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 *host_spec;
  70     char *export_name; /* An NBD server may export several devices */
  71 } BDRVNBDState;
  72
  73 static int nbd_parse_uri(BDRVNBDState *s, const char *filename)
  74 {
  75     URI *uri;
  76     const char *p;
  77     QueryParams *qp = NULL;
  78     int ret = 0;
  79
  80     uri = uri_parse(filename);
  81     if (!uri) {
  82         return -EINVAL;
  83     }
  84
  85     /* transport */
  86     if (!strcmp(uri->scheme, "nbd")) {
  87         s->is_unix = false;
  88     } else if (!strcmp(uri->scheme, "nbd+tcp")) {
  89         s->is_unix = false;
  90     } else if (!strcmp(uri->scheme, "nbd+unix")) {
  91         s->is_unix = true;
  92     } else {
  93         ret = -EINVAL;
  94         goto out;
  95     }
  96
  97     p = uri->path ? uri->path : "/";
  98     p += strspn(p, "/");
  99     if (p[0]) {
 100         s->export_name = g_strdup(p);
 101     }
 102
 103     qp = query_params_parse(uri->query);
 104     if (qp->n > 1 || (s->is_unix && !qp->n) || (!s->is_unix && qp->n)) {
 105         ret = -EINVAL;
 106         goto out;
 107     }
 108
 109     if (s->is_unix) {
 110         /* nbd+unix:///export?socket=path */
 111         if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) {
 112             ret = -EINVAL;
 113             goto out;
 114         }
 115         s->host_spec = g_strdup(qp->p[0].value);
 116     } else {
 117         /* nbd[+tcp]://host:port/export */
 118         if (!uri->server) {
 119             ret = -EINVAL;
 120             goto out;
 121         }
 122         if (!uri->port) {
 123             uri->port = NBD_DEFAULT_PORT;
 124         }
 125         s->host_spec = g_strdup_printf("%s:%d", uri->server, uri->port);
 126     }
 127
 128 out:
 129     if (qp) {
 130         query_params_free(qp);
 131     }
 132     uri_free(uri);
 133     return ret;
 134 }
 135
 136 static int nbd_config(BDRVNBDState *s, const char *filename)
 137 {
 138     char *file;
 139     char *export_name;
 140     const char *host_spec;
 141     const char *unixpath;
 142     int err = -EINVAL;
 143
 144     if (strstr(filename, "://")) {
 145         return nbd_parse_uri(s, filename);
 146     }
 147
 148     file = g_strdup(filename);
 149
 150     export_name = strstr(file, EN_OPTSTR);
 151     if (export_name) {
 152         if (export_name[strlen(EN_OPTSTR)] == 0) {
 153             goto out;
 154         }
 155         export_name[0] = 0; /* truncate 'file' */
 156         export_name += strlen(EN_OPTSTR);
 157         s->export_name = g_strdup(export_name);
 158     }
 159
 160     /* extract the host_spec - fail if it's not nbd:... */
 161     if (!strstart(file, "nbd:", &host_spec)) {
 162         goto out;
 163     }
 164
 165     /* are we a UNIX or TCP socket? */
 166     if (strstart(host_spec, "unix:", &unixpath)) {
 167         s->is_unix = true;
 168         s->host_spec = g_strdup(unixpath);
 169     } else {
 170         s->is_unix = false;
 171         s->host_spec = g_strdup(host_spec);
 172     }
 173
 174     err = 0;
 175
 176 out:
 177     g_free(file);
 178     if (err != 0) {
 179         g_free(s->export_name);
 180         g_free(s->host_spec);
 181     }
 182     return err;
 183 }
 184
 185 static void nbd_coroutine_start(BDRVNBDState *s, struct nbd_request *request)
 186 {
 187     int i;
 188
 189     /* Poor man semaphore.  The free_sema is locked when no other request
 190      * can be accepted, and unlocked after receiving one reply.  */
 191     if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
 192         qemu_co_mutex_lock(&s->free_sema);
 193         assert(s->in_flight < MAX_NBD_REQUESTS);
 194     }
 195     s->in_flight++;
 196
 197     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 198         if (s->recv_coroutine[i] == NULL) {
 199             s->recv_coroutine[i] = qemu_coroutine_self();
 200             break;
 201         }
 202     }
 203
 204     assert(i < MAX_NBD_REQUESTS);
 205     request->handle = INDEX_TO_HANDLE(s, i);
 206 }
 207
 208 static int nbd_have_request(void *opaque)
 209 {
 210     BDRVNBDState *s = opaque;
 211
 212     return s->in_flight > 0;
 213 }
 214
 215 static void nbd_reply_ready(void *opaque)
 216 {
 217     BDRVNBDState *s = opaque;
 218     uint64_t i;
 219     int ret;
 220
 221     if (s->reply.handle == 0) {
 222         /* No reply already in flight.  Fetch a header.  It is possible
 223          * that another thread has done the same thing in parallel, so
 224          * the socket is not readable anymore.
 225          */
 226         ret = nbd_receive_reply(s->sock, &s->reply);
 227         if (ret == -EAGAIN) {
 228             return;
 229         }
 230         if (ret < 0) {
 231             s->reply.handle = 0;
 232             goto fail;
 233         }
 234     }
 235
 236     /* There's no need for a mutex on the receive side, because the
 237      * handler acts as a synchronization point and ensures that only
 238      * one coroutine is called until the reply finishes.  */
 239     i = HANDLE_TO_INDEX(s, s->reply.handle);
 240     if (i >= MAX_NBD_REQUESTS) {
 241         goto fail;
 242     }
 243
 244     if (s->recv_coroutine[i]) {
 245         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 246         return;
 247     }
 248
 249 fail:
 250     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 251         if (s->recv_coroutine[i]) {
 252             qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 253         }
 254     }
 255 }
 256
 257 static void nbd_restart_write(void *opaque)
 258 {
 259     BDRVNBDState *s = opaque;
 260     qemu_coroutine_enter(s->send_coroutine, NULL);
 261 }
 262
 263 static int nbd_co_send_request(BDRVNBDState *s, struct nbd_request *request,
 264                                QEMUIOVector *qiov, int offset)
 265 {
 266     int rc, ret;
 267
 268     qemu_co_mutex_lock(&s->send_mutex);
 269     s->send_coroutine = qemu_coroutine_self();
 270     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, nbd_restart_write,
 271                             nbd_have_request, s);
 272     rc = nbd_send_request(s->sock, request);
 273     if (rc >= 0 && qiov) {
 274         ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov,
 275                             offset, request->len);
 276         if (ret != request->len) {
 277             return -EIO;
 278         }
 279     }
 280     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 281                             nbd_have_request, s);
 282     s->send_coroutine = NULL;
 283     qemu_co_mutex_unlock(&s->send_mutex);
 284     return rc;
 285 }
 286
 287 static void nbd_co_receive_reply(BDRVNBDState *s, struct nbd_request *request,
 288                                  struct nbd_reply *reply,
 289                                  QEMUIOVector *qiov, int offset)
 290 {
 291     int ret;
 292
 293     /* Wait until we're woken up by the read handler.  TODO: perhaps
 294      * peek at the next reply and avoid yielding if it's ours?  */
 295     qemu_coroutine_yield();
 296     *reply = s->reply;
 297     if (reply->handle != request->handle) {
 298         reply->error = EIO;
 299     } else {
 300         if (qiov && reply->error == 0) {
 301             ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov,
 302                                 offset, request->len);
 303             if (ret != request->len) {
 304                 reply->error = EIO;
 305             }
 306         }
 307
 308         /* Tell the read handler to read another header.  */
 309         s->reply.handle = 0;
 310     }
 311 }
 312
 313 static void nbd_coroutine_end(BDRVNBDState *s, struct nbd_request *request)
 314 {
 315     int i = HANDLE_TO_INDEX(s, request->handle);
 316     s->recv_coroutine[i] = NULL;
 317     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 318         qemu_co_mutex_unlock(&s->free_sema);
 319     }
 320 }
 321
 322 static int nbd_establish_connection(BlockDriverState *bs)
 323 {
 324     BDRVNBDState *s = bs->opaque;
 325     int sock;
 326     int ret;
 327     off_t size;
 328     size_t blocksize;
 329
 330     if (s->is_unix) {
 331         sock = unix_socket_outgoing(s->host_spec);
 332     } else {
 333         sock = tcp_socket_outgoing_spec(s->host_spec);
 334     }
 335
 336     /* Failed to establish connection */
 337     if (sock < 0) {
 338         logout("Failed to establish connection to NBD server\n");
 339         return -errno;
 340     }
 341
 342     /* NBD handshake */
 343     ret = nbd_receive_negotiate(sock, s->export_name, &s->nbdflags, &size,
 344                                 &blocksize);
 345     if (ret < 0) {
 346         logout("Failed to negotiate with the NBD server\n");
 347         closesocket(sock);
 348         return ret;
 349     }
 350
 351     /* Now that we're connected, set the socket to be non-blocking and
 352      * kick the reply mechanism.  */
 353     socket_set_nonblock(sock);
 354     qemu_aio_set_fd_handler(sock, nbd_reply_ready, NULL,
 355                             nbd_have_request, s);
 356
 357     s->sock = sock;
 358     s->size = size;
 359     s->blocksize = blocksize;
 360
 361     logout("Established connection with NBD server\n");
 362     return 0;
 363 }
 364
 365 static void nbd_teardown_connection(BlockDriverState *bs)
 366 {
 367     BDRVNBDState *s = bs->opaque;
 368     struct nbd_request request;
 369
 370     request.type = NBD_CMD_DISC;
 371     request.from = 0;
 372     request.len = 0;
 373     nbd_send_request(s->sock, &request);
 374
 375     qemu_aio_set_fd_handler(s->sock, NULL, NULL, NULL, NULL);
 376     closesocket(s->sock);
 377 }
 378
 379 static int nbd_open(BlockDriverState *bs, const char* filename, int flags)
 380 {
 381     BDRVNBDState *s = bs->opaque;
 382     int result;
 383
 384     qemu_co_mutex_init(&s->send_mutex);
 385     qemu_co_mutex_init(&s->free_sema);
 386
 387     /* Pop the config into our state object. Exit if invalid. */
 388     result = nbd_config(s, filename);
 389     if (result != 0) {
 390         return result;
 391     }
 392
 393     /* establish TCP connection, return error if it fails
 394      * TODO: Configurable retry-until-timeout behaviour.
 395      */
 396     result = nbd_establish_connection(bs);
 397
 398     return result;
 399 }
 400
 401 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 402                           int nb_sectors, QEMUIOVector *qiov,
 403                           int offset)
 404 {
 405     BDRVNBDState *s = bs->opaque;
 406     struct nbd_request request;
 407     struct nbd_reply reply;
 408     ssize_t ret;
 409
 410     request.type = NBD_CMD_READ;
 411     request.from = sector_num * 512;
 412     request.len = nb_sectors * 512;
 413
 414     nbd_coroutine_start(s, &request);
 415     ret = nbd_co_send_request(s, &request, NULL, 0);
 416     if (ret < 0) {
 417         reply.error = -ret;
 418     } else {
 419         nbd_co_receive_reply(s, &request, &reply, qiov, offset);
 420     }
 421     nbd_coroutine_end(s, &request);
 422     return -reply.error;
 423
 424 }
 425
 426 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 427                            int nb_sectors, QEMUIOVector *qiov,
 428                            int offset)
 429 {
 430     BDRVNBDState *s = bs->opaque;
 431     struct nbd_request request;
 432     struct nbd_reply reply;
 433     ssize_t ret;
 434
 435     request.type = NBD_CMD_WRITE;
 436     if (!bdrv_enable_write_cache(bs) && (s->nbdflags & NBD_FLAG_SEND_FUA)) {
 437         request.type |= NBD_CMD_FLAG_FUA;
 438     }
 439
 440     request.from = sector_num * 512;
 441     request.len = nb_sectors * 512;
 442
 443     nbd_coroutine_start(s, &request);
 444     ret = nbd_co_send_request(s, &request, qiov, offset);
 445     if (ret < 0) {
 446         reply.error = -ret;
 447     } else {
 448         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 449     }
 450     nbd_coroutine_end(s, &request);
 451     return -reply.error;
 452 }
 453
 454 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 455  * remain aligned to 4K. */
 456 #define NBD_MAX_SECTORS 2040
 457
 458 static int nbd_co_readv(BlockDriverState *bs, int64_t sector_num,
 459                         int nb_sectors, QEMUIOVector *qiov)
 460 {
 461     int offset = 0;
 462     int ret;
 463     while (nb_sectors > NBD_MAX_SECTORS) {
 464         ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 465         if (ret < 0) {
 466             return ret;
 467         }
 468         offset += NBD_MAX_SECTORS * 512;
 469         sector_num += NBD_MAX_SECTORS;
 470         nb_sectors -= NBD_MAX_SECTORS;
 471     }
 472     return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 473 }
 474
 475 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
 476                          int nb_sectors, QEMUIOVector *qiov)
 477 {
 478     int offset = 0;
 479     int ret;
 480     while (nb_sectors > NBD_MAX_SECTORS) {
 481         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 482         if (ret < 0) {
 483             return ret;
 484         }
 485         offset += NBD_MAX_SECTORS * 512;
 486         sector_num += NBD_MAX_SECTORS;
 487         nb_sectors -= NBD_MAX_SECTORS;
 488     }
 489     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 490 }
 491
 492 static int nbd_co_flush(BlockDriverState *bs)
 493 {
 494     BDRVNBDState *s = bs->opaque;
 495     struct nbd_request request;
 496     struct nbd_reply reply;
 497     ssize_t ret;
 498
 499     if (!(s->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 500         return 0;
 501     }
 502
 503     request.type = NBD_CMD_FLUSH;
 504     if (s->nbdflags & NBD_FLAG_SEND_FUA) {
 505         request.type |= NBD_CMD_FLAG_FUA;
 506     }
 507
 508     request.from = 0;
 509     request.len = 0;
 510
 511     nbd_coroutine_start(s, &request);
 512     ret = nbd_co_send_request(s, &request, NULL, 0);
 513     if (ret < 0) {
 514         reply.error = -ret;
 515     } else {
 516         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 517     }
 518     nbd_coroutine_end(s, &request);
 519     return -reply.error;
 520 }
 521
 522 static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num,
 523                           int nb_sectors)
 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_TRIM)) {
 531         return 0;
 532     }
 533     request.type = NBD_CMD_TRIM;
 534     request.from = sector_num * 512;;
 535     request.len = nb_sectors * 512;
 536
 537     nbd_coroutine_start(s, &request);
 538     ret = nbd_co_send_request(s, &request, NULL, 0);
 539     if (ret < 0) {
 540         reply.error = -ret;
 541     } else {
 542         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 543     }
 544     nbd_coroutine_end(s, &request);
 545     return -reply.error;
 546 }
 547
 548 static void nbd_close(BlockDriverState *bs)
 549 {
 550     BDRVNBDState *s = bs->opaque;
 551     g_free(s->export_name);
 552     g_free(s->host_spec);
 553
 554     nbd_teardown_connection(bs);
 555 }
 556
 557 static int64_t nbd_getlength(BlockDriverState *bs)
 558 {
 559     BDRVNBDState *s = bs->opaque;
 560
 561     return s->size;
 562 }
 563
 564 static BlockDriver bdrv_nbd = {
 565     .format_name         = "nbd",
 566     .protocol_name       = "nbd",
 567     .instance_size       = sizeof(BDRVNBDState),
 568     .bdrv_file_open      = nbd_open,
 569     .bdrv_co_readv       = nbd_co_readv,
 570     .bdrv_co_writev      = nbd_co_writev,
 571     .bdrv_close          = nbd_close,
 572     .bdrv_co_flush_to_os = nbd_co_flush,
 573     .bdrv_co_discard     = nbd_co_discard,
 574     .bdrv_getlength      = nbd_getlength,
 575 };
 576
 577 static BlockDriver bdrv_nbd_tcp = {
 578     .format_name         = "nbd",
 579     .protocol_name       = "nbd+tcp",
 580     .instance_size       = sizeof(BDRVNBDState),
 581     .bdrv_file_open      = nbd_open,
 582     .bdrv_co_readv       = nbd_co_readv,
 583     .bdrv_co_writev      = nbd_co_writev,
 584     .bdrv_close          = nbd_close,
 585     .bdrv_co_flush_to_os = nbd_co_flush,
 586     .bdrv_co_discard     = nbd_co_discard,
 587     .bdrv_getlength      = nbd_getlength,
 588 };
 589
 590 static BlockDriver bdrv_nbd_unix = {
 591     .format_name         = "nbd",
 592     .protocol_name       = "nbd+unix",
 593     .instance_size       = sizeof(BDRVNBDState),
 594     .bdrv_file_open      = nbd_open,
 595     .bdrv_co_readv       = nbd_co_readv,
 596     .bdrv_co_writev      = nbd_co_writev,
 597     .bdrv_close          = nbd_close,
 598     .bdrv_co_flush_to_os = nbd_co_flush,
 599     .bdrv_co_discard     = nbd_co_discard,
 600     .bdrv_getlength      = nbd_getlength,
 601 };
 602
 603 static void bdrv_nbd_init(void)
 604 {
 605     bdrv_register(&bdrv_nbd);
 606     bdrv_register(&bdrv_nbd_tcp);
 607     bdrv_register(&bdrv_nbd_unix);
 608 }
 609
 610 block_init(bdrv_nbd_init);