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 <rbd/librbd.h>
  17 #include "qapi/error.h"
  18 #include "qemu/error-report.h"
  19 #include "block/block_int.h"
  20 #include "crypto/secret.h"
  21 #include "qemu/cutils.h"
  22 #include "qapi/qmp/qstring.h"
  23 #include "qapi/qmp/qjson.h"
  24
  25 /*
  26  * When specifying the image filename use:
  27  *
  28  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
  29  *
  30  * poolname must be the name of an existing rados pool.
  31  *
  32  * devicename is the name of the rbd image.
  33  *
  34  * Each option given is used to configure rados, and may be any valid
  35  * Ceph option, "id", or "conf".
  36  *
  37  * The "id" option indicates what user we should authenticate as to
  38  * the Ceph cluster.  If it is excluded we will use the Ceph default
  39  * (normally 'admin').
  40  *
  41  * The "conf" option specifies a Ceph configuration file to read.  If
  42  * it is not specified, we will read from the default Ceph locations
  43  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
  44  * file, specify conf=/dev/null.
  45  *
  46  * Configuration values containing :, @, or = can be escaped with a
  47  * leading "\".
  48  */
  49
  50 /* rbd_aio_discard added in 0.1.2 */
  51 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
  52 #define LIBRBD_SUPPORTS_DISCARD
  53 #else
  54 #undef LIBRBD_SUPPORTS_DISCARD
  55 #endif
  56
  57 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
  58
  59 #define RBD_MAX_SNAPS 100
  60
  61 /* The LIBRBD_SUPPORTS_IOVEC is defined in librbd.h */
  62 #ifdef LIBRBD_SUPPORTS_IOVEC
  63 #define LIBRBD_USE_IOVEC 1
  64 #else
  65 #define LIBRBD_USE_IOVEC 0
  66 #endif
  67
  68 typedef enum {
  69     RBD_AIO_READ,
  70     RBD_AIO_WRITE,
  71     RBD_AIO_DISCARD,
  72     RBD_AIO_FLUSH
  73 } RBDAIOCmd;
  74
  75 typedef struct RBDAIOCB {
  76     BlockAIOCB common;
  77     int64_t ret;
  78     QEMUIOVector *qiov;
  79     char *bounce;
  80     RBDAIOCmd cmd;
  81     int error;
  82     struct BDRVRBDState *s;
  83 } RBDAIOCB;
  84
  85 typedef struct RADOSCB {
  86     RBDAIOCB *acb;
  87     struct BDRVRBDState *s;
  88     int64_t size;
  89     char *buf;
  90     int64_t ret;
  91 } RADOSCB;
  92
  93 typedef struct BDRVRBDState {
  94     rados_t cluster;
  95     rados_ioctx_t io_ctx;
  96     rbd_image_t image;
  97     char *image_name;
  98     char *snap;
  99 } BDRVRBDState;
 100
 101 static char *qemu_rbd_next_tok(char *src, char delim, char **p)
 102 {
 103     char *end;
 104
 105     *p = NULL;
 106
 107     for (end = src; *end; ++end) {
 108         if (*end == delim) {
 109             break;
 110         }
 111         if (*end == '\\' && end[1] != '\0') {
 112             end++;
 113         }
 114     }
 115     if (*end == delim) {
 116         *p = end + 1;
 117         *end = '\0';
 118     }
 119     return src;
 120 }
 121
 122 static void qemu_rbd_unescape(char *src)
 123 {
 124     char *p;
 125
 126     for (p = src; *src; ++src, ++p) {
 127         if (*src == '\\' && src[1] != '\0') {
 128             src++;
 129         }
 130         *p = *src;
 131     }
 132     *p = '\0';
 133 }
 134
 135 static void qemu_rbd_parse_filename(const char *filename, QDict *options,
 136                                     Error **errp)
 137 {
 138     const char *start;
 139     char *p, *buf;
 140     QList *keypairs = NULL;
 141     char *found_str;
 142
 143     if (!strstart(filename, "rbd:", &start)) {
 144         error_setg(errp, "File name must start with 'rbd:'");
 145         return;
 146     }
 147
 148     buf = g_strdup(start);
 149     p = buf;
 150
 151     found_str = qemu_rbd_next_tok(p, '/', &p);
 152     if (!p) {
 153         error_setg(errp, "Pool name is required");
 154         goto done;
 155     }
 156     qemu_rbd_unescape(found_str);
 157     qdict_put_str(options, "pool", found_str);
 158
 159     if (strchr(p, '@')) {
 160         found_str = qemu_rbd_next_tok(p, '@', &p);
 161         qemu_rbd_unescape(found_str);
 162         qdict_put_str(options, "image", found_str);
 163
 164         found_str = qemu_rbd_next_tok(p, ':', &p);
 165         qemu_rbd_unescape(found_str);
 166         qdict_put_str(options, "snapshot", found_str);
 167     } else {
 168         found_str = qemu_rbd_next_tok(p, ':', &p);
 169         qemu_rbd_unescape(found_str);
 170         qdict_put_str(options, "image", found_str);
 171     }
 172     if (!p) {
 173         goto done;
 174     }
 175
 176     /* The following are essentially all key/value pairs, and we treat
 177      * 'id' and 'conf' a bit special.  Key/value pairs may be in any order. */
 178     while (p) {
 179         char *name, *value;
 180         name = qemu_rbd_next_tok(p, '=', &p);
 181         if (!p) {
 182             error_setg(errp, "conf option %s has no value", name);
 183             break;
 184         }
 185
 186         qemu_rbd_unescape(name);
 187
 188         value = qemu_rbd_next_tok(p, ':', &p);
 189         qemu_rbd_unescape(value);
 190
 191         if (!strcmp(name, "conf")) {
 192             qdict_put_str(options, "conf", value);
 193         } else if (!strcmp(name, "id")) {
 194             qdict_put_str(options, "user", value);
 195         } else {
 196             /*
 197              * We pass these internally to qemu_rbd_set_keypairs(), so
 198              * we can get away with the simpler list of [ "key1",
 199              * "value1", "key2", "value2" ] rather than a raw dict
 200              * { "key1": "value1", "key2": "value2" } where we can't
 201              * guarantee order, or even a more correct but complex
 202              * [ { "key1": "value1" }, { "key2": "value2" } ]
 203              */
 204             if (!keypairs) {
 205                 keypairs = qlist_new();
 206             }
 207             qlist_append_str(keypairs, name);
 208             qlist_append_str(keypairs, value);
 209         }
 210     }
 211
 212     if (keypairs) {
 213         qdict_put(options, "=keyvalue-pairs",
 214                   qobject_to_json(QOBJECT(keypairs)));
 215     }
 216
 217 done:
 218     g_free(buf);
 219     QDECREF(keypairs);
 220     return;
 221 }
 222
 223
 224 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
 225                              Error **errp)
 226 {
 227     if (secretid == 0) {
 228         return 0;
 229     }
 230
 231     gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
 232                                                     errp);
 233     if (!secret) {
 234         return -1;
 235     }
 236
 237     rados_conf_set(cluster, "key", secret);
 238     g_free(secret);
 239
 240     return 0;
 241 }
 242
 243 static int qemu_rbd_set_keypairs(rados_t cluster, const char *keypairs_json,
 244                                  Error **errp)
 245 {
 246     QList *keypairs;
 247     QString *name;
 248     QString *value;
 249     const char *key;
 250     size_t remaining;
 251     int ret = 0;
 252
 253     if (!keypairs_json) {
 254         return ret;
 255     }
 256     keypairs = qobject_to_qlist(qobject_from_json(keypairs_json,
 257                                                   &error_abort));
 258     remaining = qlist_size(keypairs) / 2;
 259     assert(remaining);
 260
 261     while (remaining--) {
 262         name = qobject_to_qstring(qlist_pop(keypairs));
 263         value = qobject_to_qstring(qlist_pop(keypairs));
 264         assert(name && value);
 265         key = qstring_get_str(name);
 266
 267         ret = rados_conf_set(cluster, key, qstring_get_str(value));
 268         QDECREF(name);
 269         QDECREF(value);
 270         if (ret < 0) {
 271             error_setg_errno(errp, -ret, "invalid conf option %s", key);
 272             ret = -EINVAL;
 273             break;
 274         }
 275     }
 276
 277     QDECREF(keypairs);
 278     return ret;
 279 }
 280
 281 static void qemu_rbd_memset(RADOSCB *rcb, int64_t offs)
 282 {
 283     if (LIBRBD_USE_IOVEC) {
 284         RBDAIOCB *acb = rcb->acb;
 285         iov_memset(acb->qiov->iov, acb->qiov->niov, offs, 0,
 286                    acb->qiov->size - offs);
 287     } else {
 288         memset(rcb->buf + offs, 0, rcb->size - offs);
 289     }
 290 }
 291
 292 static QemuOptsList runtime_opts = {
 293     .name = "rbd",
 294     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 295     .desc = {
 296         {
 297             .name = "pool",
 298             .type = QEMU_OPT_STRING,
 299             .help = "Rados pool name",
 300         },
 301         {
 302             .name = "image",
 303             .type = QEMU_OPT_STRING,
 304             .help = "Image name in the pool",
 305         },
 306         {
 307             .name = "conf",
 308             .type = QEMU_OPT_STRING,
 309             .help = "Rados config file location",
 310         },
 311         {
 312             .name = "snapshot",
 313             .type = QEMU_OPT_STRING,
 314             .help = "Ceph snapshot name",
 315         },
 316         {
 317             /* maps to 'id' in rados_create() */
 318             .name = "user",
 319             .type = QEMU_OPT_STRING,
 320             .help = "Rados id name",
 321         },
 322         /*
 323          * server.* extracted manually, see qemu_rbd_mon_host()
 324          */
 325         {
 326             .name = "password-secret",
 327             .type = QEMU_OPT_STRING,
 328             .help = "ID of secret providing the password",
 329         },
 330
 331         /*
 332          * Keys for qemu_rbd_parse_filename(), not in the QAPI schema
 333          */
 334         {
 335             /*
 336              * HACK: name starts with '=' so that qemu_opts_parse()
 337              * can't set it
 338              */
 339             .name = "=keyvalue-pairs",
 340             .type = QEMU_OPT_STRING,
 341             .help = "Legacy rados key/value option parameters",
 342         },
 343         { /* end of list */ }
 344     },
 345 };
 346
 347 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 348 {
 349     Error *local_err = NULL;
 350     int64_t bytes = 0;
 351     int64_t objsize;
 352     int obj_order = 0;
 353     const char *pool, *image_name, *conf, *user, *keypairs;
 354     const char *secretid;
 355     rados_t cluster;
 356     rados_ioctx_t io_ctx;
 357     QDict *options = NULL;
 358     int ret = 0;
 359
 360     secretid = qemu_opt_get(opts, "password-secret");
 361
 362     /* Read out options */
 363     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 364                      BDRV_SECTOR_SIZE);
 365     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 366     if (objsize) {
 367         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 368             error_setg(errp, "obj size needs to be power of 2");
 369             ret = -EINVAL;
 370             goto exit;
 371         }
 372         if (objsize < 4096) {
 373             error_setg(errp, "obj size too small");
 374             ret = -EINVAL;
 375             goto exit;
 376         }
 377         obj_order = ctz32(objsize);
 378     }
 379
 380     options = qdict_new();
 381     qemu_rbd_parse_filename(filename, options, &local_err);
 382     if (local_err) {
 383         ret = -EINVAL;
 384         error_propagate(errp, local_err);
 385         goto exit;
 386     }
 387
 388     /*
 389      * Caution: while qdict_get_try_str() is fine, getting non-string
 390      * types would require more care.  When @options come from -blockdev
 391      * or blockdev_add, its members are typed according to the QAPI
 392      * schema, but when they come from -drive, they're all QString.
 393      */
 394     pool       = qdict_get_try_str(options, "pool");
 395     conf       = qdict_get_try_str(options, "conf");
 396     user       = qdict_get_try_str(options, "user");
 397     image_name = qdict_get_try_str(options, "image");
 398     keypairs   = qdict_get_try_str(options, "=keyvalue-pairs");
 399
 400     ret = rados_create(&cluster, user);
 401     if (ret < 0) {
 402         error_setg_errno(errp, -ret, "error initializing");
 403         goto exit;
 404     }
 405
 406     /* try default location when conf=NULL, but ignore failure */
 407     ret = rados_conf_read_file(cluster, conf);
 408     if (conf && ret < 0) {
 409         error_setg_errno(errp, -ret, "error reading conf file %s", conf);
 410         ret = -EIO;
 411         goto shutdown;
 412     }
 413
 414     ret = qemu_rbd_set_keypairs(cluster, keypairs, errp);
 415     if (ret < 0) {
 416         ret = -EIO;
 417         goto shutdown;
 418     }
 419
 420     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
 421         ret = -EIO;
 422         goto shutdown;
 423     }
 424
 425     ret = rados_connect(cluster);
 426     if (ret < 0) {
 427         error_setg_errno(errp, -ret, "error connecting");
 428         goto shutdown;
 429     }
 430
 431     ret = rados_ioctx_create(cluster, pool, &io_ctx);
 432     if (ret < 0) {
 433         error_setg_errno(errp, -ret, "error opening pool %s", pool);
 434         goto shutdown;
 435     }
 436
 437     ret = rbd_create(io_ctx, image_name, bytes, &obj_order);
 438     if (ret < 0) {
 439         error_setg_errno(errp, -ret, "error rbd create");
 440     }
 441
 442     rados_ioctx_destroy(io_ctx);
 443
 444 shutdown:
 445     rados_shutdown(cluster);
 446
 447 exit:
 448     QDECREF(options);
 449     return ret;
 450 }
 451
 452 /*
 453  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 454  * BH context.
 455  */
 456 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 457 {
 458     RBDAIOCB *acb = rcb->acb;
 459     int64_t r;
 460
 461     r = rcb->ret;
 462
 463     if (acb->cmd != RBD_AIO_READ) {
 464         if (r < 0) {
 465             acb->ret = r;
 466             acb->error = 1;
 467         } else if (!acb->error) {
 468             acb->ret = rcb->size;
 469         }
 470     } else {
 471         if (r < 0) {
 472             qemu_rbd_memset(rcb, 0);
 473             acb->ret = r;
 474             acb->error = 1;
 475         } else if (r < rcb->size) {
 476             qemu_rbd_memset(rcb, r);
 477             if (!acb->error) {
 478                 acb->ret = rcb->size;
 479             }
 480         } else if (!acb->error) {
 481             acb->ret = r;
 482         }
 483     }
 484
 485     g_free(rcb);
 486
 487     if (!LIBRBD_USE_IOVEC) {
 488         if (acb->cmd == RBD_AIO_READ) {
 489             qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 490         }
 491         qemu_vfree(acb->bounce);
 492     }
 493
 494     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 495
 496     qemu_aio_unref(acb);
 497 }
 498
 499 static char *qemu_rbd_mon_host(QDict *options, Error **errp)
 500 {
 501     const char **vals = g_new(const char *, qdict_size(options) + 1);
 502     char keybuf[32];
 503     const char *host, *port;
 504     char *rados_str;
 505     int i;
 506
 507     for (i = 0;; i++) {
 508         sprintf(keybuf, "server.%d.host", i);
 509         host = qdict_get_try_str(options, keybuf);
 510         qdict_del(options, keybuf);
 511         sprintf(keybuf, "server.%d.port", i);
 512         port = qdict_get_try_str(options, keybuf);
 513         qdict_del(options, keybuf);
 514         if (!host && !port) {
 515             break;
 516         }
 517         if (!host) {
 518             error_setg(errp, "Parameter server.%d.host is missing", i);
 519             rados_str = NULL;
 520             goto out;
 521         }
 522
 523         if (strchr(host, ':')) {
 524             vals[i] = port ? g_strdup_printf("[%s]:%s", host, port)
 525                 : g_strdup_printf("[%s]", host);
 526         } else {
 527             vals[i] = port ? g_strdup_printf("%s:%s", host, port)
 528                 : g_strdup(host);
 529         }
 530     }
 531     vals[i] = NULL;
 532
 533     rados_str = i ? g_strjoinv(";", (char **)vals) : NULL;
 534 out:
 535     g_strfreev((char **)vals);
 536     return rados_str;
 537 }
 538
 539 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 540                          Error **errp)
 541 {
 542     BDRVRBDState *s = bs->opaque;
 543     const char *pool, *snap, *conf, *user, *image_name, *keypairs;
 544     const char *secretid;
 545     QemuOpts *opts;
 546     Error *local_err = NULL;
 547     char *mon_host = NULL;
 548     int r;
 549
 550     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 551     qemu_opts_absorb_qdict(opts, options, &local_err);
 552     if (local_err) {
 553         error_propagate(errp, local_err);
 554         r = -EINVAL;
 555         goto failed_opts;
 556     }
 557
 558     mon_host = qemu_rbd_mon_host(options, &local_err);
 559     if (local_err) {
 560         error_propagate(errp, local_err);
 561         r = -EINVAL;
 562         goto failed_opts;
 563     }
 564
 565     secretid = qemu_opt_get(opts, "password-secret");
 566
 567     pool           = qemu_opt_get(opts, "pool");
 568     conf           = qemu_opt_get(opts, "conf");
 569     snap           = qemu_opt_get(opts, "snapshot");
 570     user           = qemu_opt_get(opts, "user");
 571     image_name     = qemu_opt_get(opts, "image");
 572     keypairs       = qemu_opt_get(opts, "=keyvalue-pairs");
 573
 574     if (!pool || !image_name) {
 575         error_setg(errp, "Parameters 'pool' and 'image' are required");
 576         r = -EINVAL;
 577         goto failed_opts;
 578     }
 579
 580     r = rados_create(&s->cluster, user);
 581     if (r < 0) {
 582         error_setg_errno(errp, -r, "error initializing");
 583         goto failed_opts;
 584     }
 585
 586     s->snap = g_strdup(snap);
 587     s->image_name = g_strdup(image_name);
 588
 589     /* try default location when conf=NULL, but ignore failure */
 590     r = rados_conf_read_file(s->cluster, conf);
 591     if (conf && r < 0) {
 592         error_setg_errno(errp, -r, "error reading conf file %s", conf);
 593         goto failed_shutdown;
 594     }
 595
 596     r = qemu_rbd_set_keypairs(s->cluster, keypairs, errp);
 597     if (r < 0) {
 598         goto failed_shutdown;
 599     }
 600
 601     if (mon_host) {
 602         r = rados_conf_set(s->cluster, "mon_host", mon_host);
 603         if (r < 0) {
 604             goto failed_shutdown;
 605         }
 606     }
 607
 608     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
 609         r = -EIO;
 610         goto failed_shutdown;
 611     }
 612
 613     /*
 614      * Fallback to more conservative semantics if setting cache
 615      * options fails. Ignore errors from setting rbd_cache because the
 616      * only possible error is that the option does not exist, and
 617      * librbd defaults to no caching. If write through caching cannot
 618      * be set up, fall back to no caching.
 619      */
 620     if (flags & BDRV_O_NOCACHE) {
 621         rados_conf_set(s->cluster, "rbd_cache", "false");
 622     } else {
 623         rados_conf_set(s->cluster, "rbd_cache", "true");
 624     }
 625
 626     r = rados_connect(s->cluster);
 627     if (r < 0) {
 628         error_setg_errno(errp, -r, "error connecting");
 629         goto failed_shutdown;
 630     }
 631
 632     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 633     if (r < 0) {
 634         error_setg_errno(errp, -r, "error opening pool %s", pool);
 635         goto failed_shutdown;
 636     }
 637
 638     /* rbd_open is always r/w */
 639     r = rbd_open(s->io_ctx, s->image_name, &s->image, s->snap);
 640     if (r < 0) {
 641         error_setg_errno(errp, -r, "error reading header from %s",
 642                          s->image_name);
 643         goto failed_open;
 644     }
 645
 646     /* If we are using an rbd snapshot, we must be r/o, otherwise
 647      * leave as-is */
 648     if (s->snap != NULL) {
 649         r = bdrv_set_read_only(bs, true, &local_err);
 650         if (r < 0) {
 651             error_propagate(errp, local_err);
 652             goto failed_open;
 653         }
 654     }
 655
 656     qemu_opts_del(opts);
 657     return 0;
 658
 659 failed_open:
 660     rados_ioctx_destroy(s->io_ctx);
 661 failed_shutdown:
 662     rados_shutdown(s->cluster);
 663     g_free(s->snap);
 664     g_free(s->image_name);
 665 failed_opts:
 666     qemu_opts_del(opts);
 667     g_free(mon_host);
 668     return r;
 669 }
 670
 671
 672 /* Since RBD is currently always opened R/W via the API,
 673  * we just need to check if we are using a snapshot or not, in
 674  * order to determine if we will allow it to be R/W */
 675 static int qemu_rbd_reopen_prepare(BDRVReopenState *state,
 676                                    BlockReopenQueue *queue, Error **errp)
 677 {
 678     BDRVRBDState *s = state->bs->opaque;
 679     int ret = 0;
 680
 681     if (s->snap && state->flags & BDRV_O_RDWR) {
 682         error_setg(errp,
 683                    "Cannot change node '%s' to r/w when using RBD snapshot",
 684                    bdrv_get_device_or_node_name(state->bs));
 685         ret = -EINVAL;
 686     }
 687
 688     return ret;
 689 }
 690
 691 static void qemu_rbd_close(BlockDriverState *bs)
 692 {
 693     BDRVRBDState *s = bs->opaque;
 694
 695     rbd_close(s->image);
 696     rados_ioctx_destroy(s->io_ctx);
 697     g_free(s->snap);
 698     g_free(s->image_name);
 699     rados_shutdown(s->cluster);
 700 }
 701
 702 static const AIOCBInfo rbd_aiocb_info = {
 703     .aiocb_size = sizeof(RBDAIOCB),
 704 };
 705
 706 static void rbd_finish_bh(void *opaque)
 707 {
 708     RADOSCB *rcb = opaque;
 709     qemu_rbd_complete_aio(rcb);
 710 }
 711
 712 /*
 713  * This is the callback function for rbd_aio_read and _write
 714  *
 715  * Note: this function is being called from a non qemu thread so
 716  * we need to be careful about what we do here. Generally we only
 717  * schedule a BH, and do the rest of the io completion handling
 718  * from rbd_finish_bh() which runs in a qemu context.
 719  */
 720 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 721 {
 722     RBDAIOCB *acb = rcb->acb;
 723
 724     rcb->ret = rbd_aio_get_return_value(c);
 725     rbd_aio_release(c);
 726
 727     aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs),
 728                             rbd_finish_bh, rcb);
 729 }
 730
 731 static int rbd_aio_discard_wrapper(rbd_image_t image,
 732                                    uint64_t off,
 733                                    uint64_t len,
 734                                    rbd_completion_t comp)
 735 {
 736 #ifdef LIBRBD_SUPPORTS_DISCARD
 737     return rbd_aio_discard(image, off, len, comp);
 738 #else
 739     return -ENOTSUP;
 740 #endif
 741 }
 742
 743 static int rbd_aio_flush_wrapper(rbd_image_t image,
 744                                  rbd_completion_t comp)
 745 {
 746 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 747     return rbd_aio_flush(image, comp);
 748 #else
 749     return -ENOTSUP;
 750 #endif
 751 }
 752
 753 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 754                                  int64_t off,
 755                                  QEMUIOVector *qiov,
 756                                  int64_t size,
 757                                  BlockCompletionFunc *cb,
 758                                  void *opaque,
 759                                  RBDAIOCmd cmd)
 760 {
 761     RBDAIOCB *acb;
 762     RADOSCB *rcb = NULL;
 763     rbd_completion_t c;
 764     int r;
 765
 766     BDRVRBDState *s = bs->opaque;
 767
 768     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 769     acb->cmd = cmd;
 770     acb->qiov = qiov;
 771     assert(!qiov || qiov->size == size);
 772
 773     rcb = g_new(RADOSCB, 1);
 774
 775     if (!LIBRBD_USE_IOVEC) {
 776         if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 777             acb->bounce = NULL;
 778         } else {
 779             acb->bounce = qemu_try_blockalign(bs, qiov->size);
 780             if (acb->bounce == NULL) {
 781                 goto failed;
 782             }
 783         }
 784         if (cmd == RBD_AIO_WRITE) {
 785             qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 786         }
 787         rcb->buf = acb->bounce;
 788     }
 789
 790     acb->ret = 0;
 791     acb->error = 0;
 792     acb->s = s;
 793
 794     rcb->acb = acb;
 795     rcb->s = acb->s;
 796     rcb->size = size;
 797     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 798     if (r < 0) {
 799         goto failed;
 800     }
 801
 802     switch (cmd) {
 803     case RBD_AIO_WRITE:
 804 #ifdef LIBRBD_SUPPORTS_IOVEC
 805             r = rbd_aio_writev(s->image, qiov->iov, qiov->niov, off, c);
 806 #else
 807             r = rbd_aio_write(s->image, off, size, rcb->buf, c);
 808 #endif
 809         break;
 810     case RBD_AIO_READ:
 811 #ifdef LIBRBD_SUPPORTS_IOVEC
 812             r = rbd_aio_readv(s->image, qiov->iov, qiov->niov, off, c);
 813 #else
 814             r = rbd_aio_read(s->image, off, size, rcb->buf, c);
 815 #endif
 816         break;
 817     case RBD_AIO_DISCARD:
 818         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 819         break;
 820     case RBD_AIO_FLUSH:
 821         r = rbd_aio_flush_wrapper(s->image, c);
 822         break;
 823     default:
 824         r = -EINVAL;
 825     }
 826
 827     if (r < 0) {
 828         goto failed_completion;
 829     }
 830     return &acb->common;
 831
 832 failed_completion:
 833     rbd_aio_release(c);
 834 failed:
 835     g_free(rcb);
 836     if (!LIBRBD_USE_IOVEC) {
 837         qemu_vfree(acb->bounce);
 838     }
 839
 840     qemu_aio_unref(acb);
 841     return NULL;
 842 }
 843
 844 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 845                                       int64_t sector_num,
 846                                       QEMUIOVector *qiov,
 847                                       int nb_sectors,
 848                                       BlockCompletionFunc *cb,
 849                                       void *opaque)
 850 {
 851     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 852                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 853                          RBD_AIO_READ);
 854 }
 855
 856 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 857                                        int64_t sector_num,
 858                                        QEMUIOVector *qiov,
 859                                        int nb_sectors,
 860                                        BlockCompletionFunc *cb,
 861                                        void *opaque)
 862 {
 863     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 864                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 865                          RBD_AIO_WRITE);
 866 }
 867
 868 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 869 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 870                                       BlockCompletionFunc *cb,
 871                                       void *opaque)
 872 {
 873     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 874 }
 875
 876 #else
 877
 878 static int qemu_rbd_co_flush(BlockDriverState *bs)
 879 {
 880 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 881     /* rbd_flush added in 0.1.1 */
 882     BDRVRBDState *s = bs->opaque;
 883     return rbd_flush(s->image);
 884 #else
 885     return 0;
 886 #endif
 887 }
 888 #endif
 889
 890 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 891 {
 892     BDRVRBDState *s = bs->opaque;
 893     rbd_image_info_t info;
 894     int r;
 895
 896     r = rbd_stat(s->image, &info, sizeof(info));
 897     if (r < 0) {
 898         return r;
 899     }
 900
 901     bdi->cluster_size = info.obj_size;
 902     return 0;
 903 }
 904
 905 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 906 {
 907     BDRVRBDState *s = bs->opaque;
 908     rbd_image_info_t info;
 909     int r;
 910
 911     r = rbd_stat(s->image, &info, sizeof(info));
 912     if (r < 0) {
 913         return r;
 914     }
 915
 916     return info.size;
 917 }
 918
 919 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset, Error **errp)
 920 {
 921     BDRVRBDState *s = bs->opaque;
 922     int r;
 923
 924     r = rbd_resize(s->image, offset);
 925     if (r < 0) {
 926         error_setg_errno(errp, -r, "Failed to resize file");
 927         return r;
 928     }
 929
 930     return 0;
 931 }
 932
 933 static int qemu_rbd_snap_create(BlockDriverState *bs,
 934                                 QEMUSnapshotInfo *sn_info)
 935 {
 936     BDRVRBDState *s = bs->opaque;
 937     int r;
 938
 939     if (sn_info->name[0] == '\0') {
 940         return -EINVAL; /* we need a name for rbd snapshots */
 941     }
 942
 943     /*
 944      * rbd snapshots are using the name as the user controlled unique identifier
 945      * we can't use the rbd snapid for that purpose, as it can't be set
 946      */
 947     if (sn_info->id_str[0] != '\0' &&
 948         strcmp(sn_info->id_str, sn_info->name) != 0) {
 949         return -EINVAL;
 950     }
 951
 952     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 953         return -ERANGE;
 954     }
 955
 956     r = rbd_snap_create(s->image, sn_info->name);
 957     if (r < 0) {
 958         error_report("failed to create snap: %s", strerror(-r));
 959         return r;
 960     }
 961
 962     return 0;
 963 }
 964
 965 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 966                                 const char *snapshot_id,
 967                                 const char *snapshot_name,
 968                                 Error **errp)
 969 {
 970     BDRVRBDState *s = bs->opaque;
 971     int r;
 972
 973     if (!snapshot_name) {
 974         error_setg(errp, "rbd need a valid snapshot name");
 975         return -EINVAL;
 976     }
 977
 978     /* If snapshot_id is specified, it must be equal to name, see
 979        qemu_rbd_snap_list() */
 980     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 981         error_setg(errp,
 982                    "rbd do not support snapshot id, it should be NULL or "
 983                    "equal to snapshot name");
 984         return -EINVAL;
 985     }
 986
 987     r = rbd_snap_remove(s->image, snapshot_name);
 988     if (r < 0) {
 989         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 990     }
 991     return r;
 992 }
 993
 994 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 995                                   const char *snapshot_name)
 996 {
 997     BDRVRBDState *s = bs->opaque;
 998
 999     return rbd_snap_rollback(s->image, snapshot_name);
1000 }
1001
1002 static int qemu_rbd_snap_list(BlockDriverState *bs,
1003                               QEMUSnapshotInfo **psn_tab)
1004 {
1005     BDRVRBDState *s = bs->opaque;
1006     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
1007     int i, snap_count;
1008     rbd_snap_info_t *snaps;
1009     int max_snaps = RBD_MAX_SNAPS;
1010
1011     do {
1012         snaps = g_new(rbd_snap_info_t, max_snaps);
1013         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
1014         if (snap_count <= 0) {
1015             g_free(snaps);
1016         }
1017     } while (snap_count == -ERANGE);
1018
1019     if (snap_count <= 0) {
1020         goto done;
1021     }
1022
1023     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
1024
1025     for (i = 0; i < snap_count; i++) {
1026         const char *snap_name = snaps[i].name;
1027
1028         sn_info = sn_tab + i;
1029         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
1030         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
1031
1032         sn_info->vm_state_size = snaps[i].size;
1033         sn_info->date_sec = 0;
1034         sn_info->date_nsec = 0;
1035         sn_info->vm_clock_nsec = 0;
1036     }
1037     rbd_snap_list_end(snaps);
1038     g_free(snaps);
1039
1040  done:
1041     *psn_tab = sn_tab;
1042     return snap_count;
1043 }
1044
1045 #ifdef LIBRBD_SUPPORTS_DISCARD
1046 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
1047                                          int64_t offset,
1048                                          int count,
1049                                          BlockCompletionFunc *cb,
1050                                          void *opaque)
1051 {
1052     return rbd_start_aio(bs, offset, NULL, count, cb, opaque,
1053                          RBD_AIO_DISCARD);
1054 }
1055 #endif
1056
1057 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1058 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
1059                                       Error **errp)
1060 {
1061     BDRVRBDState *s = bs->opaque;
1062     int r = rbd_invalidate_cache(s->image);
1063     if (r < 0) {
1064         error_setg_errno(errp, -r, "Failed to invalidate the cache");
1065     }
1066 }
1067 #endif
1068
1069 static QemuOptsList qemu_rbd_create_opts = {
1070     .name = "rbd-create-opts",
1071     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
1072     .desc = {
1073         {
1074             .name = BLOCK_OPT_SIZE,
1075             .type = QEMU_OPT_SIZE,
1076             .help = "Virtual disk size"
1077         },
1078         {
1079             .name = BLOCK_OPT_CLUSTER_SIZE,
1080             .type = QEMU_OPT_SIZE,
1081             .help = "RBD object size"
1082         },
1083         {
1084             .name = "password-secret",
1085             .type = QEMU_OPT_STRING,
1086             .help = "ID of secret providing the password",
1087         },
1088         { /* end of list */ }
1089     }
1090 };
1091
1092 static BlockDriver bdrv_rbd = {
1093     .format_name            = "rbd",
1094     .instance_size          = sizeof(BDRVRBDState),
1095     .bdrv_parse_filename    = qemu_rbd_parse_filename,
1096     .bdrv_file_open         = qemu_rbd_open,
1097     .bdrv_close             = qemu_rbd_close,
1098     .bdrv_reopen_prepare    = qemu_rbd_reopen_prepare,
1099     .bdrv_create            = qemu_rbd_create,
1100     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1101     .bdrv_get_info          = qemu_rbd_getinfo,
1102     .create_opts            = &qemu_rbd_create_opts,
1103     .bdrv_getlength         = qemu_rbd_getlength,
1104     .bdrv_truncate          = qemu_rbd_truncate,
1105     .protocol_name          = "rbd",
1106
1107     .bdrv_aio_readv         = qemu_rbd_aio_readv,
1108     .bdrv_aio_writev        = qemu_rbd_aio_writev,
1109
1110 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1111     .bdrv_aio_flush         = qemu_rbd_aio_flush,
1112 #else
1113     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
1114 #endif
1115
1116 #ifdef LIBRBD_SUPPORTS_DISCARD
1117     .bdrv_aio_pdiscard      = qemu_rbd_aio_pdiscard,
1118 #endif
1119
1120     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1121     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1122     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1123     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1124 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1125     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1126 #endif
1127 };
1128
1129 static void bdrv_rbd_init(void)
1130 {
1131     bdrv_register(&bdrv_rbd);
1132 }
1133
1134 block_init(bdrv_rbd_init);