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)
 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_report("%s too long", name);
 132         return -EINVAL;
 133     } else if (l == 0) {
 134         error_report("%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 {
 162     const char *start;
 163     char *p, *buf;
 164     int ret;
 165
 166     if (!strstart(filename, "rbd:", &start)) {
 167         return -EINVAL;
 168     }
 169
 170     buf = g_strdup(start);
 171     p = buf;
 172     *snap = '\0';
 173     *conf = '\0';
 174
 175     ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
 176     if (ret < 0 || !p) {
 177         ret = -EINVAL;
 178         goto done;
 179     }
 180     qemu_rbd_unescape(pool);
 181
 182     if (strchr(p, '@')) {
 183         ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
 184         if (ret < 0) {
 185             goto done;
 186         }
 187         ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
 188         qemu_rbd_unescape(snap);
 189     } else {
 190         ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
 191     }
 192     qemu_rbd_unescape(name);
 193     if (ret < 0 || !p) {
 194         goto done;
 195     }
 196
 197     ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
 198
 199 done:
 200     g_free(buf);
 201     return ret;
 202 }
 203
 204 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 205 {
 206     const char *p = conf;
 207
 208     while (*p) {
 209         int len;
 210         const char *end = strchr(p, ':');
 211
 212         if (end) {
 213             len = end - p;
 214         } else {
 215             len = strlen(p);
 216         }
 217
 218         if (strncmp(p, "id=", 3) == 0) {
 219             len -= 3;
 220             strncpy(clientname, p + 3, len);
 221             clientname[len] = '\0';
 222             return clientname;
 223         }
 224         if (end == NULL) {
 225             break;
 226         }
 227         p = end + 1;
 228     }
 229     return NULL;
 230 }
 231
 232 static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
 233 {
 234     char *p, *buf;
 235     char name[RBD_MAX_CONF_NAME_SIZE];
 236     char value[RBD_MAX_CONF_VAL_SIZE];
 237     int ret = 0;
 238
 239     buf = g_strdup(conf);
 240     p = buf;
 241
 242     while (p) {
 243         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 244                                 '=', "conf option name", &p);
 245         if (ret < 0) {
 246             break;
 247         }
 248         qemu_rbd_unescape(name);
 249
 250         if (!p) {
 251             error_report("conf option %s has no value", name);
 252             ret = -EINVAL;
 253             break;
 254         }
 255
 256         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 257                                 ':', "conf option value", &p);
 258         if (ret < 0) {
 259             break;
 260         }
 261         qemu_rbd_unescape(value);
 262
 263         if (strcmp(name, "conf") == 0) {
 264             ret = rados_conf_read_file(cluster, value);
 265             if (ret < 0) {
 266                 error_report("error reading conf file %s", value);
 267                 break;
 268             }
 269         } else if (strcmp(name, "id") == 0) {
 270             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 271         } else {
 272             ret = rados_conf_set(cluster, name, value);
 273             if (ret < 0) {
 274                 error_report("invalid conf option %s", name);
 275                 ret = -EINVAL;
 276                 break;
 277             }
 278         }
 279     }
 280
 281     g_free(buf);
 282     return ret;
 283 }
 284
 285 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options,
 286                            Error **errp)
 287 {
 288     int64_t bytes = 0;
 289     int64_t objsize;
 290     int obj_order = 0;
 291     char pool[RBD_MAX_POOL_NAME_SIZE];
 292     char name[RBD_MAX_IMAGE_NAME_SIZE];
 293     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 294     char conf[RBD_MAX_CONF_SIZE];
 295     char clientname_buf[RBD_MAX_CONF_SIZE];
 296     char *clientname;
 297     rados_t cluster;
 298     rados_ioctx_t io_ctx;
 299     int ret;
 300
 301     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 302                            snap_buf, sizeof(snap_buf),
 303                            name, sizeof(name),
 304                            conf, sizeof(conf)) < 0) {
 305         return -EINVAL;
 306     }
 307
 308     /* Read out options */
 309     while (options && options->name) {
 310         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 311             bytes = options->value.n;
 312         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
 313             if (options->value.n) {
 314                 objsize = options->value.n;
 315                 if ((objsize - 1) & objsize) {    /* not a power of 2? */
 316                     error_report("obj size needs to be power of 2");
 317                     return -EINVAL;
 318                 }
 319                 if (objsize < 4096) {
 320                     error_report("obj size too small");
 321                     return -EINVAL;
 322                 }
 323                 obj_order = ffs(objsize) - 1;
 324             }
 325         }
 326         options++;
 327     }
 328
 329     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 330     if (rados_create(&cluster, clientname) < 0) {
 331         error_report("error initializing");
 332         return -EIO;
 333     }
 334
 335     if (strstr(conf, "conf=") == NULL) {
 336         /* try default location, but ignore failure */
 337         rados_conf_read_file(cluster, NULL);
 338     }
 339
 340     if (conf[0] != '\0' &&
 341         qemu_rbd_set_conf(cluster, conf) < 0) {
 342         error_report("error setting config options");
 343         rados_shutdown(cluster);
 344         return -EIO;
 345     }
 346
 347     if (rados_connect(cluster) < 0) {
 348         error_report("error connecting");
 349         rados_shutdown(cluster);
 350         return -EIO;
 351     }
 352
 353     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 354         error_report("error opening pool %s", pool);
 355         rados_shutdown(cluster);
 356         return -EIO;
 357     }
 358
 359     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 360     rados_ioctx_destroy(io_ctx);
 361     rados_shutdown(cluster);
 362
 363     return ret;
 364 }
 365
 366 /*
 367  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 368  * BH context.
 369  */
 370 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 371 {
 372     RBDAIOCB *acb = rcb->acb;
 373     int64_t r;
 374
 375     r = rcb->ret;
 376
 377     if (acb->cmd != RBD_AIO_READ) {
 378         if (r < 0) {
 379             acb->ret = r;
 380             acb->error = 1;
 381         } else if (!acb->error) {
 382             acb->ret = rcb->size;
 383         }
 384     } else {
 385         if (r < 0) {
 386             memset(rcb->buf, 0, rcb->size);
 387             acb->ret = r;
 388             acb->error = 1;
 389         } else if (r < rcb->size) {
 390             memset(rcb->buf + r, 0, rcb->size - r);
 391             if (!acb->error) {
 392                 acb->ret = rcb->size;
 393             }
 394         } else if (!acb->error) {
 395             acb->ret = r;
 396         }
 397     }
 398
 399     g_free(rcb);
 400
 401     if (acb->cmd == RBD_AIO_READ) {
 402         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 403     }
 404     qemu_vfree(acb->bounce);
 405     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 406     acb->status = 0;
 407
 408     if (!acb->cancelled) {
 409         qemu_aio_release(acb);
 410     }
 411 }
 412
 413 /* TODO Convert to fine grained options */
 414 static QemuOptsList runtime_opts = {
 415     .name = "rbd",
 416     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 417     .desc = {
 418         {
 419             .name = "filename",
 420             .type = QEMU_OPT_STRING,
 421             .help = "Specification of the rbd image",
 422         },
 423         { /* end of list */ }
 424     },
 425 };
 426
 427 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 428                          Error **errp)
 429 {
 430     BDRVRBDState *s = bs->opaque;
 431     char pool[RBD_MAX_POOL_NAME_SIZE];
 432     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 433     char conf[RBD_MAX_CONF_SIZE];
 434     char clientname_buf[RBD_MAX_CONF_SIZE];
 435     char *clientname;
 436     QemuOpts *opts;
 437     Error *local_err = NULL;
 438     const char *filename;
 439     int r;
 440
 441     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 442     qemu_opts_absorb_qdict(opts, options, &local_err);
 443     if (local_err) {
 444         qerror_report_err(local_err);
 445         error_free(local_err);
 446         qemu_opts_del(opts);
 447         return -EINVAL;
 448     }
 449
 450     filename = qemu_opt_get(opts, "filename");
 451
 452     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 453                            snap_buf, sizeof(snap_buf),
 454                            s->name, sizeof(s->name),
 455                            conf, sizeof(conf)) < 0) {
 456         r = -EINVAL;
 457         goto failed_opts;
 458     }
 459
 460     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 461     r = rados_create(&s->cluster, clientname);
 462     if (r < 0) {
 463         error_report("error initializing");
 464         goto failed_opts;
 465     }
 466
 467     s->snap = NULL;
 468     if (snap_buf[0] != '\0') {
 469         s->snap = g_strdup(snap_buf);
 470     }
 471
 472     /*
 473      * Fallback to more conservative semantics if setting cache
 474      * options fails. Ignore errors from setting rbd_cache because the
 475      * only possible error is that the option does not exist, and
 476      * librbd defaults to no caching. If write through caching cannot
 477      * be set up, fall back to no caching.
 478      */
 479     if (flags & BDRV_O_NOCACHE) {
 480         rados_conf_set(s->cluster, "rbd_cache", "false");
 481     } else {
 482         rados_conf_set(s->cluster, "rbd_cache", "true");
 483     }
 484
 485     if (strstr(conf, "conf=") == NULL) {
 486         /* try default location, but ignore failure */
 487         rados_conf_read_file(s->cluster, NULL);
 488     }
 489
 490     if (conf[0] != '\0') {
 491         r = qemu_rbd_set_conf(s->cluster, conf);
 492         if (r < 0) {
 493             error_report("error setting config options");
 494             goto failed_shutdown;
 495         }
 496     }
 497
 498     r = rados_connect(s->cluster);
 499     if (r < 0) {
 500         error_report("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_report("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_report("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         qemu_aio_wait();
 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 = qemu_bh_new(rbd_finish_bh, rcb);
 585     qemu_bh_schedule(acb->bh);
 586 }
 587
 588 static int rbd_aio_discard_wrapper(rbd_image_t image,
 589                                    uint64_t off,
 590                                    uint64_t len,
 591                                    rbd_completion_t comp)
 592 {
 593 #ifdef LIBRBD_SUPPORTS_DISCARD
 594     return rbd_aio_discard(image, off, len, comp);
 595 #else
 596     return -ENOTSUP;
 597 #endif
 598 }
 599
 600 static int rbd_aio_flush_wrapper(rbd_image_t image,
 601                                  rbd_completion_t comp)
 602 {
 603 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 604     return rbd_aio_flush(image, comp);
 605 #else
 606     return -ENOTSUP;
 607 #endif
 608 }
 609
 610 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
 611                                        int64_t sector_num,
 612                                        QEMUIOVector *qiov,
 613                                        int nb_sectors,
 614                                        BlockDriverCompletionFunc *cb,
 615                                        void *opaque,
 616                                        RBDAIOCmd cmd)
 617 {
 618     RBDAIOCB *acb;
 619     RADOSCB *rcb;
 620     rbd_completion_t c;
 621     int64_t off, size;
 622     char *buf;
 623     int r;
 624
 625     BDRVRBDState *s = bs->opaque;
 626
 627     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 628     acb->cmd = cmd;
 629     acb->qiov = qiov;
 630     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 631         acb->bounce = NULL;
 632     } else {
 633         acb->bounce = qemu_blockalign(bs, qiov->size);
 634     }
 635     acb->ret = 0;
 636     acb->error = 0;
 637     acb->s = s;
 638     acb->cancelled = 0;
 639     acb->bh = NULL;
 640     acb->status = -EINPROGRESS;
 641
 642     if (cmd == RBD_AIO_WRITE) {
 643         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 644     }
 645
 646     buf = acb->bounce;
 647
 648     off = sector_num * BDRV_SECTOR_SIZE;
 649     size = nb_sectors * BDRV_SECTOR_SIZE;
 650
 651     rcb = g_malloc(sizeof(RADOSCB));
 652     rcb->done = 0;
 653     rcb->acb = acb;
 654     rcb->buf = buf;
 655     rcb->s = acb->s;
 656     rcb->size = size;
 657     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 658     if (r < 0) {
 659         goto failed;
 660     }
 661
 662     switch (cmd) {
 663     case RBD_AIO_WRITE:
 664         r = rbd_aio_write(s->image, off, size, buf, c);
 665         break;
 666     case RBD_AIO_READ:
 667         r = rbd_aio_read(s->image, off, size, buf, c);
 668         break;
 669     case RBD_AIO_DISCARD:
 670         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 671         break;
 672     case RBD_AIO_FLUSH:
 673         r = rbd_aio_flush_wrapper(s->image, c);
 674         break;
 675     default:
 676         r = -EINVAL;
 677     }
 678
 679     if (r < 0) {
 680         goto failed;
 681     }
 682
 683     return &acb->common;
 684
 685 failed:
 686     g_free(rcb);
 687     qemu_aio_release(acb);
 688     return NULL;
 689 }
 690
 691 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 692                                             int64_t sector_num,
 693                                             QEMUIOVector *qiov,
 694                                             int nb_sectors,
 695                                             BlockDriverCompletionFunc *cb,
 696                                             void *opaque)
 697 {
 698     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 699                          RBD_AIO_READ);
 700 }
 701
 702 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 703                                              int64_t sector_num,
 704                                              QEMUIOVector *qiov,
 705                                              int nb_sectors,
 706                                              BlockDriverCompletionFunc *cb,
 707                                              void *opaque)
 708 {
 709     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 710                          RBD_AIO_WRITE);
 711 }
 712
 713 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 714 static BlockDriverAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 715                                             BlockDriverCompletionFunc *cb,
 716                                             void *opaque)
 717 {
 718     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 719 }
 720
 721 #else
 722
 723 static int qemu_rbd_co_flush(BlockDriverState *bs)
 724 {
 725 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 726     /* rbd_flush added in 0.1.1 */
 727     BDRVRBDState *s = bs->opaque;
 728     return rbd_flush(s->image);
 729 #else
 730     return 0;
 731 #endif
 732 }
 733 #endif
 734
 735 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 736 {
 737     BDRVRBDState *s = bs->opaque;
 738     rbd_image_info_t info;
 739     int r;
 740
 741     r = rbd_stat(s->image, &info, sizeof(info));
 742     if (r < 0) {
 743         return r;
 744     }
 745
 746     bdi->cluster_size = info.obj_size;
 747     return 0;
 748 }
 749
 750 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 751 {
 752     BDRVRBDState *s = bs->opaque;
 753     rbd_image_info_t info;
 754     int r;
 755
 756     r = rbd_stat(s->image, &info, sizeof(info));
 757     if (r < 0) {
 758         return r;
 759     }
 760
 761     return info.size;
 762 }
 763
 764 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 765 {
 766     BDRVRBDState *s = bs->opaque;
 767     int r;
 768
 769     r = rbd_resize(s->image, offset);
 770     if (r < 0) {
 771         return r;
 772     }
 773
 774     return 0;
 775 }
 776
 777 static int qemu_rbd_snap_create(BlockDriverState *bs,
 778                                 QEMUSnapshotInfo *sn_info)
 779 {
 780     BDRVRBDState *s = bs->opaque;
 781     int r;
 782
 783     if (sn_info->name[0] == '\0') {
 784         return -EINVAL; /* we need a name for rbd snapshots */
 785     }
 786
 787     /*
 788      * rbd snapshots are using the name as the user controlled unique identifier
 789      * we can't use the rbd snapid for that purpose, as it can't be set
 790      */
 791     if (sn_info->id_str[0] != '\0' &&
 792         strcmp(sn_info->id_str, sn_info->name) != 0) {
 793         return -EINVAL;
 794     }
 795
 796     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 797         return -ERANGE;
 798     }
 799
 800     r = rbd_snap_create(s->image, sn_info->name);
 801     if (r < 0) {
 802         error_report("failed to create snap: %s", strerror(-r));
 803         return r;
 804     }
 805
 806     return 0;
 807 }
 808
 809 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 810                                 const char *snapshot_id,
 811                                 const char *snapshot_name,
 812                                 Error **errp)
 813 {
 814     BDRVRBDState *s = bs->opaque;
 815     int r;
 816
 817     if (!snapshot_name) {
 818         error_setg(errp, "rbd need a valid snapshot name");
 819         return -EINVAL;
 820     }
 821
 822     /* If snapshot_id is specified, it must be equal to name, see
 823        qemu_rbd_snap_list() */
 824     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 825         error_setg(errp,
 826                    "rbd do not support snapshot id, it should be NULL or "
 827                    "equal to snapshot name");
 828         return -EINVAL;
 829     }
 830
 831     r = rbd_snap_remove(s->image, snapshot_name);
 832     if (r < 0) {
 833         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 834     }
 835     return r;
 836 }
 837
 838 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 839                                   const char *snapshot_name)
 840 {
 841     BDRVRBDState *s = bs->opaque;
 842     int r;
 843
 844     r = rbd_snap_rollback(s->image, snapshot_name);
 845     return r;
 846 }
 847
 848 static int qemu_rbd_snap_list(BlockDriverState *bs,
 849                               QEMUSnapshotInfo **psn_tab)
 850 {
 851     BDRVRBDState *s = bs->opaque;
 852     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 853     int i, snap_count;
 854     rbd_snap_info_t *snaps;
 855     int max_snaps = RBD_MAX_SNAPS;
 856
 857     do {
 858         snaps = g_malloc(sizeof(*snaps) * max_snaps);
 859         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 860         if (snap_count <= 0) {
 861             g_free(snaps);
 862         }
 863     } while (snap_count == -ERANGE);
 864
 865     if (snap_count <= 0) {
 866         goto done;
 867     }
 868
 869     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
 870
 871     for (i = 0; i < snap_count; i++) {
 872         const char *snap_name = snaps[i].name;
 873
 874         sn_info = sn_tab + i;
 875         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 876         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 877
 878         sn_info->vm_state_size = snaps[i].size;
 879         sn_info->date_sec = 0;
 880         sn_info->date_nsec = 0;
 881         sn_info->vm_clock_nsec = 0;
 882     }
 883     rbd_snap_list_end(snaps);
 884     g_free(snaps);
 885
 886  done:
 887     *psn_tab = sn_tab;
 888     return snap_count;
 889 }
 890
 891 #ifdef LIBRBD_SUPPORTS_DISCARD
 892 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 893                                               int64_t sector_num,
 894                                               int nb_sectors,
 895                                               BlockDriverCompletionFunc *cb,
 896                                               void *opaque)
 897 {
 898     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 899                          RBD_AIO_DISCARD);
 900 }
 901 #endif
 902
 903 static QEMUOptionParameter qemu_rbd_create_options[] = {
 904     {
 905      .name = BLOCK_OPT_SIZE,
 906      .type = OPT_SIZE,
 907      .help = "Virtual disk size"
 908     },
 909     {
 910      .name = BLOCK_OPT_CLUSTER_SIZE,
 911      .type = OPT_SIZE,
 912      .help = "RBD object size"
 913     },
 914     {NULL}
 915 };
 916
 917 static BlockDriver bdrv_rbd = {
 918     .format_name        = "rbd",
 919     .instance_size      = sizeof(BDRVRBDState),
 920     .bdrv_needs_filename = true,
 921     .bdrv_file_open     = qemu_rbd_open,
 922     .bdrv_close         = qemu_rbd_close,
 923     .bdrv_create        = qemu_rbd_create,
 924     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 925     .bdrv_get_info      = qemu_rbd_getinfo,
 926     .create_options     = qemu_rbd_create_options,
 927     .bdrv_getlength     = qemu_rbd_getlength,
 928     .bdrv_truncate      = qemu_rbd_truncate,
 929     .protocol_name      = "rbd",
 930
 931     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 932     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 933
 934 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 935     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 936 #else
 937     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 938 #endif
 939
 940 #ifdef LIBRBD_SUPPORTS_DISCARD
 941     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 942 #endif
 943
 944     .bdrv_snapshot_create   = qemu_rbd_snap_create,
 945     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
 946     .bdrv_snapshot_list     = qemu_rbd_snap_list,
 947     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
 948 };
 949
 950 static void bdrv_rbd_init(void)
 951 {
 952     bdrv_register(&bdrv_rbd);
 953 }
 954
 955 block_init(bdrv_rbd_init);