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     int qemu_aio_count;
 104     char *snap;
 105     int event_reader_pos;
 106     RADOSCB *event_rcb;
 107 } BDRVRBDState;
 108
 109 static void rbd_aio_bh_cb(void *opaque);
 110
 111 static int qemu_rbd_next_tok(char *dst, int dst_len,
 112                              char *src, char delim,
 113                              const char *name,
 114                              char **p)
 115 {
 116     int l;
 117     char *end;
 118
 119     *p = NULL;
 120
 121     if (delim != '\0') {
 122         for (end = src; *end; ++end) {
 123             if (*end == delim) {
 124                 break;
 125             }
 126             if (*end == '\\' && end[1] != '\0') {
 127                 end++;
 128             }
 129         }
 130         if (*end == delim) {
 131             *p = end + 1;
 132             *end = '\0';
 133         }
 134     }
 135     l = strlen(src);
 136     if (l >= dst_len) {
 137         error_report("%s too long", name);
 138         return -EINVAL;
 139     } else if (l == 0) {
 140         error_report("%s too short", name);
 141         return -EINVAL;
 142     }
 143
 144     pstrcpy(dst, dst_len, src);
 145
 146     return 0;
 147 }
 148
 149 static void qemu_rbd_unescape(char *src)
 150 {
 151     char *p;
 152
 153     for (p = src; *src; ++src, ++p) {
 154         if (*src == '\\' && src[1] != '\0') {
 155             src++;
 156         }
 157         *p = *src;
 158     }
 159     *p = '\0';
 160 }
 161
 162 static int qemu_rbd_parsename(const char *filename,
 163                               char *pool, int pool_len,
 164                               char *snap, int snap_len,
 165                               char *name, int name_len,
 166                               char *conf, int conf_len)
 167 {
 168     const char *start;
 169     char *p, *buf;
 170     int ret;
 171
 172     if (!strstart(filename, "rbd:", &start)) {
 173         return -EINVAL;
 174     }
 175
 176     buf = g_strdup(start);
 177     p = buf;
 178     *snap = '\0';
 179     *conf = '\0';
 180
 181     ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
 182     if (ret < 0 || !p) {
 183         ret = -EINVAL;
 184         goto done;
 185     }
 186     qemu_rbd_unescape(pool);
 187
 188     if (strchr(p, '@')) {
 189         ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
 190         if (ret < 0) {
 191             goto done;
 192         }
 193         ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
 194         qemu_rbd_unescape(snap);
 195     } else {
 196         ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
 197     }
 198     qemu_rbd_unescape(name);
 199     if (ret < 0 || !p) {
 200         goto done;
 201     }
 202
 203     ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
 204
 205 done:
 206     g_free(buf);
 207     return ret;
 208 }
 209
 210 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 211 {
 212     const char *p = conf;
 213
 214     while (*p) {
 215         int len;
 216         const char *end = strchr(p, ':');
 217
 218         if (end) {
 219             len = end - p;
 220         } else {
 221             len = strlen(p);
 222         }
 223
 224         if (strncmp(p, "id=", 3) == 0) {
 225             len -= 3;
 226             strncpy(clientname, p + 3, len);
 227             clientname[len] = '\0';
 228             return clientname;
 229         }
 230         if (end == NULL) {
 231             break;
 232         }
 233         p = end + 1;
 234     }
 235     return NULL;
 236 }
 237
 238 static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
 239 {
 240     char *p, *buf;
 241     char name[RBD_MAX_CONF_NAME_SIZE];
 242     char value[RBD_MAX_CONF_VAL_SIZE];
 243     int ret = 0;
 244
 245     buf = g_strdup(conf);
 246     p = buf;
 247
 248     while (p) {
 249         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 250                                 '=', "conf option name", &p);
 251         if (ret < 0) {
 252             break;
 253         }
 254         qemu_rbd_unescape(name);
 255
 256         if (!p) {
 257             error_report("conf option %s has no value", name);
 258             ret = -EINVAL;
 259             break;
 260         }
 261
 262         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 263                                 ':', "conf option value", &p);
 264         if (ret < 0) {
 265             break;
 266         }
 267         qemu_rbd_unescape(value);
 268
 269         if (strcmp(name, "conf") == 0) {
 270             ret = rados_conf_read_file(cluster, value);
 271             if (ret < 0) {
 272                 error_report("error reading conf file %s", value);
 273                 break;
 274             }
 275         } else if (strcmp(name, "id") == 0) {
 276             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 277         } else {
 278             ret = rados_conf_set(cluster, name, value);
 279             if (ret < 0) {
 280                 error_report("invalid conf option %s", name);
 281                 ret = -EINVAL;
 282                 break;
 283             }
 284         }
 285     }
 286
 287     g_free(buf);
 288     return ret;
 289 }
 290
 291 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options)
 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                 s->qemu_aio_count--;
 432             }
 433         }
 434     } while (ret < 0 && errno == EINTR);
 435 }
 436
 437 static int qemu_rbd_aio_flush_cb(void *opaque)
 438 {
 439     BDRVRBDState *s = opaque;
 440
 441     return (s->qemu_aio_count > 0);
 442 }
 443
 444 static int qemu_rbd_open(BlockDriverState *bs, const char *filename,
 445                          QDict *options, int flags)
 446 {
 447     BDRVRBDState *s = bs->opaque;
 448     char pool[RBD_MAX_POOL_NAME_SIZE];
 449     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 450     char conf[RBD_MAX_CONF_SIZE];
 451     char clientname_buf[RBD_MAX_CONF_SIZE];
 452     char *clientname;
 453     int r;
 454
 455     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 456                            snap_buf, sizeof(snap_buf),
 457                            s->name, sizeof(s->name),
 458                            conf, sizeof(conf)) < 0) {
 459         return -EINVAL;
 460     }
 461
 462     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 463     r = rados_create(&s->cluster, clientname);
 464     if (r < 0) {
 465         error_report("error initializing");
 466         return r;
 467     }
 468
 469     s->snap = NULL;
 470     if (snap_buf[0] != '\0') {
 471         s->snap = g_strdup(snap_buf);
 472     }
 473
 474     /*
 475      * Fallback to more conservative semantics if setting cache
 476      * options fails. Ignore errors from setting rbd_cache because the
 477      * only possible error is that the option does not exist, and
 478      * librbd defaults to no caching. If write through caching cannot
 479      * be set up, fall back to no caching.
 480      */
 481     if (flags & BDRV_O_NOCACHE) {
 482         rados_conf_set(s->cluster, "rbd_cache", "false");
 483     } else {
 484         rados_conf_set(s->cluster, "rbd_cache", "true");
 485     }
 486
 487     if (strstr(conf, "conf=") == NULL) {
 488         /* try default location, but ignore failure */
 489         rados_conf_read_file(s->cluster, NULL);
 490     }
 491
 492     if (conf[0] != '\0') {
 493         r = qemu_rbd_set_conf(s->cluster, conf);
 494         if (r < 0) {
 495             error_report("error setting config options");
 496             goto failed_shutdown;
 497         }
 498     }
 499
 500     r = rados_connect(s->cluster);
 501     if (r < 0) {
 502         error_report("error connecting");
 503         goto failed_shutdown;
 504     }
 505
 506     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 507     if (r < 0) {
 508         error_report("error opening pool %s", pool);
 509         goto failed_shutdown;
 510     }
 511
 512     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 513     if (r < 0) {
 514         error_report("error reading header from %s", s->name);
 515         goto failed_open;
 516     }
 517
 518     bs->read_only = (s->snap != NULL);
 519
 520     s->event_reader_pos = 0;
 521     r = qemu_pipe(s->fds);
 522     if (r < 0) {
 523         error_report("error opening eventfd");
 524         goto failed;
 525     }
 526     fcntl(s->fds[0], F_SETFL, O_NONBLOCK);
 527     fcntl(s->fds[1], F_SETFL, O_NONBLOCK);
 528     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader,
 529                             NULL, qemu_rbd_aio_flush_cb, s);
 530
 531
 532     return 0;
 533
 534 failed:
 535     rbd_close(s->image);
 536 failed_open:
 537     rados_ioctx_destroy(s->io_ctx);
 538 failed_shutdown:
 539     rados_shutdown(s->cluster);
 540     g_free(s->snap);
 541     return r;
 542 }
 543
 544 static void qemu_rbd_close(BlockDriverState *bs)
 545 {
 546     BDRVRBDState *s = bs->opaque;
 547
 548     close(s->fds[0]);
 549     close(s->fds[1]);
 550     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL, NULL);
 551
 552     rbd_close(s->image);
 553     rados_ioctx_destroy(s->io_ctx);
 554     g_free(s->snap);
 555     rados_shutdown(s->cluster);
 556 }
 557
 558 /*
 559  * Cancel aio. Since we don't reference acb in a non qemu threads,
 560  * it is safe to access it here.
 561  */
 562 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
 563 {
 564     RBDAIOCB *acb = (RBDAIOCB *) blockacb;
 565     acb->cancelled = 1;
 566
 567     while (acb->status == -EINPROGRESS) {
 568         qemu_aio_wait();
 569     }
 570
 571     qemu_aio_release(acb);
 572 }
 573
 574 static const AIOCBInfo rbd_aiocb_info = {
 575     .aiocb_size = sizeof(RBDAIOCB),
 576     .cancel = qemu_rbd_aio_cancel,
 577 };
 578
 579 static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb)
 580 {
 581     int ret = 0;
 582     while (1) {
 583         fd_set wfd;
 584         int fd = s->fds[RBD_FD_WRITE];
 585
 586         /* send the op pointer to the qemu thread that is responsible
 587            for the aio/op completion. Must do it in a qemu thread context */
 588         ret = write(fd, (void *)&rcb, sizeof(rcb));
 589         if (ret >= 0) {
 590             break;
 591         }
 592         if (errno == EINTR) {
 593             continue;
 594         }
 595         if (errno != EAGAIN) {
 596             break;
 597         }
 598
 599         FD_ZERO(&wfd);
 600         FD_SET(fd, &wfd);
 601         do {
 602             ret = select(fd + 1, NULL, &wfd, NULL, NULL);
 603         } while (ret < 0 && errno == EINTR);
 604     }
 605
 606     return ret;
 607 }
 608
 609 /*
 610  * This is the callback function for rbd_aio_read and _write
 611  *
 612  * Note: this function is being called from a non qemu thread so
 613  * we need to be careful about what we do here. Generally we only
 614  * write to the block notification pipe, and do the rest of the
 615  * io completion handling from qemu_rbd_aio_event_reader() which
 616  * runs in a qemu context.
 617  */
 618 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 619 {
 620     int ret;
 621     rcb->ret = rbd_aio_get_return_value(c);
 622     rbd_aio_release(c);
 623     ret = qemu_rbd_send_pipe(rcb->s, rcb);
 624     if (ret < 0) {
 625         error_report("failed writing to acb->s->fds");
 626         g_free(rcb);
 627     }
 628 }
 629
 630 /* Callback when all queued rbd_aio requests are complete */
 631
 632 static void rbd_aio_bh_cb(void *opaque)
 633 {
 634     RBDAIOCB *acb = opaque;
 635
 636     if (acb->cmd == RBD_AIO_READ) {
 637         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 638     }
 639     qemu_vfree(acb->bounce);
 640     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 641     qemu_bh_delete(acb->bh);
 642     acb->bh = NULL;
 643     acb->status = 0;
 644
 645     if (!acb->cancelled) {
 646         qemu_aio_release(acb);
 647     }
 648 }
 649
 650 static int rbd_aio_discard_wrapper(rbd_image_t image,
 651                                    uint64_t off,
 652                                    uint64_t len,
 653                                    rbd_completion_t comp)
 654 {
 655 #ifdef LIBRBD_SUPPORTS_DISCARD
 656     return rbd_aio_discard(image, off, len, comp);
 657 #else
 658     return -ENOTSUP;
 659 #endif
 660 }
 661
 662 static int rbd_aio_flush_wrapper(rbd_image_t image,
 663                                  rbd_completion_t comp)
 664 {
 665 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 666     return rbd_aio_flush(image, comp);
 667 #else
 668     return -ENOTSUP;
 669 #endif
 670 }
 671
 672 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
 673                                        int64_t sector_num,
 674                                        QEMUIOVector *qiov,
 675                                        int nb_sectors,
 676                                        BlockDriverCompletionFunc *cb,
 677                                        void *opaque,
 678                                        RBDAIOCmd cmd)
 679 {
 680     RBDAIOCB *acb;
 681     RADOSCB *rcb;
 682     rbd_completion_t c;
 683     int64_t off, size;
 684     char *buf;
 685     int r;
 686
 687     BDRVRBDState *s = bs->opaque;
 688
 689     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 690     acb->cmd = cmd;
 691     acb->qiov = qiov;
 692     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 693         acb->bounce = NULL;
 694     } else {
 695         acb->bounce = qemu_blockalign(bs, qiov->size);
 696     }
 697     acb->ret = 0;
 698     acb->error = 0;
 699     acb->s = s;
 700     acb->cancelled = 0;
 701     acb->bh = NULL;
 702     acb->status = -EINPROGRESS;
 703
 704     if (cmd == RBD_AIO_WRITE) {
 705         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 706     }
 707
 708     buf = acb->bounce;
 709
 710     off = sector_num * BDRV_SECTOR_SIZE;
 711     size = nb_sectors * BDRV_SECTOR_SIZE;
 712
 713     s->qemu_aio_count++; /* All the RADOSCB */
 714
 715     rcb = g_malloc(sizeof(RADOSCB));
 716     rcb->done = 0;
 717     rcb->acb = acb;
 718     rcb->buf = buf;
 719     rcb->s = acb->s;
 720     rcb->size = size;
 721     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 722     if (r < 0) {
 723         goto failed;
 724     }
 725
 726     switch (cmd) {
 727     case RBD_AIO_WRITE:
 728         r = rbd_aio_write(s->image, off, size, buf, c);
 729         break;
 730     case RBD_AIO_READ:
 731         r = rbd_aio_read(s->image, off, size, buf, c);
 732         break;
 733     case RBD_AIO_DISCARD:
 734         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 735         break;
 736     case RBD_AIO_FLUSH:
 737         r = rbd_aio_flush_wrapper(s->image, c);
 738         break;
 739     default:
 740         r = -EINVAL;
 741     }
 742
 743     if (r < 0) {
 744         goto failed;
 745     }
 746
 747     return &acb->common;
 748
 749 failed:
 750     g_free(rcb);
 751     s->qemu_aio_count--;
 752     qemu_aio_release(acb);
 753     return NULL;
 754 }
 755
 756 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 757                                             int64_t sector_num,
 758                                             QEMUIOVector *qiov,
 759                                             int nb_sectors,
 760                                             BlockDriverCompletionFunc *cb,
 761                                             void *opaque)
 762 {
 763     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 764                          RBD_AIO_READ);
 765 }
 766
 767 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 768                                              int64_t sector_num,
 769                                              QEMUIOVector *qiov,
 770                                              int nb_sectors,
 771                                              BlockDriverCompletionFunc *cb,
 772                                              void *opaque)
 773 {
 774     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 775                          RBD_AIO_WRITE);
 776 }
 777
 778 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 779 static BlockDriverAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 780                                             BlockDriverCompletionFunc *cb,
 781                                             void *opaque)
 782 {
 783     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 784 }
 785
 786 #else
 787
 788 static int qemu_rbd_co_flush(BlockDriverState *bs)
 789 {
 790 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 791     /* rbd_flush added in 0.1.1 */
 792     BDRVRBDState *s = bs->opaque;
 793     return rbd_flush(s->image);
 794 #else
 795     return 0;
 796 #endif
 797 }
 798 #endif
 799
 800 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 801 {
 802     BDRVRBDState *s = bs->opaque;
 803     rbd_image_info_t info;
 804     int r;
 805
 806     r = rbd_stat(s->image, &info, sizeof(info));
 807     if (r < 0) {
 808         return r;
 809     }
 810
 811     bdi->cluster_size = info.obj_size;
 812     return 0;
 813 }
 814
 815 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 816 {
 817     BDRVRBDState *s = bs->opaque;
 818     rbd_image_info_t info;
 819     int r;
 820
 821     r = rbd_stat(s->image, &info, sizeof(info));
 822     if (r < 0) {
 823         return r;
 824     }
 825
 826     return info.size;
 827 }
 828
 829 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 830 {
 831     BDRVRBDState *s = bs->opaque;
 832     int r;
 833
 834     r = rbd_resize(s->image, offset);
 835     if (r < 0) {
 836         return r;
 837     }
 838
 839     return 0;
 840 }
 841
 842 static int qemu_rbd_snap_create(BlockDriverState *bs,
 843                                 QEMUSnapshotInfo *sn_info)
 844 {
 845     BDRVRBDState *s = bs->opaque;
 846     int r;
 847
 848     if (sn_info->name[0] == '\0') {
 849         return -EINVAL; /* we need a name for rbd snapshots */
 850     }
 851
 852     /*
 853      * rbd snapshots are using the name as the user controlled unique identifier
 854      * we can't use the rbd snapid for that purpose, as it can't be set
 855      */
 856     if (sn_info->id_str[0] != '\0' &&
 857         strcmp(sn_info->id_str, sn_info->name) != 0) {
 858         return -EINVAL;
 859     }
 860
 861     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 862         return -ERANGE;
 863     }
 864
 865     r = rbd_snap_create(s->image, sn_info->name);
 866     if (r < 0) {
 867         error_report("failed to create snap: %s", strerror(-r));
 868         return r;
 869     }
 870
 871     return 0;
 872 }
 873
 874 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 875                                 const char *snapshot_name)
 876 {
 877     BDRVRBDState *s = bs->opaque;
 878     int r;
 879
 880     r = rbd_snap_remove(s->image, snapshot_name);
 881     return r;
 882 }
 883
 884 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 885                                   const char *snapshot_name)
 886 {
 887     BDRVRBDState *s = bs->opaque;
 888     int r;
 889
 890     r = rbd_snap_rollback(s->image, snapshot_name);
 891     return r;
 892 }
 893
 894 static int qemu_rbd_snap_list(BlockDriverState *bs,
 895                               QEMUSnapshotInfo **psn_tab)
 896 {
 897     BDRVRBDState *s = bs->opaque;
 898     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 899     int i, snap_count;
 900     rbd_snap_info_t *snaps;
 901     int max_snaps = RBD_MAX_SNAPS;
 902
 903     do {
 904         snaps = g_malloc(sizeof(*snaps) * max_snaps);
 905         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 906         if (snap_count < 0) {
 907             g_free(snaps);
 908         }
 909     } while (snap_count == -ERANGE);
 910
 911     if (snap_count <= 0) {
 912         goto done;
 913     }
 914
 915     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
 916
 917     for (i = 0; i < snap_count; i++) {
 918         const char *snap_name = snaps[i].name;
 919
 920         sn_info = sn_tab + i;
 921         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 922         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 923
 924         sn_info->vm_state_size = snaps[i].size;
 925         sn_info->date_sec = 0;
 926         sn_info->date_nsec = 0;
 927         sn_info->vm_clock_nsec = 0;
 928     }
 929     rbd_snap_list_end(snaps);
 930
 931  done:
 932     *psn_tab = sn_tab;
 933     return snap_count;
 934 }
 935
 936 #ifdef LIBRBD_SUPPORTS_DISCARD
 937 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 938                                               int64_t sector_num,
 939                                               int nb_sectors,
 940                                               BlockDriverCompletionFunc *cb,
 941                                               void *opaque)
 942 {
 943     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 944                          RBD_AIO_DISCARD);
 945 }
 946 #endif
 947
 948 static QEMUOptionParameter qemu_rbd_create_options[] = {
 949     {
 950      .name = BLOCK_OPT_SIZE,
 951      .type = OPT_SIZE,
 952      .help = "Virtual disk size"
 953     },
 954     {
 955      .name = BLOCK_OPT_CLUSTER_SIZE,
 956      .type = OPT_SIZE,
 957      .help = "RBD object size"
 958     },
 959     {NULL}
 960 };
 961
 962 static BlockDriver bdrv_rbd = {
 963     .format_name        = "rbd",
 964     .instance_size      = sizeof(BDRVRBDState),
 965     .bdrv_file_open     = qemu_rbd_open,
 966     .bdrv_close         = qemu_rbd_close,
 967     .bdrv_create        = qemu_rbd_create,
 968     .bdrv_get_info      = qemu_rbd_getinfo,
 969     .create_options     = qemu_rbd_create_options,
 970     .bdrv_getlength     = qemu_rbd_getlength,
 971     .bdrv_truncate      = qemu_rbd_truncate,
 972     .protocol_name      = "rbd",
 973
 974     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 975     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 976
 977 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 978     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 979 #else
 980     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 981 #endif
 982
 983 #ifdef LIBRBD_SUPPORTS_DISCARD
 984     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 985 #endif
 986
 987     .bdrv_snapshot_create   = qemu_rbd_snap_create,
 988     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
 989     .bdrv_snapshot_list     = qemu_rbd_snap_list,
 990     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
 991 };
 992
 993 static void bdrv_rbd_init(void)
 994 {
 995     bdrv_register(&bdrv_rbd);
 996 }
 997
 998 block_init(bdrv_rbd_init);