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     rados_t cluster;
  99     rados_ioctx_t io_ctx;
 100     rbd_image_t image;
 101     char name[RBD_MAX_IMAGE_NAME_SIZE];
 102     char *snap;
 103 } BDRVRBDState;
 104
 105 static int qemu_rbd_next_tok(char *dst, int dst_len,
 106                              char *src, char delim,
 107                              const char *name,
 108                              char **p, Error **errp)
 109 {
 110     int l;
 111     char *end;
 112
 113     *p = NULL;
 114
 115     if (delim != '\0') {
 116         for (end = src; *end; ++end) {
 117             if (*end == delim) {
 118                 break;
 119             }
 120             if (*end == '\\' && end[1] != '\0') {
 121                 end++;
 122             }
 123         }
 124         if (*end == delim) {
 125             *p = end + 1;
 126             *end = '\0';
 127         }
 128     }
 129     l = strlen(src);
 130     if (l >= dst_len) {
 131         error_setg(errp, "%s too long", name);
 132         return -EINVAL;
 133     } else if (l == 0) {
 134         error_setg(errp, "%s too short", name);
 135         return -EINVAL;
 136     }
 137
 138     pstrcpy(dst, dst_len, src);
 139
 140     return 0;
 141 }
 142
 143 static void qemu_rbd_unescape(char *src)
 144 {
 145     char *p;
 146
 147     for (p = src; *src; ++src, ++p) {
 148         if (*src == '\\' && src[1] != '\0') {
 149             src++;
 150         }
 151         *p = *src;
 152     }
 153     *p = '\0';
 154 }
 155
 156 static int qemu_rbd_parsename(const char *filename,
 157                               char *pool, int pool_len,
 158                               char *snap, int snap_len,
 159                               char *name, int name_len,
 160                               char *conf, int conf_len,
 161                               Error **errp)
 162 {
 163     const char *start;
 164     char *p, *buf;
 165     int ret;
 166
 167     if (!strstart(filename, "rbd:", &start)) {
 168         error_setg(errp, "File name must start with 'rbd:'");
 169         return -EINVAL;
 170     }
 171
 172     buf = g_strdup(start);
 173     p = buf;
 174     *snap = '\0';
 175     *conf = '\0';
 176
 177     ret = qemu_rbd_next_tok(pool, pool_len, p,
 178                             '/', "pool name", &p, errp);
 179     if (ret < 0 || !p) {
 180         ret = -EINVAL;
 181         goto done;
 182     }
 183     qemu_rbd_unescape(pool);
 184
 185     if (strchr(p, '@')) {
 186         ret = qemu_rbd_next_tok(name, name_len, p,
 187                                 '@', "object name", &p, errp);
 188         if (ret < 0) {
 189             goto done;
 190         }
 191         ret = qemu_rbd_next_tok(snap, snap_len, p,
 192                                 ':', "snap name", &p, errp);
 193         qemu_rbd_unescape(snap);
 194     } else {
 195         ret = qemu_rbd_next_tok(name, name_len, p,
 196                                 ':', "object name", &p, errp);
 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,
 204                             '\0', "configuration", &p, errp);
 205
 206 done:
 207     g_free(buf);
 208     return ret;
 209 }
 210
 211 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 212 {
 213     const char *p = conf;
 214
 215     while (*p) {
 216         int len;
 217         const char *end = strchr(p, ':');
 218
 219         if (end) {
 220             len = end - p;
 221         } else {
 222             len = strlen(p);
 223         }
 224
 225         if (strncmp(p, "id=", 3) == 0) {
 226             len -= 3;
 227             strncpy(clientname, p + 3, len);
 228             clientname[len] = '\0';
 229             return clientname;
 230         }
 231         if (end == NULL) {
 232             break;
 233         }
 234         p = end + 1;
 235     }
 236     return NULL;
 237 }
 238
 239 static int qemu_rbd_set_conf(rados_t cluster, const char *conf, Error **errp)
 240 {
 241     char *p, *buf;
 242     char name[RBD_MAX_CONF_NAME_SIZE];
 243     char value[RBD_MAX_CONF_VAL_SIZE];
 244     int ret = 0;
 245
 246     buf = g_strdup(conf);
 247     p = buf;
 248
 249     while (p) {
 250         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 251                                 '=', "conf option name", &p, errp);
 252         if (ret < 0) {
 253             break;
 254         }
 255         qemu_rbd_unescape(name);
 256
 257         if (!p) {
 258             error_setg(errp, "conf option %s has no value", name);
 259             ret = -EINVAL;
 260             break;
 261         }
 262
 263         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 264                                 ':', "conf option value", &p, errp);
 265         if (ret < 0) {
 266             break;
 267         }
 268         qemu_rbd_unescape(value);
 269
 270         if (strcmp(name, "conf") == 0) {
 271             ret = rados_conf_read_file(cluster, value);
 272             if (ret < 0) {
 273                 error_setg(errp, "error reading conf file %s", value);
 274                 break;
 275             }
 276         } else if (strcmp(name, "id") == 0) {
 277             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 278         } else {
 279             ret = rados_conf_set(cluster, name, value);
 280             if (ret < 0) {
 281                 error_setg(errp, "invalid conf option %s", name);
 282                 ret = -EINVAL;
 283                 break;
 284             }
 285         }
 286     }
 287
 288     g_free(buf);
 289     return ret;
 290 }
 291
 292 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 293 {
 294     Error *local_err = NULL;
 295     int64_t bytes = 0;
 296     int64_t objsize;
 297     int obj_order = 0;
 298     char pool[RBD_MAX_POOL_NAME_SIZE];
 299     char name[RBD_MAX_IMAGE_NAME_SIZE];
 300     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 301     char conf[RBD_MAX_CONF_SIZE];
 302     char clientname_buf[RBD_MAX_CONF_SIZE];
 303     char *clientname;
 304     rados_t cluster;
 305     rados_ioctx_t io_ctx;
 306     int ret;
 307
 308     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 309                            snap_buf, sizeof(snap_buf),
 310                            name, sizeof(name),
 311                            conf, sizeof(conf), &local_err) < 0) {
 312         error_propagate(errp, local_err);
 313         return -EINVAL;
 314     }
 315
 316     /* Read out options */
 317     bytes = qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0);
 318     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 319     if (objsize) {
 320         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 321             error_setg(errp, "obj size needs to be power of 2");
 322             return -EINVAL;
 323         }
 324         if (objsize < 4096) {
 325             error_setg(errp, "obj size too small");
 326             return -EINVAL;
 327         }
 328         obj_order = ffs(objsize) - 1;
 329     }
 330
 331     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 332     if (rados_create(&cluster, clientname) < 0) {
 333         error_setg(errp, "error initializing");
 334         return -EIO;
 335     }
 336
 337     if (strstr(conf, "conf=") == NULL) {
 338         /* try default location, but ignore failure */
 339         rados_conf_read_file(cluster, NULL);
 340     }
 341
 342     if (conf[0] != '\0' &&
 343         qemu_rbd_set_conf(cluster, conf, &local_err) < 0) {
 344         rados_shutdown(cluster);
 345         error_propagate(errp, local_err);
 346         return -EIO;
 347     }
 348
 349     if (rados_connect(cluster) < 0) {
 350         error_setg(errp, "error connecting");
 351         rados_shutdown(cluster);
 352         return -EIO;
 353     }
 354
 355     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 356         error_setg(errp, "error opening pool %s", pool);
 357         rados_shutdown(cluster);
 358         return -EIO;
 359     }
 360
 361     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 362     rados_ioctx_destroy(io_ctx);
 363     rados_shutdown(cluster);
 364
 365     return ret;
 366 }
 367
 368 /*
 369  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 370  * BH context.
 371  */
 372 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 373 {
 374     RBDAIOCB *acb = rcb->acb;
 375     int64_t r;
 376
 377     r = rcb->ret;
 378
 379     if (acb->cmd != RBD_AIO_READ) {
 380         if (r < 0) {
 381             acb->ret = r;
 382             acb->error = 1;
 383         } else if (!acb->error) {
 384             acb->ret = rcb->size;
 385         }
 386     } else {
 387         if (r < 0) {
 388             memset(rcb->buf, 0, rcb->size);
 389             acb->ret = r;
 390             acb->error = 1;
 391         } else if (r < rcb->size) {
 392             memset(rcb->buf + r, 0, rcb->size - r);
 393             if (!acb->error) {
 394                 acb->ret = rcb->size;
 395             }
 396         } else if (!acb->error) {
 397             acb->ret = r;
 398         }
 399     }
 400
 401     g_free(rcb);
 402
 403     if (acb->cmd == RBD_AIO_READ) {
 404         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 405     }
 406     qemu_vfree(acb->bounce);
 407     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 408     acb->status = 0;
 409
 410     if (!acb->cancelled) {
 411         qemu_aio_release(acb);
 412     }
 413 }
 414
 415 /* TODO Convert to fine grained options */
 416 static QemuOptsList runtime_opts = {
 417     .name = "rbd",
 418     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 419     .desc = {
 420         {
 421             .name = "filename",
 422             .type = QEMU_OPT_STRING,
 423             .help = "Specification of the rbd image",
 424         },
 425         { /* end of list */ }
 426     },
 427 };
 428
 429 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 430                          Error **errp)
 431 {
 432     BDRVRBDState *s = bs->opaque;
 433     char pool[RBD_MAX_POOL_NAME_SIZE];
 434     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 435     char conf[RBD_MAX_CONF_SIZE];
 436     char clientname_buf[RBD_MAX_CONF_SIZE];
 437     char *clientname;
 438     QemuOpts *opts;
 439     Error *local_err = NULL;
 440     const char *filename;
 441     int r;
 442
 443     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 444     qemu_opts_absorb_qdict(opts, options, &local_err);
 445     if (local_err) {
 446         error_propagate(errp, local_err);
 447         qemu_opts_del(opts);
 448         return -EINVAL;
 449     }
 450
 451     filename = qemu_opt_get(opts, "filename");
 452
 453     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 454                            snap_buf, sizeof(snap_buf),
 455                            s->name, sizeof(s->name),
 456                            conf, sizeof(conf), errp) < 0) {
 457         r = -EINVAL;
 458         goto failed_opts;
 459     }
 460
 461     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 462     r = rados_create(&s->cluster, clientname);
 463     if (r < 0) {
 464         error_setg(&local_err, "error initializing");
 465         goto failed_opts;
 466     }
 467
 468     s->snap = NULL;
 469     if (snap_buf[0] != '\0') {
 470         s->snap = g_strdup(snap_buf);
 471     }
 472
 473     /*
 474      * Fallback to more conservative semantics if setting cache
 475      * options fails. Ignore errors from setting rbd_cache because the
 476      * only possible error is that the option does not exist, and
 477      * librbd defaults to no caching. If write through caching cannot
 478      * be set up, fall back to no caching.
 479      */
 480     if (flags & BDRV_O_NOCACHE) {
 481         rados_conf_set(s->cluster, "rbd_cache", "false");
 482     } else {
 483         rados_conf_set(s->cluster, "rbd_cache", "true");
 484     }
 485
 486     if (strstr(conf, "conf=") == NULL) {
 487         /* try default location, but ignore failure */
 488         rados_conf_read_file(s->cluster, NULL);
 489     }
 490
 491     if (conf[0] != '\0') {
 492         r = qemu_rbd_set_conf(s->cluster, conf, errp);
 493         if (r < 0) {
 494             goto failed_shutdown;
 495         }
 496     }
 497
 498     r = rados_connect(s->cluster);
 499     if (r < 0) {
 500         error_setg(&local_err, "error connecting");
 501         goto failed_shutdown;
 502     }
 503
 504     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 505     if (r < 0) {
 506         error_setg(&local_err, "error opening pool %s", pool);
 507         goto failed_shutdown;
 508     }
 509
 510     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 511     if (r < 0) {
 512         error_setg(&local_err, "error reading header from %s", s->name);
 513         goto failed_open;
 514     }
 515
 516     bs->read_only = (s->snap != NULL);
 517
 518     qemu_opts_del(opts);
 519     return 0;
 520
 521 failed_open:
 522     rados_ioctx_destroy(s->io_ctx);
 523 failed_shutdown:
 524     rados_shutdown(s->cluster);
 525     g_free(s->snap);
 526 failed_opts:
 527     qemu_opts_del(opts);
 528     return r;
 529 }
 530
 531 static void qemu_rbd_close(BlockDriverState *bs)
 532 {
 533     BDRVRBDState *s = bs->opaque;
 534
 535     rbd_close(s->image);
 536     rados_ioctx_destroy(s->io_ctx);
 537     g_free(s->snap);
 538     rados_shutdown(s->cluster);
 539 }
 540
 541 /*
 542  * Cancel aio. Since we don't reference acb in a non qemu threads,
 543  * it is safe to access it here.
 544  */
 545 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
 546 {
 547     RBDAIOCB *acb = (RBDAIOCB *) blockacb;
 548     acb->cancelled = 1;
 549
 550     while (acb->status == -EINPROGRESS) {
 551         aio_poll(bdrv_get_aio_context(acb->common.bs), true);
 552     }
 553
 554     qemu_aio_release(acb);
 555 }
 556
 557 static const AIOCBInfo rbd_aiocb_info = {
 558     .aiocb_size = sizeof(RBDAIOCB),
 559     .cancel = qemu_rbd_aio_cancel,
 560 };
 561
 562 static void rbd_finish_bh(void *opaque)
 563 {
 564     RADOSCB *rcb = opaque;
 565     qemu_bh_delete(rcb->acb->bh);
 566     qemu_rbd_complete_aio(rcb);
 567 }
 568
 569 /*
 570  * This is the callback function for rbd_aio_read and _write
 571  *
 572  * Note: this function is being called from a non qemu thread so
 573  * we need to be careful about what we do here. Generally we only
 574  * schedule a BH, and do the rest of the io completion handling
 575  * from rbd_finish_bh() which runs in a qemu context.
 576  */
 577 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 578 {
 579     RBDAIOCB *acb = rcb->acb;
 580
 581     rcb->ret = rbd_aio_get_return_value(c);
 582     rbd_aio_release(c);
 583
 584     acb->bh = aio_bh_new(bdrv_get_aio_context(acb->common.bs),
 585                          rbd_finish_bh, rcb);
 586     qemu_bh_schedule(acb->bh);
 587 }
 588
 589 static int rbd_aio_discard_wrapper(rbd_image_t image,
 590                                    uint64_t off,
 591                                    uint64_t len,
 592                                    rbd_completion_t comp)
 593 {
 594 #ifdef LIBRBD_SUPPORTS_DISCARD
 595     return rbd_aio_discard(image, off, len, comp);
 596 #else
 597     return -ENOTSUP;
 598 #endif
 599 }
 600
 601 static int rbd_aio_flush_wrapper(rbd_image_t image,
 602                                  rbd_completion_t comp)
 603 {
 604 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 605     return rbd_aio_flush(image, comp);
 606 #else
 607     return -ENOTSUP;
 608 #endif
 609 }
 610
 611 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
 612                                        int64_t sector_num,
 613                                        QEMUIOVector *qiov,
 614                                        int nb_sectors,
 615                                        BlockDriverCompletionFunc *cb,
 616                                        void *opaque,
 617                                        RBDAIOCmd cmd)
 618 {
 619     RBDAIOCB *acb;
 620     RADOSCB *rcb;
 621     rbd_completion_t c;
 622     int64_t off, size;
 623     char *buf;
 624     int r;
 625
 626     BDRVRBDState *s = bs->opaque;
 627
 628     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 629     acb->cmd = cmd;
 630     acb->qiov = qiov;
 631     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 632         acb->bounce = NULL;
 633     } else {
 634         acb->bounce = qemu_blockalign(bs, qiov->size);
 635     }
 636     acb->ret = 0;
 637     acb->error = 0;
 638     acb->s = s;
 639     acb->cancelled = 0;
 640     acb->bh = NULL;
 641     acb->status = -EINPROGRESS;
 642
 643     if (cmd == RBD_AIO_WRITE) {
 644         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 645     }
 646
 647     buf = acb->bounce;
 648
 649     off = sector_num * BDRV_SECTOR_SIZE;
 650     size = nb_sectors * BDRV_SECTOR_SIZE;
 651
 652     rcb = g_malloc(sizeof(RADOSCB));
 653     rcb->done = 0;
 654     rcb->acb = acb;
 655     rcb->buf = buf;
 656     rcb->s = acb->s;
 657     rcb->size = size;
 658     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 659     if (r < 0) {
 660         goto failed;
 661     }
 662
 663     switch (cmd) {
 664     case RBD_AIO_WRITE:
 665         r = rbd_aio_write(s->image, off, size, buf, c);
 666         break;
 667     case RBD_AIO_READ:
 668         r = rbd_aio_read(s->image, off, size, buf, c);
 669         break;
 670     case RBD_AIO_DISCARD:
 671         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 672         break;
 673     case RBD_AIO_FLUSH:
 674         r = rbd_aio_flush_wrapper(s->image, c);
 675         break;
 676     default:
 677         r = -EINVAL;
 678     }
 679
 680     if (r < 0) {
 681         goto failed_completion;
 682     }
 683
 684     return &acb->common;
 685
 686 failed_completion:
 687     rbd_aio_release(c);
 688 failed:
 689     g_free(rcb);
 690     qemu_vfree(acb->bounce);
 691     qemu_aio_release(acb);
 692     return NULL;
 693 }
 694
 695 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 696                                             int64_t sector_num,
 697                                             QEMUIOVector *qiov,
 698                                             int nb_sectors,
 699                                             BlockDriverCompletionFunc *cb,
 700                                             void *opaque)
 701 {
 702     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 703                          RBD_AIO_READ);
 704 }
 705
 706 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 707                                              int64_t sector_num,
 708                                              QEMUIOVector *qiov,
 709                                              int nb_sectors,
 710                                              BlockDriverCompletionFunc *cb,
 711                                              void *opaque)
 712 {
 713     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 714                          RBD_AIO_WRITE);
 715 }
 716
 717 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 718 static BlockDriverAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 719                                             BlockDriverCompletionFunc *cb,
 720                                             void *opaque)
 721 {
 722     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 723 }
 724
 725 #else
 726
 727 static int qemu_rbd_co_flush(BlockDriverState *bs)
 728 {
 729 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 730     /* rbd_flush added in 0.1.1 */
 731     BDRVRBDState *s = bs->opaque;
 732     return rbd_flush(s->image);
 733 #else
 734     return 0;
 735 #endif
 736 }
 737 #endif
 738
 739 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 740 {
 741     BDRVRBDState *s = bs->opaque;
 742     rbd_image_info_t info;
 743     int r;
 744
 745     r = rbd_stat(s->image, &info, sizeof(info));
 746     if (r < 0) {
 747         return r;
 748     }
 749
 750     bdi->cluster_size = info.obj_size;
 751     return 0;
 752 }
 753
 754 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 755 {
 756     BDRVRBDState *s = bs->opaque;
 757     rbd_image_info_t info;
 758     int r;
 759
 760     r = rbd_stat(s->image, &info, sizeof(info));
 761     if (r < 0) {
 762         return r;
 763     }
 764
 765     return info.size;
 766 }
 767
 768 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 769 {
 770     BDRVRBDState *s = bs->opaque;
 771     int r;
 772
 773     r = rbd_resize(s->image, offset);
 774     if (r < 0) {
 775         return r;
 776     }
 777
 778     return 0;
 779 }
 780
 781 static int qemu_rbd_snap_create(BlockDriverState *bs,
 782                                 QEMUSnapshotInfo *sn_info)
 783 {
 784     BDRVRBDState *s = bs->opaque;
 785     int r;
 786
 787     if (sn_info->name[0] == '\0') {
 788         return -EINVAL; /* we need a name for rbd snapshots */
 789     }
 790
 791     /*
 792      * rbd snapshots are using the name as the user controlled unique identifier
 793      * we can't use the rbd snapid for that purpose, as it can't be set
 794      */
 795     if (sn_info->id_str[0] != '\0' &&
 796         strcmp(sn_info->id_str, sn_info->name) != 0) {
 797         return -EINVAL;
 798     }
 799
 800     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 801         return -ERANGE;
 802     }
 803
 804     r = rbd_snap_create(s->image, sn_info->name);
 805     if (r < 0) {
 806         error_report("failed to create snap: %s", strerror(-r));
 807         return r;
 808     }
 809
 810     return 0;
 811 }
 812
 813 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 814                                 const char *snapshot_id,
 815                                 const char *snapshot_name,
 816                                 Error **errp)
 817 {
 818     BDRVRBDState *s = bs->opaque;
 819     int r;
 820
 821     if (!snapshot_name) {
 822         error_setg(errp, "rbd need a valid snapshot name");
 823         return -EINVAL;
 824     }
 825
 826     /* If snapshot_id is specified, it must be equal to name, see
 827        qemu_rbd_snap_list() */
 828     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 829         error_setg(errp,
 830                    "rbd do not support snapshot id, it should be NULL or "
 831                    "equal to snapshot name");
 832         return -EINVAL;
 833     }
 834
 835     r = rbd_snap_remove(s->image, snapshot_name);
 836     if (r < 0) {
 837         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 838     }
 839     return r;
 840 }
 841
 842 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 843                                   const char *snapshot_name)
 844 {
 845     BDRVRBDState *s = bs->opaque;
 846     int r;
 847
 848     r = rbd_snap_rollback(s->image, snapshot_name);
 849     return r;
 850 }
 851
 852 static int qemu_rbd_snap_list(BlockDriverState *bs,
 853                               QEMUSnapshotInfo **psn_tab)
 854 {
 855     BDRVRBDState *s = bs->opaque;
 856     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 857     int i, snap_count;
 858     rbd_snap_info_t *snaps;
 859     int max_snaps = RBD_MAX_SNAPS;
 860
 861     do {
 862         snaps = g_malloc(sizeof(*snaps) * max_snaps);
 863         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 864         if (snap_count <= 0) {
 865             g_free(snaps);
 866         }
 867     } while (snap_count == -ERANGE);
 868
 869     if (snap_count <= 0) {
 870         goto done;
 871     }
 872
 873     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
 874
 875     for (i = 0; i < snap_count; i++) {
 876         const char *snap_name = snaps[i].name;
 877
 878         sn_info = sn_tab + i;
 879         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 880         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 881
 882         sn_info->vm_state_size = snaps[i].size;
 883         sn_info->date_sec = 0;
 884         sn_info->date_nsec = 0;
 885         sn_info->vm_clock_nsec = 0;
 886     }
 887     rbd_snap_list_end(snaps);
 888     g_free(snaps);
 889
 890  done:
 891     *psn_tab = sn_tab;
 892     return snap_count;
 893 }
 894
 895 #ifdef LIBRBD_SUPPORTS_DISCARD
 896 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 897                                               int64_t sector_num,
 898                                               int nb_sectors,
 899                                               BlockDriverCompletionFunc *cb,
 900                                               void *opaque)
 901 {
 902     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 903                          RBD_AIO_DISCARD);
 904 }
 905 #endif
 906
 907 static QemuOptsList qemu_rbd_create_opts = {
 908     .name = "rbd-create-opts",
 909     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
 910     .desc = {
 911         {
 912             .name = BLOCK_OPT_SIZE,
 913             .type = QEMU_OPT_SIZE,
 914             .help = "Virtual disk size"
 915         },
 916         {
 917             .name = BLOCK_OPT_CLUSTER_SIZE,
 918             .type = QEMU_OPT_SIZE,
 919             .help = "RBD object size"
 920         },
 921         { /* end of list */ }
 922     }
 923 };
 924
 925 static BlockDriver bdrv_rbd = {
 926     .format_name        = "rbd",
 927     .instance_size      = sizeof(BDRVRBDState),
 928     .bdrv_needs_filename = true,
 929     .bdrv_file_open     = qemu_rbd_open,
 930     .bdrv_close         = qemu_rbd_close,
 931     .bdrv_create        = qemu_rbd_create,
 932     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 933     .bdrv_get_info      = qemu_rbd_getinfo,
 934     .create_opts        = &qemu_rbd_create_opts,
 935     .bdrv_getlength     = qemu_rbd_getlength,
 936     .bdrv_truncate      = qemu_rbd_truncate,
 937     .protocol_name      = "rbd",
 938
 939     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 940     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 941
 942 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 943     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 944 #else
 945     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 946 #endif
 947
 948 #ifdef LIBRBD_SUPPORTS_DISCARD
 949     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 950 #endif
 951
 952     .bdrv_snapshot_create   = qemu_rbd_snap_create,
 953     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
 954     .bdrv_snapshot_list     = qemu_rbd_snap_list,
 955     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
 956 };
 957
 958 static void bdrv_rbd_init(void)
 959 {
 960     bdrv_register(&bdrv_rbd);
 961 }
 962
 963 block_init(bdrv_rbd_init);