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 "qemu/osdep.h"
  15
  16 #include "qapi/error.h"
  17 #include "qemu/error-report.h"
  18 #include "block/block_int.h"
  19 #include "crypto/secret.h"
  20 #include "qemu/cutils.h"
  21
  22 #include <rbd/librbd.h>
  23
  24 /*
  25  * When specifying the image filename use:
  26  *
  27  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
  28  *
  29  * poolname must be the name of an existing rados pool.
  30  *
  31  * devicename is the name of the rbd image.
  32  *
  33  * Each option given is used to configure rados, and may be any valid
  34  * Ceph option, "id", or "conf".
  35  *
  36  * The "id" option indicates what user we should authenticate as to
  37  * the Ceph cluster.  If it is excluded we will use the Ceph default
  38  * (normally 'admin').
  39  *
  40  * The "conf" option specifies a Ceph configuration file to read.  If
  41  * it is not specified, we will read from the default Ceph locations
  42  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
  43  * file, specify conf=/dev/null.
  44  *
  45  * Configuration values containing :, @, or = can be escaped with a
  46  * leading "\".
  47  */
  48
  49 /* rbd_aio_discard added in 0.1.2 */
  50 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
  51 #define LIBRBD_SUPPORTS_DISCARD
  52 #else
  53 #undef LIBRBD_SUPPORTS_DISCARD
  54 #endif
  55
  56 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
  57
  58 #define RBD_MAX_CONF_NAME_SIZE 128
  59 #define RBD_MAX_CONF_VAL_SIZE 512
  60 #define RBD_MAX_CONF_SIZE 1024
  61 #define RBD_MAX_POOL_NAME_SIZE 128
  62 #define RBD_MAX_SNAP_NAME_SIZE 128
  63 #define RBD_MAX_SNAPS 100
  64
  65 typedef enum {
  66     RBD_AIO_READ,
  67     RBD_AIO_WRITE,
  68     RBD_AIO_DISCARD,
  69     RBD_AIO_FLUSH
  70 } RBDAIOCmd;
  71
  72 typedef struct RBDAIOCB {
  73     BlockAIOCB common;
  74     QEMUBH *bh;
  75     int64_t ret;
  76     QEMUIOVector *qiov;
  77     char *bounce;
  78     RBDAIOCmd cmd;
  79     int error;
  80     struct BDRVRBDState *s;
  81 } RBDAIOCB;
  82
  83 typedef struct RADOSCB {
  84     RBDAIOCB *acb;
  85     struct BDRVRBDState *s;
  86     int64_t size;
  87     char *buf;
  88     int64_t ret;
  89 } RADOSCB;
  90
  91 typedef struct BDRVRBDState {
  92     rados_t cluster;
  93     rados_ioctx_t io_ctx;
  94     rbd_image_t image;
  95     char name[RBD_MAX_IMAGE_NAME_SIZE];
  96     char *snap;
  97 } BDRVRBDState;
  98
  99 static int qemu_rbd_next_tok(char *dst, int dst_len,
 100                              char *src, char delim,
 101                              const char *name,
 102                              char **p, Error **errp)
 103 {
 104     int l;
 105     char *end;
 106
 107     *p = NULL;
 108
 109     if (delim != '\0') {
 110         for (end = src; *end; ++end) {
 111             if (*end == delim) {
 112                 break;
 113             }
 114             if (*end == '\\' && end[1] != '\0') {
 115                 end++;
 116             }
 117         }
 118         if (*end == delim) {
 119             *p = end + 1;
 120             *end = '\0';
 121         }
 122     }
 123     l = strlen(src);
 124     if (l >= dst_len) {
 125         error_setg(errp, "%s too long", name);
 126         return -EINVAL;
 127     } else if (l == 0) {
 128         error_setg(errp, "%s too short", name);
 129         return -EINVAL;
 130     }
 131
 132     pstrcpy(dst, dst_len, src);
 133
 134     return 0;
 135 }
 136
 137 static void qemu_rbd_unescape(char *src)
 138 {
 139     char *p;
 140
 141     for (p = src; *src; ++src, ++p) {
 142         if (*src == '\\' && src[1] != '\0') {
 143             src++;
 144         }
 145         *p = *src;
 146     }
 147     *p = '\0';
 148 }
 149
 150 static int qemu_rbd_parsename(const char *filename,
 151                               char *pool, int pool_len,
 152                               char *snap, int snap_len,
 153                               char *name, int name_len,
 154                               char *conf, int conf_len,
 155                               Error **errp)
 156 {
 157     const char *start;
 158     char *p, *buf;
 159     int ret;
 160
 161     if (!strstart(filename, "rbd:", &start)) {
 162         error_setg(errp, "File name must start with 'rbd:'");
 163         return -EINVAL;
 164     }
 165
 166     buf = g_strdup(start);
 167     p = buf;
 168     *snap = '\0';
 169     *conf = '\0';
 170
 171     ret = qemu_rbd_next_tok(pool, pool_len, p,
 172                             '/', "pool name", &p, errp);
 173     if (ret < 0 || !p) {
 174         ret = -EINVAL;
 175         goto done;
 176     }
 177     qemu_rbd_unescape(pool);
 178
 179     if (strchr(p, '@')) {
 180         ret = qemu_rbd_next_tok(name, name_len, p,
 181                                 '@', "object name", &p, errp);
 182         if (ret < 0) {
 183             goto done;
 184         }
 185         ret = qemu_rbd_next_tok(snap, snap_len, p,
 186                                 ':', "snap name", &p, errp);
 187         qemu_rbd_unescape(snap);
 188     } else {
 189         ret = qemu_rbd_next_tok(name, name_len, p,
 190                                 ':', "object name", &p, errp);
 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,
 198                             '\0', "configuration", &p, errp);
 199
 200 done:
 201     g_free(buf);
 202     return ret;
 203 }
 204
 205 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 206 {
 207     const char *p = conf;
 208
 209     while (*p) {
 210         int len;
 211         const char *end = strchr(p, ':');
 212
 213         if (end) {
 214             len = end - p;
 215         } else {
 216             len = strlen(p);
 217         }
 218
 219         if (strncmp(p, "id=", 3) == 0) {
 220             len -= 3;
 221             strncpy(clientname, p + 3, len);
 222             clientname[len] = '\0';
 223             return clientname;
 224         }
 225         if (end == NULL) {
 226             break;
 227         }
 228         p = end + 1;
 229     }
 230     return NULL;
 231 }
 232
 233
 234 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
 235                              Error **errp)
 236 {
 237     if (secretid == 0) {
 238         return 0;
 239     }
 240
 241     gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
 242                                                     errp);
 243     if (!secret) {
 244         return -1;
 245     }
 246
 247     rados_conf_set(cluster, "key", secret);
 248     g_free(secret);
 249
 250     return 0;
 251 }
 252
 253
 254 static int qemu_rbd_set_conf(rados_t cluster, const char *conf,
 255                              bool only_read_conf_file,
 256                              Error **errp)
 257 {
 258     char *p, *buf;
 259     char name[RBD_MAX_CONF_NAME_SIZE];
 260     char value[RBD_MAX_CONF_VAL_SIZE];
 261     int ret = 0;
 262
 263     buf = g_strdup(conf);
 264     p = buf;
 265
 266     while (p) {
 267         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 268                                 '=', "conf option name", &p, errp);
 269         if (ret < 0) {
 270             break;
 271         }
 272         qemu_rbd_unescape(name);
 273
 274         if (!p) {
 275             error_setg(errp, "conf option %s has no value", name);
 276             ret = -EINVAL;
 277             break;
 278         }
 279
 280         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 281                                 ':', "conf option value", &p, errp);
 282         if (ret < 0) {
 283             break;
 284         }
 285         qemu_rbd_unescape(value);
 286
 287         if (strcmp(name, "conf") == 0) {
 288             /* read the conf file alone, so it doesn't override more
 289                specific settings for a particular device */
 290             if (only_read_conf_file) {
 291                 ret = rados_conf_read_file(cluster, value);
 292                 if (ret < 0) {
 293                     error_setg(errp, "error reading conf file %s", value);
 294                     break;
 295                 }
 296             }
 297         } else if (strcmp(name, "id") == 0) {
 298             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 299         } else if (!only_read_conf_file) {
 300             ret = rados_conf_set(cluster, name, value);
 301             if (ret < 0) {
 302                 error_setg(errp, "invalid conf option %s", name);
 303                 ret = -EINVAL;
 304                 break;
 305             }
 306         }
 307     }
 308
 309     g_free(buf);
 310     return ret;
 311 }
 312
 313 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 314 {
 315     Error *local_err = NULL;
 316     int64_t bytes = 0;
 317     int64_t objsize;
 318     int obj_order = 0;
 319     char pool[RBD_MAX_POOL_NAME_SIZE];
 320     char name[RBD_MAX_IMAGE_NAME_SIZE];
 321     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 322     char conf[RBD_MAX_CONF_SIZE];
 323     char clientname_buf[RBD_MAX_CONF_SIZE];
 324     char *clientname;
 325     const char *secretid;
 326     rados_t cluster;
 327     rados_ioctx_t io_ctx;
 328     int ret;
 329
 330     secretid = qemu_opt_get(opts, "password-secret");
 331
 332     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 333                            snap_buf, sizeof(snap_buf),
 334                            name, sizeof(name),
 335                            conf, sizeof(conf), &local_err) < 0) {
 336         error_propagate(errp, local_err);
 337         return -EINVAL;
 338     }
 339
 340     /* Read out options */
 341     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 342                      BDRV_SECTOR_SIZE);
 343     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 344     if (objsize) {
 345         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 346             error_setg(errp, "obj size needs to be power of 2");
 347             return -EINVAL;
 348         }
 349         if (objsize < 4096) {
 350             error_setg(errp, "obj size too small");
 351             return -EINVAL;
 352         }
 353         obj_order = ctz32(objsize);
 354     }
 355
 356     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 357     if (rados_create(&cluster, clientname) < 0) {
 358         error_setg(errp, "error initializing");
 359         return -EIO;
 360     }
 361
 362     if (strstr(conf, "conf=") == NULL) {
 363         /* try default location, but ignore failure */
 364         rados_conf_read_file(cluster, NULL);
 365     } else if (conf[0] != '\0' &&
 366                qemu_rbd_set_conf(cluster, conf, true, &local_err) < 0) {
 367         rados_shutdown(cluster);
 368         error_propagate(errp, local_err);
 369         return -EIO;
 370     }
 371
 372     if (conf[0] != '\0' &&
 373         qemu_rbd_set_conf(cluster, conf, false, &local_err) < 0) {
 374         rados_shutdown(cluster);
 375         error_propagate(errp, local_err);
 376         return -EIO;
 377     }
 378
 379     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
 380         rados_shutdown(cluster);
 381         return -EIO;
 382     }
 383
 384     if (rados_connect(cluster) < 0) {
 385         error_setg(errp, "error connecting");
 386         rados_shutdown(cluster);
 387         return -EIO;
 388     }
 389
 390     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 391         error_setg(errp, "error opening pool %s", pool);
 392         rados_shutdown(cluster);
 393         return -EIO;
 394     }
 395
 396     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 397     rados_ioctx_destroy(io_ctx);
 398     rados_shutdown(cluster);
 399
 400     return ret;
 401 }
 402
 403 /*
 404  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 405  * BH context.
 406  */
 407 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 408 {
 409     RBDAIOCB *acb = rcb->acb;
 410     int64_t r;
 411
 412     r = rcb->ret;
 413
 414     if (acb->cmd != RBD_AIO_READ) {
 415         if (r < 0) {
 416             acb->ret = r;
 417             acb->error = 1;
 418         } else if (!acb->error) {
 419             acb->ret = rcb->size;
 420         }
 421     } else {
 422         if (r < 0) {
 423             memset(rcb->buf, 0, rcb->size);
 424             acb->ret = r;
 425             acb->error = 1;
 426         } else if (r < rcb->size) {
 427             memset(rcb->buf + r, 0, rcb->size - r);
 428             if (!acb->error) {
 429                 acb->ret = rcb->size;
 430             }
 431         } else if (!acb->error) {
 432             acb->ret = r;
 433         }
 434     }
 435
 436     g_free(rcb);
 437
 438     if (acb->cmd == RBD_AIO_READ) {
 439         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 440     }
 441     qemu_vfree(acb->bounce);
 442     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 443
 444     qemu_aio_unref(acb);
 445 }
 446
 447 /* TODO Convert to fine grained options */
 448 static QemuOptsList runtime_opts = {
 449     .name = "rbd",
 450     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 451     .desc = {
 452         {
 453             .name = "filename",
 454             .type = QEMU_OPT_STRING,
 455             .help = "Specification of the rbd image",
 456         },
 457         {
 458             .name = "password-secret",
 459             .type = QEMU_OPT_STRING,
 460             .help = "ID of secret providing the password",
 461         },
 462         { /* end of list */ }
 463     },
 464 };
 465
 466 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 467                          Error **errp)
 468 {
 469     BDRVRBDState *s = bs->opaque;
 470     char pool[RBD_MAX_POOL_NAME_SIZE];
 471     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 472     char conf[RBD_MAX_CONF_SIZE];
 473     char clientname_buf[RBD_MAX_CONF_SIZE];
 474     char *clientname;
 475     const char *secretid;
 476     QemuOpts *opts;
 477     Error *local_err = NULL;
 478     const char *filename;
 479     int r;
 480
 481     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 482     qemu_opts_absorb_qdict(opts, options, &local_err);
 483     if (local_err) {
 484         error_propagate(errp, local_err);
 485         qemu_opts_del(opts);
 486         return -EINVAL;
 487     }
 488
 489     filename = qemu_opt_get(opts, "filename");
 490     secretid = qemu_opt_get(opts, "password-secret");
 491
 492     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 493                            snap_buf, sizeof(snap_buf),
 494                            s->name, sizeof(s->name),
 495                            conf, sizeof(conf), errp) < 0) {
 496         r = -EINVAL;
 497         goto failed_opts;
 498     }
 499
 500     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 501     r = rados_create(&s->cluster, clientname);
 502     if (r < 0) {
 503         error_setg(errp, "error initializing");
 504         goto failed_opts;
 505     }
 506
 507     s->snap = NULL;
 508     if (snap_buf[0] != '\0') {
 509         s->snap = g_strdup(snap_buf);
 510     }
 511
 512     if (strstr(conf, "conf=") == NULL) {
 513         /* try default location, but ignore failure */
 514         rados_conf_read_file(s->cluster, NULL);
 515     } else if (conf[0] != '\0') {
 516         r = qemu_rbd_set_conf(s->cluster, conf, true, errp);
 517         if (r < 0) {
 518             goto failed_shutdown;
 519         }
 520     }
 521
 522     if (conf[0] != '\0') {
 523         r = qemu_rbd_set_conf(s->cluster, conf, false, errp);
 524         if (r < 0) {
 525             goto failed_shutdown;
 526         }
 527     }
 528
 529     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
 530         r = -EIO;
 531         goto failed_shutdown;
 532     }
 533
 534     /*
 535      * Fallback to more conservative semantics if setting cache
 536      * options fails. Ignore errors from setting rbd_cache because the
 537      * only possible error is that the option does not exist, and
 538      * librbd defaults to no caching. If write through caching cannot
 539      * be set up, fall back to no caching.
 540      */
 541     if (flags & BDRV_O_NOCACHE) {
 542         rados_conf_set(s->cluster, "rbd_cache", "false");
 543     } else {
 544         rados_conf_set(s->cluster, "rbd_cache", "true");
 545     }
 546
 547     r = rados_connect(s->cluster);
 548     if (r < 0) {
 549         error_setg(errp, "error connecting");
 550         goto failed_shutdown;
 551     }
 552
 553     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 554     if (r < 0) {
 555         error_setg(errp, "error opening pool %s", pool);
 556         goto failed_shutdown;
 557     }
 558
 559     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 560     if (r < 0) {
 561         error_setg(errp, "error reading header from %s", s->name);
 562         goto failed_open;
 563     }
 564
 565     bs->read_only = (s->snap != NULL);
 566
 567     qemu_opts_del(opts);
 568     return 0;
 569
 570 failed_open:
 571     rados_ioctx_destroy(s->io_ctx);
 572 failed_shutdown:
 573     rados_shutdown(s->cluster);
 574     g_free(s->snap);
 575 failed_opts:
 576     qemu_opts_del(opts);
 577     return r;
 578 }
 579
 580 static void qemu_rbd_close(BlockDriverState *bs)
 581 {
 582     BDRVRBDState *s = bs->opaque;
 583
 584     rbd_close(s->image);
 585     rados_ioctx_destroy(s->io_ctx);
 586     g_free(s->snap);
 587     rados_shutdown(s->cluster);
 588 }
 589
 590 static const AIOCBInfo rbd_aiocb_info = {
 591     .aiocb_size = sizeof(RBDAIOCB),
 592 };
 593
 594 static void rbd_finish_bh(void *opaque)
 595 {
 596     RADOSCB *rcb = opaque;
 597     qemu_bh_delete(rcb->acb->bh);
 598     qemu_rbd_complete_aio(rcb);
 599 }
 600
 601 /*
 602  * This is the callback function for rbd_aio_read and _write
 603  *
 604  * Note: this function is being called from a non qemu thread so
 605  * we need to be careful about what we do here. Generally we only
 606  * schedule a BH, and do the rest of the io completion handling
 607  * from rbd_finish_bh() which runs in a qemu context.
 608  */
 609 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 610 {
 611     RBDAIOCB *acb = rcb->acb;
 612
 613     rcb->ret = rbd_aio_get_return_value(c);
 614     rbd_aio_release(c);
 615
 616     acb->bh = aio_bh_new(bdrv_get_aio_context(acb->common.bs),
 617                          rbd_finish_bh, rcb);
 618     qemu_bh_schedule(acb->bh);
 619 }
 620
 621 static int rbd_aio_discard_wrapper(rbd_image_t image,
 622                                    uint64_t off,
 623                                    uint64_t len,
 624                                    rbd_completion_t comp)
 625 {
 626 #ifdef LIBRBD_SUPPORTS_DISCARD
 627     return rbd_aio_discard(image, off, len, comp);
 628 #else
 629     return -ENOTSUP;
 630 #endif
 631 }
 632
 633 static int rbd_aio_flush_wrapper(rbd_image_t image,
 634                                  rbd_completion_t comp)
 635 {
 636 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 637     return rbd_aio_flush(image, comp);
 638 #else
 639     return -ENOTSUP;
 640 #endif
 641 }
 642
 643 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 644                                  int64_t sector_num,
 645                                  QEMUIOVector *qiov,
 646                                  int nb_sectors,
 647                                  BlockCompletionFunc *cb,
 648                                  void *opaque,
 649                                  RBDAIOCmd cmd)
 650 {
 651     RBDAIOCB *acb;
 652     RADOSCB *rcb = NULL;
 653     rbd_completion_t c;
 654     int64_t off, size;
 655     char *buf;
 656     int r;
 657
 658     BDRVRBDState *s = bs->opaque;
 659
 660     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 661     acb->cmd = cmd;
 662     acb->qiov = qiov;
 663     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 664         acb->bounce = NULL;
 665     } else {
 666         acb->bounce = qemu_try_blockalign(bs, qiov->size);
 667         if (acb->bounce == NULL) {
 668             goto failed;
 669         }
 670     }
 671     acb->ret = 0;
 672     acb->error = 0;
 673     acb->s = s;
 674     acb->bh = NULL;
 675
 676     if (cmd == RBD_AIO_WRITE) {
 677         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 678     }
 679
 680     buf = acb->bounce;
 681
 682     off = sector_num * BDRV_SECTOR_SIZE;
 683     size = nb_sectors * BDRV_SECTOR_SIZE;
 684
 685     rcb = g_new(RADOSCB, 1);
 686     rcb->acb = acb;
 687     rcb->buf = buf;
 688     rcb->s = acb->s;
 689     rcb->size = size;
 690     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 691     if (r < 0) {
 692         goto failed;
 693     }
 694
 695     switch (cmd) {
 696     case RBD_AIO_WRITE:
 697         r = rbd_aio_write(s->image, off, size, buf, c);
 698         break;
 699     case RBD_AIO_READ:
 700         r = rbd_aio_read(s->image, off, size, buf, c);
 701         break;
 702     case RBD_AIO_DISCARD:
 703         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 704         break;
 705     case RBD_AIO_FLUSH:
 706         r = rbd_aio_flush_wrapper(s->image, c);
 707         break;
 708     default:
 709         r = -EINVAL;
 710     }
 711
 712     if (r < 0) {
 713         goto failed_completion;
 714     }
 715
 716     return &acb->common;
 717
 718 failed_completion:
 719     rbd_aio_release(c);
 720 failed:
 721     g_free(rcb);
 722     qemu_vfree(acb->bounce);
 723     qemu_aio_unref(acb);
 724     return NULL;
 725 }
 726
 727 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 728                                       int64_t sector_num,
 729                                       QEMUIOVector *qiov,
 730                                       int nb_sectors,
 731                                       BlockCompletionFunc *cb,
 732                                       void *opaque)
 733 {
 734     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 735                          RBD_AIO_READ);
 736 }
 737
 738 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 739                                        int64_t sector_num,
 740                                        QEMUIOVector *qiov,
 741                                        int nb_sectors,
 742                                        BlockCompletionFunc *cb,
 743                                        void *opaque)
 744 {
 745     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 746                          RBD_AIO_WRITE);
 747 }
 748
 749 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 750 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 751                                       BlockCompletionFunc *cb,
 752                                       void *opaque)
 753 {
 754     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 755 }
 756
 757 #else
 758
 759 static int qemu_rbd_co_flush(BlockDriverState *bs)
 760 {
 761 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 762     /* rbd_flush added in 0.1.1 */
 763     BDRVRBDState *s = bs->opaque;
 764     return rbd_flush(s->image);
 765 #else
 766     return 0;
 767 #endif
 768 }
 769 #endif
 770
 771 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 772 {
 773     BDRVRBDState *s = bs->opaque;
 774     rbd_image_info_t info;
 775     int r;
 776
 777     r = rbd_stat(s->image, &info, sizeof(info));
 778     if (r < 0) {
 779         return r;
 780     }
 781
 782     bdi->cluster_size = info.obj_size;
 783     return 0;
 784 }
 785
 786 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 787 {
 788     BDRVRBDState *s = bs->opaque;
 789     rbd_image_info_t info;
 790     int r;
 791
 792     r = rbd_stat(s->image, &info, sizeof(info));
 793     if (r < 0) {
 794         return r;
 795     }
 796
 797     return info.size;
 798 }
 799
 800 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 801 {
 802     BDRVRBDState *s = bs->opaque;
 803     int r;
 804
 805     r = rbd_resize(s->image, offset);
 806     if (r < 0) {
 807         return r;
 808     }
 809
 810     return 0;
 811 }
 812
 813 static int qemu_rbd_snap_create(BlockDriverState *bs,
 814                                 QEMUSnapshotInfo *sn_info)
 815 {
 816     BDRVRBDState *s = bs->opaque;
 817     int r;
 818
 819     if (sn_info->name[0] == '\0') {
 820         return -EINVAL; /* we need a name for rbd snapshots */
 821     }
 822
 823     /*
 824      * rbd snapshots are using the name as the user controlled unique identifier
 825      * we can't use the rbd snapid for that purpose, as it can't be set
 826      */
 827     if (sn_info->id_str[0] != '\0' &&
 828         strcmp(sn_info->id_str, sn_info->name) != 0) {
 829         return -EINVAL;
 830     }
 831
 832     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 833         return -ERANGE;
 834     }
 835
 836     r = rbd_snap_create(s->image, sn_info->name);
 837     if (r < 0) {
 838         error_report("failed to create snap: %s", strerror(-r));
 839         return r;
 840     }
 841
 842     return 0;
 843 }
 844
 845 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 846                                 const char *snapshot_id,
 847                                 const char *snapshot_name,
 848                                 Error **errp)
 849 {
 850     BDRVRBDState *s = bs->opaque;
 851     int r;
 852
 853     if (!snapshot_name) {
 854         error_setg(errp, "rbd need a valid snapshot name");
 855         return -EINVAL;
 856     }
 857
 858     /* If snapshot_id is specified, it must be equal to name, see
 859        qemu_rbd_snap_list() */
 860     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 861         error_setg(errp,
 862                    "rbd do not support snapshot id, it should be NULL or "
 863                    "equal to snapshot name");
 864         return -EINVAL;
 865     }
 866
 867     r = rbd_snap_remove(s->image, snapshot_name);
 868     if (r < 0) {
 869         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 870     }
 871     return r;
 872 }
 873
 874 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 875                                   const char *snapshot_name)
 876 {
 877     BDRVRBDState *s = bs->opaque;
 878     int r;
 879
 880     r = rbd_snap_rollback(s->image, snapshot_name);
 881     return r;
 882 }
 883
 884 static int qemu_rbd_snap_list(BlockDriverState *bs,
 885                               QEMUSnapshotInfo **psn_tab)
 886 {
 887     BDRVRBDState *s = bs->opaque;
 888     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 889     int i, snap_count;
 890     rbd_snap_info_t *snaps;
 891     int max_snaps = RBD_MAX_SNAPS;
 892
 893     do {
 894         snaps = g_new(rbd_snap_info_t, max_snaps);
 895         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 896         if (snap_count <= 0) {
 897             g_free(snaps);
 898         }
 899     } while (snap_count == -ERANGE);
 900
 901     if (snap_count <= 0) {
 902         goto done;
 903     }
 904
 905     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
 906
 907     for (i = 0; i < snap_count; i++) {
 908         const char *snap_name = snaps[i].name;
 909
 910         sn_info = sn_tab + i;
 911         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 912         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 913
 914         sn_info->vm_state_size = snaps[i].size;
 915         sn_info->date_sec = 0;
 916         sn_info->date_nsec = 0;
 917         sn_info->vm_clock_nsec = 0;
 918     }
 919     rbd_snap_list_end(snaps);
 920     g_free(snaps);
 921
 922  done:
 923     *psn_tab = sn_tab;
 924     return snap_count;
 925 }
 926
 927 #ifdef LIBRBD_SUPPORTS_DISCARD
 928 static BlockAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 929                                         int64_t sector_num,
 930                                         int nb_sectors,
 931                                         BlockCompletionFunc *cb,
 932                                         void *opaque)
 933 {
 934     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 935                          RBD_AIO_DISCARD);
 936 }
 937 #endif
 938
 939 #ifdef LIBRBD_SUPPORTS_INVALIDATE
 940 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
 941                                       Error **errp)
 942 {
 943     BDRVRBDState *s = bs->opaque;
 944     int r = rbd_invalidate_cache(s->image);
 945     if (r < 0) {
 946         error_setg_errno(errp, -r, "Failed to invalidate the cache");
 947     }
 948 }
 949 #endif
 950
 951 static QemuOptsList qemu_rbd_create_opts = {
 952     .name = "rbd-create-opts",
 953     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
 954     .desc = {
 955         {
 956             .name = BLOCK_OPT_SIZE,
 957             .type = QEMU_OPT_SIZE,
 958             .help = "Virtual disk size"
 959         },
 960         {
 961             .name = BLOCK_OPT_CLUSTER_SIZE,
 962             .type = QEMU_OPT_SIZE,
 963             .help = "RBD object size"
 964         },
 965         {
 966             .name = "password-secret",
 967             .type = QEMU_OPT_STRING,
 968             .help = "ID of secret providing the password",
 969         },
 970         { /* end of list */ }
 971     }
 972 };
 973
 974 static BlockDriver bdrv_rbd = {
 975     .format_name        = "rbd",
 976     .instance_size      = sizeof(BDRVRBDState),
 977     .bdrv_needs_filename = true,
 978     .bdrv_file_open     = qemu_rbd_open,
 979     .bdrv_close         = qemu_rbd_close,
 980     .bdrv_create        = qemu_rbd_create,
 981     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 982     .bdrv_get_info      = qemu_rbd_getinfo,
 983     .create_opts        = &qemu_rbd_create_opts,
 984     .bdrv_getlength     = qemu_rbd_getlength,
 985     .bdrv_truncate      = qemu_rbd_truncate,
 986     .protocol_name      = "rbd",
 987
 988     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 989     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 990
 991 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 992     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 993 #else
 994     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 995 #endif
 996
 997 #ifdef LIBRBD_SUPPORTS_DISCARD
 998     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 999 #endif
1000
1001     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1002     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1003     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1004     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1005 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1006     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1007 #endif
1008 };
1009
1010 static void bdrv_rbd_init(void)
1011 {
1012     bdrv_register(&bdrv_rbd);
1013 }
1014
1015 block_init(bdrv_rbd_init);