dentry *, or
1878  * an explicit ino+path.
1879  */
1880 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1881                                   const char *rpath, u64 rino,
1882                                   const char **ppath, int *pathlen,
1883                                   u64 *ino, int *freepath)
1884 {
1885         int r = 0;
1886
1887         if (rinode) {
1888                 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1889                 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1890                      ceph_snap(rinode));
1891         } else if (rdentry) {
1892                 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1893                 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1894                      *ppath);
1895         } else if (rpath || rino) {
1896                 *ino = rino;
1897                 *ppath = rpath;
1898                 *pathlen = rpath ? strlen(rpath) : 0;
1899                 dout(" path %.*s\n", *pathlen, rpath);
1900         }
1901
1902         return r;
1903 }
1904
1905 /*
1906  * called under mdsc->mutex
1907  */
1908 static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1909                                                struct ceph_mds_request *req,
1910                                                int mds, bool drop_cap_releases)
1911 {
1912         struct ceph_msg *msg;
1913         struct ceph_mds_request_head *head;
1914         const char *path1 = NULL;
1915         const char *path2 = NULL;
1916         u64 ino1 = 0, ino2 = 0;
1917         int pathlen1 = 0, pathlen2 = 0;
1918         int freepath1 = 0, freepath2 = 0;
1919         int len;
1920         u16 releases;
1921         void *p, *end;
1922         int ret;
1923
1924         ret = set_request_path_attr(req->r_inode, req->r_dentry,
1925                               req->r_path1, req->r_ino1.ino,
1926                               &path1, &pathlen1, &ino1, &freepath1);
1927         if (ret < 0) {
1928                 msg = ERR_PTR(ret);
1929                 goto out;
1930         }
1931
1932         ret = set_request_path_attr(NULL, req->r_old_dentry,
1933                               req->r_path2, req->r_ino2.ino,
1934                               &path2, &pathlen2, &ino2, &freepath2);
1935         if (ret < 0) {
1936                 msg = ERR_PTR(ret);
1937                 goto out_free1;
1938         }
1939
1940         len = sizeof(*head) +
1941                 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
1942                 sizeof(struct ceph_timespec);
1943
1944         /* calculate (max) length for cap releases */
1945         len += sizeof(struct ceph_mds_request_release) *
1946                 (!!req->r_inode_drop + !!req->r_dentry_drop +
1947                  !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1948         if (req->r_dentry_drop)
1949                 len += req->r_dentry->d_name.len;
1950         if (req->r_old_dentry_drop)
1951                 len += req->r_old_dentry->d_name.len;
1952
1953         msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
1954         if (!msg) {
1955                 msg = ERR_PTR(-ENOMEM);
1956                 goto out_free2;
1957         }
1958
1959         msg->hdr.version = cpu_to_le16(2);
1960         msg->hdr.tid = cpu_to_le64(req->r_tid);
1961
1962         head = msg->front.iov_base;
1963         p = msg->front.iov_base + sizeof(*head);
1964         end = msg->front.iov_base + msg->front.iov_len;
1965
1966         head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1967         head->op = cpu_to_le32(req->r_op);
1968         head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1969         head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
1970         head->args = req->r_args;
1971
1972         ceph_encode_filepath(&p, end, ino1, path1);
1973         ceph_encode_filepath(&p, end, ino2, path2);
1974
1975         /* make note of release offset, in case we need to replay */
1976         req->r_request_release_offset = p - msg->front.iov_base;
1977
1978         /* cap releases */
1979         releases = 0;
1980         if (req->r_inode_drop)
1981                 releases += ceph_encode_inode_release(&p,
1982                       req->r_inode ? req->r_inode : d_inode(req->r_dentry),
1983                       mds, req->r_inode_drop, req->r_inode_unless, 0);
1984         if (req->r_dentry_drop)
1985                 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1986                        mds, req->r_dentry_drop, req->r_dentry_unless);
1987         if (req->r_old_dentry_drop)
1988                 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1989                        mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1990         if (req->r_old_inode_drop)
1991                 releases += ceph_encode_inode_release(&p,
1992                       d_inode(req->r_old_dentry),
1993                       mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1994
1995         if (drop_cap_releases) {
1996                 releases = 0;
1997                 p = msg->front.iov_base + req->r_request_release_offset;
1998         }
1999
2000         head->num_releases = cpu_to_le16(releases);
2001
2002         /* time stamp */
2003         {
2004                 struct ceph_timespec ts;
2005                 ceph_encode_timespec(&ts, &req->r_stamp);
2006                 ceph_encode_copy(&p, &ts, sizeof(ts));
2007         }
2008
2009         BUG_ON(p > end);
2010         msg->front.iov_len = p - msg->front.iov_base;
2011         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2012
2013         if (req->r_pagelist) {
2014                 struct ceph_pagelist *pagelist = req->r_pagelist;
2015                 atomic_inc(&pagelist->refcnt);
2016                 ceph_msg_data_add_pagelist(msg, pagelist);
2017                 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2018         } else {
2019                 msg->hdr.data_len = 0;
2020         }
2021
2022         msg->hdr.data_off = cpu_to_le16(0);
2023
2024 out_free2:
2025         if (freepath2)
2026                 kfree((char *)path2);
2027 out_free1:
2028         if (freepath1)
2029                 kfree((char *)path1);
2030 out:
2031         return msg;
2032 }
2033
2034 /*
2035  * called under mdsc->mutex if error, under no mutex if
2036  * success.
2037  */
2038 static void complete_request(struct ceph_mds_client *mdsc,
2039                              struct ceph_mds_request *req)
2040 {
2041         if (req->r_callback)
2042                 req->r_callback(mdsc, req);
2043         else
2044                 complete_all(&req->r_completion);
2045 }
2046
2047 /*
2048  * called under mdsc->mutex
2049  */
2050 static int __prepare_send_request(struct ceph_mds_client *mdsc,
2051                                   struct ceph_mds_request *req,
2052                                   int mds, bool drop_cap_releases)
2053 {
2054         struct ceph_mds_request_head *rhead;
2055         struct ceph_msg *msg;
2056         int flags = 0;
2057
2058         req->r_attempts++;
2059         if (req->r_inode) {
2060                 struct ceph_cap *cap =
2061                         ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2062
2063                 if (cap)
2064                         req->r_sent_on_mseq = cap->mseq;
2065                 else
2066                         req->r_sent_on_mseq = -1;
2067         }
2068         dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2069              req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2070
2071         if (req->r_got_unsafe) {
2072                 void *p;
2073                 /*
2074                  * Replay.  Do not regenerate message (and rebuild
2075                  * paths, etc.); just use the original message.
2076                  * Rebuilding paths will break for renames because
2077                  * d_move mangles the src name.
2078                  */
2079                 msg = req->r_request;
2080                 rhead = msg->front.iov_base;
2081
2082                 flags = le32_to_cpu(rhead->flags);
2083                 flags |= CEPH_MDS_FLAG_REPLAY;
2084                 rhead->flags = cpu_to_le32(flags);
2085
2086                 if (req->r_target_inode)
2087                         rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2088
2089                 rhead->num_retry = req->r_attempts - 1;
2090
2091                 /* remove cap/dentry releases from message */
2092                 rhead->num_releases = 0;
2093
2094                 /* time stamp */
2095                 p = msg->front.iov_base + req->r_request_release_offset;
2096                 {
2097                         struct ceph_timespec ts;
2098                         ceph_encode_timespec(&ts, &req->r_stamp);
2099                         ceph_encode_copy(&p, &ts, sizeof(ts));
2100                 }
2101
2102                 msg->front.iov_len = p - msg->front.iov_base;
2103                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2104                 return 0;
2105         }
2106
2107         if (req->r_request) {
2108                 ceph_msg_put(req->r_request);
2109                 req->r_request = NULL;
2110         }
2111         msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2112         if (IS_ERR(msg)) {
2113                 req->r_err = PTR_ERR(msg);
2114                 return PTR_ERR(msg);
2115         }
2116         req->r_request = msg;
2117
2118         rhead = msg->front.iov_base;
2119         rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2120         if (req->r_got_unsafe)
2121                 flags |= CEPH_MDS_FLAG_REPLAY;
2122         if (req->r_locked_dir)
2123                 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2124         rhead->flags = cpu_to_le32(flags);
2125         rhead->num_fwd = req->r_num_fwd;
2126         rhead->num_retry = req->r_attempts - 1;
2127         rhead->ino = 0;
2128
2129         dout(" r_locked_dir = %p\n", req->r_locked_dir);
2130         return 0;
2131 }
2132
2133 /*
2134  * send request, or put it on the appropriate wait list.
2135  */
2136 static int __do_request(struct ceph_mds_client *mdsc,
2137                         struct ceph_mds_request *req)
2138 {
2139         struct ceph_mds_session *session = NULL;
2140         int mds = -1;
2141         int err = 0;
2142
2143         if (req->r_err || req->r_got_result) {
2144                 if (req->r_aborted)
2145                         __unregister_request(mdsc, req);
2146                 goto out;
2147         }
2148
2149         if (req->r_timeout &&
2150             time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2151                 dout("do_request timed out\n");
2152                 err = -EIO;
2153                 goto finish;
2154         }
2155         if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2156                 dout("do_request forced umount\n");
2157                 err = -EIO;
2158                 goto finish;
2159         }
2160
2161         put_request_session(req);
2162
2163         mds = __choose_mds(mdsc, req);
2164         if (mds < 0 ||
2165             ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2166                 dout("do_request no mds or not active, waiting for map\n");
2167                 list_add(&req->r_wait, &mdsc->waiting_for_map);
2168                 goto out;
2169         }
2170
2171         /* get, open session */
2172         session = __ceph_lookup_mds_session(mdsc, mds);
2173         if (!session) {
2174                 session = register_session(mdsc, mds);
2175                 if (IS_ERR(session)) {
2176                         err = PTR_ERR(session);
2177                         goto finish;
2178                 }
2179         }
2180         req->r_session = get_session(session);
2181
2182         dout("do_request mds%d session %p state %s\n", mds, session,
2183              ceph_session_state_name(session->s_state));
2184         if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2185             session->s_state != CEPH_MDS_SESSION_HUNG) {
2186                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2187                     session->s_state == CEPH_MDS_SESSION_CLOSING)
2188                         __open_session(mdsc, session);
2189                 list_add(&req->r_wait, &session->s_waiting);
2190                 goto out_session;
2191         }
2192
2193         /* send request */
2194         req->r_resend_mds = -1;   /* forget any previous mds hint */
2195
2196         if (req->r_request_started == 0)   /* note request start time */
2197                 req->r_request_started = jiffies;
2198
2199         err = __prepare_send_request(mdsc, req, mds, false);
2200         if (!err) {
2201                 ceph_msg_get(req->r_request);
2202                 ceph_con_send(&session->s_con, req->r_request);
2203         }
2204
2205 out_session:
2206         ceph_put_mds_session(session);
2207 finish:
2208         if (err) {
2209                 dout("__do_request early error %d\n", err);
2210                 req->r_err = err;
2211                 complete_request(mdsc, req);
2212                 __unregister_request(mdsc, req);
2213         }
2214 out:
2215         return err;
2216 }
2217
2218 /*
2219  * called under mdsc->mutex
2220  */
2221 static void __wake_requests(struct ceph_mds_client *mdsc,
2222                             struct list_head *head)
2223 {
2224         struct ceph_mds_request *req;
2225         LIST_HEAD(tmp_list);
2226
2227         list_splice_init(head, &tmp_list);
2228
2229         while (!list_empty(&tmp_list)) {
2230                 req = list_entry(tmp_list.next,
2231                                  struct ceph_mds_request, r_wait);
2232                 list_del_init(&req->r_wait);
2233                 dout(" wake request %p tid %llu\n", req, req->r_tid);
2234                 __do_request(mdsc, req);
2235         }
2236 }
2237
2238 /*
2239  * Wake up threads with requests pending for @mds, so that they can
2240  * resubmit their requests to a possibly different mds.
2241  */
2242 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2243 {
2244         struct ceph_mds_request *req;
2245         struct rb_node *p = rb_first(&mdsc->request_tree);
2246
2247         dout("kick_requests mds%d\n", mds);
2248         while (p) {
2249                 req = rb_entry(p, struct ceph_mds_request, r_node);
2250                 p = rb_next(p);
2251                 if (req->r_got_unsafe)
2252                         continue;
2253                 if (req->r_attempts > 0)
2254                         continue; /* only new requests */
2255                 if (req->r_session &&
2256                     req->r_session->s_mds == mds) {
2257                         dout(" kicking tid %llu\n", req->r_tid);
2258                         list_del_init(&req->r_wait);
2259                         __do_request(mdsc, req);
2260                 }
2261         }
2262 }
2263
2264 void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2265                               struct ceph_mds_request *req)
2266 {
2267         dout("submit_request on %p\n", req);
2268         mutex_lock(&mdsc->mutex);
2269         __register_request(mdsc, req, NULL);
2270         __do_request(mdsc, req);
2271         mutex_unlock(&mdsc->mutex);
2272 }
2273
2274 /*
2275  * Synchrously perform an mds request.  Take care of all of the
2276  * session setup, forwarding, retry details.
2277  */
2278 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2279                          struct inode *dir,
2280                          struct ceph_mds_request *req)
2281 {
2282         int err;
2283
2284         dout("do_request on %p\n", req);
2285
2286         /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2287         if (req->r_inode)
2288                 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2289         if (req->r_locked_dir)
2290                 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
2291         if (req->r_old_dentry_dir)
2292                 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2293                                   CEPH_CAP_PIN);
2294
2295         /* deny access to directories with pool_ns layouts */
2296         if (req->r_inode && S_ISDIR(req->r_inode->i_mode) &&
2297             ceph_inode(req->r_inode)->i_pool_ns_len)
2298                 return -EIO;
2299         if (req->r_locked_dir &&
2300             ceph_inode(req->r_locked_dir)->i_pool_ns_len)
2301                 return -EIO;
2302
2303         /* issue */
2304         mutex_lock(&mdsc->mutex);
2305         __register_request(mdsc, req, dir);
2306         __do_request(mdsc, req);
2307
2308         if (req->r_err) {
2309                 err = req->r_err;
2310                 goto out;
2311         }
2312
2313         /* wait */
2314         mutex_unlock(&mdsc->mutex);
2315         dout("do_request waiting\n");
2316         if (!req->r_timeout && req->r_wait_for_completion) {
2317                 err = req->r_wait_for_completion(mdsc, req);
2318         } else {
2319                 long timeleft = wait_for_completion_killable_timeout(
2320                                         &req->r_completion,
2321                                         ceph_timeout_jiffies(req->r_timeout));
2322                 if (timeleft > 0)
2323                         err = 0;
2324                 else if (!timeleft)
2325                         err = -EIO;  /* timed out */
2326                 else
2327                         err = timeleft;  /* killed */
2328         }
2329         dout("do_request waited, got %d\n", err);
2330         mutex_lock(&mdsc->mutex);
2331
2332         /* only abort if we didn't race with a real reply */
2333         if (req->r_got_result) {
2334                 err = le32_to_cpu(req->r_reply_info.head->result);
2335         } else if (err < 0) {
2336                 dout("aborted request %lld with %d\n", req->r_tid, err);
2337
2338                 /*
2339                  * ensure we aren't running concurrently with
2340                  * ceph_fill_trace or ceph_readdir_prepopulate, which
2341                  * rely on locks (dir mutex) held by our caller.
2342                  */
2343                 mutex_lock(&req->r_fill_mutex);
2344                 req->r_err = err;
2345                 req->r_aborted = true;
2346                 mutex_unlock(&req->r_fill_mutex);
2347
2348                 if (req->r_locked_dir &&
2349                     (req->r_op & CEPH_MDS_OP_WRITE))
2350                         ceph_invalidate_dir_request(req);
2351         } else {
2352                 err = req->r_err;
2353         }
2354
2355 out:
2356         mutex_unlock(&mdsc->mutex);
2357         dout("do_request %p done, result %d\n", req, err);
2358         return err;
2359 }
2360
2361 /*
2362  * Invalidate dir's completeness, dentry lease state on an aborted MDS
2363  * namespace request.
2364  */
2365 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2366 {
2367         struct inode *inode = req->r_locked_dir;
2368
2369         dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
2370
2371         ceph_dir_clear_complete(inode);
2372         if (req->r_dentry)
2373                 ceph_invalidate_dentry_lease(req->r_dentry);
2374         if (req->r_old_dentry)
2375                 ceph_invalidate_dentry_lease(req->r_old_dentry);
2376 }
2377
2378 /*
2379  * Handle mds reply.
2380  *
2381  * We take the session mutex and parse and process the reply immediately.
2382  * This preserves the logical ordering of replies, capabilities, etc., sent
2383  * by the MDS as they are applied to our local cache.
2384  */
2385 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2386 {
2387         struct ceph_mds_client *mdsc = session->s_mdsc;
2388         struct ceph_mds_request *req;
2389         struct ceph_mds_reply_head *head = msg->front.iov_base;
2390         struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
2391         struct ceph_snap_realm *realm;
2392         u64 tid;
2393         int err, result;
2394         int mds = session->s_mds;
2395
2396         if (msg->front.iov_len < sizeof(*head)) {
2397                 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2398                 ceph_msg_dump(msg);
2399                 return;
2400         }
2401
2402         /* get request, session */
2403         tid = le64_to_cpu(msg->hdr.tid);
2404         mutex_lock(&mdsc->mutex);
2405         req = lookup_get_request(mdsc, tid);
2406         if (!req) {
2407                 dout("handle_reply on unknown tid %llu\n", tid);
2408                 mutex_unlock(&mdsc->mutex);
2409                 return;
2410         }
2411         dout("handle_reply %p\n", req);
2412
2413         /* correct session? */
2414         if (req->r_session != session) {
2415                 pr_err("mdsc_handle_reply got %llu on session mds%d"
2416                        " not mds%d\n", tid, session->s_mds,
2417                        req->r_session ? req->r_session->s_mds : -1);
2418                 mutex_unlock(&mdsc->mutex);
2419                 goto out;
2420         }
2421
2422         /* dup? */
2423         if ((req->r_got_unsafe && !head->safe) ||
2424             (req->r_got_safe && head->safe)) {
2425                 pr_warn("got a dup %s reply on %llu from mds%d\n",
2426                            head->safe ? "safe" : "unsafe", tid, mds);
2427                 mutex_unlock(&mdsc->mutex);
2428                 goto out;
2429         }
2430         if (req->r_got_safe) {
2431                 pr_warn("got unsafe after safe on %llu from mds%d\n",
2432                            tid, mds);
2433                 mutex_unlock(&mdsc->mutex);
2434                 goto out;
2435         }
2436
2437         result = le32_to_cpu(head->result);
2438
2439         /*
2440          * Handle an ESTALE
2441          * if we're not talking to the authority, send to them
2442          * if the authority has changed while we weren't looking,
2443          * send to new authority
2444          * Otherwise we just have to return an ESTALE
2445          */
2446         if (result == -ESTALE) {
2447                 dout("got ESTALE on request %llu", req->r_tid);
2448                 req->r_resend_mds = -1;
2449                 if (req->r_direct_mode != USE_AUTH_MDS) {
2450                         dout("not using auth, setting for that now");
2451                         req->r_direct_mode = USE_AUTH_MDS;
2452                         __do_request(mdsc, req);
2453                         mutex_unlock(&mdsc->mutex);
2454                         goto out;
2455                 } else  {
2456                         int mds = __choose_mds(mdsc, req);
2457                         if (mds >= 0 && mds != req->r_session->s_mds) {
2458                                 dout("but auth changed, so resending");
2459                                 __do_request(mdsc, req);
2460                                 mutex_unlock(&mdsc->mutex);
2461                                 goto out;
2462                         }
2463                 }
2464                 dout("have to return ESTALE on request %llu", req->r_tid);
2465         }
2466
2467
2468         if (head->safe) {
2469                 req->r_got_safe = true;
2470                 __unregister_request(mdsc, req);
2471
2472                 if (req->r_got_unsafe) {
2473                         /*
2474                          * We already handled the unsafe response, now do the
2475                          * cleanup.  No need to examine the response; the MDS
2476                          * doesn't include any result info in the safe
2477                          * response.  And even if it did, there is nothing
2478                          * useful we could do with a revised return value.
2479                          */
2480                         dout("got safe reply %llu, mds%d\n", tid, mds);
2481                         list_del_init(&req->r_unsafe_item);
2482
2483                         /* last unsafe request during umount? */
2484                         if (mdsc->stopping && !__get_oldest_req(mdsc))
2485                                 complete_all(&mdsc->safe_umount_waiters);
2486                         mutex_unlock(&mdsc->mutex);
2487                         goto out;
2488                 }
2489         } else {
2490                 req->r_got_unsafe = true;
2491                 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2492                 if (req->r_unsafe_dir) {
2493                         struct ceph_inode_info *ci =
2494                                         ceph_inode(req->r_unsafe_dir);
2495                         spin_lock(&ci->i_unsafe_lock);
2496                         list_add_tail(&req->r_unsafe_dir_item,
2497                                       &ci->i_unsafe_dirops);
2498                         spin_unlock(&ci->i_unsafe_lock);
2499                 }
2500         }
2501
2502         dout("handle_reply tid %lld result %d\n", tid, result);
2503         rinfo = &req->r_reply_info;
2504         err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2505         mutex_unlock(&mdsc->mutex);
2506
2507         mutex_lock(&session->s_mutex);
2508         if (err < 0) {
2509                 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
2510                 ceph_msg_dump(msg);
2511                 goto out_err;
2512         }
2513
2514         /* snap trace */
2515         realm = NULL;
2516         if (rinfo->snapblob_len) {
2517                 down_write(&mdsc->snap_rwsem);
2518                 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2519                                 rinfo->snapblob + rinfo->snapblob_len,
2520                                 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2521                                 &realm);
2522                 downgrade_write(&mdsc->snap_rwsem);
2523         } else {
2524                 down_read(&mdsc->snap_rwsem);
2525         }
2526
2527         /* insert trace into our cache */
2528         mutex_lock(&req->r_fill_mutex);
2529         current->journal_info = req;
2530         err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2531         if (err == 0) {
2532                 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
2533                                     req->r_op == CEPH_MDS_OP_LSSNAP))
2534                         ceph_readdir_prepopulate(req, req->r_session);
2535                 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2536         }
2537         current->journal_info = NULL;
2538         mutex_unlock(&req->r_fill_mutex);
2539
2540         up_read(&mdsc->snap_rwsem);
2541         if (realm)
2542                 ceph_put_snap_realm(mdsc, realm);
2543
2544         if (err == 0 && req->r_got_unsafe && req->r_target_inode) {
2545                 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2546                 spin_lock(&ci->i_unsafe_lock);
2547                 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2548                 spin_unlock(&ci->i_unsafe_lock);
2549         }
2550 out_err:
2551         mutex_lock(&mdsc->mutex);
2552         if (!req->r_aborted) {
2553                 if (err) {
2554                         req->r_err = err;
2555                 } else {
2556                         req->r_reply =  ceph_msg_get(msg);
2557                         req->r_got_result = true;
2558                 }
2559         } else {
2560                 dout("reply arrived after request %lld was aborted\n", tid);
2561         }
2562         mutex_unlock(&mdsc->mutex);
2563
2564         mutex_unlock(&session->s_mutex);
2565
2566         /* kick calling process */
2567         complete_request(mdsc, req);
2568 out:
2569         ceph_mdsc_put_request(req);
2570         return;
2571 }
2572
2573
2574
2575 /*
2576  * handle mds notification that our request has been forwarded.
2577  */
2578 static void handle_forward(struct ceph_mds_client *mdsc,
2579                            struct ceph_mds_session *session,
2580                            struct ceph_msg *msg)
2581 {
2582         struct ceph_mds_request *req;
2583         u64 tid = le64_to_cpu(msg->hdr.tid);
2584         u32 next_mds;
2585         u32 fwd_seq;
2586         int err = -EINVAL;
2587         void *p = msg->front.iov_base;
2588         void *end = p + msg->front.iov_len;
2589
2590         ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2591         next_mds = ceph_decode_32(&p);
2592         fwd_seq = ceph_decode_32(&p);
2593
2594         mutex_lock(&mdsc->mutex);
2595         req = lookup_get_request(mdsc, tid);
2596         if (!req) {
2597                 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2598                 goto out;  /* dup reply? */
2599         }
2600
2601         if (req->r_aborted) {
2602                 dout("forward tid %llu aborted, unregistering\n", tid);
2603                 __unregister_request(mdsc, req);
2604         } else if (fwd_seq <= req->r_num_fwd) {
2605                 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2606                      tid, next_mds, req->r_num_fwd, fwd_seq);
2607         } else {
2608                 /* resend. forward race not possible; mds would drop */
2609                 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2610                 BUG_ON(req->r_err);
2611                 BUG_ON(req->r_got_result);
2612                 req->r_attempts = 0;
2613                 req->r_num_fwd = fwd_seq;
2614                 req->r_resend_mds = next_mds;
2615                 put_request_session(req);
2616                 __do_request(mdsc, req);
2617         }
2618         ceph_mdsc_put_request(req);
2619 out:
2620         mutex_unlock(&mdsc->mutex);
2621         return;
2622
2623 bad:
2624         pr_err("mdsc_handle_forward decode error err=%d\n", err);
2625 }
2626
2627 /*
2628  * handle a mds session control message
2629  */
2630 static void handle_session(struct ceph_mds_session *session,
2631                            struct ceph_msg *msg)
2632 {
2633         struct ceph_mds_client *mdsc = session->s_mdsc;
2634         u32 op;
2635         u64 seq;
2636         int mds = session->s_mds;
2637         struct ceph_mds_session_head *h = msg->front.iov_base;
2638         int wake = 0;
2639
2640         /* decode */
2641         if (msg->front.iov_len != sizeof(*h))
2642                 goto bad;
2643         op = le32_to_cpu(h->op);
2644         seq = le64_to_cpu(h->seq);
2645
2646         mutex_lock(&mdsc->mutex);
2647         if (op == CEPH_SESSION_CLOSE)
2648                 __unregister_session(mdsc, session);
2649         /* FIXME: this ttl calculation is generous */
2650         session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2651         mutex_unlock(&mdsc->mutex);
2652
2653         mutex_lock(&session->s_mutex);
2654
2655         dout("handle_session mds%d %s %p state %s seq %llu\n",
2656              mds, ceph_session_op_name(op), session,
2657              ceph_session_state_name(session->s_state), seq);
2658
2659         if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2660                 session->s_state = CEPH_MDS_SESSION_OPEN;
2661                 pr_info("mds%d came back\n", session->s_mds);
2662         }
2663
2664         switch (op) {
2665         case CEPH_SESSION_OPEN:
2666                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2667                         pr_info("mds%d reconnect success\n", session->s_mds);
2668                 session->s_state = CEPH_MDS_SESSION_OPEN;
2669                 renewed_caps(mdsc, session, 0);
2670                 wake = 1;
2671                 if (mdsc->stopping)
2672                         __close_session(mdsc, session);
2673                 break;
2674
2675         case CEPH_SESSION_RENEWCAPS:
2676                 if (session->s_renew_seq == seq)
2677                         renewed_caps(mdsc, session, 1);
2678                 break;
2679
2680         case CEPH_SESSION_CLOSE:
2681                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2682                         pr_info("mds%d reconnect denied\n", session->s_mds);
2683                 cleanup_session_requests(mdsc, session);
2684                 remove_session_caps(session);
2685                 wake = 2; /* for good measure */
2686                 wake_up_all(&mdsc->session_close_wq);
2687                 break;
2688
2689         case CEPH_SESSION_STALE:
2690                 pr_info("mds%d caps went stale, renewing\n",
2691                         session->s_mds);
2692                 spin_lock(&session->s_gen_ttl_lock);
2693                 session->s_cap_gen++;
2694                 session->s_cap_ttl = jiffies - 1;
2695                 spin_unlock(&session->s_gen_ttl_lock);
2696                 send_renew_caps(mdsc, session);
2697                 break;
2698
2699         case CEPH_SESSION_RECALL_STATE:
2700                 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2701                 break;
2702
2703         case CEPH_SESSION_FLUSHMSG:
2704                 send_flushmsg_ack(mdsc, session, seq);
2705                 break;
2706
2707         case CEPH_SESSION_FORCE_RO:
2708                 dout("force_session_readonly %p\n", session);
2709                 spin_lock(&session->s_cap_lock);
2710                 session->s_readonly = true;
2711                 spin_unlock(&session->s_cap_lock);
2712                 wake_up_session_caps(session, 0);
2713                 break;
2714
2715         default:
2716                 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2717                 WARN_ON(1);
2718         }
2719
2720         mutex_unlock(&session->s_mutex);
2721         if (wake) {
2722                 mutex_lock(&mdsc->mutex);
2723                 __wake_requests(mdsc, &session->s_waiting);
2724                 if (wake == 2)
2725                         kick_requests(mdsc, mds);
2726                 mutex_unlock(&mdsc->mutex);
2727         }
2728         return;
2729
2730 bad:
2731         pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2732                (int)msg->front.iov_len);
2733         ceph_msg_dump(msg);
2734         return;
2735 }
2736
2737
2738 /*
2739  * called under session->mutex.
2740  */
2741 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2742                                    struct ceph_mds_session *session)
2743 {
2744         struct ceph_mds_request *req, *nreq;
2745         struct rb_node *p;
2746         int err;
2747
2748         dout("replay_unsafe_requests mds%d\n", session->s_mds);
2749
2750         mutex_lock(&mdsc->mutex);
2751         list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2752                 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2753                 if (!err) {
2754                         ceph_msg_get(req->r_request);
2755                         ceph_con_send(&session->s_con, req->r_request);
2756                 }
2757         }
2758
2759         /*
2760          * also re-send old requests when MDS enters reconnect stage. So that MDS
2761          * can process completed request in clientreplay stage.
2762          */
2763         p = rb_first(&mdsc->request_tree);
2764         while (p) {
2765                 req = rb_entry(p, struct ceph_mds_request, r_node);
2766                 p = rb_next(p);
2767                 if (req->r_got_unsafe)
2768                         continue;
2769                 if (req->r_attempts == 0)
2770                         continue; /* only old requests */
2771                 if (req->r_session &&
2772                     req->r_session->s_mds == session->s_mds) {
2773                         err = __prepare_send_request(mdsc, req,
2774                                                      session->s_mds, true);
2775                         if (!err) {
2776                                 ceph_msg_get(req->r_request);
2777                                 ceph_con_send(&session->s_con, req->r_request);
2778                         }
2779                 }
2780         }
2781         mutex_unlock(&mdsc->mutex);
2782 }
2783
2784 /*
2785  * Encode information about a cap for a reconnect with the MDS.
2786  */
2787 static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2788                           void *arg)
2789 {
2790         union {
2791                 struct ceph_mds_cap_reconnect v2;
2792                 struct ceph_mds_cap_reconnect_v1 v1;
2793         } rec;
2794         size_t reclen;
2795         struct ceph_inode_info *ci;
2796         struct ceph_reconnect_state *recon_state = arg;
2797         struct ceph_pagelist *pagelist = recon_state->pagelist;
2798         char *path;
2799         int pathlen, err;
2800         u64 pathbase;
2801         struct dentry *dentry;
2802
2803         ci = cap->ci;
2804
2805         dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2806              inode, ceph_vinop(inode), cap, cap->cap_id,
2807              ceph_cap_string(cap->issued));
2808         err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2809         if (err)
2810                 return err;
2811
2812         dentry = d_find_alias(inode);
2813         if (dentry) {
2814                 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2815                 if (IS_ERR(path)) {
2816                         err = PTR_ERR(path);
2817                         goto out_dput;
2818                 }
2819         } else {
2820                 path = NULL;
2821                 pathlen = 0;
2822         }
2823         err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2824         if (err)
2825                 goto out_free;
2826
2827         spin_lock(&ci->i_ceph_lock);
2828         cap->seq = 0;        /* reset cap seq */
2829         cap->issue_seq = 0;  /* and issue_seq */
2830         cap->mseq = 0;       /* and migrate_seq */
2831         cap->cap_gen = cap->session->s_cap_gen;
2832
2833         if (recon_state->flock) {
2834                 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2835                 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2836                 rec.v2.issued = cpu_to_le32(cap->issued);
2837                 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2838                 rec.v2.pathbase = cpu_to_le64(pathbase);
2839                 rec.v2.flock_len = 0;
2840                 reclen = sizeof(rec.v2);
2841         } else {
2842                 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2843                 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2844                 rec.v1.issued = cpu_to_le32(cap->issued);
2845                 rec.v1.size = cpu_to_le64(inode->i_size);
2846                 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2847                 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2848                 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2849                 rec.v1.pathbase = cpu_to_le64(pathbase);
2850                 reclen = sizeof(rec.v1);
2851         }
2852         spin_unlock(&ci->i_ceph_lock);
2853
2854         if (recon_state->flock) {
2855                 int num_fcntl_locks, num_flock_locks;
2856                 struct ceph_filelock *flocks;
2857
2858 encode_again:
2859                 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
2860                 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2861                                  sizeof(struct ceph_filelock), GFP_NOFS);
2862                 if (!flocks) {
2863                         err = -ENOMEM;
2864                         goto out_free;
2865                 }
2866                 err = ceph_encode_locks_to_buffer(inode, flocks,
2867                                                   num_fcntl_locks,
2868                                                   num_flock_locks);
2869                 if (err) {
2870                         kfree(flocks);
2871                         if (err == -ENOSPC)
2872                                 goto encode_again;
2873                         goto out_free;
2874                 }
2875                 /*
2876                  * number of encoded locks is stable, so copy to pagelist
2877                  */
2878                 rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2879                                     (num_fcntl_locks+num_flock_locks) *
2880                                     sizeof(struct ceph_filelock));
2881                 err = ceph_pagelist_append(pagelist, &rec, reclen);
2882                 if (!err)
2883                         err = ceph_locks_to_pagelist(flocks, pagelist,
2884                                                      num_fcntl_locks,
2885                                                      num_flock_locks);
2886                 kfree(flocks);
2887         } else {
2888                 err = ceph_pagelist_append(pagelist, &rec, reclen);
2889         }
2890
2891         recon_state->nr_caps++;
2892 out_free:
2893         kfree(path);
2894 out_dput:
2895         dput(dentry);
2896         return err;
2897 }
2898
2899
2900 /*
2901  * If an MDS fails and recovers, clients need to reconnect in order to
2902  * reestablish shared state.  This includes all caps issued through
2903  * this session _and_ the snap_realm hierarchy.  Because it's not
2904  * clear which snap realms the mds cares about, we send everything we
2905  * know about.. that ensures we'll then get any new info the
2906  * recovering MDS might have.
2907  *
2908  * This is a relatively heavyweight operation, but it's rare.
2909  *
2910  * called with mdsc->mutex held.
2911  */
2912 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2913                                struct ceph_mds_session *session)
2914 {
2915         struct ceph_msg *reply;
2916         struct rb_node *p;
2917         int mds = session->s_mds;
2918         int err = -ENOMEM;
2919         int s_nr_caps;
2920         struct ceph_pagelist *pagelist;
2921         struct ceph_reconnect_state recon_state;
2922
2923         pr_info("mds%d reconnect start\n", mds);
2924
2925         pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2926         if (!pagelist)
2927                 goto fail_nopagelist;
2928         ceph_pagelist_init(pagelist);
2929
2930         reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
2931         if (!reply)
2932                 goto fail_nomsg;
2933
2934         mutex_lock(&session->s_mutex);
2935         session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2936         session->s_seq = 0;
2937
2938         dout("session %p state %s\n", session,
2939              ceph_session_state_name(session->s_state));
2940
2941         spin_lock(&session->s_gen_ttl_lock);
2942         session->s_cap_gen++;
2943         spin_unlock(&session->s_gen_ttl_lock);
2944
2945         spin_lock(&session->s_cap_lock);
2946         /* don't know if session is readonly */
2947         session->s_readonly = 0;
2948         /*
2949          * notify __ceph_remove_cap() that we are composing cap reconnect.
2950          * If a cap get released before being added to the cap reconnect,
2951          * __ceph_remove_cap() should skip queuing cap release.
2952          */
2953         session->s_cap_reconnect = 1;
2954         /* drop old cap expires; we're about to reestablish that state */
2955         cleanup_cap_releases(mdsc, session);
2956
2957         /* trim unused caps to reduce MDS's cache rejoin time */
2958         if (mdsc->fsc->sb->s_root)
2959                 shrink_dcache_parent(mdsc->fsc->sb->s_root);
2960
2961         ceph_con_close(&session->s_con);
2962         ceph_con_open(&session->s_con,
2963                       CEPH_ENTITY_TYPE_MDS, mds,
2964                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2965
2966         /* replay unsafe requests */
2967         replay_unsafe_requests(mdsc, session);
2968
2969         down_read(&mdsc->snap_rwsem);
2970
2971         /* traverse this session's caps */
2972         s_nr_caps = session->s_nr_caps;
2973         err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
2974         if (err)
2975                 goto fail;
2976
2977         recon_state.nr_caps = 0;
2978         recon_state.pagelist = pagelist;
2979         recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2980         err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2981         if (err < 0)
2982                 goto fail;
2983
2984         spin_lock(&session->s_cap_lock);
2985         session->s_cap_reconnect = 0;
2986         spin_unlock(&session->s_cap_lock);
2987
2988         /*
2989          * snaprealms.  we provide mds with the ino, seq (version), and
2990          * parent for all of our realms.  If the mds has any newer info,
2991          * it will tell us.
2992          */
2993         for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2994                 struct ceph_snap_realm *realm =
2995                         rb_entry(p, struct ceph_snap_realm, node);
2996                 struct ceph_mds_snaprealm_reconnect sr_rec;
2997
2998                 dout(" adding snap realm %llx seq %lld parent %llx\n",
2999                      realm->ino, realm->seq, realm->parent_ino);
3000                 sr_rec.ino = cpu_to_le64(realm->ino);
3001                 sr_rec.seq = cpu_to_le64(realm->seq);
3002                 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3003                 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3004                 if (err)
3005                         goto fail;
3006         }
3007
3008         if (recon_state.flock)
3009                 reply->hdr.version = cpu_to_le16(2);
3010
3011         /* raced with cap release? */
3012         if (s_nr_caps != recon_state.nr_caps) {
3013                 struct page *page = list_first_entry(&pagelist->head,
3014                                                      struct page, lru);
3015                 __le32 *addr = kmap_atomic(page);
3016                 *addr = cpu_to_le32(recon_state.nr_caps);
3017                 kunmap_atomic(addr);
3018         }
3019
3020         reply->hdr.data_len = cpu_to_le32(pagelist->length);
3021         ceph_msg_data_add_pagelist(reply, pagelist);
3022
3023         ceph_early_kick_flushing_caps(mdsc, session);
3024
3025         ceph_con_send(&session->s_con, reply);
3026
3027         mutex_unlock(&session->s_mutex);
3028
3029         mutex_lock(&mdsc->mutex);
3030         __wake_requests(mdsc, &session->s_waiting);
3031         mutex_unlock(&mdsc->mutex);
3032
3033         up_read(&mdsc->snap_rwsem);
3034         return;
3035
3036 fail:
3037         ceph_msg_put(reply);
3038         up_read(&mdsc->snap_rwsem);
3039         mutex_unlock(&session->s_mutex);
3040 fail_nomsg:
3041         ceph_pagelist_release(pagelist);
3042 fail_nopagelist:
3043         pr_err("error %d preparing reconnect for mds%d\n", err, mds);
3044         return;
3045 }
3046
3047
3048 /*
3049  * compare old and new mdsmaps, kicking requests
3050  * and closing out old connections as necessary
3051  *
3052  * called under mdsc->mutex.
3053  */
3054 static void check_new_map(struct ceph_mds_client *mdsc,
3055                           struct ceph_mdsmap *newmap,
3056                           struct ceph_mdsmap *oldmap)
3057 {
3058         int i;
3059         int oldstate, newstate;
3060         struct ceph_mds_session *s;
3061
3062         dout("check_new_map new %u old %u\n",
3063              newmap->m_epoch, oldmap->m_epoch);
3064
3065         for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
3066                 if (mdsc->sessions[i] == NULL)
3067                         continue;
3068                 s = mdsc->sessions[i];
3069                 oldstate = ceph_mdsmap_get_state(oldmap, i);
3070                 newstate = ceph_mdsmap_get_state(newmap, i);
3071
3072                 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
3073                      i, ceph_mds_state_name(oldstate),
3074                      ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
3075                      ceph_mds_state_name(newstate),
3076                      ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
3077                      ceph_session_state_name(s->s_state));
3078
3079                 if (i >= newmap->m_max_mds ||
3080                     memcmp(ceph_mdsmap_get_addr(oldmap, i),
3081                            ceph_mdsmap_get_addr(newmap, i),
3082                            sizeof(struct ceph_entity_addr))) {
3083                         if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3084                                 /* the session never opened, just close it
3085                                  * out now */
3086                                 __wake_requests(mdsc, &s->s_waiting);
3087                                 __unregister_session(mdsc, s);
3088                         } else {
3089                                 /* just close it */
3090                                 mutex_unlock(&mdsc->mutex);
3091                                 mutex_lock(&s->s_mutex);
3092                                 mutex_lock(&mdsc->mutex);
3093                                 ceph_con_close(&s->s_con);
3094                                 mutex_unlock(&s->s_mutex);
3095                                 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3096                         }
3097                 } else if (oldstate == newstate) {
3098                         continue;  /* nothing new with this mds */
3099                 }
3100
3101                 /*
3102                  * send reconnect?
3103                  */
3104                 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
3105                     newstate >= CEPH_MDS_STATE_RECONNECT) {
3106                         mutex_unlock(&mdsc->mutex);
3107                         send_mds_reconnect(mdsc, s);
3108                         mutex_lock(&mdsc->mutex);
3109                 }
3110
3111                 /*
3112                  * kick request on any mds that has gone active.
3113                  */
3114                 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3115                     newstate >= CEPH_MDS_STATE_ACTIVE) {
3116                         if (oldstate != CEPH_MDS_STATE_CREATING &&
3117                             oldstate != CEPH_MDS_STATE_STARTING)
3118                                 pr_info("mds%d recovery completed\n", s->s_mds);
3119                         kick_requests(mdsc, i);
3120                         ceph_kick_flushing_caps(mdsc, s);
3121                         wake_up_session_caps(s, 1);
3122                 }
3123         }
3124
3125         for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
3126                 s = mdsc->sessions[i];
3127                 if (!s)
3128                         continue;
3129                 if (!ceph_mdsmap_is_laggy(newmap, i))
3130                         continue;
3131                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3132                     s->s_state == CEPH_MDS_SESSION_HUNG ||
3133                     s->s_state == CEPH_MDS_SESSION_CLOSING) {
3134                         dout(" connecting to export targets of laggy mds%d\n",
3135                              i);
3136                         __open_export_target_sessions(mdsc, s);
3137                 }
3138         }
3139 }
3140
3141
3142
3143 /*
3144  * leases
3145  */
3146
3147 /*
3148  * caller must hold session s_mutex, dentry->d_lock
3149  */
3150 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3151 {
3152         struct ceph_dentry_info *di = ceph_dentry(dentry);
3153
3154         ceph_put_mds_session(di->lease_session);
3155         di->lease_session = NULL;
3156 }
3157
3158 static void handle_lease(struct ceph_mds_client *mdsc,
3159                          struct ceph_mds_session *session,
3160                          struct ceph_msg *msg)
3161 {
3162         struct super_block *sb = mdsc->fsc->sb;
3163         struct inode *inode;
3164         struct dentry *parent, *dentry;
3165         struct ceph_dentry_info *di;
3166         int mds = session->s_mds;
3167         struct ceph_mds_lease *h = msg->front.iov_base;
3168         u32 seq;
3169         struct ceph_vino vino;
3170         struct qstr dname;
3171         int release = 0;
3172
3173         dout("handle_lease from mds%d\n", mds);
3174
3175         /* decode */
3176         if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3177                 goto bad;
3178         vino.ino = le64_to_cpu(h->ino);
3179         vino.snap = CEPH_NOSNAP;
3180         seq = le32_to_cpu(h->seq);
3181         dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3182         dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3183         if (dname.len != get_unaligned_le32(h+1))
3184                 goto bad;
3185
3186         /* lookup inode */
3187         inode = ceph_find_inode(sb, vino);
3188         dout("handle_lease %s, ino %llx %p %.*s\n",
3189              ceph_lease_op_name(h->action), vino.ino, inode,
3190              dname.len, dname.name);
3191
3192         mutex_lock(&session->s_mutex);
3193         session->s_seq++;
3194
3195         if (inode == NULL) {
3196                 dout("handle_lease no inode %llx\n", vino.ino);
3197                 goto release;
3198         }
3199
3200         /* dentry */
3201         parent = d_find_alias(inode);
3202         if (!parent) {
3203                 dout("no parent dentry on inode %p\n", inode);
3204                 WARN_ON(1);
3205                 goto release;  /* hrm... */
3206         }
3207         dname.hash = full_name_hash(dname.name, dname.len);
3208         dentry = d_lookup(parent, &dname);
3209         dput(parent);
3210         if (!dentry)
3211                 goto release;
3212
3213         spin_lock(&dentry->d_lock);
3214         di = ceph_dentry(dentry);
3215         switch (h->action) {
3216         case CEPH_MDS_LEASE_REVOKE:
3217                 if (di->lease_session == session) {
3218                         if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3219                                 h->seq = cpu_to_le32(di->lease_seq);
3220                         __ceph_mdsc_drop_dentry_lease(dentry);
3221                 }
3222                 release = 1;
3223                 break;
3224
3225         case CEPH_MDS_LEASE_RENEW:
3226                 if (di->lease_session == session &&
3227                     di->lease_gen == session->s_cap_gen &&
3228                     di->lease_renew_from &&
3229                     di->lease_renew_after == 0) {
3230                         unsigned long duration =
3231                                 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
3232
3233                         di->lease_seq = seq;
3234                         dentry->d_time = di->lease_renew_from + duration;
3235                         di->lease_renew_after = di->lease_renew_from +
3236                                 (duration >> 1);
3237                         di->lease_renew_from = 0;
3238                 }
3239                 break;
3240         }
3241         spin_unlock(&dentry->d_lock);
3242         dput(dentry);
3243
3244         if (!release)
3245                 goto out;
3246
3247 release:
3248         /* let's just reuse the same message */
3249         h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3250         ceph_msg_get(msg);
3251         ceph_con_send(&session->s_con, msg);
3252
3253 out:
3254         iput(inode);
3255         mutex_unlock(&session->s_mutex);
3256         return;
3257
3258 bad:
3259         pr_err("corrupt lease message\n");
3260         ceph_msg_dump(msg);
3261 }
3262
3263 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3264                               struct inode *inode,
3265                               struct dentry *dentry, char action,
3266                               u32 seq)
3267 {
3268         struct ceph_msg *msg;
3269         struct ceph_mds_lease *lease;
3270         int len = sizeof(*lease) + sizeof(u32);
3271         int dnamelen = 0;
3272
3273         dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3274              inode, dentry, ceph_lease_op_name(action), session->s_mds);
3275         dnamelen = dentry->d_name.len;
3276         len += dnamelen;
3277
3278         msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
3279         if (!msg)
3280                 return;
3281         lease = msg->front.iov_base;
3282         lease->action = action;
3283         lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3284         lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3285         lease->seq = cpu_to_le32(seq);
3286         put_unaligned_le32(dnamelen, lease + 1);
3287         memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3288
3289         /*
3290          * if this is a preemptive lease RELEASE, no need to
3291          * flush request stream, since the actual request will
3292          * soon follow.
3293          */
3294         msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3295
3296         ceph_con_send(&session->s_con, msg);
3297 }
3298
3299 /*
3300  * Preemptively release a lease we expect to invalidate anyway.
3301  * Pass @inode always, @dentry is optional.
3302  */
3303 void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
3304                              struct dentry *dentry)
3305 {
3306         struct ceph_dentry_info *di;
3307         struct ceph_mds_session *session;
3308         u32 seq;
3309
3310         BUG_ON(inode == NULL);
3311         BUG_ON(dentry == NULL);
3312
3313         /* is dentry lease valid? */
3314         spin_lock(&dentry->d_lock);
3315         di = ceph_dentry(dentry);
3316         if (!di || !di->lease_session ||
3317             di->lease_session->s_mds < 0 ||
3318             di->lease_gen != di->lease_session->s_cap_gen ||
3319             !time_before(jiffies, dentry->d_time)) {
3320                 dout("lease_release inode %p dentry %p -- "
3321                      "no lease\n",
3322                      inode, dentry);
3323                 spin_unlock(&dentry->d_lock);
3324                 return;
3325         }
3326
3327         /* we do have a lease on this dentry; note mds and seq */
3328         session = ceph_get_mds_session(di->lease_session);
3329         seq = di->lease_seq;
3330         __ceph_mdsc_drop_dentry_lease(dentry);
3331         spin_unlock(&dentry->d_lock);
3332
3333         dout("lease_release inode %p dentry %p to mds%d\n",
3334              inode, dentry, session->s_mds);
3335         ceph_mdsc_lease_send_msg(session, inode, dentry,
3336                                  CEPH_MDS_LEASE_RELEASE, seq);
3337         ceph_put_mds_session(session);
3338 }
3339
3340 /*
3341  * drop all leases (and dentry refs) in preparation for umount
3342  */
3343 static void drop_leases(struct ceph_mds_client *mdsc)
3344 {
3345         int i;
3346
3347         dout("drop_leases\n");
3348         mutex_lock(&mdsc->mutex);
3349         for (i = 0; i < mdsc->max_sessions; i++) {
3350                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3351                 if (!s)
3352                         continue;
3353                 mutex_unlock(&mdsc->mutex);
3354                 mutex_lock(&s->s_mutex);
3355                 mutex_unlock(&s->s_mutex);
3356                 ceph_put_mds_session(s);
3357                 mutex_lock(&mdsc->mutex);
3358         }
3359         mutex_unlock(&mdsc->mutex);
3360 }
3361
3362
3363
3364 /*
3365  * delayed work -- periodically trim expired leases, renew caps with mds
3366  */
3367 static void schedule_delayed(struct ceph_mds_client *mdsc)
3368 {
3369         int delay = 5;
3370         unsigned hz = round_jiffies_relative(HZ * delay);
3371         schedule_delayed_work(&mdsc->delayed_work, hz);
3372 }
3373
3374 static void delayed_work(struct work_struct *work)
3375 {
3376         int i;
3377         struct ceph_mds_client *mdsc =
3378                 container_of(work, struct ceph_mds_client, delayed_work.work);
3379         int renew_interval;
3380         int renew_caps;
3381
3382         dout("mdsc delayed_work\n");
3383         ceph_check_delayed_caps(mdsc);
3384
3385         mutex_lock(&mdsc->mutex);
3386         renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3387         renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3388                                    mdsc->last_renew_caps);
3389         if (renew_caps)
3390                 mdsc->last_renew_caps = jiffies;
3391
3392         for (i = 0; i < mdsc->max_sessions; i++) {
3393                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3394                 if (s == NULL)
3395                         continue;
3396                 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3397                         dout("resending session close request for mds%d\n",
3398                              s->s_mds);
3399                         request_close_session(mdsc, s);
3400                         ceph_put_mds_session(s);
3401                         continue;
3402                 }
3403                 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3404                         if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3405                                 s->s_state = CEPH_MDS_SESSION_HUNG;
3406                                 pr_info("mds%d hung\n", s->s_mds);
3407                         }
3408                 }
3409                 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3410                         /* this mds is failed or recovering, just wait */
3411                         ceph_put_mds_session(s);
3412                         continue;
3413                 }
3414                 mutex_unlock(&mdsc->mutex);
3415
3416                 mutex_lock(&s->s_mutex);
3417                 if (renew_caps)
3418                         send_renew_caps(mdsc, s);
3419                 else
3420                         ceph_con_keepalive(&s->s_con);
3421                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3422                     s->s_state == CEPH_MDS_SESSION_HUNG)
3423                         ceph_send_cap_releases(mdsc, s);
3424                 mutex_unlock(&s->s_mutex);
3425                 ceph_put_mds_session(s);
3426
3427                 mutex_lock(&mdsc->mutex);
3428         }
3429         mutex_unlock(&mdsc->mutex);
3430
3431         schedule_delayed(mdsc);
3432 }
3433
3434 int ceph_mdsc_init(struct ceph_fs_client *fsc)
3435
3436 {
3437         struct ceph_mds_client *mdsc;
3438
3439         mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3440         if (!mdsc)
3441                 return -ENOMEM;
3442         mdsc->fsc = fsc;
3443         fsc->mdsc = mdsc;
3444         mutex_init(&mdsc->mutex);
3445         mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
3446         if (mdsc->mdsmap == NULL) {
3447                 kfree(mdsc);
3448                 return -ENOMEM;
3449         }
3450
3451         init_completion(&mdsc->safe_umount_waiters);
3452         init_waitqueue_head(&mdsc->session_close_wq);
3453         INIT_LIST_HEAD(&mdsc->waiting_for_map);
3454         mdsc->sessions = NULL;
3455         atomic_set(&mdsc->num_sessions, 0);
3456         mdsc->max_sessions = 0;
3457         mdsc->stopping = 0;
3458         mdsc->last_snap_seq = 0;
3459         init_rwsem(&mdsc->snap_rwsem);
3460         mdsc->snap_realms = RB_ROOT;
3461         INIT_LIST_HEAD(&mdsc->snap_empty);
3462         spin_lock_init(&mdsc->snap_empty_lock);
3463         mdsc->last_tid = 0;
3464         mdsc->oldest_tid = 0;
3465         mdsc->request_tree = RB_ROOT;
3466         INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3467         mdsc->last_renew_caps = jiffies;
3468         INIT_LIST_HEAD(&mdsc->cap_delay_list);
3469         spin_lock_init(&mdsc->cap_delay_lock);
3470         INIT_LIST_HEAD(&mdsc->snap_flush_list);
3471         spin_lock_init(&mdsc->snap_flush_lock);
3472         mdsc->last_cap_flush_tid = 1;
3473         mdsc->cap_flush_tree = RB_ROOT;
3474         INIT_LIST_HEAD(&mdsc->cap_dirty);
3475         INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
3476         mdsc->num_cap_flushing = 0;
3477         spin_lock_init(&mdsc->cap_dirty_lock);
3478         init_waitqueue_head(&mdsc->cap_flushing_wq);
3479         spin_lock_init(&mdsc->dentry_lru_lock);
3480         INIT_LIST_HEAD(&mdsc->dentry_lru);
3481
3482         ceph_caps_init(mdsc);
3483         ceph_adjust_min_caps(mdsc, fsc->min_caps);
3484
3485         init_rwsem(&mdsc->pool_perm_rwsem);
3486         mdsc->pool_perm_tree = RB_ROOT;
3487
3488         return 0;
3489 }
3490
3491 /*
3492  * Wait for safe replies on open mds requests.  If we time out, drop
3493  * all requests from the tree to avoid dangling dentry refs.
3494  */
3495 static void wait_requests(struct ceph_mds_client *mdsc)
3496 {
3497         struct ceph_options *opts = mdsc->fsc->client->options;
3498         struct ceph_mds_request *req;
3499
3500         mutex_lock(&mdsc->mutex);
3501         if (__get_oldest_req(mdsc)) {
3502                 mutex_unlock(&mdsc->mutex);
3503
3504                 dout("wait_requests waiting for requests\n");
3505                 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
3506                                     ceph_timeout_jiffies(opts->mount_timeout));
3507
3508                 /* tear down remaining requests */
3509                 mutex_lock(&mdsc->mutex);
3510                 while ((req = __get_oldest_req(mdsc))) {
3511                         dout("wait_requests timed out on tid %llu\n",
3512                              req->r_tid);
3513                         __unregister_request(mdsc, req);
3514                 }
3515         }
3516         mutex_unlock(&mdsc->mutex);
3517         dout("wait_requests done\n");
3518 }
3519
3520 /*
3521  * called before mount is ro, and before dentries are torn down.
3522  * (hmm, does this still race with new lookups?)
3523  */
3524 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3525 {
3526         dout("pre_umount\n");
3527         mdsc->stopping = 1;
3528
3529         drop_leases(mdsc);
3530         ceph_flush_dirty_caps(mdsc);
3531         wait_requests(mdsc);
3532
3533         /*
3534          * wait for reply handlers to drop their request refs and
3535          * their inode/dcache refs
3536          */
3537         ceph_msgr_flush();
3538 }
3539
3540 /*
3541  * wait for all write mds requests to flush.
3542  */
3543 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3544 {
3545         struct ceph_mds_request *req = NULL, *nextreq;
3546         struct rb_node *n;
3547
3548         mutex_lock(&mdsc->mutex);
3549         dout("wait_unsafe_requests want %lld\n", want_tid);
3550 restart:
3551         req = __get_oldest_req(mdsc);
3552         while (req && req->r_tid <= want_tid) {
3553                 /* find next request */
3554                 n = rb_next(&req->r_node);
3555                 if (n)
3556                         nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3557                 else
3558                         nextreq = NULL;
3559                 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3560                     (req->r_op & CEPH_MDS_OP_WRITE)) {
3561                         /* write op */
3562                         ceph_mdsc_get_request(req);
3563                         if (nextreq)
3564                                 ceph_mdsc_get_request(nextreq);
3565                         mutex_unlock(&mdsc->mutex);
3566                         dout("wait_unsafe_requests  wait on %llu (want %llu)\n",
3567                              req->r_tid, want_tid);
3568                         wait_for_completion(&req->r_safe_completion);
3569                         mutex_lock(&mdsc->mutex);
3570                         ceph_mdsc_put_request(req);
3571                         if (!nextreq)
3572                                 break;  /* next dne before, so we're done! */
3573                         if (RB_EMPTY_NODE(&nextreq->r_node)) {
3574                                 /* next request was removed from tree */
3575                                 ceph_mdsc_put_request(nextreq);
3576                                 goto restart;
3577                         }
3578                         ceph_mdsc_put_request(nextreq);  /* won't go away */
3579                 }
3580                 req = nextreq;
3581         }
3582         mutex_unlock(&mdsc->mutex);
3583         dout("wait_unsafe_requests done\n");
3584 }
3585
3586 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3587 {
3588         u64 want_tid, want_flush, want_snap;
3589
3590         if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
3591                 return;
3592
3593         dout("sync\n");
3594         mutex_lock(&mdsc->mutex);
3595         want_tid = mdsc->last_tid;
3596         mutex_unlock(&mdsc->mutex);
3597
3598         ceph_flush_dirty_caps(mdsc);
3599         spin_lock(&mdsc->cap_dirty_lock);
3600         want_flush = mdsc->last_cap_flush_tid;
3601         spin_unlock(&mdsc->cap_dirty_lock);
3602
3603         down_read(&mdsc->snap_rwsem);
3604         want_snap = mdsc->last_snap_seq;
3605         up_read(&mdsc->snap_rwsem);
3606
3607         dout("sync want tid %lld flush_seq %lld snap_seq %lld\n",
3608              want_tid, want_flush, want_snap);
3609
3610         wait_unsafe_requests(mdsc, want_tid);
3611         wait_caps_flush(mdsc, want_flush, want_snap);
3612 }
3613
3614 /*
3615  * true if all sessions are closed, or we force unmount
3616  */
3617 static bool done_closing_sessions(struct ceph_mds_client *mdsc)
3618 {
3619         if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
3620                 return true;
3621         return atomic_read(&mdsc->num_sessions) == 0;
3622 }
3623
3624 /*
3625  * called after sb is ro.
3626  */
3627 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3628 {
3629         struct ceph_options *opts = mdsc->fsc->client->options;
3630         struct ceph_mds_session *session;
3631         int i;
3632
3633         dout("close_sessions\n");
3634
3635         /* close sessions */
3636         mutex_lock(&mdsc->mutex);
3637         for (i = 0; i < mdsc->max_sessions; i++) {
3638                 session = __ceph_lookup_mds_session(mdsc, i);
3639                 if (!session)
3640                         continue;
3641                 mutex_unlock(&mdsc->mutex);
3642                 mutex_lock(&session->s_mutex);
3643                 __close_session(mdsc, session);
3644                 mutex_unlock(&session->s_mutex);
3645                 ceph_put_mds_session(session);
3646                 mutex_lock(&mdsc->mutex);
3647         }
3648         mutex_unlock(&mdsc->mutex);
3649
3650         dout("waiting for sessions to close\n");
3651         wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3652                            ceph_timeout_jiffies(opts->mount_timeout));
3653
3654         /* tear down remaining sessions */
3655         mutex_lock(&mdsc->mutex);
3656         for (i = 0; i < mdsc->max_sessions; i++) {
3657                 if (mdsc->sessions[i]) {
3658                         session = get_session(mdsc->sessions[i]);
3659                         __unregister_session(mdsc, session);
3660                         mutex_unlock(&mdsc->mutex);
3661                         mutex_lock(&session->s_mutex);
3662                         remove_session_caps(session);
3663                         mutex_unlock(&session->s_mutex);
3664                         ceph_put_mds_session(session);
3665                         mutex_lock(&mdsc->mutex);
3666                 }
3667         }
3668         WARN_ON(!list_empty(&mdsc->cap_delay_list));
3669         mutex_unlock(&mdsc->mutex);
3670
3671         ceph_cleanup_empty_realms(mdsc);
3672
3673         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3674
3675         dout("stopped\n");
3676 }
3677
3678 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3679 {
3680         struct ceph_mds_session *session;
3681         int mds;
3682
3683         dout("force umount\n");
3684
3685         mutex_lock(&mdsc->mutex);
3686         for (mds = 0; mds < mdsc->max_sessions; mds++) {
3687                 session = __ceph_lookup_mds_session(mdsc, mds);
3688                 if (!session)
3689                         continue;
3690                 mutex_unlock(&mdsc->mutex);
3691                 mutex_lock(&session->s_mutex);
3692                 __close_session(mdsc, session);
3693                 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3694                         cleanup_session_requests(mdsc, session);
3695                         remove_session_caps(session);
3696                 }
3697                 mutex_unlock(&session->s_mutex);
3698                 ceph_put_mds_session(session);
3699                 mutex_lock(&mdsc->mutex);
3700                 kick_requests(mdsc, mds);
3701         }
3702         __wake_requests(mdsc, &mdsc->waiting_for_map);
3703         mutex_unlock(&mdsc->mutex);
3704 }
3705
3706 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
3707 {
3708         dout("stop\n");
3709         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3710         if (mdsc->mdsmap)
3711                 ceph_mdsmap_destroy(mdsc->mdsmap);
3712         kfree(mdsc->sessions);
3713         ceph_caps_finalize(mdsc);
3714         ceph_pool_perm_destroy(mdsc);
3715 }
3716
3717 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3718 {
3719         struct ceph_mds_client *mdsc = fsc->mdsc;
3720
3721         dout("mdsc_destroy %p\n", mdsc);
3722         ceph_mdsc_stop(mdsc);
3723
3724         /* flush out any connection work with references to us */
3725         ceph_msgr_flush();
3726
3727         fsc->mdsc = NULL;
3728         kfree(mdsc);
3729         dout("mdsc_destroy %p done\n", mdsc);
3730 }
3731
3732
3733 /*
3734  * handle mds map update.
3735  */
3736 void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3737 {
3738         u32 epoch;
3739         u32 maplen;
3740         void *p = msg->front.iov_base;
3741         void *end = p + msg->front.iov_len;
3742         struct ceph_mdsmap *newmap, *oldmap;
3743         struct ceph_fsid fsid;
3744         int err = -EINVAL;
3745
3746         ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3747         ceph_decode_copy(&p, &fsid, sizeof(fsid));
3748         if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
3749                 return;
3750         epoch = ceph_decode_32(&p);
3751         maplen = ceph_decode_32(&p);
3752         dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3753
3754         /* do we need it? */
3755         mutex_lock(&mdsc->mutex);
3756         if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3757                 dout("handle_map epoch %u <= our %u\n",
3758                      epoch, mdsc->mdsmap->m_epoch);
3759                 mutex_unlock(&mdsc->mutex);
3760                 return;
3761         }
3762
3763         newmap = ceph_mdsmap_decode(&p, end);
3764         if (IS_ERR(newmap)) {
3765                 err = PTR_ERR(newmap);
3766                 goto bad_unlock;
3767         }
3768
3769         /* swap into place */
3770         if (mdsc->mdsmap) {
3771                 oldmap = mdsc->mdsmap;
3772                 mdsc->mdsmap = newmap;
3773                 check_new_map(mdsc, newmap, oldmap);
3774                 ceph_mdsmap_destroy(oldmap);
3775         } else {
3776                 mdsc->mdsmap = newmap;  /* first mds map */
3777         }
3778         mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
3779
3780         __wake_requests(mdsc, &mdsc->waiting_for_map);
3781         ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
3782                           mdsc->mdsmap->m_epoch);
3783
3784         mutex_unlock(&mdsc->mutex);
3785         schedule_delayed(mdsc);
3786         return;
3787
3788 bad_unlock:
3789         mutex_unlock(&mdsc->mutex);
3790 bad:
3791         pr_err("error decoding mdsmap %d\n", err);
3792         return;
3793 }
3794
3795 static struct ceph_connection *con_get(struct ceph_connection *con)
3796 {
3797         struct ceph_mds_session *s = con->private;
3798
3799         if (get_session(s)) {
3800                 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
3801                 return con;
3802         }
3803         dout("mdsc con_get %p FAIL\n", s);
3804         return NULL;
3805 }
3806
3807 static void con_put(struct ceph_connection *con)
3808 {
3809         struct ceph_mds_session *s = con->private;
3810
3811         dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
3812         ceph_put_mds_session(s);
3813 }
3814
3815 /*
3816  * if the client is unresponsive for long enough, the mds will kill
3817  * the session entirely.
3818  */
3819 static void peer_reset(struct ceph_connection *con)
3820 {
3821         struct ceph_mds_session *s = con->private;
3822         struct ceph_mds_client *mdsc = s->s_mdsc;
3823
3824         pr_warn("mds%d closed our session\n", s->s_mds);
3825         send_mds_reconnect(mdsc, s);
3826 }
3827
3828 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3829 {
3830         struct ceph_mds_session *s = con->private;
3831         struct ceph_mds_client *mdsc = s->s_mdsc;
3832         int type = le16_to_cpu(msg->hdr.type);
3833
3834         mutex_lock(&mdsc->mutex);
3835         if (__verify_registered_session(mdsc, s) < 0) {
3836                 mutex_unlock(&mdsc->mutex);
3837                 goto out;
3838         }
3839         mutex_unlock(&mdsc->mutex);
3840
3841         switch (type) {
3842         case CEPH_MSG_MDS_MAP:
3843                 ceph_mdsc_handle_map(mdsc, msg);
3844                 break;
3845         case CEPH_MSG_CLIENT_SESSION:
3846                 handle_session(s, msg);
3847                 break;
3848         case CEPH_MSG_CLIENT_REPLY:
3849                 handle_reply(s, msg);
3850                 break;
3851         case CEPH_MSG_CLIENT_REQUEST_FORWARD:
3852                 handle_forward(mdsc, s, msg);
3853                 break;
3854         case CEPH_MSG_CLIENT_CAPS:
3855                 ceph_handle_caps(s, msg);
3856                 break;
3857         case CEPH_MSG_CLIENT_SNAP:
3858                 ceph_handle_snap(mdsc, s, msg);
3859                 break;
3860         case CEPH_MSG_CLIENT_LEASE:
3861                 handle_lease(mdsc, s, msg);
3862                 break;
3863
3864         default:
3865                 pr_err("received unknown message type %d %s\n", type,
3866                        ceph_msg_type_name(type));
3867         }
3868 out:
3869         ceph_msg_put(msg);
3870 }
3871
3872 /*
3873  * authentication
3874  */
3875
3876 /*
3877  * Note: returned pointer is the address of a structure that's
3878  * managed separately.  Caller must *not* attempt to free it.
3879  */
3880 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
3881                                         int *proto, int force_new)
3882 {
3883         struct ceph_mds_session *s = con->private;
3884         struct ceph_mds_client *mdsc = s->s_mdsc;
3885         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3886         struct ceph_auth_handshake *auth = &s->s_auth;
3887
3888         if (force_new && auth->authorizer) {
3889                 ceph_auth_destroy_authorizer(auth->authorizer);
3890                 auth->authorizer = NULL;
3891         }
3892         if (!auth->authorizer) {
3893                 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3894                                                       auth);
3895                 if (ret)
3896                         return ERR_PTR(ret);
3897         } else {
3898                 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3899                                                       auth);
3900                 if (ret)
3901                         return ERR_PTR(ret);
3902         }
3903         *proto = ac->protocol;
3904
3905         return auth;
3906 }
3907
3908
3909 static int verify_authorizer_reply(struct ceph_connection *con, int len)
3910 {
3911         struct ceph_mds_session *s = con->private;
3912         struct ceph_mds_client *mdsc = s->s_mdsc;
3913         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3914
3915         return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
3916 }
3917
3918 static int invalidate_authorizer(struct ceph_connection *con)
3919 {
3920         struct ceph_mds_session *s = con->private;
3921         struct ceph_mds_client *mdsc = s->s_mdsc;
3922         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3923
3924         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
3925
3926         return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
3927 }
3928
3929 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3930                                 struct ceph_msg_header *hdr, int *skip)
3931 {
3932         struct ceph_msg *msg;
3933         int type = (int) le16_to_cpu(hdr->type);
3934         int front_len = (int) le32_to_cpu(hdr->front_len);
3935
3936         if (con->in_msg)
3937                 return con->in_msg;
3938
3939         *skip = 0;
3940         msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3941         if (!msg) {
3942                 pr_err("unable to allocate msg type %d len %d\n",
3943                        type, front_len);
3944                 return NULL;
3945         }
3946
3947         return msg;
3948 }
3949
3950 static int mds_sign_message(struct ceph_msg *msg)
3951 {
3952        struct ceph_mds_session *s = msg->con->private;
3953        struct ceph_auth_handshake *auth = &s->s_auth;
3954
3955        return ceph_auth_sign_message(auth, msg);
3956 }
3957
3958 static int mds_check_message_signature(struct ceph_msg *msg)
3959 {
3960        struct ceph_mds_session *s = msg->con->private;
3961        struct ceph_auth_handshake *auth = &s->s_auth;
3962
3963        return ceph_auth_check_message_signature(auth, msg);
3964 }
3965
3966 static const struct ceph_connection_operations mds_con_ops = {
3967         .get = con_get,
3968         .put = con_put,
3969         .dispatch = dispatch,
3970         .get_authorizer = get_authorizer,
3971         .verify_authorizer_reply = verify_authorizer_reply,
3972         .invalidate_authorizer = invalidate_authorizer,
3973         .peer_reset = peer_reset,
3974         .alloc_msg = mds_alloc_msg,
3975         .sign_message = mds_sign_message,
3976         .check_message_signature = mds_check_message_signature,
3977 };
3978
3979 /* eof */