block/gluster.c

   1 /*
   2  * GlusterFS backend for QEMU
   3  *
   4  * Copyright (C) 2012 Bharata B Rao <bharata@linux.vnet.ibm.com>
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   7  * See the COPYING file in the top-level directory.
   8  *
   9  */
  10 #include "qemu/osdep.h"
  11 #include <glusterfs/api/glfs.h>
  12 #include "block/block_int.h"
  13 #include "qapi/error.h"
  14 #include "qemu/uri.h"
  15
  16 typedef struct GlusterAIOCB {
  17     int64_t size;
  18     int ret;
  19     QEMUBH *bh;
  20     Coroutine *coroutine;
  21     AioContext *aio_context;
  22 } GlusterAIOCB;
  23
  24 typedef struct BDRVGlusterState {
  25     struct glfs *glfs;
  26     struct glfs_fd *fd;
  27     bool supports_seek_data;
  28     int debug_level;
  29 } BDRVGlusterState;
  30
  31 typedef struct GlusterConf {
  32     char *server;
  33     int port;
  34     char *volname;
  35     char *image;
  36     char *transport;
  37     int debug_level;
  38 } GlusterConf;
  39
  40 static void qemu_gluster_gconf_free(GlusterConf *gconf)
  41 {
  42     if (gconf) {
  43         g_free(gconf->server);
  44         g_free(gconf->volname);
  45         g_free(gconf->image);
  46         g_free(gconf->transport);
  47         g_free(gconf);
  48     }
  49 }
  50
  51 static int parse_volume_options(GlusterConf *gconf, char *path)
  52 {
  53     char *p, *q;
  54
  55     if (!path) {
  56         return -EINVAL;
  57     }
  58
  59     /* volume */
  60     p = q = path + strspn(path, "/");
  61     p += strcspn(p, "/");
  62     if (*p == '\0') {
  63         return -EINVAL;
  64     }
  65     gconf->volname = g_strndup(q, p - q);
  66
  67     /* image */
  68     p += strspn(p, "/");
  69     if (*p == '\0') {
  70         return -EINVAL;
  71     }
  72     gconf->image = g_strdup(p);
  73     return 0;
  74 }
  75
  76 /*
  77  * file=gluster[+transport]://[server[:port]]/volname/image[?socket=...]
  78  *
  79  * 'gluster' is the protocol.
  80  *
  81  * 'transport' specifies the transport type used to connect to gluster
  82  * management daemon (glusterd). Valid transport types are
  83  * tcp, unix and rdma. If a transport type isn't specified, then tcp
  84  * type is assumed.
  85  *
  86  * 'server' specifies the server where the volume file specification for
  87  * the given volume resides. This can be either hostname, ipv4 address
  88  * or ipv6 address. ipv6 address needs to be within square brackets [ ].
  89  * If transport type is 'unix', then 'server' field should not be specified.
  90  * The 'socket' field needs to be populated with the path to unix domain
  91  * socket.
  92  *
  93  * 'port' is the port number on which glusterd is listening. This is optional
  94  * and if not specified, QEMU will send 0 which will make gluster to use the
  95  * default port. If the transport type is unix, then 'port' should not be
  96  * specified.
  97  *
  98  * 'volname' is the name of the gluster volume which contains the VM image.
  99  *
 100  * 'image' is the path to the actual VM image that resides on gluster volume.
 101  *
 102  * Examples:
 103  *
 104  * file=gluster://1.2.3.4/testvol/a.img
 105  * file=gluster+tcp://1.2.3.4/testvol/a.img
 106  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
 107  * file=gluster+tcp://[1:2:3:4:5:6:7:8]/testvol/dir/a.img
 108  * file=gluster+tcp://[1:2:3:4:5:6:7:8]:24007/testvol/dir/a.img
 109  * file=gluster+tcp://server.domain.com:24007/testvol/dir/a.img
 110  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
 111  * file=gluster+rdma://1.2.3.4:24007/testvol/a.img
 112  */
 113 static int qemu_gluster_parseuri(GlusterConf *gconf, const char *filename)
 114 {
 115     URI *uri;
 116     QueryParams *qp = NULL;
 117     bool is_unix = false;
 118     int ret = 0;
 119
 120     uri = uri_parse(filename);
 121     if (!uri) {
 122         return -EINVAL;
 123     }
 124
 125     /* transport */
 126     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
 127         gconf->transport = g_strdup("tcp");
 128     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
 129         gconf->transport = g_strdup("tcp");
 130     } else if (!strcmp(uri->scheme, "gluster+unix")) {
 131         gconf->transport = g_strdup("unix");
 132         is_unix = true;
 133     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
 134         gconf->transport = g_strdup("rdma");
 135     } else {
 136         ret = -EINVAL;
 137         goto out;
 138     }
 139
 140     ret = parse_volume_options(gconf, uri->path);
 141     if (ret < 0) {
 142         goto out;
 143     }
 144
 145     qp = query_params_parse(uri->query);
 146     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
 147         ret = -EINVAL;
 148         goto out;
 149     }
 150
 151     if (is_unix) {
 152         if (uri->server || uri->port) {
 153             ret = -EINVAL;
 154             goto out;
 155         }
 156         if (strcmp(qp->p[0].name, "socket")) {
 157             ret = -EINVAL;
 158             goto out;
 159         }
 160         gconf->server = g_strdup(qp->p[0].value);
 161     } else {
 162         gconf->server = g_strdup(uri->server ? uri->server : "localhost");
 163         gconf->port = uri->port;
 164     }
 165
 166 out:
 167     if (qp) {
 168         query_params_free(qp);
 169     }
 170     uri_free(uri);
 171     return ret;
 172 }
 173
 174 static struct glfs *qemu_gluster_init(GlusterConf *gconf, const char *filename,
 175                                       Error **errp)
 176 {
 177     struct glfs *glfs = NULL;
 178     int ret;
 179     int old_errno;
 180
 181     ret = qemu_gluster_parseuri(gconf, filename);
 182     if (ret < 0) {
 183         error_setg(errp, "Usage: file=gluster[+transport]://[server[:port]]/"
 184                    "volname/image[?socket=...]");
 185         errno = -ret;
 186         goto out;
 187     }
 188
 189     glfs = glfs_new(gconf->volname);
 190     if (!glfs) {
 191         goto out;
 192     }
 193
 194     ret = glfs_set_volfile_server(glfs, gconf->transport, gconf->server,
 195             gconf->port);
 196     if (ret < 0) {
 197         goto out;
 198     }
 199
 200     ret = glfs_set_logging(glfs, "-", gconf->debug_level);
 201     if (ret < 0) {
 202         goto out;
 203     }
 204
 205     ret = glfs_init(glfs);
 206     if (ret) {
 207         error_setg_errno(errp, errno,
 208                          "Gluster connection failed for server=%s port=%d "
 209                          "volume=%s image=%s transport=%s", gconf->server,
 210                          gconf->port, gconf->volname, gconf->image,
 211                          gconf->transport);
 212
 213         /* glfs_init sometimes doesn't set errno although docs suggest that */
 214         if (errno == 0)
 215             errno = EINVAL;
 216
 217         goto out;
 218     }
 219     return glfs;
 220
 221 out:
 222     if (glfs) {
 223         old_errno = errno;
 224         glfs_fini(glfs);
 225         errno = old_errno;
 226     }
 227     return NULL;
 228 }
 229
 230 static void qemu_gluster_complete_aio(void *opaque)
 231 {
 232     GlusterAIOCB *acb = (GlusterAIOCB *)opaque;
 233
 234     qemu_bh_delete(acb->bh);
 235     acb->bh = NULL;
 236     qemu_coroutine_enter(acb->coroutine, NULL);
 237 }
 238
 239 /*
 240  * AIO callback routine called from GlusterFS thread.
 241  */
 242 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
 243 {
 244     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
 245
 246     if (!ret || ret == acb->size) {
 247         acb->ret = 0; /* Success */
 248     } else if (ret < 0) {
 249         acb->ret = -errno; /* Read/Write failed */
 250     } else {
 251         acb->ret = -EIO; /* Partial read/write - fail it */
 252     }
 253
 254     acb->bh = aio_bh_new(acb->aio_context, qemu_gluster_complete_aio, acb);
 255     qemu_bh_schedule(acb->bh);
 256 }
 257
 258 #define GLUSTER_OPT_FILENAME "filename"
 259 #define GLUSTER_OPT_DEBUG "debug"
 260 #define GLUSTER_DEBUG_DEFAULT 4
 261 #define GLUSTER_DEBUG_MAX 9
 262
 263 /* TODO Convert to fine grained options */
 264 static QemuOptsList runtime_opts = {
 265     .name = "gluster",
 266     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 267     .desc = {
 268         {
 269             .name = GLUSTER_OPT_FILENAME,
 270             .type = QEMU_OPT_STRING,
 271             .help = "URL to the gluster image",
 272         },
 273         {
 274             .name = GLUSTER_OPT_DEBUG,
 275             .type = QEMU_OPT_NUMBER,
 276             .help = "Gluster log level, valid range is 0-9",
 277         },
 278         { /* end of list */ }
 279     },
 280 };
 281
 282 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
 283 {
 284     assert(open_flags != NULL);
 285
 286     *open_flags |= O_BINARY;
 287
 288     if (bdrv_flags & BDRV_O_RDWR) {
 289         *open_flags |= O_RDWR;
 290     } else {
 291         *open_flags |= O_RDONLY;
 292     }
 293
 294     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 295         *open_flags |= O_DIRECT;
 296     }
 297 }
 298
 299 /*
 300  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
 301  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
 302  * - Corrected versions return -1 and set errno to EINVAL.
 303  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
 304  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
 305  */
 306 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
 307 {
 308     off_t ret, eof;
 309
 310     eof = glfs_lseek(fd, 0, SEEK_END);
 311     if (eof < 0) {
 312         /* this should never occur */
 313         return false;
 314     }
 315
 316     /* this should always fail with ENXIO if SEEK_DATA is supported */
 317     ret = glfs_lseek(fd, eof, SEEK_DATA);
 318     return (ret < 0) && (errno == ENXIO);
 319 }
 320
 321 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
 322                              int bdrv_flags, Error **errp)
 323 {
 324     BDRVGlusterState *s = bs->opaque;
 325     int open_flags = 0;
 326     int ret = 0;
 327     GlusterConf *gconf = g_new0(GlusterConf, 1);
 328     QemuOpts *opts;
 329     Error *local_err = NULL;
 330     const char *filename;
 331
 332     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 333     qemu_opts_absorb_qdict(opts, options, &local_err);
 334     if (local_err) {
 335         error_propagate(errp, local_err);
 336         ret = -EINVAL;
 337         goto out;
 338     }
 339
 340     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
 341
 342     s->debug_level = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
 343                                          GLUSTER_DEBUG_DEFAULT);
 344     if (s->debug_level < 0) {
 345         s->debug_level = 0;
 346     } else if (s->debug_level > GLUSTER_DEBUG_MAX) {
 347         s->debug_level = GLUSTER_DEBUG_MAX;
 348     }
 349
 350     gconf->debug_level = s->debug_level;
 351     s->glfs = qemu_gluster_init(gconf, filename, errp);
 352     if (!s->glfs) {
 353         ret = -errno;
 354         goto out;
 355     }
 356
 357 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
 358     /* Without this, if fsync fails for a recoverable reason (for instance,
 359      * ENOSPC), gluster will dump its cache, preventing retries.  This means
 360      * almost certain data loss.  Not all gluster versions support the
 361      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
 362      * discover during runtime if it is supported (this api returns success for
 363      * unknown key/value pairs) */
 364     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
 365                                           "resync-failed-syncs-after-fsync",
 366                                           "on");
 367     if (ret < 0) {
 368         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
 369         ret = -errno;
 370         goto out;
 371     }
 372 #endif
 373
 374     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
 375
 376     s->fd = glfs_open(s->glfs, gconf->image, open_flags);
 377     if (!s->fd) {
 378         ret = -errno;
 379     }
 380
 381     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
 382
 383 out:
 384     qemu_opts_del(opts);
 385     qemu_gluster_gconf_free(gconf);
 386     if (!ret) {
 387         return ret;
 388     }
 389     if (s->fd) {
 390         glfs_close(s->fd);
 391     }
 392     if (s->glfs) {
 393         glfs_fini(s->glfs);
 394     }
 395     return ret;
 396 }
 397
 398 typedef struct BDRVGlusterReopenState {
 399     struct glfs *glfs;
 400     struct glfs_fd *fd;
 401 } BDRVGlusterReopenState;
 402
 403
 404 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
 405                                        BlockReopenQueue *queue, Error **errp)
 406 {
 407     int ret = 0;
 408     BDRVGlusterState *s;
 409     BDRVGlusterReopenState *reop_s;
 410     GlusterConf *gconf = NULL;
 411     int open_flags = 0;
 412
 413     assert(state != NULL);
 414     assert(state->bs != NULL);
 415
 416     s = state->bs->opaque;
 417
 418     state->opaque = g_new0(BDRVGlusterReopenState, 1);
 419     reop_s = state->opaque;
 420
 421     qemu_gluster_parse_flags(state->flags, &open_flags);
 422
 423     gconf = g_new0(GlusterConf, 1);
 424
 425     gconf->debug_level = s->debug_level;
 426     reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, errp);
 427     if (reop_s->glfs == NULL) {
 428         ret = -errno;
 429         goto exit;
 430     }
 431
 432 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
 433     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
 434                                  "resync-failed-syncs-after-fsync", "on");
 435     if (ret < 0) {
 436         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
 437         ret = -errno;
 438         goto exit;
 439     }
 440 #endif
 441
 442     reop_s->fd = glfs_open(reop_s->glfs, gconf->image, open_flags);
 443     if (reop_s->fd == NULL) {
 444         /* reops->glfs will be cleaned up in _abort */
 445         ret = -errno;
 446         goto exit;
 447     }
 448
 449 exit:
 450     /* state->opaque will be freed in either the _abort or _commit */
 451     qemu_gluster_gconf_free(gconf);
 452     return ret;
 453 }
 454
 455 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
 456 {
 457     BDRVGlusterReopenState *reop_s = state->opaque;
 458     BDRVGlusterState *s = state->bs->opaque;
 459
 460
 461     /* close the old */
 462     if (s->fd) {
 463         glfs_close(s->fd);
 464     }
 465     if (s->glfs) {
 466         glfs_fini(s->glfs);
 467     }
 468
 469     /* use the newly opened image / connection */
 470     s->fd         = reop_s->fd;
 471     s->glfs       = reop_s->glfs;
 472
 473     g_free(state->opaque);
 474     state->opaque = NULL;
 475
 476     return;
 477 }
 478
 479
 480 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
 481 {
 482     BDRVGlusterReopenState *reop_s = state->opaque;
 483
 484     if (reop_s == NULL) {
 485         return;
 486     }
 487
 488     if (reop_s->fd) {
 489         glfs_close(reop_s->fd);
 490     }
 491
 492     if (reop_s->glfs) {
 493         glfs_fini(reop_s->glfs);
 494     }
 495
 496     g_free(state->opaque);
 497     state->opaque = NULL;
 498
 499     return;
 500 }
 501
 502 #ifdef CONFIG_GLUSTERFS_ZEROFILL
 503 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
 504         int64_t offset, int size, BdrvRequestFlags flags)
 505 {
 506     int ret;
 507     GlusterAIOCB acb;
 508     BDRVGlusterState *s = bs->opaque;
 509
 510     acb.size = size;
 511     acb.ret = 0;
 512     acb.coroutine = qemu_coroutine_self();
 513     acb.aio_context = bdrv_get_aio_context(bs);
 514
 515     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
 516     if (ret < 0) {
 517         return -errno;
 518     }
 519
 520     qemu_coroutine_yield();
 521     return acb.ret;
 522 }
 523
 524 static inline bool gluster_supports_zerofill(void)
 525 {
 526     return 1;
 527 }
 528
 529 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
 530         int64_t size)
 531 {
 532     return glfs_zerofill(fd, offset, size);
 533 }
 534
 535 #else
 536 static inline bool gluster_supports_zerofill(void)
 537 {
 538     return 0;
 539 }
 540
 541 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
 542         int64_t size)
 543 {
 544     return 0;
 545 }
 546 #endif
 547
 548 static int qemu_gluster_create(const char *filename,
 549                                QemuOpts *opts, Error **errp)
 550 {
 551     struct glfs *glfs;
 552     struct glfs_fd *fd;
 553     int ret = 0;
 554     int prealloc = 0;
 555     int64_t total_size = 0;
 556     char *tmp = NULL;
 557     GlusterConf *gconf = g_new0(GlusterConf, 1);
 558
 559     gconf->debug_level = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
 560                                                  GLUSTER_DEBUG_DEFAULT);
 561     if (gconf->debug_level < 0) {
 562         gconf->debug_level = 0;
 563     } else if (gconf->debug_level > GLUSTER_DEBUG_MAX) {
 564         gconf->debug_level = GLUSTER_DEBUG_MAX;
 565     }
 566
 567     glfs = qemu_gluster_init(gconf, filename, errp);
 568     if (!glfs) {
 569         ret = -errno;
 570         goto out;
 571     }
 572
 573     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
 574                           BDRV_SECTOR_SIZE);
 575
 576     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
 577     if (!tmp || !strcmp(tmp, "off")) {
 578         prealloc = 0;
 579     } else if (!strcmp(tmp, "full") &&
 580                gluster_supports_zerofill()) {
 581         prealloc = 1;
 582     } else {
 583         error_setg(errp, "Invalid preallocation mode: '%s'"
 584             " or GlusterFS doesn't support zerofill API",
 585             tmp);
 586         ret = -EINVAL;
 587         goto out;
 588     }
 589
 590     fd = glfs_creat(glfs, gconf->image,
 591         O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
 592     if (!fd) {
 593         ret = -errno;
 594     } else {
 595         if (!glfs_ftruncate(fd, total_size)) {
 596             if (prealloc && qemu_gluster_zerofill(fd, 0, total_size)) {
 597                 ret = -errno;
 598             }
 599         } else {
 600             ret = -errno;
 601         }
 602
 603         if (glfs_close(fd) != 0) {
 604             ret = -errno;
 605         }
 606     }
 607 out:
 608     g_free(tmp);
 609     qemu_gluster_gconf_free(gconf);
 610     if (glfs) {
 611         glfs_fini(glfs);
 612     }
 613     return ret;
 614 }
 615
 616 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
 617         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int write)
 618 {
 619     int ret;
 620     GlusterAIOCB acb;
 621     BDRVGlusterState *s = bs->opaque;
 622     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
 623     off_t offset = sector_num * BDRV_SECTOR_SIZE;
 624
 625     acb.size = size;
 626     acb.ret = 0;
 627     acb.coroutine = qemu_coroutine_self();
 628     acb.aio_context = bdrv_get_aio_context(bs);
 629
 630     if (write) {
 631         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
 632             gluster_finish_aiocb, &acb);
 633     } else {
 634         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
 635             gluster_finish_aiocb, &acb);
 636     }
 637
 638     if (ret < 0) {
 639         return -errno;
 640     }
 641
 642     qemu_coroutine_yield();
 643     return acb.ret;
 644 }
 645
 646 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset)
 647 {
 648     int ret;
 649     BDRVGlusterState *s = bs->opaque;
 650
 651     ret = glfs_ftruncate(s->fd, offset);
 652     if (ret < 0) {
 653         return -errno;
 654     }
 655
 656     return 0;
 657 }
 658
 659 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
 660         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 661 {
 662     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
 663 }
 664
 665 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
 666         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 667 {
 668     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
 669 }
 670
 671 static void qemu_gluster_close(BlockDriverState *bs)
 672 {
 673     BDRVGlusterState *s = bs->opaque;
 674
 675     if (s->fd) {
 676         glfs_close(s->fd);
 677         s->fd = NULL;
 678     }
 679     glfs_fini(s->glfs);
 680 }
 681
 682 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
 683 {
 684     int ret;
 685     GlusterAIOCB acb;
 686     BDRVGlusterState *s = bs->opaque;
 687
 688     acb.size = 0;
 689     acb.ret = 0;
 690     acb.coroutine = qemu_coroutine_self();
 691     acb.aio_context = bdrv_get_aio_context(bs);
 692
 693     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
 694     if (ret < 0) {
 695         ret = -errno;
 696         goto error;
 697     }
 698
 699     qemu_coroutine_yield();
 700     if (acb.ret < 0) {
 701         ret = acb.ret;
 702         goto error;
 703     }
 704
 705     return acb.ret;
 706
 707 error:
 708     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
 709      * after a fsync failure, so we have no way of allowing the guest to safely
 710      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
 711      * either, but will invalidate the fd on error, so this is again our only
 712      * option.
 713      *
 714      * The 'resync-failed-syncs-after-fsync' xlator option for the
 715      * write-behind cache will cause later gluster versions to retain its
 716      * cache after error, so long as the fd remains open.  However, we
 717      * currently have no way of knowing if this option is supported.
 718      *
 719      * TODO: Once gluster provides a way for us to determine if the option
 720      * is supported, bypass the closure and setting drv to NULL.  */
 721     qemu_gluster_close(bs);
 722     bs->drv = NULL;
 723     return ret;
 724 }
 725
 726 #ifdef CONFIG_GLUSTERFS_DISCARD
 727 static coroutine_fn int qemu_gluster_co_discard(BlockDriverState *bs,
 728         int64_t sector_num, int nb_sectors)
 729 {
 730     int ret;
 731     GlusterAIOCB acb;
 732     BDRVGlusterState *s = bs->opaque;
 733     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
 734     off_t offset = sector_num * BDRV_SECTOR_SIZE;
 735
 736     acb.size = 0;
 737     acb.ret = 0;
 738     acb.coroutine = qemu_coroutine_self();
 739     acb.aio_context = bdrv_get_aio_context(bs);
 740
 741     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
 742     if (ret < 0) {
 743         return -errno;
 744     }
 745
 746     qemu_coroutine_yield();
 747     return acb.ret;
 748 }
 749 #endif
 750
 751 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
 752 {
 753     BDRVGlusterState *s = bs->opaque;
 754     int64_t ret;
 755
 756     ret = glfs_lseek(s->fd, 0, SEEK_END);
 757     if (ret < 0) {
 758         return -errno;
 759     } else {
 760         return ret;
 761     }
 762 }
 763
 764 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
 765 {
 766     BDRVGlusterState *s = bs->opaque;
 767     struct stat st;
 768     int ret;
 769
 770     ret = glfs_fstat(s->fd, &st);
 771     if (ret < 0) {
 772         return -errno;
 773     } else {
 774         return st.st_blocks * 512;
 775     }
 776 }
 777
 778 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
 779 {
 780     /* GlusterFS volume could be backed by a block device */
 781     return 0;
 782 }
 783
 784 /*
 785  * Find allocation range in @bs around offset @start.
 786  * May change underlying file descriptor's file offset.
 787  * If @start is not in a hole, store @start in @data, and the
 788  * beginning of the next hole in @hole, and return 0.
 789  * If @start is in a non-trailing hole, store @start in @hole and the
 790  * beginning of the next non-hole in @data, and return 0.
 791  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
 792  * If we can't find out, return a negative errno other than -ENXIO.
 793  *
 794  * (Shamefully copied from raw-posix.c, only miniscule adaptions.)
 795  */
 796 static int find_allocation(BlockDriverState *bs, off_t start,
 797                            off_t *data, off_t *hole)
 798 {
 799     BDRVGlusterState *s = bs->opaque;
 800     off_t offs;
 801
 802     if (!s->supports_seek_data) {
 803         return -ENOTSUP;
 804     }
 805
 806     /*
 807      * SEEK_DATA cases:
 808      * D1. offs == start: start is in data
 809      * D2. offs > start: start is in a hole, next data at offs
 810      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
 811      *                              or start is beyond EOF
 812      *     If the latter happens, the file has been truncated behind
 813      *     our back since we opened it.  All bets are off then.
 814      *     Treating like a trailing hole is simplest.
 815      * D4. offs < 0, errno != ENXIO: we learned nothing
 816      */
 817     offs = glfs_lseek(s->fd, start, SEEK_DATA);
 818     if (offs < 0) {
 819         return -errno;          /* D3 or D4 */
 820     }
 821     assert(offs >= start);
 822
 823     if (offs > start) {
 824         /* D2: in hole, next data at offs */
 825         *hole = start;
 826         *data = offs;
 827         return 0;
 828     }
 829
 830     /* D1: in data, end not yet known */
 831
 832     /*
 833      * SEEK_HOLE cases:
 834      * H1. offs == start: start is in a hole
 835      *     If this happens here, a hole has been dug behind our back
 836      *     since the previous lseek().
 837      * H2. offs > start: either start is in data, next hole at offs,
 838      *                   or start is in trailing hole, EOF at offs
 839      *     Linux treats trailing holes like any other hole: offs ==
 840      *     start.  Solaris seeks to EOF instead: offs > start (blech).
 841      *     If that happens here, a hole has been dug behind our back
 842      *     since the previous lseek().
 843      * H3. offs < 0, errno = ENXIO: start is beyond EOF
 844      *     If this happens, the file has been truncated behind our
 845      *     back since we opened it.  Treat it like a trailing hole.
 846      * H4. offs < 0, errno != ENXIO: we learned nothing
 847      *     Pretend we know nothing at all, i.e. "forget" about D1.
 848      */
 849     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
 850     if (offs < 0) {
 851         return -errno;          /* D1 and (H3 or H4) */
 852     }
 853     assert(offs >= start);
 854
 855     if (offs > start) {
 856         /*
 857          * D1 and H2: either in data, next hole at offs, or it was in
 858          * data but is now in a trailing hole.  In the latter case,
 859          * all bets are off.  Treating it as if it there was data all
 860          * the way to EOF is safe, so simply do that.
 861          */
 862         *data = start;
 863         *hole = offs;
 864         return 0;
 865     }
 866
 867     /* D1 and H1 */
 868     return -EBUSY;
 869 }
 870
 871 /*
 872  * Returns the allocation status of the specified sectors.
 873  *
 874  * If 'sector_num' is beyond the end of the disk image the return value is 0
 875  * and 'pnum' is set to 0.
 876  *
 877  * 'pnum' is set to the number of sectors (including and immediately following
 878  * the specified sector) that are known to be in the same
 879  * allocated/unallocated state.
 880  *
 881  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
 882  * beyond the end of the disk image it will be clamped.
 883  *
 884  * (Based on raw_co_get_block_status() from raw-posix.c.)
 885  */
 886 static int64_t coroutine_fn qemu_gluster_co_get_block_status(
 887         BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
 888         BlockDriverState **file)
 889 {
 890     BDRVGlusterState *s = bs->opaque;
 891     off_t start, data = 0, hole = 0;
 892     int64_t total_size;
 893     int ret = -EINVAL;
 894
 895     if (!s->fd) {
 896         return ret;
 897     }
 898
 899     start = sector_num * BDRV_SECTOR_SIZE;
 900     total_size = bdrv_getlength(bs);
 901     if (total_size < 0) {
 902         return total_size;
 903     } else if (start >= total_size) {
 904         *pnum = 0;
 905         return 0;
 906     } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
 907         nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
 908     }
 909
 910     ret = find_allocation(bs, start, &data, &hole);
 911     if (ret == -ENXIO) {
 912         /* Trailing hole */
 913         *pnum = nb_sectors;
 914         ret = BDRV_BLOCK_ZERO;
 915     } else if (ret < 0) {
 916         /* No info available, so pretend there are no holes */
 917         *pnum = nb_sectors;
 918         ret = BDRV_BLOCK_DATA;
 919     } else if (data == start) {
 920         /* On a data extent, compute sectors to the end of the extent,
 921          * possibly including a partial sector at EOF. */
 922         *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
 923         ret = BDRV_BLOCK_DATA;
 924     } else {
 925         /* On a hole, compute sectors to the beginning of the next extent.  */
 926         assert(hole == start);
 927         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
 928         ret = BDRV_BLOCK_ZERO;
 929     }
 930
 931     *file = bs;
 932
 933     return ret | BDRV_BLOCK_OFFSET_VALID | start;
 934 }
 935
 936
 937 static QemuOptsList qemu_gluster_create_opts = {
 938     .name = "qemu-gluster-create-opts",
 939     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
 940     .desc = {
 941         {
 942             .name = BLOCK_OPT_SIZE,
 943             .type = QEMU_OPT_SIZE,
 944             .help = "Virtual disk size"
 945         },
 946         {
 947             .name = BLOCK_OPT_PREALLOC,
 948             .type = QEMU_OPT_STRING,
 949             .help = "Preallocation mode (allowed values: off, full)"
 950         },
 951         {
 952             .name = GLUSTER_OPT_DEBUG,
 953             .type = QEMU_OPT_NUMBER,
 954             .help = "Gluster log level, valid range is 0-9",
 955         },
 956         { /* end of list */ }
 957     }
 958 };
 959
 960 static BlockDriver bdrv_gluster = {
 961     .format_name                  = "gluster",
 962     .protocol_name                = "gluster",
 963     .instance_size                = sizeof(BDRVGlusterState),
 964     .bdrv_needs_filename          = true,
 965     .bdrv_file_open               = qemu_gluster_open,
 966     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
 967     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
 968     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
 969     .bdrv_close                   = qemu_gluster_close,
 970     .bdrv_create                  = qemu_gluster_create,
 971     .bdrv_getlength               = qemu_gluster_getlength,
 972     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
 973     .bdrv_truncate                = qemu_gluster_truncate,
 974     .bdrv_co_readv                = qemu_gluster_co_readv,
 975     .bdrv_co_writev               = qemu_gluster_co_writev,
 976     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
 977     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
 978 #ifdef CONFIG_GLUSTERFS_DISCARD
 979     .bdrv_co_discard              = qemu_gluster_co_discard,
 980 #endif
 981 #ifdef CONFIG_GLUSTERFS_ZEROFILL
 982     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
 983 #endif
 984     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
 985     .create_opts                  = &qemu_gluster_create_opts,
 986 };
 987
 988 static BlockDriver bdrv_gluster_tcp = {
 989     .format_name                  = "gluster",
 990     .protocol_name                = "gluster+tcp",
 991     .instance_size                = sizeof(BDRVGlusterState),
 992     .bdrv_needs_filename          = true,
 993     .bdrv_file_open               = qemu_gluster_open,
 994     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
 995     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
 996     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
 997     .bdrv_close                   = qemu_gluster_close,
 998     .bdrv_create                  = qemu_gluster_create,
 999     .bdrv_getlength               = qemu_gluster_getlength,
1000     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1001     .bdrv_truncate                = qemu_gluster_truncate,
1002     .bdrv_co_readv                = qemu_gluster_co_readv,
1003     .bdrv_co_writev               = qemu_gluster_co_writev,
1004     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1005     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1006 #ifdef CONFIG_GLUSTERFS_DISCARD
1007     .bdrv_co_discard              = qemu_gluster_co_discard,
1008 #endif
1009 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1010     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1011 #endif
1012     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1013     .create_opts                  = &qemu_gluster_create_opts,
1014 };
1015
1016 static BlockDriver bdrv_gluster_unix = {
1017     .format_name                  = "gluster",
1018     .protocol_name                = "gluster+unix",
1019     .instance_size                = sizeof(BDRVGlusterState),
1020     .bdrv_needs_filename          = true,
1021     .bdrv_file_open               = qemu_gluster_open,
1022     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1023     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1024     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1025     .bdrv_close                   = qemu_gluster_close,
1026     .bdrv_create                  = qemu_gluster_create,
1027     .bdrv_getlength               = qemu_gluster_getlength,
1028     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1029     .bdrv_truncate                = qemu_gluster_truncate,
1030     .bdrv_co_readv                = qemu_gluster_co_readv,
1031     .bdrv_co_writev               = qemu_gluster_co_writev,
1032     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1033     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1034 #ifdef CONFIG_GLUSTERFS_DISCARD
1035     .bdrv_co_discard              = qemu_gluster_co_discard,
1036 #endif
1037 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1038     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1039 #endif
1040     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1041     .create_opts                  = &qemu_gluster_create_opts,
1042 };
1043
1044 static BlockDriver bdrv_gluster_rdma = {
1045     .format_name                  = "gluster",
1046     .protocol_name                = "gluster+rdma",
1047     .instance_size                = sizeof(BDRVGlusterState),
1048     .bdrv_needs_filename          = true,
1049     .bdrv_file_open               = qemu_gluster_open,
1050     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1051     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1052     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1053     .bdrv_close                   = qemu_gluster_close,
1054     .bdrv_create                  = qemu_gluster_create,
1055     .bdrv_getlength               = qemu_gluster_getlength,
1056     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1057     .bdrv_truncate                = qemu_gluster_truncate,
1058     .bdrv_co_readv                = qemu_gluster_co_readv,
1059     .bdrv_co_writev               = qemu_gluster_co_writev,
1060     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1061     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1062 #ifdef CONFIG_GLUSTERFS_DISCARD
1063     .bdrv_co_discard              = qemu_gluster_co_discard,
1064 #endif
1065 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1066     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1067 #endif
1068     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1069     .create_opts                  = &qemu_gluster_create_opts,
1070 };
1071
1072 static void bdrv_gluster_init(void)
1073 {
1074     bdrv_register(&bdrv_gluster_rdma);
1075     bdrv_register(&bdrv_gluster_unix);
1076     bdrv_register(&bdrv_gluster_tcp);
1077     bdrv_register(&bdrv_gluster);
1078 }
1079
1080 block_init(bdrv_gluster_init);