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
  11 #include "qemu/osdep.h"
  12 #include <glusterfs/api/glfs.h>
  13 #include "block/block_int.h"
  14 #include "block/qdict.h"
  15 #include "qapi/error.h"
  16 #include "qapi/qmp/qdict.h"
  17 #include "qapi/qmp/qerror.h"
  18 #include "qemu/uri.h"
  19 #include "qemu/error-report.h"
  20 #include "qemu/option.h"
  21 #include "qemu/cutils.h"
  22
  23 #define GLUSTER_OPT_FILENAME        "filename"
  24 #define GLUSTER_OPT_VOLUME          "volume"
  25 #define GLUSTER_OPT_PATH            "path"
  26 #define GLUSTER_OPT_TYPE            "type"
  27 #define GLUSTER_OPT_SERVER_PATTERN  "server."
  28 #define GLUSTER_OPT_HOST            "host"
  29 #define GLUSTER_OPT_PORT            "port"
  30 #define GLUSTER_OPT_TO              "to"
  31 #define GLUSTER_OPT_IPV4            "ipv4"
  32 #define GLUSTER_OPT_IPV6            "ipv6"
  33 #define GLUSTER_OPT_SOCKET          "socket"
  34 #define GLUSTER_OPT_DEBUG           "debug"
  35 #define GLUSTER_DEFAULT_PORT        24007
  36 #define GLUSTER_DEBUG_DEFAULT       4
  37 #define GLUSTER_DEBUG_MAX           9
  38 #define GLUSTER_OPT_LOGFILE         "logfile"
  39 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
  40
  41 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
  42
  43 typedef struct GlusterAIOCB {
  44     int64_t size;
  45     int ret;
  46     Coroutine *coroutine;
  47     AioContext *aio_context;
  48 } GlusterAIOCB;
  49
  50 typedef struct BDRVGlusterState {
  51     struct glfs *glfs;
  52     struct glfs_fd *fd;
  53     char *logfile;
  54     bool supports_seek_data;
  55     int debug;
  56 } BDRVGlusterState;
  57
  58 typedef struct BDRVGlusterReopenState {
  59     struct glfs *glfs;
  60     struct glfs_fd *fd;
  61 } BDRVGlusterReopenState;
  62
  63
  64 typedef struct GlfsPreopened {
  65     char *volume;
  66     glfs_t *fs;
  67     int ref;
  68 } GlfsPreopened;
  69
  70 typedef struct ListElement {
  71     QLIST_ENTRY(ListElement) list;
  72     GlfsPreopened saved;
  73 } ListElement;
  74
  75 static QLIST_HEAD(glfs_list, ListElement) glfs_list;
  76
  77 static QemuOptsList qemu_gluster_create_opts = {
  78     .name = "qemu-gluster-create-opts",
  79     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
  80     .desc = {
  81         {
  82             .name = BLOCK_OPT_SIZE,
  83             .type = QEMU_OPT_SIZE,
  84             .help = "Virtual disk size"
  85         },
  86         {
  87             .name = BLOCK_OPT_PREALLOC,
  88             .type = QEMU_OPT_STRING,
  89             .help = "Preallocation mode (allowed values: off, full)"
  90         },
  91         {
  92             .name = GLUSTER_OPT_DEBUG,
  93             .type = QEMU_OPT_NUMBER,
  94             .help = "Gluster log level, valid range is 0-9",
  95         },
  96         {
  97             .name = GLUSTER_OPT_LOGFILE,
  98             .type = QEMU_OPT_STRING,
  99             .help = "Logfile path of libgfapi",
 100         },
 101         { /* end of list */ }
 102     }
 103 };
 104
 105 static QemuOptsList runtime_opts = {
 106     .name = "gluster",
 107     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 108     .desc = {
 109         {
 110             .name = GLUSTER_OPT_FILENAME,
 111             .type = QEMU_OPT_STRING,
 112             .help = "URL to the gluster image",
 113         },
 114         {
 115             .name = GLUSTER_OPT_DEBUG,
 116             .type = QEMU_OPT_NUMBER,
 117             .help = "Gluster log level, valid range is 0-9",
 118         },
 119         {
 120             .name = GLUSTER_OPT_LOGFILE,
 121             .type = QEMU_OPT_STRING,
 122             .help = "Logfile path of libgfapi",
 123         },
 124         { /* end of list */ }
 125     },
 126 };
 127
 128 static QemuOptsList runtime_json_opts = {
 129     .name = "gluster_json",
 130     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
 131     .desc = {
 132         {
 133             .name = GLUSTER_OPT_VOLUME,
 134             .type = QEMU_OPT_STRING,
 135             .help = "name of gluster volume where VM image resides",
 136         },
 137         {
 138             .name = GLUSTER_OPT_PATH,
 139             .type = QEMU_OPT_STRING,
 140             .help = "absolute path to image file in gluster volume",
 141         },
 142         {
 143             .name = GLUSTER_OPT_DEBUG,
 144             .type = QEMU_OPT_NUMBER,
 145             .help = "Gluster log level, valid range is 0-9",
 146         },
 147         { /* end of list */ }
 148     },
 149 };
 150
 151 static QemuOptsList runtime_type_opts = {
 152     .name = "gluster_type",
 153     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
 154     .desc = {
 155         {
 156             .name = GLUSTER_OPT_TYPE,
 157             .type = QEMU_OPT_STRING,
 158             .help = "inet|unix",
 159         },
 160         { /* end of list */ }
 161     },
 162 };
 163
 164 static QemuOptsList runtime_unix_opts = {
 165     .name = "gluster_unix",
 166     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
 167     .desc = {
 168         {
 169             .name = GLUSTER_OPT_SOCKET,
 170             .type = QEMU_OPT_STRING,
 171             .help = "socket file path (legacy)",
 172         },
 173         {
 174             .name = GLUSTER_OPT_PATH,
 175             .type = QEMU_OPT_STRING,
 176             .help = "socket file path (QAPI)",
 177         },
 178         { /* end of list */ }
 179     },
 180 };
 181
 182 static QemuOptsList runtime_inet_opts = {
 183     .name = "gluster_inet",
 184     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
 185     .desc = {
 186         {
 187             .name = GLUSTER_OPT_TYPE,
 188             .type = QEMU_OPT_STRING,
 189             .help = "inet|unix",
 190         },
 191         {
 192             .name = GLUSTER_OPT_HOST,
 193             .type = QEMU_OPT_STRING,
 194             .help = "host address (hostname/ipv4/ipv6 addresses)",
 195         },
 196         {
 197             .name = GLUSTER_OPT_PORT,
 198             .type = QEMU_OPT_STRING,
 199             .help = "port number on which glusterd is listening (default 24007)",
 200         },
 201         {
 202             .name = "to",
 203             .type = QEMU_OPT_NUMBER,
 204             .help = "max port number, not supported by gluster",
 205         },
 206         {
 207             .name = "ipv4",
 208             .type = QEMU_OPT_BOOL,
 209             .help = "ipv4 bool value, not supported by gluster",
 210         },
 211         {
 212             .name = "ipv6",
 213             .type = QEMU_OPT_BOOL,
 214             .help = "ipv6 bool value, not supported by gluster",
 215         },
 216         { /* end of list */ }
 217     },
 218 };
 219
 220 static void glfs_set_preopened(const char *volume, glfs_t *fs)
 221 {
 222     ListElement *entry = NULL;
 223
 224     entry = g_new(ListElement, 1);
 225
 226     entry->saved.volume = g_strdup(volume);
 227
 228     entry->saved.fs = fs;
 229     entry->saved.ref = 1;
 230
 231     QLIST_INSERT_HEAD(&glfs_list, entry, list);
 232 }
 233
 234 static glfs_t *glfs_find_preopened(const char *volume)
 235 {
 236     ListElement *entry = NULL;
 237
 238      QLIST_FOREACH(entry, &glfs_list, list) {
 239         if (strcmp(entry->saved.volume, volume) == 0) {
 240             entry->saved.ref++;
 241             return entry->saved.fs;
 242         }
 243      }
 244
 245     return NULL;
 246 }
 247
 248 static void glfs_clear_preopened(glfs_t *fs)
 249 {
 250     ListElement *entry = NULL;
 251     ListElement *next;
 252
 253     if (fs == NULL) {
 254         return;
 255     }
 256
 257     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
 258         if (entry->saved.fs == fs) {
 259             if (--entry->saved.ref) {
 260                 return;
 261             }
 262
 263             QLIST_REMOVE(entry, list);
 264
 265             glfs_fini(entry->saved.fs);
 266             g_free(entry->saved.volume);
 267             g_free(entry);
 268         }
 269     }
 270 }
 271
 272 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
 273 {
 274     char *p, *q;
 275
 276     if (!path) {
 277         return -EINVAL;
 278     }
 279
 280     /* volume */
 281     p = q = path + strspn(path, "/");
 282     p += strcspn(p, "/");
 283     if (*p == '\0') {
 284         return -EINVAL;
 285     }
 286     gconf->volume = g_strndup(q, p - q);
 287
 288     /* path */
 289     p += strspn(p, "/");
 290     if (*p == '\0') {
 291         return -EINVAL;
 292     }
 293     gconf->path = g_strdup(p);
 294     return 0;
 295 }
 296
 297 /*
 298  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
 299  *
 300  * 'gluster' is the protocol.
 301  *
 302  * 'transport' specifies the transport type used to connect to gluster
 303  * management daemon (glusterd). Valid transport types are
 304  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
 305  *
 306  * 'host' specifies the host where the volume file specification for
 307  * the given volume resides. This can be either hostname or ipv4 address.
 308  * If transport type is 'unix', then 'host' field should not be specified.
 309  * The 'socket' field needs to be populated with the path to unix domain
 310  * socket.
 311  *
 312  * 'port' is the port number on which glusterd is listening. This is optional
 313  * and if not specified, QEMU will send 0 which will make gluster to use the
 314  * default port. If the transport type is unix, then 'port' should not be
 315  * specified.
 316  *
 317  * 'volume' is the name of the gluster volume which contains the VM image.
 318  *
 319  * 'path' is the path to the actual VM image that resides on gluster volume.
 320  *
 321  * Examples:
 322  *
 323  * file=gluster://1.2.3.4/testvol/a.img
 324  * file=gluster+tcp://1.2.3.4/testvol/a.img
 325  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
 326  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
 327  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
 328  */
 329 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
 330                                   const char *filename)
 331 {
 332     SocketAddress *gsconf;
 333     URI *uri;
 334     QueryParams *qp = NULL;
 335     bool is_unix = false;
 336     int ret = 0;
 337
 338     uri = uri_parse(filename);
 339     if (!uri) {
 340         return -EINVAL;
 341     }
 342
 343     gconf->server = g_new0(SocketAddressList, 1);
 344     gconf->server->value = gsconf = g_new0(SocketAddress, 1);
 345
 346     /* transport */
 347     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
 348         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
 349     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
 350         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
 351     } else if (!strcmp(uri->scheme, "gluster+unix")) {
 352         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
 353         is_unix = true;
 354     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
 355         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
 356         warn_report("rdma feature is not supported, falling back to tcp");
 357     } else {
 358         ret = -EINVAL;
 359         goto out;
 360     }
 361
 362     ret = parse_volume_options(gconf, uri->path);
 363     if (ret < 0) {
 364         goto out;
 365     }
 366
 367     qp = query_params_parse(uri->query);
 368     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
 369         ret = -EINVAL;
 370         goto out;
 371     }
 372
 373     if (is_unix) {
 374         if (uri->server || uri->port) {
 375             ret = -EINVAL;
 376             goto out;
 377         }
 378         if (strcmp(qp->p[0].name, "socket")) {
 379             ret = -EINVAL;
 380             goto out;
 381         }
 382         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
 383     } else {
 384         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
 385         if (uri->port) {
 386             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
 387         } else {
 388             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
 389         }
 390     }
 391
 392 out:
 393     if (qp) {
 394         query_params_free(qp);
 395     }
 396     uri_free(uri);
 397     return ret;
 398 }
 399
 400 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
 401                                            Error **errp)
 402 {
 403     struct glfs *glfs;
 404     int ret;
 405     int old_errno;
 406     SocketAddressList *server;
 407     unsigned long long port;
 408
 409     glfs = glfs_find_preopened(gconf->volume);
 410     if (glfs) {
 411         return glfs;
 412     }
 413
 414     glfs = glfs_new(gconf->volume);
 415     if (!glfs) {
 416         goto out;
 417     }
 418
 419     glfs_set_preopened(gconf->volume, glfs);
 420
 421     for (server = gconf->server; server; server = server->next) {
 422         switch (server->value->type) {
 423         case SOCKET_ADDRESS_TYPE_UNIX:
 424             ret = glfs_set_volfile_server(glfs, "unix",
 425                                    server->value->u.q_unix.path, 0);
 426             break;
 427         case SOCKET_ADDRESS_TYPE_INET:
 428             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
 429                 port > 65535) {
 430                 error_setg(errp, "'%s' is not a valid port number",
 431                            server->value->u.inet.port);
 432                 errno = EINVAL;
 433                 goto out;
 434             }
 435             ret = glfs_set_volfile_server(glfs, "tcp",
 436                                    server->value->u.inet.host,
 437                                    (int)port);
 438             break;
 439         case SOCKET_ADDRESS_TYPE_VSOCK:
 440         case SOCKET_ADDRESS_TYPE_FD:
 441         default:
 442             abort();
 443         }
 444
 445         if (ret < 0) {
 446             goto out;
 447         }
 448     }
 449
 450     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
 451     if (ret < 0) {
 452         goto out;
 453     }
 454
 455     ret = glfs_init(glfs);
 456     if (ret) {
 457         error_setg(errp, "Gluster connection for volume %s, path %s failed"
 458                          " to connect", gconf->volume, gconf->path);
 459         for (server = gconf->server; server; server = server->next) {
 460             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
 461                 error_append_hint(errp, "hint: failed on socket %s ",
 462                                   server->value->u.q_unix.path);
 463             } else {
 464                 error_append_hint(errp, "hint: failed on host %s and port %s ",
 465                                   server->value->u.inet.host,
 466                                   server->value->u.inet.port);
 467             }
 468         }
 469
 470         error_append_hint(errp, "Please refer to gluster logs for more info\n");
 471
 472         /* glfs_init sometimes doesn't set errno although docs suggest that */
 473         if (errno == 0) {
 474             errno = EINVAL;
 475         }
 476
 477         goto out;
 478     }
 479     return glfs;
 480
 481 out:
 482     if (glfs) {
 483         old_errno = errno;
 484         glfs_clear_preopened(glfs);
 485         errno = old_errno;
 486     }
 487     return NULL;
 488 }
 489
 490 /*
 491  * Convert the json formatted command line into qapi.
 492 */
 493 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
 494                                   QDict *options, Error **errp)
 495 {
 496     QemuOpts *opts;
 497     SocketAddress *gsconf = NULL;
 498     SocketAddressList *curr = NULL;
 499     QDict *backing_options = NULL;
 500     Error *local_err = NULL;
 501     char *str = NULL;
 502     const char *ptr;
 503     int i, type, num_servers;
 504
 505     /* create opts info from runtime_json_opts list */
 506     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
 507     qemu_opts_absorb_qdict(opts, options, &local_err);
 508     if (local_err) {
 509         goto out;
 510     }
 511
 512     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
 513     if (num_servers < 1) {
 514         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
 515         goto out;
 516     }
 517
 518     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
 519     if (!ptr) {
 520         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
 521         goto out;
 522     }
 523     gconf->volume = g_strdup(ptr);
 524
 525     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
 526     if (!ptr) {
 527         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
 528         goto out;
 529     }
 530     gconf->path = g_strdup(ptr);
 531     qemu_opts_del(opts);
 532
 533     for (i = 0; i < num_servers; i++) {
 534         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
 535         qdict_extract_subqdict(options, &backing_options, str);
 536
 537         /* create opts info from runtime_type_opts list */
 538         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
 539         qemu_opts_absorb_qdict(opts, backing_options, &local_err);
 540         if (local_err) {
 541             goto out;
 542         }
 543
 544         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
 545         if (!ptr) {
 546             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
 547             error_append_hint(&local_err, GERR_INDEX_HINT, i);
 548             goto out;
 549
 550         }
 551         gsconf = g_new0(SocketAddress, 1);
 552         if (!strcmp(ptr, "tcp")) {
 553             ptr = "inet";       /* accept legacy "tcp" */
 554         }
 555         type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
 556         if (type != SOCKET_ADDRESS_TYPE_INET
 557             && type != SOCKET_ADDRESS_TYPE_UNIX) {
 558             error_setg(&local_err,
 559                        "Parameter '%s' may be 'inet' or 'unix'",
 560                        GLUSTER_OPT_TYPE);
 561             error_append_hint(&local_err, GERR_INDEX_HINT, i);
 562             goto out;
 563         }
 564         gsconf->type = type;
 565         qemu_opts_del(opts);
 566
 567         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
 568             /* create opts info from runtime_inet_opts list */
 569             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
 570             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
 571             if (local_err) {
 572                 goto out;
 573             }
 574
 575             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
 576             if (!ptr) {
 577                 error_setg(&local_err, QERR_MISSING_PARAMETER,
 578                            GLUSTER_OPT_HOST);
 579                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
 580                 goto out;
 581             }
 582             gsconf->u.inet.host = g_strdup(ptr);
 583             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
 584             if (!ptr) {
 585                 error_setg(&local_err, QERR_MISSING_PARAMETER,
 586                            GLUSTER_OPT_PORT);
 587                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
 588                 goto out;
 589             }
 590             gsconf->u.inet.port = g_strdup(ptr);
 591
 592             /* defend for unsupported fields in InetSocketAddress,
 593              * i.e. @ipv4, @ipv6  and @to
 594              */
 595             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
 596             if (ptr) {
 597                 gsconf->u.inet.has_to = true;
 598             }
 599             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
 600             if (ptr) {
 601                 gsconf->u.inet.has_ipv4 = true;
 602             }
 603             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
 604             if (ptr) {
 605                 gsconf->u.inet.has_ipv6 = true;
 606             }
 607             if (gsconf->u.inet.has_to) {
 608                 error_setg(&local_err, "Parameter 'to' not supported");
 609                 goto out;
 610             }
 611             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
 612                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
 613                 goto out;
 614             }
 615             qemu_opts_del(opts);
 616         } else {
 617             /* create opts info from runtime_unix_opts list */
 618             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
 619             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
 620             if (local_err) {
 621                 goto out;
 622             }
 623
 624             ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
 625             if (!ptr) {
 626                 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
 627             } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
 628                 error_setg(&local_err,
 629                            "Conflicting parameters 'path' and 'socket'");
 630                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
 631                 goto out;
 632             }
 633             if (!ptr) {
 634                 error_setg(&local_err, QERR_MISSING_PARAMETER,
 635                            GLUSTER_OPT_PATH);
 636                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
 637                 goto out;
 638             }
 639             gsconf->u.q_unix.path = g_strdup(ptr);
 640             qemu_opts_del(opts);
 641         }
 642
 643         if (gconf->server == NULL) {
 644             gconf->server = g_new0(SocketAddressList, 1);
 645             gconf->server->value = gsconf;
 646             curr = gconf->server;
 647         } else {
 648             curr->next = g_new0(SocketAddressList, 1);
 649             curr->next->value = gsconf;
 650             curr = curr->next;
 651         }
 652         gsconf = NULL;
 653
 654         qobject_unref(backing_options);
 655         backing_options = NULL;
 656         g_free(str);
 657         str = NULL;
 658     }
 659
 660     return 0;
 661
 662 out:
 663     error_propagate(errp, local_err);
 664     qapi_free_SocketAddress(gsconf);
 665     qemu_opts_del(opts);
 666     g_free(str);
 667     qobject_unref(backing_options);
 668     errno = EINVAL;
 669     return -errno;
 670 }
 671
 672 /* Converts options given in @filename and the @options QDict into the QAPI
 673  * object @gconf. */
 674 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
 675                               const char *filename,
 676                               QDict *options, Error **errp)
 677 {
 678     int ret;
 679     if (filename) {
 680         ret = qemu_gluster_parse_uri(gconf, filename);
 681         if (ret < 0) {
 682             error_setg(errp, "invalid URI %s", filename);
 683             error_append_hint(errp, "Usage: file=gluster[+transport]://"
 684                                     "[host[:port]]volume/path[?socket=...]"
 685                                     "[,file.debug=N]"
 686                                     "[,file.logfile=/path/filename.log]\n");
 687             return ret;
 688         }
 689     } else {
 690         ret = qemu_gluster_parse_json(gconf, options, errp);
 691         if (ret < 0) {
 692             error_append_hint(errp, "Usage: "
 693                              "-drive driver=qcow2,file.driver=gluster,"
 694                              "file.volume=testvol,file.path=/path/a.qcow2"
 695                              "[,file.debug=9]"
 696                              "[,file.logfile=/path/filename.log],"
 697                              "file.server.0.type=inet,"
 698                              "file.server.0.host=1.2.3.4,"
 699                              "file.server.0.port=24007,"
 700                              "file.server.1.transport=unix,"
 701                              "file.server.1.path=/var/run/glusterd.socket ..."
 702                              "\n");
 703             return ret;
 704         }
 705     }
 706
 707     return 0;
 708 }
 709
 710 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
 711                                       const char *filename,
 712                                       QDict *options, Error **errp)
 713 {
 714     int ret;
 715
 716     ret = qemu_gluster_parse(gconf, filename, options, errp);
 717     if (ret < 0) {
 718         errno = -ret;
 719         return NULL;
 720     }
 721
 722     return qemu_gluster_glfs_init(gconf, errp);
 723 }
 724
 725 /*
 726  * AIO callback routine called from GlusterFS thread.
 727  */
 728 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
 729 {
 730     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
 731
 732     if (!ret || ret == acb->size) {
 733         acb->ret = 0; /* Success */
 734     } else if (ret < 0) {
 735         acb->ret = -errno; /* Read/Write failed */
 736     } else {
 737         acb->ret = -EIO; /* Partial read/write - fail it */
 738     }
 739
 740     aio_co_schedule(acb->aio_context, acb->coroutine);
 741 }
 742
 743 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
 744 {
 745     assert(open_flags != NULL);
 746
 747     *open_flags |= O_BINARY;
 748
 749     if (bdrv_flags & BDRV_O_RDWR) {
 750         *open_flags |= O_RDWR;
 751     } else {
 752         *open_flags |= O_RDONLY;
 753     }
 754
 755     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 756         *open_flags |= O_DIRECT;
 757     }
 758 }
 759
 760 /*
 761  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
 762  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
 763  * - Corrected versions return -1 and set errno to EINVAL.
 764  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
 765  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
 766  */
 767 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
 768 {
 769     off_t ret = 0;
 770
 771 #if defined SEEK_HOLE && defined SEEK_DATA
 772     off_t eof;
 773
 774     eof = glfs_lseek(fd, 0, SEEK_END);
 775     if (eof < 0) {
 776         /* this should never occur */
 777         return false;
 778     }
 779
 780     /* this should always fail with ENXIO if SEEK_DATA is supported */
 781     ret = glfs_lseek(fd, eof, SEEK_DATA);
 782 #endif
 783
 784     return (ret < 0) && (errno == ENXIO);
 785 }
 786
 787 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
 788                              int bdrv_flags, Error **errp)
 789 {
 790     BDRVGlusterState *s = bs->opaque;
 791     int open_flags = 0;
 792     int ret = 0;
 793     BlockdevOptionsGluster *gconf = NULL;
 794     QemuOpts *opts;
 795     Error *local_err = NULL;
 796     const char *filename, *logfile;
 797
 798     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 799     qemu_opts_absorb_qdict(opts, options, &local_err);
 800     if (local_err) {
 801         error_propagate(errp, local_err);
 802         ret = -EINVAL;
 803         goto out;
 804     }
 805
 806     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
 807
 808     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
 809                                    GLUSTER_DEBUG_DEFAULT);
 810     if (s->debug < 0) {
 811         s->debug = 0;
 812     } else if (s->debug > GLUSTER_DEBUG_MAX) {
 813         s->debug = GLUSTER_DEBUG_MAX;
 814     }
 815
 816     gconf = g_new0(BlockdevOptionsGluster, 1);
 817     gconf->debug = s->debug;
 818     gconf->has_debug = true;
 819
 820     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
 821     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
 822
 823     gconf->logfile = g_strdup(s->logfile);
 824     gconf->has_logfile = true;
 825
 826     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
 827     if (!s->glfs) {
 828         ret = -errno;
 829         goto out;
 830     }
 831
 832 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
 833     /* Without this, if fsync fails for a recoverable reason (for instance,
 834      * ENOSPC), gluster will dump its cache, preventing retries.  This means
 835      * almost certain data loss.  Not all gluster versions support the
 836      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
 837      * discover during runtime if it is supported (this api returns success for
 838      * unknown key/value pairs) */
 839     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
 840                                           "resync-failed-syncs-after-fsync",
 841                                           "on");
 842     if (ret < 0) {
 843         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
 844         ret = -errno;
 845         goto out;
 846     }
 847 #endif
 848
 849     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
 850
 851     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
 852     if (!s->fd) {
 853         ret = -errno;
 854     }
 855
 856     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
 857
 858 out:
 859     qemu_opts_del(opts);
 860     qapi_free_BlockdevOptionsGluster(gconf);
 861     if (!ret) {
 862         return ret;
 863     }
 864     g_free(s->logfile);
 865     if (s->fd) {
 866         glfs_close(s->fd);
 867     }
 868
 869     glfs_clear_preopened(s->glfs);
 870
 871     return ret;
 872 }
 873
 874 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
 875                                        BlockReopenQueue *queue, Error **errp)
 876 {
 877     int ret = 0;
 878     BDRVGlusterState *s;
 879     BDRVGlusterReopenState *reop_s;
 880     BlockdevOptionsGluster *gconf;
 881     int open_flags = 0;
 882
 883     assert(state != NULL);
 884     assert(state->bs != NULL);
 885
 886     s = state->bs->opaque;
 887
 888     state->opaque = g_new0(BDRVGlusterReopenState, 1);
 889     reop_s = state->opaque;
 890
 891     qemu_gluster_parse_flags(state->flags, &open_flags);
 892
 893     gconf = g_new0(BlockdevOptionsGluster, 1);
 894     gconf->debug = s->debug;
 895     gconf->has_debug = true;
 896     gconf->logfile = g_strdup(s->logfile);
 897     gconf->has_logfile = true;
 898     reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
 899     if (reop_s->glfs == NULL) {
 900         ret = -errno;
 901         goto exit;
 902     }
 903
 904 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
 905     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
 906                                  "resync-failed-syncs-after-fsync", "on");
 907     if (ret < 0) {
 908         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
 909         ret = -errno;
 910         goto exit;
 911     }
 912 #endif
 913
 914     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
 915     if (reop_s->fd == NULL) {
 916         /* reops->glfs will be cleaned up in _abort */
 917         ret = -errno;
 918         goto exit;
 919     }
 920
 921 exit:
 922     /* state->opaque will be freed in either the _abort or _commit */
 923     qapi_free_BlockdevOptionsGluster(gconf);
 924     return ret;
 925 }
 926
 927 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
 928 {
 929     BDRVGlusterReopenState *reop_s = state->opaque;
 930     BDRVGlusterState *s = state->bs->opaque;
 931
 932
 933     /* close the old */
 934     if (s->fd) {
 935         glfs_close(s->fd);
 936     }
 937
 938     glfs_clear_preopened(s->glfs);
 939
 940     /* use the newly opened image / connection */
 941     s->fd         = reop_s->fd;
 942     s->glfs       = reop_s->glfs;
 943
 944     g_free(state->opaque);
 945     state->opaque = NULL;
 946
 947     return;
 948 }
 949
 950
 951 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
 952 {
 953     BDRVGlusterReopenState *reop_s = state->opaque;
 954
 955     if (reop_s == NULL) {
 956         return;
 957     }
 958
 959     if (reop_s->fd) {
 960         glfs_close(reop_s->fd);
 961     }
 962
 963     glfs_clear_preopened(reop_s->glfs);
 964
 965     g_free(state->opaque);
 966     state->opaque = NULL;
 967
 968     return;
 969 }
 970
 971 #ifdef CONFIG_GLUSTERFS_ZEROFILL
 972 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
 973                                                       int64_t offset,
 974                                                       int size,
 975                                                       BdrvRequestFlags flags)
 976 {
 977     int ret;
 978     GlusterAIOCB acb;
 979     BDRVGlusterState *s = bs->opaque;
 980
 981     acb.size = size;
 982     acb.ret = 0;
 983     acb.coroutine = qemu_coroutine_self();
 984     acb.aio_context = bdrv_get_aio_context(bs);
 985
 986     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
 987     if (ret < 0) {
 988         return -errno;
 989     }
 990
 991     qemu_coroutine_yield();
 992     return acb.ret;
 993 }
 994 #endif
 995
 996 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
 997                                     PreallocMode prealloc, Error **errp)
 998 {
 999     int64_t current_length;
1000
1001     current_length = glfs_lseek(fd, 0, SEEK_END);
1002     if (current_length < 0) {
1003         error_setg_errno(errp, errno, "Failed to determine current size");
1004         return -errno;
1005     }
1006
1007     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1008         error_setg(errp, "Cannot use preallocation for shrinking files");
1009         return -ENOTSUP;
1010     }
1011
1012     if (current_length == offset) {
1013         return 0;
1014     }
1015
1016     switch (prealloc) {
1017 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1018     case PREALLOC_MODE_FALLOC:
1019         if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1020             error_setg_errno(errp, errno, "Could not preallocate data");
1021             return -errno;
1022         }
1023         break;
1024 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1025 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1026     case PREALLOC_MODE_FULL:
1027         if (glfs_ftruncate(fd, offset)) {
1028             error_setg_errno(errp, errno, "Could not resize file");
1029             return -errno;
1030         }
1031         if (glfs_zerofill(fd, current_length, offset - current_length)) {
1032             error_setg_errno(errp, errno, "Could not zerofill the new area");
1033             return -errno;
1034         }
1035         break;
1036 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1037     case PREALLOC_MODE_OFF:
1038         if (glfs_ftruncate(fd, offset)) {
1039             error_setg_errno(errp, errno, "Could not resize file");
1040             return -errno;
1041         }
1042         break;
1043     default:
1044         error_setg(errp, "Unsupported preallocation mode: %s",
1045                    PreallocMode_str(prealloc));
1046         return -EINVAL;
1047     }
1048
1049     return 0;
1050 }
1051
1052 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1053                                   Error **errp)
1054 {
1055     BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1056     struct glfs *glfs;
1057     struct glfs_fd *fd = NULL;
1058     int ret = 0;
1059
1060     assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1061
1062     glfs = qemu_gluster_glfs_init(opts->location, errp);
1063     if (!glfs) {
1064         ret = -errno;
1065         goto out;
1066     }
1067
1068     fd = glfs_creat(glfs, opts->location->path,
1069                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1070     if (!fd) {
1071         ret = -errno;
1072         goto out;
1073     }
1074
1075     ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1076
1077 out:
1078     if (fd) {
1079         if (glfs_close(fd) != 0 && ret == 0) {
1080             ret = -errno;
1081         }
1082     }
1083     glfs_clear_preopened(glfs);
1084     return ret;
1085 }
1086
1087 static int coroutine_fn qemu_gluster_co_create_opts(const char *filename,
1088                                                     QemuOpts *opts,
1089                                                     Error **errp)
1090 {
1091     BlockdevCreateOptions *options;
1092     BlockdevCreateOptionsGluster *gopts;
1093     BlockdevOptionsGluster *gconf;
1094     char *tmp = NULL;
1095     Error *local_err = NULL;
1096     int ret;
1097
1098     options = g_new0(BlockdevCreateOptions, 1);
1099     options->driver = BLOCKDEV_DRIVER_GLUSTER;
1100     gopts = &options->u.gluster;
1101
1102     gconf = g_new0(BlockdevOptionsGluster, 1);
1103     gopts->location = gconf;
1104
1105     gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1106                            BDRV_SECTOR_SIZE);
1107
1108     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1109     gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1110                                            PREALLOC_MODE_OFF, &local_err);
1111     g_free(tmp);
1112     if (local_err) {
1113         error_propagate(errp, local_err);
1114         ret = -EINVAL;
1115         goto fail;
1116     }
1117
1118     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1119                                            GLUSTER_DEBUG_DEFAULT);
1120     if (gconf->debug < 0) {
1121         gconf->debug = 0;
1122     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1123         gconf->debug = GLUSTER_DEBUG_MAX;
1124     }
1125     gconf->has_debug = true;
1126
1127     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1128     if (!gconf->logfile) {
1129         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1130     }
1131     gconf->has_logfile = true;
1132
1133     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1134     if (ret < 0) {
1135         goto fail;
1136     }
1137
1138     ret = qemu_gluster_co_create(options, errp);
1139     if (ret < 0) {
1140         goto fail;
1141     }
1142
1143     ret = 0;
1144 fail:
1145     qapi_free_BlockdevCreateOptions(options);
1146     return ret;
1147 }
1148
1149 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1150                                            int64_t sector_num, int nb_sectors,
1151                                            QEMUIOVector *qiov, int write)
1152 {
1153     int ret;
1154     GlusterAIOCB acb;
1155     BDRVGlusterState *s = bs->opaque;
1156     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1157     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1158
1159     acb.size = size;
1160     acb.ret = 0;
1161     acb.coroutine = qemu_coroutine_self();
1162     acb.aio_context = bdrv_get_aio_context(bs);
1163
1164     if (write) {
1165         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1166                                  gluster_finish_aiocb, &acb);
1167     } else {
1168         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1169                                 gluster_finish_aiocb, &acb);
1170     }
1171
1172     if (ret < 0) {
1173         return -errno;
1174     }
1175
1176     qemu_coroutine_yield();
1177     return acb.ret;
1178 }
1179
1180 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1181                                                  int64_t offset,
1182                                                  PreallocMode prealloc,
1183                                                  Error **errp)
1184 {
1185     BDRVGlusterState *s = bs->opaque;
1186     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1187 }
1188
1189 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1190                                               int64_t sector_num,
1191                                               int nb_sectors,
1192                                               QEMUIOVector *qiov)
1193 {
1194     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1195 }
1196
1197 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1198                                                int64_t sector_num,
1199                                                int nb_sectors,
1200                                                QEMUIOVector *qiov,
1201                                                int flags)
1202 {
1203     assert(!flags);
1204     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1205 }
1206
1207 static void qemu_gluster_close(BlockDriverState *bs)
1208 {
1209     BDRVGlusterState *s = bs->opaque;
1210
1211     g_free(s->logfile);
1212     if (s->fd) {
1213         glfs_close(s->fd);
1214         s->fd = NULL;
1215     }
1216     glfs_clear_preopened(s->glfs);
1217 }
1218
1219 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1220 {
1221     int ret;
1222     GlusterAIOCB acb;
1223     BDRVGlusterState *s = bs->opaque;
1224
1225     acb.size = 0;
1226     acb.ret = 0;
1227     acb.coroutine = qemu_coroutine_self();
1228     acb.aio_context = bdrv_get_aio_context(bs);
1229
1230     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1231     if (ret < 0) {
1232         ret = -errno;
1233         goto error;
1234     }
1235
1236     qemu_coroutine_yield();
1237     if (acb.ret < 0) {
1238         ret = acb.ret;
1239         goto error;
1240     }
1241
1242     return acb.ret;
1243
1244 error:
1245     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1246      * after a fsync failure, so we have no way of allowing the guest to safely
1247      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1248      * either, but will invalidate the fd on error, so this is again our only
1249      * option.
1250      *
1251      * The 'resync-failed-syncs-after-fsync' xlator option for the
1252      * write-behind cache will cause later gluster versions to retain its
1253      * cache after error, so long as the fd remains open.  However, we
1254      * currently have no way of knowing if this option is supported.
1255      *
1256      * TODO: Once gluster provides a way for us to determine if the option
1257      * is supported, bypass the closure and setting drv to NULL.  */
1258     qemu_gluster_close(bs);
1259     bs->drv = NULL;
1260     return ret;
1261 }
1262
1263 #ifdef CONFIG_GLUSTERFS_DISCARD
1264 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1265                                                  int64_t offset, int size)
1266 {
1267     int ret;
1268     GlusterAIOCB acb;
1269     BDRVGlusterState *s = bs->opaque;
1270
1271     acb.size = 0;
1272     acb.ret = 0;
1273     acb.coroutine = qemu_coroutine_self();
1274     acb.aio_context = bdrv_get_aio_context(bs);
1275
1276     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1277     if (ret < 0) {
1278         return -errno;
1279     }
1280
1281     qemu_coroutine_yield();
1282     return acb.ret;
1283 }
1284 #endif
1285
1286 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1287 {
1288     BDRVGlusterState *s = bs->opaque;
1289     int64_t ret;
1290
1291     ret = glfs_lseek(s->fd, 0, SEEK_END);
1292     if (ret < 0) {
1293         return -errno;
1294     } else {
1295         return ret;
1296     }
1297 }
1298
1299 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1300 {
1301     BDRVGlusterState *s = bs->opaque;
1302     struct stat st;
1303     int ret;
1304
1305     ret = glfs_fstat(s->fd, &st);
1306     if (ret < 0) {
1307         return -errno;
1308     } else {
1309         return st.st_blocks * 512;
1310     }
1311 }
1312
1313 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1314 {
1315     /* GlusterFS volume could be backed by a block device */
1316     return 0;
1317 }
1318
1319 /*
1320  * Find allocation range in @bs around offset @start.
1321  * May change underlying file descriptor's file offset.
1322  * If @start is not in a hole, store @start in @data, and the
1323  * beginning of the next hole in @hole, and return 0.
1324  * If @start is in a non-trailing hole, store @start in @hole and the
1325  * beginning of the next non-hole in @data, and return 0.
1326  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1327  * If we can't find out, return a negative errno other than -ENXIO.
1328  *
1329  * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1330  */
1331 static int find_allocation(BlockDriverState *bs, off_t start,
1332                            off_t *data, off_t *hole)
1333 {
1334     BDRVGlusterState *s = bs->opaque;
1335
1336     if (!s->supports_seek_data) {
1337         goto exit;
1338     }
1339
1340 #if defined SEEK_HOLE && defined SEEK_DATA
1341     off_t offs;
1342
1343     /*
1344      * SEEK_DATA cases:
1345      * D1. offs == start: start is in data
1346      * D2. offs > start: start is in a hole, next data at offs
1347      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1348      *                              or start is beyond EOF
1349      *     If the latter happens, the file has been truncated behind
1350      *     our back since we opened it.  All bets are off then.
1351      *     Treating like a trailing hole is simplest.
1352      * D4. offs < 0, errno != ENXIO: we learned nothing
1353      */
1354     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1355     if (offs < 0) {
1356         return -errno;          /* D3 or D4 */
1357     }
1358
1359     if (offs < start) {
1360         /* This is not a valid return by lseek().  We are safe to just return
1361          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1362          *  versions of gluster server will return offs < start, so an assert
1363          *  here will unnecessarily abort QEMU. */
1364         return -EIO;
1365     }
1366
1367     if (offs > start) {
1368         /* D2: in hole, next data at offs */
1369         *hole = start;
1370         *data = offs;
1371         return 0;
1372     }
1373
1374     /* D1: in data, end not yet known */
1375
1376     /*
1377      * SEEK_HOLE cases:
1378      * H1. offs == start: start is in a hole
1379      *     If this happens here, a hole has been dug behind our back
1380      *     since the previous lseek().
1381      * H2. offs > start: either start is in data, next hole at offs,
1382      *                   or start is in trailing hole, EOF at offs
1383      *     Linux treats trailing holes like any other hole: offs ==
1384      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1385      *     If that happens here, a hole has been dug behind our back
1386      *     since the previous lseek().
1387      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1388      *     If this happens, the file has been truncated behind our
1389      *     back since we opened it.  Treat it like a trailing hole.
1390      * H4. offs < 0, errno != ENXIO: we learned nothing
1391      *     Pretend we know nothing at all, i.e. "forget" about D1.
1392      */
1393     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1394     if (offs < 0) {
1395         return -errno;          /* D1 and (H3 or H4) */
1396     }
1397
1398     if (offs < start) {
1399         /* This is not a valid return by lseek().  We are safe to just return
1400          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1401          *  versions of gluster server will return offs < start, so an assert
1402          *  here will unnecessarily abort QEMU. */
1403         return -EIO;
1404     }
1405
1406     if (offs > start) {
1407         /*
1408          * D1 and H2: either in data, next hole at offs, or it was in
1409          * data but is now in a trailing hole.  In the latter case,
1410          * all bets are off.  Treating it as if it there was data all
1411          * the way to EOF is safe, so simply do that.
1412          */
1413         *data = start;
1414         *hole = offs;
1415         return 0;
1416     }
1417
1418     /* D1 and H1 */
1419     return -EBUSY;
1420 #endif
1421
1422 exit:
1423     return -ENOTSUP;
1424 }
1425
1426 /*
1427  * Returns the allocation status of the specified offset.
1428  *
1429  * The block layer guarantees 'offset' and 'bytes' are within bounds.
1430  *
1431  * 'pnum' is set to the number of bytes (including and immediately following
1432  * the specified offset) that are known to be in the same
1433  * allocated/unallocated state.
1434  *
1435  * 'bytes' is the max value 'pnum' should be set to.
1436  *
1437  * (Based on raw_co_block_status() from file-posix.c.)
1438  */
1439 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1440                                                      bool want_zero,
1441                                                      int64_t offset,
1442                                                      int64_t bytes,
1443                                                      int64_t *pnum,
1444                                                      int64_t *map,
1445                                                      BlockDriverState **file)
1446 {
1447     BDRVGlusterState *s = bs->opaque;
1448     off_t data = 0, hole = 0;
1449     int ret = -EINVAL;
1450
1451     if (!s->fd) {
1452         return ret;
1453     }
1454
1455     if (!want_zero) {
1456         *pnum = bytes;
1457         *map = offset;
1458         *file = bs;
1459         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1460     }
1461
1462     ret = find_allocation(bs, offset, &data, &hole);
1463     if (ret == -ENXIO) {
1464         /* Trailing hole */
1465         *pnum = bytes;
1466         ret = BDRV_BLOCK_ZERO;
1467     } else if (ret < 0) {
1468         /* No info available, so pretend there are no holes */
1469         *pnum = bytes;
1470         ret = BDRV_BLOCK_DATA;
1471     } else if (data == offset) {
1472         /* On a data extent, compute bytes to the end of the extent,
1473          * possibly including a partial sector at EOF. */
1474         *pnum = MIN(bytes, hole - offset);
1475         ret = BDRV_BLOCK_DATA;
1476     } else {
1477         /* On a hole, compute bytes to the beginning of the next extent.  */
1478         assert(hole == offset);
1479         *pnum = MIN(bytes, data - offset);
1480         ret = BDRV_BLOCK_ZERO;
1481     }
1482
1483     *map = offset;
1484     *file = bs;
1485
1486     return ret | BDRV_BLOCK_OFFSET_VALID;
1487 }
1488
1489
1490 static BlockDriver bdrv_gluster = {
1491     .format_name                  = "gluster",
1492     .protocol_name                = "gluster",
1493     .instance_size                = sizeof(BDRVGlusterState),
1494     .bdrv_needs_filename          = false,
1495     .bdrv_file_open               = qemu_gluster_open,
1496     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1497     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1498     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1499     .bdrv_close                   = qemu_gluster_close,
1500     .bdrv_co_create               = qemu_gluster_co_create,
1501     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1502     .bdrv_getlength               = qemu_gluster_getlength,
1503     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1504     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1505     .bdrv_co_readv                = qemu_gluster_co_readv,
1506     .bdrv_co_writev               = qemu_gluster_co_writev,
1507     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1508     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1509 #ifdef CONFIG_GLUSTERFS_DISCARD
1510     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1511 #endif
1512 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1513     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1514 #endif
1515     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1516     .create_opts                  = &qemu_gluster_create_opts,
1517 };
1518
1519 static BlockDriver bdrv_gluster_tcp = {
1520     .format_name                  = "gluster",
1521     .protocol_name                = "gluster+tcp",
1522     .instance_size                = sizeof(BDRVGlusterState),
1523     .bdrv_needs_filename          = false,
1524     .bdrv_file_open               = qemu_gluster_open,
1525     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1526     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1527     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1528     .bdrv_close                   = qemu_gluster_close,
1529     .bdrv_co_create               = qemu_gluster_co_create,
1530     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1531     .bdrv_getlength               = qemu_gluster_getlength,
1532     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1533     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1534     .bdrv_co_readv                = qemu_gluster_co_readv,
1535     .bdrv_co_writev               = qemu_gluster_co_writev,
1536     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1537     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1538 #ifdef CONFIG_GLUSTERFS_DISCARD
1539     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1540 #endif
1541 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1542     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1543 #endif
1544     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1545     .create_opts                  = &qemu_gluster_create_opts,
1546 };
1547
1548 static BlockDriver bdrv_gluster_unix = {
1549     .format_name                  = "gluster",
1550     .protocol_name                = "gluster+unix",
1551     .instance_size                = sizeof(BDRVGlusterState),
1552     .bdrv_needs_filename          = true,
1553     .bdrv_file_open               = qemu_gluster_open,
1554     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1555     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1556     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1557     .bdrv_close                   = qemu_gluster_close,
1558     .bdrv_co_create               = qemu_gluster_co_create,
1559     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1560     .bdrv_getlength               = qemu_gluster_getlength,
1561     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1562     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1563     .bdrv_co_readv                = qemu_gluster_co_readv,
1564     .bdrv_co_writev               = qemu_gluster_co_writev,
1565     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1566     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1567 #ifdef CONFIG_GLUSTERFS_DISCARD
1568     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1569 #endif
1570 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1571     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1572 #endif
1573     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1574     .create_opts                  = &qemu_gluster_create_opts,
1575 };
1576
1577 /* rdma is deprecated (actually never supported for volfile fetch).
1578  * Let's maintain it for the protocol compatibility, to make sure things
1579  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1580  * protocol with a warning.
1581  * TODO: remove gluster+rdma interface support
1582  */
1583 static BlockDriver bdrv_gluster_rdma = {
1584     .format_name                  = "gluster",
1585     .protocol_name                = "gluster+rdma",
1586     .instance_size                = sizeof(BDRVGlusterState),
1587     .bdrv_needs_filename          = true,
1588     .bdrv_file_open               = qemu_gluster_open,
1589     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1590     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1591     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1592     .bdrv_close                   = qemu_gluster_close,
1593     .bdrv_co_create               = qemu_gluster_co_create,
1594     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1595     .bdrv_getlength               = qemu_gluster_getlength,
1596     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1597     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1598     .bdrv_co_readv                = qemu_gluster_co_readv,
1599     .bdrv_co_writev               = qemu_gluster_co_writev,
1600     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1601     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1602 #ifdef CONFIG_GLUSTERFS_DISCARD
1603     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1604 #endif
1605 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1606     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1607 #endif
1608     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1609     .create_opts                  = &qemu_gluster_create_opts,
1610 };
1611
1612 static void bdrv_gluster_init(void)
1613 {
1614     bdrv_register(&bdrv_gluster_rdma);
1615     bdrv_register(&bdrv_gluster_unix);
1616     bdrv_register(&bdrv_gluster_tcp);
1617     bdrv_register(&bdrv_gluster);
1618 }
1619
1620 block_init(bdrv_gluster_init);