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
  22 #include <rbd/librbd.h>
  23
  24 /*
  25  * When specifying the image filename use:
  26  *
  27  * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
  28  *
  29  * poolname must be the name of an existing rados pool.
  30  *
  31  * devicename is the name of the rbd image.
  32  *
  33  * Each option given is used to configure rados, and may be any valid
  34  * Ceph option, "id", or "conf".
  35  *
  36  * The "id" option indicates what user we should authenticate as to
  37  * the Ceph cluster.  If it is excluded we will use the Ceph default
  38  * (normally 'admin').
  39  *
  40  * The "conf" option specifies a Ceph configuration file to read.  If
  41  * it is not specified, we will read from the default Ceph locations
  42  * (e.g., /etc/ceph/ceph.conf).  To avoid reading _any_ configuration
  43  * file, specify conf=/dev/null.
  44  *
  45  * Configuration values containing :, @, or = can be escaped with a
  46  * leading "\".
  47  */
  48
  49 /* rbd_aio_discard added in 0.1.2 */
  50 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
  51 #define LIBRBD_SUPPORTS_DISCARD
  52 #else
  53 #undef LIBRBD_SUPPORTS_DISCARD
  54 #endif
  55
  56 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
  57
  58 #define RBD_MAX_CONF_NAME_SIZE 128
  59 #define RBD_MAX_CONF_VAL_SIZE 512
  60 #define RBD_MAX_CONF_SIZE 1024
  61 #define RBD_MAX_POOL_NAME_SIZE 128
  62 #define RBD_MAX_SNAP_NAME_SIZE 128
  63 #define RBD_MAX_SNAPS 100
  64
  65 typedef enum {
  66     RBD_AIO_READ,
  67     RBD_AIO_WRITE,
  68     RBD_AIO_DISCARD,
  69     RBD_AIO_FLUSH
  70 } RBDAIOCmd;
  71
  72 typedef struct RBDAIOCB {
  73     BlockAIOCB common;
  74     int64_t ret;
  75     QEMUIOVector *qiov;
  76     char *bounce;
  77     RBDAIOCmd cmd;
  78     int error;
  79     struct BDRVRBDState *s;
  80 } RBDAIOCB;
  81
  82 typedef struct RADOSCB {
  83     RBDAIOCB *acb;
  84     struct BDRVRBDState *s;
  85     int64_t size;
  86     char *buf;
  87     int64_t ret;
  88 } RADOSCB;
  89
  90 typedef struct BDRVRBDState {
  91     rados_t cluster;
  92     rados_ioctx_t io_ctx;
  93     rbd_image_t image;
  94     char name[RBD_MAX_IMAGE_NAME_SIZE];
  95     char *snap;
  96 } BDRVRBDState;
  97
  98 static int qemu_rbd_next_tok(char *dst, int dst_len,
  99                              char *src, char delim,
 100                              const char *name,
 101                              char **p, Error **errp)
 102 {
 103     int l;
 104     char *end;
 105
 106     *p = NULL;
 107
 108     if (delim != '\0') {
 109         for (end = src; *end; ++end) {
 110             if (*end == delim) {
 111                 break;
 112             }
 113             if (*end == '\\' && end[1] != '\0') {
 114                 end++;
 115             }
 116         }
 117         if (*end == delim) {
 118             *p = end + 1;
 119             *end = '\0';
 120         }
 121     }
 122     l = strlen(src);
 123     if (l >= dst_len) {
 124         error_setg(errp, "%s too long", name);
 125         return -EINVAL;
 126     } else if (l == 0) {
 127         error_setg(errp, "%s too short", name);
 128         return -EINVAL;
 129     }
 130
 131     pstrcpy(dst, dst_len, src);
 132
 133     return 0;
 134 }
 135
 136 static void qemu_rbd_unescape(char *src)
 137 {
 138     char *p;
 139
 140     for (p = src; *src; ++src, ++p) {
 141         if (*src == '\\' && src[1] != '\0') {
 142             src++;
 143         }
 144         *p = *src;
 145     }
 146     *p = '\0';
 147 }
 148
 149 static int qemu_rbd_parsename(const char *filename,
 150                               char *pool, int pool_len,
 151                               char *snap, int snap_len,
 152                               char *name, int name_len,
 153                               char *conf, int conf_len,
 154                               Error **errp)
 155 {
 156     const char *start;
 157     char *p, *buf;
 158     int ret;
 159
 160     if (!strstart(filename, "rbd:", &start)) {
 161         error_setg(errp, "File name must start with 'rbd:'");
 162         return -EINVAL;
 163     }
 164
 165     buf = g_strdup(start);
 166     p = buf;
 167     *snap = '\0';
 168     *conf = '\0';
 169
 170     ret = qemu_rbd_next_tok(pool, pool_len, p,
 171                             '/', "pool name", &p, errp);
 172     if (ret < 0 || !p) {
 173         ret = -EINVAL;
 174         goto done;
 175     }
 176     qemu_rbd_unescape(pool);
 177
 178     if (strchr(p, '@')) {
 179         ret = qemu_rbd_next_tok(name, name_len, p,
 180                                 '@', "object name", &p, errp);
 181         if (ret < 0) {
 182             goto done;
 183         }
 184         ret = qemu_rbd_next_tok(snap, snap_len, p,
 185                                 ':', "snap name", &p, errp);
 186         qemu_rbd_unescape(snap);
 187     } else {
 188         ret = qemu_rbd_next_tok(name, name_len, p,
 189                                 ':', "object name", &p, errp);
 190     }
 191     qemu_rbd_unescape(name);
 192     if (ret < 0 || !p) {
 193         goto done;
 194     }
 195
 196     ret = qemu_rbd_next_tok(conf, conf_len, p,
 197                             '\0', "configuration", &p, errp);
 198
 199 done:
 200     g_free(buf);
 201     return ret;
 202 }
 203
 204 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 205 {
 206     const char *p = conf;
 207
 208     while (*p) {
 209         int len;
 210         const char *end = strchr(p, ':');
 211
 212         if (end) {
 213             len = end - p;
 214         } else {
 215             len = strlen(p);
 216         }
 217
 218         if (strncmp(p, "id=", 3) == 0) {
 219             len -= 3;
 220             strncpy(clientname, p + 3, len);
 221             clientname[len] = '\0';
 222             return clientname;
 223         }
 224         if (end == NULL) {
 225             break;
 226         }
 227         p = end + 1;
 228     }
 229     return NULL;
 230 }
 231
 232
 233 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
 234                              Error **errp)
 235 {
 236     if (secretid == 0) {
 237         return 0;
 238     }
 239
 240     gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
 241                                                     errp);
 242     if (!secret) {
 243         return -1;
 244     }
 245
 246     rados_conf_set(cluster, "key", secret);
 247     g_free(secret);
 248
 249     return 0;
 250 }
 251
 252
 253 static int qemu_rbd_set_conf(rados_t cluster, const char *conf,
 254                              bool only_read_conf_file,
 255                              Error **errp)
 256 {
 257     char *p, *buf;
 258     char name[RBD_MAX_CONF_NAME_SIZE];
 259     char value[RBD_MAX_CONF_VAL_SIZE];
 260     int ret = 0;
 261
 262     buf = g_strdup(conf);
 263     p = buf;
 264
 265     while (p) {
 266         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 267                                 '=', "conf option name", &p, errp);
 268         if (ret < 0) {
 269             break;
 270         }
 271         qemu_rbd_unescape(name);
 272
 273         if (!p) {
 274             error_setg(errp, "conf option %s has no value", name);
 275             ret = -EINVAL;
 276             break;
 277         }
 278
 279         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 280                                 ':', "conf option value", &p, errp);
 281         if (ret < 0) {
 282             break;
 283         }
 284         qemu_rbd_unescape(value);
 285
 286         if (strcmp(name, "conf") == 0) {
 287             /* read the conf file alone, so it doesn't override more
 288                specific settings for a particular device */
 289             if (only_read_conf_file) {
 290                 ret = rados_conf_read_file(cluster, value);
 291                 if (ret < 0) {
 292                     error_setg_errno(errp, -ret, "error reading conf file %s",
 293                                      value);
 294                     break;
 295                 }
 296             }
 297         } else if (strcmp(name, "id") == 0) {
 298             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 299         } else if (!only_read_conf_file) {
 300             ret = rados_conf_set(cluster, name, value);
 301             if (ret < 0) {
 302                 error_setg_errno(errp, -ret, "invalid conf option %s", name);
 303                 ret = -EINVAL;
 304                 break;
 305             }
 306         }
 307     }
 308
 309     g_free(buf);
 310     return ret;
 311 }
 312
 313 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 314 {
 315     Error *local_err = NULL;
 316     int64_t bytes = 0;
 317     int64_t objsize;
 318     int obj_order = 0;
 319     char pool[RBD_MAX_POOL_NAME_SIZE];
 320     char name[RBD_MAX_IMAGE_NAME_SIZE];
 321     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 322     char conf[RBD_MAX_CONF_SIZE];
 323     char clientname_buf[RBD_MAX_CONF_SIZE];
 324     char *clientname;
 325     const char *secretid;
 326     rados_t cluster;
 327     rados_ioctx_t io_ctx;
 328     int ret;
 329
 330     secretid = qemu_opt_get(opts, "password-secret");
 331
 332     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 333                            snap_buf, sizeof(snap_buf),
 334                            name, sizeof(name),
 335                            conf, sizeof(conf), &local_err) < 0) {
 336         error_propagate(errp, local_err);
 337         return -EINVAL;
 338     }
 339
 340     /* Read out options */
 341     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 342                      BDRV_SECTOR_SIZE);
 343     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 344     if (objsize) {
 345         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 346             error_setg(errp, "obj size needs to be power of 2");
 347             return -EINVAL;
 348         }
 349         if (objsize < 4096) {
 350             error_setg(errp, "obj size too small");
 351             return -EINVAL;
 352         }
 353         obj_order = ctz32(objsize);
 354     }
 355
 356     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 357     ret = rados_create(&cluster, clientname);
 358     if (ret < 0) {
 359         error_setg_errno(errp, -ret, "error initializing");
 360         return ret;
 361     }
 362
 363     if (strstr(conf, "conf=") == NULL) {
 364         /* try default location, but ignore failure */
 365         rados_conf_read_file(cluster, NULL);
 366     } else if (conf[0] != '\0' &&
 367                qemu_rbd_set_conf(cluster, conf, true, &local_err) < 0) {
 368         rados_shutdown(cluster);
 369         error_propagate(errp, local_err);
 370         return -EIO;
 371     }
 372
 373     if (conf[0] != '\0' &&
 374         qemu_rbd_set_conf(cluster, conf, false, &local_err) < 0) {
 375         rados_shutdown(cluster);
 376         error_propagate(errp, local_err);
 377         return -EIO;
 378     }
 379
 380     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
 381         rados_shutdown(cluster);
 382         return -EIO;
 383     }
 384
 385     ret = rados_connect(cluster);
 386     if (ret < 0) {
 387         error_setg_errno(errp, -ret, "error connecting");
 388         rados_shutdown(cluster);
 389         return ret;
 390     }
 391
 392     ret = rados_ioctx_create(cluster, pool, &io_ctx);
 393     if (ret < 0) {
 394         error_setg_errno(errp, -ret, "error opening pool %s", pool);
 395         rados_shutdown(cluster);
 396         return ret;
 397     }
 398
 399     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 400     rados_ioctx_destroy(io_ctx);
 401     rados_shutdown(cluster);
 402     if (ret < 0) {
 403         error_setg_errno(errp, -ret, "error rbd create");
 404         return ret;
 405     }
 406
 407     return ret;
 408 }
 409
 410 /*
 411  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 412  * BH context.
 413  */
 414 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 415 {
 416     RBDAIOCB *acb = rcb->acb;
 417     int64_t r;
 418
 419     r = rcb->ret;
 420
 421     if (acb->cmd != RBD_AIO_READ) {
 422         if (r < 0) {
 423             acb->ret = r;
 424             acb->error = 1;
 425         } else if (!acb->error) {
 426             acb->ret = rcb->size;
 427         }
 428     } else {
 429         if (r < 0) {
 430             memset(rcb->buf, 0, rcb->size);
 431             acb->ret = r;
 432             acb->error = 1;
 433         } else if (r < rcb->size) {
 434             memset(rcb->buf + r, 0, rcb->size - r);
 435             if (!acb->error) {
 436                 acb->ret = rcb->size;
 437             }
 438         } else if (!acb->error) {
 439             acb->ret = r;
 440         }
 441     }
 442
 443     g_free(rcb);
 444
 445     if (acb->cmd == RBD_AIO_READ) {
 446         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 447     }
 448     qemu_vfree(acb->bounce);
 449     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 450
 451     qemu_aio_unref(acb);
 452 }
 453
 454 /* TODO Convert to fine grained options */
 455 static QemuOptsList runtime_opts = {
 456     .name = "rbd",
 457     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 458     .desc = {
 459         {
 460             .name = "filename",
 461             .type = QEMU_OPT_STRING,
 462             .help = "Specification of the rbd image",
 463         },
 464         {
 465             .name = "password-secret",
 466             .type = QEMU_OPT_STRING,
 467             .help = "ID of secret providing the password",
 468         },
 469         { /* end of list */ }
 470     },
 471 };
 472
 473 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 474                          Error **errp)
 475 {
 476     BDRVRBDState *s = bs->opaque;
 477     char pool[RBD_MAX_POOL_NAME_SIZE];
 478     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 479     char conf[RBD_MAX_CONF_SIZE];
 480     char clientname_buf[RBD_MAX_CONF_SIZE];
 481     char *clientname;
 482     const char *secretid;
 483     QemuOpts *opts;
 484     Error *local_err = NULL;
 485     const char *filename;
 486     int r;
 487
 488     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 489     qemu_opts_absorb_qdict(opts, options, &local_err);
 490     if (local_err) {
 491         error_propagate(errp, local_err);
 492         qemu_opts_del(opts);
 493         return -EINVAL;
 494     }
 495
 496     filename = qemu_opt_get(opts, "filename");
 497     secretid = qemu_opt_get(opts, "password-secret");
 498
 499     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 500                            snap_buf, sizeof(snap_buf),
 501                            s->name, sizeof(s->name),
 502                            conf, sizeof(conf), errp) < 0) {
 503         r = -EINVAL;
 504         goto failed_opts;
 505     }
 506
 507     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 508     r = rados_create(&s->cluster, clientname);
 509     if (r < 0) {
 510         error_setg_errno(errp, -r, "error initializing");
 511         goto failed_opts;
 512     }
 513
 514     s->snap = NULL;
 515     if (snap_buf[0] != '\0') {
 516         s->snap = g_strdup(snap_buf);
 517     }
 518
 519     if (strstr(conf, "conf=") == NULL) {
 520         /* try default location, but ignore failure */
 521         rados_conf_read_file(s->cluster, NULL);
 522     } else if (conf[0] != '\0') {
 523         r = qemu_rbd_set_conf(s->cluster, conf, true, errp);
 524         if (r < 0) {
 525             goto failed_shutdown;
 526         }
 527     }
 528
 529     if (conf[0] != '\0') {
 530         r = qemu_rbd_set_conf(s->cluster, conf, false, errp);
 531         if (r < 0) {
 532             goto failed_shutdown;
 533         }
 534     }
 535
 536     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
 537         r = -EIO;
 538         goto failed_shutdown;
 539     }
 540
 541     /*
 542      * Fallback to more conservative semantics if setting cache
 543      * options fails. Ignore errors from setting rbd_cache because the
 544      * only possible error is that the option does not exist, and
 545      * librbd defaults to no caching. If write through caching cannot
 546      * be set up, fall back to no caching.
 547      */
 548     if (flags & BDRV_O_NOCACHE) {
 549         rados_conf_set(s->cluster, "rbd_cache", "false");
 550     } else {
 551         rados_conf_set(s->cluster, "rbd_cache", "true");
 552     }
 553
 554     r = rados_connect(s->cluster);
 555     if (r < 0) {
 556         error_setg_errno(errp, -r, "error connecting");
 557         goto failed_shutdown;
 558     }
 559
 560     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 561     if (r < 0) {
 562         error_setg_errno(errp, -r, "error opening pool %s", pool);
 563         goto failed_shutdown;
 564     }
 565
 566     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 567     if (r < 0) {
 568         error_setg_errno(errp, -r, "error reading header from %s", s->name);
 569         goto failed_open;
 570     }
 571
 572     bs->read_only = (s->snap != NULL);
 573
 574     qemu_opts_del(opts);
 575     return 0;
 576
 577 failed_open:
 578     rados_ioctx_destroy(s->io_ctx);
 579 failed_shutdown:
 580     rados_shutdown(s->cluster);
 581     g_free(s->snap);
 582 failed_opts:
 583     qemu_opts_del(opts);
 584     return r;
 585 }
 586
 587 static void qemu_rbd_close(BlockDriverState *bs)
 588 {
 589     BDRVRBDState *s = bs->opaque;
 590
 591     rbd_close(s->image);
 592     rados_ioctx_destroy(s->io_ctx);
 593     g_free(s->snap);
 594     rados_shutdown(s->cluster);
 595 }
 596
 597 static const AIOCBInfo rbd_aiocb_info = {
 598     .aiocb_size = sizeof(RBDAIOCB),
 599 };
 600
 601 static void rbd_finish_bh(void *opaque)
 602 {
 603     RADOSCB *rcb = opaque;
 604     qemu_rbd_complete_aio(rcb);
 605 }
 606
 607 /*
 608  * This is the callback function for rbd_aio_read and _write
 609  *
 610  * Note: this function is being called from a non qemu thread so
 611  * we need to be careful about what we do here. Generally we only
 612  * schedule a BH, and do the rest of the io completion handling
 613  * from rbd_finish_bh() which runs in a qemu context.
 614  */
 615 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 616 {
 617     RBDAIOCB *acb = rcb->acb;
 618
 619     rcb->ret = rbd_aio_get_return_value(c);
 620     rbd_aio_release(c);
 621
 622     aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs),
 623                             rbd_finish_bh, rcb);
 624 }
 625
 626 static int rbd_aio_discard_wrapper(rbd_image_t image,
 627                                    uint64_t off,
 628                                    uint64_t len,
 629                                    rbd_completion_t comp)
 630 {
 631 #ifdef LIBRBD_SUPPORTS_DISCARD
 632     return rbd_aio_discard(image, off, len, comp);
 633 #else
 634     return -ENOTSUP;
 635 #endif
 636 }
 637
 638 static int rbd_aio_flush_wrapper(rbd_image_t image,
 639                                  rbd_completion_t comp)
 640 {
 641 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 642     return rbd_aio_flush(image, comp);
 643 #else
 644     return -ENOTSUP;
 645 #endif
 646 }
 647
 648 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 649                                  int64_t off,
 650                                  QEMUIOVector *qiov,
 651                                  int64_t size,
 652                                  BlockCompletionFunc *cb,
 653                                  void *opaque,
 654                                  RBDAIOCmd cmd)
 655 {
 656     RBDAIOCB *acb;
 657     RADOSCB *rcb = NULL;
 658     rbd_completion_t c;
 659     char *buf;
 660     int r;
 661
 662     BDRVRBDState *s = bs->opaque;
 663
 664     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 665     acb->cmd = cmd;
 666     acb->qiov = qiov;
 667     assert(!qiov || qiov->size == size);
 668     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 669         acb->bounce = NULL;
 670     } else {
 671         acb->bounce = qemu_try_blockalign(bs, qiov->size);
 672         if (acb->bounce == NULL) {
 673             goto failed;
 674         }
 675     }
 676     acb->ret = 0;
 677     acb->error = 0;
 678     acb->s = s;
 679
 680     if (cmd == RBD_AIO_WRITE) {
 681         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 682     }
 683
 684     buf = acb->bounce;
 685
 686     rcb = g_new(RADOSCB, 1);
 687     rcb->acb = acb;
 688     rcb->buf = buf;
 689     rcb->s = acb->s;
 690     rcb->size = size;
 691     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 692     if (r < 0) {
 693         goto failed;
 694     }
 695
 696     switch (cmd) {
 697     case RBD_AIO_WRITE:
 698         r = rbd_aio_write(s->image, off, size, buf, c);
 699         break;
 700     case RBD_AIO_READ:
 701         r = rbd_aio_read(s->image, off, size, buf, c);
 702         break;
 703     case RBD_AIO_DISCARD:
 704         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 705         break;
 706     case RBD_AIO_FLUSH:
 707         r = rbd_aio_flush_wrapper(s->image, c);
 708         break;
 709     default:
 710         r = -EINVAL;
 711     }
 712
 713     if (r < 0) {
 714         goto failed_completion;
 715     }
 716
 717     return &acb->common;
 718
 719 failed_completion:
 720     rbd_aio_release(c);
 721 failed:
 722     g_free(rcb);
 723     qemu_vfree(acb->bounce);
 724     qemu_aio_unref(acb);
 725     return NULL;
 726 }
 727
 728 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 729                                       int64_t sector_num,
 730                                       QEMUIOVector *qiov,
 731                                       int nb_sectors,
 732                                       BlockCompletionFunc *cb,
 733                                       void *opaque)
 734 {
 735     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 736                          nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 737                          RBD_AIO_READ);
 738 }
 739
 740 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 741                                        int64_t sector_num,
 742                                        QEMUIOVector *qiov,
 743                                        int nb_sectors,
 744                                        BlockCompletionFunc *cb,
 745                                        void *opaque)
 746 {
 747     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 748                          nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 749                          RBD_AIO_WRITE);
 750 }
 751
 752 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 753 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 754                                       BlockCompletionFunc *cb,
 755                                       void *opaque)
 756 {
 757     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 758 }
 759
 760 #else
 761
 762 static int qemu_rbd_co_flush(BlockDriverState *bs)
 763 {
 764 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 765     /* rbd_flush added in 0.1.1 */
 766     BDRVRBDState *s = bs->opaque;
 767     return rbd_flush(s->image);
 768 #else
 769     return 0;
 770 #endif
 771 }
 772 #endif
 773
 774 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 775 {
 776     BDRVRBDState *s = bs->opaque;
 777     rbd_image_info_t info;
 778     int r;
 779
 780     r = rbd_stat(s->image, &info, sizeof(info));
 781     if (r < 0) {
 782         return r;
 783     }
 784
 785     bdi->cluster_size = info.obj_size;
 786     return 0;
 787 }
 788
 789 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 790 {
 791     BDRVRBDState *s = bs->opaque;
 792     rbd_image_info_t info;
 793     int r;
 794
 795     r = rbd_stat(s->image, &info, sizeof(info));
 796     if (r < 0) {
 797         return r;
 798     }
 799
 800     return info.size;
 801 }
 802
 803 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 804 {
 805     BDRVRBDState *s = bs->opaque;
 806     int r;
 807
 808     r = rbd_resize(s->image, offset);
 809     if (r < 0) {
 810         return r;
 811     }
 812
 813     return 0;
 814 }
 815
 816 static int qemu_rbd_snap_create(BlockDriverState *bs,
 817                                 QEMUSnapshotInfo *sn_info)
 818 {
 819     BDRVRBDState *s = bs->opaque;
 820     int r;
 821
 822     if (sn_info->name[0] == '\0') {
 823         return -EINVAL; /* we need a name for rbd snapshots */
 824     }
 825
 826     /*
 827      * rbd snapshots are using the name as the user controlled unique identifier
 828      * we can't use the rbd snapid for that purpose, as it can't be set
 829      */
 830     if (sn_info->id_str[0] != '\0' &&
 831         strcmp(sn_info->id_str, sn_info->name) != 0) {
 832         return -EINVAL;
 833     }
 834
 835     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 836         return -ERANGE;
 837     }
 838
 839     r = rbd_snap_create(s->image, sn_info->name);
 840     if (r < 0) {
 841         error_report("failed to create snap: %s", strerror(-r));
 842         return r;
 843     }
 844
 845     return 0;
 846 }
 847
 848 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 849                                 const char *snapshot_id,
 850                                 const char *snapshot_name,
 851                                 Error **errp)
 852 {
 853     BDRVRBDState *s = bs->opaque;
 854     int r;
 855
 856     if (!snapshot_name) {
 857         error_setg(errp, "rbd need a valid snapshot name");
 858         return -EINVAL;
 859     }
 860
 861     /* If snapshot_id is specified, it must be equal to name, see
 862        qemu_rbd_snap_list() */
 863     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 864         error_setg(errp,
 865                    "rbd do not support snapshot id, it should be NULL or "
 866                    "equal to snapshot name");
 867         return -EINVAL;
 868     }
 869
 870     r = rbd_snap_remove(s->image, snapshot_name);
 871     if (r < 0) {
 872         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 873     }
 874     return r;
 875 }
 876
 877 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 878                                   const char *snapshot_name)
 879 {
 880     BDRVRBDState *s = bs->opaque;
 881
 882     return rbd_snap_rollback(s->image, snapshot_name);
 883 }
 884
 885 static int qemu_rbd_snap_list(BlockDriverState *bs,
 886                               QEMUSnapshotInfo **psn_tab)
 887 {
 888     BDRVRBDState *s = bs->opaque;
 889     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 890     int i, snap_count;
 891     rbd_snap_info_t *snaps;
 892     int max_snaps = RBD_MAX_SNAPS;
 893
 894     do {
 895         snaps = g_new(rbd_snap_info_t, max_snaps);
 896         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 897         if (snap_count <= 0) {
 898             g_free(snaps);
 899         }
 900     } while (snap_count == -ERANGE);
 901
 902     if (snap_count <= 0) {
 903         goto done;
 904     }
 905
 906     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
 907
 908     for (i = 0; i < snap_count; i++) {
 909         const char *snap_name = snaps[i].name;
 910
 911         sn_info = sn_tab + i;
 912         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 913         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 914
 915         sn_info->vm_state_size = snaps[i].size;
 916         sn_info->date_sec = 0;
 917         sn_info->date_nsec = 0;
 918         sn_info->vm_clock_nsec = 0;
 919     }
 920     rbd_snap_list_end(snaps);
 921     g_free(snaps);
 922
 923  done:
 924     *psn_tab = sn_tab;
 925     return snap_count;
 926 }
 927
 928 #ifdef LIBRBD_SUPPORTS_DISCARD
 929 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
 930                                          int64_t offset,
 931                                          int count,
 932                                          BlockCompletionFunc *cb,
 933                                          void *opaque)
 934 {
 935     return rbd_start_aio(bs, offset, NULL, count, cb, opaque,
 936                          RBD_AIO_DISCARD);
 937 }
 938 #endif
 939
 940 #ifdef LIBRBD_SUPPORTS_INVALIDATE
 941 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
 942                                       Error **errp)
 943 {
 944     BDRVRBDState *s = bs->opaque;
 945     int r = rbd_invalidate_cache(s->image);
 946     if (r < 0) {
 947         error_setg_errno(errp, -r, "Failed to invalidate the cache");
 948     }
 949 }
 950 #endif
 951
 952 static QemuOptsList qemu_rbd_create_opts = {
 953     .name = "rbd-create-opts",
 954     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
 955     .desc = {
 956         {
 957             .name = BLOCK_OPT_SIZE,
 958             .type = QEMU_OPT_SIZE,
 959             .help = "Virtual disk size"
 960         },
 961         {
 962             .name = BLOCK_OPT_CLUSTER_SIZE,
 963             .type = QEMU_OPT_SIZE,
 964             .help = "RBD object size"
 965         },
 966         {
 967             .name = "password-secret",
 968             .type = QEMU_OPT_STRING,
 969             .help = "ID of secret providing the password",
 970         },
 971         { /* end of list */ }
 972     }
 973 };
 974
 975 static BlockDriver bdrv_rbd = {
 976     .format_name        = "rbd",
 977     .instance_size      = sizeof(BDRVRBDState),
 978     .bdrv_needs_filename = true,
 979     .bdrv_file_open     = qemu_rbd_open,
 980     .bdrv_close         = qemu_rbd_close,
 981     .bdrv_create        = qemu_rbd_create,
 982     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 983     .bdrv_get_info      = qemu_rbd_getinfo,
 984     .create_opts        = &qemu_rbd_create_opts,
 985     .bdrv_getlength     = qemu_rbd_getlength,
 986     .bdrv_truncate      = qemu_rbd_truncate,
 987     .protocol_name      = "rbd",
 988
 989     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 990     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 991
 992 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 993     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 994 #else
 995     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
 996 #endif
 997
 998 #ifdef LIBRBD_SUPPORTS_DISCARD
 999     .bdrv_aio_pdiscard      = qemu_rbd_aio_pdiscard,
1000 #endif
1001
1002     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1003     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1004     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1005     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1006 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1007     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1008 #endif
1009 };
1010
1011 static void bdrv_rbd_init(void)
1012 {
1013     bdrv_register(&bdrv_rbd);
1014 }
1015
1016 block_init(bdrv_rbd_init);