block/rbd.c

   1 /*
   2  * QEMU Block driver for RADOS (Ceph)
   3  *
   4  * Copyright (C) 2010-2011 Christian Brunner <chb@muc.de>,
   5  *                         Josh Durgin <josh.durgin@dreamhost.com>
   6  *
   7  * This work is licensed under the terms of the GNU GPL, version 2.  See
   8  * the COPYING file in the top-level directory.
   9  *
  10  * Contributions after 2012-01-13 are licensed under the terms of the
  11  * GNU GPL, version 2 or (at your option) any later version.
  12  */
  13
  14 #include <inttypes.h>
  15
  16 #include "qemu-common.h"
  17 #include "qemu/error-report.h"
  18 #include "block/block_int.h"
  19
  20 #include <rbd/librbd.h>
  21
  22 /*
  23  * When specifying the image filename use:
  24  *
  25  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
  26  *
  27  * poolname must be the name of an existing rados pool.
  28  *
  29  * devicename is the name of the rbd image.
  30  *
  31  * Each option given is used to configure rados, and may be any valid
  32  * Ceph option, "id", or "conf".
  33  *
  34  * The "id" option indicates what user we should authenticate as to
  35  * the Ceph cluster.  If it is excluded we will use the Ceph default
  36  * (normally 'admin').
  37  *
  38  * The "conf" option specifies a Ceph configuration file to read.  If
  39  * it is not specified, we will read from the default Ceph locations
  40  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
  41  * file, specify conf=/dev/null.
  42  *
  43  * Configuration values containing :, @, or = can be escaped with a
  44  * leading "\".
  45  */
  46
  47 /* rbd_aio_discard added in 0.1.2 */
  48 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
  49 #define LIBRBD_SUPPORTS_DISCARD
  50 #else
  51 #undef LIBRBD_SUPPORTS_DISCARD
  52 #endif
  53
  54 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
  55
  56 #define RBD_MAX_CONF_NAME_SIZE 128
  57 #define RBD_MAX_CONF_VAL_SIZE 512
  58 #define RBD_MAX_CONF_SIZE 1024
  59 #define RBD_MAX_POOL_NAME_SIZE 128
  60 #define RBD_MAX_SNAP_NAME_SIZE 128
  61 #define RBD_MAX_SNAPS 100
  62
  63 typedef enum {
  64     RBD_AIO_READ,
  65     RBD_AIO_WRITE,
  66     RBD_AIO_DISCARD,
  67     RBD_AIO_FLUSH
  68 } RBDAIOCmd;
  69
  70 typedef struct RBDAIOCB {
  71     BlockDriverAIOCB common;
  72     QEMUBH *bh;
  73     int64_t ret;
  74     QEMUIOVector *qiov;
  75     char *bounce;
  76     RBDAIOCmd cmd;
  77     int64_t sector_num;
  78     int error;
  79     struct BDRVRBDState *s;
  80     int cancelled;
  81     int status;
  82 } RBDAIOCB;
  83
  84 typedef struct RADOSCB {
  85     int rcbid;
  86     RBDAIOCB *acb;
  87     struct BDRVRBDState *s;
  88     int done;
  89     int64_t size;
  90     char *buf;
  91     int64_t ret;
  92 } RADOSCB;
  93
  94 #define RBD_FD_READ 0
  95 #define RBD_FD_WRITE 1
  96
  97 typedef struct BDRVRBDState {
  98     int fds[2];
  99     rados_t cluster;
 100     rados_ioctx_t io_ctx;
 101     rbd_image_t image;
 102     char name[RBD_MAX_IMAGE_NAME_SIZE];
 103     char *snap;
 104     int event_reader_pos;
 105     RADOSCB *event_rcb;
 106 } BDRVRBDState;
 107
 108 static void rbd_aio_bh_cb(void *opaque);
 109
 110 static int qemu_rbd_next_tok(char *dst, int dst_len,
 111                              char *src, char delim,
 112                              const char *name,
 113                              char **p)
 114 {
 115     int l;
 116     char *end;
 117
 118     *p = NULL;
 119
 120     if (delim != '\0') {
 121         for (end = src; *end; ++end) {
 122             if (*end == delim) {
 123                 break;
 124             }
 125             if (*end == '\\' && end[1] != '\0') {
 126                 end++;
 127             }
 128         }
 129         if (*end == delim) {
 130             *p = end + 1;
 131             *end = '\0';
 132         }
 133     }
 134     l = strlen(src);
 135     if (l >= dst_len) {
 136         error_report("%s too long", name);
 137         return -EINVAL;
 138     } else if (l == 0) {
 139         error_report("%s too short", name);
 140         return -EINVAL;
 141     }
 142
 143     pstrcpy(dst, dst_len, src);
 144
 145     return 0;
 146 }
 147
 148 static void qemu_rbd_unescape(char *src)
 149 {
 150     char *p;
 151
 152     for (p = src; *src; ++src, ++p) {
 153         if (*src == '\\' && src[1] != '\0') {
 154             src++;
 155         }
 156         *p = *src;
 157     }
 158     *p = '\0';
 159 }
 160
 161 static int qemu_rbd_parsename(const char *filename,
 162                               char *pool, int pool_len,
 163                               char *snap, int snap_len,
 164                               char *name, int name_len,
 165                               char *conf, int conf_len)
 166 {
 167     const char *start;
 168     char *p, *buf;
 169     int ret;
 170
 171     if (!strstart(filename, "rbd:", &start)) {
 172         return -EINVAL;
 173     }
 174
 175     buf = g_strdup(start);
 176     p = buf;
 177     *snap = '\0';
 178     *conf = '\0';
 179
 180     ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
 181     if (ret < 0 || !p) {
 182         ret = -EINVAL;
 183         goto done;
 184     }
 185     qemu_rbd_unescape(pool);
 186
 187     if (strchr(p, '@')) {
 188         ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
 189         if (ret < 0) {
 190             goto done;
 191         }
 192         ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
 193         qemu_rbd_unescape(snap);
 194     } else {
 195         ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
 196     }
 197     qemu_rbd_unescape(name);
 198     if (ret < 0 || !p) {
 199         goto done;
 200     }
 201
 202     ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
 203
 204 done:
 205     g_free(buf);
 206     return ret;
 207 }
 208
 209 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 210 {
 211     const char *p = conf;
 212
 213     while (*p) {
 214         int len;
 215         const char *end = strchr(p, ':');
 216
 217         if (end) {
 218             len = end - p;
 219         } else {
 220             len = strlen(p);
 221         }
 222
 223         if (strncmp(p, "id=", 3) == 0) {
 224             len -= 3;
 225             strncpy(clientname, p + 3, len);
 226             clientname[len] = '\0';
 227             return clientname;
 228         }
 229         if (end == NULL) {
 230             break;
 231         }
 232         p = end + 1;
 233     }
 234     return NULL;
 235 }
 236
 237 static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
 238 {
 239     char *p, *buf;
 240     char name[RBD_MAX_CONF_NAME_SIZE];
 241     char value[RBD_MAX_CONF_VAL_SIZE];
 242     int ret = 0;
 243
 244     buf = g_strdup(conf);
 245     p = buf;
 246
 247     while (p) {
 248         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 249                                 '=', "conf option name", &p);
 250         if (ret < 0) {
 251             break;
 252         }
 253         qemu_rbd_unescape(name);
 254
 255         if (!p) {
 256             error_report("conf option %s has no value", name);
 257             ret = -EINVAL;
 258             break;
 259         }
 260
 261         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 262                                 ':', "conf option value", &p);
 263         if (ret < 0) {
 264             break;
 265         }
 266         qemu_rbd_unescape(value);
 267
 268         if (strcmp(name, "conf") == 0) {
 269             ret = rados_conf_read_file(cluster, value);
 270             if (ret < 0) {
 271                 error_report("error reading conf file %s", value);
 272                 break;
 273             }
 274         } else if (strcmp(name, "id") == 0) {
 275             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 276         } else {
 277             ret = rados_conf_set(cluster, name, value);
 278             if (ret < 0) {
 279                 error_report("invalid conf option %s", name);
 280                 ret = -EINVAL;
 281                 break;
 282             }
 283         }
 284     }
 285
 286     g_free(buf);
 287     return ret;
 288 }
 289
 290 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options,
 291                            Error **errp)
 292 {
 293     int64_t bytes = 0;
 294     int64_t objsize;
 295     int obj_order = 0;
 296     char pool[RBD_MAX_POOL_NAME_SIZE];
 297     char name[RBD_MAX_IMAGE_NAME_SIZE];
 298     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 299     char conf[RBD_MAX_CONF_SIZE];
 300     char clientname_buf[RBD_MAX_CONF_SIZE];
 301     char *clientname;
 302     rados_t cluster;
 303     rados_ioctx_t io_ctx;
 304     int ret;
 305
 306     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 307                            snap_buf, sizeof(snap_buf),
 308                            name, sizeof(name),
 309                            conf, sizeof(conf)) < 0) {
 310         return -EINVAL;
 311     }
 312
 313     /* Read out options */
 314     while (options && options->name) {
 315         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 316             bytes = options->value.n;
 317         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
 318             if (options->value.n) {
 319                 objsize = options->value.n;
 320                 if ((objsize - 1) & objsize) {    /* not a power of 2? */
 321                     error_report("obj size needs to be power of 2");
 322                     return -EINVAL;
 323                 }
 324                 if (objsize < 4096) {
 325                     error_report("obj size too small");
 326                     return -EINVAL;
 327                 }
 328                 obj_order = ffs(objsize) - 1;
 329             }
 330         }
 331         options++;
 332     }
 333
 334     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 335     if (rados_create(&cluster, clientname) < 0) {
 336         error_report("error initializing");
 337         return -EIO;
 338     }
 339
 340     if (strstr(conf, "conf=") == NULL) {
 341         /* try default location, but ignore failure */
 342         rados_conf_read_file(cluster, NULL);
 343     }
 344
 345     if (conf[0] != '\0' &&
 346         qemu_rbd_set_conf(cluster, conf) < 0) {
 347         error_report("error setting config options");
 348         rados_shutdown(cluster);
 349         return -EIO;
 350     }
 351
 352     if (rados_connect(cluster) < 0) {
 353         error_report("error connecting");
 354         rados_shutdown(cluster);
 355         return -EIO;
 356     }
 357
 358     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 359         error_report("error opening pool %s", pool);
 360         rados_shutdown(cluster);
 361         return -EIO;
 362     }
 363
 364     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 365     rados_ioctx_destroy(io_ctx);
 366     rados_shutdown(cluster);
 367
 368     return ret;
 369 }
 370
 371 /*
 372  * This aio completion is being called from qemu_rbd_aio_event_reader()
 373  * and runs in qemu context. It schedules a bh, but just in case the aio
 374  * was not cancelled before.
 375  */
 376 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 377 {
 378     RBDAIOCB *acb = rcb->acb;
 379     int64_t r;
 380
 381     r = rcb->ret;
 382
 383     if (acb->cmd != RBD_AIO_READ) {
 384         if (r < 0) {
 385             acb->ret = r;
 386             acb->error = 1;
 387         } else if (!acb->error) {
 388             acb->ret = rcb->size;
 389         }
 390     } else {
 391         if (r < 0) {
 392             memset(rcb->buf, 0, rcb->size);
 393             acb->ret = r;
 394             acb->error = 1;
 395         } else if (r < rcb->size) {
 396             memset(rcb->buf + r, 0, rcb->size - r);
 397             if (!acb->error) {
 398                 acb->ret = rcb->size;
 399             }
 400         } else if (!acb->error) {
 401             acb->ret = r;
 402         }
 403     }
 404     /* Note that acb->bh can be NULL in case where the aio was cancelled */
 405     acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb);
 406     qemu_bh_schedule(acb->bh);
 407     g_free(rcb);
 408 }
 409
 410 /*
 411  * aio fd read handler. It runs in the qemu context and calls the
 412  * completion handling of completed rados aio operations.
 413  */
 414 static void qemu_rbd_aio_event_reader(void *opaque)
 415 {
 416     BDRVRBDState *s = opaque;
 417
 418     ssize_t ret;
 419
 420     do {
 421         char *p = (char *)&s->event_rcb;
 422
 423         /* now read the rcb pointer that was sent from a non qemu thread */
 424         ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos,
 425                    sizeof(s->event_rcb) - s->event_reader_pos);
 426         if (ret > 0) {
 427             s->event_reader_pos += ret;
 428             if (s->event_reader_pos == sizeof(s->event_rcb)) {
 429                 s->event_reader_pos = 0;
 430                 qemu_rbd_complete_aio(s->event_rcb);
 431             }
 432         }
 433     } while (ret < 0 && errno == EINTR);
 434 }
 435
 436 /* TODO Convert to fine grained options */
 437 static QemuOptsList runtime_opts = {
 438     .name = "rbd",
 439     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 440     .desc = {
 441         {
 442             .name = "filename",
 443             .type = QEMU_OPT_STRING,
 444             .help = "Specification of the rbd image",
 445         },
 446         { /* end of list */ }
 447     },
 448 };
 449
 450 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 451                          Error **errp)
 452 {
 453     BDRVRBDState *s = bs->opaque;
 454     char pool[RBD_MAX_POOL_NAME_SIZE];
 455     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 456     char conf[RBD_MAX_CONF_SIZE];
 457     char clientname_buf[RBD_MAX_CONF_SIZE];
 458     char *clientname;
 459     QemuOpts *opts;
 460     Error *local_err = NULL;
 461     const char *filename;
 462     int r;
 463
 464     opts = qemu_opts_create_nofail(&runtime_opts);
 465     qemu_opts_absorb_qdict(opts, options, &local_err);
 466     if (error_is_set(&local_err)) {
 467         qerror_report_err(local_err);
 468         error_free(local_err);
 469         qemu_opts_del(opts);
 470         return -EINVAL;
 471     }
 472
 473     filename = qemu_opt_get(opts, "filename");
 474
 475     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 476                            snap_buf, sizeof(snap_buf),
 477                            s->name, sizeof(s->name),
 478                            conf, sizeof(conf)) < 0) {
 479         r = -EINVAL;
 480         goto failed_opts;
 481     }
 482
 483     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 484     r = rados_create(&s->cluster, clientname);
 485     if (r < 0) {
 486         error_report("error initializing");
 487         goto failed_opts;
 488     }
 489
 490     s->snap = NULL;
 491     if (snap_buf[0] != '\0') {
 492         s->snap = g_strdup(snap_buf);
 493     }
 494
 495     /*
 496      * Fallback to more conservative semantics if setting cache
 497      * options fails. Ignore errors from setting rbd_cache because the
 498      * only possible error is that the option does not exist, and
 499      * librbd defaults to no caching. If write through caching cannot
 500      * be set up, fall back to no caching.
 501      */
 502     if (flags & BDRV_O_NOCACHE) {
 503         rados_conf_set(s->cluster, "rbd_cache", "false");
 504     } else {
 505         rados_conf_set(s->cluster, "rbd_cache", "true");
 506     }
 507
 508     if (strstr(conf, "conf=") == NULL) {
 509         /* try default location, but ignore failure */
 510         rados_conf_read_file(s->cluster, NULL);
 511     }
 512
 513     if (conf[0] != '\0') {
 514         r = qemu_rbd_set_conf(s->cluster, conf);
 515         if (r < 0) {
 516             error_report("error setting config options");
 517             goto failed_shutdown;
 518         }
 519     }
 520
 521     r = rados_connect(s->cluster);
 522     if (r < 0) {
 523         error_report("error connecting");
 524         goto failed_shutdown;
 525     }
 526
 527     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 528     if (r < 0) {
 529         error_report("error opening pool %s", pool);
 530         goto failed_shutdown;
 531     }
 532
 533     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 534     if (r < 0) {
 535         error_report("error reading header from %s", s->name);
 536         goto failed_open;
 537     }
 538
 539     bs->read_only = (s->snap != NULL);
 540
 541     s->event_reader_pos = 0;
 542     r = qemu_pipe(s->fds);
 543     if (r < 0) {
 544         error_report("error opening eventfd");
 545         goto failed;
 546     }
 547     fcntl(s->fds[0], F_SETFL, O_NONBLOCK);
 548     fcntl(s->fds[1], F_SETFL, O_NONBLOCK);
 549     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader,
 550                             NULL, s);
 551
 552
 553     qemu_opts_del(opts);
 554     return 0;
 555
 556 failed:
 557     rbd_close(s->image);
 558 failed_open:
 559     rados_ioctx_destroy(s->io_ctx);
 560 failed_shutdown:
 561     rados_shutdown(s->cluster);
 562     g_free(s->snap);
 563 failed_opts:
 564     qemu_opts_del(opts);
 565     return r;
 566 }
 567
 568 static void qemu_rbd_close(BlockDriverState *bs)
 569 {
 570     BDRVRBDState *s = bs->opaque;
 571
 572     close(s->fds[0]);
 573     close(s->fds[1]);
 574     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL);
 575
 576     rbd_close(s->image);
 577     rados_ioctx_destroy(s->io_ctx);
 578     g_free(s->snap);
 579     rados_shutdown(s->cluster);
 580 }
 581
 582 /*
 583  * Cancel aio. Since we don't reference acb in a non qemu threads,
 584  * it is safe to access it here.
 585  */
 586 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
 587 {
 588     RBDAIOCB *acb = (RBDAIOCB *) blockacb;
 589     acb->cancelled = 1;
 590
 591     while (acb->status == -EINPROGRESS) {
 592         qemu_aio_wait();
 593     }
 594
 595     qemu_aio_release(acb);
 596 }
 597
 598 static const AIOCBInfo rbd_aiocb_info = {
 599     .aiocb_size = sizeof(RBDAIOCB),
 600     .cancel = qemu_rbd_aio_cancel,
 601 };
 602
 603 static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb)
 604 {
 605     int ret = 0;
 606     while (1) {
 607         fd_set wfd;
 608         int fd = s->fds[RBD_FD_WRITE];
 609
 610         /* send the op pointer to the qemu thread that is responsible
 611            for the aio/op completion. Must do it in a qemu thread context */
 612         ret = write(fd, (void *)&rcb, sizeof(rcb));
 613         if (ret >= 0) {
 614             break;
 615         }
 616         if (errno == EINTR) {
 617             continue;
 618         }
 619         if (errno != EAGAIN) {
 620             break;
 621         }
 622
 623         FD_ZERO(&wfd);
 624         FD_SET(fd, &wfd);
 625         do {
 626             ret = select(fd + 1, NULL, &wfd, NULL, NULL);
 627         } while (ret < 0 && errno == EINTR);
 628     }
 629
 630     return ret;
 631 }
 632
 633 /*
 634  * This is the callback function for rbd_aio_read and _write
 635  *
 636  * Note: this function is being called from a non qemu thread so
 637  * we need to be careful about what we do here. Generally we only
 638  * write to the block notification pipe, and do the rest of the
 639  * io completion handling from qemu_rbd_aio_event_reader() which
 640  * runs in a qemu context.
 641  */
 642 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 643 {
 644     int ret;
 645     rcb->ret = rbd_aio_get_return_value(c);
 646     rbd_aio_release(c);
 647     ret = qemu_rbd_send_pipe(rcb->s, rcb);
 648     if (ret < 0) {
 649         error_report("failed writing to acb->s->fds");
 650         g_free(rcb);
 651     }
 652 }
 653
 654 /* Callback when all queued rbd_aio requests are complete */
 655
 656 static void rbd_aio_bh_cb(void *opaque)
 657 {
 658     RBDAIOCB *acb = opaque;
 659
 660     if (acb->cmd == RBD_AIO_READ) {
 661         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 662     }
 663     qemu_vfree(acb->bounce);
 664     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 665     qemu_bh_delete(acb->bh);
 666     acb->bh = NULL;
 667     acb->status = 0;
 668
 669     if (!acb->cancelled) {
 670         qemu_aio_release(acb);
 671     }
 672 }
 673
 674 static int rbd_aio_discard_wrapper(rbd_image_t image,
 675                                    uint64_t off,
 676                                    uint64_t len,
 677                                    rbd_completion_t comp)
 678 {
 679 #ifdef LIBRBD_SUPPORTS_DISCARD
 680     return rbd_aio_discard(image, off, len, comp);
 681 #else
 682     return -ENOTSUP;
 683 #endif
 684 }
 685
 686 static int rbd_aio_flush_wrapper(rbd_image_t image,
 687                                  rbd_completion_t comp)
 688 {
 689 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 690     return rbd_aio_flush(image, comp);
 691 #else
 692     return -ENOTSUP;
 693 #endif
 694 }
 695
 696 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
 697                                        int64_t sector_num,
 698                                        QEMUIOVector *qiov,
 699                                        int nb_sectors,
 700                                        BlockDriverCompletionFunc *cb,
 701                                        void *opaque,
 702                                        RBDAIOCmd cmd)
 703 {
 704     RBDAIOCB *acb;
 705     RADOSCB *rcb;
 706     rbd_completion_t c;
 707     int64_t off, size;
 708     char *buf;
 709     int r;
 710
 711     BDRVRBDState *s = bs->opaque;
 712
 713     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 714     acb->cmd = cmd;
 715     acb->qiov = qiov;
 716     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 717         acb->bounce = NULL;
 718     } else {
 719         acb->bounce = qemu_blockalign(bs, qiov->size);
 720     }
 721     acb->ret = 0;
 722     acb->error = 0;
 723     acb->s = s;
 724     acb->cancelled = 0;
 725     acb->bh = NULL;
 726     acb->status = -EINPROGRESS;
 727
 728     if (cmd == RBD_AIO_WRITE) {
 729         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 730     }
 731
 732     buf = acb->bounce;
 733
 734     off = sector_num * BDRV_SECTOR_SIZE;
 735     size = nb_sectors * BDRV_SECTOR_SIZE;
 736
 737     rcb = g_malloc(sizeof(RADOSCB));
 738     rcb->done = 0;
 739     rcb->acb = acb;
 740     rcb->buf = buf;
 741     rcb->s = acb->s;
 742     rcb->size = size;
 743     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 744     if (r < 0) {
 745         goto failed;
 746     }
 747
 748     switch (cmd) {
 749     case RBD_AIO_WRITE:
 750         r = rbd_aio_write(s->image, off, size, buf, c);
 751         break;
 752     case RBD_AIO_READ:
 753         r = rbd_aio_read(s->image, off, size, buf, c);
 754         break;
 755     case RBD_AIO_DISCARD:
 756         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 757         break;
 758     case RBD_AIO_FLUSH:
 759         r = rbd_aio_flush_wrapper(s->image, c);
 760         break;
 761     default:
 762         r = -EINVAL;
 763     }
 764
 765     if (r < 0) {
 766         goto failed;
 767     }
 768
 769     return &acb->common;
 770
 771 failed:
 772     g_free(rcb);
 773     qemu_aio_release(acb);
 774     return NULL;
 775 }
 776
 777 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 778                                             int64_t sector_num,
 779                                             QEMUIOVector *qiov,
 780                                             int nb_sectors,
 781                                             BlockDriverCompletionFunc *cb,
 782                                             void *opaque)
 783 {
 784     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 785                          RBD_AIO_READ);
 786 }
 787
 788 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 789                                              int64_t sector_num,
 790                                              QEMUIOVector *qiov,
 791                                              int nb_sectors,
 792                                              BlockDriverCompletionFunc *cb,
 793                                              void *opaque)
 794 {
 795     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 796                          RBD_AIO_WRITE);
 797 }
 798
 799 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 800 static BlockDriverAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 801                                             BlockDriverCompletionFunc *cb,
 802                                             void *opaque)
 803 {
 804     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 805 }
 806
 807 #else
 808
 809 static int qemu_rbd_co_flush(BlockDriverState *bs)
 810 {
 811 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 812     /* rbd_flush added in 0.1.1 */
 813     BDRVRBDState *s = bs->opaque;
 814     return rbd_flush(s->image);
 815 #else
 816     return 0;
 817 #endif
 818 }
 819 #endif
 820
 821 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 822 {
 823     BDRVRBDState *s = bs->opaque;
 824     rbd_image_info_t info;
 825     int r;
 826
 827     r = rbd_stat(s->image, &info, sizeof(info));
 828     if (r < 0) {
 829         return r;
 830     }
 831
 832     bdi->cluster_size = info.obj_size;
 833     return 0;
 834 }
 835
 836 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 837 {
 838     BDRVRBDState *s = bs->opaque;
 839     rbd_image_info_t info;
 840     int r;
 841
 842     r = rbd_stat(s->image, &info, sizeof(info));
 843     if (r < 0) {
 844         return r;
 845     }
 846
 847     return info.size;
 848 }
 849
 850 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 851 {
 852     BDRVRBDState *s = bs->opaque;
 853     int r;
 854
 855     r = rbd_resize(s->image, offset);
 856     if (r < 0) {
 857         return r;
 858     }
 859
 860     return 0;
 861 }
 862
 863 static int qemu_rbd_snap_create(BlockDriverState *bs,
 864                                 QEMUSnapshotInfo *sn_info)
 865 {
 866     BDRVRBDState *s = bs->opaque;
 867     int r;
 868
 869     if (sn_info->name[0] == '\0') {
 870         return -EINVAL; /* we need a name for rbd snapshots */
 871     }
 872
 873     /*
 874      * rbd snapshots are using the name as the user controlled unique identifier
 875      * we can't use the rbd snapid for that purpose, as it can't be set
 876      */
 877     if (sn_info->id_str[0] != '\0' &&
 878         strcmp(sn_info->id_str, sn_info->name) != 0) {
 879         return -EINVAL;
 880     }
 881
 882     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 883         return -ERANGE;
 884     }
 885
 886     r = rbd_snap_create(s->image, sn_info->name);
 887     if (r < 0) {
 888         error_report("failed to create snap: %s", strerror(-r));
 889         return r;
 890     }
 891
 892     return 0;
 893 }
 894
 895 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 896                                 const char *snapshot_id,
 897                                 const char *snapshot_name,
 898                                 Error **errp)
 899 {
 900     BDRVRBDState *s = bs->opaque;
 901     int r;
 902
 903     if (!snapshot_name) {
 904         error_setg(errp, "rbd need a valid snapshot name");
 905         return -EINVAL;
 906     }
 907
 908     /* If snapshot_id is specified, it must be equal to name, see
 909        qemu_rbd_snap_list() */
 910     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 911         error_setg(errp,
 912                    "rbd do not support snapshot id, it should be NULL or "
 913                    "equal to snapshot name");
 914         return -EINVAL;
 915     }
 916
 917     r = rbd_snap_remove(s->image, snapshot_name);
 918     if (r < 0) {
 919         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 920     }
 921     return r;
 922 }
 923
 924 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 925                                   const char *snapshot_name)
 926 {
 927     BDRVRBDState *s = bs->opaque;
 928     int r;
 929
 930     r = rbd_snap_rollback(s->image, snapshot_name);
 931     return r;
 932 }
 933
 934 static int qemu_rbd_snap_list(BlockDriverState *bs,
 935                               QEMUSnapshotInfo **psn_tab)
 936 {
 937     BDRVRBDState *s = bs->opaque;
 938     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 939     int i, snap_count;
 940     rbd_snap_info_t *snaps;
 941     int max_snaps = RBD_MAX_SNAPS;
 942
 943     do {
 944         snaps = g_malloc(sizeof(*snaps) * max_snaps);
 945         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 946         if (snap_count <= 0) {
 947             g_free(snaps);
 948         }
 949     } while (snap_count == -ERANGE);
 950
 951     if (snap_count <= 0) {
 952         goto done;
 953     }
 954
 955     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
 956
 957     for (i = 0; i < snap_count; i++) {
 958         const char *snap_name = snaps[i].name;
 959
 960         sn_info = sn_tab + i;
 961         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 962         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 963
 964         sn_info->vm_state_size = snaps[i].size;
 965         sn_info->date_sec = 0;
 966         sn_info->date_nsec = 0;
 967         sn_info->vm_clock_nsec = 0;
 968     }
 969     rbd_snap_list_end(snaps);
 970     g_free(snaps);
 971
 972  done:
 973     *psn_tab = sn_tab;
 974     return snap_count;
 975 }
 976
 977 #ifdef LIBRBD_SUPPORTS_DISCARD
 978 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 979                                               int64_t sector_num,
 980                                               int nb_sectors,
 981                                               BlockDriverCompletionFunc *cb,
 982                                               void *opaque)
 983 {
 984     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 985                          RBD_AIO_DISCARD);
 986 }
 987 #endif
 988
 989 static QEMUOptionParameter qemu_rbd_create_options[] = {
 990     {
 991      .name = BLOCK_OPT_SIZE,
 992      .type = OPT_SIZE,
 993      .help = "Virtual disk size"
 994     },
 995     {
 996      .name = BLOCK_OPT_CLUSTER_SIZE,
 997      .type = OPT_SIZE,
 998      .help = "RBD object size"
 999     },
1000     {NULL}
1001 };
1002
1003 static BlockDriver bdrv_rbd = {
1004     .format_name        = "rbd",
1005     .instance_size      = sizeof(BDRVRBDState),
1006     .bdrv_needs_filename = true,
1007     .bdrv_file_open     = qemu_rbd_open,
1008     .bdrv_close         = qemu_rbd_close,
1009     .bdrv_create        = qemu_rbd_create,
1010     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1011     .bdrv_get_info      = qemu_rbd_getinfo,
1012     .create_options     = qemu_rbd_create_options,
1013     .bdrv_getlength     = qemu_rbd_getlength,
1014     .bdrv_truncate      = qemu_rbd_truncate,
1015     .protocol_name      = "rbd",
1016
1017     .bdrv_aio_readv         = qemu_rbd_aio_readv,
1018     .bdrv_aio_writev        = qemu_rbd_aio_writev,
1019
1020 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1021     .bdrv_aio_flush         = qemu_rbd_aio_flush,
1022 #else
1023     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
1024 #endif
1025
1026 #ifdef LIBRBD_SUPPORTS_DISCARD
1027     .bdrv_aio_discard       = qemu_rbd_aio_discard,
1028 #endif
1029
1030     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1031     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1032     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1033     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1034 };
1035
1036 static void bdrv_rbd_init(void)
1037 {
1038     bdrv_register(&bdrv_rbd);
1039 }
1040
1041 block_init(bdrv_rbd_init);