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 *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(options, "pool", qstring_from_str(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(options, "image", qstring_from_str(found_str));
 163
 164         found_str = qemu_rbd_next_tok(p, ':', &p);
 165         qemu_rbd_unescape(found_str);
 166         qdict_put(options, "snapshot", qstring_from_str(found_str));
 167     } else {
 168         found_str = qemu_rbd_next_tok(p, ':', &p);
 169         qemu_rbd_unescape(found_str);
 170         qdict_put(options, "image", qstring_from_str(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(options, "conf", qstring_from_str(value));
 193         } else if (!strcmp(name, "id")) {
 194             qdict_put(options, "user" , qstring_from_str(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(keypairs, qstring_from_str(name));
 208             qlist_append(keypairs, qstring_from_str(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, *name, *conf, *clientname, *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     pool       = qdict_get_try_str(options, "pool");
 389     conf       = qdict_get_try_str(options, "conf");
 390     clientname = qdict_get_try_str(options, "user");
 391     name       = qdict_get_try_str(options, "image");
 392     keypairs   = qdict_get_try_str(options, "=keyvalue-pairs");
 393
 394     ret = rados_create(&cluster, clientname);
 395     if (ret < 0) {
 396         error_setg_errno(errp, -ret, "error initializing");
 397         goto exit;
 398     }
 399
 400     /* try default location when conf=NULL, but ignore failure */
 401     ret = rados_conf_read_file(cluster, conf);
 402     if (conf && ret < 0) {
 403         error_setg_errno(errp, -ret, "error reading conf file %s", conf);
 404         ret = -EIO;
 405         goto shutdown;
 406     }
 407
 408     ret = qemu_rbd_set_keypairs(cluster, keypairs, errp);
 409     if (ret < 0) {
 410         ret = -EIO;
 411         goto shutdown;
 412     }
 413
 414     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
 415         ret = -EIO;
 416         goto shutdown;
 417     }
 418
 419     ret = rados_connect(cluster);
 420     if (ret < 0) {
 421         error_setg_errno(errp, -ret, "error connecting");
 422         goto shutdown;
 423     }
 424
 425     ret = rados_ioctx_create(cluster, pool, &io_ctx);
 426     if (ret < 0) {
 427         error_setg_errno(errp, -ret, "error opening pool %s", pool);
 428         goto shutdown;
 429     }
 430
 431     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 432     if (ret < 0) {
 433         error_setg_errno(errp, -ret, "error rbd create");
 434     }
 435
 436     rados_ioctx_destroy(io_ctx);
 437
 438 shutdown:
 439     rados_shutdown(cluster);
 440
 441 exit:
 442     QDECREF(options);
 443     return ret;
 444 }
 445
 446 /*
 447  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 448  * BH context.
 449  */
 450 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 451 {
 452     RBDAIOCB *acb = rcb->acb;
 453     int64_t r;
 454
 455     r = rcb->ret;
 456
 457     if (acb->cmd != RBD_AIO_READ) {
 458         if (r < 0) {
 459             acb->ret = r;
 460             acb->error = 1;
 461         } else if (!acb->error) {
 462             acb->ret = rcb->size;
 463         }
 464     } else {
 465         if (r < 0) {
 466             qemu_rbd_memset(rcb, 0);
 467             acb->ret = r;
 468             acb->error = 1;
 469         } else if (r < rcb->size) {
 470             qemu_rbd_memset(rcb, r);
 471             if (!acb->error) {
 472                 acb->ret = rcb->size;
 473             }
 474         } else if (!acb->error) {
 475             acb->ret = r;
 476         }
 477     }
 478
 479     g_free(rcb);
 480
 481     if (!LIBRBD_USE_IOVEC) {
 482         if (acb->cmd == RBD_AIO_READ) {
 483             qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 484         }
 485         qemu_vfree(acb->bounce);
 486     }
 487
 488     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 489
 490     qemu_aio_unref(acb);
 491 }
 492
 493 static char *qemu_rbd_mon_host(QDict *options, Error **errp)
 494 {
 495     const char **vals = g_new(const char *, qdict_size(options) + 1);
 496     char keybuf[32];
 497     const char *host, *port;
 498     char *rados_str;
 499     int i;
 500
 501     for (i = 0;; i++) {
 502         sprintf(keybuf, "server.%d.host", i);
 503         host = qdict_get_try_str(options, keybuf);
 504         qdict_del(options, keybuf);
 505         sprintf(keybuf, "server.%d.port", i);
 506         port = qdict_get_try_str(options, keybuf);
 507         qdict_del(options, keybuf);
 508         if (!host && !port) {
 509             break;
 510         }
 511         if (!host) {
 512             error_setg(errp, "Parameter server.%d.host is missing", i);
 513             rados_str = NULL;
 514             goto out;
 515         }
 516
 517         if (strchr(host, ':')) {
 518             vals[i] = port ? g_strdup_printf("[%s]:%s", host, port)
 519                 : g_strdup_printf("[%s]", host);
 520         } else {
 521             vals[i] = port ? g_strdup_printf("%s:%s", host, port)
 522                 : g_strdup(host);
 523         }
 524     }
 525     vals[i] = NULL;
 526
 527     rados_str = i ? g_strjoinv(";", (char **)vals) : NULL;
 528 out:
 529     g_strfreev((char **)vals);
 530     return rados_str;
 531 }
 532
 533 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 534                          Error **errp)
 535 {
 536     BDRVRBDState *s = bs->opaque;
 537     const char *pool, *snap, *conf, *clientname, *name, *keypairs;
 538     const char *secretid;
 539     QemuOpts *opts;
 540     Error *local_err = NULL;
 541     char *mon_host = NULL;
 542     int r;
 543
 544     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 545     qemu_opts_absorb_qdict(opts, options, &local_err);
 546     if (local_err) {
 547         error_propagate(errp, local_err);
 548         r = -EINVAL;
 549         goto failed_opts;
 550     }
 551
 552     mon_host = qemu_rbd_mon_host(options, &local_err);
 553     if (local_err) {
 554         error_propagate(errp, local_err);
 555         r = -EINVAL;
 556         goto failed_opts;
 557     }
 558
 559     secretid = qemu_opt_get(opts, "password-secret");
 560
 561     pool           = qemu_opt_get(opts, "pool");
 562     conf           = qemu_opt_get(opts, "conf");
 563     snap           = qemu_opt_get(opts, "snapshot");
 564     clientname     = qemu_opt_get(opts, "user");
 565     name           = qemu_opt_get(opts, "image");
 566     keypairs       = qemu_opt_get(opts, "=keyvalue-pairs");
 567
 568     if (!pool || !name) {
 569         error_setg(errp, "Parameters 'pool' and 'image' are required");
 570         r = -EINVAL;
 571         goto failed_opts;
 572     }
 573
 574     r = rados_create(&s->cluster, clientname);
 575     if (r < 0) {
 576         error_setg_errno(errp, -r, "error initializing");
 577         goto failed_opts;
 578     }
 579
 580     s->snap = g_strdup(snap);
 581     s->name = g_strdup(name);
 582
 583     /* try default location when conf=NULL, but ignore failure */
 584     r = rados_conf_read_file(s->cluster, conf);
 585     if (conf && r < 0) {
 586         error_setg_errno(errp, -r, "error reading conf file %s", conf);
 587         goto failed_shutdown;
 588     }
 589
 590     r = qemu_rbd_set_keypairs(s->cluster, keypairs, errp);
 591     if (r < 0) {
 592         goto failed_shutdown;
 593     }
 594
 595     if (mon_host) {
 596         r = rados_conf_set(s->cluster, "mon_host", mon_host);
 597         if (r < 0) {
 598             goto failed_shutdown;
 599         }
 600     }
 601
 602     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
 603         r = -EIO;
 604         goto failed_shutdown;
 605     }
 606
 607     /*
 608      * Fallback to more conservative semantics if setting cache
 609      * options fails. Ignore errors from setting rbd_cache because the
 610      * only possible error is that the option does not exist, and
 611      * librbd defaults to no caching. If write through caching cannot
 612      * be set up, fall back to no caching.
 613      */
 614     if (flags & BDRV_O_NOCACHE) {
 615         rados_conf_set(s->cluster, "rbd_cache", "false");
 616     } else {
 617         rados_conf_set(s->cluster, "rbd_cache", "true");
 618     }
 619
 620     r = rados_connect(s->cluster);
 621     if (r < 0) {
 622         error_setg_errno(errp, -r, "error connecting");
 623         goto failed_shutdown;
 624     }
 625
 626     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 627     if (r < 0) {
 628         error_setg_errno(errp, -r, "error opening pool %s", pool);
 629         goto failed_shutdown;
 630     }
 631
 632     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 633     if (r < 0) {
 634         error_setg_errno(errp, -r, "error reading header from %s", s->name);
 635         goto failed_open;
 636     }
 637
 638     bs->read_only = (s->snap != NULL);
 639
 640     qemu_opts_del(opts);
 641     return 0;
 642
 643 failed_open:
 644     rados_ioctx_destroy(s->io_ctx);
 645 failed_shutdown:
 646     rados_shutdown(s->cluster);
 647     g_free(s->snap);
 648     g_free(s->name);
 649 failed_opts:
 650     qemu_opts_del(opts);
 651     g_free(mon_host);
 652     return r;
 653 }
 654
 655 static void qemu_rbd_close(BlockDriverState *bs)
 656 {
 657     BDRVRBDState *s = bs->opaque;
 658
 659     rbd_close(s->image);
 660     rados_ioctx_destroy(s->io_ctx);
 661     g_free(s->snap);
 662     g_free(s->name);
 663     rados_shutdown(s->cluster);
 664 }
 665
 666 static const AIOCBInfo rbd_aiocb_info = {
 667     .aiocb_size = sizeof(RBDAIOCB),
 668 };
 669
 670 static void rbd_finish_bh(void *opaque)
 671 {
 672     RADOSCB *rcb = opaque;
 673     qemu_rbd_complete_aio(rcb);
 674 }
 675
 676 /*
 677  * This is the callback function for rbd_aio_read and _write
 678  *
 679  * Note: this function is being called from a non qemu thread so
 680  * we need to be careful about what we do here. Generally we only
 681  * schedule a BH, and do the rest of the io completion handling
 682  * from rbd_finish_bh() which runs in a qemu context.
 683  */
 684 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 685 {
 686     RBDAIOCB *acb = rcb->acb;
 687
 688     rcb->ret = rbd_aio_get_return_value(c);
 689     rbd_aio_release(c);
 690
 691     aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs),
 692                             rbd_finish_bh, rcb);
 693 }
 694
 695 static int rbd_aio_discard_wrapper(rbd_image_t image,
 696                                    uint64_t off,
 697                                    uint64_t len,
 698                                    rbd_completion_t comp)
 699 {
 700 #ifdef LIBRBD_SUPPORTS_DISCARD
 701     return rbd_aio_discard(image, off, len, comp);
 702 #else
 703     return -ENOTSUP;
 704 #endif
 705 }
 706
 707 static int rbd_aio_flush_wrapper(rbd_image_t image,
 708                                  rbd_completion_t comp)
 709 {
 710 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 711     return rbd_aio_flush(image, comp);
 712 #else
 713     return -ENOTSUP;
 714 #endif
 715 }
 716
 717 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 718                                  int64_t off,
 719                                  QEMUIOVector *qiov,
 720                                  int64_t size,
 721                                  BlockCompletionFunc *cb,
 722                                  void *opaque,
 723                                  RBDAIOCmd cmd)
 724 {
 725     RBDAIOCB *acb;
 726     RADOSCB *rcb = NULL;
 727     rbd_completion_t c;
 728     int r;
 729
 730     BDRVRBDState *s = bs->opaque;
 731
 732     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 733     acb->cmd = cmd;
 734     acb->qiov = qiov;
 735     assert(!qiov || qiov->size == size);
 736
 737     rcb = g_new(RADOSCB, 1);
 738
 739     if (!LIBRBD_USE_IOVEC) {
 740         if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 741             acb->bounce = NULL;
 742         } else {
 743             acb->bounce = qemu_try_blockalign(bs, qiov->size);
 744             if (acb->bounce == NULL) {
 745                 goto failed;
 746             }
 747         }
 748         if (cmd == RBD_AIO_WRITE) {
 749             qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 750         }
 751         rcb->buf = acb->bounce;
 752     }
 753
 754     acb->ret = 0;
 755     acb->error = 0;
 756     acb->s = s;
 757
 758     rcb->acb = acb;
 759     rcb->s = acb->s;
 760     rcb->size = size;
 761     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 762     if (r < 0) {
 763         goto failed;
 764     }
 765
 766     switch (cmd) {
 767     case RBD_AIO_WRITE:
 768 #ifdef LIBRBD_SUPPORTS_IOVEC
 769             r = rbd_aio_writev(s->image, qiov->iov, qiov->niov, off, c);
 770 #else
 771             r = rbd_aio_write(s->image, off, size, rcb->buf, c);
 772 #endif
 773         break;
 774     case RBD_AIO_READ:
 775 #ifdef LIBRBD_SUPPORTS_IOVEC
 776             r = rbd_aio_readv(s->image, qiov->iov, qiov->niov, off, c);
 777 #else
 778             r = rbd_aio_read(s->image, off, size, rcb->buf, c);
 779 #endif
 780         break;
 781     case RBD_AIO_DISCARD:
 782         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 783         break;
 784     case RBD_AIO_FLUSH:
 785         r = rbd_aio_flush_wrapper(s->image, c);
 786         break;
 787     default:
 788         r = -EINVAL;
 789     }
 790
 791     if (r < 0) {
 792         goto failed_completion;
 793     }
 794     return &acb->common;
 795
 796 failed_completion:
 797     rbd_aio_release(c);
 798 failed:
 799     g_free(rcb);
 800     if (!LIBRBD_USE_IOVEC) {
 801         qemu_vfree(acb->bounce);
 802     }
 803
 804     qemu_aio_unref(acb);
 805     return NULL;
 806 }
 807
 808 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 809                                       int64_t sector_num,
 810                                       QEMUIOVector *qiov,
 811                                       int nb_sectors,
 812                                       BlockCompletionFunc *cb,
 813                                       void *opaque)
 814 {
 815     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 816                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 817                          RBD_AIO_READ);
 818 }
 819
 820 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 821                                        int64_t sector_num,
 822                                        QEMUIOVector *qiov,
 823                                        int nb_sectors,
 824                                        BlockCompletionFunc *cb,
 825                                        void *opaque)
 826 {
 827     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 828                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 829                          RBD_AIO_WRITE);
 830 }
 831
 832 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 833 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 834                                       BlockCompletionFunc *cb,
 835                                       void *opaque)
 836 {
 837     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 838 }
 839
 840 #else
 841
 842 static int qemu_rbd_co_flush(BlockDriverState *bs)
 843 {
 844 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 845     /* rbd_flush added in 0.1.1 */
 846     BDRVRBDState *s = bs->opaque;
 847     return rbd_flush(s->image);
 848 #else
 849     return 0;
 850 #endif
 851 }
 852 #endif
 853
 854 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 855 {
 856     BDRVRBDState *s = bs->opaque;
 857     rbd_image_info_t info;
 858     int r;
 859
 860     r = rbd_stat(s->image, &info, sizeof(info));
 861     if (r < 0) {
 862         return r;
 863     }
 864
 865     bdi->cluster_size = info.obj_size;
 866     return 0;
 867 }
 868
 869 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 870 {
 871     BDRVRBDState *s = bs->opaque;
 872     rbd_image_info_t info;
 873     int r;
 874
 875     r = rbd_stat(s->image, &info, sizeof(info));
 876     if (r < 0) {
 877         return r;
 878     }
 879
 880     return info.size;
 881 }
 882
 883 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 884 {
 885     BDRVRBDState *s = bs->opaque;
 886     int r;
 887
 888     r = rbd_resize(s->image, offset);
 889     if (r < 0) {
 890         return r;
 891     }
 892
 893     return 0;
 894 }
 895
 896 static int qemu_rbd_snap_create(BlockDriverState *bs,
 897                                 QEMUSnapshotInfo *sn_info)
 898 {
 899     BDRVRBDState *s = bs->opaque;
 900     int r;
 901
 902     if (sn_info->name[0] == '\0') {
 903         return -EINVAL; /* we need a name for rbd snapshots */
 904     }
 905
 906     /*
 907      * rbd snapshots are using the name as the user controlled unique identifier
 908      * we can't use the rbd snapid for that purpose, as it can't be set
 909      */
 910     if (sn_info->id_str[0] != '\0' &&
 911         strcmp(sn_info->id_str, sn_info->name) != 0) {
 912         return -EINVAL;
 913     }
 914
 915     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 916         return -ERANGE;
 917     }
 918
 919     r = rbd_snap_create(s->image, sn_info->name);
 920     if (r < 0) {
 921         error_report("failed to create snap: %s", strerror(-r));
 922         return r;
 923     }
 924
 925     return 0;
 926 }
 927
 928 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 929                                 const char *snapshot_id,
 930                                 const char *snapshot_name,
 931                                 Error **errp)
 932 {
 933     BDRVRBDState *s = bs->opaque;
 934     int r;
 935
 936     if (!snapshot_name) {
 937         error_setg(errp, "rbd need a valid snapshot name");
 938         return -EINVAL;
 939     }
 940
 941     /* If snapshot_id is specified, it must be equal to name, see
 942        qemu_rbd_snap_list() */
 943     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 944         error_setg(errp,
 945                    "rbd do not support snapshot id, it should be NULL or "
 946                    "equal to snapshot name");
 947         return -EINVAL;
 948     }
 949
 950     r = rbd_snap_remove(s->image, snapshot_name);
 951     if (r < 0) {
 952         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 953     }
 954     return r;
 955 }
 956
 957 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 958                                   const char *snapshot_name)
 959 {
 960     BDRVRBDState *s = bs->opaque;
 961
 962     return rbd_snap_rollback(s->image, snapshot_name);
 963 }
 964
 965 static int qemu_rbd_snap_list(BlockDriverState *bs,
 966                               QEMUSnapshotInfo **psn_tab)
 967 {
 968     BDRVRBDState *s = bs->opaque;
 969     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 970     int i, snap_count;
 971     rbd_snap_info_t *snaps;
 972     int max_snaps = RBD_MAX_SNAPS;
 973
 974     do {
 975         snaps = g_new(rbd_snap_info_t, max_snaps);
 976         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 977         if (snap_count <= 0) {
 978             g_free(snaps);
 979         }
 980     } while (snap_count == -ERANGE);
 981
 982     if (snap_count <= 0) {
 983         goto done;
 984     }
 985
 986     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
 987
 988     for (i = 0; i < snap_count; i++) {
 989         const char *snap_name = snaps[i].name;
 990
 991         sn_info = sn_tab + i;
 992         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 993         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 994
 995         sn_info->vm_state_size = snaps[i].size;
 996         sn_info->date_sec = 0;
 997         sn_info->date_nsec = 0;
 998         sn_info->vm_clock_nsec = 0;
 999     }
1000     rbd_snap_list_end(snaps);
1001     g_free(snaps);
1002
1003  done:
1004     *psn_tab = sn_tab;
1005     return snap_count;
1006 }
1007
1008 #ifdef LIBRBD_SUPPORTS_DISCARD
1009 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
1010                                          int64_t offset,
1011                                          int count,
1012                                          BlockCompletionFunc *cb,
1013                                          void *opaque)
1014 {
1015     return rbd_start_aio(bs, offset, NULL, count, cb, opaque,
1016                          RBD_AIO_DISCARD);
1017 }
1018 #endif
1019
1020 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1021 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
1022                                       Error **errp)
1023 {
1024     BDRVRBDState *s = bs->opaque;
1025     int r = rbd_invalidate_cache(s->image);
1026     if (r < 0) {
1027         error_setg_errno(errp, -r, "Failed to invalidate the cache");
1028     }
1029 }
1030 #endif
1031
1032 static QemuOptsList qemu_rbd_create_opts = {
1033     .name = "rbd-create-opts",
1034     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
1035     .desc = {
1036         {
1037             .name = BLOCK_OPT_SIZE,
1038             .type = QEMU_OPT_SIZE,
1039             .help = "Virtual disk size"
1040         },
1041         {
1042             .name = BLOCK_OPT_CLUSTER_SIZE,
1043             .type = QEMU_OPT_SIZE,
1044             .help = "RBD object size"
1045         },
1046         {
1047             .name = "password-secret",
1048             .type = QEMU_OPT_STRING,
1049             .help = "ID of secret providing the password",
1050         },
1051         { /* end of list */ }
1052     }
1053 };
1054
1055 static BlockDriver bdrv_rbd = {
1056     .format_name            = "rbd",
1057     .instance_size          = sizeof(BDRVRBDState),
1058     .bdrv_parse_filename    = qemu_rbd_parse_filename,
1059     .bdrv_file_open         = qemu_rbd_open,
1060     .bdrv_close             = qemu_rbd_close,
1061     .bdrv_create            = qemu_rbd_create,
1062     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1063     .bdrv_get_info          = qemu_rbd_getinfo,
1064     .create_opts            = &qemu_rbd_create_opts,
1065     .bdrv_getlength         = qemu_rbd_getlength,
1066     .bdrv_truncate          = qemu_rbd_truncate,
1067     .protocol_name          = "rbd",
1068
1069     .bdrv_aio_readv         = qemu_rbd_aio_readv,
1070     .bdrv_aio_writev        = qemu_rbd_aio_writev,
1071
1072 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1073     .bdrv_aio_flush         = qemu_rbd_aio_flush,
1074 #else
1075     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
1076 #endif
1077
1078 #ifdef LIBRBD_SUPPORTS_DISCARD
1079     .bdrv_aio_pdiscard      = qemu_rbd_aio_pdiscard,
1080 #endif
1081
1082     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1083     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1084     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1085     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1086 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1087     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1088 #endif
1089 };
1090
1091 static void bdrv_rbd_init(void)
1092 {
1093     bdrv_register(&bdrv_rbd);
1094 }
1095
1096 block_init(bdrv_rbd_init);