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 } RBDAIOCmd;
  68
  69 typedef struct RBDAIOCB {
  70     BlockDriverAIOCB common;
  71     QEMUBH *bh;
  72     int64_t ret;
  73     QEMUIOVector *qiov;
  74     char *bounce;
  75     RBDAIOCmd cmd;
  76     int64_t sector_num;
  77     int error;
  78     struct BDRVRBDState *s;
  79     int cancelled;
  80     int status;
  81 } RBDAIOCB;
  82
  83 typedef struct RADOSCB {
  84     int rcbid;
  85     RBDAIOCB *acb;
  86     struct BDRVRBDState *s;
  87     int done;
  88     int64_t size;
  89     char *buf;
  90     int64_t ret;
  91 } RADOSCB;
  92
  93 #define RBD_FD_READ 0
  94 #define RBD_FD_WRITE 1
  95
  96 typedef struct BDRVRBDState {
  97     int fds[2];
  98     rados_t cluster;
  99     rados_ioctx_t io_ctx;
 100     rbd_image_t image;
 101     char name[RBD_MAX_IMAGE_NAME_SIZE];
 102     int qemu_aio_count;
 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 {
 292     int64_t bytes = 0;
 293     int64_t objsize;
 294     int obj_order = 0;
 295     char pool[RBD_MAX_POOL_NAME_SIZE];
 296     char name[RBD_MAX_IMAGE_NAME_SIZE];
 297     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 298     char conf[RBD_MAX_CONF_SIZE];
 299     char clientname_buf[RBD_MAX_CONF_SIZE];
 300     char *clientname;
 301     rados_t cluster;
 302     rados_ioctx_t io_ctx;
 303     int ret;
 304
 305     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 306                            snap_buf, sizeof(snap_buf),
 307                            name, sizeof(name),
 308                            conf, sizeof(conf)) < 0) {
 309         return -EINVAL;
 310     }
 311
 312     /* Read out options */
 313     while (options && options->name) {
 314         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 315             bytes = options->value.n;
 316         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
 317             if (options->value.n) {
 318                 objsize = options->value.n;
 319                 if ((objsize - 1) & objsize) {    /* not a power of 2? */
 320                     error_report("obj size needs to be power of 2");
 321                     return -EINVAL;
 322                 }
 323                 if (objsize < 4096) {
 324                     error_report("obj size too small");
 325                     return -EINVAL;
 326                 }
 327                 obj_order = ffs(objsize) - 1;
 328             }
 329         }
 330         options++;
 331     }
 332
 333     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 334     if (rados_create(&cluster, clientname) < 0) {
 335         error_report("error initializing");
 336         return -EIO;
 337     }
 338
 339     if (strstr(conf, "conf=") == NULL) {
 340         /* try default location, but ignore failure */
 341         rados_conf_read_file(cluster, NULL);
 342     }
 343
 344     if (conf[0] != '\0' &&
 345         qemu_rbd_set_conf(cluster, conf) < 0) {
 346         error_report("error setting config options");
 347         rados_shutdown(cluster);
 348         return -EIO;
 349     }
 350
 351     if (rados_connect(cluster) < 0) {
 352         error_report("error connecting");
 353         rados_shutdown(cluster);
 354         return -EIO;
 355     }
 356
 357     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 358         error_report("error opening pool %s", pool);
 359         rados_shutdown(cluster);
 360         return -EIO;
 361     }
 362
 363     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 364     rados_ioctx_destroy(io_ctx);
 365     rados_shutdown(cluster);
 366
 367     return ret;
 368 }
 369
 370 /*
 371  * This aio completion is being called from qemu_rbd_aio_event_reader()
 372  * and runs in qemu context. It schedules a bh, but just in case the aio
 373  * was not cancelled before.
 374  */
 375 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 376 {
 377     RBDAIOCB *acb = rcb->acb;
 378     int64_t r;
 379
 380     r = rcb->ret;
 381
 382     if (acb->cmd == RBD_AIO_WRITE ||
 383         acb->cmd == RBD_AIO_DISCARD) {
 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 BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
 663                                        int64_t sector_num,
 664                                        QEMUIOVector *qiov,
 665                                        int nb_sectors,
 666                                        BlockDriverCompletionFunc *cb,
 667                                        void *opaque,
 668                                        RBDAIOCmd cmd)
 669 {
 670     RBDAIOCB *acb;
 671     RADOSCB *rcb;
 672     rbd_completion_t c;
 673     int64_t off, size;
 674     char *buf;
 675     int r;
 676
 677     BDRVRBDState *s = bs->opaque;
 678
 679     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 680     acb->cmd = cmd;
 681     acb->qiov = qiov;
 682     if (cmd == RBD_AIO_DISCARD) {
 683         acb->bounce = NULL;
 684     } else {
 685         acb->bounce = qemu_blockalign(bs, qiov->size);
 686     }
 687     acb->ret = 0;
 688     acb->error = 0;
 689     acb->s = s;
 690     acb->cancelled = 0;
 691     acb->bh = NULL;
 692     acb->status = -EINPROGRESS;
 693
 694     if (cmd == RBD_AIO_WRITE) {
 695         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 696     }
 697
 698     buf = acb->bounce;
 699
 700     off = sector_num * BDRV_SECTOR_SIZE;
 701     size = nb_sectors * BDRV_SECTOR_SIZE;
 702
 703     s->qemu_aio_count++; /* All the RADOSCB */
 704
 705     rcb = g_malloc(sizeof(RADOSCB));
 706     rcb->done = 0;
 707     rcb->acb = acb;
 708     rcb->buf = buf;
 709     rcb->s = acb->s;
 710     rcb->size = size;
 711     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 712     if (r < 0) {
 713         goto failed;
 714     }
 715
 716     switch (cmd) {
 717     case RBD_AIO_WRITE:
 718         r = rbd_aio_write(s->image, off, size, buf, c);
 719         break;
 720     case RBD_AIO_READ:
 721         r = rbd_aio_read(s->image, off, size, buf, c);
 722         break;
 723     case RBD_AIO_DISCARD:
 724         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 725         break;
 726     default:
 727         r = -EINVAL;
 728     }
 729
 730     if (r < 0) {
 731         goto failed;
 732     }
 733
 734     return &acb->common;
 735
 736 failed:
 737     g_free(rcb);
 738     s->qemu_aio_count--;
 739     qemu_aio_release(acb);
 740     return NULL;
 741 }
 742
 743 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 744                                             int64_t sector_num,
 745                                             QEMUIOVector *qiov,
 746                                             int nb_sectors,
 747                                             BlockDriverCompletionFunc *cb,
 748                                             void *opaque)
 749 {
 750     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 751                          RBD_AIO_READ);
 752 }
 753
 754 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 755                                              int64_t sector_num,
 756                                              QEMUIOVector *qiov,
 757                                              int nb_sectors,
 758                                              BlockDriverCompletionFunc *cb,
 759                                              void *opaque)
 760 {
 761     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 762                          RBD_AIO_WRITE);
 763 }
 764
 765 static int qemu_rbd_co_flush(BlockDriverState *bs)
 766 {
 767 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 768     /* rbd_flush added in 0.1.1 */
 769     BDRVRBDState *s = bs->opaque;
 770     return rbd_flush(s->image);
 771 #else
 772     return 0;
 773 #endif
 774 }
 775
 776 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 777 {
 778     BDRVRBDState *s = bs->opaque;
 779     rbd_image_info_t info;
 780     int r;
 781
 782     r = rbd_stat(s->image, &info, sizeof(info));
 783     if (r < 0) {
 784         return r;
 785     }
 786
 787     bdi->cluster_size = info.obj_size;
 788     return 0;
 789 }
 790
 791 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 792 {
 793     BDRVRBDState *s = bs->opaque;
 794     rbd_image_info_t info;
 795     int r;
 796
 797     r = rbd_stat(s->image, &info, sizeof(info));
 798     if (r < 0) {
 799         return r;
 800     }
 801
 802     return info.size;
 803 }
 804
 805 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 806 {
 807     BDRVRBDState *s = bs->opaque;
 808     int r;
 809
 810     r = rbd_resize(s->image, offset);
 811     if (r < 0) {
 812         return r;
 813     }
 814
 815     return 0;
 816 }
 817
 818 static int qemu_rbd_snap_create(BlockDriverState *bs,
 819                                 QEMUSnapshotInfo *sn_info)
 820 {
 821     BDRVRBDState *s = bs->opaque;
 822     int r;
 823
 824     if (sn_info->name[0] == '\0') {
 825         return -EINVAL; /* we need a name for rbd snapshots */
 826     }
 827
 828     /*
 829      * rbd snapshots are using the name as the user controlled unique identifier
 830      * we can't use the rbd snapid for that purpose, as it can't be set
 831      */
 832     if (sn_info->id_str[0] != '\0' &&
 833         strcmp(sn_info->id_str, sn_info->name) != 0) {
 834         return -EINVAL;
 835     }
 836
 837     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 838         return -ERANGE;
 839     }
 840
 841     r = rbd_snap_create(s->image, sn_info->name);
 842     if (r < 0) {
 843         error_report("failed to create snap: %s", strerror(-r));
 844         return r;
 845     }
 846
 847     return 0;
 848 }
 849
 850 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 851                                 const char *snapshot_name)
 852 {
 853     BDRVRBDState *s = bs->opaque;
 854     int r;
 855
 856     r = rbd_snap_remove(s->image, snapshot_name);
 857     return r;
 858 }
 859
 860 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 861                                   const char *snapshot_name)
 862 {
 863     BDRVRBDState *s = bs->opaque;
 864     int r;
 865
 866     r = rbd_snap_rollback(s->image, snapshot_name);
 867     return r;
 868 }
 869
 870 static int qemu_rbd_snap_list(BlockDriverState *bs,
 871                               QEMUSnapshotInfo **psn_tab)
 872 {
 873     BDRVRBDState *s = bs->opaque;
 874     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 875     int i, snap_count;
 876     rbd_snap_info_t *snaps;
 877     int max_snaps = RBD_MAX_SNAPS;
 878
 879     do {
 880         snaps = g_malloc(sizeof(*snaps) * max_snaps);
 881         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 882         if (snap_count < 0) {
 883             g_free(snaps);
 884         }
 885     } while (snap_count == -ERANGE);
 886
 887     if (snap_count <= 0) {
 888         goto done;
 889     }
 890
 891     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
 892
 893     for (i = 0; i < snap_count; i++) {
 894         const char *snap_name = snaps[i].name;
 895
 896         sn_info = sn_tab + i;
 897         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 898         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 899
 900         sn_info->vm_state_size = snaps[i].size;
 901         sn_info->date_sec = 0;
 902         sn_info->date_nsec = 0;
 903         sn_info->vm_clock_nsec = 0;
 904     }
 905     rbd_snap_list_end(snaps);
 906
 907  done:
 908     *psn_tab = sn_tab;
 909     return snap_count;
 910 }
 911
 912 #ifdef LIBRBD_SUPPORTS_DISCARD
 913 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 914                                               int64_t sector_num,
 915                                               int nb_sectors,
 916                                               BlockDriverCompletionFunc *cb,
 917                                               void *opaque)
 918 {
 919     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 920                          RBD_AIO_DISCARD);
 921 }
 922 #endif
 923
 924 static QEMUOptionParameter qemu_rbd_create_options[] = {
 925     {
 926      .name = BLOCK_OPT_SIZE,
 927      .type = OPT_SIZE,
 928      .help = "Virtual disk size"
 929     },
 930     {
 931      .name = BLOCK_OPT_CLUSTER_SIZE,
 932      .type = OPT_SIZE,
 933      .help = "RBD object size"
 934     },
 935     {NULL}
 936 };
 937
 938 static BlockDriver bdrv_rbd = {
 939     .format_name        = "rbd",
 940     .instance_size      = sizeof(BDRVRBDState),
 941     .bdrv_file_open     = qemu_rbd_open,
 942     .bdrv_close         = qemu_rbd_close,
 943     .bdrv_create        = qemu_rbd_create,
 944     .bdrv_get_info      = qemu_rbd_getinfo,
 945     .create_options     = qemu_rbd_create_options,
 946     .bdrv_getlength     = qemu_rbd_getlength,
 947     .bdrv_truncate      = qemu_rbd_truncate,
 948     .protocol_name      = "rbd",
 949
 950     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 951     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 952     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 953
 954 #ifdef LIBRBD_SUPPORTS_DISCARD
 955     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 956 #endif
 957
 958     .bdrv_snapshot_create   = qemu_rbd_snap_create,
 959     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
 960     .bdrv_snapshot_list     = qemu_rbd_snap_list,
 961     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
 962 };
 963
 964 static void bdrv_rbd_init(void)
 965 {
 966     bdrv_register(&bdrv_rbd);
 967 }
 968
 969 block_init(bdrv_rbd_init);