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 "qemu-common.h"
  17 #include "qemu/error-report.h"
  18 #include "block/block_int.h"
  19 #include "crypto/secret.h"
  20
  21 #include <rbd/librbd.h>
  22
  23 /*
  24  * When specifying the image filename use:
  25  *
  26  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
  27  *
  28  * poolname must be the name of an existing rados pool.
  29  *
  30  * devicename is the name of the rbd image.
  31  *
  32  * Each option given is used to configure rados, and may be any valid
  33  * Ceph option, "id", or "conf".
  34  *
  35  * The "id" option indicates what user we should authenticate as to
  36  * the Ceph cluster.  If it is excluded we will use the Ceph default
  37  * (normally 'admin').
  38  *
  39  * The "conf" option specifies a Ceph configuration file to read.  If
  40  * it is not specified, we will read from the default Ceph locations
  41  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
  42  * file, specify conf=/dev/null.
  43  *
  44  * Configuration values containing :, @, or = can be escaped with a
  45  * leading "\".
  46  */
  47
  48 /* rbd_aio_discard added in 0.1.2 */
  49 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
  50 #define LIBRBD_SUPPORTS_DISCARD
  51 #else
  52 #undef LIBRBD_SUPPORTS_DISCARD
  53 #endif
  54
  55 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
  56
  57 #define RBD_MAX_CONF_NAME_SIZE 128
  58 #define RBD_MAX_CONF_VAL_SIZE 512
  59 #define RBD_MAX_CONF_SIZE 1024
  60 #define RBD_MAX_POOL_NAME_SIZE 128
  61 #define RBD_MAX_SNAP_NAME_SIZE 128
  62 #define RBD_MAX_SNAPS 100
  63
  64 typedef enum {
  65     RBD_AIO_READ,
  66     RBD_AIO_WRITE,
  67     RBD_AIO_DISCARD,
  68     RBD_AIO_FLUSH
  69 } RBDAIOCmd;
  70
  71 typedef struct RBDAIOCB {
  72     BlockAIOCB common;
  73     QEMUBH *bh;
  74     int64_t ret;
  75     QEMUIOVector *qiov;
  76     char *bounce;
  77     RBDAIOCmd cmd;
  78     int error;
  79     struct BDRVRBDState *s;
  80 } RBDAIOCB;
  81
  82 typedef struct RADOSCB {
  83     RBDAIOCB *acb;
  84     struct BDRVRBDState *s;
  85     int64_t size;
  86     char *buf;
  87     int64_t ret;
  88 } RADOSCB;
  89
  90 typedef struct BDRVRBDState {
  91     rados_t cluster;
  92     rados_ioctx_t io_ctx;
  93     rbd_image_t image;
  94     char name[RBD_MAX_IMAGE_NAME_SIZE];
  95     char *snap;
  96 } BDRVRBDState;
  97
  98 static int qemu_rbd_next_tok(char *dst, int dst_len,
  99                              char *src, char delim,
 100                              const char *name,
 101                              char **p, Error **errp)
 102 {
 103     int l;
 104     char *end;
 105
 106     *p = NULL;
 107
 108     if (delim != '\0') {
 109         for (end = src; *end; ++end) {
 110             if (*end == delim) {
 111                 break;
 112             }
 113             if (*end == '\\' && end[1] != '\0') {
 114                 end++;
 115             }
 116         }
 117         if (*end == delim) {
 118             *p = end + 1;
 119             *end = '\0';
 120         }
 121     }
 122     l = strlen(src);
 123     if (l >= dst_len) {
 124         error_setg(errp, "%s too long", name);
 125         return -EINVAL;
 126     } else if (l == 0) {
 127         error_setg(errp, "%s too short", name);
 128         return -EINVAL;
 129     }
 130
 131     pstrcpy(dst, dst_len, src);
 132
 133     return 0;
 134 }
 135
 136 static void qemu_rbd_unescape(char *src)
 137 {
 138     char *p;
 139
 140     for (p = src; *src; ++src, ++p) {
 141         if (*src == '\\' && src[1] != '\0') {
 142             src++;
 143         }
 144         *p = *src;
 145     }
 146     *p = '\0';
 147 }
 148
 149 static int qemu_rbd_parsename(const char *filename,
 150                               char *pool, int pool_len,
 151                               char *snap, int snap_len,
 152                               char *name, int name_len,
 153                               char *conf, int conf_len,
 154                               Error **errp)
 155 {
 156     const char *start;
 157     char *p, *buf;
 158     int ret;
 159
 160     if (!strstart(filename, "rbd:", &start)) {
 161         error_setg(errp, "File name must start with 'rbd:'");
 162         return -EINVAL;
 163     }
 164
 165     buf = g_strdup(start);
 166     p = buf;
 167     *snap = '\0';
 168     *conf = '\0';
 169
 170     ret = qemu_rbd_next_tok(pool, pool_len, p,
 171                             '/', "pool name", &p, errp);
 172     if (ret < 0 || !p) {
 173         ret = -EINVAL;
 174         goto done;
 175     }
 176     qemu_rbd_unescape(pool);
 177
 178     if (strchr(p, '@')) {
 179         ret = qemu_rbd_next_tok(name, name_len, p,
 180                                 '@', "object name", &p, errp);
 181         if (ret < 0) {
 182             goto done;
 183         }
 184         ret = qemu_rbd_next_tok(snap, snap_len, p,
 185                                 ':', "snap name", &p, errp);
 186         qemu_rbd_unescape(snap);
 187     } else {
 188         ret = qemu_rbd_next_tok(name, name_len, p,
 189                                 ':', "object name", &p, errp);
 190     }
 191     qemu_rbd_unescape(name);
 192     if (ret < 0 || !p) {
 193         goto done;
 194     }
 195
 196     ret = qemu_rbd_next_tok(conf, conf_len, p,
 197                             '\0', "configuration", &p, errp);
 198
 199 done:
 200     g_free(buf);
 201     return ret;
 202 }
 203
 204 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 205 {
 206     const char *p = conf;
 207
 208     while (*p) {
 209         int len;
 210         const char *end = strchr(p, ':');
 211
 212         if (end) {
 213             len = end - p;
 214         } else {
 215             len = strlen(p);
 216         }
 217
 218         if (strncmp(p, "id=", 3) == 0) {
 219             len -= 3;
 220             strncpy(clientname, p + 3, len);
 221             clientname[len] = '\0';
 222             return clientname;
 223         }
 224         if (end == NULL) {
 225             break;
 226         }
 227         p = end + 1;
 228     }
 229     return NULL;
 230 }
 231
 232
 233 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
 234                              Error **errp)
 235 {
 236     if (secretid == 0) {
 237         return 0;
 238     }
 239
 240     gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
 241                                                     errp);
 242     if (!secret) {
 243         return -1;
 244     }
 245
 246     rados_conf_set(cluster, "key", secret);
 247     g_free(secret);
 248
 249     return 0;
 250 }
 251
 252
 253 static int qemu_rbd_set_conf(rados_t cluster, const char *conf,
 254                              bool only_read_conf_file,
 255                              Error **errp)
 256 {
 257     char *p, *buf;
 258     char name[RBD_MAX_CONF_NAME_SIZE];
 259     char value[RBD_MAX_CONF_VAL_SIZE];
 260     int ret = 0;
 261
 262     buf = g_strdup(conf);
 263     p = buf;
 264
 265     while (p) {
 266         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 267                                 '=', "conf option name", &p, errp);
 268         if (ret < 0) {
 269             break;
 270         }
 271         qemu_rbd_unescape(name);
 272
 273         if (!p) {
 274             error_setg(errp, "conf option %s has no value", name);
 275             ret = -EINVAL;
 276             break;
 277         }
 278
 279         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 280                                 ':', "conf option value", &p, errp);
 281         if (ret < 0) {
 282             break;
 283         }
 284         qemu_rbd_unescape(value);
 285
 286         if (strcmp(name, "conf") == 0) {
 287             /* read the conf file alone, so it doesn't override more
 288                specific settings for a particular device */
 289             if (only_read_conf_file) {
 290                 ret = rados_conf_read_file(cluster, value);
 291                 if (ret < 0) {
 292                     error_setg(errp, "error reading conf file %s", value);
 293                     break;
 294                 }
 295             }
 296         } else if (strcmp(name, "id") == 0) {
 297             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 298         } else if (!only_read_conf_file) {
 299             ret = rados_conf_set(cluster, name, value);
 300             if (ret < 0) {
 301                 error_setg(errp, "invalid conf option %s", name);
 302                 ret = -EINVAL;
 303                 break;
 304             }
 305         }
 306     }
 307
 308     g_free(buf);
 309     return ret;
 310 }
 311
 312 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 313 {
 314     Error *local_err = NULL;
 315     int64_t bytes = 0;
 316     int64_t objsize;
 317     int obj_order = 0;
 318     char pool[RBD_MAX_POOL_NAME_SIZE];
 319     char name[RBD_MAX_IMAGE_NAME_SIZE];
 320     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 321     char conf[RBD_MAX_CONF_SIZE];
 322     char clientname_buf[RBD_MAX_CONF_SIZE];
 323     char *clientname;
 324     const char *secretid;
 325     rados_t cluster;
 326     rados_ioctx_t io_ctx;
 327     int ret;
 328
 329     secretid = qemu_opt_get(opts, "password-secret");
 330
 331     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 332                            snap_buf, sizeof(snap_buf),
 333                            name, sizeof(name),
 334                            conf, sizeof(conf), &local_err) < 0) {
 335         error_propagate(errp, local_err);
 336         return -EINVAL;
 337     }
 338
 339     /* Read out options */
 340     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 341                      BDRV_SECTOR_SIZE);
 342     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 343     if (objsize) {
 344         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 345             error_setg(errp, "obj size needs to be power of 2");
 346             return -EINVAL;
 347         }
 348         if (objsize < 4096) {
 349             error_setg(errp, "obj size too small");
 350             return -EINVAL;
 351         }
 352         obj_order = ctz32(objsize);
 353     }
 354
 355     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 356     if (rados_create(&cluster, clientname) < 0) {
 357         error_setg(errp, "error initializing");
 358         return -EIO;
 359     }
 360
 361     if (strstr(conf, "conf=") == NULL) {
 362         /* try default location, but ignore failure */
 363         rados_conf_read_file(cluster, NULL);
 364     } else if (conf[0] != '\0' &&
 365                qemu_rbd_set_conf(cluster, conf, true, &local_err) < 0) {
 366         rados_shutdown(cluster);
 367         error_propagate(errp, local_err);
 368         return -EIO;
 369     }
 370
 371     if (conf[0] != '\0' &&
 372         qemu_rbd_set_conf(cluster, conf, false, &local_err) < 0) {
 373         rados_shutdown(cluster);
 374         error_propagate(errp, local_err);
 375         return -EIO;
 376     }
 377
 378     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
 379         rados_shutdown(cluster);
 380         return -EIO;
 381     }
 382
 383     if (rados_connect(cluster) < 0) {
 384         error_setg(errp, "error connecting");
 385         rados_shutdown(cluster);
 386         return -EIO;
 387     }
 388
 389     if (rados_ioctx_create(cluster, pool, &io_ctx) < 0) {
 390         error_setg(errp, "error opening pool %s", pool);
 391         rados_shutdown(cluster);
 392         return -EIO;
 393     }
 394
 395     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 396     rados_ioctx_destroy(io_ctx);
 397     rados_shutdown(cluster);
 398
 399     return ret;
 400 }
 401
 402 /*
 403  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 404  * BH context.
 405  */
 406 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 407 {
 408     RBDAIOCB *acb = rcb->acb;
 409     int64_t r;
 410
 411     r = rcb->ret;
 412
 413     if (acb->cmd != RBD_AIO_READ) {
 414         if (r < 0) {
 415             acb->ret = r;
 416             acb->error = 1;
 417         } else if (!acb->error) {
 418             acb->ret = rcb->size;
 419         }
 420     } else {
 421         if (r < 0) {
 422             memset(rcb->buf, 0, rcb->size);
 423             acb->ret = r;
 424             acb->error = 1;
 425         } else if (r < rcb->size) {
 426             memset(rcb->buf + r, 0, rcb->size - r);
 427             if (!acb->error) {
 428                 acb->ret = rcb->size;
 429             }
 430         } else if (!acb->error) {
 431             acb->ret = r;
 432         }
 433     }
 434
 435     g_free(rcb);
 436
 437     if (acb->cmd == RBD_AIO_READ) {
 438         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 439     }
 440     qemu_vfree(acb->bounce);
 441     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 442
 443     qemu_aio_unref(acb);
 444 }
 445
 446 /* TODO Convert to fine grained options */
 447 static QemuOptsList runtime_opts = {
 448     .name = "rbd",
 449     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 450     .desc = {
 451         {
 452             .name = "filename",
 453             .type = QEMU_OPT_STRING,
 454             .help = "Specification of the rbd image",
 455         },
 456         {
 457             .name = "password-secret",
 458             .type = QEMU_OPT_STRING,
 459             .help = "ID of secret providing the password",
 460         },
 461         { /* end of list */ }
 462     },
 463 };
 464
 465 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 466                          Error **errp)
 467 {
 468     BDRVRBDState *s = bs->opaque;
 469     char pool[RBD_MAX_POOL_NAME_SIZE];
 470     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 471     char conf[RBD_MAX_CONF_SIZE];
 472     char clientname_buf[RBD_MAX_CONF_SIZE];
 473     char *clientname;
 474     const char *secretid;
 475     QemuOpts *opts;
 476     Error *local_err = NULL;
 477     const char *filename;
 478     int r;
 479
 480     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 481     qemu_opts_absorb_qdict(opts, options, &local_err);
 482     if (local_err) {
 483         error_propagate(errp, local_err);
 484         qemu_opts_del(opts);
 485         return -EINVAL;
 486     }
 487
 488     filename = qemu_opt_get(opts, "filename");
 489     secretid = qemu_opt_get(opts, "password-secret");
 490
 491     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 492                            snap_buf, sizeof(snap_buf),
 493                            s->name, sizeof(s->name),
 494                            conf, sizeof(conf), errp) < 0) {
 495         r = -EINVAL;
 496         goto failed_opts;
 497     }
 498
 499     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 500     r = rados_create(&s->cluster, clientname);
 501     if (r < 0) {
 502         error_setg(errp, "error initializing");
 503         goto failed_opts;
 504     }
 505
 506     s->snap = NULL;
 507     if (snap_buf[0] != '\0') {
 508         s->snap = g_strdup(snap_buf);
 509     }
 510
 511     if (strstr(conf, "conf=") == NULL) {
 512         /* try default location, but ignore failure */
 513         rados_conf_read_file(s->cluster, NULL);
 514     } else if (conf[0] != '\0') {
 515         r = qemu_rbd_set_conf(s->cluster, conf, true, errp);
 516         if (r < 0) {
 517             goto failed_shutdown;
 518         }
 519     }
 520
 521     if (conf[0] != '\0') {
 522         r = qemu_rbd_set_conf(s->cluster, conf, false, errp);
 523         if (r < 0) {
 524             goto failed_shutdown;
 525         }
 526     }
 527
 528     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
 529         r = -EIO;
 530         goto failed_shutdown;
 531     }
 532
 533     /*
 534      * Fallback to more conservative semantics if setting cache
 535      * options fails. Ignore errors from setting rbd_cache because the
 536      * only possible error is that the option does not exist, and
 537      * librbd defaults to no caching. If write through caching cannot
 538      * be set up, fall back to no caching.
 539      */
 540     if (flags & BDRV_O_NOCACHE) {
 541         rados_conf_set(s->cluster, "rbd_cache", "false");
 542     } else {
 543         rados_conf_set(s->cluster, "rbd_cache", "true");
 544     }
 545
 546     r = rados_connect(s->cluster);
 547     if (r < 0) {
 548         error_setg(errp, "error connecting");
 549         goto failed_shutdown;
 550     }
 551
 552     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 553     if (r < 0) {
 554         error_setg(errp, "error opening pool %s", pool);
 555         goto failed_shutdown;
 556     }
 557
 558     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 559     if (r < 0) {
 560         error_setg(errp, "error reading header from %s", s->name);
 561         goto failed_open;
 562     }
 563
 564     bs->read_only = (s->snap != NULL);
 565
 566     qemu_opts_del(opts);
 567     return 0;
 568
 569 failed_open:
 570     rados_ioctx_destroy(s->io_ctx);
 571 failed_shutdown:
 572     rados_shutdown(s->cluster);
 573     g_free(s->snap);
 574 failed_opts:
 575     qemu_opts_del(opts);
 576     return r;
 577 }
 578
 579 static void qemu_rbd_close(BlockDriverState *bs)
 580 {
 581     BDRVRBDState *s = bs->opaque;
 582
 583     rbd_close(s->image);
 584     rados_ioctx_destroy(s->io_ctx);
 585     g_free(s->snap);
 586     rados_shutdown(s->cluster);
 587 }
 588
 589 static const AIOCBInfo rbd_aiocb_info = {
 590     .aiocb_size = sizeof(RBDAIOCB),
 591 };
 592
 593 static void rbd_finish_bh(void *opaque)
 594 {
 595     RADOSCB *rcb = opaque;
 596     qemu_bh_delete(rcb->acb->bh);
 597     qemu_rbd_complete_aio(rcb);
 598 }
 599
 600 /*
 601  * This is the callback function for rbd_aio_read and _write
 602  *
 603  * Note: this function is being called from a non qemu thread so
 604  * we need to be careful about what we do here. Generally we only
 605  * schedule a BH, and do the rest of the io completion handling
 606  * from rbd_finish_bh() which runs in a qemu context.
 607  */
 608 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 609 {
 610     RBDAIOCB *acb = rcb->acb;
 611
 612     rcb->ret = rbd_aio_get_return_value(c);
 613     rbd_aio_release(c);
 614
 615     acb->bh = aio_bh_new(bdrv_get_aio_context(acb->common.bs),
 616                          rbd_finish_bh, rcb);
 617     qemu_bh_schedule(acb->bh);
 618 }
 619
 620 static int rbd_aio_discard_wrapper(rbd_image_t image,
 621                                    uint64_t off,
 622                                    uint64_t len,
 623                                    rbd_completion_t comp)
 624 {
 625 #ifdef LIBRBD_SUPPORTS_DISCARD
 626     return rbd_aio_discard(image, off, len, comp);
 627 #else
 628     return -ENOTSUP;
 629 #endif
 630 }
 631
 632 static int rbd_aio_flush_wrapper(rbd_image_t image,
 633                                  rbd_completion_t comp)
 634 {
 635 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 636     return rbd_aio_flush(image, comp);
 637 #else
 638     return -ENOTSUP;
 639 #endif
 640 }
 641
 642 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 643                                  int64_t sector_num,
 644                                  QEMUIOVector *qiov,
 645                                  int nb_sectors,
 646                                  BlockCompletionFunc *cb,
 647                                  void *opaque,
 648                                  RBDAIOCmd cmd)
 649 {
 650     RBDAIOCB *acb;
 651     RADOSCB *rcb = NULL;
 652     rbd_completion_t c;
 653     int64_t off, size;
 654     char *buf;
 655     int r;
 656
 657     BDRVRBDState *s = bs->opaque;
 658
 659     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 660     acb->cmd = cmd;
 661     acb->qiov = qiov;
 662     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 663         acb->bounce = NULL;
 664     } else {
 665         acb->bounce = qemu_try_blockalign(bs, qiov->size);
 666         if (acb->bounce == NULL) {
 667             goto failed;
 668         }
 669     }
 670     acb->ret = 0;
 671     acb->error = 0;
 672     acb->s = s;
 673     acb->bh = NULL;
 674
 675     if (cmd == RBD_AIO_WRITE) {
 676         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 677     }
 678
 679     buf = acb->bounce;
 680
 681     off = sector_num * BDRV_SECTOR_SIZE;
 682     size = nb_sectors * BDRV_SECTOR_SIZE;
 683
 684     rcb = g_new(RADOSCB, 1);
 685     rcb->acb = acb;
 686     rcb->buf = buf;
 687     rcb->s = acb->s;
 688     rcb->size = size;
 689     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 690     if (r < 0) {
 691         goto failed;
 692     }
 693
 694     switch (cmd) {
 695     case RBD_AIO_WRITE:
 696         r = rbd_aio_write(s->image, off, size, buf, c);
 697         break;
 698     case RBD_AIO_READ:
 699         r = rbd_aio_read(s->image, off, size, buf, c);
 700         break;
 701     case RBD_AIO_DISCARD:
 702         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 703         break;
 704     case RBD_AIO_FLUSH:
 705         r = rbd_aio_flush_wrapper(s->image, c);
 706         break;
 707     default:
 708         r = -EINVAL;
 709     }
 710
 711     if (r < 0) {
 712         goto failed_completion;
 713     }
 714
 715     return &acb->common;
 716
 717 failed_completion:
 718     rbd_aio_release(c);
 719 failed:
 720     g_free(rcb);
 721     qemu_vfree(acb->bounce);
 722     qemu_aio_unref(acb);
 723     return NULL;
 724 }
 725
 726 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 727                                       int64_t sector_num,
 728                                       QEMUIOVector *qiov,
 729                                       int nb_sectors,
 730                                       BlockCompletionFunc *cb,
 731                                       void *opaque)
 732 {
 733     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 734                          RBD_AIO_READ);
 735 }
 736
 737 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 738                                        int64_t sector_num,
 739                                        QEMUIOVector *qiov,
 740                                        int nb_sectors,
 741                                        BlockCompletionFunc *cb,
 742                                        void *opaque)
 743 {
 744     return rbd_start_aio(bs, sector_num, qiov, nb_sectors, cb, opaque,
 745                          RBD_AIO_WRITE);
 746 }
 747
 748 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 749 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 750                                       BlockCompletionFunc *cb,
 751                                       void *opaque)
 752 {
 753     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 754 }
 755
 756 #else
 757
 758 static int qemu_rbd_co_flush(BlockDriverState *bs)
 759 {
 760 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 761     /* rbd_flush added in 0.1.1 */
 762     BDRVRBDState *s = bs->opaque;
 763     return rbd_flush(s->image);
 764 #else
 765     return 0;
 766 #endif
 767 }
 768 #endif
 769
 770 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 771 {
 772     BDRVRBDState *s = bs->opaque;
 773     rbd_image_info_t info;
 774     int r;
 775
 776     r = rbd_stat(s->image, &info, sizeof(info));
 777     if (r < 0) {
 778         return r;
 779     }
 780
 781     bdi->cluster_size = info.obj_size;
 782     return 0;
 783 }
 784
 785 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 786 {
 787     BDRVRBDState *s = bs->opaque;
 788     rbd_image_info_t info;
 789     int r;
 790
 791     r = rbd_stat(s->image, &info, sizeof(info));
 792     if (r < 0) {
 793         return r;
 794     }
 795
 796     return info.size;
 797 }
 798
 799 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 800 {
 801     BDRVRBDState *s = bs->opaque;
 802     int r;
 803
 804     r = rbd_resize(s->image, offset);
 805     if (r < 0) {
 806         return r;
 807     }
 808
 809     return 0;
 810 }
 811
 812 static int qemu_rbd_snap_create(BlockDriverState *bs,
 813                                 QEMUSnapshotInfo *sn_info)
 814 {
 815     BDRVRBDState *s = bs->opaque;
 816     int r;
 817
 818     if (sn_info->name[0] == '\0') {
 819         return -EINVAL; /* we need a name for rbd snapshots */
 820     }
 821
 822     /*
 823      * rbd snapshots are using the name as the user controlled unique identifier
 824      * we can't use the rbd snapid for that purpose, as it can't be set
 825      */
 826     if (sn_info->id_str[0] != '\0' &&
 827         strcmp(sn_info->id_str, sn_info->name) != 0) {
 828         return -EINVAL;
 829     }
 830
 831     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 832         return -ERANGE;
 833     }
 834
 835     r = rbd_snap_create(s->image, sn_info->name);
 836     if (r < 0) {
 837         error_report("failed to create snap: %s", strerror(-r));
 838         return r;
 839     }
 840
 841     return 0;
 842 }
 843
 844 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 845                                 const char *snapshot_id,
 846                                 const char *snapshot_name,
 847                                 Error **errp)
 848 {
 849     BDRVRBDState *s = bs->opaque;
 850     int r;
 851
 852     if (!snapshot_name) {
 853         error_setg(errp, "rbd need a valid snapshot name");
 854         return -EINVAL;
 855     }
 856
 857     /* If snapshot_id is specified, it must be equal to name, see
 858        qemu_rbd_snap_list() */
 859     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 860         error_setg(errp,
 861                    "rbd do not support snapshot id, it should be NULL or "
 862                    "equal to snapshot name");
 863         return -EINVAL;
 864     }
 865
 866     r = rbd_snap_remove(s->image, snapshot_name);
 867     if (r < 0) {
 868         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 869     }
 870     return r;
 871 }
 872
 873 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 874                                   const char *snapshot_name)
 875 {
 876     BDRVRBDState *s = bs->opaque;
 877     int r;
 878
 879     r = rbd_snap_rollback(s->image, snapshot_name);
 880     return r;
 881 }
 882
 883 static int qemu_rbd_snap_list(BlockDriverState *bs,
 884                               QEMUSnapshotInfo **psn_tab)
 885 {
 886     BDRVRBDState *s = bs->opaque;
 887     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 888     int i, snap_count;
 889     rbd_snap_info_t *snaps;
 890     int max_snaps = RBD_MAX_SNAPS;
 891
 892     do {
 893         snaps = g_new(rbd_snap_info_t, max_snaps);
 894         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 895         if (snap_count <= 0) {
 896             g_free(snaps);
 897         }
 898     } while (snap_count == -ERANGE);
 899
 900     if (snap_count <= 0) {
 901         goto done;
 902     }
 903
 904     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
 905
 906     for (i = 0; i < snap_count; i++) {
 907         const char *snap_name = snaps[i].name;
 908
 909         sn_info = sn_tab + i;
 910         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 911         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 912
 913         sn_info->vm_state_size = snaps[i].size;
 914         sn_info->date_sec = 0;
 915         sn_info->date_nsec = 0;
 916         sn_info->vm_clock_nsec = 0;
 917     }
 918     rbd_snap_list_end(snaps);
 919     g_free(snaps);
 920
 921  done:
 922     *psn_tab = sn_tab;
 923     return snap_count;
 924 }
 925
 926 #ifdef LIBRBD_SUPPORTS_DISCARD
 927 static BlockAIOCB* qemu_rbd_aio_discard(BlockDriverState *bs,
 928                                         int64_t sector_num,
 929                                         int nb_sectors,
 930                                         BlockCompletionFunc *cb,
 931                                         void *opaque)
 932 {
 933     return rbd_start_aio(bs, sector_num, NULL, nb_sectors, cb, opaque,
 934                          RBD_AIO_DISCARD);
 935 }
 936 #endif
 937
 938 #ifdef LIBRBD_SUPPORTS_INVALIDATE
 939 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
 940                                       Error **errp)
 941 {
 942     BDRVRBDState *s = bs->opaque;
 943     int r = rbd_invalidate_cache(s->image);
 944     if (r < 0) {
 945         error_setg_errno(errp, -r, "Failed to invalidate the cache");
 946     }
 947 }
 948 #endif
 949
 950 static QemuOptsList qemu_rbd_create_opts = {
 951     .name = "rbd-create-opts",
 952     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
 953     .desc = {
 954         {
 955             .name = BLOCK_OPT_SIZE,
 956             .type = QEMU_OPT_SIZE,
 957             .help = "Virtual disk size"
 958         },
 959         {
 960             .name = BLOCK_OPT_CLUSTER_SIZE,
 961             .type = QEMU_OPT_SIZE,
 962             .help = "RBD object size"
 963         },
 964         {
 965             .name = "password-secret",
 966             .type = QEMU_OPT_STRING,
 967             .help = "ID of secret providing the password",
 968         },
 969         { /* end of list */ }
 970     }
 971 };
 972
 973 static BlockDriver bdrv_rbd = {
 974     .format_name        = "rbd",
 975     .instance_size      = sizeof(BDRVRBDState),
 976     .bdrv_needs_filename = true,
 977     .bdrv_file_open     = qemu_rbd_open,
 978     .bdrv_close         = qemu_rbd_close,
 979     .bdrv_create        = qemu_rbd_create,
 980     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 981     .bdrv_get_info      = qemu_rbd_getinfo,
 982     .create_opts        = &qemu_rbd_create_opts,
 983     .bdrv_getlength     = qemu_rbd_getlength,
 984     .bdrv_truncate      = qemu_rbd_truncate,
 985     .protocol_name      = "rbd",
 986
 987     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 988     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 989
 990 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 991     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 992 #else
 993     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 994 #endif
 995
 996 #ifdef LIBRBD_SUPPORTS_DISCARD
 997     .bdrv_aio_discard       = qemu_rbd_aio_discard,
 998 #endif
 999
1000     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1001     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1002     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1003     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1004 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1005     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1006 #endif
1007 };
1008
1009 static void bdrv_rbd_init(void)
1010 {
1011     bdrv_register(&bdrv_rbd);
1012 }
1013
1014 block_init(bdrv_rbd_init);