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  42  * are no longer valid.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