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     int ret;
 155
 156     if (s->reply.handle == 0) {
 157         /* No reply already in flight.  Fetch a header.  It is possible
 158          * that another thread has done the same thing in parallel, so
 159          * the socket is not readable anymore.
 160          */
 161         ret = nbd_receive_reply(s->sock, &s->reply);
 162         if (ret == -EAGAIN) {
 163             return;
 164         }
 165         if (ret < 0) {
 166             s->reply.handle = 0;
 167             goto fail;
 168         }
 169     }
 170
 171     /* There's no need for a mutex on the receive side, because the
 172      * handler acts as a synchronization point and ensures that only
 173      * one coroutine is called until the reply finishes.  */
 174     i = HANDLE_TO_INDEX(s, s->reply.handle);
 175     if (i >= MAX_NBD_REQUESTS) {
 176         goto fail;
 177     }
 178
 179     if (s->recv_coroutine[i]) {
 180         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 181         return;
 182     }
 183
 184 fail:
 185     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 186         if (s->recv_coroutine[i]) {
 187             qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 188         }
 189     }
 190 }
 191
 192 static void nbd_restart_write(void *opaque)
 193 {
 194     BDRVNBDState *s = opaque;
 195     qemu_coroutine_enter(s->send_coroutine, NULL);
 196 }
 197
 198 static int nbd_co_send_request(BDRVNBDState *s, struct nbd_request *request,
 199                                QEMUIOVector *qiov, int offset)
 200 {
 201     int rc, ret;
 202
 203     qemu_co_mutex_lock(&s->send_mutex);
 204     s->send_coroutine = qemu_coroutine_self();
 205     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, nbd_restart_write,
 206                             nbd_have_request, s);
 207     rc = nbd_send_request(s->sock, request);
 208     if (rc >= 0 && qiov) {
 209         ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov,
 210                             offset, request->len);
 211         if (ret != request->len) {
 212             return -EIO;
 213         }
 214     }
 215     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 216                             nbd_have_request, s);
 217     s->send_coroutine = NULL;
 218     qemu_co_mutex_unlock(&s->send_mutex);
 219     return rc;
 220 }
 221
 222 static void nbd_co_receive_reply(BDRVNBDState *s, struct nbd_request *request,
 223                                  struct nbd_reply *reply,
 224                                  QEMUIOVector *qiov, int offset)
 225 {
 226     int ret;
 227
 228     /* Wait until we're woken up by the read handler.  TODO: perhaps
 229      * peek at the next reply and avoid yielding if it's ours?  */
 230     qemu_coroutine_yield();
 231     *reply = s->reply;
 232     if (reply->handle != request->handle) {
 233         reply->error = EIO;
 234     } else {
 235         if (qiov && reply->error == 0) {
 236             ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov,
 237                                 offset, request->len);
 238             if (ret != request->len) {
 239                 reply->error = EIO;
 240             }
 241         }
 242
 243         /* Tell the read handler to read another header.  */
 244         s->reply.handle = 0;
 245     }
 246 }
 247
 248 static void nbd_coroutine_end(BDRVNBDState *s, struct nbd_request *request)
 249 {
 250     int i = HANDLE_TO_INDEX(s, request->handle);
 251     s->recv_coroutine[i] = NULL;
 252     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 253         qemu_co_mutex_unlock(&s->free_sema);
 254     }
 255 }
 256
 257 static int nbd_establish_connection(BlockDriverState *bs)
 258 {
 259     BDRVNBDState *s = bs->opaque;
 260     int sock;
 261     int ret;
 262     off_t size;
 263     size_t blocksize;
 264
 265     if (s->host_spec[0] == '/') {
 266         sock = unix_socket_outgoing(s->host_spec);
 267     } else {
 268         sock = tcp_socket_outgoing_spec(s->host_spec);
 269     }
 270
 271     /* Failed to establish connection */
 272     if (sock < 0) {
 273         logout("Failed to establish connection to NBD server\n");
 274         return -errno;
 275     }
 276
 277     /* NBD handshake */
 278     ret = nbd_receive_negotiate(sock, s->export_name, &s->nbdflags, &size,
 279                                 &blocksize);
 280     if (ret < 0) {
 281         logout("Failed to negotiate with the NBD server\n");
 282         closesocket(sock);
 283         return ret;
 284     }
 285
 286     /* Now that we're connected, set the socket to be non-blocking and
 287      * kick the reply mechanism.  */
 288     socket_set_nonblock(sock);
 289     qemu_aio_set_fd_handler(sock, nbd_reply_ready, NULL,
 290                             nbd_have_request, s);
 291
 292     s->sock = sock;
 293     s->size = size;
 294     s->blocksize = blocksize;
 295
 296     logout("Established connection with NBD server\n");
 297     return 0;
 298 }
 299
 300 static void nbd_teardown_connection(BlockDriverState *bs)
 301 {
 302     BDRVNBDState *s = bs->opaque;
 303     struct nbd_request request;
 304
 305     request.type = NBD_CMD_DISC;
 306     request.from = 0;
 307     request.len = 0;
 308     nbd_send_request(s->sock, &request);
 309
 310     qemu_aio_set_fd_handler(s->sock, NULL, NULL, NULL, NULL);
 311     closesocket(s->sock);
 312 }
 313
 314 static int nbd_open(BlockDriverState *bs, const char* filename, int flags)
 315 {
 316     BDRVNBDState *s = bs->opaque;
 317     int result;
 318
 319     qemu_co_mutex_init(&s->send_mutex);
 320     qemu_co_mutex_init(&s->free_sema);
 321
 322     /* Pop the config into our state object. Exit if invalid. */
 323     result = nbd_config(s, filename, flags);
 324     if (result != 0) {
 325         return result;
 326     }
 327
 328     /* establish TCP connection, return error if it fails
 329      * TODO: Configurable retry-until-timeout behaviour.
 330      */
 331     result = nbd_establish_connection(bs);
 332
 333     return result;
 334 }
 335
 336 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 337                           int nb_sectors, QEMUIOVector *qiov,
 338                           int offset)
 339 {
 340     BDRVNBDState *s = bs->opaque;
 341     struct nbd_request request;
 342     struct nbd_reply reply;
 343     ssize_t ret;
 344
 345     request.type = NBD_CMD_READ;
 346     request.from = sector_num * 512;
 347     request.len = nb_sectors * 512;
 348
 349     nbd_coroutine_start(s, &request);
 350     ret = nbd_co_send_request(s, &request, NULL, 0);
 351     if (ret < 0) {
 352         reply.error = -ret;
 353     } else {
 354         nbd_co_receive_reply(s, &request, &reply, qiov, offset);
 355     }
 356     nbd_coroutine_end(s, &request);
 357     return -reply.error;
 358
 359 }
 360
 361 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 362                            int nb_sectors, QEMUIOVector *qiov,
 363                            int offset)
 364 {
 365     BDRVNBDState *s = bs->opaque;
 366     struct nbd_request request;
 367     struct nbd_reply reply;
 368     ssize_t ret;
 369
 370     request.type = NBD_CMD_WRITE;
 371     if (!bdrv_enable_write_cache(bs) && (s->nbdflags & NBD_FLAG_SEND_FUA)) {
 372         request.type |= NBD_CMD_FLAG_FUA;
 373     }
 374
 375     request.from = sector_num * 512;
 376     request.len = nb_sectors * 512;
 377
 378     nbd_coroutine_start(s, &request);
 379     ret = nbd_co_send_request(s, &request, qiov, offset);
 380     if (ret < 0) {
 381         reply.error = -ret;
 382     } else {
 383         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 384     }
 385     nbd_coroutine_end(s, &request);
 386     return -reply.error;
 387 }
 388
 389 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 390  * remain aligned to 4K. */
 391 #define NBD_MAX_SECTORS 2040
 392
 393 static int nbd_co_readv(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_readv_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_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 408 }
 409
 410 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
 411                          int nb_sectors, QEMUIOVector *qiov)
 412 {
 413     int offset = 0;
 414     int ret;
 415     while (nb_sectors > NBD_MAX_SECTORS) {
 416         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 417         if (ret < 0) {
 418             return ret;
 419         }
 420         offset += NBD_MAX_SECTORS * 512;
 421         sector_num += NBD_MAX_SECTORS;
 422         nb_sectors -= NBD_MAX_SECTORS;
 423     }
 424     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 425 }
 426
 427 static int nbd_co_flush(BlockDriverState *bs)
 428 {
 429     BDRVNBDState *s = bs->opaque;
 430     struct nbd_request request;
 431     struct nbd_reply reply;
 432     ssize_t ret;
 433
 434     if (!(s->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 435         return 0;
 436     }
 437
 438     request.type = NBD_CMD_FLUSH;
 439     if (s->nbdflags & NBD_FLAG_SEND_FUA) {
 440         request.type |= NBD_CMD_FLAG_FUA;
 441     }
 442
 443     request.from = 0;
 444     request.len = 0;
 445
 446     nbd_coroutine_start(s, &request);
 447     ret = nbd_co_send_request(s, &request, NULL, 0);
 448     if (ret < 0) {
 449         reply.error = -ret;
 450     } else {
 451         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 452     }
 453     nbd_coroutine_end(s, &request);
 454     return -reply.error;
 455 }
 456
 457 static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num,
 458                           int nb_sectors)
 459 {
 460     BDRVNBDState *s = bs->opaque;
 461     struct nbd_request request;
 462     struct nbd_reply reply;
 463     ssize_t ret;
 464
 465     if (!(s->nbdflags & NBD_FLAG_SEND_TRIM)) {
 466         return 0;
 467     }
 468     request.type = NBD_CMD_TRIM;
 469     request.from = sector_num * 512;;
 470     request.len = nb_sectors * 512;
 471
 472     nbd_coroutine_start(s, &request);
 473     ret = nbd_co_send_request(s, &request, NULL, 0);
 474     if (ret < 0) {
 475         reply.error = -ret;
 476     } else {
 477         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 478     }
 479     nbd_coroutine_end(s, &request);
 480     return -reply.error;
 481 }
 482
 483 static void nbd_close(BlockDriverState *bs)
 484 {
 485     BDRVNBDState *s = bs->opaque;
 486     g_free(s->export_name);
 487     g_free(s->host_spec);
 488
 489     nbd_teardown_connection(bs);
 490 }
 491
 492 static int64_t nbd_getlength(BlockDriverState *bs)
 493 {
 494     BDRVNBDState *s = bs->opaque;
 495
 496     return s->size;
 497 }
 498
 499 static BlockDriver bdrv_nbd = {
 500     .format_name         = "nbd",
 501     .instance_size       = sizeof(BDRVNBDState),
 502     .bdrv_file_open      = nbd_open,
 503     .bdrv_co_readv       = nbd_co_readv,
 504     .bdrv_co_writev      = nbd_co_writev,
 505     .bdrv_close          = nbd_close,
 506     .bdrv_co_flush_to_os = nbd_co_flush,
 507     .bdrv_co_discard     = nbd_co_discard,
 508     .bdrv_getlength      = nbd_getlength,
 509     .protocol_name       = "nbd",
 510 };
 511
 512 static void bdrv_nbd_init(void)
 513 {
 514     bdrv_register(&bdrv_nbd);
 515 }
 516
 517 block_init(bdrv_nbd_init);