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.h"
  18 #include "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 } RBDAIOCmd;
  68
  69 typedef struct RBDAIOCB {
  70     BlockDriverAIOCB common;
  71     QEMUBH *bh;
  72     int64_t ret;
  73     QEMUIOVector *qiov;
  74     char *bounce;
  75     RBDAIOCmd cmd;
  76     int64_t sector_num;
  77     int error;
  78     struct BDRVRBDState *s;
  79     int cancelled;
  80 } RBDAIOCB;
  81
  82 typedef struct RADOSCB {
  83     int rcbid;
  84     RBDAIOCB *acb;
  85     struct BDRVRBDState *s;
  86     int done;
  87     int64_t size;
  88     char *buf;
  89     int64_t ret;
  90 } RADOSCB;
  91
  92 #define RBD_FD_READ 0
  93 #define RBD_FD_WRITE 1
  94
  95 typedef struct BDRVRBDState {
  96     int fds[2];
  97     rados_t cluster;
  98     rados_ioctx_t io_ctx;
  99     rbd_image_t image;
 100     char name[RBD_MAX_IMAGE_NAME_SIZE];
 101     int qemu_aio_count;
 102     char *snap;
 103     int event_reader_pos;
 104     RADOSCB *event_rcb;
 105 } BDRVRBDState;
 106
 107 static void rbd_aio_bh_cb(void *opaque);
 108
 109 static int qemu_rbd_next_tok(char *dst, int dst_len,
 110                              char *src, char delim,
 111                              const char *name,
 112                              char **p)
 113 {
 114     int l;
 115     char *end;
 116
 117     *p = NULL;
 118
 119     if (delim != '\0') {
 120         for (end = src; *end; ++end) {
 121             if (*end == delim) {
 122                 break;
 123             }
 124             if (*end == '\\' && end[1] != '\0') {
 125                 end++;
 126             }
 127         }
 128         if (*end == delim) {
 129             *p = end + 1;
 130             *end = '\0';
 131         }
 132     }
 133     l = strlen(src);
 134     if (l >= dst_len) {
 135         error_report("%s too long", name);
 136         return -EINVAL;
 137     } else if (l == 0) {
 138         error_report("%s too short", name);
 139         return -EINVAL;
 140     }
 141
 142     pstrcpy(dst, dst_len, src);
 143
 144     return 0;
 145 }
 146
 147 static void qemu_rbd_unescape(char *src)
 148 {
 149     char *p;
 150
 151     for (p = src; *src; ++src, ++p) {
 152         if (*src == '\\' && src[1] != '\0') {
 153             src++;
 154         }
 155         *p = *src;
 156     }
 157     *p = '\0';
 158 }
 159
 160 static int qemu_rbd_parsename(const char *filename,
 161                               char *pool, int pool_len,
 162                               char *snap, int snap_len,
 163                               char *name, int name_len,
 164                               char *conf, int conf_len)
 165 {
 166     const char *start;
 167     char *p, *buf;
 168     int ret;
 169
 170     if (!strstart(filename, "rbd:", &start)) {
 171         return -EINVAL;
 172     }
 173
 174     buf = g_strdup(start);
 175     p = buf;
 176     *snap = '\0';
 177     *conf = '\0';
 178
 179     ret = qemu_rbd_next_tok(pool, pool_len, p, '/', "pool name", &p);
 180     if (ret < 0 || !p) {
 181         ret = -EINVAL;
 182         goto done;
 183     }
 184     qemu_rbd_unescape(pool);
 185
 186     if (strchr(p, '@')) {
 187         ret = qemu_rbd_next_tok(name, name_len, p, '@', "object name", &p);
 188         if (ret < 0) {
 189             goto done;
 190         }
 191         ret = qemu_rbd_next_tok(snap, snap_len, p, ':', "snap name", &p);
 192         qemu_rbd_unescape(snap);
 193     } else {
 194         ret = qemu_rbd_next_tok(name, name_len, p, ':', "object name", &p);
 195     }
 196     qemu_rbd_unescape(name);
 197     if (ret < 0 || !p) {
 198         goto done;
 199     }
 200
 201     ret = qemu_rbd_next_tok(conf, conf_len, p, '\0', "configuration", &p);
 202
 203 done:
 204     g_free(buf);
 205     return ret;
 206 }
 207
 208 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 209 {
 210     const char *p = conf;
 211
 212     while (*p) {
 213         int len;
 214         const char *end = strchr(p, ':');
 215
 216         if (end) {
 217             len = end - p;
 218         } else {
 219             len = strlen(p);
 220         }
 221
 222         if (strncmp(p, "id=", 3) == 0) {
 223             len -= 3;
 224             strncpy(clientname, p + 3, len);
 225             clientname[len] = '\0';
 226             return clientname;
 227         }
 228         if (end == NULL) {
 229             break;
 230         }
 231         p = end + 1;
 232     }
 233     return NULL;
 234 }
 235
 236 static int qemu_rbd_set_conf(rados_t cluster, const char *conf)
 237 {
 238     char *p, *buf;
 239     char name[RBD_MAX_CONF_NAME_SIZE];
 240     char value[RBD_MAX_CONF_VAL_SIZE];
 241     int ret = 0;
 242
 243     buf = g_strdup(conf);
 244     p = buf;
 245
 246     while (p) {
 247         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 248                                 '=', "conf option name", &p);
 249         if (ret < 0) {
 250             break;
 251         }
 252         qemu_rbd_unescape(name);
 253
 254         if (!p) {
 255             error_report("conf option %s has no value", name);
 256             ret = -EINVAL;
 257             break;
 258         }
 259
 260         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 261                                 ':', "conf option value", &p);
 262         if (ret < 0) {
 263             break;
 264         }
 265         qemu_rbd_unescape(value);
 266
 267         if (strcmp(name, "conf") == 0) {
 268             ret = rados_conf_read_file(cluster, value);
 269             if (ret < 0) {
 270                 error_report("error reading conf file %s", value);
 271                 break;
 272             }
 273         } else if (strcmp(name, "id") == 0) {
 274             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 275         } else {
 276             ret = rados_conf_set(cluster, name, value);
 277             if (ret < 0) {
 278                 error_report("invalid conf option %s", name);
 279                 ret = -EINVAL;
 280                 break;
 281             }
 282         }
 283     }
 284
 285     g_free(buf);
 286     return ret;
 287 }
 288
 289 static int qemu_rbd_create(const char *filename, QEMUOptionParameter *options)
 290 {
 291     int64_t bytes = 0;
 292     int64_t objsize;
 293     int obj_order = 0;
 294     char pool[RBD_MAX_POOL_NAME_SIZE];
 295     char name[RBD_MAX_IMAGE_NAME_SIZE];
 296     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 297     char conf[RBD_MAX_CONF_SIZE];
 298     char clientname_buf[RBD_MAX_CONF_SIZE];
 299     char *clientname;
 300     rados_t cluster;
 301     rados_ioctx_t io_ctx;
 302     int ret;
 303
 304     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 305                            snap_buf, sizeof(snap_buf),
 306                            name, sizeof(name),
 307                            conf, sizeof(conf)) < 0) {
 308         return -EINVAL;
 309     }
 310
 311     /* Read out options */
 312     while (options && options->name) {
 313         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 314             bytes = options->value.n;
 315         } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) {
 316             if (options->value.n) {
 317                 objsize = options->value.n;
 318                 if ((objsize - 1) & objsize) {    /* not a power of 2? */
 319                     error_report("obj size needs to be power of 2");
 320                     return -EINVAL;
 321                 }
 322                 if (objsize < 4096) {
 323                     error_report("obj size too small");
 324                     return -EINVAL;
 325                 }
 326                 obj_order = ffs(objsize) - 1;
 327             }
 328         }
 329         options++;
 330     }
 331
 332     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 333     if (rados_create(&cluster, clientname) < 0) {
 334         error_report("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) < 0) {
 345         error_report("error setting config options");
 346         rados_shutdown(cluster);
 347         return -EIO;
 348     }
 349
 350     if (rados_connect(cluster) < 0) {
 351         error_report("error connecting");
 352         rados_shutdown(cluster);
 353         return -EIO;
 354     }
 355
 356     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 357         error_report("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 qemu_rbd_aio_event_reader()
 371  * and runs in qemu context. It schedules a bh, but just in case the aio
 372  * was not cancelled before.
 373  */
 374 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 375 {
 376     RBDAIOCB *acb = rcb->acb;
 377     int64_t r;
 378
 379     if (acb->cancelled) {
 380         qemu_vfree(acb->bounce);
 381         qemu_aio_release(acb);
 382         goto done;
 383     }
 384
 385     r = rcb->ret;
 386
 387     if (acb->cmd == RBD_AIO_WRITE ||
 388         acb->cmd == RBD_AIO_DISCARD) {
 389         if (r < 0) {
 390             acb->ret = r;
 391             acb->error = 1;
 392         } else if (!acb->error) {
 393             acb->ret = rcb->size;
 394         }
 395     } else {
 396         if (r < 0) {
 397             memset(rcb->buf, 0, rcb->size);
 398             acb->ret = r;
 399             acb->error = 1;
 400         } else if (r < rcb->size) {
 401             memset(rcb->buf + r, 0, rcb->size - r);
 402             if (!acb->error) {
 403                 acb->ret = rcb->size;
 404             }
 405         } else if (!acb->error) {
 406             acb->ret = r;
 407         }
 408     }
 409     /* Note that acb->bh can be NULL in case where the aio was cancelled */
 410     acb->bh = qemu_bh_new(rbd_aio_bh_cb, acb);
 411     qemu_bh_schedule(acb->bh);
 412 done:
 413     g_free(rcb);
 414 }
 415
 416 /*
 417  * aio fd read handler. It runs in the qemu context and calls the
 418  * completion handling of completed rados aio operations.
 419  */
 420 static void qemu_rbd_aio_event_reader(void *opaque)
 421 {
 422     BDRVRBDState *s = opaque;
 423
 424     ssize_t ret;
 425
 426     do {
 427         char *p = (char *)&s->event_rcb;
 428
 429         /* now read the rcb pointer that was sent from a non qemu thread */
 430         ret = read(s->fds[RBD_FD_READ], p + s->event_reader_pos,
 431                    sizeof(s->event_rcb) - s->event_reader_pos);
 432         if (ret > 0) {
 433             s->event_reader_pos += ret;
 434             if (s->event_reader_pos == sizeof(s->event_rcb)) {
 435                 s->event_reader_pos = 0;
 436                 qemu_rbd_complete_aio(s->event_rcb);
 437                 s->qemu_aio_count--;
 438             }
 439         }
 440     } while (ret < 0 && errno == EINTR);
 441 }
 442
 443 static int qemu_rbd_aio_flush_cb(void *opaque)
 444 {
 445     BDRVRBDState *s = opaque;
 446
 447     return (s->qemu_aio_count > 0);
 448 }
 449
 450 static int qemu_rbd_open(BlockDriverState *bs, const char *filename, int flags)
 451 {
 452     BDRVRBDState *s = bs->opaque;
 453     char pool[RBD_MAX_POOL_NAME_SIZE];
 454     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 455     char conf[RBD_MAX_CONF_SIZE];
 456     char clientname_buf[RBD_MAX_CONF_SIZE];
 457     char *clientname;
 458     int r;
 459
 460     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 461                            snap_buf, sizeof(snap_buf),
 462                            s->name, sizeof(s->name),
 463                            conf, sizeof(conf)) < 0) {
 464         return -EINVAL;
 465     }
 466
 467     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 468     r = rados_create(&s->cluster, clientname);
 469     if (r < 0) {
 470         error_report("error initializing");
 471         return r;
 472     }
 473
 474     s->snap = NULL;
 475     if (snap_buf[0] != '\0') {
 476         s->snap = g_strdup(snap_buf);
 477     }
 478
 479     /*
 480      * Fallback to more conservative semantics if setting cache
 481      * options fails. Ignore errors from setting rbd_cache because the
 482      * only possible error is that the option does not exist, and
 483      * librbd defaults to no caching. If write through caching cannot
 484      * be set up, fall back to no caching.
 485      */
 486     if (flags & BDRV_O_NOCACHE) {
 487         rados_conf_set(s->cluster, "rbd_cache", "false");
 488     } else {
 489         rados_conf_set(s->cluster, "rbd_cache", "true");
 490     }
 491
 492     if (strstr(conf, "conf=") == NULL) {
 493         /* try default location, but ignore failure */
 494         rados_conf_read_file(s->cluster, NULL);
 495     }
 496
 497     if (conf[0] != '\0') {
 498         r = qemu_rbd_set_conf(s->cluster, conf);
 499         if (r < 0) {
 500             error_report("error setting config options");
 501             goto failed_shutdown;
 502         }
 503     }
 504
 505     r = rados_connect(s->cluster);
 506     if (r < 0) {
 507         error_report("error connecting");
 508         goto failed_shutdown;
 509     }
 510
 511     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 512     if (r < 0) {
 513         error_report("error opening pool %s", pool);
 514         goto failed_shutdown;
 515     }
 516
 517     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 518     if (r < 0) {
 519         error_report("error reading header from %s", s->name);
 520         goto failed_open;
 521     }
 522
 523     bs->read_only = (s->snap != NULL);
 524
 525     s->event_reader_pos = 0;
 526     r = qemu_pipe(s->fds);
 527     if (r < 0) {
 528         error_report("error opening eventfd");
 529         goto failed;
 530     }
 531     fcntl(s->fds[0], F_SETFL, O_NONBLOCK);
 532     fcntl(s->fds[1], F_SETFL, O_NONBLOCK);
 533     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], qemu_rbd_aio_event_reader,
 534                             NULL, qemu_rbd_aio_flush_cb, s);
 535
 536
 537     return 0;
 538
 539 failed:
 540     rbd_close(s->image);
 541 failed_open:
 542     rados_ioctx_destroy(s->io_ctx);
 543 failed_shutdown:
 544     rados_shutdown(s->cluster);
 545     g_free(s->snap);
 546     return r;
 547 }
 548
 549 static void qemu_rbd_close(BlockDriverState *bs)
 550 {
 551     BDRVRBDState *s = bs->opaque;
 552
 553     close(s->fds[0]);
 554     close(s->fds[1]);
 555     qemu_aio_set_fd_handler(s->fds[RBD_FD_READ], NULL, NULL, NULL, NULL);
 556
 557     rbd_close(s->image);
 558     rados_ioctx_destroy(s->io_ctx);
 559     g_free(s->snap);
 560     rados_shutdown(s->cluster);
 561 }
 562
 563 /*
 564  * Cancel aio. Since we don't reference acb in a non qemu threads,
 565  * it is safe to access it here.
 566  */
 567 static void qemu_rbd_aio_cancel(BlockDriverAIOCB *blockacb)
 568 {
 569     RBDAIOCB *acb = (RBDAIOCB *) blockacb;
 570     acb->cancelled = 1;
 571 }
 572
 573 static const AIOCBInfo rbd_aiocb_info = {
 574     .aiocb_size = sizeof(RBDAIOCB),
 575     .cancel = qemu_rbd_aio_cancel,
 576 };
 577
 578 static int qemu_rbd_send_pipe(BDRVRBDState *s, RADOSCB *rcb)
 579 {
 580     int ret = 0;
 581     while (1) {
 582         fd_set wfd;
 583         int fd = s->fds[RBD_FD_WRITE];
 584
 585         /* send the op pointer to the qemu thread that is responsible
 586            for the aio/op completion. Must do it in a qemu thread context */
 587         ret = write(fd, (void *)&rcb, sizeof(rcb));
 588         if (ret >= 0) {
 589             break;
 590         }
 591         if (errno == EINTR) {
 592             continue;
 593         }
 594         if (errno != EAGAIN) {
 595             break;
 596         }
 597
 598         FD_ZERO(&wfd);
 599         FD_SET(fd, &wfd);
 600         do {
 601             ret = select(fd + 1, NULL, &wfd, NULL, NULL);
 602         } while (ret < 0 && errno == EINTR);
 603     }
 604
 605     return ret;
 606 }
 607
 608 /*
 609  * This is the callback function for rbd_aio_read and _write
 610  *
 611  * Note: this function is being called from a non qemu thread so
 612  * we need to be careful about what we do here. Generally we only
 613  * write to the block notification pipe, and do the rest of the
 614  * io completion handling from qemu_rbd_aio_event_reader() which
 615  * runs in a qemu context.
 616  */
 617 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 618 {
 619     int ret;
 620     rcb->ret = rbd_aio_get_return_value(c);
 621     rbd_aio_release(c);
 622     ret = qemu_rbd_send_pipe(rcb->s, rcb);
 623     if (ret < 0) {
 624         error_report("failed writing to acb->s->fds");
 625         g_free(rcb);
 626     }
 627 }
 628
 629 /* Callback when all queued rbd_aio requests are complete */
 630
 631 static void rbd_aio_bh_cb(void *opaque)
 632 {
 633     RBDAIOCB *acb = opaque;
 634
 635     if (acb->cmd == RBD_AIO_READ) {
 636         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 637     }
 638     qemu_vfree(acb->bounce);
 639     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 640     qemu_bh_delete(acb->bh);
 641     acb->bh = NULL;
 642
 643     qemu_aio_release(acb);
 644 }
 645
 646 static int rbd_aio_discard_wrapper(rbd_image_t image,
 647                                    uint64_t off,
 648                                    uint64_t len,
 649                                    rbd_completion_t comp)
 650 {
 651 #ifdef LIBRBD_SUPPORTS_DISCARD
 652     return rbd_aio_discard(image, off, len, comp);
 653 #else
 654     return -ENOTSUP;
 655 #endif
 656 }
 657
 658 static BlockDriverAIOCB *rbd_start_aio(BlockDriverState *bs,
 659                                        int64_t sector_num,
 660                                        QEMUIOVector *qiov,
 661                                        int nb_sectors,
 662                                        BlockDriverCompletionFunc *cb,
 663                                        void *opaque,
 664                                        RBDAIOCmd cmd)
 665 {
 666     RBDAIOCB *acb;
 667     RADOSCB *rcb;
 668     rbd_completion_t c;
 669     int64_t off, size;
 670     char *buf;
 671     int r;
 672
 673     BDRVRBDState *s = bs->opaque;
 674
 675     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 676     acb->cmd = cmd;
 677     acb->qiov = qiov;
 678     if (cmd == RBD_AIO_DISCARD) {
 679         acb->bounce = NULL;
 680     } else {
 681         acb->bounce = qemu_blockalign(bs, qiov->size);
 682     }
 683     acb->ret = 0;
 684     acb->error = 0;
 685     acb->s = s;
 686     acb->cancelled = 0;
 687     acb->bh = NULL;
 688
 689     if (cmd == RBD_AIO_WRITE) {
 690         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 691     }
 692
 693     buf = acb->bounce;
 694
 695     off = sector_num * BDRV_SECTOR_SIZE;
 696     size = nb_sectors * BDRV_SECTOR_SIZE;
 697
 698     s->qemu_aio_count++; /* All the RADOSCB */
 699
 700     rcb = g_malloc(sizeof(RADOSCB));
 701     rcb->done = 0;
 702     rcb->acb = acb;
 703     rcb->buf = buf;
 704     rcb->s = acb->s;
 705     rcb->size = size;
 706     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 707     if (r < 0) {
 708         goto failed;
 709     }
 710
 711     switch (cmd) {
 712     case RBD_AIO_WRITE:
 713         r = rbd_aio_write(s->image, off, size, buf, c);
 714         break;
 715     case RBD_AIO_READ:
 716         r = rbd_aio_read(s->image, off, size, buf, c);
 717         break;
 718     case RBD_AIO_DISCARD:
 719         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 720         break;
 721     default:
 722         r = -EINVAL;
 723     }
 724
 725     if (r < 0) {
 726         goto failed;
 727     }
 728
 729     return &acb->common;
 730
 731 failed:
 732     g_free(rcb);
 733     s->qemu_aio_count--;
 734     qemu_aio_release(acb);
 735     return NULL;
 736 }
 737
 738 static BlockDriverAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 739                                             int64_t sector_num,
 740                                             QEMUIOVector *qiov,
 741                                             int nb_sectors,
 742                                             BlockDriverCompletionFunc *cb,
 743                                             void *opaque)
 744 {
 745     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 746                          RBD_AIO_READ);
 747 }
 748
 749 static BlockDriverAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 750                                              int64_t sector_num,
 751                                              QEMUIOVector *qiov,
 752                                              int nb_sectors,
 753                                              BlockDriverCompletionFunc *cb,
 754                                              void *opaque)
 755 {
 756     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 757                          RBD_AIO_WRITE);
 758 }
 759
 760 static int qemu_rbd_co_flush(BlockDriverState *bs)
 761 {
 762 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 763     /* rbd_flush added in 0.1.1 */
 764     BDRVRBDState *s = bs->opaque;
 765     return rbd_flush(s->image);
 766 #else
 767     return 0;
 768 #endif
 769 }
 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_name)
 847 {
 848     BDRVRBDState *s = bs->opaque;
 849     int r;
 850
 851     r = rbd_snap_remove(s->image, snapshot_name);
 852     return r;
 853 }
 854
 855 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 856                                   const char *snapshot_name)
 857 {
 858     BDRVRBDState *s = bs->opaque;
 859     int r;
 860
 861     r = rbd_snap_rollback(s->image, snapshot_name);
 862     return r;
 863 }
 864
 865 static int qemu_rbd_snap_list(BlockDriverState *bs,
 866                               QEMUSnapshotInfo **psn_tab)
 867 {
 868     BDRVRBDState *s = bs->opaque;
 869     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 870     int i, snap_count;
 871     rbd_snap_info_t *snaps;
 872     int max_snaps = RBD_MAX_SNAPS;
 873
 874     do {
 875         snaps = g_malloc(sizeof(*snaps) * max_snaps);
 876         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 877         if (snap_count < 0) {
 878             g_free(snaps);
 879         }
 880     } while (snap_count == -ERANGE);
 881
 882     if (snap_count <= 0) {
 883         goto done;
 884     }
 885
 886     sn_tab = g_malloc0(snap_count * sizeof(QEMUSnapshotInfo));
 887
 888     for (i = 0; i < snap_count; i++) {
 889         const char *snap_name = snaps[i].name;
 890
 891         sn_info = sn_tab + i;
 892         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 893         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 894
 895         sn_info->vm_state_size = snaps[i].size;
 896         sn_info->date_sec = 0;
 897         sn_info->date_nsec = 0;
 898         sn_info->vm_clock_nsec = 0;
 899     }
 900     rbd_snap_list_end(snaps);
 901
 902  done:
 903     *psn_tab = sn_tab;
 904     return snap_count;
 905 }
 906
 907 #ifdef LIBRBD_SUPPORTS_DISCARD
 908 static BlockDriverAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 909                                               int64_t sector_num,
 910                                               int nb_sectors,
 911                                               BlockDriverCompletionFunc *cb,
 912                                               void *opaque)
 913 {
 914     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 915                          RBD_AIO_DISCARD);
 916 }
 917 #endif
 918
 919 static QEMUOptionParameter qemu_rbd_create_options[] = {
 920     {
 921      .name = BLOCK_OPT_SIZE,
 922      .type = OPT_SIZE,
 923      .help = "Virtual disk size"
 924     },
 925     {
 926      .name = BLOCK_OPT_CLUSTER_SIZE,
 927      .type = OPT_SIZE,
 928      .help = "RBD object size"
 929     },
 930     {NULL}
 931 };
 932
 933 static BlockDriver bdrv_rbd = {
 934     .format_name        = "rbd",
 935     .instance_size      = sizeof(BDRVRBDState),
 936     .bdrv_file_open     = qemu_rbd_open,
 937     .bdrv_close         = qemu_rbd_close,
 938     .bdrv_create        = qemu_rbd_create,
 939     .bdrv_get_info      = qemu_rbd_getinfo,
 940     .create_options     = qemu_rbd_create_options,
 941     .bdrv_getlength     = qemu_rbd_getlength,
 942     .bdrv_truncate      = qemu_rbd_truncate,
 943     .protocol_name      = "rbd",
 944
 945     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 946     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 947     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 948
 949 #ifdef LIBRBD_SUPPORTS_DISCARD
 950     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 951 #endif
 952
 953     .bdrv_snapshot_create   = qemu_rbd_snap_create,
 954     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
 955     .bdrv_snapshot_list     = qemu_rbd_snap_list,
 956     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
 957 };
 958
 959 static void bdrv_rbd_init(void)
 960 {
 961     bdrv_register(&bdrv_rbd);
 962 }
 963
 964 block_init(bdrv_rbd_init);