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     QEMUBH *bh;
  75     int64_t ret;
  76     QEMUIOVector *qiov;
  77     char *bounce;
  78     RBDAIOCmd cmd;
  79     int error;
  80     struct BDRVRBDState *s;
  81 } RBDAIOCB;
  82
  83 typedef struct RADOSCB {
  84     RBDAIOCB *acb;
  85     struct BDRVRBDState *s;
  86     int64_t size;
  87     char *buf;
  88     int64_t ret;
  89 } RADOSCB;
  90
  91 typedef struct BDRVRBDState {
  92     rados_t cluster;
  93     rados_ioctx_t io_ctx;
  94     rbd_image_t image;
  95     char name[RBD_MAX_IMAGE_NAME_SIZE];
  96     char *snap;
  97 } BDRVRBDState;
  98
  99 static int qemu_rbd_next_tok(char *dst, int dst_len,
 100                              char *src, char delim,
 101                              const char *name,
 102                              char **p, Error **errp)
 103 {
 104     int l;
 105     char *end;
 106
 107     *p = NULL;
 108
 109     if (delim != '\0') {
 110         for (end = src; *end; ++end) {
 111             if (*end == delim) {
 112                 break;
 113             }
 114             if (*end == '\\' && end[1] != '\0') {
 115                 end++;
 116             }
 117         }
 118         if (*end == delim) {
 119             *p = end + 1;
 120             *end = '\0';
 121         }
 122     }
 123     l = strlen(src);
 124     if (l >= dst_len) {
 125         error_setg(errp, "%s too long", name);
 126         return -EINVAL;
 127     } else if (l == 0) {
 128         error_setg(errp, "%s too short", name);
 129         return -EINVAL;
 130     }
 131
 132     pstrcpy(dst, dst_len, src);
 133
 134     return 0;
 135 }
 136
 137 static void qemu_rbd_unescape(char *src)
 138 {
 139     char *p;
 140
 141     for (p = src; *src; ++src, ++p) {
 142         if (*src == '\\' && src[1] != '\0') {
 143             src++;
 144         }
 145         *p = *src;
 146     }
 147     *p = '\0';
 148 }
 149
 150 static int qemu_rbd_parsename(const char *filename,
 151                               char *pool, int pool_len,
 152                               char *snap, int snap_len,
 153                               char *name, int name_len,
 154                               char *conf, int conf_len,
 155                               Error **errp)
 156 {
 157     const char *start;
 158     char *p, *buf;
 159     int ret;
 160
 161     if (!strstart(filename, "rbd:", &start)) {
 162         error_setg(errp, "File name must start with 'rbd:'");
 163         return -EINVAL;
 164     }
 165
 166     buf = g_strdup(start);
 167     p = buf;
 168     *snap = '\0';
 169     *conf = '\0';
 170
 171     ret = qemu_rbd_next_tok(pool, pool_len, p,
 172                             '/', "pool name", &p, errp);
 173     if (ret < 0 || !p) {
 174         ret = -EINVAL;
 175         goto done;
 176     }
 177     qemu_rbd_unescape(pool);
 178
 179     if (strchr(p, '@')) {
 180         ret = qemu_rbd_next_tok(name, name_len, p,
 181                                 '@', "object name", &p, errp);
 182         if (ret < 0) {
 183             goto done;
 184         }
 185         ret = qemu_rbd_next_tok(snap, snap_len, p,
 186                                 ':', "snap name", &p, errp);
 187         qemu_rbd_unescape(snap);
 188     } else {
 189         ret = qemu_rbd_next_tok(name, name_len, p,
 190                                 ':', "object name", &p, errp);
 191     }
 192     qemu_rbd_unescape(name);
 193     if (ret < 0 || !p) {
 194         goto done;
 195     }
 196
 197     ret = qemu_rbd_next_tok(conf, conf_len, p,
 198                             '\0', "configuration", &p, errp);
 199
 200 done:
 201     g_free(buf);
 202     return ret;
 203 }
 204
 205 static char *qemu_rbd_parse_clientname(const char *conf, char *clientname)
 206 {
 207     const char *p = conf;
 208
 209     while (*p) {
 210         int len;
 211         const char *end = strchr(p, ':');
 212
 213         if (end) {
 214             len = end - p;
 215         } else {
 216             len = strlen(p);
 217         }
 218
 219         if (strncmp(p, "id=", 3) == 0) {
 220             len -= 3;
 221             strncpy(clientname, p + 3, len);
 222             clientname[len] = '\0';
 223             return clientname;
 224         }
 225         if (end == NULL) {
 226             break;
 227         }
 228         p = end + 1;
 229     }
 230     return NULL;
 231 }
 232
 233
 234 static int qemu_rbd_set_auth(rados_t cluster, const char *secretid,
 235                              Error **errp)
 236 {
 237     if (secretid == 0) {
 238         return 0;
 239     }
 240
 241     gchar *secret = qcrypto_secret_lookup_as_base64(secretid,
 242                                                     errp);
 243     if (!secret) {
 244         return -1;
 245     }
 246
 247     rados_conf_set(cluster, "key", secret);
 248     g_free(secret);
 249
 250     return 0;
 251 }
 252
 253
 254 static int qemu_rbd_set_conf(rados_t cluster, const char *conf,
 255                              bool only_read_conf_file,
 256                              Error **errp)
 257 {
 258     char *p, *buf;
 259     char name[RBD_MAX_CONF_NAME_SIZE];
 260     char value[RBD_MAX_CONF_VAL_SIZE];
 261     int ret = 0;
 262
 263     buf = g_strdup(conf);
 264     p = buf;
 265
 266     while (p) {
 267         ret = qemu_rbd_next_tok(name, sizeof(name), p,
 268                                 '=', "conf option name", &p, errp);
 269         if (ret < 0) {
 270             break;
 271         }
 272         qemu_rbd_unescape(name);
 273
 274         if (!p) {
 275             error_setg(errp, "conf option %s has no value", name);
 276             ret = -EINVAL;
 277             break;
 278         }
 279
 280         ret = qemu_rbd_next_tok(value, sizeof(value), p,
 281                                 ':', "conf option value", &p, errp);
 282         if (ret < 0) {
 283             break;
 284         }
 285         qemu_rbd_unescape(value);
 286
 287         if (strcmp(name, "conf") == 0) {
 288             /* read the conf file alone, so it doesn't override more
 289                specific settings for a particular device */
 290             if (only_read_conf_file) {
 291                 ret = rados_conf_read_file(cluster, value);
 292                 if (ret < 0) {
 293                     error_setg_errno(errp, -ret, "error reading conf file %s",
 294                                      value);
 295                     break;
 296                 }
 297             }
 298         } else if (strcmp(name, "id") == 0) {
 299             /* ignore, this is parsed by qemu_rbd_parse_clientname() */
 300         } else if (!only_read_conf_file) {
 301             ret = rados_conf_set(cluster, name, value);
 302             if (ret < 0) {
 303                 error_setg_errno(errp, -ret, "invalid conf option %s", name);
 304                 ret = -EINVAL;
 305                 break;
 306             }
 307         }
 308     }
 309
 310     g_free(buf);
 311     return ret;
 312 }
 313
 314 static int qemu_rbd_create(const char *filename, QemuOpts *opts, Error **errp)
 315 {
 316     Error *local_err = NULL;
 317     int64_t bytes = 0;
 318     int64_t objsize;
 319     int obj_order = 0;
 320     char pool[RBD_MAX_POOL_NAME_SIZE];
 321     char name[RBD_MAX_IMAGE_NAME_SIZE];
 322     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 323     char conf[RBD_MAX_CONF_SIZE];
 324     char clientname_buf[RBD_MAX_CONF_SIZE];
 325     char *clientname;
 326     const char *secretid;
 327     rados_t cluster;
 328     rados_ioctx_t io_ctx;
 329     int ret;
 330
 331     secretid = qemu_opt_get(opts, "password-secret");
 332
 333     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 334                            snap_buf, sizeof(snap_buf),
 335                            name, sizeof(name),
 336                            conf, sizeof(conf), &local_err) < 0) {
 337         error_propagate(errp, local_err);
 338         return -EINVAL;
 339     }
 340
 341     /* Read out options */
 342     bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 343                      BDRV_SECTOR_SIZE);
 344     objsize = qemu_opt_get_size_del(opts, BLOCK_OPT_CLUSTER_SIZE, 0);
 345     if (objsize) {
 346         if ((objsize - 1) & objsize) {    /* not a power of 2? */
 347             error_setg(errp, "obj size needs to be power of 2");
 348             return -EINVAL;
 349         }
 350         if (objsize < 4096) {
 351             error_setg(errp, "obj size too small");
 352             return -EINVAL;
 353         }
 354         obj_order = ctz32(objsize);
 355     }
 356
 357     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 358     ret = rados_create(&cluster, clientname);
 359     if (ret < 0) {
 360         error_setg_errno(errp, -ret, "error initializing");
 361         return ret;
 362     }
 363
 364     if (strstr(conf, "conf=") == NULL) {
 365         /* try default location, but ignore failure */
 366         rados_conf_read_file(cluster, NULL);
 367     } else if (conf[0] != '\0' &&
 368                qemu_rbd_set_conf(cluster, conf, true, &local_err) < 0) {
 369         rados_shutdown(cluster);
 370         error_propagate(errp, local_err);
 371         return -EIO;
 372     }
 373
 374     if (conf[0] != '\0' &&
 375         qemu_rbd_set_conf(cluster, conf, false, &local_err) < 0) {
 376         rados_shutdown(cluster);
 377         error_propagate(errp, local_err);
 378         return -EIO;
 379     }
 380
 381     if (qemu_rbd_set_auth(cluster, secretid, errp) < 0) {
 382         rados_shutdown(cluster);
 383         return -EIO;
 384     }
 385
 386     ret = rados_connect(cluster);
 387     if (ret < 0) {
 388         error_setg_errno(errp, -ret, "error connecting");
 389         rados_shutdown(cluster);
 390         return ret;
 391     }
 392
 393     ret = rados_ioctx_create(cluster, pool, &io_ctx);
 394     if (ret < 0) {
 395         error_setg_errno(errp, -ret, "error opening pool %s", pool);
 396         rados_shutdown(cluster);
 397         return ret;
 398     }
 399
 400     ret = rbd_create(io_ctx, name, bytes, &obj_order);
 401     rados_ioctx_destroy(io_ctx);
 402     rados_shutdown(cluster);
 403     if (ret < 0) {
 404         error_setg_errno(errp, -ret, "error rbd create");
 405         return ret;
 406     }
 407
 408     return ret;
 409 }
 410
 411 /*
 412  * This aio completion is being called from rbd_finish_bh() and runs in qemu
 413  * BH context.
 414  */
 415 static void qemu_rbd_complete_aio(RADOSCB *rcb)
 416 {
 417     RBDAIOCB *acb = rcb->acb;
 418     int64_t r;
 419
 420     r = rcb->ret;
 421
 422     if (acb->cmd != RBD_AIO_READ) {
 423         if (r < 0) {
 424             acb->ret = r;
 425             acb->error = 1;
 426         } else if (!acb->error) {
 427             acb->ret = rcb->size;
 428         }
 429     } else {
 430         if (r < 0) {
 431             memset(rcb->buf, 0, rcb->size);
 432             acb->ret = r;
 433             acb->error = 1;
 434         } else if (r < rcb->size) {
 435             memset(rcb->buf + r, 0, rcb->size - r);
 436             if (!acb->error) {
 437                 acb->ret = rcb->size;
 438             }
 439         } else if (!acb->error) {
 440             acb->ret = r;
 441         }
 442     }
 443
 444     g_free(rcb);
 445
 446     if (acb->cmd == RBD_AIO_READ) {
 447         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
 448     }
 449     qemu_vfree(acb->bounce);
 450     acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
 451
 452     qemu_aio_unref(acb);
 453 }
 454
 455 /* TODO Convert to fine grained options */
 456 static QemuOptsList runtime_opts = {
 457     .name = "rbd",
 458     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 459     .desc = {
 460         {
 461             .name = "filename",
 462             .type = QEMU_OPT_STRING,
 463             .help = "Specification of the rbd image",
 464         },
 465         {
 466             .name = "password-secret",
 467             .type = QEMU_OPT_STRING,
 468             .help = "ID of secret providing the password",
 469         },
 470         { /* end of list */ }
 471     },
 472 };
 473
 474 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
 475                          Error **errp)
 476 {
 477     BDRVRBDState *s = bs->opaque;
 478     char pool[RBD_MAX_POOL_NAME_SIZE];
 479     char snap_buf[RBD_MAX_SNAP_NAME_SIZE];
 480     char conf[RBD_MAX_CONF_SIZE];
 481     char clientname_buf[RBD_MAX_CONF_SIZE];
 482     char *clientname;
 483     const char *secretid;
 484     QemuOpts *opts;
 485     Error *local_err = NULL;
 486     const char *filename;
 487     int r;
 488
 489     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 490     qemu_opts_absorb_qdict(opts, options, &local_err);
 491     if (local_err) {
 492         error_propagate(errp, local_err);
 493         qemu_opts_del(opts);
 494         return -EINVAL;
 495     }
 496
 497     filename = qemu_opt_get(opts, "filename");
 498     secretid = qemu_opt_get(opts, "password-secret");
 499
 500     if (qemu_rbd_parsename(filename, pool, sizeof(pool),
 501                            snap_buf, sizeof(snap_buf),
 502                            s->name, sizeof(s->name),
 503                            conf, sizeof(conf), errp) < 0) {
 504         r = -EINVAL;
 505         goto failed_opts;
 506     }
 507
 508     clientname = qemu_rbd_parse_clientname(conf, clientname_buf);
 509     r = rados_create(&s->cluster, clientname);
 510     if (r < 0) {
 511         error_setg_errno(errp, -r, "error initializing");
 512         goto failed_opts;
 513     }
 514
 515     s->snap = NULL;
 516     if (snap_buf[0] != '\0') {
 517         s->snap = g_strdup(snap_buf);
 518     }
 519
 520     if (strstr(conf, "conf=") == NULL) {
 521         /* try default location, but ignore failure */
 522         rados_conf_read_file(s->cluster, NULL);
 523     } else if (conf[0] != '\0') {
 524         r = qemu_rbd_set_conf(s->cluster, conf, true, errp);
 525         if (r < 0) {
 526             goto failed_shutdown;
 527         }
 528     }
 529
 530     if (conf[0] != '\0') {
 531         r = qemu_rbd_set_conf(s->cluster, conf, false, errp);
 532         if (r < 0) {
 533             goto failed_shutdown;
 534         }
 535     }
 536
 537     if (qemu_rbd_set_auth(s->cluster, secretid, errp) < 0) {
 538         r = -EIO;
 539         goto failed_shutdown;
 540     }
 541
 542     /*
 543      * Fallback to more conservative semantics if setting cache
 544      * options fails. Ignore errors from setting rbd_cache because the
 545      * only possible error is that the option does not exist, and
 546      * librbd defaults to no caching. If write through caching cannot
 547      * be set up, fall back to no caching.
 548      */
 549     if (flags & BDRV_O_NOCACHE) {
 550         rados_conf_set(s->cluster, "rbd_cache", "false");
 551     } else {
 552         rados_conf_set(s->cluster, "rbd_cache", "true");
 553     }
 554
 555     r = rados_connect(s->cluster);
 556     if (r < 0) {
 557         error_setg_errno(errp, -r, "error connecting");
 558         goto failed_shutdown;
 559     }
 560
 561     r = rados_ioctx_create(s->cluster, pool, &s->io_ctx);
 562     if (r < 0) {
 563         error_setg_errno(errp, -r, "error opening pool %s", pool);
 564         goto failed_shutdown;
 565     }
 566
 567     r = rbd_open(s->io_ctx, s->name, &s->image, s->snap);
 568     if (r < 0) {
 569         error_setg_errno(errp, -r, "error reading header from %s", s->name);
 570         goto failed_open;
 571     }
 572
 573     bs->read_only = (s->snap != NULL);
 574
 575     qemu_opts_del(opts);
 576     return 0;
 577
 578 failed_open:
 579     rados_ioctx_destroy(s->io_ctx);
 580 failed_shutdown:
 581     rados_shutdown(s->cluster);
 582     g_free(s->snap);
 583 failed_opts:
 584     qemu_opts_del(opts);
 585     return r;
 586 }
 587
 588 static void qemu_rbd_close(BlockDriverState *bs)
 589 {
 590     BDRVRBDState *s = bs->opaque;
 591
 592     rbd_close(s->image);
 593     rados_ioctx_destroy(s->io_ctx);
 594     g_free(s->snap);
 595     rados_shutdown(s->cluster);
 596 }
 597
 598 static const AIOCBInfo rbd_aiocb_info = {
 599     .aiocb_size = sizeof(RBDAIOCB),
 600 };
 601
 602 static void rbd_finish_bh(void *opaque)
 603 {
 604     RADOSCB *rcb = opaque;
 605     qemu_bh_delete(rcb->acb->bh);
 606     qemu_rbd_complete_aio(rcb);
 607 }
 608
 609 /*
 610  * This is the callback function for rbd_aio_read and _write
 611  *
 612  * Note: this function is being called from a non qemu thread so
 613  * we need to be careful about what we do here. Generally we only
 614  * schedule a BH, and do the rest of the io completion handling
 615  * from rbd_finish_bh() which runs in a qemu context.
 616  */
 617 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
 618 {
 619     RBDAIOCB *acb = rcb->acb;
 620
 621     rcb->ret = rbd_aio_get_return_value(c);
 622     rbd_aio_release(c);
 623
 624     acb->bh = aio_bh_new(bdrv_get_aio_context(acb->common.bs),
 625                          rbd_finish_bh, rcb);
 626     qemu_bh_schedule(acb->bh);
 627 }
 628
 629 static int rbd_aio_discard_wrapper(rbd_image_t image,
 630                                    uint64_t off,
 631                                    uint64_t len,
 632                                    rbd_completion_t comp)
 633 {
 634 #ifdef LIBRBD_SUPPORTS_DISCARD
 635     return rbd_aio_discard(image, off, len, comp);
 636 #else
 637     return -ENOTSUP;
 638 #endif
 639 }
 640
 641 static int rbd_aio_flush_wrapper(rbd_image_t image,
 642                                  rbd_completion_t comp)
 643 {
 644 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 645     return rbd_aio_flush(image, comp);
 646 #else
 647     return -ENOTSUP;
 648 #endif
 649 }
 650
 651 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
 652                                  int64_t off,
 653                                  QEMUIOVector *qiov,
 654                                  int64_t size,
 655                                  BlockCompletionFunc *cb,
 656                                  void *opaque,
 657                                  RBDAIOCmd cmd)
 658 {
 659     RBDAIOCB *acb;
 660     RADOSCB *rcb = NULL;
 661     rbd_completion_t c;
 662     char *buf;
 663     int r;
 664
 665     BDRVRBDState *s = bs->opaque;
 666
 667     acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
 668     acb->cmd = cmd;
 669     acb->qiov = qiov;
 670     assert(!qiov || qiov->size == size);
 671     if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
 672         acb->bounce = NULL;
 673     } else {
 674         acb->bounce = qemu_try_blockalign(bs, qiov->size);
 675         if (acb->bounce == NULL) {
 676             goto failed;
 677         }
 678     }
 679     acb->ret = 0;
 680     acb->error = 0;
 681     acb->s = s;
 682     acb->bh = NULL;
 683
 684     if (cmd == RBD_AIO_WRITE) {
 685         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
 686     }
 687
 688     buf = acb->bounce;
 689
 690     rcb = g_new(RADOSCB, 1);
 691     rcb->acb = acb;
 692     rcb->buf = buf;
 693     rcb->s = acb->s;
 694     rcb->size = size;
 695     r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
 696     if (r < 0) {
 697         goto failed;
 698     }
 699
 700     switch (cmd) {
 701     case RBD_AIO_WRITE:
 702         r = rbd_aio_write(s->image, off, size, buf, c);
 703         break;
 704     case RBD_AIO_READ:
 705         r = rbd_aio_read(s->image, off, size, buf, c);
 706         break;
 707     case RBD_AIO_DISCARD:
 708         r = rbd_aio_discard_wrapper(s->image, off, size, c);
 709         break;
 710     case RBD_AIO_FLUSH:
 711         r = rbd_aio_flush_wrapper(s->image, c);
 712         break;
 713     default:
 714         r = -EINVAL;
 715     }
 716
 717     if (r < 0) {
 718         goto failed_completion;
 719     }
 720
 721     return &acb->common;
 722
 723 failed_completion:
 724     rbd_aio_release(c);
 725 failed:
 726     g_free(rcb);
 727     qemu_vfree(acb->bounce);
 728     qemu_aio_unref(acb);
 729     return NULL;
 730 }
 731
 732 static BlockAIOCB *qemu_rbd_aio_readv(BlockDriverState *bs,
 733                                       int64_t sector_num,
 734                                       QEMUIOVector *qiov,
 735                                       int nb_sectors,
 736                                       BlockCompletionFunc *cb,
 737                                       void *opaque)
 738 {
 739     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 740                          nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 741                          RBD_AIO_READ);
 742 }
 743
 744 static BlockAIOCB *qemu_rbd_aio_writev(BlockDriverState *bs,
 745                                        int64_t sector_num,
 746                                        QEMUIOVector *qiov,
 747                                        int nb_sectors,
 748                                        BlockCompletionFunc *cb,
 749                                        void *opaque)
 750 {
 751     return rbd_start_aio(bs, sector_num << BDRV_SECTOR_BITS, qiov,
 752                          nb_sectors << BDRV_SECTOR_BITS, cb, opaque,
 753                          RBD_AIO_WRITE);
 754 }
 755
 756 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 757 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
 758                                       BlockCompletionFunc *cb,
 759                                       void *opaque)
 760 {
 761     return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
 762 }
 763
 764 #else
 765
 766 static int qemu_rbd_co_flush(BlockDriverState *bs)
 767 {
 768 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
 769     /* rbd_flush added in 0.1.1 */
 770     BDRVRBDState *s = bs->opaque;
 771     return rbd_flush(s->image);
 772 #else
 773     return 0;
 774 #endif
 775 }
 776 #endif
 777
 778 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
 779 {
 780     BDRVRBDState *s = bs->opaque;
 781     rbd_image_info_t info;
 782     int r;
 783
 784     r = rbd_stat(s->image, &info, sizeof(info));
 785     if (r < 0) {
 786         return r;
 787     }
 788
 789     bdi->cluster_size = info.obj_size;
 790     return 0;
 791 }
 792
 793 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
 794 {
 795     BDRVRBDState *s = bs->opaque;
 796     rbd_image_info_t info;
 797     int r;
 798
 799     r = rbd_stat(s->image, &info, sizeof(info));
 800     if (r < 0) {
 801         return r;
 802     }
 803
 804     return info.size;
 805 }
 806
 807 static int qemu_rbd_truncate(BlockDriverState *bs, int64_t offset)
 808 {
 809     BDRVRBDState *s = bs->opaque;
 810     int r;
 811
 812     r = rbd_resize(s->image, offset);
 813     if (r < 0) {
 814         return r;
 815     }
 816
 817     return 0;
 818 }
 819
 820 static int qemu_rbd_snap_create(BlockDriverState *bs,
 821                                 QEMUSnapshotInfo *sn_info)
 822 {
 823     BDRVRBDState *s = bs->opaque;
 824     int r;
 825
 826     if (sn_info->name[0] == '\0') {
 827         return -EINVAL; /* we need a name for rbd snapshots */
 828     }
 829
 830     /*
 831      * rbd snapshots are using the name as the user controlled unique identifier
 832      * we can't use the rbd snapid for that purpose, as it can't be set
 833      */
 834     if (sn_info->id_str[0] != '\0' &&
 835         strcmp(sn_info->id_str, sn_info->name) != 0) {
 836         return -EINVAL;
 837     }
 838
 839     if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
 840         return -ERANGE;
 841     }
 842
 843     r = rbd_snap_create(s->image, sn_info->name);
 844     if (r < 0) {
 845         error_report("failed to create snap: %s", strerror(-r));
 846         return r;
 847     }
 848
 849     return 0;
 850 }
 851
 852 static int qemu_rbd_snap_remove(BlockDriverState *bs,
 853                                 const char *snapshot_id,
 854                                 const char *snapshot_name,
 855                                 Error **errp)
 856 {
 857     BDRVRBDState *s = bs->opaque;
 858     int r;
 859
 860     if (!snapshot_name) {
 861         error_setg(errp, "rbd need a valid snapshot name");
 862         return -EINVAL;
 863     }
 864
 865     /* If snapshot_id is specified, it must be equal to name, see
 866        qemu_rbd_snap_list() */
 867     if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
 868         error_setg(errp,
 869                    "rbd do not support snapshot id, it should be NULL or "
 870                    "equal to snapshot name");
 871         return -EINVAL;
 872     }
 873
 874     r = rbd_snap_remove(s->image, snapshot_name);
 875     if (r < 0) {
 876         error_setg_errno(errp, -r, "Failed to remove the snapshot");
 877     }
 878     return r;
 879 }
 880
 881 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
 882                                   const char *snapshot_name)
 883 {
 884     BDRVRBDState *s = bs->opaque;
 885
 886     return rbd_snap_rollback(s->image, snapshot_name);
 887 }
 888
 889 static int qemu_rbd_snap_list(BlockDriverState *bs,
 890                               QEMUSnapshotInfo **psn_tab)
 891 {
 892     BDRVRBDState *s = bs->opaque;
 893     QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
 894     int i, snap_count;
 895     rbd_snap_info_t *snaps;
 896     int max_snaps = RBD_MAX_SNAPS;
 897
 898     do {
 899         snaps = g_new(rbd_snap_info_t, max_snaps);
 900         snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
 901         if (snap_count <= 0) {
 902             g_free(snaps);
 903         }
 904     } while (snap_count == -ERANGE);
 905
 906     if (snap_count <= 0) {
 907         goto done;
 908     }
 909
 910     sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
 911
 912     for (i = 0; i < snap_count; i++) {
 913         const char *snap_name = snaps[i].name;
 914
 915         sn_info = sn_tab + i;
 916         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
 917         pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
 918
 919         sn_info->vm_state_size = snaps[i].size;
 920         sn_info->date_sec = 0;
 921         sn_info->date_nsec = 0;
 922         sn_info->vm_clock_nsec = 0;
 923     }
 924     rbd_snap_list_end(snaps);
 925     g_free(snaps);
 926
 927  done:
 928     *psn_tab = sn_tab;
 929     return snap_count;
 930 }
 931
 932 #ifdef LIBRBD_SUPPORTS_DISCARD
 933 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
 934                                          int64_t offset,
 935                                          int count,
 936                                          BlockCompletionFunc *cb,
 937                                          void *opaque)
 938 {
 939     return rbd_start_aio(bs, offset, NULL, count, cb, opaque,
 940                          RBD_AIO_DISCARD);
 941 }
 942 #endif
 943
 944 #ifdef LIBRBD_SUPPORTS_INVALIDATE
 945 static void qemu_rbd_invalidate_cache(BlockDriverState *bs,
 946                                       Error **errp)
 947 {
 948     BDRVRBDState *s = bs->opaque;
 949     int r = rbd_invalidate_cache(s->image);
 950     if (r < 0) {
 951         error_setg_errno(errp, -r, "Failed to invalidate the cache");
 952     }
 953 }
 954 #endif
 955
 956 static QemuOptsList qemu_rbd_create_opts = {
 957     .name = "rbd-create-opts",
 958     .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
 959     .desc = {
 960         {
 961             .name = BLOCK_OPT_SIZE,
 962             .type = QEMU_OPT_SIZE,
 963             .help = "Virtual disk size"
 964         },
 965         {
 966             .name = BLOCK_OPT_CLUSTER_SIZE,
 967             .type = QEMU_OPT_SIZE,
 968             .help = "RBD object size"
 969         },
 970         {
 971             .name = "password-secret",
 972             .type = QEMU_OPT_STRING,
 973             .help = "ID of secret providing the password",
 974         },
 975         { /* end of list */ }
 976     }
 977 };
 978
 979 static BlockDriver bdrv_rbd = {
 980     .format_name        = "rbd",
 981     .instance_size      = sizeof(BDRVRBDState),
 982     .bdrv_needs_filename = true,
 983     .bdrv_file_open     = qemu_rbd_open,
 984     .bdrv_close         = qemu_rbd_close,
 985     .bdrv_create        = qemu_rbd_create,
 986     .bdrv_has_zero_init = bdrv_has_zero_init_1,
 987     .bdrv_get_info      = qemu_rbd_getinfo,
 988     .create_opts        = &qemu_rbd_create_opts,
 989     .bdrv_getlength     = qemu_rbd_getlength,
 990     .bdrv_truncate      = qemu_rbd_truncate,
 991     .protocol_name      = "rbd",
 992
 993     .bdrv_aio_readv         = qemu_rbd_aio_readv,
 994     .bdrv_aio_writev        = qemu_rbd_aio_writev,
 995
 996 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
 997     .bdrv_aio_flush         = qemu_rbd_aio_flush,
 998 #else
 999     .bdrv_co_flush_to_disk  = qemu_rbd_co_flush,
1000 #endif
1001
1002 #ifdef LIBRBD_SUPPORTS_DISCARD
1003     .bdrv_aio_pdiscard      = qemu_rbd_aio_pdiscard,
1004 #endif
1005
1006     .bdrv_snapshot_create   = qemu_rbd_snap_create,
1007     .bdrv_snapshot_delete   = qemu_rbd_snap_remove,
1008     .bdrv_snapshot_list     = qemu_rbd_snap_list,
1009     .bdrv_snapshot_goto     = qemu_rbd_snap_rollback,
1010 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1011     .bdrv_invalidate_cache  = qemu_rbd_invalidate_cache,
1012 #endif
1013 };
1014
1015 static void bdrv_rbd_init(void)
1016 {
1017     bdrv_register(&bdrv_rbd);
1018 }
1019
1020 block_init(bdrv_rbd_init);