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
  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     if (!client->ioc) { /* Already closed */
  51         return;
  52     }
  53
  54     /* finish any pending coroutines */
  55     qio_channel_shutdown(client->ioc,
  56                          QIO_CHANNEL_SHUTDOWN_BOTH,
  57                          NULL);
  58     nbd_recv_coroutines_enter_all(client);
  59
  60     nbd_client_detach_aio_context(bs);
  61     object_unref(OBJECT(client->sioc));
  62     client->sioc = NULL;
  63     object_unref(OBJECT(client->ioc));
  64     client->ioc = NULL;
  65 }
  66
  67 static void nbd_reply_ready(void *opaque)
  68 {
  69     BlockDriverState *bs = opaque;
  70     NbdClientSession *s = nbd_get_client_session(bs);
  71     uint64_t i;
  72     int ret;
  73
  74     if (!s->ioc) { /* Already closed */
  75         return;
  76     }
  77
  78     if (s->reply.handle == 0) {
  79         /* No reply already in flight.  Fetch a header.  It is possible
  80          * that another thread has done the same thing in parallel, so
  81          * the socket is not readable anymore.
  82          */
  83         ret = nbd_receive_reply(s->ioc, &s->reply);
  84         if (ret == -EAGAIN) {
  85             return;
  86         }
  87         if (ret < 0) {
  88             s->reply.handle = 0;
  89             goto fail;
  90         }
  91     }
  92
  93     /* There's no need for a mutex on the receive side, because the
  94      * handler acts as a synchronization point and ensures that only
  95      * one coroutine is called until the reply finishes.  */
  96     i = HANDLE_TO_INDEX(s, s->reply.handle);
  97     if (i >= MAX_NBD_REQUESTS) {
  98         goto fail;
  99     }
 100
 101     if (s->recv_coroutine[i]) {
 102         qemu_coroutine_enter(s->recv_coroutine[i], NULL);
 103         return;
 104     }
 105
 106 fail:
 107     nbd_teardown_connection(bs);
 108 }
 109
 110 static void nbd_restart_write(void *opaque)
 111 {
 112     BlockDriverState *bs = opaque;
 113
 114     qemu_coroutine_enter(nbd_get_client_session(bs)->send_coroutine, NULL);
 115 }
 116
 117 static int nbd_co_send_request(BlockDriverState *bs,
 118                                struct nbd_request *request,
 119                                QEMUIOVector *qiov, int offset)
 120 {
 121     NbdClientSession *s = nbd_get_client_session(bs);
 122     AioContext *aio_context;
 123     int rc, ret, i;
 124
 125     qemu_co_mutex_lock(&s->send_mutex);
 126
 127     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 128         if (s->recv_coroutine[i] == NULL) {
 129             s->recv_coroutine[i] = qemu_coroutine_self();
 130             break;
 131         }
 132     }
 133
 134     g_assert(qemu_in_coroutine());
 135     assert(i < MAX_NBD_REQUESTS);
 136     request->handle = INDEX_TO_HANDLE(s, i);
 137
 138     if (!s->ioc) {
 139         qemu_co_mutex_unlock(&s->send_mutex);
 140         return -EPIPE;
 141     }
 142
 143     s->send_coroutine = qemu_coroutine_self();
 144     aio_context = bdrv_get_aio_context(bs);
 145
 146     aio_set_fd_handler(aio_context, s->sioc->fd, false,
 147                        nbd_reply_ready, nbd_restart_write, bs);
 148     if (qiov) {
 149         qio_channel_set_cork(s->ioc, true);
 150         rc = nbd_send_request(s->ioc, request);
 151         if (rc >= 0) {
 152             ret = nbd_wr_syncv(s->ioc, qiov->iov, qiov->niov,
 153                                offset, request->len, 0);
 154             if (ret != request->len) {
 155                 rc = -EIO;
 156             }
 157         }
 158         qio_channel_set_cork(s->ioc, false);
 159     } else {
 160         rc = nbd_send_request(s->ioc, request);
 161     }
 162     aio_set_fd_handler(aio_context, s->sioc->fd, false,
 163                        nbd_reply_ready, NULL, bs);
 164     s->send_coroutine = NULL;
 165     qemu_co_mutex_unlock(&s->send_mutex);
 166     return rc;
 167 }
 168
 169 static void nbd_co_receive_reply(NbdClientSession *s,
 170     struct nbd_request *request, struct nbd_reply *reply,
 171     QEMUIOVector *qiov, int offset)
 172 {
 173     int ret;
 174
 175     /* Wait until we're woken up by the read handler.  TODO: perhaps
 176      * peek at the next reply and avoid yielding if it's ours?  */
 177     qemu_coroutine_yield();
 178     *reply = s->reply;
 179     if (reply->handle != request->handle ||
 180         !s->ioc) {
 181         reply->error = EIO;
 182     } else {
 183         if (qiov && reply->error == 0) {
 184             ret = nbd_wr_syncv(s->ioc, qiov->iov, qiov->niov,
 185                                offset, request->len, 1);
 186             if (ret != request->len) {
 187                 reply->error = EIO;
 188             }
 189         }
 190
 191         /* Tell the read handler to read another header.  */
 192         s->reply.handle = 0;
 193     }
 194 }
 195
 196 static void nbd_coroutine_start(NbdClientSession *s,
 197    struct nbd_request *request)
 198 {
 199     /* Poor man semaphore.  The free_sema is locked when no other request
 200      * can be accepted, and unlocked after receiving one reply.  */
 201     if (s->in_flight >= MAX_NBD_REQUESTS - 1) {
 202         qemu_co_mutex_lock(&s->free_sema);
 203         assert(s->in_flight < MAX_NBD_REQUESTS);
 204     }
 205     s->in_flight++;
 206
 207     /* s->recv_coroutine[i] is set as soon as we get the send_lock.  */
 208 }
 209
 210 static void nbd_coroutine_end(NbdClientSession *s,
 211     struct nbd_request *request)
 212 {
 213     int i = HANDLE_TO_INDEX(s, request->handle);
 214     s->recv_coroutine[i] = NULL;
 215     if (s->in_flight-- == MAX_NBD_REQUESTS) {
 216         qemu_co_mutex_unlock(&s->free_sema);
 217     }
 218 }
 219
 220 static int nbd_co_readv_1(BlockDriverState *bs, int64_t sector_num,
 221                           int nb_sectors, QEMUIOVector *qiov,
 222                           int offset)
 223 {
 224     NbdClientSession *client = nbd_get_client_session(bs);
 225     struct nbd_request request = { .type = NBD_CMD_READ };
 226     struct nbd_reply reply;
 227     ssize_t ret;
 228
 229     request.from = sector_num * 512;
 230     request.len = nb_sectors * 512;
 231
 232     nbd_coroutine_start(client, &request);
 233     ret = nbd_co_send_request(bs, &request, NULL, 0);
 234     if (ret < 0) {
 235         reply.error = -ret;
 236     } else {
 237         nbd_co_receive_reply(client, &request, &reply, qiov, offset);
 238     }
 239     nbd_coroutine_end(client, &request);
 240     return -reply.error;
 241
 242 }
 243
 244 static int nbd_co_writev_1(BlockDriverState *bs, int64_t sector_num,
 245                            int nb_sectors, QEMUIOVector *qiov,
 246                            int offset)
 247 {
 248     NbdClientSession *client = nbd_get_client_session(bs);
 249     struct nbd_request request = { .type = NBD_CMD_WRITE };
 250     struct nbd_reply reply;
 251     ssize_t ret;
 252
 253     if (!bdrv_enable_write_cache(bs) &&
 254         (client->nbdflags & NBD_FLAG_SEND_FUA)) {
 255         request.type |= NBD_CMD_FLAG_FUA;
 256     }
 257
 258     request.from = sector_num * 512;
 259     request.len = nb_sectors * 512;
 260
 261     nbd_coroutine_start(client, &request);
 262     ret = nbd_co_send_request(bs, &request, qiov, offset);
 263     if (ret < 0) {
 264         reply.error = -ret;
 265     } else {
 266         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 267     }
 268     nbd_coroutine_end(client, &request);
 269     return -reply.error;
 270 }
 271
 272 /* qemu-nbd has a limit of slightly less than 1M per request.  Try to
 273  * remain aligned to 4K. */
 274 #define NBD_MAX_SECTORS 2040
 275
 276 int nbd_client_co_readv(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_readv_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_readv_1(bs, sector_num, nb_sectors, qiov, offset);
 291 }
 292
 293 int nbd_client_co_writev(BlockDriverState *bs, int64_t sector_num,
 294                          int nb_sectors, QEMUIOVector *qiov)
 295 {
 296     int offset = 0;
 297     int ret;
 298     while (nb_sectors > NBD_MAX_SECTORS) {
 299         ret = nbd_co_writev_1(bs, sector_num, NBD_MAX_SECTORS, qiov, offset);
 300         if (ret < 0) {
 301             return ret;
 302         }
 303         offset += NBD_MAX_SECTORS * 512;
 304         sector_num += NBD_MAX_SECTORS;
 305         nb_sectors -= NBD_MAX_SECTORS;
 306     }
 307     return nbd_co_writev_1(bs, sector_num, nb_sectors, qiov, offset);
 308 }
 309
 310 int nbd_client_co_flush(BlockDriverState *bs)
 311 {
 312     NbdClientSession *client = nbd_get_client_session(bs);
 313     struct nbd_request request = { .type = NBD_CMD_FLUSH };
 314     struct nbd_reply reply;
 315     ssize_t ret;
 316
 317     if (!(client->nbdflags & NBD_FLAG_SEND_FLUSH)) {
 318         return 0;
 319     }
 320
 321     if (client->nbdflags & NBD_FLAG_SEND_FUA) {
 322         request.type |= NBD_CMD_FLAG_FUA;
 323     }
 324
 325     request.from = 0;
 326     request.len = 0;
 327
 328     nbd_coroutine_start(client, &request);
 329     ret = nbd_co_send_request(bs, &request, NULL, 0);
 330     if (ret < 0) {
 331         reply.error = -ret;
 332     } else {
 333         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 334     }
 335     nbd_coroutine_end(client, &request);
 336     return -reply.error;
 337 }
 338
 339 int nbd_client_co_discard(BlockDriverState *bs, int64_t sector_num,
 340                           int nb_sectors)
 341 {
 342     NbdClientSession *client = nbd_get_client_session(bs);
 343     struct nbd_request request = { .type = NBD_CMD_TRIM };
 344     struct nbd_reply reply;
 345     ssize_t ret;
 346
 347     if (!(client->nbdflags & NBD_FLAG_SEND_TRIM)) {
 348         return 0;
 349     }
 350     request.from = sector_num * 512;
 351     request.len = nb_sectors * 512;
 352
 353     nbd_coroutine_start(client, &request);
 354     ret = nbd_co_send_request(bs, &request, NULL, 0);
 355     if (ret < 0) {
 356         reply.error = -ret;
 357     } else {
 358         nbd_co_receive_reply(client, &request, &reply, NULL, 0);
 359     }
 360     nbd_coroutine_end(client, &request);
 361     return -reply.error;
 362
 363 }
 364
 365 void nbd_client_detach_aio_context(BlockDriverState *bs)
 366 {
 367     aio_set_fd_handler(bdrv_get_aio_context(bs),
 368                        nbd_get_client_session(bs)->sioc->fd,
 369                        false, NULL, NULL, NULL);
 370 }
 371
 372 void nbd_client_attach_aio_context(BlockDriverState *bs,
 373                                    AioContext *new_context)
 374 {
 375     aio_set_fd_handler(new_context, nbd_get_client_session(bs)->sioc->fd,
 376                        false, nbd_reply_ready, NULL, bs);
 377 }
 378
 379 void nbd_client_close(BlockDriverState *bs)
 380 {
 381     NbdClientSession *client = nbd_get_client_session(bs);
 382     struct nbd_request request = {
 383         .type = NBD_CMD_DISC,
 384         .from = 0,
 385         .len = 0
 386     };
 387
 388     if (client->ioc == NULL) {
 389         return;
 390     }
 391
 392     nbd_send_request(client->ioc, &request);
 393
 394     nbd_teardown_connection(bs);
 395 }
 396
 397 int nbd_client_init(BlockDriverState *bs,
 398                     QIOChannelSocket *sioc,
 399                     const char *export,
 400                     QCryptoTLSCreds *tlscreds,
 401                     const char *hostname,
 402                     Error **errp)
 403 {
 404     NbdClientSession *client = nbd_get_client_session(bs);
 405     int ret;
 406
 407     /* NBD handshake */
 408     logout("session init %s\n", export);
 409     qio_channel_set_blocking(QIO_CHANNEL(sioc), true, NULL);
 410
 411     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), export,
 412                                 &client->nbdflags,
 413                                 tlscreds, hostname,
 414                                 &client->ioc,
 415                                 &client->size, errp);
 416     if (ret < 0) {
 417         logout("Failed to negotiate with the NBD server\n");
 418         return ret;
 419     }
 420
 421     qemu_co_mutex_init(&client->send_mutex);
 422     qemu_co_mutex_init(&client->free_sema);
 423     client->sioc = sioc;
 424     object_ref(OBJECT(client->sioc));
 425
 426     if (!client->ioc) {
 427         client->ioc = QIO_CHANNEL(sioc);
 428         object_ref(OBJECT(client->ioc));
 429     }
 430
 431     /* Now that we're connected, set the socket to be non-blocking and
 432      * kick the reply mechanism.  */
 433     qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
 434
 435     nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
 436
 437     logout("Established connection with NBD server\n");
 438     return 0;
 439 }