block/nbd.c

   1 /*
   2  * QEMU Block driver for  NBD
   3  *
   4  * Copyright (C) 2008 Bull S.A.S.
   5  *     Author: Laurent Vivier <Laurent.Vivier@bull.net>
   6  *
   7  * Some parts:
   8  *    Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws>
   9  *
  10  * Permission is hereby granted, free of charge, to any person obtaining a copy
  11  * of this software and associated documentation files (the "Software"), to deal
  12  * in the Software without restriction, including without limitation the rights
  13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  14  * copies of the Software, and to permit persons to whom the Software is
  15  * furnished to do so, subject to the following conditions:
  16  *
  17  * The above copyright notice and this permission notice shall be included in
  18  * all copies or substantial portions of the Software.
  19  *
  20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  26  * THE SOFTWARE.
  27  */
  28
  29 #include "qemu-common.h"
  30 #include "block/nbd.h"
  31 #include "qemu/uri.h"
  32 #include "block/block_int.h"
  33 #include "qemu/module.h"
  34 #include "qemu/sockets.h"
  35
  36 #include <sys/types.h>
  37 #include <unistd.h>
  38
  39 #define EN_OPTSTR ":exportname="
  40
  41 /* #define DEBUG_NBD */
  42
  43 #if defined(DEBUG_NBD)
  44 #define logout(fmt, ...) \
  45                 fprintf(stderr, "nbd\t%-24s" fmt, __func__, ##__VA_ARGS__)
  46 #else
  47 #define logout(fmt, ...) ((void)0)
  48 #endif
  49
  50 #define MAX_NBD_REQUESTS        16
  51 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
  52 #define INDEX_TO_HANDLE(bs, index)  ((index)  ^ ((uint64_t)(intptr_t)bs))
  53
  54 typedef struct BDRVNBDState {
  55     int sock;
  56     uint32_t nbdflags;
  57     off_t size;
  58     size_t blocksize;
  59
  60     CoMutex send_mutex;
  61     CoMutex free_sema;
  62     Coroutine *send_coroutine;
  63     int in_flight;
  64
  65     Coroutine *recv_coroutine[MAX_NBD_REQUESTS];
  66     struct nbd_reply reply;
  67
  68     int is_unix;
  69     char *unix_path;
  70
  71     InetSocketAddress *inet_addr;
  72
  73     char *export_name; /* An NBD server may export several devices */
  74 } BDRVNBDState;
  75
  76 static int nbd_parse_uri(BDRVNBDState *s, const char *filename)
  77 {
  78     URI *uri;
  79     const char *p;
  80     QueryParams *qp = NULL;
  81     int ret = 0;
  82
  83     uri = uri_parse(filename);
  84     if (!uri) {
  85         return -EINVAL;
  86     }
  87
  88     /* transport */
  89     if (!strcmp(uri->scheme, "nbd")) {
  90         s->is_unix = false;
  91     } else if (!strcmp(uri->scheme, "nbd+tcp")) {
  92         s->is_unix = false;
  93     } else if (!strcmp(uri->scheme, "nbd+unix")) {
  94         s->is_unix = true;
  95     } else {
  96         ret = -EINVAL;
  97         goto out;
  98     }
  99
 100     p = uri->path ? uri->path : "/";
 101     p += strspn(p, "/");
 102     if (p[0]) {
 103         s->export_name = g_strdup(p);
 104     }
 105
 106     qp = query_params_parse(uri->query);
 107     if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) {
 108         ret = -EINVAL;
 109         goto out;
 110     }
 111
 112     if (s->is_unix) {
 113         /* nbd+unix:///export?socket=path */
 114         if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
 115             ret = -EINVAL;
 116             goto out;
 117         }
 118         s->unix_path = g_strdup(qp->p[0].value);
 119     } else {
 120         /* nbd[+tcp]://host:port/export */
 121         if (!uri->server) {
 122             ret = -EINVAL;
 123             goto out;
 124         }
 125         if (!uri->port) {
 126             uri->port = NBD_DEFAULT_PORT;
 127         }
 128
 129         s->inet_addr = g_new0(InetSocketAddress, 1);
 130         *s->inet_addr = (InetSocketAddress) {
 131             .host   = g_strdup(uri->server),
 132             .port   = g_strdup_printf("%d", uri->port),
 133         };
 134     }
 135
 136 out:
 137     if (qp) {
 138         query_params_free(qp);
 139     }
 140     uri_free(uri);
 141     return ret;
 142 }
 143
 144 static int nbd_config(BDRVNBDState *s, const char *filename)
 145 {
 146     char *file;
 147     char *export_name;
 148     const char *host_spec;
 149     const char *unixpath;
 150     int err = -EINVAL;
 151     Error *local_err = NULL;
 152
 153     if (strstr(filename, "://")) {
 154         return nbd_parse_uri(s, filename);
 155     }
 156
 157     file = g_strdup(filename);
 158
 159     export_name = strstr(file, EN_OPTSTR);
 160     if (export_name) {
 161         if (export_name[strlen(EN_OPTSTR)] == 0) {
 162             goto out;
 163         }
 164         export_name[0] = 0; /* truncate 'file' */
 165         export_name += strlen(EN_OPTSTR);
 166         s->export_name = g_strdup(export_name);
 167     }
 168
 169     /* extract the host_spec - fail if it's not nbd:... */
 170     if (!strstart(file, "nbd:", &host_spec)) {
 171         goto out;
 172     }
 173
 174     /* are we a UNIX or TCP socket? */
 175     if (strstart(host_spec, "unix:", &unixpath)) {
 176         s->is_unix = true;
 177         s->unix_path = g_strdup(unixpath);
 178     } else {
 179         s->is_unix = false;
 180         s->inet_addr = inet_parse(host_spec, &local_err);
 181         if (local_err != NULL) {
 182             qerror_report_err(local_err);
 183             error_free(local_err);
 184             goto out;
 185         }
 186     }
 187
 188     err = 0;
 189
 190 out:
 191     g_free(file);
 192     if (err != 0) {
 193         g_free(s->export_name);
 194         g_free(s->unix_path);
 195         qapi_free_InetSocketAddress(s->inet_addr);
 196     }
 197     return err;
 198 }
 199
 200 static void nbd_coroutine_start(BDRVNBDState *s, struct nbd_request *request)
 201 {
 202     int i;
 203
 204     /* Poor man semaphore.  The free_sema is locked when no other request
 205      * can be accepted, and unlocked after receiving one reply.  */
 206     if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
 207         qemu_co_mutex_lock(&s->free_sema);
 208         assert(s->in_flight < MAX_NBD_REQUESTS);
 209     }
 210     s->in_flight++;
 211
 212     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 213         if (s->recv_coroutine[i] == NULL) {
 214             s->recv_coroutine[i] = qemu_coroutine_self();
 215             break;
 216         }
 217     }
 218
 219     assert(i < MAX_NBD_REQUESTS);
 220     request->handle = INDEX_TO_HANDLE(s, i);
 221 }
 222
 223 static int nbd_have_request(void *opaque)
 224 {
 225     BDRVNBDState *s = opaque;
 226
 227     return s->in_flight > 0;
 228 }
 229
 230 static void nbd_reply_ready(void *opaque)
 231 {
 232     BDRVNBDState *s = opaque;
 233     uint64_t i;
 234     int ret;
 235
 236     if (s->reply.handle == 0) {
 237         /* No reply already in flight.  Fetch a header.  It is possible
 238          * that another thread has done the same thing in parallel, so
 239          * the socket is not readable anymore.
 240          */
 241         ret = nbd_receive_reply(s->sock, &s->reply);
 242         if (ret == -EAGAIN) {
 243             return;
 244         }
 245         if (ret < 0) {
 246             s->reply.handle = 0;
 247             goto fail;
 248         }
 249     }
 250
 251     /* There's no need for a mutex on the receive side, because the
 252      * handler acts as a synchronization point and ensures that only
 253      * one coroutine is called until the reply finishes.  */
 254     i = HANDLE_TO_INDEX(s, s->reply.handle);
 255     if (i >= MAX_NBD_REQUESTS) {
 256         goto fail;
 257     }
 258
 259     if (s->recv_coroutine[i]) {
 260         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 261         return;
 262     }
 263
 264 fail:
 265     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 266         if (s->recv_coroutine[i]) {
 267             qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 268         }
 269     }
 270 }
 271
 272 static void nbd_restart_write(void *opaque)
 273 {
 274     BDRVNBDState *s = opaque;
 275     qemu_coroutine_enter(s->send_coroutine, NULL);
 276 }
 277
 278 static int nbd_co_send_request(BDRVNBDState *s, struct nbd_request *request,
 279                                QEMUIOVector *qiov, int offset)
 280 {
 281     int rc, ret;
 282
 283     qemu_co_mutex_lock(&s->send_mutex);
 284     s->send_coroutine = qemu_coroutine_self();
 285     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, nbd_restart_write,
 286                             nbd_have_request, s);
 287     rc = nbd_send_request(s->sock, request);
 288     if (rc >= 0 && qiov) {
 289         ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov,
 290                             offset, request->len);
 291         if (ret != request->len) {
 292             return -EIO;
 293         }
 294     }
 295     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 296                             nbd_have_request, s);
 297     s->send_coroutine = NULL;
 298     qemu_co_mutex_unlock(&s->send_mutex);
 299     return rc;
 300 }
 301
 302 static void nbd_co_receive_reply(BDRVNBDState *s, struct nbd_request *request,
 303                                  struct nbd_reply *reply,
 304                                  QEMUIOVector *qiov, int offset)
 305 {
 306     int ret;
 307
 308     /* Wait until we're woken up by the read handler.  TODO: perhaps
 309      * peek at the next reply and avoid yielding if it's ours?  */
 310     qemu_coroutine_yield();
 311     *reply = s->reply;
 312     if (reply->handle != request->handle) {
 313         reply->error = EIO;
 314     } else {
 315         if (qiov && reply->error == 0) {
 316             ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov,
 317                                 offset, request->len);
 318             if (ret != request->len) {
 319                 reply->error = EIO;
 320             }
 321         }
 322
 323         /* Tell the read handler to read another header.  */
 324         s->reply.handle = 0;
 325     }
 326 }
 327
 328 static void nbd_coroutine_end(BDRVNBDState *s, struct nbd_request *request)
 329 {
 330     int i = HANDLE_TO_INDEX(s, request->handle);
 331     s->recv_coroutine[i] = NULL;
 332     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 333         qemu_co_mutex_unlock(&s->free_sema);
 334     }
 335 }
 336
 337 static int nbd_establish_connection(BlockDriverState *bs)
 338 {
 339     BDRVNBDState *s = bs->opaque;
 340     int sock;
 341     int ret;
 342     off_t size;
 343     size_t blocksize;
 344
 345     if (s->is_unix) {
 346         sock = unix_socket_outgoing(s->unix_path);
 347     } else {
 348         QemuOpts *opts = qemu_opts_create_nofail(&socket_optslist);
 349
 350         qemu_opt_set(opts, "host", s->inet_addr->host);
 351         qemu_opt_set(opts, "port", s->inet_addr->port);
 352         if (s->inet_addr->has_to) {
 353             qemu_opt_set_number(opts, "to", s->inet_addr->to);
 354         }
 355         if (s->inet_addr->has_ipv4) {
 356             qemu_opt_set_number(opts, "ipv4", s->inet_addr->ipv4);
 357         }
 358         if (s->inet_addr->has_ipv6) {
 359             qemu_opt_set_number(opts, "ipv6", s->inet_addr->ipv6);
 360         }
 361
 362         sock = tcp_socket_outgoing_opts(opts);
 363         qemu_opts_del(opts);
 364     }
 365
 366     /* Failed to establish connection */
 367     if (sock < 0) {
 368         logout("Failed to establish connection to NBD server\n");
 369         return -errno;
 370     }
 371
 372     /* NBD handshake */
 373     ret = nbd_receive_negotiate(sock, s->export_name, &s->nbdflags, &size,
 374                                 &blocksize);
 375     if (ret < 0) {
 376         logout("Failed to negotiate with the NBD server\n");
 377         closesocket(sock);
 378         return ret;
 379     }
 380
 381     /* Now that we're connected, set the socket to be non-blocking and
 382      * kick the reply mechanism.  */
 383     socket_set_nonblock(sock);
 384     qemu_aio_set_fd_handler(sock, nbd_reply_ready, NULL,
 385                             nbd_have_request, s);
 386
 387     s->sock = sock;
 388     s->size = size;
 389     s->blocksize = blocksize;
 390
 391     logout("Established connection with NBD server\n");
 392     return 0;
 393 }
 394
 395 static void nbd_teardown_connection(BlockDriverState *bs)
 396 {
 397     BDRVNBDState *s = bs->opaque;
 398     struct nbd_request request;
 399
 400     request.type = NBD_CMD_DISC;
 401     request.from = 0;
 402     request.len = 0;
 403     nbd_send_request(s->sock, &request);
 404
 405     qemu_aio_set_fd_handler(s->sock, NULL, NULL, NULL, NULL);
 406     closesocket(s->sock);
 407 }
 408
 409 static int nbd_open(BlockDriverState *bs, const char* filename,
 410                     QDict *options, int flags)
 411 {
 412     BDRVNBDState *s = bs->opaque;
 413     int result;
 414
 415     qemu_co_mutex_init(&s->send_mutex);
 416     qemu_co_mutex_init(&s->free_sema);
 417
 418     /* Pop the config into our state object. Exit if invalid. */
 419     result = nbd_config(s, filename);
 420     if (result != 0) {
 421         return result;
 422     }
 423
 424     /* establish TCP connection, return error if it fails
 425      * TODO: Configurable retry-until-timeout behaviour.
 426      */
 427     result = nbd_establish_connection(bs);
 428
 429     return result;
 430 }
 431
 432 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 433                           int nb_sectors, QEMUIOVector *qiov,
 434                           int offset)
 435 {
 436     BDRVNBDState *s = bs->opaque;
 437     struct nbd_request request;
 438     struct nbd_reply reply;
 439     ssize_t ret;
 440
 441     request.type = NBD_CMD_READ;
 442     request.from = sector_num * 512;
 443     request.len = nb_sectors * 512;
 444
 445     nbd_coroutine_start(s, &request);
 446     ret = nbd_co_send_request(s, &request, NULL, 0);
 447     if (ret < 0) {
 448         reply.error = -ret;
 449     } else {
 450         nbd_co_receive_reply(s, &request, &reply, qiov, offset);
 451     }
 452     nbd_coroutine_end(s, &request);
 453     return -reply.error;
 454
 455 }
 456
 457 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 458                            int nb_sectors, QEMUIOVector *qiov,
 459                            int offset)
 460 {
 461     BDRVNBDState *s = bs->opaque;
 462     struct nbd_request request;
 463     struct nbd_reply reply;
 464     ssize_t ret;
 465
 466     request.type = NBD_CMD_WRITE;
 467     if (!bdrv_enable_write_cache(bs) && (s->nbdflags & NBD_FLAG_SEND_FUA)) {
 468         request.type |= NBD_CMD_FLAG_FUA;
 469     }
 470
 471     request.from = sector_num * 512;
 472     request.len = nb_sectors * 512;
 473
 474     nbd_coroutine_start(s, &request);
 475     ret = nbd_co_send_request(s, &request, qiov, offset);
 476     if (ret < 0) {
 477         reply.error = -ret;
 478     } else {
 479         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 480     }
 481     nbd_coroutine_end(s, &request);
 482     return -reply.error;
 483 }
 484
 485 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 486  * remain aligned to 4K. */
 487 #define NBD_MAX_SECTORS 2040
 488
 489 static int nbd_co_readv(BlockDriverState *bs, int64_t sector_num,
 490                         int nb_sectors, QEMUIOVector *qiov)
 491 {
 492     int offset = 0;
 493     int ret;
 494     while (nb_sectors > NBD_MAX_SECTORS) {
 495         ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 496         if (ret < 0) {
 497             return ret;
 498         }
 499         offset += NBD_MAX_SECTORS * 512;
 500         sector_num += NBD_MAX_SECTORS;
 501         nb_sectors -= NBD_MAX_SECTORS;
 502     }
 503     return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 504 }
 505
 506 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
 507                          int nb_sectors, QEMUIOVector *qiov)
 508 {
 509     int offset = 0;
 510     int ret;
 511     while (nb_sectors > NBD_MAX_SECTORS) {
 512         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 513         if (ret < 0) {
 514             return ret;
 515         }
 516         offset += NBD_MAX_SECTORS * 512;
 517         sector_num += NBD_MAX_SECTORS;
 518         nb_sectors -= NBD_MAX_SECTORS;
 519     }
 520     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 521 }
 522
 523 static int nbd_co_flush(BlockDriverState *bs)
 524 {
 525     BDRVNBDState *s = bs->opaque;
 526     struct nbd_request request;
 527     struct nbd_reply reply;
 528     ssize_t ret;
 529
 530     if (!(s->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 531         return 0;
 532     }
 533
 534     request.type = NBD_CMD_FLUSH;
 535     if (s->nbdflags & NBD_FLAG_SEND_FUA) {
 536         request.type |= NBD_CMD_FLAG_FUA;
 537     }
 538
 539     request.from = 0;
 540     request.len = 0;
 541
 542     nbd_coroutine_start(s, &request);
 543     ret = nbd_co_send_request(s, &request, NULL, 0);
 544     if (ret < 0) {
 545         reply.error = -ret;
 546     } else {
 547         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 548     }
 549     nbd_coroutine_end(s, &request);
 550     return -reply.error;
 551 }
 552
 553 static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num,
 554                           int nb_sectors)
 555 {
 556     BDRVNBDState *s = bs->opaque;
 557     struct nbd_request request;
 558     struct nbd_reply reply;
 559     ssize_t ret;
 560
 561     if (!(s->nbdflags & NBD_FLAG_SEND_TRIM)) {
 562         return 0;
 563     }
 564     request.type = NBD_CMD_TRIM;
 565     request.from = sector_num * 512;;
 566     request.len = nb_sectors * 512;
 567
 568     nbd_coroutine_start(s, &request);
 569     ret = nbd_co_send_request(s, &request, NULL, 0);
 570     if (ret < 0) {
 571         reply.error = -ret;
 572     } else {
 573         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 574     }
 575     nbd_coroutine_end(s, &request);
 576     return -reply.error;
 577 }
 578
 579 static void nbd_close(BlockDriverState *bs)
 580 {
 581     BDRVNBDState *s = bs->opaque;
 582     g_free(s->export_name);
 583     g_free(s->unix_path);
 584     qapi_free_InetSocketAddress(s->inet_addr);
 585
 586     nbd_teardown_connection(bs);
 587 }
 588
 589 static int64_t nbd_getlength(BlockDriverState *bs)
 590 {
 591     BDRVNBDState *s = bs->opaque;
 592
 593     return s->size;
 594 }
 595
 596 static BlockDriver bdrv_nbd = {
 597     .format_name         = "nbd",
 598     .protocol_name       = "nbd",
 599     .instance_size       = sizeof(BDRVNBDState),
 600     .bdrv_file_open      = nbd_open,
 601     .bdrv_co_readv       = nbd_co_readv,
 602     .bdrv_co_writev      = nbd_co_writev,
 603     .bdrv_close          = nbd_close,
 604     .bdrv_co_flush_to_os = nbd_co_flush,
 605     .bdrv_co_discard     = nbd_co_discard,
 606     .bdrv_getlength      = nbd_getlength,
 607 };
 608
 609 static BlockDriver bdrv_nbd_tcp = {
 610     .format_name         = "nbd",
 611     .protocol_name       = "nbd+tcp",
 612     .instance_size       = sizeof(BDRVNBDState),
 613     .bdrv_file_open      = nbd_open,
 614     .bdrv_co_readv       = nbd_co_readv,
 615     .bdrv_co_writev      = nbd_co_writev,
 616     .bdrv_close          = nbd_close,
 617     .bdrv_co_flush_to_os = nbd_co_flush,
 618     .bdrv_co_discard     = nbd_co_discard,
 619     .bdrv_getlength      = nbd_getlength,
 620 };
 621
 622 static BlockDriver bdrv_nbd_unix = {
 623     .format_name         = "nbd",
 624     .protocol_name       = "nbd+unix",
 625     .instance_size       = sizeof(BDRVNBDState),
 626     .bdrv_file_open      = nbd_open,
 627     .bdrv_co_readv       = nbd_co_readv,
 628     .bdrv_co_writev      = nbd_co_writev,
 629     .bdrv_close          = nbd_close,
 630     .bdrv_co_flush_to_os = nbd_co_flush,
 631     .bdrv_co_discard     = nbd_co_discard,
 632     .bdrv_getlength      = nbd_getlength,
 633 };
 634
 635 static void bdrv_nbd_init(void)
 636 {
 637     bdrv_register(&bdrv_nbd);
 638     bdrv_register(&bdrv_nbd_tcp);
 639     bdrv_register(&bdrv_nbd_unix);
 640 }
 641
 642 block_init(bdrv_nbd_init);