net/ceph/mon_client.c

   1 #include <linux/ceph/ceph_debug.h>
   2
   3 #include <linux/module.h>
   4 #include <linux/types.h>
   5 #include <linux/slab.h>
   6 #include <linux/random.h>
   7 #include <linux/sched.h>
   8
   9 #include <linux/ceph/mon_client.h>
  10 #include <linux/ceph/libceph.h>
  11 #include <linux/ceph/decode.h>
  12
  13 #include <linux/ceph/auth.h>
  14
  15 /*
  16  * Interact with Ceph monitor cluster.  Handle requests for new map
  17  * versions, and periodically resend as needed.  Also implement
  18  * statfs() and umount().
  19  *
  20  * A small cluster of Ceph "monitors" are responsible for managing critical
  21  * cluster configuration and state information.  An odd number (e.g., 3, 5)
  22  * of cmon daemons use a modified version of the Paxos part-time parliament
  23  * algorithm to manage the MDS map (mds cluster membership), OSD map, and
  24  * list of clients who have mounted the file system.
  25  *
  26  * We maintain an open, active session with a monitor at all times in order to
  27  * receive timely MDSMap updates.  We periodically send a keepalive byte on the
  28  * TCP socket to ensure we detect a failure.  If the connection does break, we
  29  * randomly hunt for a new monitor.  Once the connection is reestablished, we
  30  * resend any outstanding requests.
  31  */
  32
  33 static const struct ceph_connection_operations mon_con_ops;
  34
  35 static int __validate_auth(struct ceph_mon_client *monc);
  36
  37 /*
  38  * Decode a monmap blob (e.g., during mount).
  39  */
  40 struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
  41 {
  42         struct ceph_monmap *m = NULL;
  43         int i, err = -EINVAL;
  44         struct ceph_fsid fsid;
  45         u32 epoch, num_mon;
  46         u16 version;
  47         u32 len;
  48
  49         ceph_decode_32_safe(&p, end, len, bad);
  50         ceph_decode_need(&p, end, len, bad);
  51
  52         dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
  53
  54         ceph_decode_16_safe(&p, end, version, bad);
  55
  56         ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad);
  57         ceph_decode_copy(&p, &fsid, sizeof(fsid));
  58         epoch = ceph_decode_32(&p);
  59
  60         num_mon = ceph_decode_32(&p);
  61         ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
  62
  63         if (num_mon >= CEPH_MAX_MON)
  64                 goto bad;
  65         m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
  66         if (m == NULL)
  67                 return ERR_PTR(-ENOMEM);
  68         m->fsid = fsid;
  69         m->epoch = epoch;
  70         m->num_mon = num_mon;
  71         ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0]));
  72         for (i = 0; i < num_mon; i++)
  73                 ceph_decode_addr(&m->mon_inst[i].addr);
  74
  75         dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
  76              m->num_mon);
  77         for (i = 0; i < m->num_mon; i++)
  78                 dout("monmap_decode  mon%d is %s\n", i,
  79                      ceph_pr_addr(&m->mon_inst[i].addr.in_addr));
  80         return m;
  81
  82 bad:
  83         dout("monmap_decode failed with %d\n", err);
  84         kfree(m);
  85         return ERR_PTR(err);
  86 }
  87
  88 /*
  89  * return true if *addr is included in the monmap.
  90  */
  91 int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
  92 {
  93         int i;
  94
  95         for (i = 0; i < m->num_mon; i++)
  96                 if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
  97                         return 1;
  98         return 0;
  99 }
 100
 101 /*
 102  * Send an auth request.
 103  */
 104 static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
 105 {
 106         monc->pending_auth = 1;
 107         monc->m_auth->front.iov_len = len;
 108         monc->m_auth->hdr.front_len = cpu_to_le32(len);
 109         ceph_con_revoke(monc->con, monc->m_auth);
 110         ceph_msg_get(monc->m_auth);  /* keep our ref */
 111         ceph_con_send(monc->con, monc->m_auth);
 112 }
 113
 114 /*
 115  * Close monitor session, if any.
 116  */
 117 static void __close_session(struct ceph_mon_client *monc)
 118 {
 119         dout("__close_session closing mon%d\n", monc->cur_mon);
 120         ceph_con_revoke(monc->con, monc->m_auth);
 121         ceph_con_close(monc->con);
 122         monc->cur_mon = -1;
 123         monc->pending_auth = 0;
 124         ceph_auth_reset(monc->auth);
 125 }
 126
 127 /*
 128  * Open a session with a (new) monitor.
 129  */
 130 static int __open_session(struct ceph_mon_client *monc)
 131 {
 132         char r;
 133         int ret;
 134
 135         if (monc->cur_mon < 0) {
 136                 get_random_bytes(&r, 1);
 137                 monc->cur_mon = r % monc->monmap->num_mon;
 138                 dout("open_session num=%d r=%d -> mon%d\n",
 139                      monc->monmap->num_mon, r, monc->cur_mon);
 140                 monc->sub_sent = 0;
 141                 monc->sub_renew_after = jiffies;  /* i.e., expired */
 142                 monc->want_next_osdmap = !!monc->want_next_osdmap;
 143
 144                 dout("open_session mon%d opening\n", monc->cur_mon);
 145                 monc->con->peer_name.type = CEPH_ENTITY_TYPE_MON;
 146                 monc->con->peer_name.num = cpu_to_le64(monc->cur_mon);
 147                 ceph_con_open(monc->con,
 148                               &monc->monmap->mon_inst[monc->cur_mon].addr);
 149
 150                 /* initiatiate authentication handshake */
 151                 ret = ceph_auth_build_hello(monc->auth,
 152                                             monc->m_auth->front.iov_base,
 153                                             monc->m_auth->front_max);
 154                 __send_prepared_auth_request(monc, ret);
 155         } else {
 156                 dout("open_session mon%d already open\n", monc->cur_mon);
 157         }
 158         return 0;
 159 }
 160
 161 static bool __sub_expired(struct ceph_mon_client *monc)
 162 {
 163         return time_after_eq(jiffies, monc->sub_renew_after);
 164 }
 165
 166 /*
 167  * Reschedule delayed work timer.
 168  */
 169 static void __schedule_delayed(struct ceph_mon_client *monc)
 170 {
 171         unsigned delay;
 172
 173         if (monc->cur_mon < 0 || __sub_expired(monc))
 174                 delay = 10 * HZ;
 175         else
 176                 delay = 20 * HZ;
 177         dout("__schedule_delayed after %u\n", delay);
 178         schedule_delayed_work(&monc->delayed_work, delay);
 179 }
 180
 181 /*
 182  * Send subscribe request for mdsmap and/or osdmap.
 183  */
 184 static void __send_subscribe(struct ceph_mon_client *monc)
 185 {
 186         dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
 187              (unsigned)monc->sub_sent, __sub_expired(monc),
 188              monc->want_next_osdmap);
 189         if ((__sub_expired(monc) && !monc->sub_sent) ||
 190             monc->want_next_osdmap == 1) {
 191                 struct ceph_msg *msg = monc->m_subscribe;
 192                 struct ceph_mon_subscribe_item *i;
 193                 void *p, *end;
 194                 int num;
 195
 196                 p = msg->front.iov_base;
 197                 end = p + msg->front_max;
 198
 199                 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
 200                 ceph_encode_32(&p, num);
 201
 202                 if (monc->want_next_osdmap) {
 203                         dout("__send_subscribe to 'osdmap' %u\n",
 204                              (unsigned)monc->have_osdmap);
 205                         ceph_encode_string(&p, end, "osdmap", 6);
 206                         i = p;
 207                         i->have = cpu_to_le64(monc->have_osdmap);
 208                         i->onetime = 1;
 209                         p += sizeof(*i);
 210                         monc->want_next_osdmap = 2;  /* requested */
 211                 }
 212                 if (monc->want_mdsmap) {
 213                         dout("__send_subscribe to 'mdsmap' %u+\n",
 214                              (unsigned)monc->have_mdsmap);
 215                         ceph_encode_string(&p, end, "mdsmap", 6);
 216                         i = p;
 217                         i->have = cpu_to_le64(monc->have_mdsmap);
 218                         i->onetime = 0;
 219                         p += sizeof(*i);
 220                 }
 221                 ceph_encode_string(&p, end, "monmap", 6);
 222                 i = p;
 223                 i->have = 0;
 224                 i->onetime = 0;
 225                 p += sizeof(*i);
 226
 227                 msg->front.iov_len = p - msg->front.iov_base;
 228                 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
 229                 ceph_con_revoke(monc->con, msg);
 230                 ceph_con_send(monc->con, ceph_msg_get(msg));
 231
 232                 monc->sub_sent = jiffies | 1;  /* never 0 */
 233         }
 234 }
 235
 236 static void handle_subscribe_ack(struct ceph_mon_client *monc,
 237                                  struct ceph_msg *msg)
 238 {
 239         unsigned seconds;
 240         struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
 241
 242         if (msg->front.iov_len < sizeof(*h))
 243                 goto bad;
 244         seconds = le32_to_cpu(h->duration);
 245
 246         mutex_lock(&monc->mutex);
 247         if (monc->hunting) {
 248                 pr_info("mon%d %s session established\n",
 249                         monc->cur_mon,
 250                         ceph_pr_addr(&monc->con->peer_addr.in_addr));
 251                 monc->hunting = false;
 252         }
 253         dout("handle_subscribe_ack after %d seconds\n", seconds);
 254         monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
 255         monc->sub_sent = 0;
 256         mutex_unlock(&monc->mutex);
 257         return;
 258 bad:
 259         pr_err("got corrupt subscribe-ack msg\n");
 260         ceph_msg_dump(msg);
 261 }
 262
 263 /*
 264  * Keep track of which maps we have
 265  */
 266 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
 267 {
 268         mutex_lock(&monc->mutex);
 269         monc->have_mdsmap = got;
 270         mutex_unlock(&monc->mutex);
 271         return 0;
 272 }
 273 EXPORT_SYMBOL(ceph_monc_got_mdsmap);
 274
 275 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
 276 {
 277         mutex_lock(&monc->mutex);
 278         monc->have_osdmap = got;
 279         monc->want_next_osdmap = 0;
 280         mutex_unlock(&monc->mutex);
 281         return 0;
 282 }
 283
 284 /*
 285  * Register interest in the next osdmap
 286  */
 287 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
 288 {
 289         dout("request_next_osdmap have %u\n", monc->have_osdmap);
 290         mutex_lock(&monc->mutex);
 291         if (!monc->want_next_osdmap)
 292                 monc->want_next_osdmap = 1;
 293         if (monc->want_next_osdmap < 2)
 294                 __send_subscribe(monc);
 295         mutex_unlock(&monc->mutex);
 296 }
 297
 298 /*
 299  *
 300  */
 301 int ceph_monc_open_session(struct ceph_mon_client *monc)
 302 {
 303         mutex_lock(&monc->mutex);
 304         __open_session(monc);
 305         __schedule_delayed(monc);
 306         mutex_unlock(&monc->mutex);
 307         return 0;
 308 }
 309 EXPORT_SYMBOL(ceph_monc_open_session);
 310
 311 /*
 312  * The monitor responds with mount ack indicate mount success.  The
 313  * included client ticket allows the client to talk to MDSs and OSDs.
 314  */
 315 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
 316                                  struct ceph_msg *msg)
 317 {
 318         struct ceph_client *client = monc->client;
 319         struct ceph_monmap *monmap = NULL, *old = monc->monmap;
 320         void *p, *end;
 321
 322         mutex_lock(&monc->mutex);
 323
 324         dout("handle_monmap\n");
 325         p = msg->front.iov_base;
 326         end = p + msg->front.iov_len;
 327
 328         monmap = ceph_monmap_decode(p, end);
 329         if (IS_ERR(monmap)) {
 330                 pr_err("problem decoding monmap, %d\n",
 331                        (int)PTR_ERR(monmap));
 332                 goto out;
 333         }
 334
 335         if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
 336                 kfree(monmap);
 337                 goto out;
 338         }
 339
 340         client->monc.monmap = monmap;
 341         kfree(old);
 342
 343 out:
 344         mutex_unlock(&monc->mutex);
 345         wake_up_all(&client->auth_wq);
 346 }
 347
 348 /*
 349  * generic requests (e.g., statfs, poolop)
 350  */
 351 static struct ceph_mon_generic_request *__lookup_generic_req(
 352         struct ceph_mon_client *monc, u64 tid)
 353 {
 354         struct ceph_mon_generic_request *req;
 355         struct rb_node *n = monc->generic_request_tree.rb_node;
 356
 357         while (n) {
 358                 req = rb_entry(n, struct ceph_mon_generic_request, node);
 359                 if (tid < req->tid)
 360                         n = n->rb_left;
 361                 else if (tid > req->tid)
 362                         n = n->rb_right;
 363                 else
 364                         return req;
 365         }
 366         return NULL;
 367 }
 368
 369 static void __insert_generic_request(struct ceph_mon_client *monc,
 370                             struct ceph_mon_generic_request *new)
 371 {
 372         struct rb_node **p = &monc->generic_request_tree.rb_node;
 373         struct rb_node *parent = NULL;
 374         struct ceph_mon_generic_request *req = NULL;
 375
 376         while (*p) {
 377                 parent = *p;
 378                 req = rb_entry(parent, struct ceph_mon_generic_request, node);
 379                 if (new->tid < req->tid)
 380                         p = &(*p)->rb_left;
 381                 else if (new->tid > req->tid)
 382                         p = &(*p)->rb_right;
 383                 else
 384                         BUG();
 385         }
 386
 387         rb_link_node(&new->node, parent, p);
 388         rb_insert_color(&new->node, &monc->generic_request_tree);
 389 }
 390
 391 static void release_generic_request(struct kref *kref)
 392 {
 393         struct ceph_mon_generic_request *req =
 394                 container_of(kref, struct ceph_mon_generic_request, kref);
 395
 396         if (req->reply)
 397                 ceph_msg_put(req->reply);
 398         if (req->request)
 399                 ceph_msg_put(req->request);
 400
 401         kfree(req);
 402 }
 403
 404 static void put_generic_request(struct ceph_mon_generic_request *req)
 405 {
 406         kref_put(&req->kref, release_generic_request);
 407 }
 408
 409 static void get_generic_request(struct ceph_mon_generic_request *req)
 410 {
 411         kref_get(&req->kref);
 412 }
 413
 414 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
 415                                          struct ceph_msg_header *hdr,
 416                                          int *skip)
 417 {
 418         struct ceph_mon_client *monc = con->private;
 419         struct ceph_mon_generic_request *req;
 420         u64 tid = le64_to_cpu(hdr->tid);
 421         struct ceph_msg *m;
 422
 423         mutex_lock(&monc->mutex);
 424         req = __lookup_generic_req(monc, tid);
 425         if (!req) {
 426                 dout("get_generic_reply %lld dne\n", tid);
 427                 *skip = 1;
 428                 m = NULL;
 429         } else {
 430                 dout("get_generic_reply %lld got %p\n", tid, req->reply);
 431                 m = ceph_msg_get(req->reply);
 432                 /*
 433                  * we don't need to track the connection reading into
 434                  * this reply because we only have one open connection
 435                  * at a time, ever.
 436                  */
 437         }
 438         mutex_unlock(&monc->mutex);
 439         return m;
 440 }
 441
 442 static int do_generic_request(struct ceph_mon_client *monc,
 443                               struct ceph_mon_generic_request *req)
 444 {
 445         int err;
 446
 447         /* register request */
 448         mutex_lock(&monc->mutex);
 449         req->tid = ++monc->last_tid;
 450         req->request->hdr.tid = cpu_to_le64(req->tid);
 451         __insert_generic_request(monc, req);
 452         monc->num_generic_requests++;
 453         ceph_con_send(monc->con, ceph_msg_get(req->request));
 454         mutex_unlock(&monc->mutex);
 455
 456         err = wait_for_completion_interruptible(&req->completion);
 457
 458         mutex_lock(&monc->mutex);
 459         rb_erase(&req->node, &monc->generic_request_tree);
 460         monc->num_generic_requests--;
 461         mutex_unlock(&monc->mutex);
 462
 463         if (!err)
 464                 err = req->result;
 465         return err;
 466 }
 467
 468 /*
 469  * statfs
 470  */
 471 static void handle_statfs_reply(struct ceph_mon_client *monc,
 472                                 struct ceph_msg *msg)
 473 {
 474         struct ceph_mon_generic_request *req;
 475         struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
 476         u64 tid = le64_to_cpu(msg->hdr.tid);
 477
 478         if (msg->front.iov_len != sizeof(*reply))
 479                 goto bad;
 480         dout("handle_statfs_reply %p tid %llu\n", msg, tid);
 481
 482         mutex_lock(&monc->mutex);
 483         req = __lookup_generic_req(monc, tid);
 484         if (req) {
 485                 *(struct ceph_statfs *)req->buf = reply->st;
 486                 req->result = 0;
 487                 get_generic_request(req);
 488         }
 489         mutex_unlock(&monc->mutex);
 490         if (req) {
 491                 complete_all(&req->completion);
 492                 put_generic_request(req);
 493         }
 494         return;
 495
 496 bad:
 497         pr_err("corrupt generic reply, tid %llu\n", tid);
 498         ceph_msg_dump(msg);
 499 }
 500
 501 /*
 502  * Do a synchronous statfs().
 503  */
 504 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
 505 {
 506         struct ceph_mon_generic_request *req;
 507         struct ceph_mon_statfs *h;
 508         int err;
 509
 510         req = kzalloc(sizeof(*req), GFP_NOFS);
 511         if (!req)
 512                 return -ENOMEM;
 513
 514         kref_init(&req->kref);
 515         req->buf = buf;
 516         req->buf_len = sizeof(*buf);
 517         init_completion(&req->completion);
 518
 519         err = -ENOMEM;
 520         req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
 521                                     true);
 522         if (!req->request)
 523                 goto out;
 524         req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
 525                                   true);
 526         if (!req->reply)
 527                 goto out;
 528
 529         /* fill out request */
 530         h = req->request->front.iov_base;
 531         h->monhdr.have_version = 0;
 532         h->monhdr.session_mon = cpu_to_le16(-1);
 533         h->monhdr.session_mon_tid = 0;
 534         h->fsid = monc->monmap->fsid;
 535
 536         err = do_generic_request(monc, req);
 537
 538 out:
 539         kref_put(&req->kref, release_generic_request);
 540         return err;
 541 }
 542 EXPORT_SYMBOL(ceph_monc_do_statfs);
 543
 544 /*
 545  * pool ops
 546  */
 547 static int get_poolop_reply_buf(const char *src, size_t src_len,
 548                                 char *dst, size_t dst_len)
 549 {
 550         u32 buf_len;
 551
 552         if (src_len != sizeof(u32) + dst_len)
 553                 return -EINVAL;
 554
 555         buf_len = le32_to_cpu(*(u32 *)src);
 556         if (buf_len != dst_len)
 557                 return -EINVAL;
 558
 559         memcpy(dst, src + sizeof(u32), dst_len);
 560         return 0;
 561 }
 562
 563 static void handle_poolop_reply(struct ceph_mon_client *monc,
 564                                 struct ceph_msg *msg)
 565 {
 566         struct ceph_mon_generic_request *req;
 567         struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
 568         u64 tid = le64_to_cpu(msg->hdr.tid);
 569
 570         if (msg->front.iov_len < sizeof(*reply))
 571                 goto bad;
 572         dout("handle_poolop_reply %p tid %llu\n", msg, tid);
 573
 574         mutex_lock(&monc->mutex);
 575         req = __lookup_generic_req(monc, tid);
 576         if (req) {
 577                 if (req->buf_len &&
 578                     get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
 579                                      msg->front.iov_len - sizeof(*reply),
 580                                      req->buf, req->buf_len) < 0) {
 581                         mutex_unlock(&monc->mutex);
 582                         goto bad;
 583                 }
 584                 req->result = le32_to_cpu(reply->reply_code);
 585                 get_generic_request(req);
 586         }
 587         mutex_unlock(&monc->mutex);
 588         if (req) {
 589                 complete(&req->completion);
 590                 put_generic_request(req);
 591         }
 592         return;
 593
 594 bad:
 595         pr_err("corrupt generic reply, tid %llu\n", tid);
 596         ceph_msg_dump(msg);
 597 }
 598
 599 /*
 600  * Do a synchronous pool op.
 601  */
 602 int ceph_monc_do_poolop(struct ceph_mon_client *monc, u32 op,
 603                         u32 pool, u64 snapid,
 604                         char *buf, int len)
 605 {
 606         struct ceph_mon_generic_request *req;
 607         struct ceph_mon_poolop *h;
 608         int err;
 609
 610         req = kzalloc(sizeof(*req), GFP_NOFS);
 611         if (!req)
 612                 return -ENOMEM;
 613
 614         kref_init(&req->kref);
 615         req->buf = buf;
 616         req->buf_len = len;
 617         init_completion(&req->completion);
 618
 619         err = -ENOMEM;
 620         req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
 621                                     true);
 622         if (!req->request)
 623                 goto out;
 624         req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
 625                                   true);
 626         if (!req->reply)
 627                 goto out;
 628
 629         /* fill out request */
 630         req->request->hdr.version = cpu_to_le16(2);
 631         h = req->request->front.iov_base;
 632         h->monhdr.have_version = 0;
 633         h->monhdr.session_mon = cpu_to_le16(-1);
 634         h->monhdr.session_mon_tid = 0;
 635         h->fsid = monc->monmap->fsid;
 636         h->pool = cpu_to_le32(pool);
 637         h->op = cpu_to_le32(op);
 638         h->auid = 0;
 639         h->snapid = cpu_to_le64(snapid);
 640         h->name_len = 0;
 641
 642         err = do_generic_request(monc, req);
 643
 644 out:
 645         kref_put(&req->kref, release_generic_request);
 646         return err;
 647 }
 648
 649 int ceph_monc_create_snapid(struct ceph_mon_client *monc,
 650                             u32 pool, u64 *snapid)
 651 {
 652         return ceph_monc_do_poolop(monc,  POOL_OP_CREATE_UNMANAGED_SNAP,
 653                                    pool, 0, (char *)snapid, sizeof(*snapid));
 654
 655 }
 656 EXPORT_SYMBOL(ceph_monc_create_snapid);
 657
 658 int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
 659                             u32 pool, u64 snapid)
 660 {
 661         return ceph_monc_do_poolop(monc,  POOL_OP_CREATE_UNMANAGED_SNAP,
 662                                    pool, snapid, 0, 0);
 663
 664 }
 665
 666 /*
 667  * Resend pending generic requests.
 668  */
 669 static void __resend_generic_request(struct ceph_mon_client *monc)
 670 {
 671         struct ceph_mon_generic_request *req;
 672         struct rb_node *p;
 673
 674         for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
 675                 req = rb_entry(p, struct ceph_mon_generic_request, node);
 676                 ceph_con_revoke(monc->con, req->request);
 677                 ceph_con_send(monc->con, ceph_msg_get(req->request));
 678         }
 679 }
 680
 681 /*
 682  * Delayed work.  If we haven't mounted yet, retry.  Otherwise,
 683  * renew/retry subscription as needed (in case it is timing out, or we
 684  * got an ENOMEM).  And keep the monitor connection alive.
 685  */
 686 static void delayed_work(struct work_struct *work)
 687 {
 688         struct ceph_mon_client *monc =
 689                 container_of(work, struct ceph_mon_client, delayed_work.work);
 690
 691         dout("monc delayed_work\n");
 692         mutex_lock(&monc->mutex);
 693         if (monc->hunting) {
 694                 __close_session(monc);
 695                 __open_session(monc);  /* continue hunting */
 696         } else {
 697                 ceph_con_keepalive(monc->con);
 698
 699                 __validate_auth(monc);
 700
 701                 if (monc->auth->ops->is_authenticated(monc->auth))
 702                         __send_subscribe(monc);
 703         }
 704         __schedule_delayed(monc);
 705         mutex_unlock(&monc->mutex);
 706 }
 707
 708 /*
 709  * On startup, we build a temporary monmap populated with the IPs
 710  * provided by mount(2).
 711  */
 712 static int build_initial_monmap(struct ceph_mon_client *monc)
 713 {
 714         struct ceph_options *opt = monc->client->options;
 715         struct ceph_entity_addr *mon_addr = opt->mon_addr;
 716         int num_mon = opt->num_mon;
 717         int i;
 718
 719         /* build initial monmap */
 720         monc->monmap = kzalloc(sizeof(*monc->monmap) +
 721                                num_mon*sizeof(monc->monmap->mon_inst[0]),
 722                                GFP_KERNEL);
 723         if (!monc->monmap)
 724                 return -ENOMEM;
 725         for (i = 0; i < num_mon; i++) {
 726                 monc->monmap->mon_inst[i].addr = mon_addr[i];
 727                 monc->monmap->mon_inst[i].addr.nonce = 0;
 728                 monc->monmap->mon_inst[i].name.type =
 729                         CEPH_ENTITY_TYPE_MON;
 730                 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
 731         }
 732         monc->monmap->num_mon = num_mon;
 733         monc->have_fsid = false;
 734         return 0;
 735 }
 736
 737 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
 738 {
 739         int err = 0;
 740
 741         dout("init\n");
 742         memset(monc, 0, sizeof(*monc));
 743         monc->client = cl;
 744         monc->monmap = NULL;
 745         mutex_init(&monc->mutex);
 746
 747         err = build_initial_monmap(monc);
 748         if (err)
 749                 goto out;
 750
 751         /* connection */
 752         monc->con = kmalloc(sizeof(*monc->con), GFP_KERNEL);
 753         if (!monc->con)
 754                 goto out_monmap;
 755         ceph_con_init(monc->client->msgr, monc->con);
 756         monc->con->private = monc;
 757         monc->con->ops = &mon_con_ops;
 758
 759         /* authentication */
 760         monc->auth = ceph_auth_init(cl->options->name,
 761                                     cl->options->key);
 762         if (IS_ERR(monc->auth)) {
 763                 err = PTR_ERR(monc->auth);
 764                 goto out_con;
 765         }
 766         monc->auth->want_keys =
 767                 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
 768                 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
 769
 770         /* msgs */
 771         err = -ENOMEM;
 772         monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
 773                                      sizeof(struct ceph_mon_subscribe_ack),
 774                                      GFP_NOFS, true);
 775         if (!monc->m_subscribe_ack)
 776                 goto out_auth;
 777
 778         monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
 779                                          true);
 780         if (!monc->m_subscribe)
 781                 goto out_subscribe_ack;
 782
 783         monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
 784                                           true);
 785         if (!monc->m_auth_reply)
 786                 goto out_subscribe;
 787
 788         monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
 789         monc->pending_auth = 0;
 790         if (!monc->m_auth)
 791                 goto out_auth_reply;
 792
 793         monc->cur_mon = -1;
 794         monc->hunting = true;
 795         monc->sub_renew_after = jiffies;
 796         monc->sub_sent = 0;
 797
 798         INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
 799         monc->generic_request_tree = RB_ROOT;
 800         monc->num_generic_requests = 0;
 801         monc->last_tid = 0;
 802
 803         monc->have_mdsmap = 0;
 804         monc->have_osdmap = 0;
 805         monc->want_next_osdmap = 1;
 806         return 0;
 807
 808 out_auth_reply:
 809         ceph_msg_put(monc->m_auth_reply);
 810 out_subscribe:
 811         ceph_msg_put(monc->m_subscribe);
 812 out_subscribe_ack:
 813         ceph_msg_put(monc->m_subscribe_ack);
 814 out_auth:
 815         ceph_auth_destroy(monc->auth);
 816 out_con:
 817         monc->con->ops->put(monc->con);
 818 out_monmap:
 819         kfree(monc->monmap);
 820 out:
 821         return err;
 822 }
 823 EXPORT_SYMBOL(ceph_monc_init);
 824
 825 void ceph_monc_stop(struct ceph_mon_client *monc)
 826 {
 827         dout("stop\n");
 828         cancel_delayed_work_sync(&monc->delayed_work);
 829
 830         mutex_lock(&monc->mutex);
 831         __close_session(monc);
 832
 833         monc->con->private = NULL;
 834         monc->con->ops->put(monc->con);
 835         monc->con = NULL;
 836
 837         mutex_unlock(&monc->mutex);
 838
 839         ceph_auth_destroy(monc->auth);
 840
 841         ceph_msg_put(monc->m_auth);
 842         ceph_msg_put(monc->m_auth_reply);
 843         ceph_msg_put(monc->m_subscribe);
 844         ceph_msg_put(monc->m_subscribe_ack);
 845
 846         kfree(monc->monmap);
 847 }
 848 EXPORT_SYMBOL(ceph_monc_stop);
 849
 850 static void handle_auth_reply(struct ceph_mon_client *monc,
 851                               struct ceph_msg *msg)
 852 {
 853         int ret;
 854         int was_auth = 0;
 855
 856         mutex_lock(&monc->mutex);
 857         if (monc->auth->ops)
 858                 was_auth = monc->auth->ops->is_authenticated(monc->auth);
 859         monc->pending_auth = 0;
 860         ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
 861                                      msg->front.iov_len,
 862                                      monc->m_auth->front.iov_base,
 863                                      monc->m_auth->front_max);
 864         if (ret < 0) {
 865                 monc->client->auth_err = ret;
 866                 wake_up_all(&monc->client->auth_wq);
 867         } else if (ret > 0) {
 868                 __send_prepared_auth_request(monc, ret);
 869         } else if (!was_auth && monc->auth->ops->is_authenticated(monc->auth)) {
 870                 dout("authenticated, starting session\n");
 871
 872                 monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
 873                 monc->client->msgr->inst.name.num =
 874                                         cpu_to_le64(monc->auth->global_id);
 875
 876                 __send_subscribe(monc);
 877                 __resend_generic_request(monc);
 878         }
 879         mutex_unlock(&monc->mutex);
 880 }
 881
 882 static int __validate_auth(struct ceph_mon_client *monc)
 883 {
 884         int ret;
 885
 886         if (monc->pending_auth)
 887                 return 0;
 888
 889         ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
 890                               monc->m_auth->front_max);
 891         if (ret <= 0)
 892                 return ret; /* either an error, or no need to authenticate */
 893         __send_prepared_auth_request(monc, ret);
 894         return 0;
 895 }
 896
 897 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
 898 {
 899         int ret;
 900
 901         mutex_lock(&monc->mutex);
 902         ret = __validate_auth(monc);
 903         mutex_unlock(&monc->mutex);
 904         return ret;
 905 }
 906 EXPORT_SYMBOL(ceph_monc_validate_auth);
 907
 908 /*
 909  * handle incoming message
 910  */
 911 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
 912 {
 913         struct ceph_mon_client *monc = con->private;
 914         int type = le16_to_cpu(msg->hdr.type);
 915
 916         if (!monc)
 917                 return;
 918
 919         switch (type) {
 920         case CEPH_MSG_AUTH_REPLY:
 921                 handle_auth_reply(monc, msg);
 922                 break;
 923
 924         case CEPH_MSG_MON_SUBSCRIBE_ACK:
 925                 handle_subscribe_ack(monc, msg);
 926                 break;
 927
 928         case CEPH_MSG_STATFS_REPLY:
 929                 handle_statfs_reply(monc, msg);
 930                 break;
 931
 932         case CEPH_MSG_POOLOP_REPLY:
 933                 handle_poolop_reply(monc, msg);
 934                 break;
 935
 936         case CEPH_MSG_MON_MAP:
 937                 ceph_monc_handle_map(monc, msg);
 938                 break;
 939
 940         case CEPH_MSG_OSD_MAP:
 941                 ceph_osdc_handle_map(&monc->client->osdc, msg);
 942                 break;
 943
 944         default:
 945                 /* can the chained handler handle it? */
 946                 if (monc->client->extra_mon_dispatch &&
 947                     monc->client->extra_mon_dispatch(monc->client, msg) == 0)
 948                         break;
 949
 950                 pr_err("received unknown message type %d %s\n", type,
 951                        ceph_msg_type_name(type));
 952         }
 953         ceph_msg_put(msg);
 954 }
 955
 956 /*
 957  * Allocate memory for incoming message
 958  */
 959 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
 960                                       struct ceph_msg_header *hdr,
 961                                       int *skip)
 962 {
 963         struct ceph_mon_client *monc = con->private;
 964         int type = le16_to_cpu(hdr->type);
 965         int front_len = le32_to_cpu(hdr->front_len);
 966         struct ceph_msg *m = NULL;
 967
 968         *skip = 0;
 969
 970         switch (type) {
 971         case CEPH_MSG_MON_SUBSCRIBE_ACK:
 972                 m = ceph_msg_get(monc->m_subscribe_ack);
 973                 break;
 974         case CEPH_MSG_POOLOP_REPLY:
 975         case CEPH_MSG_STATFS_REPLY:
 976                 return get_generic_reply(con, hdr, skip);
 977         case CEPH_MSG_AUTH_REPLY:
 978                 m = ceph_msg_get(monc->m_auth_reply);
 979                 break;
 980         case CEPH_MSG_MON_MAP:
 981         case CEPH_MSG_MDS_MAP:
 982         case CEPH_MSG_OSD_MAP:
 983                 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
 984                 break;
 985         }
 986
 987         if (!m) {
 988                 pr_info("alloc_msg unknown type %d\n", type);
 989                 *skip = 1;
 990         }
 991         return m;
 992 }
 993
 994 /*
 995  * If the monitor connection resets, pick a new monitor and resubmit
 996  * any pending requests.
 997  */
 998 static void mon_fault(struct ceph_connection *con)
 999 {
1000         struct ceph_mon_client *monc = con->private;
1001
1002         if (!monc)
1003                 return;
1004
1005         dout("mon_fault\n");
1006         mutex_lock(&monc->mutex);
1007         if (!con->private)
1008                 goto out;
1009
1010         if (!monc->hunting)
1011                 pr_info("mon%d %s session lost, "
1012                         "hunting for new mon\n", monc->cur_mon,
1013                         ceph_pr_addr(&monc->con->peer_addr.in_addr));
1014
1015         __close_session(monc);
1016         if (!monc->hunting) {
1017                 /* start hunting */
1018                 monc->hunting = true;
1019                 __open_session(monc);
1020         } else {
1021                 /* already hunting, let's wait a bit */
1022                 __schedule_delayed(monc);
1023         }
1024 out:
1025         mutex_unlock(&monc->mutex);
1026 }
1027
1028 static const struct ceph_connection_operations mon_con_ops = {
1029         .get = ceph_con_get,
1030         .put = ceph_con_put,
1031         .dispatch = dispatch,
1032         .fault = mon_fault,
1033         .alloc_msg = mon_alloc_msg,
1034 };