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                                struct iovec *iov, 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 && iov) {
 209         ret = qemu_co_sendv(s->sock, iov, request->len, offset);
 210         if (ret != request->len) {
 211             return -EIO;
 212         }
 213     }
 214     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 215                             nbd_have_request, s);
 216     s->send_coroutine = NULL;
 217     qemu_co_mutex_unlock(&s->send_mutex);
 218     return rc;
 219 }
 220
 221 static void nbd_co_receive_reply(BDRVNBDState *s, struct nbd_request *request,
 222                                  struct nbd_reply *reply,
 223                                  struct iovec *iov, int offset)
 224 {
 225     int ret;
 226
 227     /* Wait until we're woken up by the read handler.  TODO: perhaps
 228      * peek at the next reply and avoid yielding if it's ours?  */
 229     qemu_coroutine_yield();
 230     *reply = s->reply;
 231     if (reply->handle != request->handle) {
 232         reply->error = EIO;
 233     } else {
 234         if (iov && reply->error == 0) {
 235             ret = qemu_co_recvv(s->sock, iov, request->len, offset);
 236             if (ret != request->len) {
 237                 reply->error = EIO;
 238             }
 239         }
 240
 241         /* Tell the read handler to read another header.  */
 242         s->reply.handle = 0;
 243     }
 244 }
 245
 246 static void nbd_coroutine_end(BDRVNBDState *s, struct nbd_request *request)
 247 {
 248     int i = HANDLE_TO_INDEX(s, request->handle);
 249     s->recv_coroutine[i] = NULL;
 250     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 251         qemu_co_mutex_unlock(&s->free_sema);
 252     }
 253 }
 254
 255 static int nbd_establish_connection(BlockDriverState *bs)
 256 {
 257     BDRVNBDState *s = bs->opaque;
 258     int sock;
 259     int ret;
 260     off_t size;
 261     size_t blocksize;
 262
 263     if (s->host_spec[0] == '/') {
 264         sock = unix_socket_outgoing(s->host_spec);
 265     } else {
 266         sock = tcp_socket_outgoing_spec(s->host_spec);
 267     }
 268
 269     /* Failed to establish connection */
 270     if (sock < 0) {
 271         logout("Failed to establish connection to NBD server\n");
 272         return -errno;
 273     }
 274
 275     /* NBD handshake */
 276     ret = nbd_receive_negotiate(sock, s->export_name, &s->nbdflags, &size,
 277                                 &blocksize);
 278     if (ret < 0) {
 279         logout("Failed to negotiate with the NBD server\n");
 280         closesocket(sock);
 281         return ret;
 282     }
 283
 284     /* Now that we're connected, set the socket to be non-blocking and
 285      * kick the reply mechanism.  */
 286     socket_set_nonblock(sock);
 287     qemu_aio_set_fd_handler(s->sock, nbd_reply_ready, NULL,
 288                             nbd_have_request, s);
 289
 290     s->sock = sock;
 291     s->size = size;
 292     s->blocksize = blocksize;
 293
 294     logout("Established connection with NBD server\n");
 295     return 0;
 296 }
 297
 298 static void nbd_teardown_connection(BlockDriverState *bs)
 299 {
 300     BDRVNBDState *s = bs->opaque;
 301     struct nbd_request request;
 302
 303     request.type = NBD_CMD_DISC;
 304     request.from = 0;
 305     request.len = 0;
 306     nbd_send_request(s->sock, &request);
 307
 308     qemu_aio_set_fd_handler(s->sock, NULL, NULL, NULL, NULL);
 309     closesocket(s->sock);
 310 }
 311
 312 static int nbd_open(BlockDriverState *bs, const char* filename, int flags)
 313 {
 314     BDRVNBDState *s = bs->opaque;
 315     int result;
 316
 317     qemu_co_mutex_init(&s->send_mutex);
 318     qemu_co_mutex_init(&s->free_sema);
 319
 320     /* Pop the config into our state object. Exit if invalid. */
 321     result = nbd_config(s, filename, flags);
 322     if (result != 0) {
 323         return result;
 324     }
 325
 326     /* establish TCP connection, return error if it fails
 327      * TODO: Configurable retry-until-timeout behaviour.
 328      */
 329     result = nbd_establish_connection(bs);
 330
 331     return result;
 332 }
 333
 334 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 335                           int nb_sectors, QEMUIOVector *qiov,
 336                           int offset)
 337 {
 338     BDRVNBDState *s = bs->opaque;
 339     struct nbd_request request;
 340     struct nbd_reply reply;
 341     ssize_t ret;
 342
 343     request.type = NBD_CMD_READ;
 344     request.from = sector_num * 512;
 345     request.len = nb_sectors * 512;
 346
 347     nbd_coroutine_start(s, &request);
 348     ret = nbd_co_send_request(s, &request, NULL, 0);
 349     if (ret < 0) {
 350         reply.error = -ret;
 351     } else {
 352         nbd_co_receive_reply(s, &request, &reply, qiov->iov, offset);
 353     }
 354     nbd_coroutine_end(s, &request);
 355     return -reply.error;
 356
 357 }
 358
 359 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 360                            int nb_sectors, QEMUIOVector *qiov,
 361                            int offset)
 362 {
 363     BDRVNBDState *s = bs->opaque;
 364     struct nbd_request request;
 365     struct nbd_reply reply;
 366     ssize_t ret;
 367
 368     request.type = NBD_CMD_WRITE;
 369     if (!bdrv_enable_write_cache(bs) && (s->nbdflags & NBD_FLAG_SEND_FUA)) {
 370         request.type |= NBD_CMD_FLAG_FUA;
 371     }
 372
 373     request.from = sector_num * 512;
 374     request.len = nb_sectors * 512;
 375
 376     nbd_coroutine_start(s, &request);
 377     ret = nbd_co_send_request(s, &request, qiov->iov, offset);
 378     if (ret < 0) {
 379         reply.error = -ret;
 380     } else {
 381         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 382     }
 383     nbd_coroutine_end(s, &request);
 384     return -reply.error;
 385 }
 386
 387 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 388  * remain aligned to 4K. */
 389 #define NBD_MAX_SECTORS 2040
 390
 391 static int nbd_co_readv(BlockDriverState *bs, int64_t sector_num,
 392                         int nb_sectors, QEMUIOVector *qiov)
 393 {
 394     int offset = 0;
 395     int ret;
 396     while (nb_sectors > NBD_MAX_SECTORS) {
 397         ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 398         if (ret < 0) {
 399             return ret;
 400         }
 401         offset += NBD_MAX_SECTORS * 512;
 402         sector_num += NBD_MAX_SECTORS;
 403         nb_sectors -= NBD_MAX_SECTORS;
 404     }
 405     return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 406 }
 407
 408 static int nbd_co_writev(BlockDriverState *bs, int64_t sector_num,
 409                          int nb_sectors, QEMUIOVector *qiov)
 410 {
 411     int offset = 0;
 412     int ret;
 413     while (nb_sectors > NBD_MAX_SECTORS) {
 414         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 415         if (ret < 0) {
 416             return ret;
 417         }
 418         offset += NBD_MAX_SECTORS * 512;
 419         sector_num += NBD_MAX_SECTORS;
 420         nb_sectors -= NBD_MAX_SECTORS;
 421     }
 422     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 423 }
 424
 425 static int nbd_co_flush(BlockDriverState *bs)
 426 {
 427     BDRVNBDState *s = bs->opaque;
 428     struct nbd_request request;
 429     struct nbd_reply reply;
 430     ssize_t ret;
 431
 432     if (!(s->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 433         return 0;
 434     }
 435
 436     request.type = NBD_CMD_FLUSH;
 437     if (s->nbdflags & NBD_FLAG_SEND_FUA) {
 438         request.type |= NBD_CMD_FLAG_FUA;
 439     }
 440
 441     request.from = 0;
 442     request.len = 0;
 443
 444     nbd_coroutine_start(s, &request);
 445     ret = nbd_co_send_request(s, &request, NULL, 0);
 446     if (ret < 0) {
 447         reply.error = -ret;
 448     } else {
 449         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 450     }
 451     nbd_coroutine_end(s, &request);
 452     return -reply.error;
 453 }
 454
 455 static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num,
 456                           int nb_sectors)
 457 {
 458     BDRVNBDState *s = bs->opaque;
 459     struct nbd_request request;
 460     struct nbd_reply reply;
 461     ssize_t ret;
 462
 463     if (!(s->nbdflags & NBD_FLAG_SEND_TRIM)) {
 464         return 0;
 465     }
 466     request.type = NBD_CMD_TRIM;
 467     request.from = sector_num * 512;;
 468     request.len = nb_sectors * 512;
 469
 470     nbd_coroutine_start(s, &request);
 471     ret = nbd_co_send_request(s, &request, NULL, 0);
 472     if (ret < 0) {
 473         reply.error = -ret;
 474     } else {
 475         nbd_co_receive_reply(s, &request, &reply, NULL, 0);
 476     }
 477     nbd_coroutine_end(s, &request);
 478     return -reply.error;
 479 }
 480
 481 static void nbd_close(BlockDriverState *bs)
 482 {
 483     BDRVNBDState *s = bs->opaque;
 484     g_free(s->export_name);
 485     g_free(s->host_spec);
 486
 487     nbd_teardown_connection(bs);
 488 }
 489
 490 static int64_t nbd_getlength(BlockDriverState *bs)
 491 {
 492     BDRVNBDState *s = bs->opaque;
 493
 494     return s->size;
 495 }
 496
 497 static BlockDriver bdrv_nbd = {
 498     .format_name         = "nbd",
 499     .instance_size       = sizeof(BDRVNBDState),
 500     .bdrv_file_open      = nbd_open,
 501     .bdrv_co_readv       = nbd_co_readv,
 502     .bdrv_co_writev      = nbd_co_writev,
 503     .bdrv_close          = nbd_close,
 504     .bdrv_co_flush_to_os = nbd_co_flush,
 505     .bdrv_co_discard     = nbd_co_discard,
 506     .bdrv_getlength      = nbd_getlength,
 507     .protocol_name       = "nbd",
 508 };
 509
 510 static void bdrv_nbd_init(void)
 511 {
 512     bdrv_register(&bdrv_nbd);
 513 }
 514
 515 block_init(bdrv_nbd_init);