fs/ceph/mds_client.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/fs.h>
   4 #include <linux/wait.h>
   5 #include <linux/slab.h>
   6 #include <linux/sched.h>
   7 #include <linux/debugfs.h>
   8 #include <linux/seq_file.h>
   9
  10 #include "super.h"
  11 #include "mds_client.h"
  12
  13 #include <linux/ceph/ceph_features.h>
  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  42  * are no longer valid.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                                int features)
  66 {
  67         int err = -EIO;
  68
  69         info->in = *p;
  70         *p += sizeof(struct ceph_mds_reply_inode) +
  71                 sizeof(*info->in->fragtree.splits) *
  72                 le32_to_cpu(info->in->fragtree.nsplits);
  73
  74         ceph_decode_32_safe(p, end, info->symlink_len, bad);
  75         ceph_decode_need(p, end, info->symlink_len, bad);
  76         info->symlink = *p;
  77         *p += info->symlink_len;
  78
  79         if (features & CEPH_FEATURE_DIRLAYOUTHASH)
  80                 ceph_decode_copy_safe(p, end, &info->dir_layout,
  81                                       sizeof(info->dir_layout), bad);
  82         else
  83                 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
  84
  85         ceph_decode_32_safe(p, end, info->xattr_len, bad);
  86         ceph_decode_need(p, end, info->xattr_len, bad);
  87         info->xattr_data = *p;
  88         *p += info->xattr_len;
  89         return 0;
  90 bad:
  91         return err;
  92 }
  93
  94 /*
  95  * parse a normal reply, which may contain a (dir+)dentry and/or a
  96  * target inode.
  97  */
  98 static int parse_reply_info_trace(void **p, void *end,
  99                                   struct ceph_mds_reply_info_parsed *info,
 100                                   int features)
 101 {
 102         int err;
 103
 104         if (info->head->is_dentry) {
 105                 err = parse_reply_info_in(p, end, &info->diri, features);
 106                 if (err < 0)
 107                         goto out_bad;
 108
 109                 if (unlikely(*p + sizeof(*info->dirfrag) > end))
 110                         goto bad;
 111                 info->dirfrag = *p;
 112                 *p += sizeof(*info->dirfrag) +
 113                         sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
 114                 if (unlikely(*p > end))
 115                         goto bad;
 116
 117                 ceph_decode_32_safe(p, end, info->dname_len, bad);
 118                 ceph_decode_need(p, end, info->dname_len, bad);
 119                 info->dname = *p;
 120                 *p += info->dname_len;
 121                 info->dlease = *p;
 122                 *p += sizeof(*info->dlease);
 123         }
 124
 125         if (info->head->is_target) {
 126                 err = parse_reply_info_in(p, end, &info->targeti, features);
 127                 if (err < 0)
 128                         goto out_bad;
 129         }
 130
 131         if (unlikely(*p != end))
 132                 goto bad;
 133         return 0;
 134
 135 bad:
 136         err = -EIO;
 137 out_bad:
 138         pr_err("problem parsing mds trace %d\n", err);
 139         return err;
 140 }
 141
 142 /*
 143  * parse readdir results
 144  */
 145 static int parse_reply_info_dir(void **p, void *end,
 146                                 struct ceph_mds_reply_info_parsed *info,
 147                                 int features)
 148 {
 149         u32 num, i = 0;
 150         int err;
 151
 152         info->dir_dir = *p;
 153         if (*p + sizeof(*info->dir_dir) > end)
 154                 goto bad;
 155         *p += sizeof(*info->dir_dir) +
 156                 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
 157         if (*p > end)
 158                 goto bad;
 159
 160         ceph_decode_need(p, end, sizeof(num) + 2, bad);
 161         num = ceph_decode_32(p);
 162         info->dir_end = ceph_decode_8(p);
 163         info->dir_complete = ceph_decode_8(p);
 164         if (num == 0)
 165                 goto done;
 166
 167         /* alloc large array */
 168         info->dir_nr = num;
 169         info->dir_in = kcalloc(num, sizeof(*info->dir_in) +
 170                                sizeof(*info->dir_dname) +
 171                                sizeof(*info->dir_dname_len) +
 172                                sizeof(*info->dir_dlease),
 173                                GFP_NOFS);
 174         if (info->dir_in == NULL) {
 175                 err = -ENOMEM;
 176                 goto out_bad;
 177         }
 178         info->dir_dname = (void *)(info->dir_in + num);
 179         info->dir_dname_len = (void *)(info->dir_dname + num);
 180         info->dir_dlease = (void *)(info->dir_dname_len + num);
 181
 182         while (num) {
 183                 /* dentry */
 184                 ceph_decode_need(p, end, sizeof(u32)*2, bad);
 185                 info->dir_dname_len[i] = ceph_decode_32(p);
 186                 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
 187                 info->dir_dname[i] = *p;
 188                 *p += info->dir_dname_len[i];
 189                 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
 190                      info->dir_dname[i]);
 191                 info->dir_dlease[i] = *p;
 192                 *p += sizeof(struct ceph_mds_reply_lease);
 193
 194                 /* inode */
 195                 err = parse_reply_info_in(p, end, &info->dir_in[i], features);
 196                 if (err < 0)
 197                         goto out_bad;
 198                 i++;
 199                 num--;
 200         }
 201
 202 done:
 203         if (*p != end)
 204                 goto bad;
 205         return 0;
 206
 207 bad:
 208         err = -EIO;
 209 out_bad:
 210         pr_err("problem parsing dir contents %d\n", err);
 211         return err;
 212 }
 213
 214 /*
 215  * parse fcntl F_GETLK results
 216  */
 217 static int parse_reply_info_filelock(void **p, void *end,
 218                                      struct ceph_mds_reply_info_parsed *info,
 219                                      int features)
 220 {
 221         if (*p + sizeof(*info->filelock_reply) > end)
 222                 goto bad;
 223
 224         info->filelock_reply = *p;
 225         *p += sizeof(*info->filelock_reply);
 226
 227         if (unlikely(*p != end))
 228                 goto bad;
 229         return 0;
 230
 231 bad:
 232         return -EIO;
 233 }
 234
 235 /*
 236  * parse create results
 237  */
 238 static int parse_reply_info_create(void **p, void *end,
 239                                   struct ceph_mds_reply_info_parsed *info,
 240                                   int features)
 241 {
 242         if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
 243                 if (*p == end) {
 244                         info->has_create_ino = false;
 245                 } else {
 246                         info->has_create_ino = true;
 247                         info->ino = ceph_decode_64(p);
 248                 }
 249         }
 250
 251         if (unlikely(*p != end))
 252                 goto bad;
 253         return 0;
 254
 255 bad:
 256         return -EIO;
 257 }
 258
 259 /*
 260  * parse extra results
 261  */
 262 static int parse_reply_info_extra(void **p, void *end,
 263                                   struct ceph_mds_reply_info_parsed *info,
 264                                   int features)
 265 {
 266         if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
 267                 return parse_reply_info_filelock(p, end, info, features);
 268         else if (info->head->op == CEPH_MDS_OP_READDIR ||
 269                  info->head->op == CEPH_MDS_OP_LSSNAP)
 270                 return parse_reply_info_dir(p, end, info, features);
 271         else if (info->head->op == CEPH_MDS_OP_CREATE)
 272                 return parse_reply_info_create(p, end, info, features);
 273         else
 274                 return -EIO;
 275 }
 276
 277 /*
 278  * parse entire mds reply
 279  */
 280 static int parse_reply_info(struct ceph_msg *msg,
 281                             struct ceph_mds_reply_info_parsed *info,
 282                             int features)
 283 {
 284         void *p, *end;
 285         u32 len;
 286         int err;
 287
 288         info->head = msg->front.iov_base;
 289         p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
 290         end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
 291
 292         /* trace */
 293         ceph_decode_32_safe(&p, end, len, bad);
 294         if (len > 0) {
 295                 ceph_decode_need(&p, end, len, bad);
 296                 err = parse_reply_info_trace(&p, p+len, info, features);
 297                 if (err < 0)
 298                         goto out_bad;
 299         }
 300
 301         /* extra */
 302         ceph_decode_32_safe(&p, end, len, bad);
 303         if (len > 0) {
 304                 ceph_decode_need(&p, end, len, bad);
 305                 err = parse_reply_info_extra(&p, p+len, info, features);
 306                 if (err < 0)
 307                         goto out_bad;
 308         }
 309
 310         /* snap blob */
 311         ceph_decode_32_safe(&p, end, len, bad);
 312         info->snapblob_len = len;
 313         info->snapblob = p;
 314         p += len;
 315
 316         if (p != end)
 317                 goto bad;
 318         return 0;
 319
 320 bad:
 321         err = -EIO;
 322 out_bad:
 323         pr_err("mds parse_reply err %d\n", err);
 324         return err;
 325 }
 326
 327 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
 328 {
 329         kfree(info->dir_in);
 330 }
 331
 332
 333 /*
 334  * sessions
 335  */
 336 static const char *session_state_name(int s)
 337 {
 338         switch (s) {
 339         case CEPH_MDS_SESSION_NEW: return "new";
 340         case CEPH_MDS_SESSION_OPENING: return "opening";
 341         case CEPH_MDS_SESSION_OPEN: return "open";
 342         case CEPH_MDS_SESSION_HUNG: return "hung";
 343         case CEPH_MDS_SESSION_CLOSING: return "closing";
 344         case CEPH_MDS_SESSION_RESTARTING: return "restarting";
 345         case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
 346         default: return "???";
 347         }
 348 }
 349
 350 static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
 351 {
 352         if (atomic_inc_not_zero(&s->s_ref)) {
 353                 dout("mdsc get_session %p %d -> %d\n", s,
 354                      atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
 355                 return s;
 356         } else {
 357                 dout("mdsc get_session %p 0 -- FAIL", s);
 358                 return NULL;
 359         }
 360 }
 361
 362 void ceph_put_mds_session(struct ceph_mds_session *s)
 363 {
 364         dout("mdsc put_session %p %d -> %d\n", s,
 365              atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
 366         if (atomic_dec_and_test(&s->s_ref)) {
 367                 if (s->s_auth.authorizer)
 368                         ceph_auth_destroy_authorizer(
 369                                 s->s_mdsc->fsc->client->monc.auth,
 370                                 s->s_auth.authorizer);
 371                 kfree(s);
 372         }
 373 }
 374
 375 /*
 376  * called under mdsc->mutex
 377  */
 378 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
 379                                                    int mds)
 380 {
 381         struct ceph_mds_session *session;
 382
 383         if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
 384                 return NULL;
 385         session = mdsc->sessions[mds];
 386         dout("lookup_mds_session %p %d\n", session,
 387              atomic_read(&session->s_ref));
 388         get_session(session);
 389         return session;
 390 }
 391
 392 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
 393 {
 394         if (mds >= mdsc->max_sessions)
 395                 return false;
 396         return mdsc->sessions[mds];
 397 }
 398
 399 static int __verify_registered_session(struct ceph_mds_client *mdsc,
 400                                        struct ceph_mds_session *s)
 401 {
 402         if (s->s_mds >= mdsc->max_sessions ||
 403             mdsc->sessions[s->s_mds] != s)
 404                 return -ENOENT;
 405         return 0;
 406 }
 407
 408 /*
 409  * create+register a new session for given mds.
 410  * called under mdsc->mutex.
 411  */
 412 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
 413                                                  int mds)
 414 {
 415         struct ceph_mds_session *s;
 416
 417         if (mds >= mdsc->mdsmap->m_max_mds)
 418                 return ERR_PTR(-EINVAL);
 419
 420         s = kzalloc(sizeof(*s), GFP_NOFS);
 421         if (!s)
 422                 return ERR_PTR(-ENOMEM);
 423         s->s_mdsc = mdsc;
 424         s->s_mds = mds;
 425         s->s_state = CEPH_MDS_SESSION_NEW;
 426         s->s_ttl = 0;
 427         s->s_seq = 0;
 428         mutex_init(&s->s_mutex);
 429
 430         ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
 431
 432         spin_lock_init(&s->s_gen_ttl_lock);
 433         s->s_cap_gen = 0;
 434         s->s_cap_ttl = jiffies - 1;
 435
 436         spin_lock_init(&s->s_cap_lock);
 437         s->s_renew_requested = 0;
 438         s->s_renew_seq = 0;
 439         INIT_LIST_HEAD(&s->s_caps);
 440         s->s_nr_caps = 0;
 441         s->s_trim_caps = 0;
 442         atomic_set(&s->s_ref, 1);
 443         INIT_LIST_HEAD(&s->s_waiting);
 444         INIT_LIST_HEAD(&s->s_unsafe);
 445         s->s_num_cap_releases = 0;
 446         s->s_cap_iterator = NULL;
 447         INIT_LIST_HEAD(&s->s_cap_releases);
 448         INIT_LIST_HEAD(&s->s_cap_releases_done);
 449         INIT_LIST_HEAD(&s->s_cap_flushing);
 450         INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
 451
 452         dout("register_session mds%d\n", mds);
 453         if (mds >= mdsc->max_sessions) {
 454                 int newmax = 1 << get_count_order(mds+1);
 455                 struct ceph_mds_session **sa;
 456
 457                 dout("register_session realloc to %d\n", newmax);
 458                 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
 459                 if (sa == NULL)
 460                         goto fail_realloc;
 461                 if (mdsc->sessions) {
 462                         memcpy(sa, mdsc->sessions,
 463                                mdsc->max_sessions * sizeof(void *));
 464                         kfree(mdsc->sessions);
 465                 }
 466                 mdsc->sessions = sa;
 467                 mdsc->max_sessions = newmax;
 468         }
 469         mdsc->sessions[mds] = s;
 470         atomic_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
 471
 472         ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
 473                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
 474
 475         return s;
 476
 477 fail_realloc:
 478         kfree(s);
 479         return ERR_PTR(-ENOMEM);
 480 }
 481
 482 /*
 483  * called under mdsc->mutex
 484  */
 485 static void __unregister_session(struct ceph_mds_client *mdsc,
 486                                struct ceph_mds_session *s)
 487 {
 488         dout("__unregister_session mds%d %p\n", s->s_mds, s);
 489         BUG_ON(mdsc->sessions[s->s_mds] != s);
 490         mdsc->sessions[s->s_mds] = NULL;
 491         ceph_con_close(&s->s_con);
 492         ceph_put_mds_session(s);
 493 }
 494
 495 /*
 496  * drop session refs in request.
 497  *
 498  * should be last request ref, or hold mdsc->mutex
 499  */
 500 static void put_request_session(struct ceph_mds_request *req)
 501 {
 502         if (req->r_session) {
 503                 ceph_put_mds_session(req->r_session);
 504                 req->r_session = NULL;
 505         }
 506 }
 507
 508 void ceph_mdsc_release_request(struct kref *kref)
 509 {
 510         struct ceph_mds_request *req = container_of(kref,
 511                                                     struct ceph_mds_request,
 512                                                     r_kref);
 513         if (req->r_request)
 514                 ceph_msg_put(req->r_request);
 515         if (req->r_reply) {
 516                 ceph_msg_put(req->r_reply);
 517                 destroy_reply_info(&req->r_reply_info);
 518         }
 519         if (req->r_inode) {
 520                 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
 521                 iput(req->r_inode);
 522         }
 523         if (req->r_locked_dir)
 524                 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
 525         if (req->r_target_inode)
 526                 iput(req->r_target_inode);
 527         if (req->r_dentry)
 528                 dput(req->r_dentry);
 529         if (req->r_old_dentry) {
 530                 /*
 531                  * track (and drop pins for) r_old_dentry_dir
 532                  * separately, since r_old_dentry's d_parent may have
 533                  * changed between the dir mutex being dropped and
 534                  * this request being freed.
 535                  */
 536                 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
 537                                   CEPH_CAP_PIN);
 538                 dput(req->r_old_dentry);
 539                 iput(req->r_old_dentry_dir);
 540         }
 541         kfree(req->r_path1);
 542         kfree(req->r_path2);
 543         put_request_session(req);
 544         ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
 545         kfree(req);
 546 }
 547
 548 /*
 549  * lookup session, bump ref if found.
 550  *
 551  * called under mdsc->mutex.
 552  */
 553 static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
 554                                              u64 tid)
 555 {
 556         struct ceph_mds_request *req;
 557         struct rb_node *n = mdsc->request_tree.rb_node;
 558
 559         while (n) {
 560                 req = rb_entry(n, struct ceph_mds_request, r_node);
 561                 if (tid < req->r_tid)
 562                         n = n->rb_left;
 563                 else if (tid > req->r_tid)
 564                         n = n->rb_right;
 565                 else {
 566                         ceph_mdsc_get_request(req);
 567                         return req;
 568                 }
 569         }
 570         return NULL;
 571 }
 572
 573 static void __insert_request(struct ceph_mds_client *mdsc,
 574                              struct ceph_mds_request *new)
 575 {
 576         struct rb_node **p = &mdsc->request_tree.rb_node;
 577         struct rb_node *parent = NULL;
 578         struct ceph_mds_request *req = NULL;
 579
 580         while (*p) {
 581                 parent = *p;
 582                 req = rb_entry(parent, struct ceph_mds_request, r_node);
 583                 if (new->r_tid < req->r_tid)
 584                         p = &(*p)->rb_left;
 585                 else if (new->r_tid > req->r_tid)
 586                         p = &(*p)->rb_right;
 587                 else
 588                         BUG();
 589         }
 590
 591         rb_link_node(&new->r_node, parent, p);
 592         rb_insert_color(&new->r_node, &mdsc->request_tree);
 593 }
 594
 595 /*
 596  * Register an in-flight request, and assign a tid.  Link to directory
 597  * are modifying (if any).
 598  *
 599  * Called under mdsc->mutex.
 600  */
 601 static void __register_request(struct ceph_mds_client *mdsc,
 602                                struct ceph_mds_request *req,
 603                                struct inode *dir)
 604 {
 605         req->r_tid = ++mdsc->last_tid;
 606         if (req->r_num_caps)
 607                 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
 608                                   req->r_num_caps);
 609         dout("__register_request %p tid %lld\n", req, req->r_tid);
 610         ceph_mdsc_get_request(req);
 611         __insert_request(mdsc, req);
 612
 613         req->r_uid = current_fsuid();
 614         req->r_gid = current_fsgid();
 615
 616         if (dir) {
 617                 struct ceph_inode_info *ci = ceph_inode(dir);
 618
 619                 ihold(dir);
 620                 spin_lock(&ci->i_unsafe_lock);
 621                 req->r_unsafe_dir = dir;
 622                 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
 623                 spin_unlock(&ci->i_unsafe_lock);
 624         }
 625 }
 626
 627 static void __unregister_request(struct ceph_mds_client *mdsc,
 628                                  struct ceph_mds_request *req)
 629 {
 630         dout("__unregister_request %p tid %lld\n", req, req->r_tid);
 631         rb_erase(&req->r_node, &mdsc->request_tree);
 632         RB_CLEAR_NODE(&req->r_node);
 633
 634         if (req->r_unsafe_dir) {
 635                 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
 636
 637                 spin_lock(&ci->i_unsafe_lock);
 638                 list_del_init(&req->r_unsafe_dir_item);
 639                 spin_unlock(&ci->i_unsafe_lock);
 640
 641                 iput(req->r_unsafe_dir);
 642                 req->r_unsafe_dir = NULL;
 643         }
 644
 645         ceph_mdsc_put_request(req);
 646 }
 647
 648 /*
 649  * Choose mds to send request to next.  If there is a hint set in the
 650  * request (e.g., due to a prior forward hint from the mds), use that.
 651  * Otherwise, consult frag tree and/or caps to identify the
 652  * appropriate mds.  If all else fails, choose randomly.
 653  *
 654  * Called under mdsc->mutex.
 655  */
 656 static struct dentry *get_nonsnap_parent(struct dentry *dentry)
 657 {
 658         /*
 659          * we don't need to worry about protecting the d_parent access
 660          * here because we never renaming inside the snapped namespace
 661          * except to resplice to another snapdir, and either the old or new
 662          * result is a valid result.
 663          */
 664         while (!IS_ROOT(dentry) && ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
 665                 dentry = dentry->d_parent;
 666         return dentry;
 667 }
 668
 669 static int __choose_mds(struct ceph_mds_client *mdsc,
 670                         struct ceph_mds_request *req)
 671 {
 672         struct inode *inode;
 673         struct ceph_inode_info *ci;
 674         struct ceph_cap *cap;
 675         int mode = req->r_direct_mode;
 676         int mds = -1;
 677         u32 hash = req->r_direct_hash;
 678         bool is_hash = req->r_direct_is_hash;
 679
 680         /*
 681          * is there a specific mds we should try?  ignore hint if we have
 682          * no session and the mds is not up (active or recovering).
 683          */
 684         if (req->r_resend_mds >= 0 &&
 685             (__have_session(mdsc, req->r_resend_mds) ||
 686              ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
 687                 dout("choose_mds using resend_mds mds%d\n",
 688                      req->r_resend_mds);
 689                 return req->r_resend_mds;
 690         }
 691
 692         if (mode == USE_RANDOM_MDS)
 693                 goto random;
 694
 695         inode = NULL;
 696         if (req->r_inode) {
 697                 inode = req->r_inode;
 698         } else if (req->r_dentry) {
 699                 /* ignore race with rename; old or new d_parent is okay */
 700                 struct dentry *parent = req->r_dentry->d_parent;
 701                 struct inode *dir = parent->d_inode;
 702
 703                 if (dir->i_sb != mdsc->fsc->sb) {
 704                         /* not this fs! */
 705                         inode = req->r_dentry->d_inode;
 706                 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
 707                         /* direct snapped/virtual snapdir requests
 708                          * based on parent dir inode */
 709                         struct dentry *dn = get_nonsnap_parent(parent);
 710                         inode = dn->d_inode;
 711                         dout("__choose_mds using nonsnap parent %p\n", inode);
 712                 } else if (req->r_dentry->d_inode) {
 713                         /* dentry target */
 714                         inode = req->r_dentry->d_inode;
 715                 } else {
 716                         /* dir + name */
 717                         inode = dir;
 718                         hash = ceph_dentry_hash(dir, req->r_dentry);
 719                         is_hash = true;
 720                 }
 721         }
 722
 723         dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
 724              (int)hash, mode);
 725         if (!inode)
 726                 goto random;
 727         ci = ceph_inode(inode);
 728
 729         if (is_hash && S_ISDIR(inode->i_mode)) {
 730                 struct ceph_inode_frag frag;
 731                 int found;
 732
 733                 ceph_choose_frag(ci, hash, &frag, &found);
 734                 if (found) {
 735                         if (mode == USE_ANY_MDS && frag.ndist > 0) {
 736                                 u8 r;
 737
 738                                 /* choose a random replica */
 739                                 get_random_bytes(&r, 1);
 740                                 r %= frag.ndist;
 741                                 mds = frag.dist[r];
 742                                 dout("choose_mds %p %llx.%llx "
 743                                      "frag %u mds%d (%d/%d)\n",
 744                                      inode, ceph_vinop(inode),
 745                                      frag.frag, mds,
 746                                      (int)r, frag.ndist);
 747                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
 748                                     CEPH_MDS_STATE_ACTIVE)
 749                                         return mds;
 750                         }
 751
 752                         /* since this file/dir wasn't known to be
 753                          * replicated, then we want to look for the
 754                          * authoritative mds. */
 755                         mode = USE_AUTH_MDS;
 756                         if (frag.mds >= 0) {
 757                                 /* choose auth mds */
 758                                 mds = frag.mds;
 759                                 dout("choose_mds %p %llx.%llx "
 760                                      "frag %u mds%d (auth)\n",
 761                                      inode, ceph_vinop(inode), frag.frag, mds);
 762                                 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
 763                                     CEPH_MDS_STATE_ACTIVE)
 764                                         return mds;
 765                         }
 766                 }
 767         }
 768
 769         spin_lock(&ci->i_ceph_lock);
 770         cap = NULL;
 771         if (mode == USE_AUTH_MDS)
 772                 cap = ci->i_auth_cap;
 773         if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
 774                 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
 775         if (!cap) {
 776                 spin_unlock(&ci->i_ceph_lock);
 777                 goto random;
 778         }
 779         mds = cap->session->s_mds;
 780         dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
 781              inode, ceph_vinop(inode), mds,
 782              cap == ci->i_auth_cap ? "auth " : "", cap);
 783         spin_unlock(&ci->i_ceph_lock);
 784         return mds;
 785
 786 random:
 787         mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
 788         dout("choose_mds chose random mds%d\n", mds);
 789         return mds;
 790 }
 791
 792
 793 /*
 794  * session messages
 795  */
 796 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
 797 {
 798         struct ceph_msg *msg;
 799         struct ceph_mds_session_head *h;
 800
 801         msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
 802                            false);
 803         if (!msg) {
 804                 pr_err("create_session_msg ENOMEM creating msg\n");
 805                 return NULL;
 806         }
 807         h = msg->front.iov_base;
 808         h->op = cpu_to_le32(op);
 809         h->seq = cpu_to_le64(seq);
 810         return msg;
 811 }
 812
 813 /*
 814  * send session open request.
 815  *
 816  * called under mdsc->mutex
 817  */
 818 static int __open_session(struct ceph_mds_client *mdsc,
 819                           struct ceph_mds_session *session)
 820 {
 821         struct ceph_msg *msg;
 822         int mstate;
 823         int mds = session->s_mds;
 824
 825         /* wait for mds to go active? */
 826         mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
 827         dout("open_session to mds%d (%s)\n", mds,
 828              ceph_mds_state_name(mstate));
 829         session->s_state = CEPH_MDS_SESSION_OPENING;
 830         session->s_renew_requested = jiffies;
 831
 832         /* send connect message */
 833         msg = create_session_msg(CEPH_SESSION_REQUEST_OPEN, session->s_seq);
 834         if (!msg)
 835                 return -ENOMEM;
 836         ceph_con_send(&session->s_con, msg);
 837         return 0;
 838 }
 839
 840 /*
 841  * open sessions for any export targets for the given mds
 842  *
 843  * called under mdsc->mutex
 844  */
 845 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
 846                                           struct ceph_mds_session *session)
 847 {
 848         struct ceph_mds_info *mi;
 849         struct ceph_mds_session *ts;
 850         int i, mds = session->s_mds;
 851         int target;
 852
 853         if (mds >= mdsc->mdsmap->m_max_mds)
 854                 return;
 855         mi = &mdsc->mdsmap->m_info[mds];
 856         dout("open_export_target_sessions for mds%d (%d targets)\n",
 857              session->s_mds, mi->num_export_targets);
 858
 859         for (i = 0; i < mi->num_export_targets; i++) {
 860                 target = mi->export_targets[i];
 861                 ts = __ceph_lookup_mds_session(mdsc, target);
 862                 if (!ts) {
 863                         ts = register_session(mdsc, target);
 864                         if (IS_ERR(ts))
 865                                 return;
 866                 }
 867                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
 868                     session->s_state == CEPH_MDS_SESSION_CLOSING)
 869                         __open_session(mdsc, session);
 870                 else
 871                         dout(" mds%d target mds%d %p is %s\n", session->s_mds,
 872                              i, ts, session_state_name(ts->s_state));
 873                 ceph_put_mds_session(ts);
 874         }
 875 }
 876
 877 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
 878                                            struct ceph_mds_session *session)
 879 {
 880         mutex_lock(&mdsc->mutex);
 881         __open_export_target_sessions(mdsc, session);
 882         mutex_unlock(&mdsc->mutex);
 883 }
 884
 885 /*
 886  * session caps
 887  */
 888
 889 /*
 890  * Free preallocated cap messages assigned to this session
 891  */
 892 static void cleanup_cap_releases(struct ceph_mds_session *session)
 893 {
 894         struct ceph_msg *msg;
 895
 896         spin_lock(&session->s_cap_lock);
 897         while (!list_empty(&session->s_cap_releases)) {
 898                 msg = list_first_entry(&session->s_cap_releases,
 899                                        struct ceph_msg, list_head);
 900                 list_del_init(&msg->list_head);
 901                 ceph_msg_put(msg);
 902         }
 903         while (!list_empty(&session->s_cap_releases_done)) {
 904                 msg = list_first_entry(&session->s_cap_releases_done,
 905                                        struct ceph_msg, list_head);
 906                 list_del_init(&msg->list_head);
 907                 ceph_msg_put(msg);
 908         }
 909         spin_unlock(&session->s_cap_lock);
 910 }
 911
 912 /*
 913  * Helper to safely iterate over all caps associated with a session, with
 914  * special care taken to handle a racing __ceph_remove_cap().
 915  *
 916  * Caller must hold session s_mutex.
 917  */
 918 static int iterate_session_caps(struct ceph_mds_session *session,
 919                                  int (*cb)(struct inode *, struct ceph_cap *,
 920                                             void *), void *arg)
 921 {
 922         struct list_head *p;
 923         struct ceph_cap *cap;
 924         struct inode *inode, *last_inode = NULL;
 925         struct ceph_cap *old_cap = NULL;
 926         int ret;
 927
 928         dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
 929         spin_lock(&session->s_cap_lock);
 930         p = session->s_caps.next;
 931         while (p != &session->s_caps) {
 932                 cap = list_entry(p, struct ceph_cap, session_caps);
 933                 inode = igrab(&cap->ci->vfs_inode);
 934                 if (!inode) {
 935                         p = p->next;
 936                         continue;
 937                 }
 938                 session->s_cap_iterator = cap;
 939                 spin_unlock(&session->s_cap_lock);
 940
 941                 if (last_inode) {
 942                         iput(last_inode);
 943                         last_inode = NULL;
 944                 }
 945                 if (old_cap) {
 946                         ceph_put_cap(session->s_mdsc, old_cap);
 947                         old_cap = NULL;
 948                 }
 949
 950                 ret = cb(inode, cap, arg);
 951                 last_inode = inode;
 952
 953                 spin_lock(&session->s_cap_lock);
 954                 p = p->next;
 955                 if (cap->ci == NULL) {
 956                         dout("iterate_session_caps  finishing cap %p removal\n",
 957                              cap);
 958                         BUG_ON(cap->session != session);
 959                         list_del_init(&cap->session_caps);
 960                         session->s_nr_caps--;
 961                         cap->session = NULL;
 962                         old_cap = cap;  /* put_cap it w/o locks held */
 963                 }
 964                 if (ret < 0)
 965                         goto out;
 966         }
 967         ret = 0;
 968 out:
 969         session->s_cap_iterator = NULL;
 970         spin_unlock(&session->s_cap_lock);
 971
 972         if (last_inode)
 973                 iput(last_inode);
 974         if (old_cap)
 975                 ceph_put_cap(session->s_mdsc, old_cap);
 976
 977         return ret;
 978 }
 979
 980 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
 981                                   void *arg)
 982 {
 983         struct ceph_inode_info *ci = ceph_inode(inode);
 984         int drop = 0;
 985
 986         dout("removing cap %p, ci is %p, inode is %p\n",
 987              cap, ci, &ci->vfs_inode);
 988         spin_lock(&ci->i_ceph_lock);
 989         __ceph_remove_cap(cap);
 990         if (!__ceph_is_any_real_caps(ci)) {
 991                 struct ceph_mds_client *mdsc =
 992                         ceph_sb_to_client(inode->i_sb)->mdsc;
 993
 994                 spin_lock(&mdsc->cap_dirty_lock);
 995                 if (!list_empty(&ci->i_dirty_item)) {
 996                         pr_info(" dropping dirty %s state for %p %lld\n",
 997                                 ceph_cap_string(ci->i_dirty_caps),
 998                                 inode, ceph_ino(inode));
 999                         ci->i_dirty_caps = 0;
1000                         list_del_init(&ci->i_dirty_item);
1001                         drop = 1;
1002                 }
1003                 if (!list_empty(&ci->i_flushing_item)) {
1004                         pr_info(" dropping dirty+flushing %s state for %p %lld\n",
1005                                 ceph_cap_string(ci->i_flushing_caps),
1006                                 inode, ceph_ino(inode));
1007                         ci->i_flushing_caps = 0;
1008                         list_del_init(&ci->i_flushing_item);
1009                         mdsc->num_cap_flushing--;
1010                         drop = 1;
1011                 }
1012                 if (drop && ci->i_wrbuffer_ref) {
1013                         pr_info(" dropping dirty data for %p %lld\n",
1014                                 inode, ceph_ino(inode));
1015                         ci->i_wrbuffer_ref = 0;
1016                         ci->i_wrbuffer_ref_head = 0;
1017                         drop++;
1018                 }
1019                 spin_unlock(&mdsc->cap_dirty_lock);
1020         }
1021         spin_unlock(&ci->i_ceph_lock);
1022         while (drop--)
1023                 iput(inode);
1024         return 0;
1025 }
1026
1027 /*
1028  * caller must hold session s_mutex
1029  */
1030 static void remove_session_caps(struct ceph_mds_session *session)
1031 {
1032         dout("remove_session_caps on %p\n", session);
1033         iterate_session_caps(session, remove_session_caps_cb, NULL);
1034
1035         spin_lock(&session->s_cap_lock);
1036         if (session->s_nr_caps > 0) {
1037                 struct super_block *sb = session->s_mdsc->fsc->sb;
1038                 struct inode *inode;
1039                 struct ceph_cap *cap, *prev = NULL;
1040                 struct ceph_vino vino;
1041                 /*
1042                  * iterate_session_caps() skips inodes that are being
1043                  * deleted, we need to wait until deletions are complete.
1044                  * __wait_on_freeing_inode() is designed for the job,
1045                  * but it is not exported, so use lookup inode function
1046                  * to access it.
1047                  */
1048                 while (!list_empty(&session->s_caps)) {
1049                         cap = list_entry(session->s_caps.next,
1050                                          struct ceph_cap, session_caps);
1051                         if (cap == prev)
1052                                 break;
1053                         prev = cap;
1054                         vino = cap->ci->i_vino;
1055                         spin_unlock(&session->s_cap_lock);
1056
1057                         inode = ceph_lookup_inode(sb, vino);
1058                         iput(inode);
1059
1060                         spin_lock(&session->s_cap_lock);
1061                 }
1062         }
1063         spin_unlock(&session->s_cap_lock);
1064
1065         BUG_ON(session->s_nr_caps > 0);
1066         BUG_ON(!list_empty(&session->s_cap_flushing));
1067         cleanup_cap_releases(session);
1068 }
1069
1070 /*
1071  * wake up any threads waiting on this session's caps.  if the cap is
1072  * old (didn't get renewed on the client reconnect), remove it now.
1073  *
1074  * caller must hold s_mutex.
1075  */
1076 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1077                               void *arg)
1078 {
1079         struct ceph_inode_info *ci = ceph_inode(inode);
1080
1081         wake_up_all(&ci->i_cap_wq);
1082         if (arg) {
1083                 spin_lock(&ci->i_ceph_lock);
1084                 ci->i_wanted_max_size = 0;
1085                 ci->i_requested_max_size = 0;
1086                 spin_unlock(&ci->i_ceph_lock);
1087         }
1088         return 0;
1089 }
1090
1091 static void wake_up_session_caps(struct ceph_mds_session *session,
1092                                  int reconnect)
1093 {
1094         dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1095         iterate_session_caps(session, wake_up_session_cb,
1096                              (void *)(unsigned long)reconnect);
1097 }
1098
1099 /*
1100  * Send periodic message to MDS renewing all currently held caps.  The
1101  * ack will reset the expiration for all caps from this session.
1102  *
1103  * caller holds s_mutex
1104  */
1105 static int send_renew_caps(struct ceph_mds_client *mdsc,
1106                            struct ceph_mds_session *session)
1107 {
1108         struct ceph_msg *msg;
1109         int state;
1110
1111         if (time_after_eq(jiffies, session->s_cap_ttl) &&
1112             time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1113                 pr_info("mds%d caps stale\n", session->s_mds);
1114         session->s_renew_requested = jiffies;
1115
1116         /* do not try to renew caps until a recovering mds has reconnected
1117          * with its clients. */
1118         state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1119         if (state < CEPH_MDS_STATE_RECONNECT) {
1120                 dout("send_renew_caps ignoring mds%d (%s)\n",
1121                      session->s_mds, ceph_mds_state_name(state));
1122                 return 0;
1123         }
1124
1125         dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1126                 ceph_mds_state_name(state));
1127         msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1128                                  ++session->s_renew_seq);
1129         if (!msg)
1130                 return -ENOMEM;
1131         ceph_con_send(&session->s_con, msg);
1132         return 0;
1133 }
1134
1135 /*
1136  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1137  *
1138  * Called under session->s_mutex
1139  */
1140 static void renewed_caps(struct ceph_mds_client *mdsc,
1141                          struct ceph_mds_session *session, int is_renew)
1142 {
1143         int was_stale;
1144         int wake = 0;
1145
1146         spin_lock(&session->s_cap_lock);
1147         was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1148
1149         session->s_cap_ttl = session->s_renew_requested +
1150                 mdsc->mdsmap->m_session_timeout*HZ;
1151
1152         if (was_stale) {
1153                 if (time_before(jiffies, session->s_cap_ttl)) {
1154                         pr_info("mds%d caps renewed\n", session->s_mds);
1155                         wake = 1;
1156                 } else {
1157                         pr_info("mds%d caps still stale\n", session->s_mds);
1158                 }
1159         }
1160         dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1161              session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1162              time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1163         spin_unlock(&session->s_cap_lock);
1164
1165         if (wake)
1166                 wake_up_session_caps(session, 0);
1167 }
1168
1169 /*
1170  * send a session close request
1171  */
1172 static int request_close_session(struct ceph_mds_client *mdsc,
1173                                  struct ceph_mds_session *session)
1174 {
1175         struct ceph_msg *msg;
1176
1177         dout("request_close_session mds%d state %s seq %lld\n",
1178              session->s_mds, session_state_name(session->s_state),
1179              session->s_seq);
1180         msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1181         if (!msg)
1182                 return -ENOMEM;
1183         ceph_con_send(&session->s_con, msg);
1184         return 0;
1185 }
1186
1187 /*
1188  * Called with s_mutex held.
1189  */
1190 static int __close_session(struct ceph_mds_client *mdsc,
1191                          struct ceph_mds_session *session)
1192 {
1193         if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1194                 return 0;
1195         session->s_state = CEPH_MDS_SESSION_CLOSING;
1196         return request_close_session(mdsc, session);
1197 }
1198
1199 /*
1200  * Trim old(er) caps.
1201  *
1202  * Because we can't cache an inode without one or more caps, we do
1203  * this indirectly: if a cap is unused, we prune its aliases, at which
1204  * point the inode will hopefully get dropped to.
1205  *
1206  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1207  * memory pressure from the MDS, though, so it needn't be perfect.
1208  */
1209 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1210 {
1211         struct ceph_mds_session *session = arg;
1212         struct ceph_inode_info *ci = ceph_inode(inode);
1213         int used, oissued, mine;
1214
1215         if (session->s_trim_caps <= 0)
1216                 return -1;
1217
1218         spin_lock(&ci->i_ceph_lock);
1219         mine = cap->issued | cap->implemented;
1220         used = __ceph_caps_used(ci);
1221         oissued = __ceph_caps_issued_other(ci, cap);
1222
1223         dout("trim_caps_cb %p cap %p mine %s oissued %s used %s\n",
1224              inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1225              ceph_cap_string(used));
1226         if (ci->i_dirty_caps)
1227                 goto out;   /* dirty caps */
1228         if ((used & ~oissued) & mine)
1229                 goto out;   /* we need these caps */
1230
1231         session->s_trim_caps--;
1232         if (oissued) {
1233                 /* we aren't the only cap.. just remove us */
1234                 __queue_cap_release(session, ceph_ino(inode), cap->cap_id,
1235                                     cap->mseq, cap->issue_seq);
1236                 __ceph_remove_cap(cap);
1237         } else {
1238                 /* try to drop referring dentries */
1239                 spin_unlock(&ci->i_ceph_lock);
1240                 d_prune_aliases(inode);
1241                 dout("trim_caps_cb %p cap %p  pruned, count now %d\n",
1242                      inode, cap, atomic_read(&inode->i_count));
1243                 return 0;
1244         }
1245
1246 out:
1247         spin_unlock(&ci->i_ceph_lock);
1248         return 0;
1249 }
1250
1251 /*
1252  * Trim session cap count down to some max number.
1253  */
1254 static int trim_caps(struct ceph_mds_client *mdsc,
1255                      struct ceph_mds_session *session,
1256                      int max_caps)
1257 {
1258         int trim_caps = session->s_nr_caps - max_caps;
1259
1260         dout("trim_caps mds%d start: %d / %d, trim %d\n",
1261              session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1262         if (trim_caps > 0) {
1263                 session->s_trim_caps = trim_caps;
1264                 iterate_session_caps(session, trim_caps_cb, session);
1265                 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1266                      session->s_mds, session->s_nr_caps, max_caps,
1267                         trim_caps - session->s_trim_caps);
1268                 session->s_trim_caps = 0;
1269         }
1270         return 0;
1271 }
1272
1273 /*
1274  * Allocate cap_release messages.  If there is a partially full message
1275  * in the queue, try to allocate enough to cover it's remainder, so that
1276  * we can send it immediately.
1277  *
1278  * Called under s_mutex.
1279  */
1280 int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
1281                           struct ceph_mds_session *session)
1282 {
1283         struct ceph_msg *msg, *partial = NULL;
1284         struct ceph_mds_cap_release *head;
1285         int err = -ENOMEM;
1286         int extra = mdsc->fsc->mount_options->cap_release_safety;
1287         int num;
1288
1289         dout("add_cap_releases %p mds%d extra %d\n", session, session->s_mds,
1290              extra);
1291
1292         spin_lock(&session->s_cap_lock);
1293
1294         if (!list_empty(&session->s_cap_releases)) {
1295                 msg = list_first_entry(&session->s_cap_releases,
1296                                        struct ceph_msg,
1297                                  list_head);
1298                 head = msg->front.iov_base;
1299                 num = le32_to_cpu(head->num);
1300                 if (num) {
1301                         dout(" partial %p with (%d/%d)\n", msg, num,
1302                              (int)CEPH_CAPS_PER_RELEASE);
1303                         extra += CEPH_CAPS_PER_RELEASE - num;
1304                         partial = msg;
1305                 }
1306         }
1307         while (session->s_num_cap_releases < session->s_nr_caps + extra) {
1308                 spin_unlock(&session->s_cap_lock);
1309                 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
1310                                    GFP_NOFS, false);
1311                 if (!msg)
1312                         goto out_unlocked;
1313                 dout("add_cap_releases %p msg %p now %d\n", session, msg,
1314                      (int)msg->front.iov_len);
1315                 head = msg->front.iov_base;
1316                 head->num = cpu_to_le32(0);
1317                 msg->front.iov_len = sizeof(*head);
1318                 spin_lock(&session->s_cap_lock);
1319                 list_add(&msg->list_head, &session->s_cap_releases);
1320                 session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE;
1321         }
1322
1323         if (partial) {
1324                 head = partial->front.iov_base;
1325                 num = le32_to_cpu(head->num);
1326                 dout(" queueing partial %p with %d/%d\n", partial, num,
1327                      (int)CEPH_CAPS_PER_RELEASE);
1328                 list_move_tail(&partial->list_head,
1329                                &session->s_cap_releases_done);
1330                 session->s_num_cap_releases -= CEPH_CAPS_PER_RELEASE - num;
1331         }
1332         err = 0;
1333         spin_unlock(&session->s_cap_lock);
1334 out_unlocked:
1335         return err;
1336 }
1337
1338 /*
1339  * flush all dirty inode data to disk.
1340  *
1341  * returns true if we've flushed through want_flush_seq
1342  */
1343 static int check_cap_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq)
1344 {
1345         int mds, ret = 1;
1346
1347         dout("check_cap_flush want %lld\n", want_flush_seq);
1348         mutex_lock(&mdsc->mutex);
1349         for (mds = 0; ret && mds < mdsc->max_sessions; mds++) {
1350                 struct ceph_mds_session *session = mdsc->sessions[mds];
1351
1352                 if (!session)
1353                         continue;
1354                 get_session(session);
1355                 mutex_unlock(&mdsc->mutex);
1356
1357                 mutex_lock(&session->s_mutex);
1358                 if (!list_empty(&session->s_cap_flushing)) {
1359                         struct ceph_inode_info *ci =
1360                                 list_entry(session->s_cap_flushing.next,
1361                                            struct ceph_inode_info,
1362                                            i_flushing_item);
1363                         struct inode *inode = &ci->vfs_inode;
1364
1365                         spin_lock(&ci->i_ceph_lock);
1366                         if (ci->i_cap_flush_seq <= want_flush_seq) {
1367                                 dout("check_cap_flush still flushing %p "
1368                                      "seq %lld <= %lld to mds%d\n", inode,
1369                                      ci->i_cap_flush_seq, want_flush_seq,
1370                                      session->s_mds);
1371                                 ret = 0;
1372                         }
1373                         spin_unlock(&ci->i_ceph_lock);
1374                 }
1375                 mutex_unlock(&session->s_mutex);
1376                 ceph_put_mds_session(session);
1377
1378                 if (!ret)
1379                         return ret;
1380                 mutex_lock(&mdsc->mutex);
1381         }
1382
1383         mutex_unlock(&mdsc->mutex);
1384         dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
1385         return ret;
1386 }
1387
1388 /*
1389  * called under s_mutex
1390  */
1391 void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1392                             struct ceph_mds_session *session)
1393 {
1394         struct ceph_msg *msg;
1395
1396         dout("send_cap_releases mds%d\n", session->s_mds);
1397         spin_lock(&session->s_cap_lock);
1398         while (!list_empty(&session->s_cap_releases_done)) {
1399                 msg = list_first_entry(&session->s_cap_releases_done,
1400                                  struct ceph_msg, list_head);
1401                 list_del_init(&msg->list_head);
1402                 spin_unlock(&session->s_cap_lock);
1403                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1404                 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1405                 ceph_con_send(&session->s_con, msg);
1406                 spin_lock(&session->s_cap_lock);
1407         }
1408         spin_unlock(&session->s_cap_lock);
1409 }
1410
1411 static void discard_cap_releases(struct ceph_mds_client *mdsc,
1412                                  struct ceph_mds_session *session)
1413 {
1414         struct ceph_msg *msg;
1415         struct ceph_mds_cap_release *head;
1416         unsigned num;
1417
1418         dout("discard_cap_releases mds%d\n", session->s_mds);
1419         spin_lock(&session->s_cap_lock);
1420
1421         /* zero out the in-progress message */
1422         msg = list_first_entry(&session->s_cap_releases,
1423                                struct ceph_msg, list_head);
1424         head = msg->front.iov_base;
1425         num = le32_to_cpu(head->num);
1426         dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg, num);
1427         head->num = cpu_to_le32(0);
1428         msg->front.iov_len = sizeof(*head);
1429         session->s_num_cap_releases += num;
1430
1431         /* requeue completed messages */
1432         while (!list_empty(&session->s_cap_releases_done)) {
1433                 msg = list_first_entry(&session->s_cap_releases_done,
1434                                  struct ceph_msg, list_head);
1435                 list_del_init(&msg->list_head);
1436
1437                 head = msg->front.iov_base;
1438                 num = le32_to_cpu(head->num);
1439                 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg,
1440                      num);
1441                 session->s_num_cap_releases += num;
1442                 head->num = cpu_to_le32(0);
1443                 msg->front.iov_len = sizeof(*head);
1444                 list_add(&msg->list_head, &session->s_cap_releases);
1445         }
1446
1447         spin_unlock(&session->s_cap_lock);
1448 }
1449
1450 /*
1451  * requests
1452  */
1453
1454 /*
1455  * Create an mds request.
1456  */
1457 struct ceph_mds_request *
1458 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1459 {
1460         struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1461
1462         if (!req)
1463                 return ERR_PTR(-ENOMEM);
1464
1465         mutex_init(&req->r_fill_mutex);
1466         req->r_mdsc = mdsc;
1467         req->r_started = jiffies;
1468         req->r_resend_mds = -1;
1469         INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1470         req->r_fmode = -1;
1471         kref_init(&req->r_kref);
1472         INIT_LIST_HEAD(&req->r_wait);
1473         init_completion(&req->r_completion);
1474         init_completion(&req->r_safe_completion);
1475         INIT_LIST_HEAD(&req->r_unsafe_item);
1476
1477         req->r_op = op;
1478         req->r_direct_mode = mode;
1479         return req;
1480 }
1481
1482 /*
1483  * return oldest (lowest) request, tid in request tree, 0 if none.
1484  *
1485  * called under mdsc->mutex.
1486  */
1487 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1488 {
1489         if (RB_EMPTY_ROOT(&mdsc->request_tree))
1490                 return NULL;
1491         return rb_entry(rb_first(&mdsc->request_tree),
1492                         struct ceph_mds_request, r_node);
1493 }
1494
1495 static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1496 {
1497         struct ceph_mds_request *req = __get_oldest_req(mdsc);
1498
1499         if (req)
1500                 return req->r_tid;
1501         return 0;
1502 }
1503
1504 /*
1505  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
1506  * on build_path_from_dentry in fs/cifs/dir.c.
1507  *
1508  * If @stop_on_nosnap, generate path relative to the first non-snapped
1509  * inode.
1510  *
1511  * Encode hidden .snap dirs as a double /, i.e.
1512  *   foo/.snap/bar -> foo//bar
1513  */
1514 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1515                            int stop_on_nosnap)
1516 {
1517         struct dentry *temp;
1518         char *path;
1519         int len, pos;
1520         unsigned seq;
1521
1522         if (dentry == NULL)
1523                 return ERR_PTR(-EINVAL);
1524
1525 retry:
1526         len = 0;
1527         seq = read_seqbegin(&rename_lock);
1528         rcu_read_lock();
1529         for (temp = dentry; !IS_ROOT(temp);) {
1530                 struct inode *inode = temp->d_inode;
1531                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1532                         len++;  /* slash only */
1533                 else if (stop_on_nosnap && inode &&
1534                          ceph_snap(inode) == CEPH_NOSNAP)
1535                         break;
1536                 else
1537                         len += 1 + temp->d_name.len;
1538                 temp = temp->d_parent;
1539         }
1540         rcu_read_unlock();
1541         if (len)
1542                 len--;  /* no leading '/' */
1543
1544         path = kmalloc(len+1, GFP_NOFS);
1545         if (path == NULL)
1546                 return ERR_PTR(-ENOMEM);
1547         pos = len;
1548         path[pos] = 0;  /* trailing null */
1549         rcu_read_lock();
1550         for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1551                 struct inode *inode;
1552
1553                 spin_lock(&temp->d_lock);
1554                 inode = temp->d_inode;
1555                 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1556                         dout("build_path path+%d: %p SNAPDIR\n",
1557                              pos, temp);
1558                 } else if (stop_on_nosnap && inode &&
1559                            ceph_snap(inode) == CEPH_NOSNAP) {
1560                         spin_unlock(&temp->d_lock);
1561                         break;
1562                 } else {
1563                         pos -= temp->d_name.len;
1564                         if (pos < 0) {
1565                                 spin_unlock(&temp->d_lock);
1566                                 break;
1567                         }
1568                         strncpy(path + pos, temp->d_name.name,
1569                                 temp->d_name.len);
1570                 }
1571                 spin_unlock(&temp->d_lock);
1572                 if (pos)
1573                         path[--pos] = '/';
1574                 temp = temp->d_parent;
1575         }
1576         rcu_read_unlock();
1577         if (pos != 0 || read_seqretry(&rename_lock, seq)) {
1578                 pr_err("build_path did not end path lookup where "
1579                        "expected, namelen is %d, pos is %d\n", len, pos);
1580                 /* presumably this is only possible if racing with a
1581                    rename of one of the parent directories (we can not
1582                    lock the dentries above us to prevent this, but
1583                    retrying should be harmless) */
1584                 kfree(path);
1585                 goto retry;
1586         }
1587
1588         *base = ceph_ino(temp->d_inode);
1589         *plen = len;
1590         dout("build_path on %p %d built %llx '%.*s'\n",
1591              dentry, d_count(dentry), *base, len, path);
1592         return path;
1593 }
1594
1595 static int build_dentry_path(struct dentry *dentry,
1596                              const char **ppath, int *ppathlen, u64 *pino,
1597                              int *pfreepath)
1598 {
1599         char *path;
1600
1601         if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) {
1602                 *pino = ceph_ino(dentry->d_parent->d_inode);
1603                 *ppath = dentry->d_name.name;
1604                 *ppathlen = dentry->d_name.len;
1605                 return 0;
1606         }
1607         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1608         if (IS_ERR(path))
1609                 return PTR_ERR(path);
1610         *ppath = path;
1611         *pfreepath = 1;
1612         return 0;
1613 }
1614
1615 static int build_inode_path(struct inode *inode,
1616                             const char **ppath, int *ppathlen, u64 *pino,
1617                             int *pfreepath)
1618 {
1619         struct dentry *dentry;
1620         char *path;
1621
1622         if (ceph_snap(inode) == CEPH_NOSNAP) {
1623                 *pino = ceph_ino(inode);
1624                 *ppathlen = 0;
1625                 return 0;
1626         }
1627         dentry = d_find_alias(inode);
1628         path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1629         dput(dentry);
1630         if (IS_ERR(path))
1631                 return PTR_ERR(path);
1632         *ppath = path;
1633         *pfreepath = 1;
1634         return 0;
1635 }
1636
1637 /*
1638  * request arguments may be specified via an inode *, a dentry *, or
1639  * an explicit ino+path.
1640  */
1641 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1642                                   const char *rpath, u64 rino,
1643                                   const char **ppath, int *pathlen,
1644                                   u64 *ino, int *freepath)
1645 {
1646         int r = 0;
1647
1648         if (rinode) {
1649                 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1650                 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1651                      ceph_snap(rinode));
1652         } else if (rdentry) {
1653                 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1654                 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1655                      *ppath);
1656         } else if (rpath || rino) {
1657                 *ino = rino;
1658                 *ppath = rpath;
1659                 *pathlen = rpath ? strlen(rpath) : 0;
1660                 dout(" path %.*s\n", *pathlen, rpath);
1661         }
1662
1663         return r;
1664 }
1665
1666 /*
1667  * called under mdsc->mutex
1668  */
1669 static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1670                                                struct ceph_mds_request *req,
1671                                                int mds)
1672 {
1673         struct ceph_msg *msg;
1674         struct ceph_mds_request_head *head;
1675         const char *path1 = NULL;
1676         const char *path2 = NULL;
1677         u64 ino1 = 0, ino2 = 0;
1678         int pathlen1 = 0, pathlen2 = 0;
1679         int freepath1 = 0, freepath2 = 0;
1680         int len;
1681         u16 releases;
1682         void *p, *end;
1683         int ret;
1684
1685         ret = set_request_path_attr(req->r_inode, req->r_dentry,
1686                               req->r_path1, req->r_ino1.ino,
1687                               &path1, &pathlen1, &ino1, &freepath1);
1688         if (ret < 0) {
1689                 msg = ERR_PTR(ret);
1690                 goto out;
1691         }
1692
1693         ret = set_request_path_attr(NULL, req->r_old_dentry,
1694                               req->r_path2, req->r_ino2.ino,
1695                               &path2, &pathlen2, &ino2, &freepath2);
1696         if (ret < 0) {
1697                 msg = ERR_PTR(ret);
1698                 goto out_free1;
1699         }
1700
1701         len = sizeof(*head) +
1702                 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64));
1703
1704         /* calculate (max) length for cap releases */
1705         len += sizeof(struct ceph_mds_request_release) *
1706                 (!!req->r_inode_drop + !!req->r_dentry_drop +
1707                  !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1708         if (req->r_dentry_drop)
1709                 len += req->r_dentry->d_name.len;
1710         if (req->r_old_dentry_drop)
1711                 len += req->r_old_dentry->d_name.len;
1712
1713         msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
1714         if (!msg) {
1715                 msg = ERR_PTR(-ENOMEM);
1716                 goto out_free2;
1717         }
1718
1719         msg->hdr.tid = cpu_to_le64(req->r_tid);
1720
1721         head = msg->front.iov_base;
1722         p = msg->front.iov_base + sizeof(*head);
1723         end = msg->front.iov_base + msg->front.iov_len;
1724
1725         head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1726         head->op = cpu_to_le32(req->r_op);
1727         head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1728         head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
1729         head->args = req->r_args;
1730
1731         ceph_encode_filepath(&p, end, ino1, path1);
1732         ceph_encode_filepath(&p, end, ino2, path2);
1733
1734         /* make note of release offset, in case we need to replay */
1735         req->r_request_release_offset = p - msg->front.iov_base;
1736
1737         /* cap releases */
1738         releases = 0;
1739         if (req->r_inode_drop)
1740                 releases += ceph_encode_inode_release(&p,
1741                       req->r_inode ? req->r_inode : req->r_dentry->d_inode,
1742                       mds, req->r_inode_drop, req->r_inode_unless, 0);
1743         if (req->r_dentry_drop)
1744                 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1745                        mds, req->r_dentry_drop, req->r_dentry_unless);
1746         if (req->r_old_dentry_drop)
1747                 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1748                        mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1749         if (req->r_old_inode_drop)
1750                 releases += ceph_encode_inode_release(&p,
1751                       req->r_old_dentry->d_inode,
1752                       mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1753         head->num_releases = cpu_to_le16(releases);
1754
1755         BUG_ON(p > end);
1756         msg->front.iov_len = p - msg->front.iov_base;
1757         msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1758
1759         if (req->r_data_len) {
1760                 /* outbound data set only by ceph_sync_setxattr() */
1761                 BUG_ON(!req->r_pages);
1762                 ceph_msg_data_add_pages(msg, req->r_pages, req->r_data_len, 0);
1763         }
1764
1765         msg->hdr.data_len = cpu_to_le32(req->r_data_len);
1766         msg->hdr.data_off = cpu_to_le16(0);
1767
1768 out_free2:
1769         if (freepath2)
1770                 kfree((char *)path2);
1771 out_free1:
1772         if (freepath1)
1773                 kfree((char *)path1);
1774 out:
1775         return msg;
1776 }
1777
1778 /*
1779  * called under mdsc->mutex if error, under no mutex if
1780  * success.
1781  */
1782 static void complete_request(struct ceph_mds_client *mdsc,
1783                              struct ceph_mds_request *req)
1784 {
1785         if (req->r_callback)
1786                 req->r_callback(mdsc, req);
1787         else
1788                 complete_all(&req->r_completion);
1789 }
1790
1791 /*
1792  * called under mdsc->mutex
1793  */
1794 static int __prepare_send_request(struct ceph_mds_client *mdsc,
1795                                   struct ceph_mds_request *req,
1796                                   int mds)
1797 {
1798         struct ceph_mds_request_head *rhead;
1799         struct ceph_msg *msg;
1800         int flags = 0;
1801
1802         req->r_attempts++;
1803         if (req->r_inode) {
1804                 struct ceph_cap *cap =
1805                         ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
1806
1807                 if (cap)
1808                         req->r_sent_on_mseq = cap->mseq;
1809                 else
1810                         req->r_sent_on_mseq = -1;
1811         }
1812         dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
1813              req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
1814
1815         if (req->r_got_unsafe) {
1816                 /*
1817                  * Replay.  Do not regenerate message (and rebuild
1818                  * paths, etc.); just use the original message.
1819                  * Rebuilding paths will break for renames because
1820                  * d_move mangles the src name.
1821                  */
1822                 msg = req->r_request;
1823                 rhead = msg->front.iov_base;
1824
1825                 flags = le32_to_cpu(rhead->flags);
1826                 flags |= CEPH_MDS_FLAG_REPLAY;
1827                 rhead->flags = cpu_to_le32(flags);
1828
1829                 if (req->r_target_inode)
1830                         rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
1831
1832                 rhead->num_retry = req->r_attempts - 1;
1833
1834                 /* remove cap/dentry releases from message */
1835                 rhead->num_releases = 0;
1836                 msg->hdr.front_len = cpu_to_le32(req->r_request_release_offset);
1837                 msg->front.iov_len = req->r_request_release_offset;
1838                 return 0;
1839         }
1840
1841         if (req->r_request) {
1842                 ceph_msg_put(req->r_request);
1843                 req->r_request = NULL;
1844         }
1845         msg = create_request_message(mdsc, req, mds);
1846         if (IS_ERR(msg)) {
1847                 req->r_err = PTR_ERR(msg);
1848                 complete_request(mdsc, req);
1849                 return PTR_ERR(msg);
1850         }
1851         req->r_request = msg;
1852
1853         rhead = msg->front.iov_base;
1854         rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
1855         if (req->r_got_unsafe)
1856                 flags |= CEPH_MDS_FLAG_REPLAY;
1857         if (req->r_locked_dir)
1858                 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
1859         rhead->flags = cpu_to_le32(flags);
1860         rhead->num_fwd = req->r_num_fwd;
1861         rhead->num_retry = req->r_attempts - 1;
1862         rhead->ino = 0;
1863
1864         dout(" r_locked_dir = %p\n", req->r_locked_dir);
1865         return 0;
1866 }
1867
1868 /*
1869  * send request, or put it on the appropriate wait list.
1870  */
1871 static int __do_request(struct ceph_mds_client *mdsc,
1872                         struct ceph_mds_request *req)
1873 {
1874         struct ceph_mds_session *session = NULL;
1875         int mds = -1;
1876         int err = -EAGAIN;
1877
1878         if (req->r_err || req->r_got_result)
1879                 goto out;
1880
1881         if (req->r_timeout &&
1882             time_after_eq(jiffies, req->r_started + req->r_timeout)) {
1883                 dout("do_request timed out\n");
1884                 err = -EIO;
1885                 goto finish;
1886         }
1887
1888         put_request_session(req);
1889
1890         mds = __choose_mds(mdsc, req);
1891         if (mds < 0 ||
1892             ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
1893                 dout("do_request no mds or not active, waiting for map\n");
1894                 list_add(&req->r_wait, &mdsc->waiting_for_map);
1895                 goto out;
1896         }
1897
1898         /* get, open session */
1899         session = __ceph_lookup_mds_session(mdsc, mds);
1900         if (!session) {
1901                 session = register_session(mdsc, mds);
1902                 if (IS_ERR(session)) {
1903                         err = PTR_ERR(session);
1904                         goto finish;
1905                 }
1906         }
1907         req->r_session = get_session(session);
1908
1909         dout("do_request mds%d session %p state %s\n", mds, session,
1910              session_state_name(session->s_state));
1911         if (session->s_state != CEPH_MDS_SESSION_OPEN &&
1912             session->s_state != CEPH_MDS_SESSION_HUNG) {
1913                 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1914                     session->s_state == CEPH_MDS_SESSION_CLOSING)
1915                         __open_session(mdsc, session);
1916                 list_add(&req->r_wait, &session->s_waiting);
1917                 goto out_session;
1918         }
1919
1920         /* send request */
1921         req->r_resend_mds = -1;   /* forget any previous mds hint */
1922
1923         if (req->r_request_started == 0)   /* note request start time */
1924                 req->r_request_started = jiffies;
1925
1926         err = __prepare_send_request(mdsc, req, mds);
1927         if (!err) {
1928                 ceph_msg_get(req->r_request);
1929                 ceph_con_send(&session->s_con, req->r_request);
1930         }
1931
1932 out_session:
1933         ceph_put_mds_session(session);
1934 out:
1935         return err;
1936
1937 finish:
1938         req->r_err = err;
1939         complete_request(mdsc, req);
1940         goto out;
1941 }
1942
1943 /*
1944  * called under mdsc->mutex
1945  */
1946 static void __wake_requests(struct ceph_mds_client *mdsc,
1947                             struct list_head *head)
1948 {
1949         struct ceph_mds_request *req;
1950         LIST_HEAD(tmp_list);
1951
1952         list_splice_init(head, &tmp_list);
1953
1954         while (!list_empty(&tmp_list)) {
1955                 req = list_entry(tmp_list.next,
1956                                  struct ceph_mds_request, r_wait);
1957                 list_del_init(&req->r_wait);
1958                 dout(" wake request %p tid %llu\n", req, req->r_tid);
1959                 __do_request(mdsc, req);
1960         }
1961 }
1962
1963 /*
1964  * Wake up threads with requests pending for @mds, so that they can
1965  * resubmit their requests to a possibly different mds.
1966  */
1967 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
1968 {
1969         struct ceph_mds_request *req;
1970         struct rb_node *p;
1971
1972         dout("kick_requests mds%d\n", mds);
1973         for (p = rb_first(&mdsc->request_tree); p; p = rb_next(p)) {
1974                 req = rb_entry(p, struct ceph_mds_request, r_node);
1975                 if (req->r_got_unsafe)
1976                         continue;
1977                 if (req->r_session &&
1978                     req->r_session->s_mds == mds) {
1979                         dout(" kicking tid %llu\n", req->r_tid);
1980                         __do_request(mdsc, req);
1981                 }
1982         }
1983 }
1984
1985 void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
1986                               struct ceph_mds_request *req)
1987 {
1988         dout("submit_request on %p\n", req);
1989         mutex_lock(&mdsc->mutex);
1990         __register_request(mdsc, req, NULL);
1991         __do_request(mdsc, req);
1992         mutex_unlock(&mdsc->mutex);
1993 }
1994
1995 /*
1996  * Synchrously perform an mds request.  Take care of all of the
1997  * session setup, forwarding, retry details.
1998  */
1999 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2000                          struct inode *dir,
2001                          struct ceph_mds_request *req)
2002 {
2003         int err;
2004
2005         dout("do_request on %p\n", req);
2006
2007         /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2008         if (req->r_inode)
2009                 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2010         if (req->r_locked_dir)
2011                 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
2012         if (req->r_old_dentry)
2013                 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2014                                   CEPH_CAP_PIN);
2015
2016         /* issue */
2017         mutex_lock(&mdsc->mutex);
2018         __register_request(mdsc, req, dir);
2019         __do_request(mdsc, req);
2020
2021         if (req->r_err) {
2022                 err = req->r_err;
2023                 __unregister_request(mdsc, req);
2024                 dout("do_request early error %d\n", err);
2025                 goto out;
2026         }
2027
2028         /* wait */
2029         mutex_unlock(&mdsc->mutex);
2030         dout("do_request waiting\n");
2031         if (req->r_timeout) {
2032                 err = (long)wait_for_completion_killable_timeout(
2033                         &req->r_completion, req->r_timeout);
2034                 if (err == 0)
2035                         err = -EIO;
2036         } else {
2037                 err = wait_for_completion_killable(&req->r_completion);
2038         }
2039         dout("do_request waited, got %d\n", err);
2040         mutex_lock(&mdsc->mutex);
2041
2042         /* only abort if we didn't race with a real reply */
2043         if (req->r_got_result) {
2044                 err = le32_to_cpu(req->r_reply_info.head->result);
2045         } else if (err < 0) {
2046                 dout("aborted request %lld with %d\n", req->r_tid, err);
2047
2048                 /*
2049                  * ensure we aren't running concurrently with
2050                  * ceph_fill_trace or ceph_readdir_prepopulate, which
2051                  * rely on locks (dir mutex) held by our caller.
2052                  */
2053                 mutex_lock(&req->r_fill_mutex);
2054                 req->r_err = err;
2055                 req->r_aborted = true;
2056                 mutex_unlock(&req->r_fill_mutex);
2057
2058                 if (req->r_locked_dir &&
2059                     (req->r_op & CEPH_MDS_OP_WRITE))
2060                         ceph_invalidate_dir_request(req);
2061         } else {
2062                 err = req->r_err;
2063         }
2064
2065 out:
2066         mutex_unlock(&mdsc->mutex);
2067         dout("do_request %p done, result %d\n", req, err);
2068         return err;
2069 }
2070
2071 /*
2072  * Invalidate dir's completeness, dentry lease state on an aborted MDS
2073  * namespace request.
2074  */
2075 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2076 {
2077         struct inode *inode = req->r_locked_dir;
2078
2079         dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
2080
2081         ceph_dir_clear_complete(inode);
2082         if (req->r_dentry)
2083                 ceph_invalidate_dentry_lease(req->r_dentry);
2084         if (req->r_old_dentry)
2085                 ceph_invalidate_dentry_lease(req->r_old_dentry);
2086 }
2087
2088 /*
2089  * Handle mds reply.
2090  *
2091  * We take the session mutex and parse and process the reply immediately.
2092  * This preserves the logical ordering of replies, capabilities, etc., sent
2093  * by the MDS as they are applied to our local cache.
2094  */
2095 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2096 {
2097         struct ceph_mds_client *mdsc = session->s_mdsc;
2098         struct ceph_mds_request *req;
2099         struct ceph_mds_reply_head *head = msg->front.iov_base;
2100         struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
2101         u64 tid;
2102         int err, result;
2103         int mds = session->s_mds;
2104
2105         if (msg->front.iov_len < sizeof(*head)) {
2106                 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2107                 ceph_msg_dump(msg);
2108                 return;
2109         }
2110
2111         /* get request, session */
2112         tid = le64_to_cpu(msg->hdr.tid);
2113         mutex_lock(&mdsc->mutex);
2114         req = __lookup_request(mdsc, tid);
2115         if (!req) {
2116                 dout("handle_reply on unknown tid %llu\n", tid);
2117                 mutex_unlock(&mdsc->mutex);
2118                 return;
2119         }
2120         dout("handle_reply %p\n", req);
2121
2122         /* correct session? */
2123         if (req->r_session != session) {
2124                 pr_err("mdsc_handle_reply got %llu on session mds%d"
2125                        " not mds%d\n", tid, session->s_mds,
2126                        req->r_session ? req->r_session->s_mds : -1);
2127                 mutex_unlock(&mdsc->mutex);
2128                 goto out;
2129         }
2130
2131         /* dup? */
2132         if ((req->r_got_unsafe && !head->safe) ||
2133             (req->r_got_safe && head->safe)) {
2134                 pr_warning("got a dup %s reply on %llu from mds%d\n",
2135                            head->safe ? "safe" : "unsafe", tid, mds);
2136                 mutex_unlock(&mdsc->mutex);
2137                 goto out;
2138         }
2139         if (req->r_got_safe && !head->safe) {
2140                 pr_warning("got unsafe after safe on %llu from mds%d\n",
2141                            tid, mds);
2142                 mutex_unlock(&mdsc->mutex);
2143                 goto out;
2144         }
2145
2146         result = le32_to_cpu(head->result);
2147
2148         /*
2149          * Handle an ESTALE
2150          * if we're not talking to the authority, send to them
2151          * if the authority has changed while we weren't looking,
2152          * send to new authority
2153          * Otherwise we just have to return an ESTALE
2154          */
2155         if (result == -ESTALE) {
2156                 dout("got ESTALE on request %llu", req->r_tid);
2157                 if (!req->r_inode) {
2158                         /* do nothing; not an authority problem */
2159                 } else if (req->r_direct_mode != USE_AUTH_MDS) {
2160                         dout("not using auth, setting for that now");
2161                         req->r_direct_mode = USE_AUTH_MDS;
2162                         __do_request(mdsc, req);
2163                         mutex_unlock(&mdsc->mutex);
2164                         goto out;
2165                 } else  {
2166                         struct ceph_inode_info *ci = ceph_inode(req->r_inode);
2167                         struct ceph_cap *cap = NULL;
2168
2169                         if (req->r_session)
2170                                 cap = ceph_get_cap_for_mds(ci,
2171                                                    req->r_session->s_mds);
2172
2173                         dout("already using auth");
2174                         if ((!cap || cap != ci->i_auth_cap) ||
2175                             (cap->mseq != req->r_sent_on_mseq)) {
2176                                 dout("but cap changed, so resending");
2177                                 __do_request(mdsc, req);
2178                                 mutex_unlock(&mdsc->mutex);
2179                                 goto out;
2180                         }
2181                 }
2182                 dout("have to return ESTALE on request %llu", req->r_tid);
2183         }
2184
2185
2186         if (head->safe) {
2187                 req->r_got_safe = true;
2188                 __unregister_request(mdsc, req);
2189                 complete_all(&req->r_safe_completion);
2190
2191                 if (req->r_got_unsafe) {
2192                         /*
2193                          * We already handled the unsafe response, now do the
2194                          * cleanup.  No need to examine the response; the MDS
2195                          * doesn't include any result info in the safe
2196                          * response.  And even if it did, there is nothing
2197                          * useful we could do with a revised return value.
2198                          */
2199                         dout("got safe reply %llu, mds%d\n", tid, mds);
2200                         list_del_init(&req->r_unsafe_item);
2201
2202                         /* last unsafe request during umount? */
2203                         if (mdsc->stopping && !__get_oldest_req(mdsc))
2204                                 complete_all(&mdsc->safe_umount_waiters);
2205                         mutex_unlock(&mdsc->mutex);
2206                         goto out;
2207                 }
2208         } else {
2209                 req->r_got_unsafe = true;
2210                 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2211         }
2212
2213         dout("handle_reply tid %lld result %d\n", tid, result);
2214         rinfo = &req->r_reply_info;
2215         err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2216         mutex_unlock(&mdsc->mutex);
2217
2218         mutex_lock(&session->s_mutex);
2219         if (err < 0) {
2220                 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
2221                 ceph_msg_dump(msg);
2222                 goto out_err;
2223         }
2224
2225         /* snap trace */
2226         if (rinfo->snapblob_len) {
2227                 down_write(&mdsc->snap_rwsem);
2228                 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2229                                rinfo->snapblob + rinfo->snapblob_len,
2230                                le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP);
2231                 downgrade_write(&mdsc->snap_rwsem);
2232         } else {
2233                 down_read(&mdsc->snap_rwsem);
2234         }
2235
2236         /* insert trace into our cache */
2237         mutex_lock(&req->r_fill_mutex);
2238         err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2239         if (err == 0) {
2240                 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
2241                                     req->r_op == CEPH_MDS_OP_LSSNAP) &&
2242                     rinfo->dir_nr)
2243                         ceph_readdir_prepopulate(req, req->r_session);
2244                 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2245         }
2246         mutex_unlock(&req->r_fill_mutex);
2247
2248         up_read(&mdsc->snap_rwsem);
2249 out_err:
2250         mutex_lock(&mdsc->mutex);
2251         if (!req->r_aborted) {
2252                 if (err) {
2253                         req->r_err = err;
2254                 } else {
2255                         req->r_reply = msg;
2256                         ceph_msg_get(msg);
2257                         req->r_got_result = true;
2258                 }
2259         } else {
2260                 dout("reply arrived after request %lld was aborted\n", tid);
2261         }
2262         mutex_unlock(&mdsc->mutex);
2263
2264         ceph_add_cap_releases(mdsc, req->r_session);
2265         mutex_unlock(&session->s_mutex);
2266
2267         /* kick calling process */
2268         complete_request(mdsc, req);
2269 out:
2270         ceph_mdsc_put_request(req);
2271         return;
2272 }
2273
2274
2275
2276 /*
2277  * handle mds notification that our request has been forwarded.
2278  */
2279 static void handle_forward(struct ceph_mds_client *mdsc,
2280                            struct ceph_mds_session *session,
2281                            struct ceph_msg *msg)
2282 {
2283         struct ceph_mds_request *req;
2284         u64 tid = le64_to_cpu(msg->hdr.tid);
2285         u32 next_mds;
2286         u32 fwd_seq;
2287         int err = -EINVAL;
2288         void *p = msg->front.iov_base;
2289         void *end = p + msg->front.iov_len;
2290
2291         ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2292         next_mds = ceph_decode_32(&p);
2293         fwd_seq = ceph_decode_32(&p);
2294
2295         mutex_lock(&mdsc->mutex);
2296         req = __lookup_request(mdsc, tid);
2297         if (!req) {
2298                 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2299                 goto out;  /* dup reply? */
2300         }
2301
2302         if (req->r_aborted) {
2303                 dout("forward tid %llu aborted, unregistering\n", tid);
2304                 __unregister_request(mdsc, req);
2305         } else if (fwd_seq <= req->r_num_fwd) {
2306                 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2307                      tid, next_mds, req->r_num_fwd, fwd_seq);
2308         } else {
2309                 /* resend. forward race not possible; mds would drop */
2310                 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2311                 BUG_ON(req->r_err);
2312                 BUG_ON(req->r_got_result);
2313                 req->r_num_fwd = fwd_seq;
2314                 req->r_resend_mds = next_mds;
2315                 put_request_session(req);
2316                 __do_request(mdsc, req);
2317         }
2318         ceph_mdsc_put_request(req);
2319 out:
2320         mutex_unlock(&mdsc->mutex);
2321         return;
2322
2323 bad:
2324         pr_err("mdsc_handle_forward decode error err=%d\n", err);
2325 }
2326
2327 /*
2328  * handle a mds session control message
2329  */
2330 static void handle_session(struct ceph_mds_session *session,
2331                            struct ceph_msg *msg)
2332 {
2333         struct ceph_mds_client *mdsc = session->s_mdsc;
2334         u32 op;
2335         u64 seq;
2336         int mds = session->s_mds;
2337         struct ceph_mds_session_head *h = msg->front.iov_base;
2338         int wake = 0;
2339
2340         /* decode */
2341         if (msg->front.iov_len != sizeof(*h))
2342                 goto bad;
2343         op = le32_to_cpu(h->op);
2344         seq = le64_to_cpu(h->seq);
2345
2346         mutex_lock(&mdsc->mutex);
2347         if (op == CEPH_SESSION_CLOSE)
2348                 __unregister_session(mdsc, session);
2349         /* FIXME: this ttl calculation is generous */
2350         session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2351         mutex_unlock(&mdsc->mutex);
2352
2353         mutex_lock(&session->s_mutex);
2354
2355         dout("handle_session mds%d %s %p state %s seq %llu\n",
2356              mds, ceph_session_op_name(op), session,
2357              session_state_name(session->s_state), seq);
2358
2359         if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2360                 session->s_state = CEPH_MDS_SESSION_OPEN;
2361                 pr_info("mds%d came back\n", session->s_mds);
2362         }
2363
2364         switch (op) {
2365         case CEPH_SESSION_OPEN:
2366                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2367                         pr_info("mds%d reconnect success\n", session->s_mds);
2368                 session->s_state = CEPH_MDS_SESSION_OPEN;
2369                 renewed_caps(mdsc, session, 0);
2370                 wake = 1;
2371                 if (mdsc->stopping)
2372                         __close_session(mdsc, session);
2373                 break;
2374
2375         case CEPH_SESSION_RENEWCAPS:
2376                 if (session->s_renew_seq == seq)
2377                         renewed_caps(mdsc, session, 1);
2378                 break;
2379
2380         case CEPH_SESSION_CLOSE:
2381                 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2382                         pr_info("mds%d reconnect denied\n", session->s_mds);
2383                 remove_session_caps(session);
2384                 wake = 1; /* for good measure */
2385                 wake_up_all(&mdsc->session_close_wq);
2386                 kick_requests(mdsc, mds);
2387                 break;
2388
2389         case CEPH_SESSION_STALE:
2390                 pr_info("mds%d caps went stale, renewing\n",
2391                         session->s_mds);
2392                 spin_lock(&session->s_gen_ttl_lock);
2393                 session->s_cap_gen++;
2394                 session->s_cap_ttl = jiffies - 1;
2395                 spin_unlock(&session->s_gen_ttl_lock);
2396                 send_renew_caps(mdsc, session);
2397                 break;
2398
2399         case CEPH_SESSION_RECALL_STATE:
2400                 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2401                 break;
2402
2403         default:
2404                 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2405                 WARN_ON(1);
2406         }
2407
2408         mutex_unlock(&session->s_mutex);
2409         if (wake) {
2410                 mutex_lock(&mdsc->mutex);
2411                 __wake_requests(mdsc, &session->s_waiting);
2412                 mutex_unlock(&mdsc->mutex);
2413         }
2414         return;
2415
2416 bad:
2417         pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2418                (int)msg->front.iov_len);
2419         ceph_msg_dump(msg);
2420         return;
2421 }
2422
2423
2424 /*
2425  * called under session->mutex.
2426  */
2427 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2428                                    struct ceph_mds_session *session)
2429 {
2430         struct ceph_mds_request *req, *nreq;
2431         int err;
2432
2433         dout("replay_unsafe_requests mds%d\n", session->s_mds);
2434
2435         mutex_lock(&mdsc->mutex);
2436         list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2437                 err = __prepare_send_request(mdsc, req, session->s_mds);
2438                 if (!err) {
2439                         ceph_msg_get(req->r_request);
2440                         ceph_con_send(&session->s_con, req->r_request);
2441                 }
2442         }
2443         mutex_unlock(&mdsc->mutex);
2444 }
2445
2446 /*
2447  * Encode information about a cap for a reconnect with the MDS.
2448  */
2449 static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2450                           void *arg)
2451 {
2452         union {
2453                 struct ceph_mds_cap_reconnect v2;
2454                 struct ceph_mds_cap_reconnect_v1 v1;
2455         } rec;
2456         size_t reclen;
2457         struct ceph_inode_info *ci;
2458         struct ceph_reconnect_state *recon_state = arg;
2459         struct ceph_pagelist *pagelist = recon_state->pagelist;
2460         char *path;
2461         int pathlen, err;
2462         u64 pathbase;
2463         struct dentry *dentry;
2464
2465         ci = cap->ci;
2466
2467         dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2468              inode, ceph_vinop(inode), cap, cap->cap_id,
2469              ceph_cap_string(cap->issued));
2470         err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2471         if (err)
2472                 return err;
2473
2474         dentry = d_find_alias(inode);
2475         if (dentry) {
2476                 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2477                 if (IS_ERR(path)) {
2478                         err = PTR_ERR(path);
2479                         goto out_dput;
2480                 }
2481         } else {
2482                 path = NULL;
2483                 pathlen = 0;
2484         }
2485         err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2486         if (err)
2487                 goto out_free;
2488
2489         spin_lock(&ci->i_ceph_lock);
2490         cap->seq = 0;        /* reset cap seq */
2491         cap->issue_seq = 0;  /* and issue_seq */
2492         cap->mseq = 0;       /* and migrate_seq */
2493
2494         if (recon_state->flock) {
2495                 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2496                 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2497                 rec.v2.issued = cpu_to_le32(cap->issued);
2498                 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2499                 rec.v2.pathbase = cpu_to_le64(pathbase);
2500                 rec.v2.flock_len = 0;
2501                 reclen = sizeof(rec.v2);
2502         } else {
2503                 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2504                 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2505                 rec.v1.issued = cpu_to_le32(cap->issued);
2506                 rec.v1.size = cpu_to_le64(inode->i_size);
2507                 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2508                 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2509                 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2510                 rec.v1.pathbase = cpu_to_le64(pathbase);
2511                 reclen = sizeof(rec.v1);
2512         }
2513         spin_unlock(&ci->i_ceph_lock);
2514
2515         if (recon_state->flock) {
2516                 int num_fcntl_locks, num_flock_locks;
2517                 struct ceph_filelock *flocks;
2518
2519 encode_again:
2520                 spin_lock(&inode->i_lock);
2521                 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
2522                 spin_unlock(&inode->i_lock);
2523                 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2524                                  sizeof(struct ceph_filelock), GFP_NOFS);
2525                 if (!flocks) {
2526                         err = -ENOMEM;
2527                         goto out_free;
2528                 }
2529                 spin_lock(&inode->i_lock);
2530                 err = ceph_encode_locks_to_buffer(inode, flocks,
2531                                                   num_fcntl_locks,
2532                                                   num_flock_locks);
2533                 spin_unlock(&inode->i_lock);
2534                 if (err) {
2535                         kfree(flocks);
2536                         if (err == -ENOSPC)
2537                                 goto encode_again;
2538                         goto out_free;
2539                 }
2540                 /*
2541                  * number of encoded locks is stable, so copy to pagelist
2542                  */
2543                 rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2544                                     (num_fcntl_locks+num_flock_locks) *
2545                                     sizeof(struct ceph_filelock));
2546                 err = ceph_pagelist_append(pagelist, &rec, reclen);
2547                 if (!err)
2548                         err = ceph_locks_to_pagelist(flocks, pagelist,
2549                                                      num_fcntl_locks,
2550                                                      num_flock_locks);
2551                 kfree(flocks);
2552         } else {
2553                 err = ceph_pagelist_append(pagelist, &rec, reclen);
2554         }
2555 out_free:
2556         kfree(path);
2557 out_dput:
2558         dput(dentry);
2559         return err;
2560 }
2561
2562
2563 /*
2564  * If an MDS fails and recovers, clients need to reconnect in order to
2565  * reestablish shared state.  This includes all caps issued through
2566  * this session _and_ the snap_realm hierarchy.  Because it's not
2567  * clear which snap realms the mds cares about, we send everything we
2568  * know about.. that ensures we'll then get any new info the
2569  * recovering MDS might have.
2570  *
2571  * This is a relatively heavyweight operation, but it's rare.
2572  *
2573  * called with mdsc->mutex held.
2574  */
2575 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2576                                struct ceph_mds_session *session)
2577 {
2578         struct ceph_msg *reply;
2579         struct rb_node *p;
2580         int mds = session->s_mds;
2581         int err = -ENOMEM;
2582         struct ceph_pagelist *pagelist;
2583         struct ceph_reconnect_state recon_state;
2584
2585         pr_info("mds%d reconnect start\n", mds);
2586
2587         pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2588         if (!pagelist)
2589                 goto fail_nopagelist;
2590         ceph_pagelist_init(pagelist);
2591
2592         reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
2593         if (!reply)
2594                 goto fail_nomsg;
2595
2596         mutex_lock(&session->s_mutex);
2597         session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2598         session->s_seq = 0;
2599
2600         ceph_con_close(&session->s_con);
2601         ceph_con_open(&session->s_con,
2602                       CEPH_ENTITY_TYPE_MDS, mds,
2603                       ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2604
2605         /* replay unsafe requests */
2606         replay_unsafe_requests(mdsc, session);
2607
2608         down_read(&mdsc->snap_rwsem);
2609
2610         dout("session %p state %s\n", session,
2611              session_state_name(session->s_state));
2612
2613         /* drop old cap expires; we're about to reestablish that state */
2614         discard_cap_releases(mdsc, session);
2615
2616         /* traverse this session's caps */
2617         err = ceph_pagelist_encode_32(pagelist, session->s_nr_caps);
2618         if (err)
2619                 goto fail;
2620
2621         recon_state.pagelist = pagelist;
2622         recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2623         err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2624         if (err < 0)
2625                 goto fail;
2626
2627         /*
2628          * snaprealms.  we provide mds with the ino, seq (version), and
2629          * parent for all of our realms.  If the mds has any newer info,
2630          * it will tell us.
2631          */
2632         for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2633                 struct ceph_snap_realm *realm =
2634                         rb_entry(p, struct ceph_snap_realm, node);
2635                 struct ceph_mds_snaprealm_reconnect sr_rec;
2636
2637                 dout(" adding snap realm %llx seq %lld parent %llx\n",
2638                      realm->ino, realm->seq, realm->parent_ino);
2639                 sr_rec.ino = cpu_to_le64(realm->ino);
2640                 sr_rec.seq = cpu_to_le64(realm->seq);
2641                 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2642                 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2643                 if (err)
2644                         goto fail;
2645         }
2646
2647         if (recon_state.flock)
2648                 reply->hdr.version = cpu_to_le16(2);
2649         if (pagelist->length) {
2650                 /* set up outbound data if we have any */
2651                 reply->hdr.data_len = cpu_to_le32(pagelist->length);
2652                 ceph_msg_data_add_pagelist(reply, pagelist);
2653         }
2654         ceph_con_send(&session->s_con, reply);
2655
2656         mutex_unlock(&session->s_mutex);
2657
2658         mutex_lock(&mdsc->mutex);
2659         __wake_requests(mdsc, &session->s_waiting);
2660         mutex_unlock(&mdsc->mutex);
2661
2662         up_read(&mdsc->snap_rwsem);
2663         return;
2664
2665 fail:
2666         ceph_msg_put(reply);
2667         up_read(&mdsc->snap_rwsem);
2668         mutex_unlock(&session->s_mutex);
2669 fail_nomsg:
2670         ceph_pagelist_release(pagelist);
2671         kfree(pagelist);
2672 fail_nopagelist:
2673         pr_err("error %d preparing reconnect for mds%d\n", err, mds);
2674         return;
2675 }
2676
2677
2678 /*
2679  * compare old and new mdsmaps, kicking requests
2680  * and closing out old connections as necessary
2681  *
2682  * called under mdsc->mutex.
2683  */
2684 static void check_new_map(struct ceph_mds_client *mdsc,
2685                           struct ceph_mdsmap *newmap,
2686                           struct ceph_mdsmap *oldmap)
2687 {
2688         int i;
2689         int oldstate, newstate;
2690         struct ceph_mds_session *s;
2691
2692         dout("check_new_map new %u old %u\n",
2693              newmap->m_epoch, oldmap->m_epoch);
2694
2695         for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2696                 if (mdsc->sessions[i] == NULL)
2697                         continue;
2698                 s = mdsc->sessions[i];
2699                 oldstate = ceph_mdsmap_get_state(oldmap, i);
2700                 newstate = ceph_mdsmap_get_state(newmap, i);
2701
2702                 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2703                      i, ceph_mds_state_name(oldstate),
2704                      ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2705                      ceph_mds_state_name(newstate),
2706                      ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
2707                      session_state_name(s->s_state));
2708
2709                 if (i >= newmap->m_max_mds ||
2710                     memcmp(ceph_mdsmap_get_addr(oldmap, i),
2711                            ceph_mdsmap_get_addr(newmap, i),
2712                            sizeof(struct ceph_entity_addr))) {
2713                         if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2714                                 /* the session never opened, just close it
2715                                  * out now */
2716                                 __wake_requests(mdsc, &s->s_waiting);
2717                                 __unregister_session(mdsc, s);
2718                         } else {
2719                                 /* just close it */
2720                                 mutex_unlock(&mdsc->mutex);
2721                                 mutex_lock(&s->s_mutex);
2722                                 mutex_lock(&mdsc->mutex);
2723                                 ceph_con_close(&s->s_con);
2724                                 mutex_unlock(&s->s_mutex);
2725                                 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2726                         }
2727
2728                         /* kick any requests waiting on the recovering mds */
2729                         kick_requests(mdsc, i);
2730                 } else if (oldstate == newstate) {
2731                         continue;  /* nothing new with this mds */
2732                 }
2733
2734                 /*
2735                  * send reconnect?
2736                  */
2737                 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
2738                     newstate >= CEPH_MDS_STATE_RECONNECT) {
2739                         mutex_unlock(&mdsc->mutex);
2740                         send_mds_reconnect(mdsc, s);
2741                         mutex_lock(&mdsc->mutex);
2742                 }
2743
2744                 /*
2745                  * kick request on any mds that has gone active.
2746                  */
2747                 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2748                     newstate >= CEPH_MDS_STATE_ACTIVE) {
2749                         if (oldstate != CEPH_MDS_STATE_CREATING &&
2750                             oldstate != CEPH_MDS_STATE_STARTING)
2751                                 pr_info("mds%d recovery completed\n", s->s_mds);
2752                         kick_requests(mdsc, i);
2753                         ceph_kick_flushing_caps(mdsc, s);
2754                         wake_up_session_caps(s, 1);
2755                 }
2756         }
2757
2758         for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
2759                 s = mdsc->sessions[i];
2760                 if (!s)
2761                         continue;
2762                 if (!ceph_mdsmap_is_laggy(newmap, i))
2763                         continue;
2764                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2765                     s->s_state == CEPH_MDS_SESSION_HUNG ||
2766                     s->s_state == CEPH_MDS_SESSION_CLOSING) {
2767                         dout(" connecting to export targets of laggy mds%d\n",
2768                              i);
2769                         __open_export_target_sessions(mdsc, s);
2770                 }
2771         }
2772 }
2773
2774
2775
2776 /*
2777  * leases
2778  */
2779
2780 /*
2781  * caller must hold session s_mutex, dentry->d_lock
2782  */
2783 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
2784 {
2785         struct ceph_dentry_info *di = ceph_dentry(dentry);
2786
2787         ceph_put_mds_session(di->lease_session);
2788         di->lease_session = NULL;
2789 }
2790
2791 static void handle_lease(struct ceph_mds_client *mdsc,
2792                          struct ceph_mds_session *session,
2793                          struct ceph_msg *msg)
2794 {
2795         struct super_block *sb = mdsc->fsc->sb;
2796         struct inode *inode;
2797         struct dentry *parent, *dentry;
2798         struct ceph_dentry_info *di;
2799         int mds = session->s_mds;
2800         struct ceph_mds_lease *h = msg->front.iov_base;
2801         u32 seq;
2802         struct ceph_vino vino;
2803         struct qstr dname;
2804         int release = 0;
2805
2806         dout("handle_lease from mds%d\n", mds);
2807
2808         /* decode */
2809         if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
2810                 goto bad;
2811         vino.ino = le64_to_cpu(h->ino);
2812         vino.snap = CEPH_NOSNAP;
2813         seq = le32_to_cpu(h->seq);
2814         dname.name = (void *)h + sizeof(*h) + sizeof(u32);
2815         dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
2816         if (dname.len != get_unaligned_le32(h+1))
2817                 goto bad;
2818
2819         mutex_lock(&session->s_mutex);
2820         session->s_seq++;
2821
2822         /* lookup inode */
2823         inode = ceph_find_inode(sb, vino);
2824         dout("handle_lease %s, ino %llx %p %.*s\n",
2825              ceph_lease_op_name(h->action), vino.ino, inode,
2826              dname.len, dname.name);
2827         if (inode == NULL) {
2828                 dout("handle_lease no inode %llx\n", vino.ino);
2829                 goto release;
2830         }
2831
2832         /* dentry */
2833         parent = d_find_alias(inode);
2834         if (!parent) {
2835                 dout("no parent dentry on inode %p\n", inode);
2836                 WARN_ON(1);
2837                 goto release;  /* hrm... */
2838         }
2839         dname.hash = full_name_hash(dname.name, dname.len);
2840         dentry = d_lookup(parent, &dname);
2841         dput(parent);
2842         if (!dentry)
2843                 goto release;
2844
2845         spin_lock(&dentry->d_lock);
2846         di = ceph_dentry(dentry);
2847         switch (h->action) {
2848         case CEPH_MDS_LEASE_REVOKE:
2849                 if (di->lease_session == session) {
2850                         if (ceph_seq_cmp(di->lease_seq, seq) > 0)
2851                                 h->seq = cpu_to_le32(di->lease_seq);
2852                         __ceph_mdsc_drop_dentry_lease(dentry);
2853                 }
2854                 release = 1;
2855                 break;
2856
2857         case CEPH_MDS_LEASE_RENEW:
2858                 if (di->lease_session == session &&
2859                     di->lease_gen == session->s_cap_gen &&
2860                     di->lease_renew_from &&
2861                     di->lease_renew_after == 0) {
2862                         unsigned long duration =
2863                                 le32_to_cpu(h->duration_ms) * HZ / 1000;
2864
2865                         di->lease_seq = seq;
2866                         dentry->d_time = di->lease_renew_from + duration;
2867                         di->lease_renew_after = di->lease_renew_from +
2868                                 (duration >> 1);
2869                         di->lease_renew_from = 0;
2870                 }
2871                 break;
2872         }
2873         spin_unlock(&dentry->d_lock);
2874         dput(dentry);
2875
2876         if (!release)
2877                 goto out;
2878
2879 release:
2880         /* let's just reuse the same message */
2881         h->action = CEPH_MDS_LEASE_REVOKE_ACK;
2882         ceph_msg_get(msg);
2883         ceph_con_send(&session->s_con, msg);
2884
2885 out:
2886         iput(inode);
2887         mutex_unlock(&session->s_mutex);
2888         return;
2889
2890 bad:
2891         pr_err("corrupt lease message\n");
2892         ceph_msg_dump(msg);
2893 }
2894
2895 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
2896                               struct inode *inode,
2897                               struct dentry *dentry, char action,
2898                               u32 seq)
2899 {
2900         struct ceph_msg *msg;
2901         struct ceph_mds_lease *lease;
2902         int len = sizeof(*lease) + sizeof(u32);
2903         int dnamelen = 0;
2904
2905         dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
2906              inode, dentry, ceph_lease_op_name(action), session->s_mds);
2907         dnamelen = dentry->d_name.len;
2908         len += dnamelen;
2909
2910         msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
2911         if (!msg)
2912                 return;
2913         lease = msg->front.iov_base;
2914         lease->action = action;
2915         lease->ino = cpu_to_le64(ceph_vino(inode).ino);
2916         lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
2917         lease->seq = cpu_to_le32(seq);
2918         put_unaligned_le32(dnamelen, lease + 1);
2919         memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
2920
2921         /*
2922          * if this is a preemptive lease RELEASE, no need to
2923          * flush request stream, since the actual request will
2924          * soon follow.
2925          */
2926         msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
2927
2928         ceph_con_send(&session->s_con, msg);
2929 }
2930
2931 /*
2932  * Preemptively release a lease we expect to invalidate anyway.
2933  * Pass @inode always, @dentry is optional.
2934  */
2935 void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
2936                              struct dentry *dentry)
2937 {
2938         struct ceph_dentry_info *di;
2939         struct ceph_mds_session *session;
2940         u32 seq;
2941
2942         BUG_ON(inode == NULL);
2943         BUG_ON(dentry == NULL);
2944
2945         /* is dentry lease valid? */
2946         spin_lock(&dentry->d_lock);
2947         di = ceph_dentry(dentry);
2948         if (!di || !di->lease_session ||
2949             di->lease_session->s_mds < 0 ||
2950             di->lease_gen != di->lease_session->s_cap_gen ||
2951             !time_before(jiffies, dentry->d_time)) {
2952                 dout("lease_release inode %p dentry %p -- "
2953                      "no lease\n",
2954                      inode, dentry);
2955                 spin_unlock(&dentry->d_lock);
2956                 return;
2957         }
2958
2959         /* we do have a lease on this dentry; note mds and seq */
2960         session = ceph_get_mds_session(di->lease_session);
2961         seq = di->lease_seq;
2962         __ceph_mdsc_drop_dentry_lease(dentry);
2963         spin_unlock(&dentry->d_lock);
2964
2965         dout("lease_release inode %p dentry %p to mds%d\n",
2966              inode, dentry, session->s_mds);
2967         ceph_mdsc_lease_send_msg(session, inode, dentry,
2968                                  CEPH_MDS_LEASE_RELEASE, seq);
2969         ceph_put_mds_session(session);
2970 }
2971
2972 /*
2973  * drop all leases (and dentry refs) in preparation for umount
2974  */
2975 static void drop_leases(struct ceph_mds_client *mdsc)
2976 {
2977         int i;
2978
2979         dout("drop_leases\n");
2980         mutex_lock(&mdsc->mutex);
2981         for (i = 0; i < mdsc->max_sessions; i++) {
2982                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2983                 if (!s)
2984                         continue;
2985                 mutex_unlock(&mdsc->mutex);
2986                 mutex_lock(&s->s_mutex);
2987                 mutex_unlock(&s->s_mutex);
2988                 ceph_put_mds_session(s);
2989                 mutex_lock(&mdsc->mutex);
2990         }
2991         mutex_unlock(&mdsc->mutex);
2992 }
2993
2994
2995
2996 /*
2997  * delayed work -- periodically trim expired leases, renew caps with mds
2998  */
2999 static void schedule_delayed(struct ceph_mds_client *mdsc)
3000 {
3001         int delay = 5;
3002         unsigned hz = round_jiffies_relative(HZ * delay);
3003         schedule_delayed_work(&mdsc->delayed_work, hz);
3004 }
3005
3006 static void delayed_work(struct work_struct *work)
3007 {
3008         int i;
3009         struct ceph_mds_client *mdsc =
3010                 container_of(work, struct ceph_mds_client, delayed_work.work);
3011         int renew_interval;
3012         int renew_caps;
3013
3014         dout("mdsc delayed_work\n");
3015         ceph_check_delayed_caps(mdsc);
3016
3017         mutex_lock(&mdsc->mutex);
3018         renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3019         renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3020                                    mdsc->last_renew_caps);
3021         if (renew_caps)
3022                 mdsc->last_renew_caps = jiffies;
3023
3024         for (i = 0; i < mdsc->max_sessions; i++) {
3025                 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3026                 if (s == NULL)
3027                         continue;
3028                 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3029                         dout("resending session close request for mds%d\n",
3030                              s->s_mds);
3031                         request_close_session(mdsc, s);
3032                         ceph_put_mds_session(s);
3033                         continue;
3034                 }
3035                 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3036                         if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3037                                 s->s_state = CEPH_MDS_SESSION_HUNG;
3038                                 pr_info("mds%d hung\n", s->s_mds);
3039                         }
3040                 }
3041                 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3042                         /* this mds is failed or recovering, just wait */
3043                         ceph_put_mds_session(s);
3044                         continue;
3045                 }
3046                 mutex_unlock(&mdsc->mutex);
3047
3048                 mutex_lock(&s->s_mutex);
3049                 if (renew_caps)
3050                         send_renew_caps(mdsc, s);
3051                 else
3052                         ceph_con_keepalive(&s->s_con);
3053                 ceph_add_cap_releases(mdsc, s);
3054                 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3055                     s->s_state == CEPH_MDS_SESSION_HUNG)
3056                         ceph_send_cap_releases(mdsc, s);
3057                 mutex_unlock(&s->s_mutex);
3058                 ceph_put_mds_session(s);
3059
3060                 mutex_lock(&mdsc->mutex);
3061         }
3062         mutex_unlock(&mdsc->mutex);
3063
3064         schedule_delayed(mdsc);
3065 }
3066
3067 int ceph_mdsc_init(struct ceph_fs_client *fsc)
3068
3069 {
3070         struct ceph_mds_client *mdsc;
3071
3072         mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3073         if (!mdsc)
3074                 return -ENOMEM;
3075         mdsc->fsc = fsc;
3076         fsc->mdsc = mdsc;
3077         mutex_init(&mdsc->mutex);
3078         mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
3079         if (mdsc->mdsmap == NULL) {
3080                 kfree(mdsc);
3081                 return -ENOMEM;
3082         }
3083
3084         init_completion(&mdsc->safe_umount_waiters);
3085         init_waitqueue_head(&mdsc->session_close_wq);
3086         INIT_LIST_HEAD(&mdsc->waiting_for_map);
3087         mdsc->sessions = NULL;
3088         mdsc->max_sessions = 0;
3089         mdsc->stopping = 0;
3090         init_rwsem(&mdsc->snap_rwsem);
3091         mdsc->snap_realms = RB_ROOT;
3092         INIT_LIST_HEAD(&mdsc->snap_empty);
3093         spin_lock_init(&mdsc->snap_empty_lock);
3094         mdsc->last_tid = 0;
3095         mdsc->request_tree = RB_ROOT;
3096         INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3097         mdsc->last_renew_caps = jiffies;
3098         INIT_LIST_HEAD(&mdsc->cap_delay_list);
3099         spin_lock_init(&mdsc->cap_delay_lock);
3100         INIT_LIST_HEAD(&mdsc->snap_flush_list);
3101         spin_lock_init(&mdsc->snap_flush_lock);
3102         mdsc->cap_flush_seq = 0;
3103         INIT_LIST_HEAD(&mdsc->cap_dirty);
3104         INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
3105         mdsc->num_cap_flushing = 0;
3106         spin_lock_init(&mdsc->cap_dirty_lock);
3107         init_waitqueue_head(&mdsc->cap_flushing_wq);
3108         spin_lock_init(&mdsc->dentry_lru_lock);
3109         INIT_LIST_HEAD(&mdsc->dentry_lru);
3110
3111         ceph_caps_init(mdsc);
3112         ceph_adjust_min_caps(mdsc, fsc->min_caps);
3113
3114         return 0;
3115 }
3116
3117 /*
3118  * Wait for safe replies on open mds requests.  If we time out, drop
3119  * all requests from the tree to avoid dangling dentry refs.
3120  */
3121 static void wait_requests(struct ceph_mds_client *mdsc)
3122 {
3123         struct ceph_mds_request *req;
3124         struct ceph_fs_client *fsc = mdsc->fsc;
3125
3126         mutex_lock(&mdsc->mutex);
3127         if (__get_oldest_req(mdsc)) {
3128                 mutex_unlock(&mdsc->mutex);
3129
3130                 dout("wait_requests waiting for requests\n");
3131                 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
3132                                     fsc->client->options->mount_timeout * HZ);
3133
3134                 /* tear down remaining requests */
3135                 mutex_lock(&mdsc->mutex);
3136                 while ((req = __get_oldest_req(mdsc))) {
3137                         dout("wait_requests timed out on tid %llu\n",
3138                              req->r_tid);
3139                         __unregister_request(mdsc, req);
3140                 }
3141         }
3142         mutex_unlock(&mdsc->mutex);
3143         dout("wait_requests done\n");
3144 }
3145
3146 /*
3147  * called before mount is ro, and before dentries are torn down.
3148  * (hmm, does this still race with new lookups?)
3149  */
3150 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3151 {
3152         dout("pre_umount\n");
3153         mdsc->stopping = 1;
3154
3155         drop_leases(mdsc);
3156         ceph_flush_dirty_caps(mdsc);
3157         wait_requests(mdsc);
3158
3159         /*
3160          * wait for reply handlers to drop their request refs and
3161          * their inode/dcache refs
3162          */
3163         ceph_msgr_flush();
3164 }
3165
3166 /*
3167  * wait for all write mds requests to flush.
3168  */
3169 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3170 {
3171         struct ceph_mds_request *req = NULL, *nextreq;
3172         struct rb_node *n;
3173
3174         mutex_lock(&mdsc->mutex);
3175         dout("wait_unsafe_requests want %lld\n", want_tid);
3176 restart:
3177         req = __get_oldest_req(mdsc);
3178         while (req && req->r_tid <= want_tid) {
3179                 /* find next request */
3180                 n = rb_next(&req->r_node);
3181                 if (n)
3182                         nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3183                 else
3184                         nextreq = NULL;
3185                 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
3186                         /* write op */
3187                         ceph_mdsc_get_request(req);
3188                         if (nextreq)
3189                                 ceph_mdsc_get_request(nextreq);
3190                         mutex_unlock(&mdsc->mutex);
3191                         dout("wait_unsafe_requests  wait on %llu (want %llu)\n",
3192                              req->r_tid, want_tid);
3193                         wait_for_completion(&req->r_safe_completion);
3194                         mutex_lock(&mdsc->mutex);
3195                         ceph_mdsc_put_request(req);
3196                         if (!nextreq)
3197                                 break;  /* next dne before, so we're done! */
3198                         if (RB_EMPTY_NODE(&nextreq->r_node)) {
3199                                 /* next request was removed from tree */
3200                                 ceph_mdsc_put_request(nextreq);
3201                                 goto restart;
3202                         }
3203                         ceph_mdsc_put_request(nextreq);  /* won't go away */
3204                 }
3205                 req = nextreq;
3206         }
3207         mutex_unlock(&mdsc->mutex);
3208         dout("wait_unsafe_requests done\n");
3209 }
3210
3211 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3212 {
3213         u64 want_tid, want_flush;
3214
3215         if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
3216                 return;
3217
3218         dout("sync\n");
3219         mutex_lock(&mdsc->mutex);
3220         want_tid = mdsc->last_tid;
3221         want_flush = mdsc->cap_flush_seq;
3222         mutex_unlock(&mdsc->mutex);
3223         dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
3224
3225         ceph_flush_dirty_caps(mdsc);
3226
3227         wait_unsafe_requests(mdsc, want_tid);
3228         wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush));
3229 }
3230
3231 /*
3232  * true if all sessions are closed, or we force unmount
3233  */
3234 static bool done_closing_sessions(struct ceph_mds_client *mdsc)
3235 {
3236         int i, n = 0;
3237
3238         if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
3239                 return true;
3240
3241         mutex_lock(&mdsc->mutex);
3242         for (i = 0; i < mdsc->max_sessions; i++)
3243                 if (mdsc->sessions[i])
3244                         n++;
3245         mutex_unlock(&mdsc->mutex);
3246         return n == 0;
3247 }
3248
3249 /*
3250  * called after sb is ro.
3251  */
3252 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3253 {
3254         struct ceph_mds_session *session;
3255         int i;
3256         struct ceph_fs_client *fsc = mdsc->fsc;
3257         unsigned long timeout = fsc->client->options->mount_timeout * HZ;
3258
3259         dout("close_sessions\n");
3260
3261         /* close sessions */
3262         mutex_lock(&mdsc->mutex);
3263         for (i = 0; i < mdsc->max_sessions; i++) {
3264                 session = __ceph_lookup_mds_session(mdsc, i);
3265                 if (!session)
3266                         continue;
3267                 mutex_unlock(&mdsc->mutex);
3268                 mutex_lock(&session->s_mutex);
3269                 __close_session(mdsc, session);
3270                 mutex_unlock(&session->s_mutex);
3271                 ceph_put_mds_session(session);
3272                 mutex_lock(&mdsc->mutex);
3273         }
3274         mutex_unlock(&mdsc->mutex);
3275
3276         dout("waiting for sessions to close\n");
3277         wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3278                            timeout);
3279
3280         /* tear down remaining sessions */
3281         mutex_lock(&mdsc->mutex);
3282         for (i = 0; i < mdsc->max_sessions; i++) {
3283                 if (mdsc->sessions[i]) {
3284                         session = get_session(mdsc->sessions[i]);
3285                         __unregister_session(mdsc, session);
3286                         mutex_unlock(&mdsc->mutex);
3287                         mutex_lock(&session->s_mutex);
3288                         remove_session_caps(session);
3289                         mutex_unlock(&session->s_mutex);
3290                         ceph_put_mds_session(session);
3291                         mutex_lock(&mdsc->mutex);
3292                 }
3293         }
3294         WARN_ON(!list_empty(&mdsc->cap_delay_list));
3295         mutex_unlock(&mdsc->mutex);
3296
3297         ceph_cleanup_empty_realms(mdsc);
3298
3299         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3300
3301         dout("stopped\n");
3302 }
3303
3304 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
3305 {
3306         dout("stop\n");
3307         cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3308         if (mdsc->mdsmap)
3309                 ceph_mdsmap_destroy(mdsc->mdsmap);
3310         kfree(mdsc->sessions);
3311         ceph_caps_finalize(mdsc);
3312 }
3313
3314 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3315 {
3316         struct ceph_mds_client *mdsc = fsc->mdsc;
3317
3318         dout("mdsc_destroy %p\n", mdsc);
3319         ceph_mdsc_stop(mdsc);
3320
3321         /* flush out any connection work with references to us */
3322         ceph_msgr_flush();
3323
3324         fsc->mdsc = NULL;
3325         kfree(mdsc);
3326         dout("mdsc_destroy %p done\n", mdsc);
3327 }
3328
3329
3330 /*
3331  * handle mds map update.
3332  */
3333 void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3334 {
3335         u32 epoch;
3336         u32 maplen;
3337         void *p = msg->front.iov_base;
3338         void *end = p + msg->front.iov_len;
3339         struct ceph_mdsmap *newmap, *oldmap;
3340         struct ceph_fsid fsid;
3341         int err = -EINVAL;
3342
3343         ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3344         ceph_decode_copy(&p, &fsid, sizeof(fsid));
3345         if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
3346                 return;
3347         epoch = ceph_decode_32(&p);
3348         maplen = ceph_decode_32(&p);
3349         dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3350
3351         /* do we need it? */
3352         ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
3353         mutex_lock(&mdsc->mutex);
3354         if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3355                 dout("handle_map epoch %u <= our %u\n",
3356                      epoch, mdsc->mdsmap->m_epoch);
3357                 mutex_unlock(&mdsc->mutex);
3358                 return;
3359         }
3360
3361         newmap = ceph_mdsmap_decode(&p, end);
3362         if (IS_ERR(newmap)) {
3363                 err = PTR_ERR(newmap);
3364                 goto bad_unlock;
3365         }
3366
3367         /* swap into place */
3368         if (mdsc->mdsmap) {
3369                 oldmap = mdsc->mdsmap;
3370                 mdsc->mdsmap = newmap;
3371                 check_new_map(mdsc, newmap, oldmap);
3372                 ceph_mdsmap_destroy(oldmap);
3373         } else {
3374                 mdsc->mdsmap = newmap;  /* first mds map */
3375         }
3376         mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
3377
3378         __wake_requests(mdsc, &mdsc->waiting_for_map);
3379
3380         mutex_unlock(&mdsc->mutex);
3381         schedule_delayed(mdsc);
3382         return;
3383
3384 bad_unlock:
3385         mutex_unlock(&mdsc->mutex);
3386 bad:
3387         pr_err("error decoding mdsmap %d\n", err);
3388         return;
3389 }
3390
3391 static struct ceph_connection *con_get(struct ceph_connection *con)
3392 {
3393         struct ceph_mds_session *s = con->private;
3394
3395         if (get_session(s)) {
3396                 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
3397                 return con;
3398         }
3399         dout("mdsc con_get %p FAIL\n", s);
3400         return NULL;
3401 }
3402
3403 static void con_put(struct ceph_connection *con)
3404 {
3405         struct ceph_mds_session *s = con->private;
3406
3407         dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
3408         ceph_put_mds_session(s);
3409 }
3410
3411 /*
3412  * if the client is unresponsive for long enough, the mds will kill
3413  * the session entirely.
3414  */
3415 static void peer_reset(struct ceph_connection *con)
3416 {
3417         struct ceph_mds_session *s = con->private;
3418         struct ceph_mds_client *mdsc = s->s_mdsc;
3419
3420         pr_warning("mds%d closed our session\n", s->s_mds);
3421         send_mds_reconnect(mdsc, s);
3422 }
3423
3424 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3425 {
3426         struct ceph_mds_session *s = con->private;
3427         struct ceph_mds_client *mdsc = s->s_mdsc;
3428         int type = le16_to_cpu(msg->hdr.type);
3429
3430         mutex_lock(&mdsc->mutex);
3431         if (__verify_registered_session(mdsc, s) < 0) {
3432                 mutex_unlock(&mdsc->mutex);
3433                 goto out;
3434         }
3435         mutex_unlock(&mdsc->mutex);
3436
3437         switch (type) {
3438         case CEPH_MSG_MDS_MAP:
3439                 ceph_mdsc_handle_map(mdsc, msg);
3440                 break;
3441         case CEPH_MSG_CLIENT_SESSION:
3442                 handle_session(s, msg);
3443                 break;
3444         case CEPH_MSG_CLIENT_REPLY:
3445                 handle_reply(s, msg);
3446                 break;
3447         case CEPH_MSG_CLIENT_REQUEST_FORWARD:
3448                 handle_forward(mdsc, s, msg);
3449                 break;
3450         case CEPH_MSG_CLIENT_CAPS:
3451                 ceph_handle_caps(s, msg);
3452                 break;
3453         case CEPH_MSG_CLIENT_SNAP:
3454                 ceph_handle_snap(mdsc, s, msg);
3455                 break;
3456         case CEPH_MSG_CLIENT_LEASE:
3457                 handle_lease(mdsc, s, msg);
3458                 break;
3459
3460         default:
3461                 pr_err("received unknown message type %d %s\n", type,
3462                        ceph_msg_type_name(type));
3463         }
3464 out:
3465         ceph_msg_put(msg);
3466 }
3467
3468 /*
3469  * authentication
3470  */
3471
3472 /*
3473  * Note: returned pointer is the address of a structure that's
3474  * managed separately.  Caller must *not* attempt to free it.
3475  */
3476 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
3477                                         int *proto, int force_new)
3478 {
3479         struct ceph_mds_session *s = con->private;
3480         struct ceph_mds_client *mdsc = s->s_mdsc;
3481         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3482         struct ceph_auth_handshake *auth = &s->s_auth;
3483
3484         if (force_new && auth->authorizer) {
3485                 ceph_auth_destroy_authorizer(ac, auth->authorizer);
3486                 auth->authorizer = NULL;
3487         }
3488         if (!auth->authorizer) {
3489                 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3490                                                       auth);
3491                 if (ret)
3492                         return ERR_PTR(ret);
3493         } else {
3494                 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3495                                                       auth);
3496                 if (ret)
3497                         return ERR_PTR(ret);
3498         }
3499         *proto = ac->protocol;
3500
3501         return auth;
3502 }
3503
3504
3505 static int verify_authorizer_reply(struct ceph_connection *con, int len)
3506 {
3507         struct ceph_mds_session *s = con->private;
3508         struct ceph_mds_client *mdsc = s->s_mdsc;
3509         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3510
3511         return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
3512 }
3513
3514 static int invalidate_authorizer(struct ceph_connection *con)
3515 {
3516         struct ceph_mds_session *s = con->private;
3517         struct ceph_mds_client *mdsc = s->s_mdsc;
3518         struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3519
3520         ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
3521
3522         return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
3523 }
3524
3525 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3526                                 struct ceph_msg_header *hdr, int *skip)
3527 {
3528         struct ceph_msg *msg;
3529         int type = (int) le16_to_cpu(hdr->type);
3530         int front_len = (int) le32_to_cpu(hdr->front_len);
3531
3532         if (con->in_msg)
3533                 return con->in_msg;
3534
3535         *skip = 0;
3536         msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3537         if (!msg) {
3538                 pr_err("unable to allocate msg type %d len %d\n",
3539                        type, front_len);
3540                 return NULL;
3541         }
3542
3543         return msg;
3544 }
3545
3546 static const struct ceph_connection_operations mds_con_ops = {
3547         .get = con_get,
3548         .put = con_put,
3549         .dispatch = dispatch,
3550         .get_authorizer = get_authorizer,
3551         .verify_authorizer_reply = verify_authorizer_reply,
3552         .invalidate_authorizer = invalidate_authorizer,
3553         .peer_reset = peer_reset,
3554         .alloc_msg = mds_alloc_msg,
3555 };
3556
3557 /* eof */