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 "qemu/osdep.h"
  30 #include "nbd-client.h"
  31 #include "qemu/sockets.h"
  32
  33 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ ((uint64_t)(intptr_t)bs))
  34 #define INDEX_TO_HANDLE(bs, index)  ((index)  ^ ((uint64_t)(intptr_t)bs))
  35
  36 static void nbd_recv_coroutines_enter_all(NbdClientSession *s)
  37 {
  38     int i;
  39
  40     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
  41         if (s->recv_coroutine[i]) {
  42             qemu_coroutine_enter(s->recv_coroutine[i], NULL);
  43         }
  44     }
  45 }
  46
  47 static void nbd_teardown_connection(BlockDriverState *bs)
  48 {
  49     NbdClientSession *client = nbd_get_client_session(bs);
  50
  51     /* finish any pending coroutines */
  52     shutdown(client->sock, 2);
  53     nbd_recv_coroutines_enter_all(client);
  54
  55     nbd_client_detach_aio_context(bs);
  56     closesocket(client->sock);
  57     client->sock = -1;
  58 }
  59
  60 static void nbd_reply_ready(void *opaque)
  61 {
  62     BlockDriverState *bs = opaque;
  63     NbdClientSession *s = nbd_get_client_session(bs);
  64     uint64_t i;
  65     int ret;
  66
  67     if (s->reply.handle == 0) {
  68         /* No reply already in flight.  Fetch a header.  It is possible
  69          * that another thread has done the same thing in parallel, so
  70          * the socket is not readable anymore.
  71          */
  72         ret = nbd_receive_reply(s->sock, &s->reply);
  73         if (ret == -EAGAIN) {
  74             return;
  75         }
  76         if (ret < 0) {
  77             s->reply.handle = 0;
  78             goto fail;
  79         }
  80     }
  81
  82     /* There's no need for a mutex on the receive side, because the
  83      * handler acts as a synchronization point and ensures that only
  84      * one coroutine is called until the reply finishes.  */
  85     i = HANDLE_TO_INDEX(s, s->reply.handle);
  86     if (i >= MAX_NBD_REQUESTS) {
  87         goto fail;
  88     }
  89
  90     if (s->recv_coroutine[i]) {
  91         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
  92         return;
  93     }
  94
  95 fail:
  96     nbd_teardown_connection(bs);
  97 }
  98
  99 static void nbd_restart_write(void *opaque)
 100 {
 101     BlockDriverState *bs = opaque;
 102
 103     qemu_coroutine_enter(nbd_get_client_session(bs)->send_coroutine, NULL);
 104 }
 105
 106 static int nbd_co_send_request(BlockDriverState *bs,
 107                                struct nbd_request *request,
 108                                QEMUIOVector *qiov, int offset)
 109 {
 110     NbdClientSession *s = nbd_get_client_session(bs);
 111     AioContext *aio_context;
 112     int rc, ret, i;
 113
 114     qemu_co_mutex_lock(&s->send_mutex);
 115
 116     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 117         if (s->recv_coroutine[i] == NULL) {
 118             s->recv_coroutine[i] = qemu_coroutine_self();
 119             break;
 120         }
 121     }
 122
 123     assert(i < MAX_NBD_REQUESTS);
 124     request->handle = INDEX_TO_HANDLE(s, i);
 125     s->send_coroutine = qemu_coroutine_self();
 126     aio_context = bdrv_get_aio_context(bs);
 127
 128     aio_set_fd_handler(aio_context, s->sock, false,
 129                        nbd_reply_ready, nbd_restart_write, bs);
 130     if (qiov) {
 131         if (!s->is_unix) {
 132             socket_set_cork(s->sock, 1);
 133         }
 134         rc = nbd_send_request(s->sock, request);
 135         if (rc >= 0) {
 136             ret = qemu_co_sendv(s->sock, qiov->iov, qiov->niov,
 137                                 offset, request->len);
 138             if (ret != request->len) {
 139                 rc = -EIO;
 140             }
 141         }
 142         if (!s->is_unix) {
 143             socket_set_cork(s->sock, 0);
 144         }
 145     } else {
 146         rc = nbd_send_request(s->sock, request);
 147     }
 148     aio_set_fd_handler(aio_context, s->sock, false,
 149                        nbd_reply_ready, NULL, bs);
 150     s->send_coroutine = NULL;
 151     qemu_co_mutex_unlock(&s->send_mutex);
 152     return rc;
 153 }
 154
 155 static void nbd_co_receive_reply(NbdClientSession *s,
 156     struct nbd_request *request, struct nbd_reply *reply,
 157     QEMUIOVector *qiov, int offset)
 158 {
 159     int ret;
 160
 161     /* Wait until we're woken up by the read handler.  TODO: perhaps
 162      * peek at the next reply and avoid yielding if it's ours?  */
 163     qemu_coroutine_yield();
 164     *reply = s->reply;
 165     if (reply->handle != request->handle) {
 166         reply->error = EIO;
 167     } else {
 168         if (qiov && reply->error == 0) {
 169             ret = qemu_co_recvv(s->sock, qiov->iov, qiov->niov,
 170                                 offset, request->len);
 171             if (ret != request->len) {
 172                 reply->error = EIO;
 173             }
 174         }
 175
 176         /* Tell the read handler to read another header.  */
 177         s->reply.handle = 0;
 178     }
 179 }
 180
 181 static void nbd_coroutine_start(NbdClientSession *s,
 182    struct nbd_request *request)
 183 {
 184     /* Poor man semaphore.  The free_sema is locked when no other request
 185      * can be accepted, and unlocked after receiving one reply.  */
 186     if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
 187         qemu_co_mutex_lock(&s->free_sema);
 188         assert(s->in_flight < MAX_NBD_REQUESTS);
 189     }
 190     s->in_flight++;
 191
 192     /* s->recv_coroutine[i] is set as soon as we get the send_lock.  */
 193 }
 194
 195 static void nbd_coroutine_end(NbdClientSession *s,
 196     struct nbd_request *request)
 197 {
 198     int i = HANDLE_TO_INDEX(s, request->handle);
 199     s->recv_coroutine[i] = NULL;
 200     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 201         qemu_co_mutex_unlock(&s->free_sema);
 202     }
 203 }
 204
 205 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 206                           int nb_sectors, QEMUIOVector *qiov,
 207                           int offset)
 208 {
 209     NbdClientSession *client = nbd_get_client_session(bs);
 210     struct nbd_request request = { .type = NBD_CMD_READ };
 211     struct nbd_reply reply;
 212     ssize_t ret;
 213
 214     request.from = sector_num * 512;
 215     request.len = nb_sectors * 512;
 216
 217     nbd_coroutine_start(client, &request);
 218     ret = nbd_co_send_request(bs, &request, NULL, 0);
 219     if (ret < 0) {
 220         reply.error = -ret;
 221     } else {
 222         nbd_co_receive_reply(client, &request, &reply, qiov, offset);
 223     }
 224     nbd_coroutine_end(client, &request);
 225     return -reply.error;
 226
 227 }
 228
 229 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 230                            int nb_sectors, QEMUIOVector *qiov,
 231                            int offset)
 232 {
 233     NbdClientSession *client = nbd_get_client_session(bs);
 234     struct nbd_request request = { .type = NBD_CMD_WRITE };
 235     struct nbd_reply reply;
 236     ssize_t ret;
 237
 238     if (!bdrv_enable_write_cache(bs) &&
 239         (client->nbdflags & NBD_FLAG_SEND_FUA)) {
 240         request.type |= NBD_CMD_FLAG_FUA;
 241     }
 242
 243     request.from = sector_num * 512;
 244     request.len = nb_sectors * 512;
 245
 246     nbd_coroutine_start(client, &request);
 247     ret = nbd_co_send_request(bs, &request, qiov, offset);
 248     if (ret < 0) {
 249         reply.error = -ret;
 250     } else {
 251         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 252     }
 253     nbd_coroutine_end(client, &request);
 254     return -reply.error;
 255 }
 256
 257 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 258  * remain aligned to 4K. */
 259 #define NBD_MAX_SECTORS 2040
 260
 261 int nbd_client_co_readv(BlockDriverState *bs, int64_t sector_num,
 262                         int nb_sectors, QEMUIOVector *qiov)
 263 {
 264     int offset = 0;
 265     int ret;
 266     while (nb_sectors > NBD_MAX_SECTORS) {
 267         ret = nbd_co_readv_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 268         if (ret < 0) {
 269             return ret;
 270         }
 271         offset += NBD_MAX_SECTORS * 512;
 272         sector_num += NBD_MAX_SECTORS;
 273         nb_sectors -= NBD_MAX_SECTORS;
 274     }
 275     return nbd_co_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 276 }
 277
 278 int nbd_client_co_writev(BlockDriverState *bs, int64_t sector_num,
 279                          int nb_sectors, QEMUIOVector *qiov)
 280 {
 281     int offset = 0;
 282     int ret;
 283     while (nb_sectors > NBD_MAX_SECTORS) {
 284         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 285         if (ret < 0) {
 286             return ret;
 287         }
 288         offset += NBD_MAX_SECTORS * 512;
 289         sector_num += NBD_MAX_SECTORS;
 290         nb_sectors -= NBD_MAX_SECTORS;
 291     }
 292     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 293 }
 294
 295 int nbd_client_co_flush(BlockDriverState *bs)
 296 {
 297     NbdClientSession *client = nbd_get_client_session(bs);
 298     struct nbd_request request = { .type = NBD_CMD_FLUSH };
 299     struct nbd_reply reply;
 300     ssize_t ret;
 301
 302     if (!(client->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 303         return 0;
 304     }
 305
 306     if (client->nbdflags & NBD_FLAG_SEND_FUA) {
 307         request.type |= NBD_CMD_FLAG_FUA;
 308     }
 309
 310     request.from = 0;
 311     request.len = 0;
 312
 313     nbd_coroutine_start(client, &request);
 314     ret = nbd_co_send_request(bs, &request, NULL, 0);
 315     if (ret < 0) {
 316         reply.error = -ret;
 317     } else {
 318         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 319     }
 320     nbd_coroutine_end(client, &request);
 321     return -reply.error;
 322 }
 323
 324 int nbd_client_co_discard(BlockDriverState *bs, int64_t sector_num,
 325                           int nb_sectors)
 326 {
 327     NbdClientSession *client = nbd_get_client_session(bs);
 328     struct nbd_request request = { .type = NBD_CMD_TRIM };
 329     struct nbd_reply reply;
 330     ssize_t ret;
 331
 332     if (!(client->nbdflags & NBD_FLAG_SEND_TRIM)) {
 333         return 0;
 334     }
 335     request.from = sector_num * 512;
 336     request.len = nb_sectors * 512;
 337
 338     nbd_coroutine_start(client, &request);
 339     ret = nbd_co_send_request(bs, &request, NULL, 0);
 340     if (ret < 0) {
 341         reply.error = -ret;
 342     } else {
 343         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 344     }
 345     nbd_coroutine_end(client, &request);
 346     return -reply.error;
 347
 348 }
 349
 350 void nbd_client_detach_aio_context(BlockDriverState *bs)
 351 {
 352     aio_set_fd_handler(bdrv_get_aio_context(bs),
 353                        nbd_get_client_session(bs)->sock,
 354                        false, NULL, NULL, NULL);
 355 }
 356
 357 void nbd_client_attach_aio_context(BlockDriverState *bs,
 358                                    AioContext *new_context)
 359 {
 360     aio_set_fd_handler(new_context, nbd_get_client_session(bs)->sock,
 361                        false, nbd_reply_ready, NULL, bs);
 362 }
 363
 364 void nbd_client_close(BlockDriverState *bs)
 365 {
 366     NbdClientSession *client = nbd_get_client_session(bs);
 367     struct nbd_request request = {
 368         .type = NBD_CMD_DISC,
 369         .from = 0,
 370         .len = 0
 371     };
 372
 373     if (client->sock == -1) {
 374         return;
 375     }
 376
 377     nbd_send_request(client->sock, &request);
 378
 379     nbd_teardown_connection(bs);
 380 }
 381
 382 int nbd_client_init(BlockDriverState *bs, int sock, const char *export,
 383                     Error **errp)
 384 {
 385     NbdClientSession *client = nbd_get_client_session(bs);
 386     int ret;
 387
 388     /* NBD handshake */
 389     logout("session init %s\n", export);
 390     qemu_set_block(sock);
 391     ret = nbd_receive_negotiate(sock, export,
 392                                 &client->nbdflags, &client->size, errp);
 393     if (ret < 0) {
 394         logout("Failed to negotiate with the NBD server\n");
 395         closesocket(sock);
 396         return ret;
 397     }
 398
 399     qemu_co_mutex_init(&client->send_mutex);
 400     qemu_co_mutex_init(&client->free_sema);
 401     client->sock = sock;
 402
 403     /* Now that we're connected, set the socket to be non-blocking and
 404      * kick the reply mechanism.  */
 405     qemu_set_nonblock(sock);
 406     nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
 407
 408     logout("Established connection with NBD server\n");
 409     return 0;
 410 }