block/gluster.c

   1 /*
   2  * GlusterFS backend for QEMU
   3  *
   4  * Copyright (C) 2012 Bharata B Rao <[email protected]>
   5  *
   6  * Pipe handling mechanism in AIO implementation is derived from
   7  * block/rbd.c. Hence,
   8  *
   9  * Copyright (C) 2010-2011 Christian Brunner <[email protected]>,
  10  *                         Josh Durgin <[email protected]>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2.  See
  13  * the COPYING file in the top-level directory.
  14  *
  15  * Contributions after 2012-01-13 are licensed under the terms of the
  16  * GNU GPL, version 2 or (at your option) any later version.
  17  */
  18 #include <glusterfs/api/glfs.h>
  19 #include "block/block_int.h"
  20 #include "qemu/sockets.h"
  21 #include "qemu/uri.h"
  22
  23 typedef struct GlusterAIOCB {
  24     BlockDriverAIOCB common;
  25     int64_t size;
  26     int ret;
  27     bool *finished;
  28     QEMUBH *bh;
  29 } GlusterAIOCB;
  30
  31 typedef struct BDRVGlusterState {
  32     struct glfs *glfs;
  33     int fds[2];
  34     struct glfs_fd *fd;
  35     int qemu_aio_count;
  36     int event_reader_pos;
  37     GlusterAIOCB *event_acb;
  38 } BDRVGlusterState;
  39
  40 #define GLUSTER_FD_READ  0
  41 #define GLUSTER_FD_WRITE 1
  42
  43 typedef struct GlusterConf {
  44     char *server;
  45     int port;
  46     char *volname;
  47     char *image;
  48     char *transport;
  49 } GlusterConf;
  50
  51 static void qemu_gluster_gconf_free(GlusterConf *gconf)
  52 {
  53     g_free(gconf->server);
  54     g_free(gconf->volname);
  55     g_free(gconf->image);
  56     g_free(gconf->transport);
  57     g_free(gconf);
  58 }
  59
  60 static int parse_volume_options(GlusterConf *gconf, char *path)
  61 {
  62     char *p, *q;
  63
  64     if (!path) {
  65         return -EINVAL;
  66     }
  67
  68     /* volume */
  69     p = q = path + strspn(path, "/");
  70     p += strcspn(p, "/");
  71     if (*p == '\0') {
  72         return -EINVAL;
  73     }
  74     gconf->volname = g_strndup(q, p - q);
  75
  76     /* image */
  77     p += strspn(p, "/");
  78     if (*p == '\0') {
  79         return -EINVAL;
  80     }
  81     gconf->image = g_strdup(p);
  82     return 0;
  83 }
  84
  85 /*
  86  * file=gluster[+transport]://[server[:port]]/volname/image[?socket=...]
  87  *
  88  * 'gluster' is the protocol.
  89  *
  90  * 'transport' specifies the transport type used to connect to gluster
  91  * management daemon (glusterd). Valid transport types are
  92  * tcp, unix and rdma. If a transport type isn't specified, then tcp
  93  * type is assumed.
  94  *
  95  * 'server' specifies the server where the volume file specification for
  96  * the given volume resides. This can be either hostname, ipv4 address
  97  * or ipv6 address. ipv6 address needs to be within square brackets [ ].
  98  * If transport type is 'unix', then 'server' field should not be specifed.
  99  * The 'socket' field needs to be populated with the path to unix domain
 100  * socket.
 101  *
 102  * 'port' is the port number on which glusterd is listening. This is optional
 103  * and if not specified, QEMU will send 0 which will make gluster to use the
 104  * default port. If the transport type is unix, then 'port' should not be
 105  * specified.
 106  *
 107  * 'volname' is the name of the gluster volume which contains the VM image.
 108  *
 109  * 'image' is the path to the actual VM image that resides on gluster volume.
 110  *
 111  * Examples:
 112  *
 113  * file=gluster://1.2.3.4/testvol/a.img
 114  * file=gluster+tcp://1.2.3.4/testvol/a.img
 115  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
 116  * file=gluster+tcp://[1:2:3:4:5:6:7:8]/testvol/dir/a.img
 117  * file=gluster+tcp://[1:2:3:4:5:6:7:8]:24007/testvol/dir/a.img
 118  * file=gluster+tcp://server.domain.com:24007/testvol/dir/a.img
 119  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
 120  * file=gluster+rdma://1.2.3.4:24007/testvol/a.img
 121  */
 122 static int qemu_gluster_parseuri(GlusterConf *gconf, const char *filename)
 123 {
 124     URI *uri;
 125     QueryParams *qp = NULL;
 126     bool is_unix = false;
 127     int ret = 0;
 128
 129     uri = uri_parse(filename);
 130     if (!uri) {
 131         return -EINVAL;
 132     }
 133
 134     /* transport */
 135     if (!strcmp(uri->scheme, "gluster")) {
 136         gconf->transport = g_strdup("tcp");
 137     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
 138         gconf->transport = g_strdup("tcp");
 139     } else if (!strcmp(uri->scheme, "gluster+unix")) {
 140         gconf->transport = g_strdup("unix");
 141         is_unix = true;
 142     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
 143         gconf->transport = g_strdup("rdma");
 144     } else {
 145         ret = -EINVAL;
 146         goto out;
 147     }
 148
 149     ret = parse_volume_options(gconf, uri->path);
 150     if (ret < 0) {
 151         goto out;
 152     }
 153
 154     qp = query_params_parse(uri->query);
 155     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
 156         ret = -EINVAL;
 157         goto out;
 158     }
 159
 160     if (is_unix) {
 161         if (uri->server || uri->port) {
 162             ret = -EINVAL;
 163             goto out;
 164         }
 165         if (strcmp(qp->p[0].name, "socket")) {
 166             ret = -EINVAL;
 167             goto out;
 168         }
 169         gconf->server = g_strdup(qp->p[0].value);
 170     } else {
 171         gconf->server = g_strdup(uri->server);
 172         gconf->port = uri->port;
 173     }
 174
 175 out:
 176     if (qp) {
 177         query_params_free(qp);
 178     }
 179     uri_free(uri);
 180     return ret;
 181 }
 182
 183 static struct glfs *qemu_gluster_init(GlusterConf *gconf, const char *filename)
 184 {
 185     struct glfs *glfs = NULL;
 186     int ret;
 187     int old_errno;
 188
 189     ret = qemu_gluster_parseuri(gconf, filename);
 190     if (ret < 0) {
 191         error_report("Usage: file=gluster[+transport]://[server[:port]]/"
 192             "volname/image[?socket=...]");
 193         errno = -ret;
 194         goto out;
 195     }
 196
 197     glfs = glfs_new(gconf->volname);
 198     if (!glfs) {
 199         goto out;
 200     }
 201
 202     ret = glfs_set_volfile_server(glfs, gconf->transport, gconf->server,
 203             gconf->port);
 204     if (ret < 0) {
 205         goto out;
 206     }
 207
 208     /*
 209      * TODO: Use GF_LOG_ERROR instead of hard code value of 4 here when
 210      * GlusterFS makes GF_LOG_* macros available to libgfapi users.
 211      */
 212     ret = glfs_set_logging(glfs, "-", 4);
 213     if (ret < 0) {
 214         goto out;
 215     }
 216
 217     ret = glfs_init(glfs);
 218     if (ret) {
 219         error_report("Gluster connection failed for server=%s port=%d "
 220              "volume=%s image=%s transport=%s\n", gconf->server, gconf->port,
 221              gconf->volname, gconf->image, gconf->transport);
 222         goto out;
 223     }
 224     return glfs;
 225
 226 out:
 227     if (glfs) {
 228         old_errno = errno;
 229         glfs_fini(glfs);
 230         errno = old_errno;
 231     }
 232     return NULL;
 233 }
 234
 235 static void qemu_gluster_complete_aio(GlusterAIOCB *acb, BDRVGlusterState *s)
 236 {
 237     int ret;
 238     bool *finished = acb->finished;
 239     BlockDriverCompletionFunc *cb = acb->common.cb;
 240     void *opaque = acb->common.opaque;
 241
 242     if (!acb->ret || acb->ret == acb->size) {
 243         ret = 0; /* Success */
 244     } else if (acb->ret < 0) {
 245         ret = acb->ret; /* Read/Write failed */
 246     } else {
 247         ret = -EIO; /* Partial read/write - fail it */
 248     }
 249
 250     s->qemu_aio_count--;
 251     qemu_aio_release(acb);
 252     cb(opaque, ret);
 253     if (finished) {
 254         *finished = true;
 255     }
 256 }
 257
 258 static void qemu_gluster_aio_event_reader(void *opaque)
 259 {
 260     BDRVGlusterState *s = opaque;
 261     ssize_t ret;
 262
 263     do {
 264         char *p = (char *)&s->event_acb;
 265
 266         ret = read(s->fds[GLUSTER_FD_READ], p + s->event_reader_pos,
 267                    sizeof(s->event_acb) - s->event_reader_pos);
 268         if (ret > 0) {
 269             s->event_reader_pos += ret;
 270             if (s->event_reader_pos == sizeof(s->event_acb)) {
 271                 s->event_reader_pos = 0;
 272                 qemu_gluster_complete_aio(s->event_acb, s);
 273             }
 274         }
 275     } while (ret < 0 && errno == EINTR);
 276 }
 277
 278 static int qemu_gluster_aio_flush_cb(void *opaque)
 279 {
 280     BDRVGlusterState *s = opaque;
 281
 282     return (s->qemu_aio_count > 0);
 283 }
 284
 285 static int qemu_gluster_open(BlockDriverState *bs, const char *filename,
 286     int bdrv_flags)
 287 {
 288     BDRVGlusterState *s = bs->opaque;
 289     int open_flags = O_BINARY;
 290     int ret = 0;
 291     GlusterConf *gconf = g_malloc0(sizeof(GlusterConf));
 292
 293     s->glfs = qemu_gluster_init(gconf, filename);
 294     if (!s->glfs) {
 295         ret = -errno;
 296         goto out;
 297     }
 298
 299     if (bdrv_flags & BDRV_O_RDWR) {
 300         open_flags |= O_RDWR;
 301     } else {
 302         open_flags |= O_RDONLY;
 303     }
 304
 305     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 306         open_flags |= O_DIRECT;
 307     }
 308
 309     s->fd = glfs_open(s->glfs, gconf->image, open_flags);
 310     if (!s->fd) {
 311         ret = -errno;
 312         goto out;
 313     }
 314
 315     ret = qemu_pipe(s->fds);
 316     if (ret < 0) {
 317         ret = -errno;
 318         goto out;
 319     }
 320     fcntl(s->fds[GLUSTER_FD_READ], F_SETFL, O_NONBLOCK);
 321     qemu_aio_set_fd_handler(s->fds[GLUSTER_FD_READ],
 322         qemu_gluster_aio_event_reader, NULL, qemu_gluster_aio_flush_cb, s);
 323
 324 out:
 325     qemu_gluster_gconf_free(gconf);
 326     if (!ret) {
 327         return ret;
 328     }
 329     if (s->fd) {
 330         glfs_close(s->fd);
 331     }
 332     if (s->glfs) {
 333         glfs_fini(s->glfs);
 334     }
 335     return ret;
 336 }
 337
 338 static int qemu_gluster_create(const char *filename,
 339         QEMUOptionParameter *options)
 340 {
 341     struct glfs *glfs;
 342     struct glfs_fd *fd;
 343     int ret = 0;
 344     int64_t total_size = 0;
 345     GlusterConf *gconf = g_malloc0(sizeof(GlusterConf));
 346
 347     glfs = qemu_gluster_init(gconf, filename);
 348     if (!glfs) {
 349         ret = -errno;
 350         goto out;
 351     }
 352
 353     while (options && options->name) {
 354         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 355             total_size = options->value.n / BDRV_SECTOR_SIZE;
 356         }
 357         options++;
 358     }
 359
 360     fd = glfs_creat(glfs, gconf->image,
 361         O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
 362     if (!fd) {
 363         ret = -errno;
 364     } else {
 365         if (glfs_ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
 366             ret = -errno;
 367         }
 368         if (glfs_close(fd) != 0) {
 369             ret = -errno;
 370         }
 371     }
 372 out:
 373     qemu_gluster_gconf_free(gconf);
 374     if (glfs) {
 375         glfs_fini(glfs);
 376     }
 377     return ret;
 378 }
 379
 380 static void qemu_gluster_aio_cancel(BlockDriverAIOCB *blockacb)
 381 {
 382     GlusterAIOCB *acb = (GlusterAIOCB *)blockacb;
 383     bool finished = false;
 384
 385     acb->finished = &finished;
 386     while (!finished) {
 387         qemu_aio_wait();
 388     }
 389 }
 390
 391 static const AIOCBInfo gluster_aiocb_info = {
 392     .aiocb_size = sizeof(GlusterAIOCB),
 393     .cancel = qemu_gluster_aio_cancel,
 394 };
 395
 396 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
 397 {
 398     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
 399     BlockDriverState *bs = acb->common.bs;
 400     BDRVGlusterState *s = bs->opaque;
 401     int retval;
 402
 403     acb->ret = ret;
 404     retval = qemu_write_full(s->fds[GLUSTER_FD_WRITE], &acb, sizeof(acb));
 405     if (retval != sizeof(acb)) {
 406         /*
 407          * Gluster AIO callback thread failed to notify the waiting
 408          * QEMU thread about IO completion.
 409          *
 410          * Complete this IO request and make the disk inaccessible for
 411          * subsequent reads and writes.
 412          */
 413         error_report("Gluster failed to notify QEMU about IO completion");
 414
 415         qemu_mutex_lock_iothread(); /* We are in gluster thread context */
 416         acb->common.cb(acb->common.opaque, -EIO);
 417         qemu_aio_release(acb);
 418         s->qemu_aio_count--;
 419         close(s->fds[GLUSTER_FD_READ]);
 420         close(s->fds[GLUSTER_FD_WRITE]);
 421         qemu_aio_set_fd_handler(s->fds[GLUSTER_FD_READ], NULL, NULL, NULL,
 422             NULL);
 423         bs->drv = NULL; /* Make the disk inaccessible */
 424         qemu_mutex_unlock_iothread();
 425     }
 426 }
 427
 428 static BlockDriverAIOCB *qemu_gluster_aio_rw(BlockDriverState *bs,
 429         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 430         BlockDriverCompletionFunc *cb, void *opaque, int write)
 431 {
 432     int ret;
 433     GlusterAIOCB *acb;
 434     BDRVGlusterState *s = bs->opaque;
 435     size_t size;
 436     off_t offset;
 437
 438     offset = sector_num * BDRV_SECTOR_SIZE;
 439     size = nb_sectors * BDRV_SECTOR_SIZE;
 440     s->qemu_aio_count++;
 441
 442     acb = qemu_aio_get(&gluster_aiocb_info, bs, cb, opaque);
 443     acb->size = size;
 444     acb->ret = 0;
 445     acb->finished = NULL;
 446
 447     if (write) {
 448         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
 449             &gluster_finish_aiocb, acb);
 450     } else {
 451         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
 452             &gluster_finish_aiocb, acb);
 453     }
 454
 455     if (ret < 0) {
 456         goto out;
 457     }
 458     return &acb->common;
 459
 460 out:
 461     s->qemu_aio_count--;
 462     qemu_aio_release(acb);
 463     return NULL;
 464 }
 465
 466 static BlockDriverAIOCB *qemu_gluster_aio_readv(BlockDriverState *bs,
 467         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 468         BlockDriverCompletionFunc *cb, void *opaque)
 469 {
 470     return qemu_gluster_aio_rw(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
 471 }
 472
 473 static BlockDriverAIOCB *qemu_gluster_aio_writev(BlockDriverState *bs,
 474         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 475         BlockDriverCompletionFunc *cb, void *opaque)
 476 {
 477     return qemu_gluster_aio_rw(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
 478 }
 479
 480 static BlockDriverAIOCB *qemu_gluster_aio_flush(BlockDriverState *bs,
 481         BlockDriverCompletionFunc *cb, void *opaque)
 482 {
 483     int ret;
 484     GlusterAIOCB *acb;
 485     BDRVGlusterState *s = bs->opaque;
 486
 487     acb = qemu_aio_get(&gluster_aiocb_info, bs, cb, opaque);
 488     acb->size = 0;
 489     acb->ret = 0;
 490     acb->finished = NULL;
 491     s->qemu_aio_count++;
 492
 493     ret = glfs_fsync_async(s->fd, &gluster_finish_aiocb, acb);
 494     if (ret < 0) {
 495         goto out;
 496     }
 497     return &acb->common;
 498
 499 out:
 500     s->qemu_aio_count--;
 501     qemu_aio_release(acb);
 502     return NULL;
 503 }
 504
 505 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
 506 {
 507     BDRVGlusterState *s = bs->opaque;
 508     int64_t ret;
 509
 510     ret = glfs_lseek(s->fd, 0, SEEK_END);
 511     if (ret < 0) {
 512         return -errno;
 513     } else {
 514         return ret;
 515     }
 516 }
 517
 518 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
 519 {
 520     BDRVGlusterState *s = bs->opaque;
 521     struct stat st;
 522     int ret;
 523
 524     ret = glfs_fstat(s->fd, &st);
 525     if (ret < 0) {
 526         return -errno;
 527     } else {
 528         return st.st_blocks * 512;
 529     }
 530 }
 531
 532 static void qemu_gluster_close(BlockDriverState *bs)
 533 {
 534     BDRVGlusterState *s = bs->opaque;
 535
 536     close(s->fds[GLUSTER_FD_READ]);
 537     close(s->fds[GLUSTER_FD_WRITE]);
 538     qemu_aio_set_fd_handler(s->fds[GLUSTER_FD_READ], NULL, NULL, NULL, NULL);
 539
 540     if (s->fd) {
 541         glfs_close(s->fd);
 542         s->fd = NULL;
 543     }
 544     glfs_fini(s->glfs);
 545 }
 546
 547 static QEMUOptionParameter qemu_gluster_create_options[] = {
 548     {
 549         .name = BLOCK_OPT_SIZE,
 550         .type = OPT_SIZE,
 551         .help = "Virtual disk size"
 552     },
 553     { NULL }
 554 };
 555
 556 static BlockDriver bdrv_gluster = {
 557     .format_name                  = "gluster",
 558     .protocol_name                = "gluster",
 559     .instance_size                = sizeof(BDRVGlusterState),
 560     .bdrv_file_open               = qemu_gluster_open,
 561     .bdrv_close                   = qemu_gluster_close,
 562     .bdrv_create                  = qemu_gluster_create,
 563     .bdrv_getlength               = qemu_gluster_getlength,
 564     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
 565     .bdrv_aio_readv               = qemu_gluster_aio_readv,
 566     .bdrv_aio_writev              = qemu_gluster_aio_writev,
 567     .bdrv_aio_flush               = qemu_gluster_aio_flush,
 568     .create_options               = qemu_gluster_create_options,
 569 };
 570
 571 static BlockDriver bdrv_gluster_tcp = {
 572     .format_name                  = "gluster",
 573     .protocol_name                = "gluster+tcp",
 574     .instance_size                = sizeof(BDRVGlusterState),
 575     .bdrv_file_open               = qemu_gluster_open,
 576     .bdrv_close                   = qemu_gluster_close,
 577     .bdrv_create                  = qemu_gluster_create,
 578     .bdrv_getlength               = qemu_gluster_getlength,
 579     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
 580     .bdrv_aio_readv               = qemu_gluster_aio_readv,
 581     .bdrv_aio_writev              = qemu_gluster_aio_writev,
 582     .bdrv_aio_flush               = qemu_gluster_aio_flush,
 583     .create_options               = qemu_gluster_create_options,
 584 };
 585
 586 static BlockDriver bdrv_gluster_unix = {
 587     .format_name                  = "gluster",
 588     .protocol_name                = "gluster+unix",
 589     .instance_size                = sizeof(BDRVGlusterState),
 590     .bdrv_file_open               = qemu_gluster_open,
 591     .bdrv_close                   = qemu_gluster_close,
 592     .bdrv_create                  = qemu_gluster_create,
 593     .bdrv_getlength               = qemu_gluster_getlength,
 594     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
 595     .bdrv_aio_readv               = qemu_gluster_aio_readv,
 596     .bdrv_aio_writev              = qemu_gluster_aio_writev,
 597     .bdrv_aio_flush               = qemu_gluster_aio_flush,
 598     .create_options               = qemu_gluster_create_options,
 599 };
 600
 601 static BlockDriver bdrv_gluster_rdma = {
 602     .format_name                  = "gluster",
 603     .protocol_name                = "gluster+rdma",
 604     .instance_size                = sizeof(BDRVGlusterState),
 605     .bdrv_file_open               = qemu_gluster_open,
 606     .bdrv_close                   = qemu_gluster_close,
 607     .bdrv_create                  = qemu_gluster_create,
 608     .bdrv_getlength               = qemu_gluster_getlength,
 609     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
 610     .bdrv_aio_readv               = qemu_gluster_aio_readv,
 611     .bdrv_aio_writev              = qemu_gluster_aio_writev,
 612     .bdrv_aio_flush               = qemu_gluster_aio_flush,
 613     .create_options               = qemu_gluster_create_options,
 614 };
 615
 616 static void bdrv_gluster_init(void)
 617 {
 618     bdrv_register(&bdrv_gluster_rdma);
 619     bdrv_register(&bdrv_gluster_unix);
 620     bdrv_register(&bdrv_gluster_tcp);
 621     bdrv_register(&bdrv_gluster);
 622 }
 623
 624 block_init(bdrv_gluster_init);