block/nbd-client.c

   1 /*
   2  * QEMU Block driver for  NBD
   3  *
   4  * Copyright (C) 2016 Red Hat, Inc.
   5  * Copyright (C) 2008 Bull S.A.S.
   6  *     Author: Laurent Vivier <Laurent.Vivier@bull.net>
   7  *
   8  * Some parts:
   9  *    Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws>
  10  *
  11  * Permission is hereby granted, free of charge, to any person obtaining a copy
  12  * of this software and associated documentation files (the "Software"), to deal
  13  * in the Software without restriction, including without limitation the rights
  14  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  15  * copies of the Software, and to permit persons to whom the Software is
  16  * furnished to do so, subject to the following conditions:
  17  *
  18  * The above copyright notice and this permission notice shall be included in
  19  * all copies or substantial portions of the Software.
  20  *
  21  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  22  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  23  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  24  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  25  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  26  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  27  * THE SOFTWARE.
  28  */
  29
  30 #include "qemu/osdep.h"
  31 #include "qapi/error.h"
  32 #include "nbd-client.h"
  33
  34 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs))
  35 #define INDEX_TO_HANDLE(bs, index)  ((index)  ^ (uint64_t)(intptr_t)(bs))
  36
  37 static void nbd_recv_coroutines_wake_all(NBDClientSession *s)
  38 {
  39     int i;
  40
  41     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
  42         NBDClientRequest *req = &s->requests[i];
  43
  44         if (req->coroutine && req->receiving) {
  45             aio_co_wake(req->coroutine);
  46         }
  47     }
  48 }
  49
  50 static void nbd_teardown_connection(BlockDriverState *bs)
  51 {
  52     NBDClientSession *client = nbd_get_client_session(bs);
  53
  54     if (!client->ioc) { /* Already closed */
  55         return;
  56     }
  57
  58     /* finish any pending coroutines */
  59     qio_channel_shutdown(client->ioc,
  60                          QIO_CHANNEL_SHUTDOWN_BOTH,
  61                          NULL);
  62     BDRV_POLL_WHILE(bs, client->read_reply_co);
  63
  64     nbd_client_detach_aio_context(bs);
  65     object_unref(OBJECT(client->sioc));
  66     client->sioc = NULL;
  67     object_unref(OBJECT(client->ioc));
  68     client->ioc = NULL;
  69 }
  70
  71 static coroutine_fn void nbd_read_reply_entry(void *opaque)
  72 {
  73     NBDClientSession *s = opaque;
  74     uint64_t i;
  75     int ret = 0;
  76     Error *local_err = NULL;
  77
  78     while (!s->quit) {
  79         assert(s->reply.handle == 0);
  80         ret = nbd_receive_reply(s->ioc, &s->reply, &local_err);
  81         if (ret < 0) {
  82             error_report_err(local_err);
  83         }
  84         if (ret <= 0) {
  85             break;
  86         }
  87
  88         /* There's no need for a mutex on the receive side, because the
  89          * handler acts as a synchronization point and ensures that only
  90          * one coroutine is called until the reply finishes.
  91          */
  92         i = HANDLE_TO_INDEX(s, s->reply.handle);
  93         if (i >= MAX_NBD_REQUESTS ||
  94             !s->requests[i].coroutine ||
  95             !s->requests[i].receiving) {
  96             break;
  97         }
  98
  99         /* We're woken up again by the request itself.  Note that there
 100          * is no race between yielding and reentering read_reply_co.  This
 101          * is because:
 102          *
 103          * - if the request runs on the same AioContext, it is only
 104          *   entered after we yield
 105          *
 106          * - if the request runs on a different AioContext, reentering
 107          *   read_reply_co happens through a bottom half, which can only
 108          *   run after we yield.
 109          */
 110         aio_co_wake(s->requests[i].coroutine);
 111         qemu_coroutine_yield();
 112     }
 113
 114     s->quit = true;
 115     nbd_recv_coroutines_wake_all(s);
 116     s->read_reply_co = NULL;
 117 }
 118
 119 static int nbd_co_send_request(BlockDriverState *bs,
 120                                NBDRequest *request,
 121                                QEMUIOVector *qiov)
 122 {
 123     NBDClientSession *s = nbd_get_client_session(bs);
 124     int rc, i;
 125
 126     qemu_co_mutex_lock(&s->send_mutex);
 127     while (s->in_flight == MAX_NBD_REQUESTS) {
 128         qemu_co_queue_wait(&s->free_sema, &s->send_mutex);
 129     }
 130     s->in_flight++;
 131
 132     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 133         if (s->requests[i].coroutine == NULL) {
 134             break;
 135         }
 136     }
 137
 138     g_assert(qemu_in_coroutine());
 139     assert(i < MAX_NBD_REQUESTS);
 140
 141     s->requests[i].coroutine = qemu_coroutine_self();
 142     s->requests[i].receiving = false;
 143
 144     request->handle = INDEX_TO_HANDLE(s, i);
 145
 146     if (s->quit) {
 147         rc = -EIO;
 148         goto err;
 149     }
 150     if (!s->ioc) {
 151         rc = -EPIPE;
 152         goto err;
 153     }
 154
 155     if (qiov) {
 156         qio_channel_set_cork(s->ioc, true);
 157         rc = nbd_send_request(s->ioc, request);
 158         if (rc >= 0 && !s->quit) {
 159             assert(request->len == iov_size(qiov->iov, qiov->niov));
 160             if (qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov,
 161                                        NULL) < 0) {
 162                 rc = -EIO;
 163             }
 164         } else if (rc >= 0) {
 165             rc = -EIO;
 166         }
 167         qio_channel_set_cork(s->ioc, false);
 168     } else {
 169         rc = nbd_send_request(s->ioc, request);
 170     }
 171
 172 err:
 173     if (rc < 0) {
 174         s->quit = true;
 175         s->requests[i].coroutine = NULL;
 176         s->in_flight--;
 177         qemu_co_queue_next(&s->free_sema);
 178     }
 179     qemu_co_mutex_unlock(&s->send_mutex);
 180     return rc;
 181 }
 182
 183 static int nbd_co_receive_reply(NBDClientSession *s,
 184                                 NBDRequest *request,
 185                                 QEMUIOVector *qiov)
 186 {
 187     int ret;
 188     int i = HANDLE_TO_INDEX(s, request->handle);
 189
 190     /* Wait until we're woken up by nbd_read_reply_entry.  */
 191     s->requests[i].receiving = true;
 192     qemu_coroutine_yield();
 193     s->requests[i].receiving = false;
 194     if (!s->ioc || s->quit) {
 195         ret = -EIO;
 196     } else {
 197         assert(s->reply.handle == request->handle);
 198         ret = -s->reply.error;
 199         if (qiov && s->reply.error == 0) {
 200             assert(request->len == iov_size(qiov->iov, qiov->niov));
 201             if (qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
 202                                       NULL) < 0) {
 203                 ret = -EIO;
 204                 s->quit = true;
 205             }
 206         }
 207
 208         /* Tell the read handler to read another header.  */
 209         s->reply.handle = 0;
 210     }
 211
 212     s->requests[i].coroutine = NULL;
 213
 214     /* Kick the read_reply_co to get the next reply.  */
 215     if (s->read_reply_co) {
 216         aio_co_wake(s->read_reply_co);
 217     }
 218
 219     qemu_co_mutex_lock(&s->send_mutex);
 220     s->in_flight--;
 221     qemu_co_queue_next(&s->free_sema);
 222     qemu_co_mutex_unlock(&s->send_mutex);
 223
 224     return ret;
 225 }
 226
 227 static int nbd_co_request(BlockDriverState *bs,
 228                           NBDRequest *request,
 229                           QEMUIOVector *qiov)
 230 {
 231     NBDClientSession *client = nbd_get_client_session(bs);
 232     int ret;
 233
 234     assert(!qiov || request->type == NBD_CMD_WRITE ||
 235            request->type == NBD_CMD_READ);
 236     ret = nbd_co_send_request(bs, request,
 237                               request->type == NBD_CMD_WRITE ? qiov : NULL);
 238     if (ret < 0) {
 239         return ret;
 240     }
 241
 242     return nbd_co_receive_reply(client, request,
 243                                 request->type == NBD_CMD_READ ? qiov : NULL);
 244 }
 245
 246 int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
 247                          uint64_t bytes, QEMUIOVector *qiov, int flags)
 248 {
 249     NBDRequest request = {
 250         .type = NBD_CMD_READ,
 251         .from = offset,
 252         .len = bytes,
 253     };
 254
 255     assert(bytes <= NBD_MAX_BUFFER_SIZE);
 256     assert(!flags);
 257
 258     return nbd_co_request(bs, &request, qiov);
 259 }
 260
 261 int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset,
 262                           uint64_t bytes, QEMUIOVector *qiov, int flags)
 263 {
 264     NBDClientSession *client = nbd_get_client_session(bs);
 265     NBDRequest request = {
 266         .type = NBD_CMD_WRITE,
 267         .from = offset,
 268         .len = bytes,
 269     };
 270
 271     if (flags & BDRV_REQ_FUA) {
 272         assert(client->info.flags & NBD_FLAG_SEND_FUA);
 273         request.flags |= NBD_CMD_FLAG_FUA;
 274     }
 275
 276     assert(bytes <= NBD_MAX_BUFFER_SIZE);
 277
 278     return nbd_co_request(bs, &request, qiov);
 279 }
 280
 281 int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
 282                                 int bytes, BdrvRequestFlags flags)
 283 {
 284     NBDClientSession *client = nbd_get_client_session(bs);
 285     NBDRequest request = {
 286         .type = NBD_CMD_WRITE_ZEROES,
 287         .from = offset,
 288         .len = bytes,
 289     };
 290
 291     if (!(client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
 292         return -ENOTSUP;
 293     }
 294
 295     if (flags & BDRV_REQ_FUA) {
 296         assert(client->info.flags & NBD_FLAG_SEND_FUA);
 297         request.flags |= NBD_CMD_FLAG_FUA;
 298     }
 299     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
 300         request.flags |= NBD_CMD_FLAG_NO_HOLE;
 301     }
 302
 303     return nbd_co_request(bs, &request, NULL);
 304 }
 305
 306 int nbd_client_co_flush(BlockDriverState *bs)
 307 {
 308     NBDClientSession *client = nbd_get_client_session(bs);
 309     NBDRequest request = { .type = NBD_CMD_FLUSH };
 310
 311     if (!(client->info.flags & NBD_FLAG_SEND_FLUSH)) {
 312         return 0;
 313     }
 314
 315     request.from = 0;
 316     request.len = 0;
 317
 318     return nbd_co_request(bs, &request, NULL);
 319 }
 320
 321 int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
 322 {
 323     NBDClientSession *client = nbd_get_client_session(bs);
 324     NBDRequest request = {
 325         .type = NBD_CMD_TRIM,
 326         .from = offset,
 327         .len = bytes,
 328     };
 329
 330     if (!(client->info.flags & NBD_FLAG_SEND_TRIM)) {
 331         return 0;
 332     }
 333
 334     return nbd_co_request(bs, &request, NULL);
 335 }
 336
 337 void nbd_client_detach_aio_context(BlockDriverState *bs)
 338 {
 339     NBDClientSession *client = nbd_get_client_session(bs);
 340     qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
 341 }
 342
 343 void nbd_client_attach_aio_context(BlockDriverState *bs,
 344                                    AioContext *new_context)
 345 {
 346     NBDClientSession *client = nbd_get_client_session(bs);
 347     qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc), new_context);
 348     aio_co_schedule(new_context, client->read_reply_co);
 349 }
 350
 351 void nbd_client_close(BlockDriverState *bs)
 352 {
 353     NBDClientSession *client = nbd_get_client_session(bs);
 354     NBDRequest request = { .type = NBD_CMD_DISC };
 355
 356     if (client->ioc == NULL) {
 357         return;
 358     }
 359
 360     nbd_send_request(client->ioc, &request);
 361
 362     nbd_teardown_connection(bs);
 363 }
 364
 365 int nbd_client_init(BlockDriverState *bs,
 366                     QIOChannelSocket *sioc,
 367                     const char *export,
 368                     QCryptoTLSCreds *tlscreds,
 369                     const char *hostname,
 370                     Error **errp)
 371 {
 372     NBDClientSession *client = nbd_get_client_session(bs);
 373     int ret;
 374
 375     /* NBD handshake */
 376     logout("session init %s\n", export);
 377     qio_channel_set_blocking(QIO_CHANNEL(sioc), true, NULL);
 378
 379     client->info.request_sizes = true;
 380     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), export,
 381                                 tlscreds, hostname,
 382                                 &client->ioc, &client->info, errp);
 383     if (ret < 0) {
 384         logout("Failed to negotiate with the NBD server\n");
 385         return ret;
 386     }
 387     if (client->info.flags & NBD_FLAG_SEND_FUA) {
 388         bs->supported_write_flags = BDRV_REQ_FUA;
 389         bs->supported_zero_flags |= BDRV_REQ_FUA;
 390     }
 391     if (client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
 392         bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
 393     }
 394     if (client->info.min_block > bs->bl.request_alignment) {
 395         bs->bl.request_alignment = client->info.min_block;
 396     }
 397
 398     qemu_co_mutex_init(&client->send_mutex);
 399     qemu_co_queue_init(&client->free_sema);
 400     client->sioc = sioc;
 401     object_ref(OBJECT(client->sioc));
 402
 403     if (!client->ioc) {
 404         client->ioc = QIO_CHANNEL(sioc);
 405         object_ref(OBJECT(client->ioc));
 406     }
 407
 408     /* Now that we're connected, set the socket to be non-blocking and
 409      * kick the reply mechanism.  */
 410     qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
 411     client->read_reply_co = qemu_coroutine_create(nbd_read_reply_entry, client);
 412     nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
 413
 414     logout("Established connection with NBD server\n");
 415     return 0;
 416 }