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