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     int 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 (s->recv_coroutine[i]) {
 168         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 169         return;
 170     }
 171
 172 fail:
 173     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 174         if (s->recv_coroutine[i]) {
 175             qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 176         }
 177     }
 178 }
 179
 180 static void nbd_restart_write(void *opaque)
 181 {
 182     BDRVNBDState *s = opaque;
 183     qemu_coroutine_enter(s->send_coroutine, NULL);
 184 }
 185
 186 static int nbd_co_send_request(BDRVNBDState *s, struct nbd_request *request,
 187                                struct iovec *iov, int offset)
 188 {
 189     int rc, ret;
 190
 191     qemu_co_mutex_lock(&s->send_mutex);
 192     s->send_coroutine = qemu_coroutine_self();
 193     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, nbd_restart_write,
 194                             nbd_have_request, s);
 195     rc = nbd_send_request(s->sock, request);
 196     if (rc != -1 && iov) {
 197         ret = qemu_co_sendv(s->sock, iov, request->len, offset);
 198         if (ret != request->len) {
 199             errno = -EIO;
 200             rc = -1;
 201         }
 202     }
 203     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 204                             nbd_have_request, s);
 205     s->send_coroutine = NULL;
 206     qemu_co_mutex_unlock(&s->send_mutex);
 207     return rc;
 208 }
 209
 210 static void nbd_co_receive_reply(BDRVNBDState *s, struct nbd_request *request,
 211                                  struct nbd_reply *reply,
 212                                  struct iovec *iov, int offset)
 213 {
 214     int ret;
 215
 216     /* Wait until we're woken up by the read handler.  TODO: perhaps
 217      * peek at the next reply and avoid yielding if it's ours?  */
 218     qemu_coroutine_yield();
 219     *reply = s->reply;
 220     if (reply->handle != request->handle) {
 221         reply->error = EIO;
 222     } else {
 223         if (iov && reply->error == 0) {
 224             ret = qemu_co_recvv(s->sock, iov, request->len, offset);
 225             if (ret != request->len) {
 226                 reply->error = EIO;
 227             }
 228         }
 229
 230         /* Tell the read handler to read another header.  */
 231         s->reply.handle = 0;
 232     }
 233 }
 234
 235 static void nbd_coroutine_end(BDRVNBDState *s, struct nbd_request *request)
 236 {
 237     int i = HANDLE_TO_INDEX(s, request->handle);
 238     s->recv_coroutine[i] = NULL;
 239     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 240         qemu_co_mutex_unlock(&s->free_sema);
 241     }
 242 }
 243
 244 static int nbd_establish_connection(BlockDriverState *bs)
 245 {
 246     BDRVNBDState *s = bs->opaque;
 247     int sock;
 248     int ret;
 249     off_t size;
 250     size_t blocksize;
 251
 252     if (s->host_spec[0] == '/') {
 253         sock = unix_socket_outgoing(s->host_spec);
 254     } else {
 255         sock = tcp_socket_outgoing_spec(s->host_spec);
 256     }
 257
 258     /* Failed to establish connection */
 259     if (sock == -1) {
 260         logout("Failed to establish connection to NBD server\n");
 261         return -errno;
 262     }
 263
 264     /* NBD handshake */
 265     ret = nbd_receive_negotiate(sock, s->export_name, &s->nbdflags, &size,
 266                                 &blocksize);
 267     if (ret == -1) {
 268         logout("Failed to negotiate with the NBD server\n");
 269         closesocket(sock);
 270         return -errno;
 271     }
 272
 273     /* Now that we're connected, set the socket to be non-blocking and
 274      * kick the reply mechanism.  */
 275     socket_set_nonblock(sock);
 276     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 277                             nbd_have_request, s);
 278
 279     s->sock = sock;
 280     s->size = size;
 281     s->blocksize = blocksize;
 282
 283     logout("Established connection with NBD server\n");
 284     return 0;
 285 }
 286
 287 static void nbd_teardown_connection(BlockDriverState *bs)
 288 {
 289     BDRVNBDState *s = bs->opaque;
 290     struct nbd_request request;
 291
 292     request.type = NBD_CMD_DISC;
 293     request.from = 0;
 294     request.len = 0;
 295     nbd_send_request(s->sock, &request);
 296
 297     qemu_aio_set_fd_handler(s->sock, NULL, NULL, NULL, NULL);
 298     closesocket(s->sock);
 299 }
 300
 301 static int nbd_open(BlockDriverState *bs, const char* filename, int flags)
 302 {
 303     BDRVNBDState *s = bs->opaque;
 304     int result;
 305
 306     qemu_co_mutex_init(&s->send_mutex);
 307     qemu_co_mutex_init(&s->free_sema);
 308
 309     /* Pop the config into our state object. Exit if invalid. */
 310     result = nbd_config(s, filename, flags);
 311     if (result != 0) {
 312         return result;
 313     }
 314
 315     /* establish TCP connection, return error if it fails
 316      * TODO: Configurable retry-until-timeout behaviour.
 317      */
 318     result = nbd_establish_connection(bs);
 319
 320     return result;
 321 }
 322
 323 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 324                           int nb_sectors, QEMUIOVector *qiov,
 325                           int offset)
 326 {
 327     BDRVNBDState *s = bs->opaque;
 328     struct nbd_request request;
 329     struct nbd_reply reply;
 330
 331     request.type = NBD_CMD_READ;
 332     request.from = sector_num * 512;
 333     request.len = nb_sectors * 512;
 334
 335     nbd_coroutine_start(s, &request);
 336     if (nbd_co_send_request(s, &request, NULL, 0) == -1) {
 337         reply.error = errno;
 338     } else {
 339         nbd_co_receive_reply(s, &request, &reply, qiov->iov, offset);
 340     }
 341     nbd_coroutine_end(s, &request);
 342     return -reply.error;
 343
 344 }
 345
 346 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 347                            int nb_sectors, QEMUIOVector *qiov,
 348                            int offset)
 349 {
 350     BDRVNBDState *s = bs->opaque;
 351     struct nbd_request request;
 352     struct nbd_reply reply;
 353
 354     request.type = NBD_CMD_WRITE;
 355     if (!bdrv_enable_write_cache(bs) && (s->nbdflags & NBD_FLAG_SEND_FUA)) {
 356         request.type |= NBD_CMD_FLAG_FUA;
 357     }
 358
 359     request.from = sector_num * 512;
 360     request.len = nb_sectors * 512;
 361
 362     nbd_coroutine_start(s, &request);
 363     if (nbd_co_send_request(s, &request, qiov->iov, offset) == -1) {
 364         reply.error = errno;
 365     } else {
 366         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 367     }
 368     nbd_coroutine_end(s, &request);
 369     return -reply.error;
 370 }
 371
 372 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 373  * remain aligned to 4K. */
 374 #define NBD_MAX_SECTORS 2040
 375
 376 static int nbd_co_readv(BlockDriverState *bs, int64_t sector_num,
 377                         int nb_sectors, QEMUIOVector *qiov)
 378 {
 379     int offset = 0;
 380     int ret;
 381     while (nb_sectors > NBD_MAX_SECTORS) {
 382         ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 383         if (ret < 0) {
 384             return ret;
 385         }
 386         offset += NBD_MAX_SECTORS * 512;
 387         sector_num += NBD_MAX_SECTORS;
 388         nb_sectors -= NBD_MAX_SECTORS;
 389     }
 390     return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 391 }
 392
 393 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
 394                          int nb_sectors, QEMUIOVector *qiov)
 395 {
 396     int offset = 0;
 397     int ret;
 398     while (nb_sectors > NBD_MAX_SECTORS) {
 399         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 400         if (ret < 0) {
 401             return ret;
 402         }
 403         offset += NBD_MAX_SECTORS * 512;
 404         sector_num += NBD_MAX_SECTORS;
 405         nb_sectors -= NBD_MAX_SECTORS;
 406     }
 407     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 408 }
 409
 410 static int nbd_co_flush(BlockDriverState *bs)
 411 {
 412     BDRVNBDState *s = bs->opaque;
 413     struct nbd_request request;
 414     struct nbd_reply reply;
 415
 416     if (!(s->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 417         return 0;
 418     }
 419
 420     request.type = NBD_CMD_FLUSH;
 421     if (s->nbdflags & NBD_FLAG_SEND_FUA) {
 422         request.type |= NBD_CMD_FLAG_FUA;
 423     }
 424
 425     request.from = 0;
 426     request.len = 0;
 427
 428     nbd_coroutine_start(s, &request);
 429     if (nbd_co_send_request(s, &request, NULL, 0) == -1) {
 430         reply.error = errno;
 431     } else {
 432         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 433     }
 434     nbd_coroutine_end(s, &request);
 435     return -reply.error;
 436 }
 437
 438 static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num,
 439                           int nb_sectors)
 440 {
 441     BDRVNBDState *s = bs->opaque;
 442     struct nbd_request request;
 443     struct nbd_reply reply;
 444
 445     if (!(s->nbdflags & NBD_FLAG_SEND_TRIM)) {
 446         return 0;
 447     }
 448     request.type = NBD_CMD_TRIM;
 449     request.from = sector_num * 512;;
 450     request.len = nb_sectors * 512;
 451
 452     nbd_coroutine_start(s, &request);
 453     if (nbd_co_send_request(s, &request, NULL, 0) == -1) {
 454         reply.error = errno;
 455     } else {
 456         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 457     }
 458     nbd_coroutine_end(s, &request);
 459     return -reply.error;
 460 }
 461
 462 static void nbd_close(BlockDriverState *bs)
 463 {
 464     BDRVNBDState *s = bs->opaque;
 465     g_free(s->export_name);
 466     g_free(s->host_spec);
 467
 468     nbd_teardown_connection(bs);
 469 }
 470
 471 static int64_t nbd_getlength(BlockDriverState *bs)
 472 {
 473     BDRVNBDState *s = bs->opaque;
 474
 475     return s->size;
 476 }
 477
 478 static BlockDriver bdrv_nbd = {
 479     .format_name         = "nbd",
 480     .instance_size       = sizeof(BDRVNBDState),
 481     .bdrv_file_open      = nbd_open,
 482     .bdrv_co_readv       = nbd_co_readv,
 483     .bdrv_co_writev      = nbd_co_writev,
 484     .bdrv_close          = nbd_close,
 485     .bdrv_co_flush_to_os = nbd_co_flush,
 486     .bdrv_co_discard     = nbd_co_discard,
 487     .bdrv_getlength      = nbd_getlength,
 488     .protocol_name       = "nbd",
 489 };
 490
 491 static void bdrv_nbd_init(void)
 492 {
 493     bdrv_register(&bdrv_nbd);
 494 }
 495
 496 block_init(bdrv_nbd_init);