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     char *export_name; /* An NBD server may export several devices */
  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     /* If it begins with  '/', this is a UNIX domain socket. Otherwise,
  69      * it's a string of the form <hostname|ip4|\[ip6\]>:port
  70      */
  71     char *host_spec;
  72 } BDRVNBDState;
  73
  74 static int nbd_config(BDRVNBDState *s, const char *filename, int flags)
  75 {
  76     char *file;
  77     char *export_name;
  78     const char *host_spec;
  79     const char *unixpath;
  80     int err = -EINVAL;
  81
  82     file = g_strdup(filename);
  83
  84     export_name = strstr(file, EN_OPTSTR);
  85     if (export_name) {
  86         if (export_name[strlen(EN_OPTSTR)] == 0) {
  87             goto out;
  88         }
  89         export_name[0] = 0; /* truncate 'file' */
  90         export_name += strlen(EN_OPTSTR);
  91         s->export_name = g_strdup(export_name);
  92     }
  93
  94     /* extract the host_spec - fail if it's not nbd:... */
  95     if (!strstart(file, "nbd:", &host_spec)) {
  96         goto out;
  97     }
  98
  99     /* are we a UNIX or TCP socket? */
 100     if (strstart(host_spec, "unix:", &unixpath)) {
 101         if (unixpath[0] != '/') { /* We demand  an absolute path*/
 102             goto out;
 103         }
 104         s->host_spec = g_strdup(unixpath);
 105     } else {
 106         s->host_spec = g_strdup(host_spec);
 107     }
 108
 109     err = 0;
 110
 111 out:
 112     g_free(file);
 113     if (err != 0) {
 114         g_free(s->export_name);
 115         g_free(s->host_spec);
 116     }
 117     return err;
 118 }
 119
 120 static void nbd_coroutine_start(BDRVNBDState *s, struct nbd_request *request)
 121 {
 122     int i;
 123
 124     /* Poor man semaphore.  The free_sema is locked when no other request
 125      * can be accepted, and unlocked after receiving one reply.  */
 126     if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
 127         qemu_co_mutex_lock(&s->free_sema);
 128         assert(s->in_flight < MAX_NBD_REQUESTS);
 129     }
 130     s->in_flight++;
 131
 132     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 133         if (s->recv_coroutine[i] == NULL) {
 134             s->recv_coroutine[i] = qemu_coroutine_self();
 135             break;
 136         }
 137     }
 138
 139     assert(i < MAX_NBD_REQUESTS);
 140     request->handle = INDEX_TO_HANDLE(s, i);
 141 }
 142
 143 static int nbd_have_request(void *opaque)
 144 {
 145     BDRVNBDState *s = opaque;
 146
 147     return s->in_flight > 0;
 148 }
 149
 150 static void nbd_reply_ready(void *opaque)
 151 {
 152     BDRVNBDState *s = opaque;
 153     uint64_t i;
 154
 155     if (s->reply.handle == 0) {
 156         /* No reply already in flight.  Fetch a header.  */
 157         if (nbd_receive_reply(s->sock, &s->reply) < 0) {
 158             s->reply.handle = 0;
 159             goto fail;
 160         }
 161     }
 162
 163     /* There's no need for a mutex on the receive side, because the
 164      * handler acts as a synchronization point and ensures that only
 165      * one coroutine is called until the reply finishes.  */
 166     i = HANDLE_TO_INDEX(s, s->reply.handle);
 167     if (i >= MAX_NBD_REQUESTS) {
 168         goto fail;
 169     }
 170
 171     if (s->recv_coroutine[i]) {
 172         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 173         return;
 174     }
 175
 176 fail:
 177     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 178         if (s->recv_coroutine[i]) {
 179             qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 180         }
 181     }
 182 }
 183
 184 static void nbd_restart_write(void *opaque)
 185 {
 186     BDRVNBDState *s = opaque;
 187     qemu_coroutine_enter(s->send_coroutine, NULL);
 188 }
 189
 190 static int nbd_co_send_request(BDRVNBDState *s, struct nbd_request *request,
 191                                struct iovec *iov, int offset)
 192 {
 193     int rc, ret;
 194
 195     qemu_co_mutex_lock(&s->send_mutex);
 196     s->send_coroutine = qemu_coroutine_self();
 197     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, nbd_restart_write,
 198                             nbd_have_request, NULL, s);
 199     rc = nbd_send_request(s->sock, request);
 200     if (rc != -1 && iov) {
 201         ret = qemu_co_sendv(s->sock, iov, request->len, offset);
 202         if (ret != request->len) {
 203             errno = -EIO;
 204             rc = -1;
 205         }
 206     }
 207     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 208                             nbd_have_request, NULL, s);
 209     s->send_coroutine = NULL;
 210     qemu_co_mutex_unlock(&s->send_mutex);
 211     return rc;
 212 }
 213
 214 static void nbd_co_receive_reply(BDRVNBDState *s, struct nbd_request *request,
 215                                  struct nbd_reply *reply,
 216                                  struct iovec *iov, int offset)
 217 {
 218     int ret;
 219
 220     /* Wait until we're woken up by the read handler.  TODO: perhaps
 221      * peek at the next reply and avoid yielding if it's ours?  */
 222     qemu_coroutine_yield();
 223     *reply = s->reply;
 224     if (reply->handle != request->handle) {
 225         reply->error = EIO;
 226     } else {
 227         if (iov && reply->error == 0) {
 228             ret = qemu_co_recvv(s->sock, iov, request->len, offset);
 229             if (ret != request->len) {
 230                 reply->error = EIO;
 231             }
 232         }
 233
 234         /* Tell the read handler to read another header.  */
 235         s->reply.handle = 0;
 236     }
 237 }
 238
 239 static void nbd_coroutine_end(BDRVNBDState *s, struct nbd_request *request)
 240 {
 241     int i = HANDLE_TO_INDEX(s, request->handle);
 242     s->recv_coroutine[i] = NULL;
 243     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 244         qemu_co_mutex_unlock(&s->free_sema);
 245     }
 246 }
 247
 248 static int nbd_establish_connection(BlockDriverState *bs)
 249 {
 250     BDRVNBDState *s = bs->opaque;
 251     int sock;
 252     int ret;
 253     off_t size;
 254     size_t blocksize;
 255
 256     if (s->host_spec[0] == '/') {
 257         sock = unix_socket_outgoing(s->host_spec);
 258     } else {
 259         sock = tcp_socket_outgoing_spec(s->host_spec);
 260     }
 261
 262     /* Failed to establish connection */
 263     if (sock == -1) {
 264         logout("Failed to establish connection to NBD server\n");
 265         return -errno;
 266     }
 267
 268     /* NBD handshake */
 269     ret = nbd_receive_negotiate(sock, s->export_name, &s->nbdflags, &size,
 270                                 &blocksize);
 271     if (ret == -1) {
 272         logout("Failed to negotiate with the NBD server\n");
 273         closesocket(sock);
 274         return -errno;
 275     }
 276
 277     /* Now that we're connected, set the socket to be non-blocking and
 278      * kick the reply mechanism.  */
 279     socket_set_nonblock(sock);
 280     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 281                             nbd_have_request, NULL, s);
 282
 283     s->sock = sock;
 284     s->size = size;
 285     s->blocksize = blocksize;
 286
 287     logout("Established connection with NBD server\n");
 288     return 0;
 289 }
 290
 291 static void nbd_teardown_connection(BlockDriverState *bs)
 292 {
 293     BDRVNBDState *s = bs->opaque;
 294     struct nbd_request request;
 295
 296     request.type = NBD_CMD_DISC;
 297     request.from = 0;
 298     request.len = 0;
 299     nbd_send_request(s->sock, &request);
 300
 301     qemu_aio_set_fd_handler(s->sock, NULL, NULL, NULL, NULL, NULL);
 302     closesocket(s->sock);
 303 }
 304
 305 static int nbd_open(BlockDriverState *bs, const char* filename, int flags)
 306 {
 307     BDRVNBDState *s = bs->opaque;
 308     int result;
 309
 310     qemu_co_mutex_init(&s->send_mutex);
 311     qemu_co_mutex_init(&s->free_sema);
 312
 313     /* Pop the config into our state object. Exit if invalid. */
 314     result = nbd_config(s, filename, flags);
 315     if (result != 0) {
 316         return result;
 317     }
 318
 319     /* establish TCP connection, return error if it fails
 320      * TODO: Configurable retry-until-timeout behaviour.
 321      */
 322     result = nbd_establish_connection(bs);
 323
 324     return result;
 325 }
 326
 327 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 328                           int nb_sectors, QEMUIOVector *qiov,
 329                           int offset)
 330 {
 331     BDRVNBDState *s = bs->opaque;
 332     struct nbd_request request;
 333     struct nbd_reply reply;
 334
 335     request.type = NBD_CMD_READ;
 336     request.from = sector_num * 512;
 337     request.len = nb_sectors * 512;
 338
 339     nbd_coroutine_start(s, &request);
 340     if (nbd_co_send_request(s, &request, NULL, 0) == -1) {
 341         reply.error = errno;
 342     } else {
 343         nbd_co_receive_reply(s, &request, &reply, qiov->iov, offset);
 344     }
 345     nbd_coroutine_end(s, &request);
 346     return -reply.error;
 347
 348 }
 349
 350 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 351                            int nb_sectors, QEMUIOVector *qiov,
 352                            int offset)
 353 {
 354     BDRVNBDState *s = bs->opaque;
 355     struct nbd_request request;
 356     struct nbd_reply reply;
 357
 358     request.type = NBD_CMD_WRITE;
 359     if (!bdrv_enable_write_cache(bs) && (s->nbdflags & NBD_FLAG_SEND_FUA)) {
 360         request.type |= NBD_CMD_FLAG_FUA;
 361     }
 362
 363     request.from = sector_num * 512;
 364     request.len = nb_sectors * 512;
 365
 366     nbd_coroutine_start(s, &request);
 367     if (nbd_co_send_request(s, &request, qiov->iov, offset) == -1) {
 368         reply.error = errno;
 369     } else {
 370         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 371     }
 372     nbd_coroutine_end(s, &request);
 373     return -reply.error;
 374 }
 375
 376 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 377  * remain aligned to 4K. */
 378 #define NBD_MAX_SECTORS 2040
 379
 380 static int nbd_co_readv(BlockDriverState *bs, int64_t sector_num,
 381                         int nb_sectors, QEMUIOVector *qiov)
 382 {
 383     int offset = 0;
 384     int ret;
 385     while (nb_sectors > NBD_MAX_SECTORS) {
 386         ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 387         if (ret < 0) {
 388             return ret;
 389         }
 390         offset += NBD_MAX_SECTORS * 512;
 391         sector_num += NBD_MAX_SECTORS;
 392         nb_sectors -= NBD_MAX_SECTORS;
 393     }
 394     return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 395 }
 396
 397 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
 398                          int nb_sectors, QEMUIOVector *qiov)
 399 {
 400     int offset = 0;
 401     int ret;
 402     while (nb_sectors > NBD_MAX_SECTORS) {
 403         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 404         if (ret < 0) {
 405             return ret;
 406         }
 407         offset += NBD_MAX_SECTORS * 512;
 408         sector_num += NBD_MAX_SECTORS;
 409         nb_sectors -= NBD_MAX_SECTORS;
 410     }
 411     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 412 }
 413
 414 static int nbd_co_flush(BlockDriverState *bs)
 415 {
 416     BDRVNBDState *s = bs->opaque;
 417     struct nbd_request request;
 418     struct nbd_reply reply;
 419
 420     if (!(s->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 421         return 0;
 422     }
 423
 424     request.type = NBD_CMD_FLUSH;
 425     if (s->nbdflags & NBD_FLAG_SEND_FUA) {
 426         request.type |= NBD_CMD_FLAG_FUA;
 427     }
 428
 429     request.from = 0;
 430     request.len = 0;
 431
 432     nbd_coroutine_start(s, &request);
 433     if (nbd_co_send_request(s, &request, NULL, 0) == -1) {
 434         reply.error = errno;
 435     } else {
 436         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 437     }
 438     nbd_coroutine_end(s, &request);
 439     return -reply.error;
 440 }
 441
 442 static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num,
 443                           int nb_sectors)
 444 {
 445     BDRVNBDState *s = bs->opaque;
 446     struct nbd_request request;
 447     struct nbd_reply reply;
 448
 449     if (!(s->nbdflags & NBD_FLAG_SEND_TRIM)) {
 450         return 0;
 451     }
 452     request.type = NBD_CMD_TRIM;
 453     request.from = sector_num * 512;;
 454     request.len = nb_sectors * 512;
 455
 456     nbd_coroutine_start(s, &request);
 457     if (nbd_co_send_request(s, &request, NULL, 0) == -1) {
 458         reply.error = errno;
 459     } else {
 460         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 461     }
 462     nbd_coroutine_end(s, &request);
 463     return -reply.error;
 464 }
 465
 466 static void nbd_close(BlockDriverState *bs)
 467 {
 468     BDRVNBDState *s = bs->opaque;
 469     g_free(s->export_name);
 470     g_free(s->host_spec);
 471
 472     nbd_teardown_connection(bs);
 473 }
 474
 475 static int64_t nbd_getlength(BlockDriverState *bs)
 476 {
 477     BDRVNBDState *s = bs->opaque;
 478
 479     return s->size;
 480 }
 481
 482 static BlockDriver bdrv_nbd = {
 483     .format_name         = "nbd",
 484     .instance_size       = sizeof(BDRVNBDState),
 485     .bdrv_file_open      = nbd_open,
 486     .bdrv_co_readv       = nbd_co_readv,
 487     .bdrv_co_writev      = nbd_co_writev,
 488     .bdrv_close          = nbd_close,
 489     .bdrv_co_flush_to_os = nbd_co_flush,
 490     .bdrv_co_discard     = nbd_co_discard,
 491     .bdrv_getlength      = nbd_getlength,
 492     .protocol_name       = "nbd",
 493 };
 494
 495 static void bdrv_nbd_init(void)
 496 {
 497     bdrv_register(&bdrv_nbd);
 498 }
 499
 500 block_init(bdrv_nbd_init);