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/sched.h>
   7 #include <linux/debugfs.h>
   8 #include <linux/seq_file.h>
   9
  10 #include "super.h"
  11 #include "mds_client.h"
  12
  13 #include <linux/ceph/messenger.h>
  14 #include <linux/ceph/decode.h>
  15 #include <linux/ceph/pagelist.h>
  16 #include <linux/ceph/auth.h>
  17 #include <linux/ceph/debugfs.h>
  18
  19 /*
  20  * A cluster of MDS (metadata server) daemons is responsible for
  21  * managing the file system namespace (the directory hierarchy and
  22  * inodes) and for coordinating shared access to storage.  Metadata is
  23  * partitioning hierarchically across a number of servers, and that
  24  * partition varies over time as the cluster adjusts the distribution
  25  * in order to balance load.
  26  *
  27  * The MDS client is primarily responsible to managing synchronous
  28  * metadata requests for operations like open, unlink, and so forth.
  29  * If there is a MDS failure, we find out about it when we (possibly
  30  * request and) receive a new MDS map, and can resubmit affected
  31  * requests.
  32  *
  33  * For the most part, though, we take advantage of a lossless
  34  * communications channel to the MDS, and do not need to worry about
  35  * timing out or resubmitting requests.
  36  *
  37  * We maintain a stateful "session" with each MDS we interact with.
  38  * Within each session, we sent periodic heartbeat messages to ensure
  39  * any capabilities or leases we have been issues remain valid.  If
  40  * the session times out and goes stale, our leases and capabilities
  41  * are no longer valid.
  42  */
  43
  44 struct ceph_reconnect_state {
  45         struct ceph_pagelist *pagelist;
  46         bool flock;
  47 };
  48
  49 static void __wake_requests(struct ceph_mds_client *mdsc,
  50                             struct list_head *head);
  51
  52 static const struct ceph_connection_operations mds_con_ops;
  53
  54
  55 /*
  56  * mds reply parsing
  57  */
  58
  59 /*
  60  * parse individual inode info
  61  */
  62 static int parse_reply_info_in(void **p, void *end,
  63                                struct ceph_mds_reply_info_in *info,
  64                                int features)
  65 {
  66         int err = -EIO;
  67
  68         info->in = *p;
  69         *p += sizeof(struct ceph_mds_reply_inode) +
  70                 sizeof(*info->in->fragtree.splits) *
  71                 le32_to_cpu(info->in->fragtree.nsplits);
  72
  73         ceph_decode_32_safe(p, end, info->symlink_len, bad);
  74         ceph_decode_need(p, end, info->symlink_len, bad);
  75         info->symlink = *p;
  76         *p += info->symlink_len;
  77
  78         if (features & CEPH_FEATURE_DIRLAYOUTHASH)
  79                 ceph_decode_copy_safe(p, end, &info->dir_layout,
  80                                       sizeof(info->dir_layout), bad);
  81         else
  82                 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
  83
  84         ceph_decode_32_safe(p, end, info->xattr_len, bad);
  85         ceph_decode_need(p, end, info->xattr_len, bad);
  86         info->xattr_data = *p;
  87         *p += info->xattr_len;
  88         return 0;
  89 bad:
  90         return err;
  91 }
  92
  93 /*
  94  * parse a normal reply, which may contain a (dir+)dentry and/or a
  95  * target inode.
  96  */
  97 static int parse_reply_info_trace(void **p, void *end,
  98                                   struct ceph_mds_reply_info_parsed *info,
  99                                   int features)
 100 {
 101         int err;
 102
 103         if (info->head->is_dentry) {
 104                 err = parse_reply_info_in(p, end, &info->diri, features);
 105                 if (err < 0)
 106                         goto out_bad;
 107
 108                 if (unlikely(*p + sizeof(*info->dirfrag) > end))
 109                         goto bad;
 110                 info->dirfrag = *p;
 111                 *p += sizeof(*info->dirfrag) +
 112                         sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
 113                 if (unlikely(*p > end))
 114                         goto bad;
 115
 116                 ceph_decode_32_safe(p, end, info->dname_len, bad);
 117                 ceph_decode_need(p, end, info->dname_len, bad);
 118                 info->dname = *p;
 119                 *p += info->dname_len;
 120                 info->dlease = *p;
 121                 *p += sizeof(*info->dlease);
 122         }
 123
 124         if (info->head->is_target) {
 125                 err = parse_reply_info_in(p, end, &info->targeti, features);
 126                 if (err < 0)
 127                         goto out_bad;
 128         }
 129
 130         if (unlikely(*p != end))
 131                 goto bad;
 132         return 0;
 133
 134 bad:
 135         err = -EIO;
 136 out_bad:
 137         pr_err("problem parsing mds trace %d\n", err);
 138         return err;
 139 }
 140
 141 /*
 142  * parse readdir results
 143  */
 144 static int parse_reply_info_dir(void **p, void *end,
 145                                 struct ceph_mds_reply_info_parsed *info,
 146                                 int features)
 147 {
 148         u32 num, i = 0;
 149         int err;
 150
 151         info->dir_dir = *p;
 152         if (*p + sizeof(*info->dir_dir) > end)
 153                 goto bad;
 154         *p += sizeof(*info->dir_dir) +
 155                 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
 156         if (*p > end)
 157                 goto bad;
 158
 159         ceph_decode_need(p, end, sizeof(num) + 2, bad);
 160         num = ceph_decode_32(p);
 161         info->dir_end = ceph_decode_8(p);
 162         info->dir_complete = ceph_decode_8(p);
 163         if (num == 0)
 164                 goto done;
 165
 166         /* alloc large array */
 167         info->dir_nr = num;
 168         info->dir_in = kcalloc(num, sizeof(*info->dir_in) +
 169                                sizeof(*info->dir_dname) +
 170                                sizeof(*info->dir_dname_len) +
 171                                sizeof(*info->dir_dlease),
 172                                GFP_NOFS);
 173         if (info->dir_in == NULL) {
 174                 err = -ENOMEM;
 175                 goto out_bad;
 176         }
 177         info->dir_dname = (void *)(info->dir_in + num);
 178         info->dir_dname_len = (void *)(info->dir_dname + num);
 179         info->dir_dlease = (void *)(info->dir_dname_len + num);
 180
 181         while (num) {
 182                 /* dentry */
 183                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 184                 info->dir_dname_len[i] = ceph_decode_32(p);
 185                 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
 186                 info->dir_dname[i] = *p;
 187                 *p += info->dir_dname_len[i];
 188                 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
 189                      info->dir_dname[i]);
 190                 info->dir_dlease[i] = *p;
 191                 *p += sizeof(struct ceph_mds_reply_lease);
 192
 193                 /* inode */
 194                 err = parse_reply_info_in(p, end, &info->dir_in[i], features);
 195                 if (err < 0)
 196                         goto out_bad;
 197                 i++;
 198                 num--;
 199         }
 200
 201 done:
 202         if (*p != end)
 203                 goto bad;
 204         return 0;
 205
 206 bad:
 207         err = -EIO;
 208 out_bad:
 209         pr_err("problem parsing dir contents %d\n", err);
 210         return err;
 211 }
 212
 213 /*
 214  * parse fcntl F_GETLK results
 215  */
 216 static int parse_reply_info_filelock(void **p, void *end,
 217                                      struct ceph_mds_reply_info_parsed *info,
 218                                      int features)
 219 {
 220         if (*p + sizeof(*info->filelock_reply) > end)
 221                 goto bad;
 222
 223         info->filelock_reply = *p;
 224         *p += sizeof(*info->filelock_reply);
 225
 226         if (unlikely(*p != end))
 227                 goto bad;
 228         return 0;
 229
 230 bad:
 231         return -EIO;
 232 }
 233
 234 /*
 235  * parse extra results
 236  */
 237 static int parse_reply_info_extra(void **p, void *end,
 238                                   struct ceph_mds_reply_info_parsed *info,
 239                                   int features)
 240 {
 241         if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
 242                 return parse_reply_info_filelock(p, end, info, features);
 243         else
 244                 return parse_reply_info_dir(p, end, info, features);
 245 }
 246
 247 /*
 248  * parse entire mds reply
 249  */
 250 static int parse_reply_info(struct ceph_msg *msg,
 251                             struct ceph_mds_reply_info_parsed *info,
 252                             int features)
 253 {
 254         void *p, *end;
 255         u32 len;
 256         int err;
 257
 258         info->head = msg->front.iov_base;
 259         p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
 260         end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
 261
 262         /* trace */
 263         ceph_decode_32_safe(&p, end, len, bad);
 264         if (len > 0) {
 265                 ceph_decode_need(&p, end, len, bad);
 266                 err = parse_reply_info_trace(&p, p+len, info, features);
 267                 if (err < 0)
 268                         goto out_bad;
 269         }
 270
 271         /* extra */
 272         ceph_decode_32_safe(&p, end, len, bad);
 273         if (len > 0) {
 274                 ceph_decode_need(&p, end, len, bad);
 275                 err = parse_reply_info_extra(&p, p+len, info, features);
 276                 if (err < 0)
 277                         goto out_bad;
 278         }
 279
 280         /* snap blob */
 281         ceph_decode_32_safe(&p, end, len, bad);
 282         info->snapblob_len = len;
 283         info->snapblob = p;
 284         p += len;
 285
 286         if (p != end)
 287                 goto bad;
 288         return 0;
 289
 290 bad:
 291         err = -EIO;
 292 out_bad:
 293         pr_err("mds parse_reply err %d\n", err);
 294         return err;
 295 }
 296
 297 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
 298 {
 299         kfree(info->dir_in);
 300 }
 301
 302
 303 /*
 304  * sessions
 305  */
 306 static const char *session_state_name(int s)
 307 {
 308         switch (s) {
 309         case CEPH_MDS_SESSION_NEW: return "new";
 310         case CEPH_MDS_SESSION_OPENING: return "opening";
 311         case CEPH_MDS_SESSION_OPEN: return "open";
 312         case CEPH_MDS_SESSION_HUNG: return "hung";
 313         case CEPH_MDS_SESSION_CLOSING: return "closing";
 314         case CEPH_MDS_SESSION_RESTARTING: return "restarting";
 315         case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
 316         default: return "???";
 317         }
 318 }
 319
 320 static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
 321 {
 322         if (atomic_inc_not_zero(&s->s_ref)) {
 323                 dout("mdsc get_session %p %d -> %d\n", s,
 324                      atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
 325                 return s;
 326         } else {
 327                 dout("mdsc get_session %p 0 -- FAIL", s);
 328                 return NULL;
 329         }
 330 }
 331
 332 void ceph_put_mds_session(struct ceph_mds_session *s)
 333 {
 334         dout("mdsc put_session %p %d -> %d\n", s,
 335              atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
 336         if (atomic_dec_and_test(&s->s_ref)) {
 337                 if (s->s_authorizer)
 338                      s->s_mdsc->fsc->client->monc.auth->ops->destroy_authorizer(
 339                              s->s_mdsc->fsc->client->monc.auth,
 340                              s->s_authorizer);
 341                 kfree(s);
 342         }
 343 }
 344
 345 /*
 346  * called under mdsc->mutex
 347  */
 348 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
 349                                                    int mds)
 350 {
 351         struct ceph_mds_session *session;
 352
 353         if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
 354                 return NULL;
 355         session = mdsc->sessions[mds];
 356         dout("lookup_mds_session %p %d\n", session,
 357              atomic_read(&session->s_ref));
 358         get_session(session);
 359         return session;
 360 }
 361
 362 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
 363 {
 364         if (mds >= mdsc->max_sessions)
 365                 return false;
 366         return mdsc->sessions[mds];
 367 }
 368
 369 static int __verify_registered_session(struct ceph_mds_client *mdsc,
 370                                        struct ceph_mds_session *s)
 371 {
 372         if (s->s_mds >= mdsc->max_sessions ||
 373             mdsc->sessions[s->s_mds] != s)
 374                 return -ENOENT;
 375         return 0;
 376 }
 377
 378 /*
 379  * create+register a new session for given mds.
 380  * called under mdsc->mutex.
 381  */
 382 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
 383                                                  int mds)
 384 {
 385         struct ceph_mds_session *s;
 386
 387         s = kzalloc(sizeof(*s), GFP_NOFS);
 388         if (!s)
 389                 return ERR_PTR(-ENOMEM);
 390         s->s_mdsc = mdsc;
 391         s->s_mds = mds;
 392         s->s_state = CEPH_MDS_SESSION_NEW;
 393         s->s_ttl = 0;
 394         s->s_seq = 0;
 395         mutex_init(&s->s_mutex);
 396
 397         ceph_con_init(mdsc->fsc->client->msgr, &s->s_con);
 398         s->s_con.private = s;
 399         s->s_con.ops = &mds_con_ops;
 400         s->s_con.peer_name.type = CEPH_ENTITY_TYPE_MDS;
 401         s->s_con.peer_name.num = cpu_to_le64(mds);
 402
 403         spin_lock_init(&s->s_gen_ttl_lock);
 404         s->s_cap_gen = 0;
 405         s->s_cap_ttl = jiffies - 1;
 406
 407         spin_lock_init(&s->s_cap_lock);
 408         s->s_renew_requested = 0;
 409         s->s_renew_seq = 0;
 410         INIT_LIST_HEAD(&s->s_caps);
 411         s->s_nr_caps = 0;
 412         s->s_trim_caps = 0;
 413         atomic_set(&s->s_ref, 1);
 414         INIT_LIST_HEAD(&s->s_waiting);
 415         INIT_LIST_HEAD(&s->s_unsafe);
 416         s->s_num_cap_releases = 0;
 417         s->s_cap_iterator = NULL;
 418         INIT_LIST_HEAD(&s->s_cap_releases);
 419         INIT_LIST_HEAD(&s->s_cap_releases_done);
 420         INIT_LIST_HEAD(&s->s_cap_flushing);
 421         INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
 422
 423         dout("register_session mds%d\n", mds);
 424         if (mds >= mdsc->max_sessions) {
 425                 int newmax = 1 << get_count_order(mds+1);
 426                 struct ceph_mds_session **sa;
 427
 428                 dout("register_session realloc to %d\n", newmax);
 429                 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
 430                 if (sa == NULL)
 431                         goto fail_realloc;
 432                 if (mdsc->sessions) {
 433                         memcpy(sa, mdsc->sessions,
 434                                mdsc->max_sessions * sizeof(void *));
 435                         kfree(mdsc->sessions);
 436                 }
 437                 mdsc->sessions = sa;
 438                 mdsc->max_sessions = newmax;
 439         }
 440         mdsc->sessions[mds] = s;
 441         atomic_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
 442
 443         ceph_con_open(&s->s_con, ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
 444
 445         return s;
 446
 447 fail_realloc:
 448         kfree(s);
 449         return ERR_PTR(-ENOMEM);
 450 }
 451
 452 /*
 453  * called under mdsc->mutex
 454  */
 455 static void __unregister_session(struct ceph_mds_client *mdsc,
 456                                struct ceph_mds_session *s)
 457 {
 458         dout("__unregister_session mds%d %p\n", s->s_mds, s);
 459         BUG_ON(mdsc->sessions[s->s_mds] != s);
 460         mdsc->sessions[s->s_mds] = NULL;
 461         ceph_con_close(&s->s_con);
 462         ceph_put_mds_session(s);
 463 }
 464
 465 /*
 466  * drop session refs in request.
 467  *
 468  * should be last request ref, or hold mdsc->mutex
 469  */
 470 static void put_request_session(struct ceph_mds_request *req)
 471 {
 472         if (req->r_session) {
 473                 ceph_put_mds_session(req->r_session);
 474                 req->r_session = NULL;
 475         }
 476 }
 477
 478 void ceph_mdsc_release_request(struct kref *kref)
 479 {
 480         struct ceph_mds_request *req = container_of(kref,
 481                                                     struct ceph_mds_request,
 482                                                     r_kref);
 483         if (req->r_request)
 484                 ceph_msg_put(req->r_request);
 485         if (req->r_reply) {
 486                 ceph_msg_put(req->r_reply);
 487                 destroy_reply_info(&req->r_reply_info);
 488         }
 489         if (req->r_inode) {
 490                 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
 491                 iput(req->r_inode);
 492         }
 493         if (req->r_locked_dir)
 494                 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
 495         if (req->r_target_inode)
 496                 iput(req->r_target_inode);
 497         if (req->r_dentry)
 498                 dput(req->r_dentry);
 499         if (req->r_old_dentry) {
 500                 /*
 501                  * track (and drop pins for) r_old_dentry_dir
 502                  * separately, since r_old_dentry's d_parent may have
 503                  * changed between the dir mutex being dropped and
 504                  * this request being freed.
 505                  */
 506                 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
 507                                   CEPH_CAP_PIN);
 508                 dput(req->r_old_dentry);
 509                 iput(req->r_old_dentry_dir);
 510         }
 511         kfree(req->r_path1);
 512         kfree(req->r_path2);
 513         put_request_session(req);
 514         ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
 515         kfree(req);
 516 }
 517
 518 /*
 519  * lookup session, bump ref if found.
 520  *
 521  * called under mdsc->mutex.
 522  */
 523 static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
 524                                              u64 tid)
 525 {
 526         struct ceph_mds_request *req;
 527         struct rb_node *n = mdsc->request_tree.rb_node;
 528
 529         while (n) {
 530                 req = rb_entry(n, struct ceph_mds_request, r_node);
 531                 if (tid < req->r_tid)
 532                         n = n->rb_left;
 533                 else if (tid > req->r_tid)
 534                         n = n->rb_right;
 535                 else {
 536                         ceph_mdsc_get_request(req);
 537                         return req;
 538                 }
 539         }
 540         return NULL;
 541 }
 542
 543 static void __insert_request(struct ceph_mds_client *mdsc,
 544                              struct ceph_mds_request *new)
 545 {
 546         struct rb_node **p = &mdsc->request_tree.rb_node;
 547         struct rb_node *parent = NULL;
 548         struct ceph_mds_request *req = NULL;
 549
 550         while (*p) {
 551                 parent = *p;
 552                 req = rb_entry(parent, struct ceph_mds_request, r_node);
 553                 if (new->r_tid < req->r_tid)
 554                         p = &(*p)->rb_left;
 555                 else if (new->r_tid > req->r_tid)
 556                         p = &(*p)->rb_right;
 557                 else
 558                         BUG();
 559         }
 560
 561         rb_link_node(&new->r_node, parent, p);
 562         rb_insert_color(&new->r_node, &mdsc->request_tree);
 563 }
 564
 565 /*
 566  * Register an in-flight request, and assign a tid.  Link to directory
 567  * are modifying (if any).
 568  *
 569  * Called under mdsc->mutex.
 570  */
 571 static void __register_request(struct ceph_mds_client *mdsc,
 572                                struct ceph_mds_request *req,
 573                                struct inode *dir)
 574 {
 575         req->r_tid = ++mdsc->last_tid;
 576         if (req->r_num_caps)
 577                 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
 578                                   req->r_num_caps);
 579         dout("__register_request %p tid %lld\n", req, req->r_tid);
 580         ceph_mdsc_get_request(req);
 581         __insert_request(mdsc, req);
 582
 583         req->r_uid = current_fsuid();
 584         req->r_gid = current_fsgid();
 585
 586         if (dir) {
 587                 struct ceph_inode_info *ci = ceph_inode(dir);
 588
 589                 ihold(dir);
 590                 spin_lock(&ci->i_unsafe_lock);
 591                 req->r_unsafe_dir = dir;
 592                 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
 593                 spin_unlock(&ci->i_unsafe_lock);
 594         }
 595 }
 596
 597 static void __unregister_request(struct ceph_mds_client *mdsc,
 598                                  struct ceph_mds_request *req)
 599 {
 600         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
 601         rb_erase(&req->r_node, &mdsc->request_tree);
 602         RB_CLEAR_NODE(&req->r_node);
 603
 604         if (req->r_unsafe_dir) {
 605                 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
 606
 607                 spin_lock(&ci->i_unsafe_lock);
 608                 list_del_init(&req->r_unsafe_dir_item);
 609                 spin_unlock(&ci->i_unsafe_lock);
 610
 611                 iput(req->r_unsafe_dir);
 612                 req->r_unsafe_dir = NULL;
 613         }
 614
 615         ceph_mdsc_put_request(req);
 616 }
 617
 618 /*
 619  * Choose mds to send request to next.  If there is a hint set in the
 620  * request (e.g., due to a prior forward hint from the mds), use that.
 621  * Otherwise, consult frag tree and/or caps to identify the
 622  * appropriate mds.  If all else fails, choose randomly.
 623  *
 624  * Called under mdsc->mutex.
 625  */
 626 static struct dentry *get_nonsnap_parent(struct dentry *dentry)
 627 {
 628         /*
 629          * we don't need to worry about protecting the d_parent access
 630          * here because we never renaming inside the snapped namespace
 631          * except to resplice to another snapdir, and either the old or new
 632          * result is a valid result.
 633          */
 634         while (!IS_ROOT(dentry) && ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
 635                 dentry = dentry->d_parent;
 636         return dentry;
 637 }
 638
 639 static int __choose_mds(struct ceph_mds_client *mdsc,
 640                         struct ceph_mds_request *req)
 641 {
 642         struct inode *inode;
 643         struct ceph_inode_info *ci;
 644         struct ceph_cap *cap;
 645         int mode = req->r_direct_mode;
 646         int mds = -1;
 647         u32 hash = req->r_direct_hash;
 648         bool is_hash = req->r_direct_is_hash;
 649
 650         /*
 651          * is there a specific mds we should try?  ignore hint if we have
 652          * no session and the mds is not up (active or recovering).
 653          */
 654         if (req->r_resend_mds >= 0 &&
 655             (__have_session(mdsc, req->r_resend_mds) ||
 656              ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
 657                 dout("choose_mds using resend_mds mds%d\n",
 658                      req->r_resend_mds);
 659                 return req->r_resend_mds;
 660         }
 661
 662         if (mode == USE_RANDOM_MDS)
 663                 goto random;
 664
 665         inode = NULL;
 666         if (req->r_inode) {
 667                 inode = req->r_inode;
 668         } else if (req->r_dentry) {
 669                 /* ignore race with rename; old or new d_parent is okay */
 670                 struct dentry *parent = req->r_dentry->d_parent;
 671                 struct inode *dir = parent->d_inode;
 672
 673                 if (dir->i_sb != mdsc->fsc->sb) {
 674                         /* not this fs! */
 675                         inode = req->r_dentry->d_inode;
 676                 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
 677                         /* direct snapped/virtual snapdir requests
 678                          * based on parent dir inode */
 679                         struct dentry *dn = get_nonsnap_parent(parent);
 680                         inode = dn->d_inode;
 681                         dout("__choose_mds using nonsnap parent %p\n", inode);
 682                 } else if (req->r_dentry->d_inode) {
 683                         /* dentry target */
 684                         inode = req->r_dentry->d_inode;
 685                 } else {
 686                         /* dir + name */
 687                         inode = dir;
 688                         hash = ceph_dentry_hash(dir, req->r_dentry);
 689                         is_hash = true;
 690                 }
 691         }
 692
 693         dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
 694              (int)hash, mode);
 695         if (!inode)
 696                 goto random;
 697         ci = ceph_inode(inode);
 698
 699         if (is_hash && S_ISDIR(inode->i_mode)) {
 700                 struct ceph_inode_frag frag;
 701                 int found;
 702
 703                 ceph_choose_frag(ci, hash, &frag, &found);
 704                 if (found) {
 705                         if (mode == USE_ANY_MDS && frag.ndist > 0) {
 706                                 u8 r;
 707
 708                                 /* choose a random replica */
 709                                 get_random_bytes(&r, 1);
 710                                 r %= frag.ndist;
 711                                 mds = frag.dist[r];
 712                                 dout("choose_mds %p %llx.%llx "
 713                                      "frag %u mds%d (%d/%d)\n",
 714                                      inode, ceph_vinop(inode),
 715                                      frag.frag, mds,
 716                                      (int)r, frag.ndist);
 717                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
 718                                     CEPH_MDS_STATE_ACTIVE)
 719                                         return mds;
 720                         }
 721
 722                         /* since this file/dir wasn't known to be
 723                          * replicated, then we want to look for the
 724                          * authoritative mds. */
 725                         mode = USE_AUTH_MDS;
 726                         if (frag.mds >= 0) {
 727                                 /* choose auth mds */
 728                                 mds = frag.mds;
 729                                 dout("choose_mds %p %llx.%llx "
 730                                      "frag %u mds%d (auth)\n",
 731                                      inode, ceph_vinop(inode), frag.frag, mds);
 732                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
 733                                     CEPH_MDS_STATE_ACTIVE)
 734                                         return mds;
 735                         }
 736                 }
 737         }
 738
 739         spin_lock(&ci->i_ceph_lock);
 740         cap = NULL;
 741         if (mode == USE_AUTH_MDS)
 742                 cap = ci->i_auth_cap;
 743         if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
 744                 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
 745         if (!cap) {
 746                 spin_unlock(&ci->i_ceph_lock);
 747                 goto random;
 748         }
 749         mds = cap->session->s_mds;
 750         dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
 751              inode, ceph_vinop(inode), mds,
 752              cap == ci->i_auth_cap ? "auth " : "", cap);
 753         spin_unlock(&ci->i_ceph_lock);
 754         return mds;
 755
 756 random:
 757         mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
 758         dout("choose_mds chose random mds%d\n", mds);
 759         return mds;
 760 }
 761
 762
 763 /*
 764  * session messages
 765  */
 766 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
 767 {
 768         struct ceph_msg *msg;
 769         struct ceph_mds_session_head *h;
 770
 771         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
 772                            false);
 773         if (!msg) {
 774                 pr_err("create_session_msg ENOMEM creating msg\n");
 775                 return NULL;
 776         }
 777         h = msg->front.iov_base;
 778         h->op = cpu_to_le32(op);
 779         h->seq = cpu_to_le64(seq);
 780         return msg;
 781 }
 782
 783 /*
 784  * send session open request.
 785  *
 786  * called under mdsc->mutex
 787  */
 788 static int __open_session(struct ceph_mds_client *mdsc,
 789                           struct ceph_mds_session *session)
 790 {
 791         struct ceph_msg *msg;
 792         int mstate;
 793         int mds = session->s_mds;
 794
 795         /* wait for mds to go active? */
 796         mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
 797         dout("open_session to mds%d (%s)\n", mds,
 798              ceph_mds_state_name(mstate));
 799         session->s_state = CEPH_MDS_SESSION_OPENING;
 800         session->s_renew_requested = jiffies;
 801
 802         /* send connect message */
 803         msg = create_session_msg(CEPH_SESSION_REQUEST_OPEN, session->s_seq);
 804         if (!msg)
 805                 return -ENOMEM;
 806         ceph_con_send(&session->s_con, msg);
 807         return 0;
 808 }
 809
 810 /*
 811  * open sessions for any export targets for the given mds
 812  *
 813  * called under mdsc->mutex
 814  */
 815 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
 816                                           struct ceph_mds_session *session)
 817 {
 818         struct ceph_mds_info *mi;
 819         struct ceph_mds_session *ts;
 820         int i, mds = session->s_mds;
 821         int target;
 822
 823         if (mds >= mdsc->mdsmap->m_max_mds)
 824                 return;
 825         mi = &mdsc->mdsmap->m_info[mds];
 826         dout("open_export_target_sessions for mds%d (%d targets)\n",
 827              session->s_mds, mi->num_export_targets);
 828
 829         for (i = 0; i < mi->num_export_targets; i++) {
 830                 target = mi->export_targets[i];
 831                 ts = __ceph_lookup_mds_session(mdsc, target);
 832                 if (!ts) {
 833                         ts = register_session(mdsc, target);
 834                         if (IS_ERR(ts))
 835                                 return;
 836                 }
 837                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
 838                     session->s_state == CEPH_MDS_SESSION_CLOSING)
 839                         __open_session(mdsc, session);
 840                 else
 841                         dout(" mds%d target mds%d %p is %s\n", session->s_mds,
 842                              i, ts, session_state_name(ts->s_state));
 843                 ceph_put_mds_session(ts);
 844         }
 845 }
 846
 847 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 848                                            struct ceph_mds_session *session)
 849 {
 850         mutex_lock(&mdsc->mutex);
 851         __open_export_target_sessions(mdsc, session);
 852         mutex_unlock(&mdsc->mutex);
 853 }
 854
 855 /*
 856  * session caps
 857  */
 858
 859 /*
 860  * Free preallocated cap messages assigned to this session
 861  */
 862 static void cleanup_cap_releases(struct ceph_mds_session *session)
 863 {
 864         struct ceph_msg *msg;
 865
 866         spin_lock(&session->s_cap_lock);
 867         while (!list_empty(&session->s_cap_releases)) {
 868                 msg = list_first_entry(&session->s_cap_releases,
 869                                        struct ceph_msg, list_head);
 870                 list_del_init(&msg->list_head);
 871                 ceph_msg_put(msg);
 872         }
 873         while (!list_empty(&session->s_cap_releases_done)) {
 874                 msg = list_first_entry(&session->s_cap_releases_done,
 875                                        struct ceph_msg, list_head);
 876                 list_del_init(&msg->list_head);
 877                 ceph_msg_put(msg);
 878         }
 879         spin_unlock(&session->s_cap_lock);
 880 }
 881
 882 /*
 883  * Helper to safely iterate over all caps associated with a session, with
 884  * special care taken to handle a racing __ceph_remove_cap().
 885  *
 886  * Caller must hold session s_mutex.
 887  */
 888 static int iterate_session_caps(struct ceph_mds_session *session,
 889                                  int (*cb)(struct inode *, struct ceph_cap *,
 890                                             void *), void *arg)
 891 {
 892         struct list_head *p;
 893         struct ceph_cap *cap;
 894         struct inode *inode, *last_inode = NULL;
 895         struct ceph_cap *old_cap = NULL;
 896         int ret;
 897
 898         dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
 899         spin_lock(&session->s_cap_lock);
 900         p = session->s_caps.next;
 901         while (p != &session->s_caps) {
 902                 cap = list_entry(p, struct ceph_cap, session_caps);
 903                 inode = igrab(&cap->ci->vfs_inode);
 904                 if (!inode) {
 905                         p = p->next;
 906                         continue;
 907                 }
 908                 session->s_cap_iterator = cap;
 909                 spin_unlock(&session->s_cap_lock);
 910
 911                 if (last_inode) {
 912                         iput(last_inode);
 913                         last_inode = NULL;
 914                 }
 915                 if (old_cap) {
 916                         ceph_put_cap(session->s_mdsc, old_cap);
 917                         old_cap = NULL;
 918                 }
 919
 920                 ret = cb(inode, cap, arg);
 921                 last_inode = inode;
 922
 923                 spin_lock(&session->s_cap_lock);
 924                 p = p->next;
 925                 if (cap->ci == NULL) {
 926                         dout("iterate_session_caps  finishing cap %p removal\n",
 927                              cap);
 928                         BUG_ON(cap->session != session);
 929                         list_del_init(&cap->session_caps);
 930                         session->s_nr_caps--;
 931                         cap->session = NULL;
 932                         old_cap = cap;  /* put_cap it w/o locks held */
 933                 }
 934                 if (ret < 0)
 935                         goto out;
 936         }
 937         ret = 0;
 938 out:
 939         session->s_cap_iterator = NULL;
 940         spin_unlock(&session->s_cap_lock);
 941
 942         if (last_inode)
 943                 iput(last_inode);
 944         if (old_cap)
 945                 ceph_put_cap(session->s_mdsc, old_cap);
 946
 947         return ret;
 948 }
 949
 950 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
 951                                   void *arg)
 952 {
 953         struct ceph_inode_info *ci = ceph_inode(inode);
 954         int drop = 0;
 955
 956         dout("removing cap %p, ci is %p, inode is %p\n",
 957              cap, ci, &ci->vfs_inode);
 958         spin_lock(&ci->i_ceph_lock);
 959         __ceph_remove_cap(cap);
 960         if (!__ceph_is_any_real_caps(ci)) {
 961                 struct ceph_mds_client *mdsc =
 962                         ceph_sb_to_client(inode->i_sb)->mdsc;
 963
 964                 spin_lock(&mdsc->cap_dirty_lock);
 965                 if (!list_empty(&ci->i_dirty_item)) {
 966                         pr_info(" dropping dirty %s state for %p %lld\n",
 967                                 ceph_cap_string(ci->i_dirty_caps),
 968                                 inode, ceph_ino(inode));
 969                         ci->i_dirty_caps = 0;
 970                         list_del_init(&ci->i_dirty_item);
 971                         drop = 1;
 972                 }
 973                 if (!list_empty(&ci->i_flushing_item)) {
 974                         pr_info(" dropping dirty+flushing %s state for %p %lld\n",
 975                                 ceph_cap_string(ci->i_flushing_caps),
 976                                 inode, ceph_ino(inode));
 977                         ci->i_flushing_caps = 0;
 978                         list_del_init(&ci->i_flushing_item);
 979                         mdsc->num_cap_flushing--;
 980                         drop = 1;
 981                 }
 982                 if (drop && ci->i_wrbuffer_ref) {
 983                         pr_info(" dropping dirty data for %p %lld\n",
 984                                 inode, ceph_ino(inode));
 985                         ci->i_wrbuffer_ref = 0;
 986                         ci->i_wrbuffer_ref_head = 0;
 987                         drop++;
 988                 }
 989                 spin_unlock(&mdsc->cap_dirty_lock);
 990         }
 991         spin_unlock(&ci->i_ceph_lock);
 992         while (drop--)
 993                 iput(inode);
 994         return 0;
 995 }
 996
 997 /*
 998  * caller must hold session s_mutex
 999  */
1000 static void remove_session_caps(struct ceph_mds_session *session)
1001 {
1002         dout("remove_session_caps on %p\n", session);
1003         iterate_session_caps(session, remove_session_caps_cb, NULL);
1004         BUG_ON(session->s_nr_caps > 0);
1005         BUG_ON(!list_empty(&session->s_cap_flushing));
1006         cleanup_cap_releases(session);
1007 }
1008
1009 /*
1010  * wake up any threads waiting on this session's caps.  if the cap is
1011  * old (didn't get renewed on the client reconnect), remove it now.
1012  *
1013  * caller must hold s_mutex.
1014  */
1015 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1016                               void *arg)
1017 {
1018         struct ceph_inode_info *ci = ceph_inode(inode);
1019
1020         wake_up_all(&ci->i_cap_wq);
1021         if (arg) {
1022                 spin_lock(&ci->i_ceph_lock);
1023                 ci->i_wanted_max_size = 0;
1024                 ci->i_requested_max_size = 0;
1025                 spin_unlock(&ci->i_ceph_lock);
1026         }
1027         return 0;
1028 }
1029
1030 static void wake_up_session_caps(struct ceph_mds_session *session,
1031                                  int reconnect)
1032 {
1033         dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1034         iterate_session_caps(session, wake_up_session_cb,
1035                              (void *)(unsigned long)reconnect);
1036 }
1037
1038 /*
1039  * Send periodic message to MDS renewing all currently held caps.  The
1040  * ack will reset the expiration for all caps from this session.
1041  *
1042  * caller holds s_mutex
1043  */
1044 static int send_renew_caps(struct ceph_mds_client *mdsc,
1045                            struct ceph_mds_session *session)
1046 {
1047         struct ceph_msg *msg;
1048         int state;
1049
1050         if (time_after_eq(jiffies, session->s_cap_ttl) &&
1051             time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1052                 pr_info("mds%d caps stale\n", session->s_mds);
1053         session->s_renew_requested = jiffies;
1054
1055         /* do not try to renew caps until a recovering mds has reconnected
1056          * with its clients. */
1057         state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1058         if (state < CEPH_MDS_STATE_RECONNECT) {
1059                 dout("send_renew_caps ignoring mds%d (%s)\n",
1060                      session->s_mds, ceph_mds_state_name(state));
1061                 return 0;
1062         }
1063
1064         dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1065                 ceph_mds_state_name(state));
1066         msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1067                                  ++session->s_renew_seq);
1068         if (!msg)
1069                 return -ENOMEM;
1070         ceph_con_send(&session->s_con, msg);
1071         return 0;
1072 }
1073
1074 /*
1075  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1076  *
1077  * Called under session->s_mutex
1078  */
1079 static void renewed_caps(struct ceph_mds_client *mdsc,
1080                          struct ceph_mds_session *session, int is_renew)
1081 {
1082         int was_stale;
1083         int wake = 0;
1084
1085         spin_lock(&session->s_cap_lock);
1086         was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1087
1088         session->s_cap_ttl = session->s_renew_requested +
1089                 mdsc->mdsmap->m_session_timeout*HZ;
1090
1091         if (was_stale) {
1092                 if (time_before(jiffies, session->s_cap_ttl)) {
1093                         pr_info("mds%d caps renewed\n", session->s_mds);
1094                         wake = 1;
1095                 } else {
1096                         pr_info("mds%d caps still stale\n", session->s_mds);
1097                 }
1098         }
1099         dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1100              session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1101              time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1102         spin_unlock(&session->s_cap_lock);
1103
1104         if (wake)
1105                 wake_up_session_caps(session, 0);
1106 }
1107
1108 /*
1109  * send a session close request
1110  */
1111 static int request_close_session(struct ceph_mds_client *mdsc,
1112                                  struct ceph_mds_session *session)
1113 {
1114         struct ceph_msg *msg;
1115
1116         dout("request_close_session mds%d state %s seq %lld\n",
1117              session->s_mds, session_state_name(session->s_state),
1118              session->s_seq);
1119         msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1120         if (!msg)
1121                 return -ENOMEM;
1122         ceph_con_send(&session->s_con, msg);
1123         return 0;
1124 }
1125
1126 /*
1127  * Called with s_mutex held.
1128  */
1129 static int __close_session(struct ceph_mds_client *mdsc,
1130                          struct ceph_mds_session *session)
1131 {
1132         if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1133                 return 0;
1134         session->s_state = CEPH_MDS_SESSION_CLOSING;
1135         return request_close_session(mdsc, session);
1136 }
1137
1138 /*
1139  * Trim old(er) caps.
1140  *
1141  * Because we can't cache an inode without one or more caps, we do
1142  * this indirectly: if a cap is unused, we prune its aliases, at which
1143  * point the inode will hopefully get dropped to.
1144  *
1145  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1146  * memory pressure from the MDS, though, so it needn't be perfect.
1147  */
1148 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1149 {
1150         struct ceph_mds_session *session = arg;
1151         struct ceph_inode_info *ci = ceph_inode(inode);
1152         int used, oissued, mine;
1153
1154         if (session->s_trim_caps <= 0)
1155                 return -1;
1156
1157         spin_lock(&ci->i_ceph_lock);
1158         mine = cap->issued | cap->implemented;
1159         used = __ceph_caps_used(ci);
1160         oissued = __ceph_caps_issued_other(ci, cap);
1161
1162         dout("trim_caps_cb %p cap %p mine %s oissued %s used %s\n",
1163              inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1164              ceph_cap_string(used));
1165         if (ci->i_dirty_caps)
1166                 goto out;   /* dirty caps */
1167         if ((used & ~oissued) & mine)
1168                 goto out;   /* we need these caps */
1169
1170         session->s_trim_caps--;
1171         if (oissued) {
1172                 /* we aren't the only cap.. just remove us */
1173                 __ceph_remove_cap(cap);
1174         } else {
1175                 /* try to drop referring dentries */
1176                 spin_unlock(&ci->i_ceph_lock);
1177                 d_prune_aliases(inode);
1178                 dout("trim_caps_cb %p cap %p  pruned, count now %d\n",
1179                      inode, cap, atomic_read(&inode->i_count));
1180                 return 0;
1181         }
1182
1183 out:
1184         spin_unlock(&ci->i_ceph_lock);
1185         return 0;
1186 }
1187
1188 /*
1189  * Trim session cap count down to some max number.
1190  */
1191 static int trim_caps(struct ceph_mds_client *mdsc,
1192                      struct ceph_mds_session *session,
1193                      int max_caps)
1194 {
1195         int trim_caps = session->s_nr_caps - max_caps;
1196
1197         dout("trim_caps mds%d start: %d / %d, trim %d\n",
1198              session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1199         if (trim_caps > 0) {
1200                 session->s_trim_caps = trim_caps;
1201                 iterate_session_caps(session, trim_caps_cb, session);
1202                 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1203                      session->s_mds, session->s_nr_caps, max_caps,
1204                         trim_caps - session->s_trim_caps);
1205                 session->s_trim_caps = 0;
1206         }
1207         return 0;
1208 }
1209
1210 /*
1211  * Allocate cap_release messages.  If there is a partially full message
1212  * in the queue, try to allocate enough to cover it's remainder, so that
1213  * we can send it immediately.
1214  *
1215  * Called under s_mutex.
1216  */
1217 int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
1218                           struct ceph_mds_session *session)
1219 {
1220         struct ceph_msg *msg, *partial = NULL;
1221         struct ceph_mds_cap_release *head;
1222         int err = -ENOMEM;
1223         int extra = mdsc->fsc->mount_options->cap_release_safety;
1224         int num;
1225
1226         dout("add_cap_releases %p mds%d extra %d\n", session, session->s_mds,
1227              extra);
1228
1229         spin_lock(&session->s_cap_lock);
1230
1231         if (!list_empty(&session->s_cap_releases)) {
1232                 msg = list_first_entry(&session->s_cap_releases,
1233                                        struct ceph_msg,
1234                                  list_head);
1235                 head = msg->front.iov_base;
1236                 num = le32_to_cpu(head->num);
1237                 if (num) {
1238                         dout(" partial %p with (%d/%d)\n", msg, num,
1239                              (int)CEPH_CAPS_PER_RELEASE);
1240                         extra += CEPH_CAPS_PER_RELEASE - num;
1241                         partial = msg;
1242                 }
1243         }
1244         while (session->s_num_cap_releases < session->s_nr_caps + extra) {
1245                 spin_unlock(&session->s_cap_lock);
1246                 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
1247                                    GFP_NOFS, false);
1248                 if (!msg)
1249                         goto out_unlocked;
1250                 dout("add_cap_releases %p msg %p now %d\n", session, msg,
1251                      (int)msg->front.iov_len);
1252                 head = msg->front.iov_base;
1253                 head->num = cpu_to_le32(0);
1254                 msg->front.iov_len = sizeof(*head);
1255                 spin_lock(&session->s_cap_lock);
1256                 list_add(&msg->list_head, &session->s_cap_releases);
1257                 session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE;
1258         }
1259
1260         if (partial) {
1261                 head = partial->front.iov_base;
1262                 num = le32_to_cpu(head->num);
1263                 dout(" queueing partial %p with %d/%d\n", partial, num,
1264                      (int)CEPH_CAPS_PER_RELEASE);
1265                 list_move_tail(&partial->list_head,
1266                                &session->s_cap_releases_done);
1267                 session->s_num_cap_releases -= CEPH_CAPS_PER_RELEASE - num;
1268         }
1269         err = 0;
1270         spin_unlock(&session->s_cap_lock);
1271 out_unlocked:
1272         return err;
1273 }
1274
1275 /*
1276  * flush all dirty inode data to disk.
1277  *
1278  * returns true if we've flushed through want_flush_seq
1279  */
1280 static int check_cap_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq)
1281 {
1282         int mds, ret = 1;
1283
1284         dout("check_cap_flush want %lld\n", want_flush_seq);
1285         mutex_lock(&mdsc->mutex);
1286         for (mds = 0; ret && mds < mdsc->max_sessions; mds++) {
1287                 struct ceph_mds_session *session = mdsc->sessions[mds];
1288
1289                 if (!session)
1290                         continue;
1291                 get_session(session);
1292                 mutex_unlock(&mdsc->mutex);
1293
1294                 mutex_lock(&session->s_mutex);
1295                 if (!list_empty(&session->s_cap_flushing)) {
1296                         struct ceph_inode_info *ci =
1297                                 list_entry(session->s_cap_flushing.next,
1298                                            struct ceph_inode_info,
1299                                            i_flushing_item);
1300                         struct inode *inode = &ci->vfs_inode;
1301
1302                         spin_lock(&ci->i_ceph_lock);
1303                         if (ci->i_cap_flush_seq <= want_flush_seq) {
1304                                 dout("check_cap_flush still flushing %p "
1305                                      "seq %lld <= %lld to mds%d\n", inode,
1306                                      ci->i_cap_flush_seq, want_flush_seq,
1307                                      session->s_mds);
1308                                 ret = 0;
1309                         }
1310                         spin_unlock(&ci->i_ceph_lock);
1311                 }
1312                 mutex_unlock(&session->s_mutex);
1313                 ceph_put_mds_session(session);
1314
1315                 if (!ret)
1316                         return ret;
1317                 mutex_lock(&mdsc->mutex);
1318         }
1319
1320         mutex_unlock(&mdsc->mutex);
1321         dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
1322         return ret;
1323 }
1324
1325 /*
1326  * called under s_mutex
1327  */
1328 void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1329                             struct ceph_mds_session *session)
1330 {
1331         struct ceph_msg *msg;
1332
1333         dout("send_cap_releases mds%d\n", session->s_mds);
1334         spin_lock(&session->s_cap_lock);
1335         while (!list_empty(&session->s_cap_releases_done)) {
1336                 msg = list_first_entry(&session->s_cap_releases_done,
1337                                  struct ceph_msg, list_head);
1338                 list_del_init(&msg->list_head);
1339                 spin_unlock(&session->s_cap_lock);
1340                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1341                 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1342                 ceph_con_send(&session->s_con, msg);
1343                 spin_lock(&session->s_cap_lock);
1344         }
1345         spin_unlock(&session->s_cap_lock);
1346 }
1347
1348 static void discard_cap_releases(struct ceph_mds_client *mdsc,
1349                                  struct ceph_mds_session *session)
1350 {
1351         struct ceph_msg *msg;
1352         struct ceph_mds_cap_release *head;
1353         unsigned num;
1354
1355         dout("discard_cap_releases mds%d\n", session->s_mds);
1356         spin_lock(&session->s_cap_lock);
1357
1358         /* zero out the in-progress message */
1359         msg = list_first_entry(&session->s_cap_releases,
1360                                struct ceph_msg, list_head);
1361         head = msg->front.iov_base;
1362         num = le32_to_cpu(head->num);
1363         dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg, num);
1364         head->num = cpu_to_le32(0);
1365         session->s_num_cap_releases += num;
1366
1367         /* requeue completed messages */
1368         while (!list_empty(&session->s_cap_releases_done)) {
1369                 msg = list_first_entry(&session->s_cap_releases_done,
1370                                  struct ceph_msg, list_head);
1371                 list_del_init(&msg->list_head);
1372
1373                 head = msg->front.iov_base;
1374                 num = le32_to_cpu(head->num);
1375                 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg,
1376                      num);
1377                 session->s_num_cap_releases += num;
1378                 head->num = cpu_to_le32(0);
1379                 msg->front.iov_len = sizeof(*head);
1380                 list_add(&msg->list_head, &session->s_cap_releases);
1381         }
1382
1383         spin_unlock(&session->s_cap_lock);
1384 }
1385
1386 /*
1387  * requests
1388  */
1389
1390 /*
1391  * Create an mds request.
1392  */
1393 struct ceph_mds_request *
1394 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1395 {
1396         struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1397
1398         if (!req)
1399                 return ERR_PTR(-ENOMEM);
1400
1401         mutex_init(&req->r_fill_mutex);
1402         req->r_mdsc = mdsc;
1403         req->r_started = jiffies;
1404         req->r_resend_mds = -1;
1405         INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1406         req->r_fmode = -1;
1407         kref_init(&req->r_kref);
1408         INIT_LIST_HEAD(&req->r_wait);
1409         init_completion(&req->r_completion);
1410         init_completion(&req->r_safe_completion);
1411         INIT_LIST_HEAD(&req->r_unsafe_item);
1412
1413         req->r_op = op;
1414         req->r_direct_mode = mode;
1415         return req;
1416 }
1417
1418 /*
1419  * return oldest (lowest) request, tid in request tree, 0 if none.
1420  *
1421  * called under mdsc->mutex.
1422  */
1423 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1424 {
1425         if (RB_EMPTY_ROOT(&mdsc->request_tree))
1426                 return NULL;
1427         return rb_entry(rb_first(&mdsc->request_tree),
1428                         struct ceph_mds_request, r_node);
1429 }
1430
1431 static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1432 {
1433         struct ceph_mds_request *req = __get_oldest_req(mdsc);
1434
1435         if (req)
1436                 return req->r_tid;
1437         return 0;
1438 }
1439
1440 /*
1441  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
1442  * on build_path_from_dentry in fs/cifs/dir.c.
1443  *
1444  * If @stop_on_nosnap, generate path relative to the first non-snapped
1445  * inode.
1446  *
1447  * Encode hidden .snap dirs as a double /, i.e.
1448  *   foo/.snap/bar -> foo//bar
1449  */
1450 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1451                            int stop_on_nosnap)
1452 {
1453         struct dentry *temp;
1454         char *path;
1455         int len, pos;
1456         unsigned seq;
1457
1458         if (dentry == NULL)
1459                 return ERR_PTR(-EINVAL);
1460
1461 retry:
1462         len = 0;
1463         seq = read_seqbegin(&rename_lock);
1464         rcu_read_lock();
1465         for (temp = dentry; !IS_ROOT(temp);) {
1466                 struct inode *inode = temp->d_inode;
1467                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1468                         len++;  /* slash only */
1469                 else if (stop_on_nosnap && inode &&
1470                          ceph_snap(inode) == CEPH_NOSNAP)
1471                         break;
1472                 else
1473                         len += 1 + temp->d_name.len;
1474                 temp = temp->d_parent;
1475                 if (temp == NULL) {
1476                         rcu_read_unlock();
1477                         pr_err("build_path corrupt dentry %p\n", dentry);
1478                         return ERR_PTR(-EINVAL);
1479                 }
1480         }
1481         rcu_read_unlock();
1482         if (len)
1483                 len--;  /* no leading '/' */
1484
1485         path = kmalloc(len+1, GFP_NOFS);
1486         if (path == NULL)
1487                 return ERR_PTR(-ENOMEM);
1488         pos = len;
1489         path[pos] = 0;  /* trailing null */
1490         rcu_read_lock();
1491         for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1492                 struct inode *inode;
1493
1494                 spin_lock(&temp->d_lock);
1495                 inode = temp->d_inode;
1496                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1497                         dout("build_path path+%d: %p SNAPDIR\n",
1498                              pos, temp);
1499                 } else if (stop_on_nosnap && inode &&
1500                            ceph_snap(inode) == CEPH_NOSNAP) {
1501                         spin_unlock(&temp->d_lock);
1502                         break;
1503                 } else {
1504                         pos -= temp->d_name.len;
1505                         if (pos < 0) {
1506                                 spin_unlock(&temp->d_lock);
1507                                 break;
1508                         }
1509                         strncpy(path + pos, temp->d_name.name,
1510                                 temp->d_name.len);
1511                 }
1512                 spin_unlock(&temp->d_lock);
1513                 if (pos)
1514                         path[--pos] = '/';
1515                 temp = temp->d_parent;
1516                 if (temp == NULL) {
1517                         rcu_read_unlock();
1518                         pr_err("build_path corrupt dentry\n");
1519                         kfree(path);
1520                         return ERR_PTR(-EINVAL);
1521                 }
1522         }
1523         rcu_read_unlock();
1524         if (pos != 0 || read_seqretry(&rename_lock, seq)) {
1525                 pr_err("build_path did not end path lookup where "
1526                        "expected, namelen is %d, pos is %d\n", len, pos);
1527                 /* presumably this is only possible if racing with a
1528                    rename of one of the parent directories (we can not
1529                    lock the dentries above us to prevent this, but
1530                    retrying should be harmless) */
1531                 kfree(path);
1532                 goto retry;
1533         }
1534
1535         *base = ceph_ino(temp->d_inode);
1536         *plen = len;
1537         dout("build_path on %p %d built %llx '%.*s'\n",
1538              dentry, dentry->d_count, *base, len, path);
1539         return path;
1540 }
1541
1542 static int build_dentry_path(struct dentry *dentry,
1543                              const char **ppath, int *ppathlen, u64 *pino,
1544                              int *pfreepath)
1545 {
1546         char *path;
1547
1548         if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) {
1549                 *pino = ceph_ino(dentry->d_parent->d_inode);
1550                 *ppath = dentry->d_name.name;
1551                 *ppathlen = dentry->d_name.len;
1552                 return 0;
1553         }
1554         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1555         if (IS_ERR(path))
1556                 return PTR_ERR(path);
1557         *ppath = path;
1558         *pfreepath = 1;
1559         return 0;
1560 }
1561
1562 static int build_inode_path(struct inode *inode,
1563                             const char **ppath, int *ppathlen, u64 *pino,
1564                             int *pfreepath)
1565 {
1566         struct dentry *dentry;
1567         char *path;
1568
1569         if (ceph_snap(inode) == CEPH_NOSNAP) {
1570                 *pino = ceph_ino(inode);
1571                 *ppathlen = 0;
1572                 return 0;
1573         }
1574         dentry = d_find_alias(inode);
1575         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1576         dput(dentry);
1577         if (IS_ERR(path))
1578                 return PTR_ERR(path);
1579         *ppath = path;
1580         *pfreepath = 1;
1581         return 0;
1582 }
1583
1584 /*
1585  * request arguments may be specified via an inode *, a dentry *, or
1586  * an explicit ino+path.
1587  */
1588 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1589                                   const char *rpath, u64 rino,
1590                                   const char **ppath, int *pathlen,
1591                                   u64 *ino, int *freepath)
1592 {
1593         int r = 0;
1594
1595         if (rinode) {
1596                 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1597                 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1598                      ceph_snap(rinode));
1599         } else if (rdentry) {
1600                 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1601                 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1602                      *ppath);
1603         } else if (rpath || rino) {
1604                 *ino = rino;
1605                 *ppath = rpath;
1606                 *pathlen = strlen(rpath);
1607                 dout(" path %.*s\n", *pathlen, rpath);
1608         }
1609
1610         return r;
1611 }
1612
1613 /*
1614  * called under mdsc->mutex
1615  */
1616 static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1617                                                struct ceph_mds_request *req,
1618                                                int mds)
1619 {
1620         struct ceph_msg *msg;
1621         struct ceph_mds_request_head *head;
1622         const char *path1 = NULL;
1623         const char *path2 = NULL;
1624         u64 ino1 = 0, ino2 = 0;
1625         int pathlen1 = 0, pathlen2 = 0;
1626         int freepath1 = 0, freepath2 = 0;
1627         int len;
1628         u16 releases;
1629         void *p, *end;
1630         int ret;
1631
1632         ret = set_request_path_attr(req->r_inode, req->r_dentry,
1633                               req->r_path1, req->r_ino1.ino,
1634                               &path1, &pathlen1, &ino1, &freepath1);
1635         if (ret < 0) {
1636                 msg = ERR_PTR(ret);
1637                 goto out;
1638         }
1639
1640         ret = set_request_path_attr(NULL, req->r_old_dentry,
1641                               req->r_path2, req->r_ino2.ino,
1642                               &path2, &pathlen2, &ino2, &freepath2);
1643         if (ret < 0) {
1644                 msg = ERR_PTR(ret);
1645                 goto out_free1;
1646         }
1647
1648         len = sizeof(*head) +
1649                 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64));
1650
1651         /* calculate (max) length for cap releases */
1652         len += sizeof(struct ceph_mds_request_release) *
1653                 (!!req->r_inode_drop + !!req->r_dentry_drop +
1654                  !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1655         if (req->r_dentry_drop)
1656                 len += req->r_dentry->d_name.len;
1657         if (req->r_old_dentry_drop)
1658                 len += req->r_old_dentry->d_name.len;
1659
1660         msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
1661         if (!msg) {
1662                 msg = ERR_PTR(-ENOMEM);
1663                 goto out_free2;
1664         }
1665
1666         msg->hdr.tid = cpu_to_le64(req->r_tid);
1667
1668         head = msg->front.iov_base;
1669         p = msg->front.iov_base + sizeof(*head);
1670         end = msg->front.iov_base + msg->front.iov_len;
1671
1672         head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1673         head->op = cpu_to_le32(req->r_op);
1674         head->caller_uid = cpu_to_le32(req->r_uid);
1675         head->caller_gid = cpu_to_le32(req->r_gid);
1676         head->args = req->r_args;
1677
1678         ceph_encode_filepath(&p, end, ino1, path1);
1679         ceph_encode_filepath(&p, end, ino2, path2);
1680
1681         /* make note of release offset, in case we need to replay */
1682         req->r_request_release_offset = p - msg->front.iov_base;
1683
1684         /* cap releases */
1685         releases = 0;
1686         if (req->r_inode_drop)
1687                 releases += ceph_encode_inode_release(&p,
1688                       req->r_inode ? req->r_inode : req->r_dentry->d_inode,
1689                       mds, req->r_inode_drop, req->r_inode_unless, 0);
1690         if (req->r_dentry_drop)
1691                 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1692                        mds, req->r_dentry_drop, req->r_dentry_unless);
1693         if (req->r_old_dentry_drop)
1694                 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1695                        mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1696         if (req->r_old_inode_drop)
1697                 releases += ceph_encode_inode_release(&p,
1698                       req->r_old_dentry->d_inode,
1699                       mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1700         head->num_releases = cpu_to_le16(releases);
1701
1702         BUG_ON(p > end);
1703         msg->front.iov_len = p - msg->front.iov_base;
1704         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1705
1706         msg->pages = req->r_pages;
1707         msg->nr_pages = req->r_num_pages;
1708         msg->hdr.data_len = cpu_to_le32(req->r_data_len);
1709         msg->hdr.data_off = cpu_to_le16(0);
1710
1711 out_free2:
1712         if (freepath2)
1713                 kfree((char *)path2);
1714 out_free1:
1715         if (freepath1)
1716                 kfree((char *)path1);
1717 out:
1718         return msg;
1719 }
1720
1721 /*
1722  * called under mdsc->mutex if error, under no mutex if
1723  * success.
1724  */
1725 static void complete_request(struct ceph_mds_client *mdsc,
1726                              struct ceph_mds_request *req)
1727 {
1728         if (req->r_callback)
1729                 req->r_callback(mdsc, req);
1730         else
1731                 complete_all(&req->r_completion);
1732 }
1733
1734 /*
1735  * called under mdsc->mutex
1736  */
1737 static int __prepare_send_request(struct ceph_mds_client *mdsc,
1738                                   struct ceph_mds_request *req,
1739                                   int mds)
1740 {
1741         struct ceph_mds_request_head *rhead;
1742         struct ceph_msg *msg;
1743         int flags = 0;
1744
1745         req->r_attempts++;
1746         if (req->r_inode) {
1747                 struct ceph_cap *cap =
1748                         ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
1749
1750                 if (cap)
1751                         req->r_sent_on_mseq = cap->mseq;
1752                 else
1753                         req->r_sent_on_mseq = -1;
1754         }
1755         dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
1756              req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
1757
1758         if (req->r_got_unsafe) {
1759                 /*
1760                  * Replay.  Do not regenerate message (and rebuild
1761                  * paths, etc.); just use the original message.
1762                  * Rebuilding paths will break for renames because
1763                  * d_move mangles the src name.
1764                  */
1765                 msg = req->r_request;
1766                 rhead = msg->front.iov_base;
1767
1768                 flags = le32_to_cpu(rhead->flags);
1769                 flags |= CEPH_MDS_FLAG_REPLAY;
1770                 rhead->flags = cpu_to_le32(flags);
1771
1772                 if (req->r_target_inode)
1773                         rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
1774
1775                 rhead->num_retry = req->r_attempts - 1;
1776
1777                 /* remove cap/dentry releases from message */
1778                 rhead->num_releases = 0;
1779                 msg->hdr.front_len = cpu_to_le32(req->r_request_release_offset);
1780                 msg->front.iov_len = req->r_request_release_offset;
1781                 return 0;
1782         }
1783
1784         if (req->r_request) {
1785                 ceph_msg_put(req->r_request);
1786                 req->r_request = NULL;
1787         }
1788         msg = create_request_message(mdsc, req, mds);
1789         if (IS_ERR(msg)) {
1790                 req->r_err = PTR_ERR(msg);
1791                 complete_request(mdsc, req);
1792                 return PTR_ERR(msg);
1793         }
1794         req->r_request = msg;
1795
1796         rhead = msg->front.iov_base;
1797         rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
1798         if (req->r_got_unsafe)
1799                 flags |= CEPH_MDS_FLAG_REPLAY;
1800         if (req->r_locked_dir)
1801                 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
1802         rhead->flags = cpu_to_le32(flags);
1803         rhead->num_fwd = req->r_num_fwd;
1804         rhead->num_retry = req->r_attempts - 1;
1805         rhead->ino = 0;
1806
1807         dout(" r_locked_dir = %p\n", req->r_locked_dir);
1808         return 0;
1809 }
1810
1811 /*
1812  * send request, or put it on the appropriate wait list.
1813  */
1814 static int __do_request(struct ceph_mds_client *mdsc,
1815                         struct ceph_mds_request *req)
1816 {
1817         struct ceph_mds_session *session = NULL;
1818         int mds = -1;
1819         int err = -EAGAIN;
1820
1821         if (req->r_err || req->r_got_result)
1822                 goto out;
1823
1824         if (req->r_timeout &&
1825             time_after_eq(jiffies, req->r_started + req->r_timeout)) {
1826                 dout("do_request timed out\n");
1827                 err = -EIO;
1828                 goto finish;
1829         }
1830
1831         put_request_session(req);
1832
1833         mds = __choose_mds(mdsc, req);
1834         if (mds < 0 ||
1835             ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
1836                 dout("do_request no mds or not active, waiting for map\n");
1837                 list_add(&req->r_wait, &mdsc->waiting_for_map);
1838                 goto out;
1839         }
1840
1841         /* get, open session */
1842         session = __ceph_lookup_mds_session(mdsc, mds);
1843         if (!session) {
1844                 session = register_session(mdsc, mds);
1845                 if (IS_ERR(session)) {
1846                         err = PTR_ERR(session);
1847                         goto finish;
1848                 }
1849         }
1850         req->r_session = get_session(session);
1851
1852         dout("do_request mds%d session %p state %s\n", mds, session,
1853              session_state_name(session->s_state));
1854         if (session->s_state != CEPH_MDS_SESSION_OPEN &&
1855             session->s_state != CEPH_MDS_SESSION_HUNG) {
1856                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1857                     session->s_state == CEPH_MDS_SESSION_CLOSING)
1858                         __open_session(mdsc, session);
1859                 list_add(&req->r_wait, &session->s_waiting);
1860                 goto out_session;
1861         }
1862
1863         /* send request */
1864         req->r_resend_mds = -1;   /* forget any previous mds hint */
1865
1866         if (req->r_request_started == 0)   /* note request start time */
1867                 req->r_request_started = jiffies;
1868
1869         err = __prepare_send_request(mdsc, req, mds);
1870         if (!err) {
1871                 ceph_msg_get(req->r_request);
1872                 ceph_con_send(&session->s_con, req->r_request);
1873         }
1874
1875 out_session:
1876         ceph_put_mds_session(session);
1877 out:
1878         return err;
1879
1880 finish:
1881         req->r_err = err;
1882         complete_request(mdsc, req);
1883         goto out;
1884 }
1885
1886 /*
1887  * called under mdsc->mutex
1888  */
1889 static void __wake_requests(struct ceph_mds_client *mdsc,
1890                             struct list_head *head)
1891 {
1892         struct ceph_mds_request *req, *nreq;
1893
1894         list_for_each_entry_safe(req, nreq, head, r_wait) {
1895                 list_del_init(&req->r_wait);
1896                 __do_request(mdsc, req);
1897         }
1898 }
1899
1900 /*
1901  * Wake up threads with requests pending for @mds, so that they can
1902  * resubmit their requests to a possibly different mds.
1903  */
1904 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
1905 {
1906         struct ceph_mds_request *req;
1907         struct rb_node *p;
1908
1909         dout("kick_requests mds%d\n", mds);
1910         for (p = rb_first(&mdsc->request_tree); p; p = rb_next(p)) {
1911                 req = rb_entry(p, struct ceph_mds_request, r_node);
1912                 if (req->r_got_unsafe)
1913                         continue;
1914                 if (req->r_session &&
1915                     req->r_session->s_mds == mds) {
1916                         dout(" kicking tid %llu\n", req->r_tid);
1917                         __do_request(mdsc, req);
1918                 }
1919         }
1920 }
1921
1922 void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
1923                               struct ceph_mds_request *req)
1924 {
1925         dout("submit_request on %p\n", req);
1926         mutex_lock(&mdsc->mutex);
1927         __register_request(mdsc, req, NULL);
1928         __do_request(mdsc, req);
1929         mutex_unlock(&mdsc->mutex);
1930 }
1931
1932 /*
1933  * Synchrously perform an mds request.  Take care of all of the
1934  * session setup, forwarding, retry details.
1935  */
1936 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
1937                          struct inode *dir,
1938                          struct ceph_mds_request *req)
1939 {
1940         int err;
1941
1942         dout("do_request on %p\n", req);
1943
1944         /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
1945         if (req->r_inode)
1946                 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
1947         if (req->r_locked_dir)
1948                 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
1949         if (req->r_old_dentry)
1950                 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
1951                                   CEPH_CAP_PIN);
1952
1953         /* issue */
1954         mutex_lock(&mdsc->mutex);
1955         __register_request(mdsc, req, dir);
1956         __do_request(mdsc, req);
1957
1958         if (req->r_err) {
1959                 err = req->r_err;
1960                 __unregister_request(mdsc, req);
1961                 dout("do_request early error %d\n", err);
1962                 goto out;
1963         }
1964
1965         /* wait */
1966         mutex_unlock(&mdsc->mutex);
1967         dout("do_request waiting\n");
1968         if (req->r_timeout) {
1969                 err = (long)wait_for_completion_killable_timeout(
1970                         &req->r_completion, req->r_timeout);
1971                 if (err == 0)
1972                         err = -EIO;
1973         } else {
1974                 err = wait_for_completion_killable(&req->r_completion);
1975         }
1976         dout("do_request waited, got %d\n", err);
1977         mutex_lock(&mdsc->mutex);
1978
1979         /* only abort if we didn't race with a real reply */
1980         if (req->r_got_result) {
1981                 err = le32_to_cpu(req->r_reply_info.head->result);
1982         } else if (err < 0) {
1983                 dout("aborted request %lld with %d\n", req->r_tid, err);
1984
1985                 /*
1986                  * ensure we aren't running concurrently with
1987                  * ceph_fill_trace or ceph_readdir_prepopulate, which
1988                  * rely on locks (dir mutex) held by our caller.
1989                  */
1990                 mutex_lock(&req->r_fill_mutex);
1991                 req->r_err = err;
1992                 req->r_aborted = true;
1993                 mutex_unlock(&req->r_fill_mutex);
1994
1995                 if (req->r_locked_dir &&
1996                     (req->r_op & CEPH_MDS_OP_WRITE))
1997                         ceph_invalidate_dir_request(req);
1998         } else {
1999                 err = req->r_err;
2000         }
2001
2002 out:
2003         mutex_unlock(&mdsc->mutex);
2004         dout("do_request %p done, result %d\n", req, err);
2005         return err;
2006 }
2007
2008 /*
2009  * Invalidate dir D_COMPLETE, dentry lease state on an aborted MDS
2010  * namespace request.
2011  */
2012 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2013 {
2014         struct inode *inode = req->r_locked_dir;
2015         struct ceph_inode_info *ci = ceph_inode(inode);
2016
2017         dout("invalidate_dir_request %p (D_COMPLETE, lease(s))\n", inode);
2018         spin_lock(&ci->i_ceph_lock);
2019         ceph_dir_clear_complete(inode);
2020         ci->i_release_count++;
2021         spin_unlock(&ci->i_ceph_lock);
2022
2023         if (req->r_dentry)
2024                 ceph_invalidate_dentry_lease(req->r_dentry);
2025         if (req->r_old_dentry)
2026                 ceph_invalidate_dentry_lease(req->r_old_dentry);
2027 }
2028
2029 /*
2030  * Handle mds reply.
2031  *
2032  * We take the session mutex and parse and process the reply immediately.
2033  * This preserves the logical ordering of replies, capabilities, etc., sent
2034  * by the MDS as they are applied to our local cache.
2035  */
2036 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2037 {
2038         struct ceph_mds_client *mdsc = session->s_mdsc;
2039         struct ceph_mds_request *req;
2040         struct ceph_mds_reply_head *head = msg->front.iov_base;
2041         struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
2042         u64 tid;
2043         int err, result;
2044         int mds = session->s_mds;
2045
2046         if (msg->front.iov_len < sizeof(*head)) {
2047                 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2048                 ceph_msg_dump(msg);
2049                 return;
2050         }
2051
2052         /* get request, session */
2053         tid = le64_to_cpu(msg->hdr.tid);
2054         mutex_lock(&mdsc->mutex);
2055         req = __lookup_request(mdsc, tid);
2056         if (!req) {
2057                 dout("handle_reply on unknown tid %llu\n", tid);
2058                 mutex_unlock(&mdsc->mutex);
2059                 return;
2060         }
2061         dout("handle_reply %p\n", req);
2062
2063         /* correct session? */
2064         if (req->r_session != session) {
2065                 pr_err("mdsc_handle_reply got %llu on session mds%d"
2066                        " not mds%d\n", tid, session->s_mds,
2067                        req->r_session ? req->r_session->s_mds : -1);
2068                 mutex_unlock(&mdsc->mutex);
2069                 goto out;
2070         }
2071
2072         /* dup? */
2073         if ((req->r_got_unsafe && !head->safe) ||
2074             (req->r_got_safe && head->safe)) {
2075                 pr_warning("got a dup %s reply on %llu from mds%d\n",
2076                            head->safe ? "safe" : "unsafe", tid, mds);
2077                 mutex_unlock(&mdsc->mutex);
2078                 goto out;
2079         }
2080         if (req->r_got_safe && !head->safe) {
2081                 pr_warning("got unsafe after safe on %llu from mds%d\n",
2082                            tid, mds);
2083                 mutex_unlock(&mdsc->mutex);
2084                 goto out;
2085         }
2086
2087         result = le32_to_cpu(head->result);
2088
2089         /*
2090          * Handle an ESTALE
2091          * if we're not talking to the authority, send to them
2092          * if the authority has changed while we weren't looking,
2093          * send to new authority
2094          * Otherwise we just have to return an ESTALE
2095          */
2096         if (result == -ESTALE) {
2097                 dout("got ESTALE on request %llu", req->r_tid);
2098                 if (!req->r_inode) {
2099                         /* do nothing; not an authority problem */
2100                 } else if (req->r_direct_mode != USE_AUTH_MDS) {
2101                         dout("not using auth, setting for that now");
2102                         req->r_direct_mode = USE_AUTH_MDS;
2103                         __do_request(mdsc, req);
2104                         mutex_unlock(&mdsc->mutex);
2105                         goto out;
2106                 } else  {
2107                         struct ceph_inode_info *ci = ceph_inode(req->r_inode);
2108                         struct ceph_cap *cap = NULL;
2109
2110                         if (req->r_session)
2111                                 cap = ceph_get_cap_for_mds(ci,
2112                                                    req->r_session->s_mds);
2113
2114                         dout("already using auth");
2115                         if ((!cap || cap != ci->i_auth_cap) ||
2116                             (cap->mseq != req->r_sent_on_mseq)) {
2117                                 dout("but cap changed, so resending");
2118                                 __do_request(mdsc, req);
2119                                 mutex_unlock(&mdsc->mutex);
2120                                 goto out;
2121                         }
2122                 }
2123                 dout("have to return ESTALE on request %llu", req->r_tid);
2124         }
2125
2126
2127         if (head->safe) {
2128                 req->r_got_safe = true;
2129                 __unregister_request(mdsc, req);
2130                 complete_all(&req->r_safe_completion);
2131
2132                 if (req->r_got_unsafe) {
2133                         /*
2134                          * We already handled the unsafe response, now do the
2135                          * cleanup.  No need to examine the response; the MDS
2136                          * doesn't include any result info in the safe
2137                          * response.  And even if it did, there is nothing
2138                          * useful we could do with a revised return value.
2139                          */
2140                         dout("got safe reply %llu, mds%d\n", tid, mds);
2141                         list_del_init(&req->r_unsafe_item);
2142
2143                         /* last unsafe request during umount? */
2144                         if (mdsc->stopping && !__get_oldest_req(mdsc))
2145                                 complete_all(&mdsc->safe_umount_waiters);
2146                         mutex_unlock(&mdsc->mutex);
2147                         goto out;
2148                 }
2149         } else {
2150                 req->r_got_unsafe = true;
2151                 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2152         }
2153
2154         dout("handle_reply tid %lld result %d\n", tid, result);
2155         rinfo = &req->r_reply_info;
2156         err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2157         mutex_unlock(&mdsc->mutex);
2158
2159         mutex_lock(&session->s_mutex);
2160         if (err < 0) {
2161                 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
2162                 ceph_msg_dump(msg);
2163                 goto out_err;
2164         }
2165
2166         /* snap trace */
2167         if (rinfo->snapblob_len) {
2168                 down_write(&mdsc->snap_rwsem);
2169                 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2170                                rinfo->snapblob + rinfo->snapblob_len,
2171                                le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP);
2172                 downgrade_write(&mdsc->snap_rwsem);
2173         } else {
2174                 down_read(&mdsc->snap_rwsem);
2175         }
2176
2177         /* insert trace into our cache */
2178         mutex_lock(&req->r_fill_mutex);
2179         err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2180         if (err == 0) {
2181                 if (result == 0 && req->r_op != CEPH_MDS_OP_GETFILELOCK &&
2182                     rinfo->dir_nr)
2183                         ceph_readdir_prepopulate(req, req->r_session);
2184                 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2185         }
2186         mutex_unlock(&req->r_fill_mutex);
2187
2188         up_read(&mdsc->snap_rwsem);
2189 out_err:
2190         mutex_lock(&mdsc->mutex);
2191         if (!req->r_aborted) {
2192                 if (err) {
2193                         req->r_err = err;
2194                 } else {
2195                         req->r_reply = msg;
2196                         ceph_msg_get(msg);
2197                         req->r_got_result = true;
2198                 }
2199         } else {
2200                 dout("reply arrived after request %lld was aborted\n", tid);
2201         }
2202         mutex_unlock(&mdsc->mutex);
2203
2204         ceph_add_cap_releases(mdsc, req->r_session);
2205         mutex_unlock(&session->s_mutex);
2206
2207         /* kick calling process */
2208         complete_request(mdsc, req);
2209 out:
2210         ceph_mdsc_put_request(req);
2211         return;
2212 }
2213
2214
2215
2216 /*
2217  * handle mds notification that our request has been forwarded.
2218  */
2219 static void handle_forward(struct ceph_mds_client *mdsc,
2220                            struct ceph_mds_session *session,
2221                            struct ceph_msg *msg)
2222 {
2223         struct ceph_mds_request *req;
2224         u64 tid = le64_to_cpu(msg->hdr.tid);
2225         u32 next_mds;
2226         u32 fwd_seq;
2227         int err = -EINVAL;
2228         void *p = msg->front.iov_base;
2229         void *end = p + msg->front.iov_len;
2230
2231         ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2232         next_mds = ceph_decode_32(&p);
2233         fwd_seq = ceph_decode_32(&p);
2234
2235         mutex_lock(&mdsc->mutex);
2236         req = __lookup_request(mdsc, tid);
2237         if (!req) {
2238                 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2239                 goto out;  /* dup reply? */
2240         }
2241
2242         if (req->r_aborted) {
2243                 dout("forward tid %llu aborted, unregistering\n", tid);
2244                 __unregister_request(mdsc, req);
2245         } else if (fwd_seq <= req->r_num_fwd) {
2246                 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2247                      tid, next_mds, req->r_num_fwd, fwd_seq);
2248         } else {
2249                 /* resend. forward race not possible; mds would drop */
2250                 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2251                 BUG_ON(req->r_err);
2252                 BUG_ON(req->r_got_result);
2253                 req->r_num_fwd = fwd_seq;
2254                 req->r_resend_mds = next_mds;
2255                 put_request_session(req);
2256                 __do_request(mdsc, req);
2257         }
2258         ceph_mdsc_put_request(req);
2259 out:
2260         mutex_unlock(&mdsc->mutex);
2261         return;
2262
2263 bad:
2264         pr_err("mdsc_handle_forward decode error err=%d\n", err);
2265 }
2266
2267 /*
2268  * handle a mds session control message
2269  */
2270 static void handle_session(struct ceph_mds_session *session,
2271                            struct ceph_msg *msg)
2272 {
2273         struct ceph_mds_client *mdsc = session->s_mdsc;
2274         u32 op;
2275         u64 seq;
2276         int mds = session->s_mds;
2277         struct ceph_mds_session_head *h = msg->front.iov_base;
2278         int wake = 0;
2279
2280         /* decode */
2281         if (msg->front.iov_len != sizeof(*h))
2282                 goto bad;
2283         op = le32_to_cpu(h->op);
2284         seq = le64_to_cpu(h->seq);
2285
2286         mutex_lock(&mdsc->mutex);
2287         if (op == CEPH_SESSION_CLOSE)
2288                 __unregister_session(mdsc, session);
2289         /* FIXME: this ttl calculation is generous */
2290         session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2291         mutex_unlock(&mdsc->mutex);
2292
2293         mutex_lock(&session->s_mutex);
2294
2295         dout("handle_session mds%d %s %p state %s seq %llu\n",
2296              mds, ceph_session_op_name(op), session,
2297              session_state_name(session->s_state), seq);
2298
2299         if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2300                 session->s_state = CEPH_MDS_SESSION_OPEN;
2301                 pr_info("mds%d came back\n", session->s_mds);
2302         }
2303
2304         switch (op) {
2305         case CEPH_SESSION_OPEN:
2306                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2307                         pr_info("mds%d reconnect success\n", session->s_mds);
2308                 session->s_state = CEPH_MDS_SESSION_OPEN;
2309                 renewed_caps(mdsc, session, 0);
2310                 wake = 1;
2311                 if (mdsc->stopping)
2312                         __close_session(mdsc, session);
2313                 break;
2314
2315         case CEPH_SESSION_RENEWCAPS:
2316                 if (session->s_renew_seq == seq)
2317                         renewed_caps(mdsc, session, 1);
2318                 break;
2319
2320         case CEPH_SESSION_CLOSE:
2321                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2322                         pr_info("mds%d reconnect denied\n", session->s_mds);
2323                 remove_session_caps(session);
2324                 wake = 1; /* for good measure */
2325                 wake_up_all(&mdsc->session_close_wq);
2326                 kick_requests(mdsc, mds);
2327                 break;
2328
2329         case CEPH_SESSION_STALE:
2330                 pr_info("mds%d caps went stale, renewing\n",
2331                         session->s_mds);
2332                 spin_lock(&session->s_gen_ttl_lock);
2333                 session->s_cap_gen++;
2334                 session->s_cap_ttl = jiffies - 1;
2335                 spin_unlock(&session->s_gen_ttl_lock);
2336                 send_renew_caps(mdsc, session);
2337                 break;
2338
2339         case CEPH_SESSION_RECALL_STATE:
2340                 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2341                 break;
2342
2343         default:
2344                 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2345                 WARN_ON(1);
2346         }
2347
2348         mutex_unlock(&session->s_mutex);
2349         if (wake) {
2350                 mutex_lock(&mdsc->mutex);
2351                 __wake_requests(mdsc, &session->s_waiting);
2352                 mutex_unlock(&mdsc->mutex);
2353         }
2354         return;
2355
2356 bad:
2357         pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2358                (int)msg->front.iov_len);
2359         ceph_msg_dump(msg);
2360         return;
2361 }
2362
2363
2364 /*
2365  * called under session->mutex.
2366  */
2367 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2368                                    struct ceph_mds_session *session)
2369 {
2370         struct ceph_mds_request *req, *nreq;
2371         int err;
2372
2373         dout("replay_unsafe_requests mds%d\n", session->s_mds);
2374
2375         mutex_lock(&mdsc->mutex);
2376         list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2377                 err = __prepare_send_request(mdsc, req, session->s_mds);
2378                 if (!err) {
2379                         ceph_msg_get(req->r_request);
2380                         ceph_con_send(&session->s_con, req->r_request);
2381                 }
2382         }
2383         mutex_unlock(&mdsc->mutex);
2384 }
2385
2386 /*
2387  * Encode information about a cap for a reconnect with the MDS.
2388  */
2389 static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2390                           void *arg)
2391 {
2392         union {
2393                 struct ceph_mds_cap_reconnect v2;
2394                 struct ceph_mds_cap_reconnect_v1 v1;
2395         } rec;
2396         size_t reclen;
2397         struct ceph_inode_info *ci;
2398         struct ceph_reconnect_state *recon_state = arg;
2399         struct ceph_pagelist *pagelist = recon_state->pagelist;
2400         char *path;
2401         int pathlen, err;
2402         u64 pathbase;
2403         struct dentry *dentry;
2404
2405         ci = cap->ci;
2406
2407         dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2408              inode, ceph_vinop(inode), cap, cap->cap_id,
2409              ceph_cap_string(cap->issued));
2410         err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2411         if (err)
2412                 return err;
2413
2414         dentry = d_find_alias(inode);
2415         if (dentry) {
2416                 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2417                 if (IS_ERR(path)) {
2418                         err = PTR_ERR(path);
2419                         goto out_dput;
2420                 }
2421         } else {
2422                 path = NULL;
2423                 pathlen = 0;
2424         }
2425         err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2426         if (err)
2427                 goto out_free;
2428
2429         spin_lock(&ci->i_ceph_lock);
2430         cap->seq = 0;        /* reset cap seq */
2431         cap->issue_seq = 0;  /* and issue_seq */
2432
2433         if (recon_state->flock) {
2434                 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2435                 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2436                 rec.v2.issued = cpu_to_le32(cap->issued);
2437                 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2438                 rec.v2.pathbase = cpu_to_le64(pathbase);
2439                 rec.v2.flock_len = 0;
2440                 reclen = sizeof(rec.v2);
2441         } else {
2442                 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2443                 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2444                 rec.v1.issued = cpu_to_le32(cap->issued);
2445                 rec.v1.size = cpu_to_le64(inode->i_size);
2446                 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2447                 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2448                 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2449                 rec.v1.pathbase = cpu_to_le64(pathbase);
2450                 reclen = sizeof(rec.v1);
2451         }
2452         spin_unlock(&ci->i_ceph_lock);
2453
2454         if (recon_state->flock) {
2455                 int num_fcntl_locks, num_flock_locks;
2456                 struct ceph_pagelist_cursor trunc_point;
2457
2458                 ceph_pagelist_set_cursor(pagelist, &trunc_point);
2459                 do {
2460                         lock_flocks();
2461                         ceph_count_locks(inode, &num_fcntl_locks,
2462                                          &num_flock_locks);
2463                         rec.v2.flock_len = (2*sizeof(u32) +
2464                                             (num_fcntl_locks+num_flock_locks) *
2465                                             sizeof(struct ceph_filelock));
2466                         unlock_flocks();
2467
2468                         /* pre-alloc pagelist */
2469                         ceph_pagelist_truncate(pagelist, &trunc_point);
2470                         err = ceph_pagelist_append(pagelist, &rec, reclen);
2471                         if (!err)
2472                                 err = ceph_pagelist_reserve(pagelist,
2473                                                             rec.v2.flock_len);
2474
2475                         /* encode locks */
2476                         if (!err) {
2477                                 lock_flocks();
2478                                 err = ceph_encode_locks(inode,
2479                                                         pagelist,
2480                                                         num_fcntl_locks,
2481                                                         num_flock_locks);
2482                                 unlock_flocks();
2483                         }
2484                 } while (err == -ENOSPC);
2485         } else {
2486                 err = ceph_pagelist_append(pagelist, &rec, reclen);
2487         }
2488
2489 out_free:
2490         kfree(path);
2491 out_dput:
2492         dput(dentry);
2493         return err;
2494 }
2495
2496
2497 /*
2498  * If an MDS fails and recovers, clients need to reconnect in order to
2499  * reestablish shared state.  This includes all caps issued through
2500  * this session _and_ the snap_realm hierarchy.  Because it's not
2501  * clear which snap realms the mds cares about, we send everything we
2502  * know about.. that ensures we'll then get any new info the
2503  * recovering MDS might have.
2504  *
2505  * This is a relatively heavyweight operation, but it's rare.
2506  *
2507  * called with mdsc->mutex held.
2508  */
2509 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2510                                struct ceph_mds_session *session)
2511 {
2512         struct ceph_msg *reply;
2513         struct rb_node *p;
2514         int mds = session->s_mds;
2515         int err = -ENOMEM;
2516         struct ceph_pagelist *pagelist;
2517         struct ceph_reconnect_state recon_state;
2518
2519         pr_info("mds%d reconnect start\n", mds);
2520
2521         pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2522         if (!pagelist)
2523                 goto fail_nopagelist;
2524         ceph_pagelist_init(pagelist);
2525
2526         reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
2527         if (!reply)
2528                 goto fail_nomsg;
2529
2530         mutex_lock(&session->s_mutex);
2531         session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2532         session->s_seq = 0;
2533
2534         ceph_con_open(&session->s_con,
2535                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2536
2537         /* replay unsafe requests */
2538         replay_unsafe_requests(mdsc, session);
2539
2540         down_read(&mdsc->snap_rwsem);
2541
2542         dout("session %p state %s\n", session,
2543              session_state_name(session->s_state));
2544
2545         /* drop old cap expires; we're about to reestablish that state */
2546         discard_cap_releases(mdsc, session);
2547
2548         /* traverse this session's caps */
2549         err = ceph_pagelist_encode_32(pagelist, session->s_nr_caps);
2550         if (err)
2551                 goto fail;
2552
2553         recon_state.pagelist = pagelist;
2554         recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2555         err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2556         if (err < 0)
2557                 goto fail;
2558
2559         /*
2560          * snaprealms.  we provide mds with the ino, seq (version), and
2561          * parent for all of our realms.  If the mds has any newer info,
2562          * it will tell us.
2563          */
2564         for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2565                 struct ceph_snap_realm *realm =
2566                         rb_entry(p, struct ceph_snap_realm, node);
2567                 struct ceph_mds_snaprealm_reconnect sr_rec;
2568
2569                 dout(" adding snap realm %llx seq %lld parent %llx\n",
2570                      realm->ino, realm->seq, realm->parent_ino);
2571                 sr_rec.ino = cpu_to_le64(realm->ino);
2572                 sr_rec.seq = cpu_to_le64(realm->seq);
2573                 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2574                 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2575                 if (err)
2576                         goto fail;
2577         }
2578
2579         reply->pagelist = pagelist;
2580         if (recon_state.flock)
2581                 reply->hdr.version = cpu_to_le16(2);
2582         reply->hdr.data_len = cpu_to_le32(pagelist->length);
2583         reply->nr_pages = calc_pages_for(0, pagelist->length);
2584         ceph_con_send(&session->s_con, reply);
2585
2586         mutex_unlock(&session->s_mutex);
2587
2588         mutex_lock(&mdsc->mutex);
2589         __wake_requests(mdsc, &session->s_waiting);
2590         mutex_unlock(&mdsc->mutex);
2591
2592         up_read(&mdsc->snap_rwsem);
2593         return;
2594
2595 fail:
2596         ceph_msg_put(reply);
2597         up_read(&mdsc->snap_rwsem);
2598         mutex_unlock(&session->s_mutex);
2599 fail_nomsg:
2600         ceph_pagelist_release(pagelist);
2601         kfree(pagelist);
2602 fail_nopagelist:
2603         pr_err("error %d preparing reconnect for mds%d\n", err, mds);
2604         return;
2605 }
2606
2607
2608 /*
2609  * compare old and new mdsmaps, kicking requests
2610  * and closing out old connections as necessary
2611  *
2612  * called under mdsc->mutex.
2613  */
2614 static void check_new_map(struct ceph_mds_client *mdsc,
2615                           struct ceph_mdsmap *newmap,
2616                           struct ceph_mdsmap *oldmap)
2617 {
2618         int i;
2619         int oldstate, newstate;
2620         struct ceph_mds_session *s;
2621
2622         dout("check_new_map new %u old %u\n",
2623              newmap->m_epoch, oldmap->m_epoch);
2624
2625         for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2626                 if (mdsc->sessions[i] == NULL)
2627                         continue;
2628                 s = mdsc->sessions[i];
2629                 oldstate = ceph_mdsmap_get_state(oldmap, i);
2630                 newstate = ceph_mdsmap_get_state(newmap, i);
2631
2632                 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2633                      i, ceph_mds_state_name(oldstate),
2634                      ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2635                      ceph_mds_state_name(newstate),
2636                      ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
2637                      session_state_name(s->s_state));
2638
2639                 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
2640                            ceph_mdsmap_get_addr(newmap, i),
2641                            sizeof(struct ceph_entity_addr))) {
2642                         if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2643                                 /* the session never opened, just close it
2644                                  * out now */
2645                                 __wake_requests(mdsc, &s->s_waiting);
2646                                 __unregister_session(mdsc, s);
2647                         } else {
2648                                 /* just close it */
2649                                 mutex_unlock(&mdsc->mutex);
2650                                 mutex_lock(&s->s_mutex);
2651                                 mutex_lock(&mdsc->mutex);
2652                                 ceph_con_close(&s->s_con);
2653                                 mutex_unlock(&s->s_mutex);
2654                                 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2655                         }
2656
2657                         /* kick any requests waiting on the recovering mds */
2658                         kick_requests(mdsc, i);
2659                 } else if (oldstate == newstate) {
2660                         continue;  /* nothing new with this mds */
2661                 }
2662
2663                 /*
2664                  * send reconnect?
2665                  */
2666                 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
2667                     newstate >= CEPH_MDS_STATE_RECONNECT) {
2668                         mutex_unlock(&mdsc->mutex);
2669                         send_mds_reconnect(mdsc, s);
2670                         mutex_lock(&mdsc->mutex);
2671                 }
2672
2673                 /*
2674                  * kick request on any mds that has gone active.
2675                  */
2676                 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2677                     newstate >= CEPH_MDS_STATE_ACTIVE) {
2678                         if (oldstate != CEPH_MDS_STATE_CREATING &&
2679                             oldstate != CEPH_MDS_STATE_STARTING)
2680                                 pr_info("mds%d recovery completed\n", s->s_mds);
2681                         kick_requests(mdsc, i);
2682                         ceph_kick_flushing_caps(mdsc, s);
2683                         wake_up_session_caps(s, 1);
2684                 }
2685         }
2686
2687         for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
2688                 s = mdsc->sessions[i];
2689                 if (!s)
2690                         continue;
2691                 if (!ceph_mdsmap_is_laggy(newmap, i))
2692                         continue;
2693                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2694                     s->s_state == CEPH_MDS_SESSION_HUNG ||
2695                     s->s_state == CEPH_MDS_SESSION_CLOSING) {
2696                         dout(" connecting to export targets of laggy mds%d\n",
2697                              i);
2698                         __open_export_target_sessions(mdsc, s);
2699                 }
2700         }
2701 }
2702
2703
2704
2705 /*
2706  * leases
2707  */
2708
2709 /*
2710  * caller must hold session s_mutex, dentry->d_lock
2711  */
2712 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
2713 {
2714         struct ceph_dentry_info *di = ceph_dentry(dentry);
2715
2716         ceph_put_mds_session(di->lease_session);
2717         di->lease_session = NULL;
2718 }
2719
2720 static void handle_lease(struct ceph_mds_client *mdsc,
2721                          struct ceph_mds_session *session,
2722                          struct ceph_msg *msg)
2723 {
2724         struct super_block *sb = mdsc->fsc->sb;
2725         struct inode *inode;
2726         struct dentry *parent, *dentry;
2727         struct ceph_dentry_info *di;
2728         int mds = session->s_mds;
2729         struct ceph_mds_lease *h = msg->front.iov_base;
2730         u32 seq;
2731         struct ceph_vino vino;
2732         struct qstr dname;
2733         int release = 0;
2734
2735         dout("handle_lease from mds%d\n", mds);
2736
2737         /* decode */
2738         if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
2739                 goto bad;
2740         vino.ino = le64_to_cpu(h->ino);
2741         vino.snap = CEPH_NOSNAP;
2742         seq = le32_to_cpu(h->seq);
2743         dname.name = (void *)h + sizeof(*h) + sizeof(u32);
2744         dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
2745         if (dname.len != get_unaligned_le32(h+1))
2746                 goto bad;
2747
2748         mutex_lock(&session->s_mutex);
2749         session->s_seq++;
2750
2751         /* lookup inode */
2752         inode = ceph_find_inode(sb, vino);
2753         dout("handle_lease %s, ino %llx %p %.*s\n",
2754              ceph_lease_op_name(h->action), vino.ino, inode,
2755              dname.len, dname.name);
2756         if (inode == NULL) {
2757                 dout("handle_lease no inode %llx\n", vino.ino);
2758                 goto release;
2759         }
2760
2761         /* dentry */
2762         parent = d_find_alias(inode);
2763         if (!parent) {
2764                 dout("no parent dentry on inode %p\n", inode);
2765                 WARN_ON(1);
2766                 goto release;  /* hrm... */
2767         }
2768         dname.hash = full_name_hash(dname.name, dname.len);
2769         dentry = d_lookup(parent, &dname);
2770         dput(parent);
2771         if (!dentry)
2772                 goto release;
2773
2774         spin_lock(&dentry->d_lock);
2775         di = ceph_dentry(dentry);
2776         switch (h->action) {
2777         case CEPH_MDS_LEASE_REVOKE:
2778                 if (di->lease_session == session) {
2779                         if (ceph_seq_cmp(di->lease_seq, seq) > 0)
2780                                 h->seq = cpu_to_le32(di->lease_seq);
2781                         __ceph_mdsc_drop_dentry_lease(dentry);
2782                 }
2783                 release = 1;
2784                 break;
2785
2786         case CEPH_MDS_LEASE_RENEW:
2787                 if (di->lease_session == session &&
2788                     di->lease_gen == session->s_cap_gen &&
2789                     di->lease_renew_from &&
2790                     di->lease_renew_after == 0) {
2791                         unsigned long duration =
2792                                 le32_to_cpu(h->duration_ms) * HZ / 1000;
2793
2794                         di->lease_seq = seq;
2795                         dentry->d_time = di->lease_renew_from + duration;
2796                         di->lease_renew_after = di->lease_renew_from +
2797                                 (duration >> 1);
2798                         di->lease_renew_from = 0;
2799                 }
2800                 break;
2801         }
2802         spin_unlock(&dentry->d_lock);
2803         dput(dentry);
2804
2805         if (!release)
2806                 goto out;
2807
2808 release:
2809         /* let's just reuse the same message */
2810         h->action = CEPH_MDS_LEASE_REVOKE_ACK;
2811         ceph_msg_get(msg);
2812         ceph_con_send(&session->s_con, msg);
2813
2814 out:
2815         iput(inode);
2816         mutex_unlock(&session->s_mutex);
2817         return;
2818
2819 bad:
2820         pr_err("corrupt lease message\n");
2821         ceph_msg_dump(msg);
2822 }
2823
2824 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
2825                               struct inode *inode,
2826                               struct dentry *dentry, char action,
2827                               u32 seq)
2828 {
2829         struct ceph_msg *msg;
2830         struct ceph_mds_lease *lease;
2831         int len = sizeof(*lease) + sizeof(u32);
2832         int dnamelen = 0;
2833
2834         dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
2835              inode, dentry, ceph_lease_op_name(action), session->s_mds);
2836         dnamelen = dentry->d_name.len;
2837         len += dnamelen;
2838
2839         msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
2840         if (!msg)
2841                 return;
2842         lease = msg->front.iov_base;
2843         lease->action = action;
2844         lease->ino = cpu_to_le64(ceph_vino(inode).ino);
2845         lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
2846         lease->seq = cpu_to_le32(seq);
2847         put_unaligned_le32(dnamelen, lease + 1);
2848         memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
2849
2850         /*
2851          * if this is a preemptive lease RELEASE, no need to
2852          * flush request stream, since the actual request will
2853          * soon follow.
2854          */
2855         msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
2856
2857         ceph_con_send(&session->s_con, msg);
2858 }
2859
2860 /*
2861  * Preemptively release a lease we expect to invalidate anyway.
2862  * Pass @inode always, @dentry is optional.
2863  */
2864 void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
2865                              struct dentry *dentry)
2866 {
2867         struct ceph_dentry_info *di;
2868         struct ceph_mds_session *session;
2869         u32 seq;
2870
2871         BUG_ON(inode == NULL);
2872         BUG_ON(dentry == NULL);
2873
2874         /* is dentry lease valid? */
2875         spin_lock(&dentry->d_lock);
2876         di = ceph_dentry(dentry);
2877         if (!di || !di->lease_session ||
2878             di->lease_session->s_mds < 0 ||
2879             di->lease_gen != di->lease_session->s_cap_gen ||
2880             !time_before(jiffies, dentry->d_time)) {
2881                 dout("lease_release inode %p dentry %p -- "
2882                      "no lease\n",
2883                      inode, dentry);
2884                 spin_unlock(&dentry->d_lock);
2885                 return;
2886         }
2887
2888         /* we do have a lease on this dentry; note mds and seq */
2889         session = ceph_get_mds_session(di->lease_session);
2890         seq = di->lease_seq;
2891         __ceph_mdsc_drop_dentry_lease(dentry);
2892         spin_unlock(&dentry->d_lock);
2893
2894         dout("lease_release inode %p dentry %p to mds%d\n",
2895              inode, dentry, session->s_mds);
2896         ceph_mdsc_lease_send_msg(session, inode, dentry,
2897                                  CEPH_MDS_LEASE_RELEASE, seq);
2898         ceph_put_mds_session(session);
2899 }
2900
2901 /*
2902  * drop all leases (and dentry refs) in preparation for umount
2903  */
2904 static void drop_leases(struct ceph_mds_client *mdsc)
2905 {
2906         int i;
2907
2908         dout("drop_leases\n");
2909         mutex_lock(&mdsc->mutex);
2910         for (i = 0; i < mdsc->max_sessions; i++) {
2911                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2912                 if (!s)
2913                         continue;
2914                 mutex_unlock(&mdsc->mutex);
2915                 mutex_lock(&s->s_mutex);
2916                 mutex_unlock(&s->s_mutex);
2917                 ceph_put_mds_session(s);
2918                 mutex_lock(&mdsc->mutex);
2919         }
2920         mutex_unlock(&mdsc->mutex);
2921 }
2922
2923
2924
2925 /*
2926  * delayed work -- periodically trim expired leases, renew caps with mds
2927  */
2928 static void schedule_delayed(struct ceph_mds_client *mdsc)
2929 {
2930         int delay = 5;
2931         unsigned hz = round_jiffies_relative(HZ * delay);
2932         schedule_delayed_work(&mdsc->delayed_work, hz);
2933 }
2934
2935 static void delayed_work(struct work_struct *work)
2936 {
2937         int i;
2938         struct ceph_mds_client *mdsc =
2939                 container_of(work, struct ceph_mds_client, delayed_work.work);
2940         int renew_interval;
2941         int renew_caps;
2942
2943         dout("mdsc delayed_work\n");
2944         ceph_check_delayed_caps(mdsc);
2945
2946         mutex_lock(&mdsc->mutex);
2947         renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
2948         renew_caps = time_after_eq(jiffies, HZ*renew_interval +
2949                                    mdsc->last_renew_caps);
2950         if (renew_caps)
2951                 mdsc->last_renew_caps = jiffies;
2952
2953         for (i = 0; i < mdsc->max_sessions; i++) {
2954                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2955                 if (s == NULL)
2956                         continue;
2957                 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
2958                         dout("resending session close request for mds%d\n",
2959                              s->s_mds);
2960                         request_close_session(mdsc, s);
2961                         ceph_put_mds_session(s);
2962                         continue;
2963                 }
2964                 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
2965                         if (s->s_state == CEPH_MDS_SESSION_OPEN) {
2966                                 s->s_state = CEPH_MDS_SESSION_HUNG;
2967                                 pr_info("mds%d hung\n", s->s_mds);
2968                         }
2969                 }
2970                 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
2971                         /* this mds is failed or recovering, just wait */
2972                         ceph_put_mds_session(s);
2973                         continue;
2974                 }
2975                 mutex_unlock(&mdsc->mutex);
2976
2977                 mutex_lock(&s->s_mutex);
2978                 if (renew_caps)
2979                         send_renew_caps(mdsc, s);
2980                 else
2981                         ceph_con_keepalive(&s->s_con);
2982                 ceph_add_cap_releases(mdsc, s);
2983                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2984                     s->s_state == CEPH_MDS_SESSION_HUNG)
2985                         ceph_send_cap_releases(mdsc, s);
2986                 mutex_unlock(&s->s_mutex);
2987                 ceph_put_mds_session(s);
2988
2989                 mutex_lock(&mdsc->mutex);
2990         }
2991         mutex_unlock(&mdsc->mutex);
2992
2993         schedule_delayed(mdsc);
2994 }
2995
2996 int ceph_mdsc_init(struct ceph_fs_client *fsc)
2997
2998 {
2999         struct ceph_mds_client *mdsc;
3000
3001         mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3002         if (!mdsc)
3003                 return -ENOMEM;
3004         mdsc->fsc = fsc;
3005         fsc->mdsc = mdsc;
3006         mutex_init(&mdsc->mutex);
3007         mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
3008         if (mdsc->mdsmap == NULL)
3009                 return -ENOMEM;
3010
3011         init_completion(&mdsc->safe_umount_waiters);
3012         init_waitqueue_head(&mdsc->session_close_wq);
3013         INIT_LIST_HEAD(&mdsc->waiting_for_map);
3014         mdsc->sessions = NULL;
3015         mdsc->max_sessions = 0;
3016         mdsc->stopping = 0;
3017         init_rwsem(&mdsc->snap_rwsem);
3018         mdsc->snap_realms = RB_ROOT;
3019         INIT_LIST_HEAD(&mdsc->snap_empty);
3020         spin_lock_init(&mdsc->snap_empty_lock);
3021         mdsc->last_tid = 0;
3022         mdsc->request_tree = RB_ROOT;
3023         INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3024         mdsc->last_renew_caps = jiffies;
3025         INIT_LIST_HEAD(&mdsc->cap_delay_list);
3026         spin_lock_init(&mdsc->cap_delay_lock);
3027         INIT_LIST_HEAD(&mdsc->snap_flush_list);
3028         spin_lock_init(&mdsc->snap_flush_lock);
3029         mdsc->cap_flush_seq = 0;
3030         INIT_LIST_HEAD(&mdsc->cap_dirty);
3031         INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
3032         mdsc->num_cap_flushing = 0;
3033         spin_lock_init(&mdsc->cap_dirty_lock);
3034         init_waitqueue_head(&mdsc->cap_flushing_wq);
3035         spin_lock_init(&mdsc->dentry_lru_lock);
3036         INIT_LIST_HEAD(&mdsc->dentry_lru);
3037
3038         ceph_caps_init(mdsc);
3039         ceph_adjust_min_caps(mdsc, fsc->min_caps);
3040
3041         return 0;
3042 }
3043
3044 /*
3045  * Wait for safe replies on open mds requests.  If we time out, drop
3046  * all requests from the tree to avoid dangling dentry refs.
3047  */
3048 static void wait_requests(struct ceph_mds_client *mdsc)
3049 {
3050         struct ceph_mds_request *req;
3051         struct ceph_fs_client *fsc = mdsc->fsc;
3052
3053         mutex_lock(&mdsc->mutex);
3054         if (__get_oldest_req(mdsc)) {
3055                 mutex_unlock(&mdsc->mutex);
3056
3057                 dout("wait_requests waiting for requests\n");
3058                 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
3059                                     fsc->client->options->mount_timeout * HZ);
3060
3061                 /* tear down remaining requests */
3062                 mutex_lock(&mdsc->mutex);
3063                 while ((req = __get_oldest_req(mdsc))) {
3064                         dout("wait_requests timed out on tid %llu\n",
3065                              req->r_tid);
3066                         __unregister_request(mdsc, req);
3067                 }
3068         }
3069         mutex_unlock(&mdsc->mutex);
3070         dout("wait_requests done\n");
3071 }
3072
3073 /*
3074  * called before mount is ro, and before dentries are torn down.
3075  * (hmm, does this still race with new lookups?)
3076  */
3077 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3078 {
3079         dout("pre_umount\n");
3080         mdsc->stopping = 1;
3081
3082         drop_leases(mdsc);
3083         ceph_flush_dirty_caps(mdsc);
3084         wait_requests(mdsc);
3085
3086         /*
3087          * wait for reply handlers to drop their request refs and
3088          * their inode/dcache refs
3089          */
3090         ceph_msgr_flush();
3091 }
3092
3093 /*
3094  * wait for all write mds requests to flush.
3095  */
3096 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3097 {
3098         struct ceph_mds_request *req = NULL, *nextreq;
3099         struct rb_node *n;
3100
3101         mutex_lock(&mdsc->mutex);
3102         dout("wait_unsafe_requests want %lld\n", want_tid);
3103 restart:
3104         req = __get_oldest_req(mdsc);
3105         while (req && req->r_tid <= want_tid) {
3106                 /* find next request */
3107                 n = rb_next(&req->r_node);
3108                 if (n)
3109                         nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3110                 else
3111                         nextreq = NULL;
3112                 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
3113                         /* write op */
3114                         ceph_mdsc_get_request(req);
3115                         if (nextreq)
3116                                 ceph_mdsc_get_request(nextreq);
3117                         mutex_unlock(&mdsc->mutex);
3118                         dout("wait_unsafe_requests  wait on %llu (want %llu)\n",
3119                              req->r_tid, want_tid);
3120                         wait_for_completion(&req->r_safe_completion);
3121                         mutex_lock(&mdsc->mutex);
3122                         ceph_mdsc_put_request(req);
3123                         if (!nextreq)
3124                                 break;  /* next dne before, so we're done! */
3125                         if (RB_EMPTY_NODE(&nextreq->r_node)) {
3126                                 /* next request was removed from tree */
3127                                 ceph_mdsc_put_request(nextreq);
3128                                 goto restart;
3129                         }
3130                         ceph_mdsc_put_request(nextreq);  /* won't go away */
3131                 }
3132                 req = nextreq;
3133         }
3134         mutex_unlock(&mdsc->mutex);
3135         dout("wait_unsafe_requests done\n");
3136 }
3137
3138 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3139 {
3140         u64 want_tid, want_flush;
3141
3142         if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
3143                 return;
3144
3145         dout("sync\n");
3146         mutex_lock(&mdsc->mutex);
3147         want_tid = mdsc->last_tid;
3148         want_flush = mdsc->cap_flush_seq;
3149         mutex_unlock(&mdsc->mutex);
3150         dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
3151
3152         ceph_flush_dirty_caps(mdsc);
3153
3154         wait_unsafe_requests(mdsc, want_tid);
3155         wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush));
3156 }
3157
3158 /*
3159  * true if all sessions are closed, or we force unmount
3160  */
3161 static bool done_closing_sessions(struct ceph_mds_client *mdsc)
3162 {
3163         int i, n = 0;
3164
3165         if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
3166                 return true;
3167
3168         mutex_lock(&mdsc->mutex);
3169         for (i = 0; i < mdsc->max_sessions; i++)
3170                 if (mdsc->sessions[i])
3171                         n++;
3172         mutex_unlock(&mdsc->mutex);
3173         return n == 0;
3174 }
3175
3176 /*
3177  * called after sb is ro.
3178  */
3179 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3180 {
3181         struct ceph_mds_session *session;
3182         int i;
3183         struct ceph_fs_client *fsc = mdsc->fsc;
3184         unsigned long timeout = fsc->client->options->mount_timeout * HZ;
3185
3186         dout("close_sessions\n");
3187
3188         /* close sessions */
3189         mutex_lock(&mdsc->mutex);
3190         for (i = 0; i < mdsc->max_sessions; i++) {
3191                 session = __ceph_lookup_mds_session(mdsc, i);
3192                 if (!session)
3193                         continue;
3194                 mutex_unlock(&mdsc->mutex);
3195                 mutex_lock(&session->s_mutex);
3196                 __close_session(mdsc, session);
3197                 mutex_unlock(&session->s_mutex);
3198                 ceph_put_mds_session(session);
3199                 mutex_lock(&mdsc->mutex);
3200         }
3201         mutex_unlock(&mdsc->mutex);
3202
3203         dout("waiting for sessions to close\n");
3204         wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3205                            timeout);
3206
3207         /* tear down remaining sessions */
3208         mutex_lock(&mdsc->mutex);
3209         for (i = 0; i < mdsc->max_sessions; i++) {
3210                 if (mdsc->sessions[i]) {
3211                         session = get_session(mdsc->sessions[i]);
3212                         __unregister_session(mdsc, session);
3213                         mutex_unlock(&mdsc->mutex);
3214                         mutex_lock(&session->s_mutex);
3215                         remove_session_caps(session);
3216                         mutex_unlock(&session->s_mutex);
3217                         ceph_put_mds_session(session);
3218                         mutex_lock(&mdsc->mutex);
3219                 }
3220         }
3221         WARN_ON(!list_empty(&mdsc->cap_delay_list));
3222         mutex_unlock(&mdsc->mutex);
3223
3224         ceph_cleanup_empty_realms(mdsc);
3225
3226         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3227
3228         dout("stopped\n");
3229 }
3230
3231 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
3232 {
3233         dout("stop\n");
3234         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3235         if (mdsc->mdsmap)
3236                 ceph_mdsmap_destroy(mdsc->mdsmap);
3237         kfree(mdsc->sessions);
3238         ceph_caps_finalize(mdsc);
3239 }
3240
3241 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3242 {
3243         struct ceph_mds_client *mdsc = fsc->mdsc;
3244
3245         dout("mdsc_destroy %p\n", mdsc);
3246         ceph_mdsc_stop(mdsc);
3247
3248         /* flush out any connection work with references to us */
3249         ceph_msgr_flush();
3250
3251         fsc->mdsc = NULL;
3252         kfree(mdsc);
3253         dout("mdsc_destroy %p done\n", mdsc);
3254 }
3255
3256
3257 /*
3258  * handle mds map update.
3259  */
3260 void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3261 {
3262         u32 epoch;
3263         u32 maplen;
3264         void *p = msg->front.iov_base;
3265         void *end = p + msg->front.iov_len;
3266         struct ceph_mdsmap *newmap, *oldmap;
3267         struct ceph_fsid fsid;
3268         int err = -EINVAL;
3269
3270         ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3271         ceph_decode_copy(&p, &fsid, sizeof(fsid));
3272         if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
3273                 return;
3274         epoch = ceph_decode_32(&p);
3275         maplen = ceph_decode_32(&p);
3276         dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3277
3278         /* do we need it? */
3279         ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
3280         mutex_lock(&mdsc->mutex);
3281         if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3282                 dout("handle_map epoch %u <= our %u\n",
3283                      epoch, mdsc->mdsmap->m_epoch);
3284                 mutex_unlock(&mdsc->mutex);
3285                 return;
3286         }
3287
3288         newmap = ceph_mdsmap_decode(&p, end);
3289         if (IS_ERR(newmap)) {
3290                 err = PTR_ERR(newmap);
3291                 goto bad_unlock;
3292         }
3293
3294         /* swap into place */
3295         if (mdsc->mdsmap) {
3296                 oldmap = mdsc->mdsmap;
3297                 mdsc->mdsmap = newmap;
3298                 check_new_map(mdsc, newmap, oldmap);
3299                 ceph_mdsmap_destroy(oldmap);
3300         } else {
3301                 mdsc->mdsmap = newmap;  /* first mds map */
3302         }
3303         mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
3304
3305         __wake_requests(mdsc, &mdsc->waiting_for_map);
3306
3307         mutex_unlock(&mdsc->mutex);
3308         schedule_delayed(mdsc);
3309         return;
3310
3311 bad_unlock:
3312         mutex_unlock(&mdsc->mutex);
3313 bad:
3314         pr_err("error decoding mdsmap %d\n", err);
3315         return;
3316 }
3317
3318 static struct ceph_connection *con_get(struct ceph_connection *con)
3319 {
3320         struct ceph_mds_session *s = con->private;
3321
3322         if (get_session(s)) {
3323                 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
3324                 return con;
3325         }
3326         dout("mdsc con_get %p FAIL\n", s);
3327         return NULL;
3328 }
3329
3330 static void con_put(struct ceph_connection *con)
3331 {
3332         struct ceph_mds_session *s = con->private;
3333
3334         dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
3335         ceph_put_mds_session(s);
3336 }
3337
3338 /*
3339  * if the client is unresponsive for long enough, the mds will kill
3340  * the session entirely.
3341  */
3342 static void peer_reset(struct ceph_connection *con)
3343 {
3344         struct ceph_mds_session *s = con->private;
3345         struct ceph_mds_client *mdsc = s->s_mdsc;
3346
3347         pr_warning("mds%d closed our session\n", s->s_mds);
3348         send_mds_reconnect(mdsc, s);
3349 }
3350
3351 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3352 {
3353         struct ceph_mds_session *s = con->private;
3354         struct ceph_mds_client *mdsc = s->s_mdsc;
3355         int type = le16_to_cpu(msg->hdr.type);
3356
3357         mutex_lock(&mdsc->mutex);
3358         if (__verify_registered_session(mdsc, s) < 0) {
3359                 mutex_unlock(&mdsc->mutex);
3360                 goto out;
3361         }
3362         mutex_unlock(&mdsc->mutex);
3363
3364         switch (type) {
3365         case CEPH_MSG_MDS_MAP:
3366                 ceph_mdsc_handle_map(mdsc, msg);
3367                 break;
3368         case CEPH_MSG_CLIENT_SESSION:
3369                 handle_session(s, msg);
3370                 break;
3371         case CEPH_MSG_CLIENT_REPLY:
3372                 handle_reply(s, msg);
3373                 break;
3374         case CEPH_MSG_CLIENT_REQUEST_FORWARD:
3375                 handle_forward(mdsc, s, msg);
3376                 break;
3377         case CEPH_MSG_CLIENT_CAPS:
3378                 ceph_handle_caps(s, msg);
3379                 break;
3380         case CEPH_MSG_CLIENT_SNAP:
3381                 ceph_handle_snap(mdsc, s, msg);
3382                 break;
3383         case CEPH_MSG_CLIENT_LEASE:
3384                 handle_lease(mdsc, s, msg);
3385                 break;
3386
3387         default:
3388                 pr_err("received unknown message type %d %s\n", type,
3389                        ceph_msg_type_name(type));
3390         }
3391 out:
3392         ceph_msg_put(msg);
3393 }
3394
3395 /*
3396  * authentication
3397  */
3398 static int get_authorizer(struct ceph_connection *con,
3399                           void **buf, int *len, int *proto,
3400                           void **reply_buf, int *reply_len, int force_new)
3401 {
3402         struct ceph_mds_session *s = con->private;
3403         struct ceph_mds_client *mdsc = s->s_mdsc;
3404         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3405         int ret = 0;
3406
3407         if (force_new && s->s_authorizer) {
3408                 ac->ops->destroy_authorizer(ac, s->s_authorizer);
3409                 s->s_authorizer = NULL;
3410         }
3411         if (s->s_authorizer == NULL) {
3412                 if (ac->ops->create_authorizer) {
3413                         ret = ac->ops->create_authorizer(
3414                                 ac, CEPH_ENTITY_TYPE_MDS,
3415                                 &s->s_authorizer,
3416                                 &s->s_authorizer_buf,
3417                                 &s->s_authorizer_buf_len,
3418                                 &s->s_authorizer_reply_buf,
3419                                 &s->s_authorizer_reply_buf_len);
3420                         if (ret)
3421                                 return ret;
3422                 }
3423         }
3424
3425         *proto = ac->protocol;
3426         *buf = s->s_authorizer_buf;
3427         *len = s->s_authorizer_buf_len;
3428         *reply_buf = s->s_authorizer_reply_buf;
3429         *reply_len = s->s_authorizer_reply_buf_len;
3430         return 0;
3431 }
3432
3433
3434 static int verify_authorizer_reply(struct ceph_connection *con, int len)
3435 {
3436         struct ceph_mds_session *s = con->private;
3437         struct ceph_mds_client *mdsc = s->s_mdsc;
3438         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3439
3440         return ac->ops->verify_authorizer_reply(ac, s->s_authorizer, len);
3441 }
3442
3443 static int invalidate_authorizer(struct ceph_connection *con)
3444 {
3445         struct ceph_mds_session *s = con->private;
3446         struct ceph_mds_client *mdsc = s->s_mdsc;
3447         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3448
3449         if (ac->ops->invalidate_authorizer)
3450                 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
3451
3452         return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
3453 }
3454
3455 static const struct ceph_connection_operations mds_con_ops = {
3456         .get = con_get,
3457         .put = con_put,
3458         .dispatch = dispatch,
3459         .get_authorizer = get_authorizer,
3460         .verify_authorizer_reply = verify_authorizer_reply,
3461         .invalidate_authorizer = invalidate_authorizer,
3462         .peer_reset = peer_reset,
3463 };
3464
3465 /* eof */