block/nbd-client.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 "nbd-client.h"
  30 #include "qemu/sockets.h"
  31
  32 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
  33 #define INDEX_TO_HANDLE(bs, index)  ((index)  ^ ((uint64_t)(intptr_t)bs))
  34
  35 static void nbd_recv_coroutines_enter_all(NbdClientSession *s)
  36 {
  37     int i;
  38
  39     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
  40         if (s->recv_coroutine[i]) {
  41             qemu_coroutine_enter(s->recv_coroutine[i], NULL);
  42         }
  43     }
  44 }
  45
  46 static void nbd_teardown_connection(BlockDriverState *bs)
  47 {
  48     NbdClientSession *client = nbd_get_client_session(bs);
  49
  50     /* finish any pending coroutines */
  51     shutdown(client->sock, 2);
  52     nbd_recv_coroutines_enter_all(client);
  53
  54     nbd_client_detach_aio_context(bs);
  55     closesocket(client->sock);
  56     client->sock = -1;
  57 }
  58
  59 static void nbd_reply_ready(void *opaque)
  60 {
  61     BlockDriverState *bs = opaque;
  62     NbdClientSession *s = nbd_get_client_session(bs);
  63     uint64_t i;
  64     int ret;
  65
  66     if (s->reply.handle == 0) {
  67         /* No reply already in flight.  Fetch a header.  It is possible
  68          * that another thread has done the same thing in parallel, so
  69          * the socket is not readable anymore.
  70          */
  71         ret = nbd_receive_reply(s->sock, &s->reply);
  72         if (ret == -EAGAIN) {
  73             return;
  74         }
  75         if (ret < 0) {
  76             s->reply.handle = 0;
  77             goto fail;
  78         }
  79     }
  80
  81     /* There's no need for a mutex on the receive side, because the
  82      * handler acts as a synchronization point and ensures that only
  83      * one coroutine is called until the reply finishes.  */
  84     i = HANDLE_TO_INDEX(s, s->reply.handle);
  85     if (i >= MAX_NBD_REQUESTS) {
  86         goto fail;
  87     }
  88
  89     if (s->recv_coroutine[i]) {
  90         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
  91         return;
  92     }
  93
  94 fail:
  95     nbd_teardown_connection(bs);
  96 }
  97
  98 static void nbd_restart_write(void *opaque)
  99 {
 100     BlockDriverState *bs = opaque;
 101
 102     qemu_coroutine_enter(nbd_get_client_session(bs)->send_coroutine, NULL);
 103 }
 104
 105 static int nbd_co_send_request(BlockDriverState *bs,
 106                                struct nbd_request *request,
 107                                QEMUIOVector *qiov, int offset)
 108 {
 109     NbdClientSession *s = nbd_get_client_session(bs);
 110     AioContext *aio_context;
 111     int rc, ret, i;
 112
 113     qemu_co_mutex_lock(&s->send_mutex);
 114
 115     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 116         if (s->recv_coroutine[i] == NULL) {
 117             s->recv_coroutine[i] = qemu_coroutine_self();
 118             break;
 119         }
 120     }
 121
 122     assert(i < MAX_NBD_REQUESTS);
 123     request->handle = INDEX_TO_HANDLE(s, i);
 124     s->send_coroutine = qemu_coroutine_self();
 125     aio_context = bdrv_get_aio_context(bs);
 126
 127     aio_set_fd_handler(aio_context, s->sock,
 128                        nbd_reply_ready, nbd_restart_write, bs);
 129     if (qiov) {
 130         if (!s->is_unix) {
 131             socket_set_cork(s->sock, 1);
 132         }
 133         rc = nbd_send_request(s->sock, request);
 134         if (rc >= 0) {
 135             ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov,
 136                                 offset, request->len);
 137             if (ret != request->len) {
 138                 rc = -EIO;
 139             }
 140         }
 141         if (!s->is_unix) {
 142             socket_set_cork(s->sock, 0);
 143         }
 144     } else {
 145         rc = nbd_send_request(s->sock, request);
 146     }
 147     aio_set_fd_handler(aio_context, s->sock, nbd_reply_ready, NULL, bs);
 148     s->send_coroutine = NULL;
 149     qemu_co_mutex_unlock(&s->send_mutex);
 150     return rc;
 151 }
 152
 153 static void nbd_co_receive_reply(NbdClientSession *s,
 154     struct nbd_request *request, struct nbd_reply *reply,
 155     QEMUIOVector *qiov, int offset)
 156 {
 157     int ret;
 158
 159     /* Wait until we're woken up by the read handler.  TODO: perhaps
 160      * peek at the next reply and avoid yielding if it's ours?  */
 161     qemu_coroutine_yield();
 162     *reply = s->reply;
 163     if (reply->handle != request->handle) {
 164         reply->error = EIO;
 165     } else {
 166         if (qiov && reply->error == 0) {
 167             ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov,
 168                                 offset, request->len);
 169             if (ret != request->len) {
 170                 reply->error = EIO;
 171             }
 172         }
 173
 174         /* Tell the read handler to read another header.  */
 175         s->reply.handle = 0;
 176     }
 177 }
 178
 179 static void nbd_coroutine_start(NbdClientSession *s,
 180    struct nbd_request *request)
 181 {
 182     /* Poor man semaphore.  The free_sema is locked when no other request
 183      * can be accepted, and unlocked after receiving one reply.  */
 184     if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
 185         qemu_co_mutex_lock(&s->free_sema);
 186         assert(s->in_flight < MAX_NBD_REQUESTS);
 187     }
 188     s->in_flight++;
 189
 190     /* s->recv_coroutine[i] is set as soon as we get the send_lock.  */
 191 }
 192
 193 static void nbd_coroutine_end(NbdClientSession *s,
 194     struct nbd_request *request)
 195 {
 196     int i = HANDLE_TO_INDEX(s, request->handle);
 197     s->recv_coroutine[i] = NULL;
 198     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 199         qemu_co_mutex_unlock(&s->free_sema);
 200     }
 201 }
 202
 203 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 204                           int nb_sectors, QEMUIOVector *qiov,
 205                           int offset)
 206 {
 207     NbdClientSession *client = nbd_get_client_session(bs);
 208     struct nbd_request request = { .type = NBD_CMD_READ };
 209     struct nbd_reply reply;
 210     ssize_t ret;
 211
 212     request.from = sector_num * 512;
 213     request.len = nb_sectors * 512;
 214
 215     nbd_coroutine_start(client, &request);
 216     ret = nbd_co_send_request(bs, &request, NULL, 0);
 217     if (ret < 0) {
 218         reply.error = -ret;
 219     } else {
 220         nbd_co_receive_reply(client, &request, &reply, qiov, offset);
 221     }
 222     nbd_coroutine_end(client, &request);
 223     return -reply.error;
 224
 225 }
 226
 227 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 228                            int nb_sectors, QEMUIOVector *qiov,
 229                            int offset)
 230 {
 231     NbdClientSession *client = nbd_get_client_session(bs);
 232     struct nbd_request request = { .type = NBD_CMD_WRITE };
 233     struct nbd_reply reply;
 234     ssize_t ret;
 235
 236     if (!bdrv_enable_write_cache(bs) &&
 237         (client->nbdflags & NBD_FLAG_SEND_FUA)) {
 238         request.type |= NBD_CMD_FLAG_FUA;
 239     }
 240
 241     request.from = sector_num * 512;
 242     request.len = nb_sectors * 512;
 243
 244     nbd_coroutine_start(client, &request);
 245     ret = nbd_co_send_request(bs, &request, qiov, offset);
 246     if (ret < 0) {
 247         reply.error = -ret;
 248     } else {
 249         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 250     }
 251     nbd_coroutine_end(client, &request);
 252     return -reply.error;
 253 }
 254
 255 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 256  * remain aligned to 4K. */
 257 #define NBD_MAX_SECTORS 2040
 258
 259 int nbd_client_co_readv(BlockDriverState *bs, int64_t sector_num,
 260                         int nb_sectors, QEMUIOVector *qiov)
 261 {
 262     int offset = 0;
 263     int ret;
 264     while (nb_sectors > NBD_MAX_SECTORS) {
 265         ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 266         if (ret < 0) {
 267             return ret;
 268         }
 269         offset += NBD_MAX_SECTORS * 512;
 270         sector_num += NBD_MAX_SECTORS;
 271         nb_sectors -= NBD_MAX_SECTORS;
 272     }
 273     return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 274 }
 275
 276 int nbd_client_co_writev(BlockDriverState *bs, int64_t sector_num,
 277                          int nb_sectors, QEMUIOVector *qiov)
 278 {
 279     int offset = 0;
 280     int ret;
 281     while (nb_sectors > NBD_MAX_SECTORS) {
 282         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 283         if (ret < 0) {
 284             return ret;
 285         }
 286         offset += NBD_MAX_SECTORS * 512;
 287         sector_num += NBD_MAX_SECTORS;
 288         nb_sectors -= NBD_MAX_SECTORS;
 289     }
 290     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 291 }
 292
 293 int nbd_client_co_flush(BlockDriverState *bs)
 294 {
 295     NbdClientSession *client = nbd_get_client_session(bs);
 296     struct nbd_request request = { .type = NBD_CMD_FLUSH };
 297     struct nbd_reply reply;
 298     ssize_t ret;
 299
 300     if (!(client->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 301         return 0;
 302     }
 303
 304     if (client->nbdflags & NBD_FLAG_SEND_FUA) {
 305         request.type |= NBD_CMD_FLAG_FUA;
 306     }
 307
 308     request.from = 0;
 309     request.len = 0;
 310
 311     nbd_coroutine_start(client, &request);
 312     ret = nbd_co_send_request(bs, &request, NULL, 0);
 313     if (ret < 0) {
 314         reply.error = -ret;
 315     } else {
 316         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 317     }
 318     nbd_coroutine_end(client, &request);
 319     return -reply.error;
 320 }
 321
 322 int nbd_client_co_discard(BlockDriverState *bs, int64_t sector_num,
 323                           int nb_sectors)
 324 {
 325     NbdClientSession *client = nbd_get_client_session(bs);
 326     struct nbd_request request = { .type = NBD_CMD_TRIM };
 327     struct nbd_reply reply;
 328     ssize_t ret;
 329
 330     if (!(client->nbdflags & NBD_FLAG_SEND_TRIM)) {
 331         return 0;
 332     }
 333     request.from = sector_num * 512;
 334     request.len = nb_sectors * 512;
 335
 336     nbd_coroutine_start(client, &request);
 337     ret = nbd_co_send_request(bs, &request, NULL, 0);
 338     if (ret < 0) {
 339         reply.error = -ret;
 340     } else {
 341         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 342     }
 343     nbd_coroutine_end(client, &request);
 344     return -reply.error;
 345
 346 }
 347
 348 void nbd_client_detach_aio_context(BlockDriverState *bs)
 349 {
 350     aio_set_fd_handler(bdrv_get_aio_context(bs),
 351                        nbd_get_client_session(bs)->sock, NULL, NULL, NULL);
 352 }
 353
 354 void nbd_client_attach_aio_context(BlockDriverState *bs,
 355                                    AioContext *new_context)
 356 {
 357     aio_set_fd_handler(new_context, nbd_get_client_session(bs)->sock,
 358                        nbd_reply_ready, NULL, bs);
 359 }
 360
 361 void nbd_client_close(BlockDriverState *bs)
 362 {
 363     NbdClientSession *client = nbd_get_client_session(bs);
 364     struct nbd_request request = {
 365         .type = NBD_CMD_DISC,
 366         .from = 0,
 367         .len = 0
 368     };
 369
 370     if (client->sock == -1) {
 371         return;
 372     }
 373
 374     nbd_send_request(client->sock, &request);
 375
 376     nbd_teardown_connection(bs);
 377 }
 378
 379 int nbd_client_init(BlockDriverState *bs, int sock, const char *export,
 380                     Error **errp)
 381 {
 382     NbdClientSession *client = nbd_get_client_session(bs);
 383     int ret;
 384
 385     /* NBD handshake */
 386     logout("session init %s\n", export);
 387     qemu_set_block(sock);
 388     ret = nbd_receive_negotiate(sock, export,
 389                                 &client->nbdflags, &client->size,
 390                                 &client->blocksize, errp);
 391     if (ret < 0) {
 392         logout("Failed to negotiate with the NBD server\n");
 393         closesocket(sock);
 394         return ret;
 395     }
 396
 397     qemu_co_mutex_init(&client->send_mutex);
 398     qemu_co_mutex_init(&client->free_sema);
 399     client->sock = sock;
 400
 401     /* Now that we're connected, set the socket to be non-blocking and
 402      * kick the reply mechanism.  */
 403     qemu_set_nonblock(sock);
 404     nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
 405
 406     logout("Established connection with NBD server\n");
 407     return 0;
 408 }