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 "qapi/error.h"
  17 #include "qemu/error-report.h"
  18 #include "block/block_int.h"
  19 #include "crypto/secret.h"
  20 #include "qemu/cutils.h"
  21 #include "qapi/qmp/qstring.h"
  22
  23 #include <rbd/librbd.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, *keypairs;
 140     char *found_str;
 141     size_t max_keypair_size;
 142
 143     if (!strstart(filename, "rbd:", &start)) {
 144         error_setg(errp, "File name must start with 'rbd:'");
 145         return;
 146     }
 147
 148     max_keypair_size = strlen(start) + 1;
 149     buf = g_strdup(start);
 150     keypairs = g_malloc0(max_keypair_size);
 151     p = buf;
 152
 153     found_str = qemu_rbd_next_tok(p, '/', &p);
 154     if (!p) {
 155         error_setg(errp, "Pool name is required");
 156         goto done;
 157     }
 158     qemu_rbd_unescape(found_str);
 159     qdict_put(options, "pool", qstring_from_str(found_str));
 160
 161     if (strchr(p, '@')) {
 162         found_str = qemu_rbd_next_tok(p, '@', &p);
 163         qemu_rbd_unescape(found_str);
 164         qdict_put(options, "image", qstring_from_str(found_str));
 165
 166         found_str = qemu_rbd_next_tok(p, ':', &p);
 167         qemu_rbd_unescape(found_str);
 168         qdict_put(options, "snapshot", qstring_from_str(found_str));
 169     } else {
 170         found_str = qemu_rbd_next_tok(p, ':', &p);
 171         qemu_rbd_unescape(found_str);
 172         qdict_put(options, "image", qstring_from_str(found_str));
 173     }
 174     if (!p) {
 175         goto done;
 176     }
 177
 178     /* The following are essentially all key/value pairs, and we treat
 179      * 'id' and 'conf' a bit special.  Key/value pairs may be in any order. */
 180     while (p) {
 181         char *name, *value;
 182         name = qemu_rbd_next_tok(p, '=', &p);
 183         if (!p) {
 184             error_setg(errp, "conf option %s has no value", name);
 185             break;
 186         }
 187
 188         qemu_rbd_unescape(name);
 189
 190         value = qemu_rbd_next_tok(p, ':', &p);
 191         qemu_rbd_unescape(value);
 192
 193         if (!strcmp(name, "conf")) {
 194             qdict_put(options, "conf", qstring_from_str(value));
 195         } else if (!strcmp(name, "id")) {
 196             qdict_put(options, "user" , qstring_from_str(value));
 197         } else {
 198             /* FIXME: This is pretty ugly, and not the right way to do this.
 199              *        These should be contained in a structure, and then
 200              *        passed explicitly as individual key/value pairs to
 201              *        rados.  Consider this legacy code that needs to be
 202              *        updated. */
 203             char *tmp = g_malloc0(max_keypair_size);
 204             /* only use a delimiter if it is not the first keypair found */
 205             /* These are sets of unknown key/value pairs we'll pass along
 206              * to ceph */
 207             if (keypairs[0]) {
 208                 snprintf(tmp, max_keypair_size, ":%s=%s", name, value);
 209                 pstrcat(keypairs, max_keypair_size, tmp);
 210             } else {
 211                 snprintf(keypairs, max_keypair_size, "%s=%s", name, value);
 212             }
 213             g_free(tmp);
 214         }
 215     }
 216
 217     if (keypairs[0]) {
 218         qdict_put(options, "=keyvalue-pairs", qstring_from_str(keypairs));
 219     }
 220
 221
 222 done:
 223     g_free(buf);
 224     g_free(keypairs);
 225     return;
 226 }
 227
 228
 229 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
 230                              Error **errp)
 231 {
 232     if (secretid == 0) {
 233         return 0;
 234     }
 235
 236     gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
 237                                                     errp);
 238     if (!secret) {
 239         return -1;
 240     }
 241
 242     rados_conf_set(cluster, "key", secret);
 243     g_free(secret);
 244
 245     return 0;
 246 }
 247
 248 static int qemu_rbd_set_keypairs(rados_t cluster, const char *keypairs,
 249                                  Error **errp)
 250 {
 251     char *p, *buf;
 252     char *name;
 253     char *value;
 254     int ret = 0;
 255
 256     buf = g_strdup(keypairs);
 257     p = buf;
 258
 259     while (p) {
 260         name = qemu_rbd_next_tok(p, '=', &p);
 261         if (!p) {
 262             error_setg(errp, "conf option %s has no value", name);
 263             ret = -EINVAL;
 264             break;
 265         }
 266
 267         value = qemu_rbd_next_tok(p, ':', &p);
 268
 269         ret = rados_conf_set(cluster, name, value);
 270         if (ret < 0) {
 271             error_setg_errno(errp, -ret, "invalid conf option %s", name);
 272             ret = -EINVAL;
 273             break;
 274         }
 275     }
 276
 277     g_free(buf);
 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 = "filename",
 298             .type = QEMU_OPT_STRING,
 299             .help = "Specification of the rbd image",
 300         },
 301         {
 302             .name = "password-secret",
 303             .type = QEMU_OPT_STRING,
 304             .help = "ID of secret providing the password",
 305         },
 306         {
 307             .name = "conf",
 308             .type = QEMU_OPT_STRING,
 309             .help = "Rados config file location",
 310         },
 311         {
 312             .name = "pool",
 313             .type = QEMU_OPT_STRING,
 314             .help = "Rados pool name",
 315         },
 316         {
 317             .name = "image",
 318             .type = QEMU_OPT_STRING,
 319             .help = "Image name in the pool",
 320         },
 321         {
 322             .name = "snapshot",
 323             .type = QEMU_OPT_STRING,
 324             .help = "Ceph snapshot name",
 325         },
 326         {
 327             /* maps to 'id' in rados_create() */
 328             .name = "user",
 329             .type = QEMU_OPT_STRING,
 330             .help = "Rados id name",
 331         },
 332         {
 333             /*
 334              * HACK: name starts with '=' so that qemu_opts_parse()
 335              * can't set it
 336              */
 337             .name = "=keyvalue-pairs",
 338             .type = QEMU_OPT_STRING,
 339             .help = "Legacy rados key/value option parameters",
 340         },
 341         {
 342             .name = "host",
 343             .type = QEMU_OPT_STRING,
 344         },
 345         {
 346             .name = "port",
 347             .type = QEMU_OPT_STRING,
 348         },
 349         {
 350             .name = "auth",
 351             .type = QEMU_OPT_STRING,
 352             .help = "Supported authentication method, either cephx or none",
 353         },
 354         { /* end of list */ }
 355     },
 356 };
 357
 358 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 359 {
 360     Error *local_err = NULL;
 361     int64_t bytes = 0;
 362     int64_t objsize;
 363     int obj_order = 0;
 364     const char *pool, *name, *conf, *clientname, *keypairs;
 365     const char *secretid;
 366     rados_t cluster;
 367     rados_ioctx_t io_ctx;
 368     QDict *options = NULL;
 369     QemuOpts *rbd_opts = NULL;
 370     int ret = 0;
 371
 372     secretid = qemu_opt_get(opts, "password-secret");
 373
 374     /* Read out options */
 375     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 376                      BDRV_SECTOR_SIZE);
 377     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 378     if (objsize) {
 379         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 380             error_setg(errp, "obj size needs to be power of 2");
 381             ret = -EINVAL;
 382             goto exit;
 383         }
 384         if (objsize < 4096) {
 385             error_setg(errp, "obj size too small");
 386             ret = -EINVAL;
 387             goto exit;
 388         }
 389         obj_order = ctz32(objsize);
 390     }
 391
 392     options = qdict_new();
 393     qemu_rbd_parse_filename(filename, options, &local_err);
 394     if (local_err) {
 395         ret = -EINVAL;
 396         error_propagate(errp, local_err);
 397         goto exit;
 398     }
 399
 400     rbd_opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 401     qemu_opts_absorb_qdict(rbd_opts, options, &local_err);
 402     if (local_err) {
 403         error_propagate(errp, local_err);
 404         ret = -EINVAL;
 405         goto exit;
 406     }
 407
 408     pool       = qemu_opt_get(rbd_opts, "pool");
 409     conf       = qemu_opt_get(rbd_opts, "conf");
 410     clientname = qemu_opt_get(rbd_opts, "user");
 411     name       = qemu_opt_get(rbd_opts, "image");
 412     keypairs   = qemu_opt_get(rbd_opts, "=keyvalue-pairs");
 413
 414     ret = rados_create(&cluster, clientname);
 415     if (ret < 0) {
 416         error_setg_errno(errp, -ret, "error initializing");
 417         goto exit;
 418     }
 419
 420     /* try default location when conf=NULL, but ignore failure */
 421     ret = rados_conf_read_file(cluster, conf);
 422     if (conf && ret < 0) {
 423         error_setg_errno(errp, -ret, "error reading conf file %s", conf);
 424         ret = -EIO;
 425         goto shutdown;
 426     }
 427
 428     ret = qemu_rbd_set_keypairs(cluster, keypairs, errp);
 429     if (ret < 0) {
 430         ret = -EIO;
 431         goto shutdown;
 432     }
 433
 434     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
 435         ret = -EIO;
 436         goto shutdown;
 437     }
 438
 439     ret = rados_connect(cluster);
 440     if (ret < 0) {
 441         error_setg_errno(errp, -ret, "error connecting");
 442         goto shutdown;
 443     }
 444
 445     ret = rados_ioctx_create(cluster, pool, &io_ctx);
 446     if (ret < 0) {
 447         error_setg_errno(errp, -ret, "error opening pool %s", pool);
 448         goto shutdown;
 449     }
 450
 451     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 452     if (ret < 0) {
 453         error_setg_errno(errp, -ret, "error rbd create");
 454     }
 455
 456     rados_ioctx_destroy(io_ctx);
 457
 458 shutdown:
 459     rados_shutdown(cluster);
 460
 461 exit:
 462     QDECREF(options);
 463     qemu_opts_del(rbd_opts);
 464     return ret;
 465 }
 466
 467 /*
 468  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 469  * BH context.
 470  */
 471 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 472 {
 473     RBDAIOCB *acb = rcb->acb;
 474     int64_t r;
 475
 476     r = rcb->ret;
 477
 478     if (acb->cmd != RBD_AIO_READ) {
 479         if (r < 0) {
 480             acb->ret = r;
 481             acb->error = 1;
 482         } else if (!acb->error) {
 483             acb->ret = rcb->size;
 484         }
 485     } else {
 486         if (r < 0) {
 487             qemu_rbd_memset(rcb, 0);
 488             acb->ret = r;
 489             acb->error = 1;
 490         } else if (r < rcb->size) {
 491             qemu_rbd_memset(rcb, r);
 492             if (!acb->error) {
 493                 acb->ret = rcb->size;
 494             }
 495         } else if (!acb->error) {
 496             acb->ret = r;
 497         }
 498     }
 499
 500     g_free(rcb);
 501
 502     if (!LIBRBD_USE_IOVEC) {
 503         if (acb->cmd == RBD_AIO_READ) {
 504             qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 505         }
 506         qemu_vfree(acb->bounce);
 507     }
 508
 509     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 510
 511     qemu_aio_unref(acb);
 512 }
 513
 514 #define RBD_MON_HOST          0
 515 #define RBD_AUTH_SUPPORTED    1
 516
 517 static char *qemu_rbd_array_opts(QDict *options, const char *prefix, int type,
 518                                  Error **errp)
 519 {
 520     int num_entries;
 521     QemuOpts *opts = NULL;
 522     QDict *sub_options;
 523     const char *host;
 524     const char *port;
 525     char *str;
 526     char *rados_str = NULL;
 527     Error *local_err = NULL;
 528     int i;
 529
 530     assert(type == RBD_MON_HOST || type == RBD_AUTH_SUPPORTED);
 531
 532     num_entries = qdict_array_entries(options, prefix);
 533
 534     if (num_entries < 0) {
 535         error_setg(errp, "Parse error on RBD QDict array");
 536         return NULL;
 537     }
 538
 539     for (i = 0; i < num_entries; i++) {
 540         char *strbuf = NULL;
 541         const char *value;
 542         char *rados_str_tmp;
 543
 544         str = g_strdup_printf("%s%d.", prefix, i);
 545         qdict_extract_subqdict(options, &sub_options, str);
 546         g_free(str);
 547
 548         opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 549         qemu_opts_absorb_qdict(opts, sub_options, &local_err);
 550         QDECREF(sub_options);
 551         if (local_err) {
 552             error_propagate(errp, local_err);
 553             g_free(rados_str);
 554             rados_str = NULL;
 555             goto exit;
 556         }
 557
 558         if (type == RBD_MON_HOST) {
 559             host = qemu_opt_get(opts, "host");
 560             port = qemu_opt_get(opts, "port");
 561
 562             value = host;
 563             if (port) {
 564                 /* check for ipv6 */
 565                 if (strchr(host, ':')) {
 566                     strbuf = g_strdup_printf("[%s]:%s", host, port);
 567                 } else {
 568                     strbuf = g_strdup_printf("%s:%s", host, port);
 569                 }
 570                 value = strbuf;
 571             } else if (strchr(host, ':')) {
 572                 strbuf = g_strdup_printf("[%s]", host);
 573                 value = strbuf;
 574             }
 575         } else {
 576             value = qemu_opt_get(opts, "auth");
 577         }
 578
 579
 580         /* each iteration in the for loop will build upon the string, and if
 581          * rados_str is NULL then it is our first pass */
 582         if (rados_str) {
 583             /* separate options with ';', as that  is what rados_conf_set()
 584              * requires */
 585             rados_str_tmp = rados_str;
 586             rados_str = g_strdup_printf("%s;%s", rados_str_tmp, value);
 587             g_free(rados_str_tmp);
 588         } else {
 589             rados_str = g_strdup(value);
 590         }
 591
 592         g_free(strbuf);
 593         qemu_opts_del(opts);
 594         opts = NULL;
 595     }
 596
 597 exit:
 598     qemu_opts_del(opts);
 599     return rados_str;
 600 }
 601
 602 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 603                          Error **errp)
 604 {
 605     BDRVRBDState *s = bs->opaque;
 606     const char *pool, *snap, *conf, *clientname, *name, *keypairs;
 607     const char *secretid;
 608     QemuOpts *opts;
 609     Error *local_err = NULL;
 610     char *mon_host = NULL;
 611     char *auth_supported = NULL;
 612     int r;
 613
 614     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 615     qemu_opts_absorb_qdict(opts, options, &local_err);
 616     if (local_err) {
 617         error_propagate(errp, local_err);
 618         qemu_opts_del(opts);
 619         return -EINVAL;
 620     }
 621
 622     auth_supported = qemu_rbd_array_opts(options, "auth-supported.",
 623                                          RBD_AUTH_SUPPORTED, &local_err);
 624     if (local_err) {
 625         error_propagate(errp, local_err);
 626         r = -EINVAL;
 627         goto failed_opts;
 628     }
 629
 630     mon_host = qemu_rbd_array_opts(options, "server.",
 631                                    RBD_MON_HOST, &local_err);
 632     if (local_err) {
 633         error_propagate(errp, local_err);
 634         r = -EINVAL;
 635         goto failed_opts;
 636     }
 637
 638     secretid = qemu_opt_get(opts, "password-secret");
 639
 640     pool           = qemu_opt_get(opts, "pool");
 641     conf           = qemu_opt_get(opts, "conf");
 642     snap           = qemu_opt_get(opts, "snapshot");
 643     clientname     = qemu_opt_get(opts, "user");
 644     name           = qemu_opt_get(opts, "image");
 645     keypairs       = qemu_opt_get(opts, "=keyvalue-pairs");
 646
 647     if (!pool || !name) {
 648         error_setg(errp, "Parameters 'pool' and 'image' are required");
 649         r = -EINVAL;
 650         goto failed_opts;
 651     }
 652
 653     r = rados_create(&s->cluster, clientname);
 654     if (r < 0) {
 655         error_setg_errno(errp, -r, "error initializing");
 656         goto failed_opts;
 657     }
 658
 659     s->snap = g_strdup(snap);
 660     s->name = g_strdup(name);
 661
 662     /* try default location when conf=NULL, but ignore failure */
 663     r = rados_conf_read_file(s->cluster, conf);
 664     if (conf && r < 0) {
 665         error_setg_errno(errp, -r, "error reading conf file %s", conf);
 666         goto failed_shutdown;
 667     }
 668
 669     r = qemu_rbd_set_keypairs(s->cluster, keypairs, errp);
 670     if (r < 0) {
 671         goto failed_shutdown;
 672     }
 673
 674     if (mon_host) {
 675         r = rados_conf_set(s->cluster, "mon_host", mon_host);
 676         if (r < 0) {
 677             goto failed_shutdown;
 678         }
 679     }
 680
 681     if (auth_supported) {
 682         r = rados_conf_set(s->cluster, "auth_supported", auth_supported);
 683         if (r < 0) {
 684             goto failed_shutdown;
 685         }
 686     }
 687
 688     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
 689         r = -EIO;
 690         goto failed_shutdown;
 691     }
 692
 693     /*
 694      * Fallback to more conservative semantics if setting cache
 695      * options fails. Ignore errors from setting rbd_cache because the
 696      * only possible error is that the option does not exist, and
 697      * librbd defaults to no caching. If write through caching cannot
 698      * be set up, fall back to no caching.
 699      */
 700     if (flags & BDRV_O_NOCACHE) {
 701         rados_conf_set(s->cluster, "rbd_cache", "false");
 702     } else {
 703         rados_conf_set(s->cluster, "rbd_cache", "true");
 704     }
 705
 706     r = rados_connect(s->cluster);
 707     if (r < 0) {
 708         error_setg_errno(errp, -r, "error connecting");
 709         goto failed_shutdown;
 710     }
 711
 712     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 713     if (r < 0) {
 714         error_setg_errno(errp, -r, "error opening pool %s", pool);
 715         goto failed_shutdown;
 716     }
 717
 718     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 719     if (r < 0) {
 720         error_setg_errno(errp, -r, "error reading header from %s", s->name);
 721         goto failed_open;
 722     }
 723
 724     bs->read_only = (s->snap != NULL);
 725
 726     qemu_opts_del(opts);
 727     return 0;
 728
 729 failed_open:
 730     rados_ioctx_destroy(s->io_ctx);
 731 failed_shutdown:
 732     rados_shutdown(s->cluster);
 733     g_free(s->snap);
 734     g_free(s->name);
 735 failed_opts:
 736     qemu_opts_del(opts);
 737     g_free(mon_host);
 738     g_free(auth_supported);
 739     return r;
 740 }
 741
 742 static void qemu_rbd_close(BlockDriverState *bs)
 743 {
 744     BDRVRBDState *s = bs->opaque;
 745
 746     rbd_close(s->image);
 747     rados_ioctx_destroy(s->io_ctx);
 748     g_free(s->snap);
 749     g_free(s->name);
 750     rados_shutdown(s->cluster);
 751 }
 752
 753 static const AIOCBInfo rbd_aiocb_info = {
 754     .aiocb_size = sizeof(RBDAIOCB),
 755 };
 756
 757 static void rbd_finish_bh(void *opaque)
 758 {
 759     RADOSCB *rcb = opaque;
 760     qemu_rbd_complete_aio(rcb);
 761 }
 762
 763 /*
 764  * This is the callback function for rbd_aio_read and _write
 765  *
 766  * Note: this function is being called from a non qemu thread so
 767  * we need to be careful about what we do here. Generally we only
 768  * schedule a BH, and do the rest of the io completion handling
 769  * from rbd_finish_bh() which runs in a qemu context.
 770  */
 771 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 772 {
 773     RBDAIOCB *acb = rcb->acb;
 774
 775     rcb->ret = rbd_aio_get_return_value(c);
 776     rbd_aio_release(c);
 777
 778     aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs),
 779                             rbd_finish_bh, rcb);
 780 }
 781
 782 static int rbd_aio_discard_wrapper(rbd_image_t image,
 783                                    uint64_t off,
 784                                    uint64_t len,
 785                                    rbd_completion_t comp)
 786 {
 787 #ifdef LIBRBD_SUPPORTS_DISCARD
 788     return rbd_aio_discard(image, off, len, comp);
 789 #else
 790     return -ENOTSUP;
 791 #endif
 792 }
 793
 794 static int rbd_aio_flush_wrapper(rbd_image_t image,
 795                                  rbd_completion_t comp)
 796 {
 797 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 798     return rbd_aio_flush(image, comp);
 799 #else
 800     return -ENOTSUP;
 801 #endif
 802 }
 803
 804 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 805                                  int64_t off,
 806                                  QEMUIOVector *qiov,
 807                                  int64_t size,
 808                                  BlockCompletionFunc *cb,
 809                                  void *opaque,
 810                                  RBDAIOCmd cmd)
 811 {
 812     RBDAIOCB *acb;
 813     RADOSCB *rcb = NULL;
 814     rbd_completion_t c;
 815     int r;
 816
 817     BDRVRBDState *s = bs->opaque;
 818
 819     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 820     acb->cmd = cmd;
 821     acb->qiov = qiov;
 822     assert(!qiov || qiov->size == size);
 823
 824     rcb = g_new(RADOSCB, 1);
 825
 826     if (!LIBRBD_USE_IOVEC) {
 827         if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 828             acb->bounce = NULL;
 829         } else {
 830             acb->bounce = qemu_try_blockalign(bs, qiov->size);
 831             if (acb->bounce == NULL) {
 832                 goto failed;
 833             }
 834         }
 835         if (cmd == RBD_AIO_WRITE) {
 836             qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 837         }
 838         rcb->buf = acb->bounce;
 839     }
 840
 841     acb->ret = 0;
 842     acb->error = 0;
 843     acb->s = s;
 844
 845     rcb->acb = acb;
 846     rcb->s = acb->s;
 847     rcb->size = size;
 848     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 849     if (r < 0) {
 850         goto failed;
 851     }
 852
 853     switch (cmd) {
 854     case RBD_AIO_WRITE:
 855 #ifdef LIBRBD_SUPPORTS_IOVEC
 856             r = rbd_aio_writev(s->image, qiov->iov, qiov->niov, off, c);
 857 #else
 858             r = rbd_aio_write(s->image, off, size, rcb->buf, c);
 859 #endif
 860         break;
 861     case RBD_AIO_READ:
 862 #ifdef LIBRBD_SUPPORTS_IOVEC
 863             r = rbd_aio_readv(s->image, qiov->iov, qiov->niov, off, c);
 864 #else
 865             r = rbd_aio_read(s->image, off, size, rcb->buf, c);
 866 #endif
 867         break;
 868     case RBD_AIO_DISCARD:
 869         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 870         break;
 871     case RBD_AIO_FLUSH:
 872         r = rbd_aio_flush_wrapper(s->image, c);
 873         break;
 874     default:
 875         r = -EINVAL;
 876     }
 877
 878     if (r < 0) {
 879         goto failed_completion;
 880     }
 881     return &acb->common;
 882
 883 failed_completion:
 884     rbd_aio_release(c);
 885 failed:
 886     g_free(rcb);
 887     if (!LIBRBD_USE_IOVEC) {
 888         qemu_vfree(acb->bounce);
 889     }
 890
 891     qemu_aio_unref(acb);
 892     return NULL;
 893 }
 894
 895 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 896                                       int64_t sector_num,
 897                                       QEMUIOVector *qiov,
 898                                       int nb_sectors,
 899                                       BlockCompletionFunc *cb,
 900                                       void *opaque)
 901 {
 902     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 903                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 904                          RBD_AIO_READ);
 905 }
 906
 907 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 908                                        int64_t sector_num,
 909                                        QEMUIOVector *qiov,
 910                                        int nb_sectors,
 911                                        BlockCompletionFunc *cb,
 912                                        void *opaque)
 913 {
 914     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 915                          (int64_t) nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 916                          RBD_AIO_WRITE);
 917 }
 918
 919 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 920 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 921                                       BlockCompletionFunc *cb,
 922                                       void *opaque)
 923 {
 924     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 925 }
 926
 927 #else
 928
 929 static int qemu_rbd_co_flush(BlockDriverState *bs)
 930 {
 931 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 932     /* rbd_flush added in 0.1.1 */
 933     BDRVRBDState *s = bs->opaque;
 934     return rbd_flush(s->image);
 935 #else
 936     return 0;
 937 #endif
 938 }
 939 #endif
 940
 941 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 942 {
 943     BDRVRBDState *s = bs->opaque;
 944     rbd_image_info_t info;
 945     int r;
 946
 947     r = rbd_stat(s->image, &info, sizeof(info));
 948     if (r < 0) {
 949         return r;
 950     }
 951
 952     bdi->cluster_size = info.obj_size;
 953     return 0;
 954 }
 955
 956 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 957 {
 958     BDRVRBDState *s = bs->opaque;
 959     rbd_image_info_t info;
 960     int r;
 961
 962     r = rbd_stat(s->image, &info, sizeof(info));
 963     if (r < 0) {
 964         return r;
 965     }
 966
 967     return info.size;
 968 }
 969
 970 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 971 {
 972     BDRVRBDState *s = bs->opaque;
 973     int r;
 974
 975     r = rbd_resize(s->image, offset);
 976     if (r < 0) {
 977         return r;
 978     }
 979
 980     return 0;
 981 }
 982
 983 static int qemu_rbd_snap_create(BlockDriverState *bs,
 984                                 QEMUSnapshotInfo *sn_info)
 985 {
 986     BDRVRBDState *s = bs->opaque;
 987     int r;
 988
 989     if (sn_info->name[0] == '\0') {
 990         return -EINVAL; /* we need a name for rbd snapshots */
 991     }
 992
 993     /*
 994      * rbd snapshots are using the name as the user controlled unique identifier
 995      * we can't use the rbd snapid for that purpose, as it can't be set
 996      */
 997     if (sn_info->id_str[0] != '\0' &&
 998         strcmp(sn_info->id_str, sn_info->name) != 0) {
 999         return -EINVAL;
1000     }
1001
1002     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
1003         return -ERANGE;
1004     }
1005
1006     r = rbd_snap_create(s->image, sn_info->name);
1007     if (r < 0) {
1008         error_report("failed to create snap: %s", strerror(-r));
1009         return r;
1010     }
1011
1012     return 0;
1013 }
1014
1015 static int qemu_rbd_snap_remove(BlockDriverState *bs,
1016                                 const char *snapshot_id,
1017                                 const char *snapshot_name,
1018                                 Error **errp)
1019 {
1020     BDRVRBDState *s = bs->opaque;
1021     int r;
1022
1023     if (!snapshot_name) {
1024         error_setg(errp, "rbd need a valid snapshot name");
1025         return -EINVAL;
1026     }
1027
1028     /* If snapshot_id is specified, it must be equal to name, see
1029        qemu_rbd_snap_list() */
1030     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
1031         error_setg(errp,
1032                    "rbd do not support snapshot id, it should be NULL or "
1033                    "equal to snapshot name");
1034         return -EINVAL;
1035     }
1036
1037     r = rbd_snap_remove(s->image, snapshot_name);
1038     if (r < 0) {
1039         error_setg_errno(errp, -r, "Failed to remove the snapshot");
1040     }
1041     return r;
1042 }
1043
1044 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
1045                                   const char *snapshot_name)
1046 {
1047     BDRVRBDState *s = bs->opaque;
1048
1049     return rbd_snap_rollback(s->image, snapshot_name);
1050 }
1051
1052 static int qemu_rbd_snap_list(BlockDriverState *bs,
1053                               QEMUSnapshotInfo **psn_tab)
1054 {
1055     BDRVRBDState *s = bs->opaque;
1056     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
1057     int i, snap_count;
1058     rbd_snap_info_t *snaps;
1059     int max_snaps = RBD_MAX_SNAPS;
1060
1061     do {
1062         snaps = g_new(rbd_snap_info_t, max_snaps);
1063         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
1064         if (snap_count <= 0) {
1065             g_free(snaps);
1066         }
1067     } while (snap_count == -ERANGE);
1068
1069     if (snap_count <= 0) {
1070         goto done;
1071     }
1072
1073     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
1074
1075     for (i = 0; i < snap_count; i++) {
1076         const char *snap_name = snaps[i].name;
1077
1078         sn_info = sn_tab + i;
1079         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
1080         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
1081
1082         sn_info->vm_state_size = snaps[i].size;
1083         sn_info->date_sec = 0;
1084         sn_info->date_nsec = 0;
1085         sn_info->vm_clock_nsec = 0;
1086     }
1087     rbd_snap_list_end(snaps);
1088     g_free(snaps);
1089
1090  done:
1091     *psn_tab = sn_tab;
1092     return snap_count;
1093 }
1094
1095 #ifdef LIBRBD_SUPPORTS_DISCARD
1096 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
1097                                          int64_t offset,
1098                                          int count,
1099                                          BlockCompletionFunc *cb,
1100                                          void *opaque)
1101 {
1102     return rbd_start_aio(bs, offset, NULL, count, cb, opaque,
1103                          RBD_AIO_DISCARD);
1104 }
1105 #endif
1106
1107 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1108 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
1109                                       Error **errp)
1110 {
1111     BDRVRBDState *s = bs->opaque;
1112     int r = rbd_invalidate_cache(s->image);
1113     if (r < 0) {
1114         error_setg_errno(errp, -r, "Failed to invalidate the cache");
1115     }
1116 }
1117 #endif
1118
1119 static QemuOptsList qemu_rbd_create_opts = {
1120     .name = "rbd-create-opts",
1121     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
1122     .desc = {
1123         {
1124             .name = BLOCK_OPT_SIZE,
1125             .type = QEMU_OPT_SIZE,
1126             .help = "Virtual disk size"
1127         },
1128         {
1129             .name = BLOCK_OPT_CLUSTER_SIZE,
1130             .type = QEMU_OPT_SIZE,
1131             .help = "RBD object size"
1132         },
1133         {
1134             .name = "password-secret",
1135             .type = QEMU_OPT_STRING,
1136             .help = "ID of secret providing the password",
1137         },
1138         { /* end of list */ }
1139     }
1140 };
1141
1142 static BlockDriver bdrv_rbd = {
1143     .format_name            = "rbd",
1144     .instance_size          = sizeof(BDRVRBDState),
1145     .bdrv_parse_filename    = qemu_rbd_parse_filename,
1146     .bdrv_file_open         = qemu_rbd_open,
1147     .bdrv_close             = qemu_rbd_close,
1148     .bdrv_create            = qemu_rbd_create,
1149     .bdrv_has_zero_init     = bdrv_has_zero_init_1,
1150     .bdrv_get_info          = qemu_rbd_getinfo,
1151     .create_opts            = &qemu_rbd_create_opts,
1152     .bdrv_getlength         = qemu_rbd_getlength,
1153     .bdrv_truncate          = qemu_rbd_truncate,
1154     .protocol_name          = "rbd",
1155
1156     .bdrv_aio_readv         = qemu_rbd_aio_readv,
1157     .bdrv_aio_writev        = qemu_rbd_aio_writev,
1158
1159 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1160     .bdrv_aio_flush         = qemu_rbd_aio_flush,
1161 #else
1162     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
1163 #endif
1164
1165 #ifdef LIBRBD_SUPPORTS_DISCARD
1166     .bdrv_aio_pdiscard      = qemu_rbd_aio_pdiscard,
1167 #endif
1168
1169     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1170     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1171     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1172     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1173 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1174     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1175 #endif
1176 };
1177
1178 static void bdrv_rbd_init(void)
1179 {
1180     bdrv_register(&bdrv_rbd);
1181 }
1182
1183 block_init(bdrv_rbd_init);