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     BlockAIOCB 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 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     rados_t cluster;
  98     rados_ioctx_t io_ctx;
  99     rbd_image_t image;
 100     char name[RBD_MAX_IMAGE_NAME_SIZE];
 101     char *snap;
 102 } BDRVRBDState;
 103
 104 static int qemu_rbd_next_tok(char *dst, int dst_len,
 105                              char *src, char delim,
 106                              const char *name,
 107                              char **p, Error **errp)
 108 {
 109     int l;
 110     char *end;
 111
 112     *p = NULL;
 113
 114     if (delim != '\0') {
 115         for (end = src; *end; ++end) {
 116             if (*end == delim) {
 117                 break;
 118             }
 119             if (*end == '\\' && end[1] != '\0') {
 120                 end++;
 121             }
 122         }
 123         if (*end == delim) {
 124             *p = end + 1;
 125             *end = '\0';
 126         }
 127     }
 128     l = strlen(src);
 129     if (l >= dst_len) {
 130         error_setg(errp, "%s too long", name);
 131         return -EINVAL;
 132     } else if (l == 0) {
 133         error_setg(errp, "%s too short", name);
 134         return -EINVAL;
 135     }
 136
 137     pstrcpy(dst, dst_len, src);
 138
 139     return 0;
 140 }
 141
 142 static void qemu_rbd_unescape(char *src)
 143 {
 144     char *p;
 145
 146     for (p = src; *src; ++src, ++p) {
 147         if (*src == '\\' && src[1] != '\0') {
 148             src++;
 149         }
 150         *p = *src;
 151     }
 152     *p = '\0';
 153 }
 154
 155 static int qemu_rbd_parsename(const char *filename,
 156                               char *pool, int pool_len,
 157                               char *snap, int snap_len,
 158                               char *name, int name_len,
 159                               char *conf, int conf_len,
 160                               Error **errp)
 161 {
 162     const char *start;
 163     char *p, *buf;
 164     int ret;
 165
 166     if (!strstart(filename, "rbd:", &start)) {
 167         error_setg(errp, "File name must start with 'rbd:'");
 168         return -EINVAL;
 169     }
 170
 171     buf = g_strdup(start);
 172     p = buf;
 173     *snap = '\0';
 174     *conf = '\0';
 175
 176     ret = qemu_rbd_next_tok(pool, pool_len, p,
 177                             '/', "pool name", &p, errp);
 178     if (ret < 0 || !p) {
 179         ret = -EINVAL;
 180         goto done;
 181     }
 182     qemu_rbd_unescape(pool);
 183
 184     if (strchr(p, '@')) {
 185         ret = qemu_rbd_next_tok(name, name_len, p,
 186                                 '@', "object name", &p, errp);
 187         if (ret < 0) {
 188             goto done;
 189         }
 190         ret = qemu_rbd_next_tok(snap, snap_len, p,
 191                                 ':', "snap name", &p, errp);
 192         qemu_rbd_unescape(snap);
 193     } else {
 194         ret = qemu_rbd_next_tok(name, name_len, p,
 195                                 ':', "object name", &p, errp);
 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,
 203                             '\0', "configuration", &p, errp);
 204
 205 done:
 206     g_free(buf);
 207     return ret;
 208 }
 209
 210 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 211 {
 212     const char *p = conf;
 213
 214     while (*p) {
 215         int len;
 216         const char *end = strchr(p, ':');
 217
 218         if (end) {
 219             len = end - p;
 220         } else {
 221             len = strlen(p);
 222         }
 223
 224         if (strncmp(p, "id=", 3) == 0) {
 225             len -= 3;
 226             strncpy(clientname, p + 3, len);
 227             clientname[len] = '\0';
 228             return clientname;
 229         }
 230         if (end == NULL) {
 231             break;
 232         }
 233         p = end + 1;
 234     }
 235     return NULL;
 236 }
 237
 238 static int qemu_rbd_set_conf(rados_t cluster, const char *conf, Error **errp)
 239 {
 240     char *p, *buf;
 241     char name[RBD_MAX_CONF_NAME_SIZE];
 242     char value[RBD_MAX_CONF_VAL_SIZE];
 243     int ret = 0;
 244
 245     buf = g_strdup(conf);
 246     p = buf;
 247
 248     while (p) {
 249         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 250                                 '=', "conf option name", &p, errp);
 251         if (ret < 0) {
 252             break;
 253         }
 254         qemu_rbd_unescape(name);
 255
 256         if (!p) {
 257             error_setg(errp, "conf option %s has no value", name);
 258             ret = -EINVAL;
 259             break;
 260         }
 261
 262         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 263                                 ':', "conf option value", &p, errp);
 264         if (ret < 0) {
 265             break;
 266         }
 267         qemu_rbd_unescape(value);
 268
 269         if (strcmp(name, "conf") == 0) {
 270             ret = rados_conf_read_file(cluster, value);
 271             if (ret < 0) {
 272                 error_setg(errp, "error reading conf file %s", value);
 273                 break;
 274             }
 275         } else if (strcmp(name, "id") == 0) {
 276             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 277         } else {
 278             ret = rados_conf_set(cluster, name, value);
 279             if (ret < 0) {
 280                 error_setg(errp, "invalid conf option %s", name);
 281                 ret = -EINVAL;
 282                 break;
 283             }
 284         }
 285     }
 286
 287     g_free(buf);
 288     return ret;
 289 }
 290
 291 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 292 {
 293     Error *local_err = NULL;
 294     int64_t bytes = 0;
 295     int64_t objsize;
 296     int obj_order = 0;
 297     char pool[RBD_MAX_POOL_NAME_SIZE];
 298     char name[RBD_MAX_IMAGE_NAME_SIZE];
 299     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 300     char conf[RBD_MAX_CONF_SIZE];
 301     char clientname_buf[RBD_MAX_CONF_SIZE];
 302     char *clientname;
 303     rados_t cluster;
 304     rados_ioctx_t io_ctx;
 305     int ret;
 306
 307     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 308                            snap_buf, sizeof(snap_buf),
 309                            name, sizeof(name),
 310                            conf, sizeof(conf), &local_err) < 0) {
 311         error_propagate(errp, local_err);
 312         return -EINVAL;
 313     }
 314
 315     /* Read out options */
 316     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 317                      BDRV_SECTOR_SIZE);
 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 = ctz32(objsize);
 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     qemu_aio_unref(acb);
 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         error_propagate(errp, local_err);
 445         qemu_opts_del(opts);
 446         return -EINVAL;
 447     }
 448
 449     filename = qemu_opt_get(opts, "filename");
 450
 451     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 452                            snap_buf, sizeof(snap_buf),
 453                            s->name, sizeof(s->name),
 454                            conf, sizeof(conf), errp) < 0) {
 455         r = -EINVAL;
 456         goto failed_opts;
 457     }
 458
 459     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 460     r = rados_create(&s->cluster, clientname);
 461     if (r < 0) {
 462         error_setg(errp, "error initializing");
 463         goto failed_opts;
 464     }
 465
 466     s->snap = NULL;
 467     if (snap_buf[0] != '\0') {
 468         s->snap = g_strdup(snap_buf);
 469     }
 470
 471     /*
 472      * Fallback to more conservative semantics if setting cache
 473      * options fails. Ignore errors from setting rbd_cache because the
 474      * only possible error is that the option does not exist, and
 475      * librbd defaults to no caching. If write through caching cannot
 476      * be set up, fall back to no caching.
 477      */
 478     if (flags & BDRV_O_NOCACHE) {
 479         rados_conf_set(s->cluster, "rbd_cache", "false");
 480     } else {
 481         rados_conf_set(s->cluster, "rbd_cache", "true");
 482     }
 483
 484     if (strstr(conf, "conf=") == NULL) {
 485         /* try default location, but ignore failure */
 486         rados_conf_read_file(s->cluster, NULL);
 487     }
 488
 489     if (conf[0] != '\0') {
 490         r = qemu_rbd_set_conf(s->cluster, conf, errp);
 491         if (r < 0) {
 492             goto failed_shutdown;
 493         }
 494     }
 495
 496     r = rados_connect(s->cluster);
 497     if (r < 0) {
 498         error_setg(errp, "error connecting");
 499         goto failed_shutdown;
 500     }
 501
 502     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 503     if (r < 0) {
 504         error_setg(errp, "error opening pool %s", pool);
 505         goto failed_shutdown;
 506     }
 507
 508     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 509     if (r < 0) {
 510         error_setg(errp, "error reading header from %s", s->name);
 511         goto failed_open;
 512     }
 513
 514     bs->read_only = (s->snap != NULL);
 515
 516     qemu_opts_del(opts);
 517     return 0;
 518
 519 failed_open:
 520     rados_ioctx_destroy(s->io_ctx);
 521 failed_shutdown:
 522     rados_shutdown(s->cluster);
 523     g_free(s->snap);
 524 failed_opts:
 525     qemu_opts_del(opts);
 526     return r;
 527 }
 528
 529 static void qemu_rbd_close(BlockDriverState *bs)
 530 {
 531     BDRVRBDState *s = bs->opaque;
 532
 533     rbd_close(s->image);
 534     rados_ioctx_destroy(s->io_ctx);
 535     g_free(s->snap);
 536     rados_shutdown(s->cluster);
 537 }
 538
 539 static const AIOCBInfo rbd_aiocb_info = {
 540     .aiocb_size = sizeof(RBDAIOCB),
 541 };
 542
 543 static void rbd_finish_bh(void *opaque)
 544 {
 545     RADOSCB *rcb = opaque;
 546     qemu_bh_delete(rcb->acb->bh);
 547     qemu_rbd_complete_aio(rcb);
 548 }
 549
 550 /*
 551  * This is the callback function for rbd_aio_read and _write
 552  *
 553  * Note: this function is being called from a non qemu thread so
 554  * we need to be careful about what we do here. Generally we only
 555  * schedule a BH, and do the rest of the io completion handling
 556  * from rbd_finish_bh() which runs in a qemu context.
 557  */
 558 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 559 {
 560     RBDAIOCB *acb = rcb->acb;
 561
 562     rcb->ret = rbd_aio_get_return_value(c);
 563     rbd_aio_release(c);
 564
 565     acb->bh = aio_bh_new(bdrv_get_aio_context(acb->common.bs),
 566                          rbd_finish_bh, rcb);
 567     qemu_bh_schedule(acb->bh);
 568 }
 569
 570 static int rbd_aio_discard_wrapper(rbd_image_t image,
 571                                    uint64_t off,
 572                                    uint64_t len,
 573                                    rbd_completion_t comp)
 574 {
 575 #ifdef LIBRBD_SUPPORTS_DISCARD
 576     return rbd_aio_discard(image, off, len, comp);
 577 #else
 578     return -ENOTSUP;
 579 #endif
 580 }
 581
 582 static int rbd_aio_flush_wrapper(rbd_image_t image,
 583                                  rbd_completion_t comp)
 584 {
 585 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 586     return rbd_aio_flush(image, comp);
 587 #else
 588     return -ENOTSUP;
 589 #endif
 590 }
 591
 592 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 593                                  int64_t sector_num,
 594                                  QEMUIOVector *qiov,
 595                                  int nb_sectors,
 596                                  BlockCompletionFunc *cb,
 597                                  void *opaque,
 598                                  RBDAIOCmd cmd)
 599 {
 600     RBDAIOCB *acb;
 601     RADOSCB *rcb = NULL;
 602     rbd_completion_t c;
 603     int64_t off, size;
 604     char *buf;
 605     int r;
 606
 607     BDRVRBDState *s = bs->opaque;
 608
 609     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 610     acb->cmd = cmd;
 611     acb->qiov = qiov;
 612     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 613         acb->bounce = NULL;
 614     } else {
 615         acb->bounce = qemu_try_blockalign(bs, qiov->size);
 616         if (acb->bounce == NULL) {
 617             goto failed;
 618         }
 619     }
 620     acb->ret = 0;
 621     acb->error = 0;
 622     acb->s = s;
 623     acb->bh = NULL;
 624     acb->status = -EINPROGRESS;
 625
 626     if (cmd == RBD_AIO_WRITE) {
 627         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 628     }
 629
 630     buf = acb->bounce;
 631
 632     off = sector_num * BDRV_SECTOR_SIZE;
 633     size = nb_sectors * BDRV_SECTOR_SIZE;
 634
 635     rcb = g_new(RADOSCB, 1);
 636     rcb->done = 0;
 637     rcb->acb = acb;
 638     rcb->buf = buf;
 639     rcb->s = acb->s;
 640     rcb->size = size;
 641     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 642     if (r < 0) {
 643         goto failed;
 644     }
 645
 646     switch (cmd) {
 647     case RBD_AIO_WRITE:
 648         r = rbd_aio_write(s->image, off, size, buf, c);
 649         break;
 650     case RBD_AIO_READ:
 651         r = rbd_aio_read(s->image, off, size, buf, c);
 652         break;
 653     case RBD_AIO_DISCARD:
 654         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 655         break;
 656     case RBD_AIO_FLUSH:
 657         r = rbd_aio_flush_wrapper(s->image, c);
 658         break;
 659     default:
 660         r = -EINVAL;
 661     }
 662
 663     if (r < 0) {
 664         goto failed_completion;
 665     }
 666
 667     return &acb->common;
 668
 669 failed_completion:
 670     rbd_aio_release(c);
 671 failed:
 672     g_free(rcb);
 673     qemu_vfree(acb->bounce);
 674     qemu_aio_unref(acb);
 675     return NULL;
 676 }
 677
 678 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 679                                       int64_t sector_num,
 680                                       QEMUIOVector *qiov,
 681                                       int nb_sectors,
 682                                       BlockCompletionFunc *cb,
 683                                       void *opaque)
 684 {
 685     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 686                          RBD_AIO_READ);
 687 }
 688
 689 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 690                                        int64_t sector_num,
 691                                        QEMUIOVector *qiov,
 692                                        int nb_sectors,
 693                                        BlockCompletionFunc *cb,
 694                                        void *opaque)
 695 {
 696     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 697                          RBD_AIO_WRITE);
 698 }
 699
 700 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 701 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 702                                       BlockCompletionFunc *cb,
 703                                       void *opaque)
 704 {
 705     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 706 }
 707
 708 #else
 709
 710 static int qemu_rbd_co_flush(BlockDriverState *bs)
 711 {
 712 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 713     /* rbd_flush added in 0.1.1 */
 714     BDRVRBDState *s = bs->opaque;
 715     return rbd_flush(s->image);
 716 #else
 717     return 0;
 718 #endif
 719 }
 720 #endif
 721
 722 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 723 {
 724     BDRVRBDState *s = bs->opaque;
 725     rbd_image_info_t info;
 726     int r;
 727
 728     r = rbd_stat(s->image, &info, sizeof(info));
 729     if (r < 0) {
 730         return r;
 731     }
 732
 733     bdi->cluster_size = info.obj_size;
 734     return 0;
 735 }
 736
 737 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 738 {
 739     BDRVRBDState *s = bs->opaque;
 740     rbd_image_info_t info;
 741     int r;
 742
 743     r = rbd_stat(s->image, &info, sizeof(info));
 744     if (r < 0) {
 745         return r;
 746     }
 747
 748     return info.size;
 749 }
 750
 751 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 752 {
 753     BDRVRBDState *s = bs->opaque;
 754     int r;
 755
 756     r = rbd_resize(s->image, offset);
 757     if (r < 0) {
 758         return r;
 759     }
 760
 761     return 0;
 762 }
 763
 764 static int qemu_rbd_snap_create(BlockDriverState *bs,
 765                                 QEMUSnapshotInfo *sn_info)
 766 {
 767     BDRVRBDState *s = bs->opaque;
 768     int r;
 769
 770     if (sn_info->name[0] == '\0') {
 771         return -EINVAL; /* we need a name for rbd snapshots */
 772     }
 773
 774     /*
 775      * rbd snapshots are using the name as the user controlled unique identifier
 776      * we can't use the rbd snapid for that purpose, as it can't be set
 777      */
 778     if (sn_info->id_str[0] != '\0' &&
 779         strcmp(sn_info->id_str, sn_info->name) != 0) {
 780         return -EINVAL;
 781     }
 782
 783     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 784         return -ERANGE;
 785     }
 786
 787     r = rbd_snap_create(s->image, sn_info->name);
 788     if (r < 0) {
 789         error_report("failed to create snap: %s", strerror(-r));
 790         return r;
 791     }
 792
 793     return 0;
 794 }
 795
 796 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 797                                 const char *snapshot_id,
 798                                 const char *snapshot_name,
 799                                 Error **errp)
 800 {
 801     BDRVRBDState *s = bs->opaque;
 802     int r;
 803
 804     if (!snapshot_name) {
 805         error_setg(errp, "rbd need a valid snapshot name");
 806         return -EINVAL;
 807     }
 808
 809     /* If snapshot_id is specified, it must be equal to name, see
 810        qemu_rbd_snap_list() */
 811     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 812         error_setg(errp,
 813                    "rbd do not support snapshot id, it should be NULL or "
 814                    "equal to snapshot name");
 815         return -EINVAL;
 816     }
 817
 818     r = rbd_snap_remove(s->image, snapshot_name);
 819     if (r < 0) {
 820         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 821     }
 822     return r;
 823 }
 824
 825 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 826                                   const char *snapshot_name)
 827 {
 828     BDRVRBDState *s = bs->opaque;
 829     int r;
 830
 831     r = rbd_snap_rollback(s->image, snapshot_name);
 832     return r;
 833 }
 834
 835 static int qemu_rbd_snap_list(BlockDriverState *bs,
 836                               QEMUSnapshotInfo **psn_tab)
 837 {
 838     BDRVRBDState *s = bs->opaque;
 839     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 840     int i, snap_count;
 841     rbd_snap_info_t *snaps;
 842     int max_snaps = RBD_MAX_SNAPS;
 843
 844     do {
 845         snaps = g_new(rbd_snap_info_t, max_snaps);
 846         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 847         if (snap_count <= 0) {
 848             g_free(snaps);
 849         }
 850     } while (snap_count == -ERANGE);
 851
 852     if (snap_count <= 0) {
 853         goto done;
 854     }
 855
 856     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
 857
 858     for (i = 0; i < snap_count; i++) {
 859         const char *snap_name = snaps[i].name;
 860
 861         sn_info = sn_tab + i;
 862         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 863         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 864
 865         sn_info->vm_state_size = snaps[i].size;
 866         sn_info->date_sec = 0;
 867         sn_info->date_nsec = 0;
 868         sn_info->vm_clock_nsec = 0;
 869     }
 870     rbd_snap_list_end(snaps);
 871     g_free(snaps);
 872
 873  done:
 874     *psn_tab = sn_tab;
 875     return snap_count;
 876 }
 877
 878 #ifdef LIBRBD_SUPPORTS_DISCARD
 879 static BlockAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 880                                         int64_t sector_num,
 881                                         int nb_sectors,
 882                                         BlockCompletionFunc *cb,
 883                                         void *opaque)
 884 {
 885     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 886                          RBD_AIO_DISCARD);
 887 }
 888 #endif
 889
 890 #ifdef LIBRBD_SUPPORTS_INVALIDATE
 891 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
 892                                       Error **errp)
 893 {
 894     BDRVRBDState *s = bs->opaque;
 895     int r = rbd_invalidate_cache(s->image);
 896     if (r < 0) {
 897         error_setg_errno(errp, -r, "Failed to invalidate the cache");
 898     }
 899 }
 900 #endif
 901
 902 static QemuOptsList qemu_rbd_create_opts = {
 903     .name = "rbd-create-opts",
 904     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
 905     .desc = {
 906         {
 907             .name = BLOCK_OPT_SIZE,
 908             .type = QEMU_OPT_SIZE,
 909             .help = "Virtual disk size"
 910         },
 911         {
 912             .name = BLOCK_OPT_CLUSTER_SIZE,
 913             .type = QEMU_OPT_SIZE,
 914             .help = "RBD object size"
 915         },
 916         { /* end of list */ }
 917     }
 918 };
 919
 920 static BlockDriver bdrv_rbd = {
 921     .format_name        = "rbd",
 922     .instance_size      = sizeof(BDRVRBDState),
 923     .bdrv_needs_filename = true,
 924     .bdrv_file_open     = qemu_rbd_open,
 925     .bdrv_close         = qemu_rbd_close,
 926     .bdrv_create        = qemu_rbd_create,
 927     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 928     .bdrv_get_info      = qemu_rbd_getinfo,
 929     .create_opts        = &qemu_rbd_create_opts,
 930     .bdrv_getlength     = qemu_rbd_getlength,
 931     .bdrv_truncate      = qemu_rbd_truncate,
 932     .protocol_name      = "rbd",
 933
 934     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 935     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 936
 937 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 938     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 939 #else
 940     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 941 #endif
 942
 943 #ifdef LIBRBD_SUPPORTS_DISCARD
 944     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 945 #endif
 946
 947     .bdrv_snapshot_create   = qemu_rbd_snap_create,
 948     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
 949     .bdrv_snapshot_list     = qemu_rbd_snap_list,
 950     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
 951 #ifdef LIBRBD_SUPPORTS_INVALIDATE
 952     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
 953 #endif
 954 };
 955
 956 static void bdrv_rbd_init(void)
 957 {
 958     bdrv_register(&bdrv_rbd);
 959 }
 960
 961 block_init(bdrv_rbd_init);