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     int error;
  78     struct BDRVRBDState *s;
  79 } RBDAIOCB;
  80
  81 typedef struct RADOSCB {
  82     RBDAIOCB *acb;
  83     struct BDRVRBDState *s;
  84     int64_t size;
  85     char *buf;
  86     int64_t ret;
  87 } RADOSCB;
  88
  89 typedef struct BDRVRBDState {
  90     rados_t cluster;
  91     rados_ioctx_t io_ctx;
  92     rbd_image_t image;
  93     char name[RBD_MAX_IMAGE_NAME_SIZE];
  94     char *snap;
  95 } BDRVRBDState;
  96
  97 static int qemu_rbd_next_tok(char *dst, int dst_len,
  98                              char *src, char delim,
  99                              const char *name,
 100                              char **p, Error **errp)
 101 {
 102     int l;
 103     char *end;
 104
 105     *p = NULL;
 106
 107     if (delim != '\0') {
 108         for (end = src; *end; ++end) {
 109             if (*end == delim) {
 110                 break;
 111             }
 112             if (*end == '\\' && end[1] != '\0') {
 113                 end++;
 114             }
 115         }
 116         if (*end == delim) {
 117             *p = end + 1;
 118             *end = '\0';
 119         }
 120     }
 121     l = strlen(src);
 122     if (l >= dst_len) {
 123         error_setg(errp, "%s too long", name);
 124         return -EINVAL;
 125     } else if (l == 0) {
 126         error_setg(errp, "%s too short", name);
 127         return -EINVAL;
 128     }
 129
 130     pstrcpy(dst, dst_len, src);
 131
 132     return 0;
 133 }
 134
 135 static void qemu_rbd_unescape(char *src)
 136 {
 137     char *p;
 138
 139     for (p = src; *src; ++src, ++p) {
 140         if (*src == '\\' && src[1] != '\0') {
 141             src++;
 142         }
 143         *p = *src;
 144     }
 145     *p = '\0';
 146 }
 147
 148 static int qemu_rbd_parsename(const char *filename,
 149                               char *pool, int pool_len,
 150                               char *snap, int snap_len,
 151                               char *name, int name_len,
 152                               char *conf, int conf_len,
 153                               Error **errp)
 154 {
 155     const char *start;
 156     char *p, *buf;
 157     int ret;
 158
 159     if (!strstart(filename, "rbd:", &start)) {
 160         error_setg(errp, "File name must start with 'rbd:'");
 161         return -EINVAL;
 162     }
 163
 164     buf = g_strdup(start);
 165     p = buf;
 166     *snap = '\0';
 167     *conf = '\0';
 168
 169     ret = qemu_rbd_next_tok(pool, pool_len, p,
 170                             '/', "pool name", &p, errp);
 171     if (ret < 0 || !p) {
 172         ret = -EINVAL;
 173         goto done;
 174     }
 175     qemu_rbd_unescape(pool);
 176
 177     if (strchr(p, '@')) {
 178         ret = qemu_rbd_next_tok(name, name_len, p,
 179                                 '@', "object name", &p, errp);
 180         if (ret < 0) {
 181             goto done;
 182         }
 183         ret = qemu_rbd_next_tok(snap, snap_len, p,
 184                                 ':', "snap name", &p, errp);
 185         qemu_rbd_unescape(snap);
 186     } else {
 187         ret = qemu_rbd_next_tok(name, name_len, p,
 188                                 ':', "object name", &p, errp);
 189     }
 190     qemu_rbd_unescape(name);
 191     if (ret < 0 || !p) {
 192         goto done;
 193     }
 194
 195     ret = qemu_rbd_next_tok(conf, conf_len, p,
 196                             '\0', "configuration", &p, errp);
 197
 198 done:
 199     g_free(buf);
 200     return ret;
 201 }
 202
 203 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 204 {
 205     const char *p = conf;
 206
 207     while (*p) {
 208         int len;
 209         const char *end = strchr(p, ':');
 210
 211         if (end) {
 212             len = end - p;
 213         } else {
 214             len = strlen(p);
 215         }
 216
 217         if (strncmp(p, "id=", 3) == 0) {
 218             len -= 3;
 219             strncpy(clientname, p + 3, len);
 220             clientname[len] = '\0';
 221             return clientname;
 222         }
 223         if (end == NULL) {
 224             break;
 225         }
 226         p = end + 1;
 227     }
 228     return NULL;
 229 }
 230
 231 static int qemu_rbd_set_conf(rados_t cluster, const char *conf, Error **errp)
 232 {
 233     char *p, *buf;
 234     char name[RBD_MAX_CONF_NAME_SIZE];
 235     char value[RBD_MAX_CONF_VAL_SIZE];
 236     int ret = 0;
 237
 238     buf = g_strdup(conf);
 239     p = buf;
 240
 241     while (p) {
 242         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 243                                 '=', "conf option name", &p, errp);
 244         if (ret < 0) {
 245             break;
 246         }
 247         qemu_rbd_unescape(name);
 248
 249         if (!p) {
 250             error_setg(errp, "conf option %s has no value", name);
 251             ret = -EINVAL;
 252             break;
 253         }
 254
 255         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 256                                 ':', "conf option value", &p, errp);
 257         if (ret < 0) {
 258             break;
 259         }
 260         qemu_rbd_unescape(value);
 261
 262         if (strcmp(name, "conf") == 0) {
 263             ret = rados_conf_read_file(cluster, value);
 264             if (ret < 0) {
 265                 error_setg(errp, "error reading conf file %s", value);
 266                 break;
 267             }
 268         } else if (strcmp(name, "id") == 0) {
 269             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 270         } else {
 271             ret = rados_conf_set(cluster, name, value);
 272             if (ret < 0) {
 273                 error_setg(errp, "invalid conf option %s", name);
 274                 ret = -EINVAL;
 275                 break;
 276             }
 277         }
 278     }
 279
 280     g_free(buf);
 281     return ret;
 282 }
 283
 284 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 285 {
 286     Error *local_err = NULL;
 287     int64_t bytes = 0;
 288     int64_t objsize;
 289     int obj_order = 0;
 290     char pool[RBD_MAX_POOL_NAME_SIZE];
 291     char name[RBD_MAX_IMAGE_NAME_SIZE];
 292     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 293     char conf[RBD_MAX_CONF_SIZE];
 294     char clientname_buf[RBD_MAX_CONF_SIZE];
 295     char *clientname;
 296     rados_t cluster;
 297     rados_ioctx_t io_ctx;
 298     int ret;
 299
 300     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 301                            snap_buf, sizeof(snap_buf),
 302                            name, sizeof(name),
 303                            conf, sizeof(conf), &local_err) < 0) {
 304         error_propagate(errp, local_err);
 305         return -EINVAL;
 306     }
 307
 308     /* Read out options */
 309     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 310                      BDRV_SECTOR_SIZE);
 311     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 312     if (objsize) {
 313         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 314             error_setg(errp, "obj size needs to be power of 2");
 315             return -EINVAL;
 316         }
 317         if (objsize < 4096) {
 318             error_setg(errp, "obj size too small");
 319             return -EINVAL;
 320         }
 321         obj_order = ctz32(objsize);
 322     }
 323
 324     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 325     if (rados_create(&cluster, clientname) < 0) {
 326         error_setg(errp, "error initializing");
 327         return -EIO;
 328     }
 329
 330     if (strstr(conf, "conf=") == NULL) {
 331         /* try default location, but ignore failure */
 332         rados_conf_read_file(cluster, NULL);
 333     }
 334
 335     if (conf[0] != '\0' &&
 336         qemu_rbd_set_conf(cluster, conf, &local_err) < 0) {
 337         rados_shutdown(cluster);
 338         error_propagate(errp, local_err);
 339         return -EIO;
 340     }
 341
 342     if (rados_connect(cluster) < 0) {
 343         error_setg(errp, "error connecting");
 344         rados_shutdown(cluster);
 345         return -EIO;
 346     }
 347
 348     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 349         error_setg(errp, "error opening pool %s", pool);
 350         rados_shutdown(cluster);
 351         return -EIO;
 352     }
 353
 354     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 355     rados_ioctx_destroy(io_ctx);
 356     rados_shutdown(cluster);
 357
 358     return ret;
 359 }
 360
 361 /*
 362  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 363  * BH context.
 364  */
 365 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 366 {
 367     RBDAIOCB *acb = rcb->acb;
 368     int64_t r;
 369
 370     r = rcb->ret;
 371
 372     if (acb->cmd != RBD_AIO_READ) {
 373         if (r < 0) {
 374             acb->ret = r;
 375             acb->error = 1;
 376         } else if (!acb->error) {
 377             acb->ret = rcb->size;
 378         }
 379     } else {
 380         if (r < 0) {
 381             memset(rcb->buf, 0, rcb->size);
 382             acb->ret = r;
 383             acb->error = 1;
 384         } else if (r < rcb->size) {
 385             memset(rcb->buf + r, 0, rcb->size - r);
 386             if (!acb->error) {
 387                 acb->ret = rcb->size;
 388             }
 389         } else if (!acb->error) {
 390             acb->ret = r;
 391         }
 392     }
 393
 394     g_free(rcb);
 395
 396     if (acb->cmd == RBD_AIO_READ) {
 397         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 398     }
 399     qemu_vfree(acb->bounce);
 400     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 401
 402     qemu_aio_unref(acb);
 403 }
 404
 405 /* TODO Convert to fine grained options */
 406 static QemuOptsList runtime_opts = {
 407     .name = "rbd",
 408     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 409     .desc = {
 410         {
 411             .name = "filename",
 412             .type = QEMU_OPT_STRING,
 413             .help = "Specification of the rbd image",
 414         },
 415         { /* end of list */ }
 416     },
 417 };
 418
 419 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 420                          Error **errp)
 421 {
 422     BDRVRBDState *s = bs->opaque;
 423     char pool[RBD_MAX_POOL_NAME_SIZE];
 424     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 425     char conf[RBD_MAX_CONF_SIZE];
 426     char clientname_buf[RBD_MAX_CONF_SIZE];
 427     char *clientname;
 428     QemuOpts *opts;
 429     Error *local_err = NULL;
 430     const char *filename;
 431     int r;
 432
 433     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 434     qemu_opts_absorb_qdict(opts, options, &local_err);
 435     if (local_err) {
 436         error_propagate(errp, local_err);
 437         qemu_opts_del(opts);
 438         return -EINVAL;
 439     }
 440
 441     filename = qemu_opt_get(opts, "filename");
 442
 443     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 444                            snap_buf, sizeof(snap_buf),
 445                            s->name, sizeof(s->name),
 446                            conf, sizeof(conf), errp) < 0) {
 447         r = -EINVAL;
 448         goto failed_opts;
 449     }
 450
 451     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 452     r = rados_create(&s->cluster, clientname);
 453     if (r < 0) {
 454         error_setg(errp, "error initializing");
 455         goto failed_opts;
 456     }
 457
 458     s->snap = NULL;
 459     if (snap_buf[0] != '\0') {
 460         s->snap = g_strdup(snap_buf);
 461     }
 462
 463     if (strstr(conf, "conf=") == NULL) {
 464         /* try default location, but ignore failure */
 465         rados_conf_read_file(s->cluster, NULL);
 466     }
 467
 468     if (conf[0] != '\0') {
 469         r = qemu_rbd_set_conf(s->cluster, conf, errp);
 470         if (r < 0) {
 471             goto failed_shutdown;
 472         }
 473     }
 474
 475     /*
 476      * Fallback to more conservative semantics if setting cache
 477      * options fails. Ignore errors from setting rbd_cache because the
 478      * only possible error is that the option does not exist, and
 479      * librbd defaults to no caching. If write through caching cannot
 480      * be set up, fall back to no caching.
 481      */
 482     if (flags & BDRV_O_NOCACHE) {
 483         rados_conf_set(s->cluster, "rbd_cache", "false");
 484     } else {
 485         rados_conf_set(s->cluster, "rbd_cache", "true");
 486     }
 487
 488     r = rados_connect(s->cluster);
 489     if (r < 0) {
 490         error_setg(errp, "error connecting");
 491         goto failed_shutdown;
 492     }
 493
 494     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 495     if (r < 0) {
 496         error_setg(errp, "error opening pool %s", pool);
 497         goto failed_shutdown;
 498     }
 499
 500     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 501     if (r < 0) {
 502         error_setg(errp, "error reading header from %s", s->name);
 503         goto failed_open;
 504     }
 505
 506     bs->read_only = (s->snap != NULL);
 507
 508     qemu_opts_del(opts);
 509     return 0;
 510
 511 failed_open:
 512     rados_ioctx_destroy(s->io_ctx);
 513 failed_shutdown:
 514     rados_shutdown(s->cluster);
 515     g_free(s->snap);
 516 failed_opts:
 517     qemu_opts_del(opts);
 518     return r;
 519 }
 520
 521 static void qemu_rbd_close(BlockDriverState *bs)
 522 {
 523     BDRVRBDState *s = bs->opaque;
 524
 525     rbd_close(s->image);
 526     rados_ioctx_destroy(s->io_ctx);
 527     g_free(s->snap);
 528     rados_shutdown(s->cluster);
 529 }
 530
 531 static const AIOCBInfo rbd_aiocb_info = {
 532     .aiocb_size = sizeof(RBDAIOCB),
 533 };
 534
 535 static void rbd_finish_bh(void *opaque)
 536 {
 537     RADOSCB *rcb = opaque;
 538     qemu_bh_delete(rcb->acb->bh);
 539     qemu_rbd_complete_aio(rcb);
 540 }
 541
 542 /*
 543  * This is the callback function for rbd_aio_read and _write
 544  *
 545  * Note: this function is being called from a non qemu thread so
 546  * we need to be careful about what we do here. Generally we only
 547  * schedule a BH, and do the rest of the io completion handling
 548  * from rbd_finish_bh() which runs in a qemu context.
 549  */
 550 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 551 {
 552     RBDAIOCB *acb = rcb->acb;
 553
 554     rcb->ret = rbd_aio_get_return_value(c);
 555     rbd_aio_release(c);
 556
 557     acb->bh = aio_bh_new(bdrv_get_aio_context(acb->common.bs),
 558                          rbd_finish_bh, rcb);
 559     qemu_bh_schedule(acb->bh);
 560 }
 561
 562 static int rbd_aio_discard_wrapper(rbd_image_t image,
 563                                    uint64_t off,
 564                                    uint64_t len,
 565                                    rbd_completion_t comp)
 566 {
 567 #ifdef LIBRBD_SUPPORTS_DISCARD
 568     return rbd_aio_discard(image, off, len, comp);
 569 #else
 570     return -ENOTSUP;
 571 #endif
 572 }
 573
 574 static int rbd_aio_flush_wrapper(rbd_image_t image,
 575                                  rbd_completion_t comp)
 576 {
 577 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 578     return rbd_aio_flush(image, comp);
 579 #else
 580     return -ENOTSUP;
 581 #endif
 582 }
 583
 584 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 585                                  int64_t sector_num,
 586                                  QEMUIOVector *qiov,
 587                                  int nb_sectors,
 588                                  BlockCompletionFunc *cb,
 589                                  void *opaque,
 590                                  RBDAIOCmd cmd)
 591 {
 592     RBDAIOCB *acb;
 593     RADOSCB *rcb = NULL;
 594     rbd_completion_t c;
 595     int64_t off, size;
 596     char *buf;
 597     int r;
 598
 599     BDRVRBDState *s = bs->opaque;
 600
 601     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 602     acb->cmd = cmd;
 603     acb->qiov = qiov;
 604     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 605         acb->bounce = NULL;
 606     } else {
 607         acb->bounce = qemu_try_blockalign(bs, qiov->size);
 608         if (acb->bounce == NULL) {
 609             goto failed;
 610         }
 611     }
 612     acb->ret = 0;
 613     acb->error = 0;
 614     acb->s = s;
 615     acb->bh = NULL;
 616
 617     if (cmd == RBD_AIO_WRITE) {
 618         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 619     }
 620
 621     buf = acb->bounce;
 622
 623     off = sector_num * BDRV_SECTOR_SIZE;
 624     size = nb_sectors * BDRV_SECTOR_SIZE;
 625
 626     rcb = g_new(RADOSCB, 1);
 627     rcb->acb = acb;
 628     rcb->buf = buf;
 629     rcb->s = acb->s;
 630     rcb->size = size;
 631     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 632     if (r < 0) {
 633         goto failed;
 634     }
 635
 636     switch (cmd) {
 637     case RBD_AIO_WRITE:
 638         r = rbd_aio_write(s->image, off, size, buf, c);
 639         break;
 640     case RBD_AIO_READ:
 641         r = rbd_aio_read(s->image, off, size, buf, c);
 642         break;
 643     case RBD_AIO_DISCARD:
 644         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 645         break;
 646     case RBD_AIO_FLUSH:
 647         r = rbd_aio_flush_wrapper(s->image, c);
 648         break;
 649     default:
 650         r = -EINVAL;
 651     }
 652
 653     if (r < 0) {
 654         goto failed_completion;
 655     }
 656
 657     return &acb->common;
 658
 659 failed_completion:
 660     rbd_aio_release(c);
 661 failed:
 662     g_free(rcb);
 663     qemu_vfree(acb->bounce);
 664     qemu_aio_unref(acb);
 665     return NULL;
 666 }
 667
 668 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 669                                       int64_t sector_num,
 670                                       QEMUIOVector *qiov,
 671                                       int nb_sectors,
 672                                       BlockCompletionFunc *cb,
 673                                       void *opaque)
 674 {
 675     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 676                          RBD_AIO_READ);
 677 }
 678
 679 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 680                                        int64_t sector_num,
 681                                        QEMUIOVector *qiov,
 682                                        int nb_sectors,
 683                                        BlockCompletionFunc *cb,
 684                                        void *opaque)
 685 {
 686     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 687                          RBD_AIO_WRITE);
 688 }
 689
 690 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 691 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 692                                       BlockCompletionFunc *cb,
 693                                       void *opaque)
 694 {
 695     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 696 }
 697
 698 #else
 699
 700 static int qemu_rbd_co_flush(BlockDriverState *bs)
 701 {
 702 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 703     /* rbd_flush added in 0.1.1 */
 704     BDRVRBDState *s = bs->opaque;
 705     return rbd_flush(s->image);
 706 #else
 707     return 0;
 708 #endif
 709 }
 710 #endif
 711
 712 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 713 {
 714     BDRVRBDState *s = bs->opaque;
 715     rbd_image_info_t info;
 716     int r;
 717
 718     r = rbd_stat(s->image, &info, sizeof(info));
 719     if (r < 0) {
 720         return r;
 721     }
 722
 723     bdi->cluster_size = info.obj_size;
 724     return 0;
 725 }
 726
 727 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 728 {
 729     BDRVRBDState *s = bs->opaque;
 730     rbd_image_info_t info;
 731     int r;
 732
 733     r = rbd_stat(s->image, &info, sizeof(info));
 734     if (r < 0) {
 735         return r;
 736     }
 737
 738     return info.size;
 739 }
 740
 741 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 742 {
 743     BDRVRBDState *s = bs->opaque;
 744     int r;
 745
 746     r = rbd_resize(s->image, offset);
 747     if (r < 0) {
 748         return r;
 749     }
 750
 751     return 0;
 752 }
 753
 754 static int qemu_rbd_snap_create(BlockDriverState *bs,
 755                                 QEMUSnapshotInfo *sn_info)
 756 {
 757     BDRVRBDState *s = bs->opaque;
 758     int r;
 759
 760     if (sn_info->name[0] == '\0') {
 761         return -EINVAL; /* we need a name for rbd snapshots */
 762     }
 763
 764     /*
 765      * rbd snapshots are using the name as the user controlled unique identifier
 766      * we can't use the rbd snapid for that purpose, as it can't be set
 767      */
 768     if (sn_info->id_str[0] != '\0' &&
 769         strcmp(sn_info->id_str, sn_info->name) != 0) {
 770         return -EINVAL;
 771     }
 772
 773     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 774         return -ERANGE;
 775     }
 776
 777     r = rbd_snap_create(s->image, sn_info->name);
 778     if (r < 0) {
 779         error_report("failed to create snap: %s", strerror(-r));
 780         return r;
 781     }
 782
 783     return 0;
 784 }
 785
 786 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 787                                 const char *snapshot_id,
 788                                 const char *snapshot_name,
 789                                 Error **errp)
 790 {
 791     BDRVRBDState *s = bs->opaque;
 792     int r;
 793
 794     if (!snapshot_name) {
 795         error_setg(errp, "rbd need a valid snapshot name");
 796         return -EINVAL;
 797     }
 798
 799     /* If snapshot_id is specified, it must be equal to name, see
 800        qemu_rbd_snap_list() */
 801     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 802         error_setg(errp,
 803                    "rbd do not support snapshot id, it should be NULL or "
 804                    "equal to snapshot name");
 805         return -EINVAL;
 806     }
 807
 808     r = rbd_snap_remove(s->image, snapshot_name);
 809     if (r < 0) {
 810         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 811     }
 812     return r;
 813 }
 814
 815 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 816                                   const char *snapshot_name)
 817 {
 818     BDRVRBDState *s = bs->opaque;
 819     int r;
 820
 821     r = rbd_snap_rollback(s->image, snapshot_name);
 822     return r;
 823 }
 824
 825 static int qemu_rbd_snap_list(BlockDriverState *bs,
 826                               QEMUSnapshotInfo **psn_tab)
 827 {
 828     BDRVRBDState *s = bs->opaque;
 829     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 830     int i, snap_count;
 831     rbd_snap_info_t *snaps;
 832     int max_snaps = RBD_MAX_SNAPS;
 833
 834     do {
 835         snaps = g_new(rbd_snap_info_t, max_snaps);
 836         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 837         if (snap_count <= 0) {
 838             g_free(snaps);
 839         }
 840     } while (snap_count == -ERANGE);
 841
 842     if (snap_count <= 0) {
 843         goto done;
 844     }
 845
 846     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
 847
 848     for (i = 0; i < snap_count; i++) {
 849         const char *snap_name = snaps[i].name;
 850
 851         sn_info = sn_tab + i;
 852         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 853         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 854
 855         sn_info->vm_state_size = snaps[i].size;
 856         sn_info->date_sec = 0;
 857         sn_info->date_nsec = 0;
 858         sn_info->vm_clock_nsec = 0;
 859     }
 860     rbd_snap_list_end(snaps);
 861     g_free(snaps);
 862
 863  done:
 864     *psn_tab = sn_tab;
 865     return snap_count;
 866 }
 867
 868 #ifdef LIBRBD_SUPPORTS_DISCARD
 869 static BlockAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 870                                         int64_t sector_num,
 871                                         int nb_sectors,
 872                                         BlockCompletionFunc *cb,
 873                                         void *opaque)
 874 {
 875     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 876                          RBD_AIO_DISCARD);
 877 }
 878 #endif
 879
 880 #ifdef LIBRBD_SUPPORTS_INVALIDATE
 881 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
 882                                       Error **errp)
 883 {
 884     BDRVRBDState *s = bs->opaque;
 885     int r = rbd_invalidate_cache(s->image);
 886     if (r < 0) {
 887         error_setg_errno(errp, -r, "Failed to invalidate the cache");
 888     }
 889 }
 890 #endif
 891
 892 static QemuOptsList qemu_rbd_create_opts = {
 893     .name = "rbd-create-opts",
 894     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
 895     .desc = {
 896         {
 897             .name = BLOCK_OPT_SIZE,
 898             .type = QEMU_OPT_SIZE,
 899             .help = "Virtual disk size"
 900         },
 901         {
 902             .name = BLOCK_OPT_CLUSTER_SIZE,
 903             .type = QEMU_OPT_SIZE,
 904             .help = "RBD object size"
 905         },
 906         { /* end of list */ }
 907     }
 908 };
 909
 910 static BlockDriver bdrv_rbd = {
 911     .format_name        = "rbd",
 912     .instance_size      = sizeof(BDRVRBDState),
 913     .bdrv_needs_filename = true,
 914     .bdrv_file_open     = qemu_rbd_open,
 915     .bdrv_close         = qemu_rbd_close,
 916     .bdrv_create        = qemu_rbd_create,
 917     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 918     .bdrv_get_info      = qemu_rbd_getinfo,
 919     .create_opts        = &qemu_rbd_create_opts,
 920     .bdrv_getlength     = qemu_rbd_getlength,
 921     .bdrv_truncate      = qemu_rbd_truncate,
 922     .protocol_name      = "rbd",
 923
 924     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 925     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 926
 927 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 928     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 929 #else
 930     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 931 #endif
 932
 933 #ifdef LIBRBD_SUPPORTS_DISCARD
 934     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 935 #endif
 936
 937     .bdrv_snapshot_create   = qemu_rbd_snap_create,
 938     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
 939     .bdrv_snapshot_list     = qemu_rbd_snap_list,
 940     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
 941 #ifdef LIBRBD_SUPPORTS_INVALIDATE
 942     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
 943 #endif
 944 };
 945
 946 static void bdrv_rbd_init(void)
 947 {
 948     bdrv_register(&bdrv_rbd);
 949 }
 950
 951 block_init(bdrv_rbd_init);