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  41  * are no longer valid.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