block/rbd.c

   1 /*
   2  * QEMU Block driver for RADOS (Ceph)
   3  *
   4  * Copyright (C) 2010-2011 Christian Brunner <chb@muc.de>,
   5  *                         Josh Durgin <josh.durgin@dreamhost.com>
   6  *
   7  * This work is licensed under the terms of the GNU GPL, version 2.  See
   8  * the COPYING file in the top-level directory.
   9  *
  10  * Contributions after 2012-01-13 are licensed under the terms of the
  11  * GNU GPL, version 2 or (at your option) any later version.
  12  */
  13
  14 #include <inttypes.h>
  15
  16 #include "qemu-common.h"
  17 #include "qemu/error-report.h"
  18 #include "block/block_int.h"
  19
  20 #include <rbd/librbd.h>
  21
  22 /*
  23  * When specifying the image filename use:
  24  *
  25  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
  26  *
  27  * poolname must be the name of an existing rados pool.
  28  *
  29  * devicename is the name of the rbd image.
  30  *
  31  * Each option given is used to configure rados, and may be any valid
  32  * Ceph option, "id", or "conf".
  33  *
  34  * The "id" option indicates what user we should authenticate as to
  35  * the Ceph cluster.  If it is excluded we will use the Ceph default
  36  * (normally 'admin').
  37  *
  38  * The "conf" option specifies a Ceph configuration file to read.  If
  39  * it is not specified, we will read from the default Ceph locations
  40  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
  41  * file, specify conf=/dev/null.
  42  *
  43  * Configuration values containing :, @, or = can be escaped with a
  44  * leading "\".
  45  */
  46
  47 /* rbd_aio_discard added in 0.1.2 */
  48 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
  49 #define LIBRBD_SUPPORTS_DISCARD
  50 #else
  51 #undef LIBRBD_SUPPORTS_DISCARD
  52 #endif
  53
  54 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
  55
  56 #define RBD_MAX_CONF_NAME_SIZE 128
  57 #define RBD_MAX_CONF_VAL_SIZE 512
  58 #define RBD_MAX_CONF_SIZE 1024
  59 #define RBD_MAX_POOL_NAME_SIZE 128
  60 #define RBD_MAX_SNAP_NAME_SIZE 128
  61 #define RBD_MAX_SNAPS 100
  62
  63 typedef enum {
  64     RBD_AIO_READ,
  65     RBD_AIO_WRITE,
  66     RBD_AIO_DISCARD,
  67     RBD_AIO_FLUSH
  68 } RBDAIOCmd;
  69
  70 typedef struct RBDAIOCB {
  71     BlockDriverAIOCB common;
  72     QEMUBH *bh;
  73     int64_t ret;
  74     QEMUIOVector *qiov;
  75     char *bounce;
  76     RBDAIOCmd cmd;
  77     int64_t sector_num;
  78     int error;
  79     struct BDRVRBDState *s;
  80     int cancelled;
  81     int status;
  82 } RBDAIOCB;
  83
  84 typedef struct RADOSCB {
  85     int rcbid;
  86     RBDAIOCB *acb;
  87     struct BDRVRBDState *s;
  88     int done;
  89     int64_t size;
  90     char *buf;
  91     int64_t ret;
  92 } RADOSCB;
  93
  94 #define RBD_FD_READ 0
  95 #define RBD_FD_WRITE 1
  96
  97 typedef struct BDRVRBDState {
  98     rados_t cluster;
  99     rados_ioctx_t io_ctx;
 100     rbd_image_t image;
 101     char name[RBD_MAX_IMAGE_NAME_SIZE];
 102     char *snap;
 103 } BDRVRBDState;
 104
 105 static int qemu_rbd_next_tok(char *dst, int dst_len,
 106                              char *src, char delim,
 107                              const char *name,
 108                              char **p, Error **errp)
 109 {
 110     int l;
 111     char *end;
 112
 113     *p = NULL;
 114
 115     if (delim != '\0') {
 116         for (end = src; *end; ++end) {
 117             if (*end == delim) {
 118                 break;
 119             }
 120             if (*end == '\\' && end[1] != '\0') {
 121                 end++;
 122             }
 123         }
 124         if (*end == delim) {
 125             *p = end + 1;
 126             *end = '\0';
 127         }
 128     }
 129     l = strlen(src);
 130     if (l >= dst_len) {
 131         error_setg(errp, "%s too long", name);
 132         return -EINVAL;
 133     } else if (l == 0) {
 134         error_setg(errp, "%s too short", name);
 135         return -EINVAL;
 136     }
 137
 138     pstrcpy(dst, dst_len, src);
 139
 140     return 0;
 141 }
 142
 143 static void qemu_rbd_unescape(char *src)
 144 {
 145     char *p;
 146
 147     for (p = src; *src; ++src, ++p) {
 148         if (*src == '\\' && src[1] != '\0') {
 149             src++;
 150         }
 151         *p = *src;
 152     }
 153     *p = '\0';
 154 }
 155
 156 static int qemu_rbd_parsename(const char *filename,
 157                               char *pool, int pool_len,
 158                               char *snap, int snap_len,
 159                               char *name, int name_len,
 160                               char *conf, int conf_len,
 161                               Error **errp)
 162 {
 163     const char *start;
 164     char *p, *buf;
 165     int ret;
 166
 167     if (!strstart(filename, "rbd:", &start)) {
 168         error_setg(errp, "File name must start with 'rbd:'");
 169         return -EINVAL;
 170     }
 171
 172     buf = g_strdup(start);
 173     p = buf;
 174     *snap = '\0';
 175     *conf = '\0';
 176
 177     ret = qemu_rbd_next_tok(pool, pool_len, p,
 178                             '/', "pool name", &p, errp);
 179     if (ret < 0 || !p) {
 180         ret = -EINVAL;
 181         goto done;
 182     }
 183     qemu_rbd_unescape(pool);
 184
 185     if (strchr(p, '@')) {
 186         ret = qemu_rbd_next_tok(name, name_len, p,
 187                                 '@', "object name", &p, errp);
 188         if (ret < 0) {
 189             goto done;
 190         }
 191         ret = qemu_rbd_next_tok(snap, snap_len, p,
 192                                 ':', "snap name", &p, errp);
 193         qemu_rbd_unescape(snap);
 194     } else {
 195         ret = qemu_rbd_next_tok(name, name_len, p,
 196                                 ':', "object name", &p, errp);
 197     }
 198     qemu_rbd_unescape(name);
 199     if (ret < 0 || !p) {
 200         goto done;
 201     }
 202
 203     ret = qemu_rbd_next_tok(conf, conf_len, p,
 204                             '\0', "configuration", &p, errp);
 205
 206 done:
 207     g_free(buf);
 208     return ret;
 209 }
 210
 211 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 212 {
 213     const char *p = conf;
 214
 215     while (*p) {
 216         int len;
 217         const char *end = strchr(p, ':');
 218
 219         if (end) {
 220             len = end - p;
 221         } else {
 222             len = strlen(p);
 223         }
 224
 225         if (strncmp(p, "id=", 3) == 0) {
 226             len -= 3;
 227             strncpy(clientname, p + 3, len);
 228             clientname[len] = '\0';
 229             return clientname;
 230         }
 231         if (end == NULL) {
 232             break;
 233         }
 234         p = end + 1;
 235     }
 236     return NULL;
 237 }
 238
 239 static int qemu_rbd_set_conf(rados_t cluster, const char *conf, Error **errp)
 240 {
 241     char *p, *buf;
 242     char name[RBD_MAX_CONF_NAME_SIZE];
 243     char value[RBD_MAX_CONF_VAL_SIZE];
 244     int ret = 0;
 245
 246     buf = g_strdup(conf);
 247     p = buf;
 248
 249     while (p) {
 250         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 251                                 '=', "conf option name", &p, errp);
 252         if (ret < 0) {
 253             break;
 254         }
 255         qemu_rbd_unescape(name);
 256
 257         if (!p) {
 258             error_setg(errp, "conf option %s has no value", name);
 259             ret = -EINVAL;
 260             break;
 261         }
 262
 263         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 264                                 ':', "conf option value", &p, errp);
 265         if (ret < 0) {
 266             break;
 267         }
 268         qemu_rbd_unescape(value);
 269
 270         if (strcmp(name, "conf") == 0) {
 271             ret = rados_conf_read_file(cluster, value);
 272             if (ret < 0) {
 273                 error_setg(errp, "error reading conf file %s", value);
 274                 break;
 275             }
 276         } else if (strcmp(name, "id") == 0) {
 277             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 278         } else {
 279             ret = rados_conf_set(cluster, name, value);
 280             if (ret < 0) {
 281                 error_setg(errp, "invalid conf option %s", name);
 282                 ret = -EINVAL;
 283                 break;
 284             }
 285         }
 286     }
 287
 288     g_free(buf);
 289     return ret;
 290 }
 291
 292 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 293 {
 294     Error *local_err = NULL;
 295     int64_t bytes = 0;
 296     int64_t objsize;
 297     int obj_order = 0;
 298     char pool[RBD_MAX_POOL_NAME_SIZE];
 299     char name[RBD_MAX_IMAGE_NAME_SIZE];
 300     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 301     char conf[RBD_MAX_CONF_SIZE];
 302     char clientname_buf[RBD_MAX_CONF_SIZE];
 303     char *clientname;
 304     rados_t cluster;
 305     rados_ioctx_t io_ctx;
 306     int ret;
 307
 308     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 309                            snap_buf, sizeof(snap_buf),
 310                            name, sizeof(name),
 311                            conf, sizeof(conf), &local_err) < 0) {
 312         error_propagate(errp, local_err);
 313         return -EINVAL;
 314     }
 315
 316     /* Read out options */
 317     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 318                      BDRV_SECTOR_SIZE);
 319     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 320     if (objsize) {
 321         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 322             error_setg(errp, "obj size needs to be power of 2");
 323             return -EINVAL;
 324         }
 325         if (objsize < 4096) {
 326             error_setg(errp, "obj size too small");
 327             return -EINVAL;
 328         }
 329         obj_order = ffs(objsize) - 1;
 330     }
 331
 332     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 333     if (rados_create(&cluster, clientname) < 0) {
 334         error_setg(errp, "error initializing");
 335         return -EIO;
 336     }
 337
 338     if (strstr(conf, "conf=") == NULL) {
 339         /* try default location, but ignore failure */
 340         rados_conf_read_file(cluster, NULL);
 341     }
 342
 343     if (conf[0] != '\0' &&
 344         qemu_rbd_set_conf(cluster, conf, &local_err) < 0) {
 345         rados_shutdown(cluster);
 346         error_propagate(errp, local_err);
 347         return -EIO;
 348     }
 349
 350     if (rados_connect(cluster) < 0) {
 351         error_setg(errp, "error connecting");
 352         rados_shutdown(cluster);
 353         return -EIO;
 354     }
 355
 356     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 357         error_setg(errp, "error opening pool %s", pool);
 358         rados_shutdown(cluster);
 359         return -EIO;
 360     }
 361
 362     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 363     rados_ioctx_destroy(io_ctx);
 364     rados_shutdown(cluster);
 365
 366     return ret;
 367 }
 368
 369 /*
 370  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 371  * BH context.
 372  */
 373 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 374 {
 375     RBDAIOCB *acb = rcb->acb;
 376     int64_t r;
 377
 378     r = rcb->ret;
 379
 380     if (acb->cmd != RBD_AIO_READ) {
 381         if (r < 0) {
 382             acb->ret = r;
 383             acb->error = 1;
 384         } else if (!acb->error) {
 385             acb->ret = rcb->size;
 386         }
 387     } else {
 388         if (r < 0) {
 389             memset(rcb->buf, 0, rcb->size);
 390             acb->ret = r;
 391             acb->error = 1;
 392         } else if (r < rcb->size) {
 393             memset(rcb->buf + r, 0, rcb->size - r);
 394             if (!acb->error) {
 395                 acb->ret = rcb->size;
 396             }
 397         } else if (!acb->error) {
 398             acb->ret = r;
 399         }
 400     }
 401
 402     g_free(rcb);
 403
 404     if (acb->cmd == RBD_AIO_READ) {
 405         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 406     }
 407     qemu_vfree(acb->bounce);
 408     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 409     acb->status = 0;
 410
 411     if (!acb->cancelled) {
 412         qemu_aio_release(acb);
 413     }
 414 }
 415
 416 /* TODO Convert to fine grained options */
 417 static QemuOptsList runtime_opts = {
 418     .name = "rbd",
 419     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 420     .desc = {
 421         {
 422             .name = "filename",
 423             .type = QEMU_OPT_STRING,
 424             .help = "Specification of the rbd image",
 425         },
 426         { /* end of list */ }
 427     },
 428 };
 429
 430 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 431                          Error **errp)
 432 {
 433     BDRVRBDState *s = bs->opaque;
 434     char pool[RBD_MAX_POOL_NAME_SIZE];
 435     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 436     char conf[RBD_MAX_CONF_SIZE];
 437     char clientname_buf[RBD_MAX_CONF_SIZE];
 438     char *clientname;
 439     QemuOpts *opts;
 440     Error *local_err = NULL;
 441     const char *filename;
 442     int r;
 443
 444     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 445     qemu_opts_absorb_qdict(opts, options, &local_err);
 446     if (local_err) {
 447         error_propagate(errp, local_err);
 448         qemu_opts_del(opts);
 449         return -EINVAL;
 450     }
 451
 452     filename = qemu_opt_get(opts, "filename");
 453
 454     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 455                            snap_buf, sizeof(snap_buf),
 456                            s->name, sizeof(s->name),
 457                            conf, sizeof(conf), errp) < 0) {
 458         r = -EINVAL;
 459         goto failed_opts;
 460     }
 461
 462     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 463     r = rados_create(&s->cluster, clientname);
 464     if (r < 0) {
 465         error_setg(&local_err, "error initializing");
 466         goto failed_opts;
 467     }
 468
 469     s->snap = NULL;
 470     if (snap_buf[0] != '\0') {
 471         s->snap = g_strdup(snap_buf);
 472     }
 473
 474     /*
 475      * Fallback to more conservative semantics if setting cache
 476      * options fails. Ignore errors from setting rbd_cache because the
 477      * only possible error is that the option does not exist, and
 478      * librbd defaults to no caching. If write through caching cannot
 479      * be set up, fall back to no caching.
 480      */
 481     if (flags & BDRV_O_NOCACHE) {
 482         rados_conf_set(s->cluster, "rbd_cache", "false");
 483     } else {
 484         rados_conf_set(s->cluster, "rbd_cache", "true");
 485     }
 486
 487     if (strstr(conf, "conf=") == NULL) {
 488         /* try default location, but ignore failure */
 489         rados_conf_read_file(s->cluster, NULL);
 490     }
 491
 492     if (conf[0] != '\0') {
 493         r = qemu_rbd_set_conf(s->cluster, conf, errp);
 494         if (r < 0) {
 495             goto failed_shutdown;
 496         }
 497     }
 498
 499     r = rados_connect(s->cluster);
 500     if (r < 0) {
 501         error_setg(&local_err, "error connecting");
 502         goto failed_shutdown;
 503     }
 504
 505     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 506     if (r < 0) {
 507         error_setg(&local_err, "error opening pool %s", pool);
 508         goto failed_shutdown;
 509     }
 510
 511     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 512     if (r < 0) {
 513         error_setg(&local_err, "error reading header from %s", s->name);
 514         goto failed_open;
 515     }
 516
 517     bs->read_only = (s->snap != NULL);
 518
 519     qemu_opts_del(opts);
 520     return 0;
 521
 522 failed_open:
 523     rados_ioctx_destroy(s->io_ctx);
 524 failed_shutdown:
 525     rados_shutdown(s->cluster);
 526     g_free(s->snap);
 527 failed_opts:
 528     qemu_opts_del(opts);
 529     return r;
 530 }
 531
 532 static void qemu_rbd_close(BlockDriverState *bs)
 533 {
 534     BDRVRBDState *s = bs->opaque;
 535
 536     rbd_close(s->image);
 537     rados_ioctx_destroy(s->io_ctx);
 538     g_free(s->snap);
 539     rados_shutdown(s->cluster);
 540 }
 541
 542 /*
 543  * Cancel aio. Since we don't reference acb in a non qemu threads,
 544  * it is safe to access it here.
 545  */
 546 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
 547 {
 548     RBDAIOCB *acb = (RBDAIOCB *) blockacb;
 549     acb->cancelled = 1;
 550
 551     while (acb->status == -EINPROGRESS) {
 552         aio_poll(bdrv_get_aio_context(acb->common.bs), true);
 553     }
 554
 555     qemu_aio_release(acb);
 556 }
 557
 558 static const AIOCBInfo rbd_aiocb_info = {
 559     .aiocb_size = sizeof(RBDAIOCB),
 560     .cancel = qemu_rbd_aio_cancel,
 561 };
 562
 563 static void rbd_finish_bh(void *opaque)
 564 {
 565     RADOSCB *rcb = opaque;
 566     qemu_bh_delete(rcb->acb->bh);
 567     qemu_rbd_complete_aio(rcb);
 568 }
 569
 570 /*
 571  * This is the callback function for rbd_aio_read and _write
 572  *
 573  * Note: this function is being called from a non qemu thread so
 574  * we need to be careful about what we do here. Generally we only
 575  * schedule a BH, and do the rest of the io completion handling
 576  * from rbd_finish_bh() which runs in a qemu context.
 577  */
 578 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 579 {
 580     RBDAIOCB *acb = rcb->acb;
 581
 582     rcb->ret = rbd_aio_get_return_value(c);
 583     rbd_aio_release(c);
 584
 585     acb->bh = aio_bh_new(bdrv_get_aio_context(acb->common.bs),
 586                          rbd_finish_bh, rcb);
 587     qemu_bh_schedule(acb->bh);
 588 }
 589
 590 static int rbd_aio_discard_wrapper(rbd_image_t image,
 591                                    uint64_t off,
 592                                    uint64_t len,
 593                                    rbd_completion_t comp)
 594 {
 595 #ifdef LIBRBD_SUPPORTS_DISCARD
 596     return rbd_aio_discard(image, off, len, comp);
 597 #else
 598     return -ENOTSUP;
 599 #endif
 600 }
 601
 602 static int rbd_aio_flush_wrapper(rbd_image_t image,
 603                                  rbd_completion_t comp)
 604 {
 605 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 606     return rbd_aio_flush(image, comp);
 607 #else
 608     return -ENOTSUP;
 609 #endif
 610 }
 611
 612 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
 613                                        int64_t sector_num,
 614                                        QEMUIOVector *qiov,
 615                                        int nb_sectors,
 616                                        BlockDriverCompletionFunc *cb,
 617                                        void *opaque,
 618                                        RBDAIOCmd cmd)
 619 {
 620     RBDAIOCB *acb;
 621     RADOSCB *rcb = NULL;
 622     rbd_completion_t c;
 623     int64_t off, size;
 624     char *buf;
 625     int r;
 626
 627     BDRVRBDState *s = bs->opaque;
 628
 629     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 630     acb->cmd = cmd;
 631     acb->qiov = qiov;
 632     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 633         acb->bounce = NULL;
 634     } else {
 635         acb->bounce = qemu_try_blockalign(bs, qiov->size);
 636         if (acb->bounce == NULL) {
 637             goto failed;
 638         }
 639     }
 640     acb->ret = 0;
 641     acb->error = 0;
 642     acb->s = s;
 643     acb->cancelled = 0;
 644     acb->bh = NULL;
 645     acb->status = -EINPROGRESS;
 646
 647     if (cmd == RBD_AIO_WRITE) {
 648         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 649     }
 650
 651     buf = acb->bounce;
 652
 653     off = sector_num * BDRV_SECTOR_SIZE;
 654     size = nb_sectors * BDRV_SECTOR_SIZE;
 655
 656     rcb = g_new(RADOSCB, 1);
 657     rcb->done = 0;
 658     rcb->acb = acb;
 659     rcb->buf = buf;
 660     rcb->s = acb->s;
 661     rcb->size = size;
 662     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 663     if (r < 0) {
 664         goto failed;
 665     }
 666
 667     switch (cmd) {
 668     case RBD_AIO_WRITE:
 669         r = rbd_aio_write(s->image, off, size, buf, c);
 670         break;
 671     case RBD_AIO_READ:
 672         r = rbd_aio_read(s->image, off, size, buf, c);
 673         break;
 674     case RBD_AIO_DISCARD:
 675         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 676         break;
 677     case RBD_AIO_FLUSH:
 678         r = rbd_aio_flush_wrapper(s->image, c);
 679         break;
 680     default:
 681         r = -EINVAL;
 682     }
 683
 684     if (r < 0) {
 685         goto failed_completion;
 686     }
 687
 688     return &acb->common;
 689
 690 failed_completion:
 691     rbd_aio_release(c);
 692 failed:
 693     g_free(rcb);
 694     qemu_vfree(acb->bounce);
 695     qemu_aio_release(acb);
 696     return NULL;
 697 }
 698
 699 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 700                                             int64_t sector_num,
 701                                             QEMUIOVector *qiov,
 702                                             int nb_sectors,
 703                                             BlockDriverCompletionFunc *cb,
 704                                             void *opaque)
 705 {
 706     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 707                          RBD_AIO_READ);
 708 }
 709
 710 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 711                                              int64_t sector_num,
 712                                              QEMUIOVector *qiov,
 713                                              int nb_sectors,
 714                                              BlockDriverCompletionFunc *cb,
 715                                              void *opaque)
 716 {
 717     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 718                          RBD_AIO_WRITE);
 719 }
 720
 721 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 722 static BlockDriverAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 723                                             BlockDriverCompletionFunc *cb,
 724                                             void *opaque)
 725 {
 726     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 727 }
 728
 729 #else
 730
 731 static int qemu_rbd_co_flush(BlockDriverState *bs)
 732 {
 733 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 734     /* rbd_flush added in 0.1.1 */
 735     BDRVRBDState *s = bs->opaque;
 736     return rbd_flush(s->image);
 737 #else
 738     return 0;
 739 #endif
 740 }
 741 #endif
 742
 743 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 744 {
 745     BDRVRBDState *s = bs->opaque;
 746     rbd_image_info_t info;
 747     int r;
 748
 749     r = rbd_stat(s->image, &info, sizeof(info));
 750     if (r < 0) {
 751         return r;
 752     }
 753
 754     bdi->cluster_size = info.obj_size;
 755     return 0;
 756 }
 757
 758 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 759 {
 760     BDRVRBDState *s = bs->opaque;
 761     rbd_image_info_t info;
 762     int r;
 763
 764     r = rbd_stat(s->image, &info, sizeof(info));
 765     if (r < 0) {
 766         return r;
 767     }
 768
 769     return info.size;
 770 }
 771
 772 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 773 {
 774     BDRVRBDState *s = bs->opaque;
 775     int r;
 776
 777     r = rbd_resize(s->image, offset);
 778     if (r < 0) {
 779         return r;
 780     }
 781
 782     return 0;
 783 }
 784
 785 static int qemu_rbd_snap_create(BlockDriverState *bs,
 786                                 QEMUSnapshotInfo *sn_info)
 787 {
 788     BDRVRBDState *s = bs->opaque;
 789     int r;
 790
 791     if (sn_info->name[0] == '\0') {
 792         return -EINVAL; /* we need a name for rbd snapshots */
 793     }
 794
 795     /*
 796      * rbd snapshots are using the name as the user controlled unique identifier
 797      * we can't use the rbd snapid for that purpose, as it can't be set
 798      */
 799     if (sn_info->id_str[0] != '\0' &&
 800         strcmp(sn_info->id_str, sn_info->name) != 0) {
 801         return -EINVAL;
 802     }
 803
 804     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 805         return -ERANGE;
 806     }
 807
 808     r = rbd_snap_create(s->image, sn_info->name);
 809     if (r < 0) {
 810         error_report("failed to create snap: %s", strerror(-r));
 811         return r;
 812     }
 813
 814     return 0;
 815 }
 816
 817 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 818                                 const char *snapshot_id,
 819                                 const char *snapshot_name,
 820                                 Error **errp)
 821 {
 822     BDRVRBDState *s = bs->opaque;
 823     int r;
 824
 825     if (!snapshot_name) {
 826         error_setg(errp, "rbd need a valid snapshot name");
 827         return -EINVAL;
 828     }
 829
 830     /* If snapshot_id is specified, it must be equal to name, see
 831        qemu_rbd_snap_list() */
 832     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 833         error_setg(errp,
 834                    "rbd do not support snapshot id, it should be NULL or "
 835                    "equal to snapshot name");
 836         return -EINVAL;
 837     }
 838
 839     r = rbd_snap_remove(s->image, snapshot_name);
 840     if (r < 0) {
 841         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 842     }
 843     return r;
 844 }
 845
 846 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 847                                   const char *snapshot_name)
 848 {
 849     BDRVRBDState *s = bs->opaque;
 850     int r;
 851
 852     r = rbd_snap_rollback(s->image, snapshot_name);
 853     return r;
 854 }
 855
 856 static int qemu_rbd_snap_list(BlockDriverState *bs,
 857                               QEMUSnapshotInfo **psn_tab)
 858 {
 859     BDRVRBDState *s = bs->opaque;
 860     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 861     int i, snap_count;
 862     rbd_snap_info_t *snaps;
 863     int max_snaps = RBD_MAX_SNAPS;
 864
 865     do {
 866         snaps = g_new(rbd_snap_info_t, max_snaps);
 867         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 868         if (snap_count <= 0) {
 869             g_free(snaps);
 870         }
 871     } while (snap_count == -ERANGE);
 872
 873     if (snap_count <= 0) {
 874         goto done;
 875     }
 876
 877     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
 878
 879     for (i = 0; i < snap_count; i++) {
 880         const char *snap_name = snaps[i].name;
 881
 882         sn_info = sn_tab + i;
 883         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 884         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 885
 886         sn_info->vm_state_size = snaps[i].size;
 887         sn_info->date_sec = 0;
 888         sn_info->date_nsec = 0;
 889         sn_info->vm_clock_nsec = 0;
 890     }
 891     rbd_snap_list_end(snaps);
 892     g_free(snaps);
 893
 894  done:
 895     *psn_tab = sn_tab;
 896     return snap_count;
 897 }
 898
 899 #ifdef LIBRBD_SUPPORTS_DISCARD
 900 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 901                                               int64_t sector_num,
 902                                               int nb_sectors,
 903                                               BlockDriverCompletionFunc *cb,
 904                                               void *opaque)
 905 {
 906     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 907                          RBD_AIO_DISCARD);
 908 }
 909 #endif
 910
 911 static QemuOptsList qemu_rbd_create_opts = {
 912     .name = "rbd-create-opts",
 913     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
 914     .desc = {
 915         {
 916             .name = BLOCK_OPT_SIZE,
 917             .type = QEMU_OPT_SIZE,
 918             .help = "Virtual disk size"
 919         },
 920         {
 921             .name = BLOCK_OPT_CLUSTER_SIZE,
 922             .type = QEMU_OPT_SIZE,
 923             .help = "RBD object size"
 924         },
 925         { /* end of list */ }
 926     }
 927 };
 928
 929 static BlockDriver bdrv_rbd = {
 930     .format_name        = "rbd",
 931     .instance_size      = sizeof(BDRVRBDState),
 932     .bdrv_needs_filename = true,
 933     .bdrv_file_open     = qemu_rbd_open,
 934     .bdrv_close         = qemu_rbd_close,
 935     .bdrv_create        = qemu_rbd_create,
 936     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 937     .bdrv_get_info      = qemu_rbd_getinfo,
 938     .create_opts        = &qemu_rbd_create_opts,
 939     .bdrv_getlength     = qemu_rbd_getlength,
 940     .bdrv_truncate      = qemu_rbd_truncate,
 941     .protocol_name      = "rbd",
 942
 943     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 944     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 945
 946 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 947     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 948 #else
 949     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 950 #endif
 951
 952 #ifdef LIBRBD_SUPPORTS_DISCARD
 953     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 954 #endif
 955
 956     .bdrv_snapshot_create   = qemu_rbd_snap_create,
 957     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
 958     .bdrv_snapshot_list     = qemu_rbd_snap_list,
 959     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
 960 };
 961
 962 static void bdrv_rbd_init(void)
 963 {
 964     bdrv_register(&bdrv_rbd);
 965 }
 966
 967 block_init(bdrv_rbd_init);