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 "nbd.h"
  31 #include "block_int.h"
  32 #include "module.h"
  33 #include "qemu_socket.h"
  34
  35 #include <sys/types.h>
  36 #include <unistd.h>
  37
  38 #define EN_OPTSTR ":exportname="
  39
  40 /* #define DEBUG_NBD */
  41
  42 #if defined(DEBUG_NBD)
  43 #define logout(fmt, ...) \
  44                 fprintf(stderr, "nbd\t%-24s" fmt, __func__, ##__VA_ARGS__)
  45 #else
  46 #define logout(fmt, ...) ((void)0)
  47 #endif
  48
  49 #define MAX_NBD_REQUESTS        16
  50 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
  51 #define INDEX_TO_HANDLE(bs, index)  ((index)  ^ ((uint64_t)(intptr_t)bs))
  52
  53 typedef struct BDRVNBDState {
  54     int sock;
  55     uint32_t nbdflags;
  56     off_t size;
  57     size_t blocksize;
  58
  59     CoMutex send_mutex;
  60     CoMutex free_sema;
  61     Coroutine *send_coroutine;
  62     int in_flight;
  63
  64     Coroutine *recv_coroutine[MAX_NBD_REQUESTS];
  65     struct nbd_reply reply;
  66
  67     int is_unix;
  68     char *host_spec;
  69     char *export_name; /* An NBD server may export several devices */
  70 } BDRVNBDState;
  71
  72 static int nbd_config(BDRVNBDState *s, const char *filename)
  73 {
  74     char *file;
  75     char *export_name;
  76     const char *host_spec;
  77     const char *unixpath;
  78     int err = -EINVAL;
  79
  80     file = g_strdup(filename);
  81
  82     export_name = strstr(file, EN_OPTSTR);
  83     if (export_name) {
  84         if (export_name[strlen(EN_OPTSTR)] == 0) {
  85             goto out;
  86         }
  87         export_name[0] = 0; /* truncate 'file' */
  88         export_name += strlen(EN_OPTSTR);
  89         s->export_name = g_strdup(export_name);
  90     }
  91
  92     /* extract the host_spec - fail if it's not nbd:... */
  93     if (!strstart(file, "nbd:", &host_spec)) {
  94         goto out;
  95     }
  96
  97     /* are we a UNIX or TCP socket? */
  98     if (strstart(host_spec, "unix:", &unixpath)) {
  99         s->is_unix = true;
 100         s->host_spec = g_strdup(unixpath);
 101     } else {
 102         s->is_unix = false;
 103         s->host_spec = g_strdup(host_spec);
 104     }
 105
 106     err = 0;
 107
 108 out:
 109     g_free(file);
 110     if (err != 0) {
 111         g_free(s->export_name);
 112         g_free(s->host_spec);
 113     }
 114     return err;
 115 }
 116
 117 static void nbd_coroutine_start(BDRVNBDState *s, struct nbd_request *request)
 118 {
 119     int i;
 120
 121     /* Poor man semaphore.  The free_sema is locked when no other request
 122      * can be accepted, and unlocked after receiving one reply.  */
 123     if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
 124         qemu_co_mutex_lock(&s->free_sema);
 125         assert(s->in_flight < MAX_NBD_REQUESTS);
 126     }
 127     s->in_flight++;
 128
 129     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 130         if (s->recv_coroutine[i] == NULL) {
 131             s->recv_coroutine[i] = qemu_coroutine_self();
 132             break;
 133         }
 134     }
 135
 136     assert(i < MAX_NBD_REQUESTS);
 137     request->handle = INDEX_TO_HANDLE(s, i);
 138 }
 139
 140 static int nbd_have_request(void *opaque)
 141 {
 142     BDRVNBDState *s = opaque;
 143
 144     return s->in_flight > 0;
 145 }
 146
 147 static void nbd_reply_ready(void *opaque)
 148 {
 149     BDRVNBDState *s = opaque;
 150     uint64_t i;
 151     int ret;
 152
 153     if (s->reply.handle == 0) {
 154         /* No reply already in flight.  Fetch a header.  It is possible
 155          * that another thread has done the same thing in parallel, so
 156          * the socket is not readable anymore.
 157          */
 158         ret = nbd_receive_reply(s->sock, &s->reply);
 159         if (ret == -EAGAIN) {
 160             return;
 161         }
 162         if (ret < 0) {
 163             s->reply.handle = 0;
 164             goto fail;
 165         }
 166     }
 167
 168     /* There's no need for a mutex on the receive side, because the
 169      * handler acts as a synchronization point and ensures that only
 170      * one coroutine is called until the reply finishes.  */
 171     i = HANDLE_TO_INDEX(s, s->reply.handle);
 172     if (i >= MAX_NBD_REQUESTS) {
 173         goto fail;
 174     }
 175
 176     if (s->recv_coroutine[i]) {
 177         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 178         return;
 179     }
 180
 181 fail:
 182     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 183         if (s->recv_coroutine[i]) {
 184             qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 185         }
 186     }
 187 }
 188
 189 static void nbd_restart_write(void *opaque)
 190 {
 191     BDRVNBDState *s = opaque;
 192     qemu_coroutine_enter(s->send_coroutine, NULL);
 193 }
 194
 195 static int nbd_co_send_request(BDRVNBDState *s, struct nbd_request *request,
 196                                QEMUIOVector *qiov, int offset)
 197 {
 198     int rc, ret;
 199
 200     qemu_co_mutex_lock(&s->send_mutex);
 201     s->send_coroutine = qemu_coroutine_self();
 202     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, nbd_restart_write,
 203                             nbd_have_request, s);
 204     rc = nbd_send_request(s->sock, request);
 205     if (rc >= 0 && qiov) {
 206         ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov,
 207                             offset, request->len);
 208         if (ret != request->len) {
 209             return -EIO;
 210         }
 211     }
 212     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 213                             nbd_have_request, s);
 214     s->send_coroutine = NULL;
 215     qemu_co_mutex_unlock(&s->send_mutex);
 216     return rc;
 217 }
 218
 219 static void nbd_co_receive_reply(BDRVNBDState *s, struct nbd_request *request,
 220                                  struct nbd_reply *reply,
 221                                  QEMUIOVector *qiov, int offset)
 222 {
 223     int ret;
 224
 225     /* Wait until we're woken up by the read handler.  TODO: perhaps
 226      * peek at the next reply and avoid yielding if it's ours?  */
 227     qemu_coroutine_yield();
 228     *reply = s->reply;
 229     if (reply->handle != request->handle) {
 230         reply->error = EIO;
 231     } else {
 232         if (qiov && reply->error == 0) {
 233             ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov,
 234                                 offset, request->len);
 235             if (ret != request->len) {
 236                 reply->error = EIO;
 237             }
 238         }
 239
 240         /* Tell the read handler to read another header.  */
 241         s->reply.handle = 0;
 242     }
 243 }
 244
 245 static void nbd_coroutine_end(BDRVNBDState *s, struct nbd_request *request)
 246 {
 247     int i = HANDLE_TO_INDEX(s, request->handle);
 248     s->recv_coroutine[i] = NULL;
 249     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 250         qemu_co_mutex_unlock(&s->free_sema);
 251     }
 252 }
 253
 254 static int nbd_establish_connection(BlockDriverState *bs)
 255 {
 256     BDRVNBDState *s = bs->opaque;
 257     int sock;
 258     int ret;
 259     off_t size;
 260     size_t blocksize;
 261
 262     if (s->is_unix) {
 263         sock = unix_socket_outgoing(s->host_spec);
 264     } else {
 265         sock = tcp_socket_outgoing_spec(s->host_spec);
 266     }
 267
 268     /* Failed to establish connection */
 269     if (sock < 0) {
 270         logout("Failed to establish connection to NBD server\n");
 271         return -errno;
 272     }
 273
 274     /* NBD handshake */
 275     ret = nbd_receive_negotiate(sock, s->export_name, &s->nbdflags, &size,
 276                                 &blocksize);
 277     if (ret < 0) {
 278         logout("Failed to negotiate with the NBD server\n");
 279         closesocket(sock);
 280         return ret;
 281     }
 282
 283     /* Now that we're connected, set the socket to be non-blocking and
 284      * kick the reply mechanism.  */
 285     socket_set_nonblock(sock);
 286     qemu_aio_set_fd_handler(sock, nbd_reply_ready, NULL,
 287                             nbd_have_request, s);
 288
 289     s->sock = sock;
 290     s->size = size;
 291     s->blocksize = blocksize;
 292
 293     logout("Established connection with NBD server\n");
 294     return 0;
 295 }
 296
 297 static void nbd_teardown_connection(BlockDriverState *bs)
 298 {
 299     BDRVNBDState *s = bs->opaque;
 300     struct nbd_request request;
 301
 302     request.type = NBD_CMD_DISC;
 303     request.from = 0;
 304     request.len = 0;
 305     nbd_send_request(s->sock, &request);
 306
 307     qemu_aio_set_fd_handler(s->sock, NULL, NULL, NULL, NULL);
 308     closesocket(s->sock);
 309 }
 310
 311 static int nbd_open(BlockDriverState *bs, const char* filename, int flags)
 312 {
 313     BDRVNBDState *s = bs->opaque;
 314     int result;
 315
 316     qemu_co_mutex_init(&s->send_mutex);
 317     qemu_co_mutex_init(&s->free_sema);
 318
 319     /* Pop the config into our state object. Exit if invalid. */
 320     result = nbd_config(s, filename);
 321     if (result != 0) {
 322         return result;
 323     }
 324
 325     /* establish TCP connection, return error if it fails
 326      * TODO: Configurable retry-until-timeout behaviour.
 327      */
 328     result = nbd_establish_connection(bs);
 329
 330     return result;
 331 }
 332
 333 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 334                           int nb_sectors, QEMUIOVector *qiov,
 335                           int offset)
 336 {
 337     BDRVNBDState *s = bs->opaque;
 338     struct nbd_request request;
 339     struct nbd_reply reply;
 340     ssize_t ret;
 341
 342     request.type = NBD_CMD_READ;
 343     request.from = sector_num * 512;
 344     request.len = nb_sectors * 512;
 345
 346     nbd_coroutine_start(s, &request);
 347     ret = nbd_co_send_request(s, &request, NULL, 0);
 348     if (ret < 0) {
 349         reply.error = -ret;
 350     } else {
 351         nbd_co_receive_reply(s, &request, &reply, qiov, offset);
 352     }
 353     nbd_coroutine_end(s, &request);
 354     return -reply.error;
 355
 356 }
 357
 358 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 359                            int nb_sectors, QEMUIOVector *qiov,
 360                            int offset)
 361 {
 362     BDRVNBDState *s = bs->opaque;
 363     struct nbd_request request;
 364     struct nbd_reply reply;
 365     ssize_t ret;
 366
 367     request.type = NBD_CMD_WRITE;
 368     if (!bdrv_enable_write_cache(bs) && (s->nbdflags & NBD_FLAG_SEND_FUA)) {
 369         request.type |= NBD_CMD_FLAG_FUA;
 370     }
 371
 372     request.from = sector_num * 512;
 373     request.len = nb_sectors * 512;
 374
 375     nbd_coroutine_start(s, &request);
 376     ret = nbd_co_send_request(s, &request, qiov, offset);
 377     if (ret < 0) {
 378         reply.error = -ret;
 379     } else {
 380         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 381     }
 382     nbd_coroutine_end(s, &request);
 383     return -reply.error;
 384 }
 385
 386 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 387  * remain aligned to 4K. */
 388 #define NBD_MAX_SECTORS 2040
 389
 390 static int nbd_co_readv(BlockDriverState *bs, int64_t sector_num,
 391                         int nb_sectors, QEMUIOVector *qiov)
 392 {
 393     int offset = 0;
 394     int ret;
 395     while (nb_sectors > NBD_MAX_SECTORS) {
 396         ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 397         if (ret < 0) {
 398             return ret;
 399         }
 400         offset += NBD_MAX_SECTORS * 512;
 401         sector_num += NBD_MAX_SECTORS;
 402         nb_sectors -= NBD_MAX_SECTORS;
 403     }
 404     return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 405 }
 406
 407 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
 408                          int nb_sectors, QEMUIOVector *qiov)
 409 {
 410     int offset = 0;
 411     int ret;
 412     while (nb_sectors > NBD_MAX_SECTORS) {
 413         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 414         if (ret < 0) {
 415             return ret;
 416         }
 417         offset += NBD_MAX_SECTORS * 512;
 418         sector_num += NBD_MAX_SECTORS;
 419         nb_sectors -= NBD_MAX_SECTORS;
 420     }
 421     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 422 }
 423
 424 static int nbd_co_flush(BlockDriverState *bs)
 425 {
 426     BDRVNBDState *s = bs->opaque;
 427     struct nbd_request request;
 428     struct nbd_reply reply;
 429     ssize_t ret;
 430
 431     if (!(s->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 432         return 0;
 433     }
 434
 435     request.type = NBD_CMD_FLUSH;
 436     if (s->nbdflags & NBD_FLAG_SEND_FUA) {
 437         request.type |= NBD_CMD_FLAG_FUA;
 438     }
 439
 440     request.from = 0;
 441     request.len = 0;
 442
 443     nbd_coroutine_start(s, &request);
 444     ret = nbd_co_send_request(s, &request, NULL, 0);
 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 static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num,
 455                           int nb_sectors)
 456 {
 457     BDRVNBDState *s = bs->opaque;
 458     struct nbd_request request;
 459     struct nbd_reply reply;
 460     ssize_t ret;
 461
 462     if (!(s->nbdflags & NBD_FLAG_SEND_TRIM)) {
 463         return 0;
 464     }
 465     request.type = NBD_CMD_TRIM;
 466     request.from = sector_num * 512;;
 467     request.len = nb_sectors * 512;
 468
 469     nbd_coroutine_start(s, &request);
 470     ret = nbd_co_send_request(s, &request, NULL, 0);
 471     if (ret < 0) {
 472         reply.error = -ret;
 473     } else {
 474         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 475     }
 476     nbd_coroutine_end(s, &request);
 477     return -reply.error;
 478 }
 479
 480 static void nbd_close(BlockDriverState *bs)
 481 {
 482     BDRVNBDState *s = bs->opaque;
 483     g_free(s->export_name);
 484     g_free(s->host_spec);
 485
 486     nbd_teardown_connection(bs);
 487 }
 488
 489 static int64_t nbd_getlength(BlockDriverState *bs)
 490 {
 491     BDRVNBDState *s = bs->opaque;
 492
 493     return s->size;
 494 }
 495
 496 static BlockDriver bdrv_nbd = {
 497     .format_name         = "nbd",
 498     .instance_size       = sizeof(BDRVNBDState),
 499     .bdrv_file_open      = nbd_open,
 500     .bdrv_co_readv       = nbd_co_readv,
 501     .bdrv_co_writev      = nbd_co_writev,
 502     .bdrv_close          = nbd_close,
 503     .bdrv_co_flush_to_os = nbd_co_flush,
 504     .bdrv_co_discard     = nbd_co_discard,
 505     .bdrv_getlength      = nbd_getlength,
 506     .protocol_name       = "nbd",
 507 };
 508
 509 static void bdrv_nbd_init(void)
 510 {
 511     bdrv_register(&bdrv_nbd);
 512 }
 513
 514 block_init(bdrv_nbd_init);