nbd.c

   1 /*
   2  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
   3  *
   4  *  Network Block Device
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; under version 2 of the License.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  You should have received a copy of the GNU General Public License
  16  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
  17  */
  18
  19 #include "nbd.h"
  20 #include "block.h"
  21
  22 #include "qemu-coroutine.h"
  23
  24 #include <errno.h>
  25 #include <string.h>
  26 #ifndef _WIN32
  27 #include <sys/ioctl.h>
  28 #endif
  29 #if defined(__sun__) || defined(__HAIKU__)
  30 #include <sys/ioccom.h>
  31 #endif
  32 #include <ctype.h>
  33 #include <inttypes.h>
  34
  35 #ifdef __linux__
  36 #include <linux/fs.h>
  37 #endif
  38
  39 #include "qemu_socket.h"
  40 #include "qemu-queue.h"
  41
  42 //#define DEBUG_NBD
  43
  44 #ifdef DEBUG_NBD
  45 #define TRACE(msg, ...) do { \
  46     LOG(msg, ## __VA_ARGS__); \
  47 } while(0)
  48 #else
  49 #define TRACE(msg, ...) \
  50     do { } while (0)
  51 #endif
  52
  53 #define LOG(msg, ...) do { \
  54     fprintf(stderr, "%s:%s():L%d: " msg "\n", \
  55             __FILE__, __FUNCTION__, __LINE__, ## __VA_ARGS__); \
  56 } while(0)
  57
  58 /* This is all part of the "official" NBD API */
  59
  60 #define NBD_REPLY_SIZE          (4 + 4 + 8)
  61 #define NBD_REQUEST_MAGIC       0x25609513
  62 #define NBD_REPLY_MAGIC         0x67446698
  63
  64 #define NBD_SET_SOCK            _IO(0xab, 0)
  65 #define NBD_SET_BLKSIZE         _IO(0xab, 1)
  66 #define NBD_SET_SIZE            _IO(0xab, 2)
  67 #define NBD_DO_IT               _IO(0xab, 3)
  68 #define NBD_CLEAR_SOCK          _IO(0xab, 4)
  69 #define NBD_CLEAR_QUE           _IO(0xab, 5)
  70 #define NBD_PRINT_DEBUG         _IO(0xab, 6)
  71 #define NBD_SET_SIZE_BLOCKS     _IO(0xab, 7)
  72 #define NBD_DISCONNECT          _IO(0xab, 8)
  73 #define NBD_SET_TIMEOUT         _IO(0xab, 9)
  74 #define NBD_SET_FLAGS           _IO(0xab, 10)
  75
  76 #define NBD_OPT_EXPORT_NAME     (1 << 0)
  77
  78 /* That's all folks */
  79
  80 ssize_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
  81 {
  82     size_t offset = 0;
  83     int err;
  84
  85     if (qemu_in_coroutine()) {
  86         if (do_read) {
  87             return qemu_co_recv(fd, buffer, size);
  88         } else {
  89             return qemu_co_send(fd, buffer, size);
  90         }
  91     }
  92
  93     while (offset < size) {
  94         ssize_t len;
  95
  96         if (do_read) {
  97             len = qemu_recv(fd, buffer + offset, size - offset, 0);
  98         } else {
  99             len = send(fd, buffer + offset, size - offset, 0);
 100         }
 101
 102         if (len < 0) {
 103             err = socket_error();
 104
 105             /* recoverable error */
 106             if (err == EINTR || (offset > 0 && err == EAGAIN)) {
 107                 continue;
 108             }
 109
 110             /* unrecoverable error */
 111             return -err;
 112         }
 113
 114         /* eof */
 115         if (len == 0) {
 116             break;
 117         }
 118
 119         offset += len;
 120     }
 121
 122     return offset;
 123 }
 124
 125 static ssize_t read_sync(int fd, void *buffer, size_t size)
 126 {
 127     /* Sockets are kept in blocking mode in the negotiation phase.  After
 128      * that, a non-readable socket simply means that another thread stole
 129      * our request/reply.  Synchronization is done with recv_coroutine, so
 130      * that this is coroutine-safe.
 131      */
 132     return nbd_wr_sync(fd, buffer, size, true);
 133 }
 134
 135 static ssize_t write_sync(int fd, void *buffer, size_t size)
 136 {
 137     int ret;
 138     do {
 139         /* For writes, we do expect the socket to be writable.  */
 140         ret = nbd_wr_sync(fd, buffer, size, false);
 141     } while (ret == -EAGAIN);
 142     return ret;
 143 }
 144
 145 static void combine_addr(char *buf, size_t len, const char* address,
 146                          uint16_t port)
 147 {
 148     /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
 149     if (strstr(address, ":")) {
 150         snprintf(buf, len, "[%s]:%u", address, port);
 151     } else {
 152         snprintf(buf, len, "%s:%u", address, port);
 153     }
 154 }
 155
 156 int tcp_socket_outgoing(const char *address, uint16_t port)
 157 {
 158     char address_and_port[128];
 159     combine_addr(address_and_port, 128, address, port);
 160     return tcp_socket_outgoing_spec(address_and_port);
 161 }
 162
 163 int tcp_socket_outgoing_spec(const char *address_and_port)
 164 {
 165     return inet_connect(address_and_port, true, NULL);
 166 }
 167
 168 int tcp_socket_incoming(const char *address, uint16_t port)
 169 {
 170     char address_and_port[128];
 171     combine_addr(address_and_port, 128, address, port);
 172     return tcp_socket_incoming_spec(address_and_port);
 173 }
 174
 175 int tcp_socket_incoming_spec(const char *address_and_port)
 176 {
 177     char *ostr  = NULL;
 178     int olen = 0;
 179     return inet_listen(address_and_port, ostr, olen, SOCK_STREAM, 0, NULL);
 180 }
 181
 182 int unix_socket_incoming(const char *path)
 183 {
 184     char *ostr = NULL;
 185     int olen = 0;
 186
 187     return unix_listen(path, ostr, olen);
 188 }
 189
 190 int unix_socket_outgoing(const char *path)
 191 {
 192     return unix_connect(path);
 193 }
 194
 195 /* Basic flow
 196
 197    Server         Client
 198
 199    Negotiate
 200                   Request
 201    Response
 202                   Request
 203    Response
 204                   ...
 205    ...
 206                   Request (type == 2)
 207 */
 208
 209 static int nbd_send_negotiate(int csock, off_t size, uint32_t flags)
 210 {
 211     char buf[8 + 8 + 8 + 128];
 212     int rc;
 213
 214     /* Negotiate
 215         [ 0 ..   7]   passwd   ("NBDMAGIC")
 216         [ 8 ..  15]   magic    (0x00420281861253)
 217         [16 ..  23]   size
 218         [24 ..  27]   flags
 219         [28 .. 151]   reserved (0)
 220      */
 221
 222     socket_set_block(csock);
 223     rc = -EINVAL;
 224
 225     TRACE("Beginning negotiation.");
 226     memcpy(buf, "NBDMAGIC", 8);
 227     cpu_to_be64w((uint64_t*)(buf + 8), 0x00420281861253LL);
 228     cpu_to_be64w((uint64_t*)(buf + 16), size);
 229     cpu_to_be32w((uint32_t*)(buf + 24),
 230                  flags | NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
 231                  NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
 232     memset(buf + 28, 0, 124);
 233
 234     if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
 235         LOG("write failed");
 236         goto fail;
 237     }
 238
 239     TRACE("Negotiation succeeded.");
 240     rc = 0;
 241 fail:
 242     socket_set_nonblock(csock);
 243     return rc;
 244 }
 245
 246 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
 247                           off_t *size, size_t *blocksize)
 248 {
 249     char buf[256];
 250     uint64_t magic, s;
 251     uint16_t tmp;
 252     int rc;
 253
 254     TRACE("Receiving negotiation.");
 255
 256     socket_set_block(csock);
 257     rc = -EINVAL;
 258
 259     if (read_sync(csock, buf, 8) != 8) {
 260         LOG("read failed");
 261         goto fail;
 262     }
 263
 264     buf[8] = '\0';
 265     if (strlen(buf) == 0) {
 266         LOG("server connection closed");
 267         goto fail;
 268     }
 269
 270     TRACE("Magic is %c%c%c%c%c%c%c%c",
 271           qemu_isprint(buf[0]) ? buf[0] : '.',
 272           qemu_isprint(buf[1]) ? buf[1] : '.',
 273           qemu_isprint(buf[2]) ? buf[2] : '.',
 274           qemu_isprint(buf[3]) ? buf[3] : '.',
 275           qemu_isprint(buf[4]) ? buf[4] : '.',
 276           qemu_isprint(buf[5]) ? buf[5] : '.',
 277           qemu_isprint(buf[6]) ? buf[6] : '.',
 278           qemu_isprint(buf[7]) ? buf[7] : '.');
 279
 280     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 281         LOG("Invalid magic received");
 282         goto fail;
 283     }
 284
 285     if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 286         LOG("read failed");
 287         goto fail;
 288     }
 289     magic = be64_to_cpu(magic);
 290     TRACE("Magic is 0x%" PRIx64, magic);
 291
 292     if (name) {
 293         uint32_t reserved = 0;
 294         uint32_t opt;
 295         uint32_t namesize;
 296
 297         TRACE("Checking magic (opts_magic)");
 298         if (magic != 0x49484156454F5054LL) {
 299             LOG("Bad magic received");
 300             goto fail;
 301         }
 302         if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 303             LOG("flags read failed");
 304             goto fail;
 305         }
 306         *flags = be16_to_cpu(tmp) << 16;
 307         /* reserved for future use */
 308         if (write_sync(csock, &reserved, sizeof(reserved)) !=
 309             sizeof(reserved)) {
 310             LOG("write failed (reserved)");
 311             goto fail;
 312         }
 313         /* write the export name */
 314         magic = cpu_to_be64(magic);
 315         if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 316             LOG("write failed (magic)");
 317             goto fail;
 318         }
 319         opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
 320         if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
 321             LOG("write failed (opt)");
 322             goto fail;
 323         }
 324         namesize = cpu_to_be32(strlen(name));
 325         if (write_sync(csock, &namesize, sizeof(namesize)) !=
 326             sizeof(namesize)) {
 327             LOG("write failed (namesize)");
 328             goto fail;
 329         }
 330         if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
 331             LOG("write failed (name)");
 332             goto fail;
 333         }
 334     } else {
 335         TRACE("Checking magic (cli_magic)");
 336
 337         if (magic != 0x00420281861253LL) {
 338             LOG("Bad magic received");
 339             goto fail;
 340         }
 341     }
 342
 343     if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
 344         LOG("read failed");
 345         goto fail;
 346     }
 347     *size = be64_to_cpu(s);
 348     *blocksize = 1024;
 349     TRACE("Size is %" PRIu64, *size);
 350
 351     if (!name) {
 352         if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
 353             LOG("read failed (flags)");
 354             goto fail;
 355         }
 356         *flags = be32_to_cpup(flags);
 357     } else {
 358         if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 359             LOG("read failed (tmp)");
 360             goto fail;
 361         }
 362         *flags |= be32_to_cpu(tmp);
 363     }
 364     if (read_sync(csock, &buf, 124) != 124) {
 365         LOG("read failed (buf)");
 366         goto fail;
 367     }
 368     rc = 0;
 369
 370 fail:
 371     socket_set_nonblock(csock);
 372     return rc;
 373 }
 374
 375 #ifdef __linux__
 376 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 377 {
 378     TRACE("Setting NBD socket");
 379
 380     if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
 381         int serrno = errno;
 382         LOG("Failed to set NBD socket");
 383         return -serrno;
 384     }
 385
 386     TRACE("Setting block size to %lu", (unsigned long)blocksize);
 387
 388     if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
 389         int serrno = errno;
 390         LOG("Failed setting NBD block size");
 391         return -serrno;
 392     }
 393
 394         TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
 395
 396     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
 397         int serrno = errno;
 398         LOG("Failed setting size (in blocks)");
 399         return -serrno;
 400     }
 401
 402     if (flags & NBD_FLAG_READ_ONLY) {
 403         int read_only = 1;
 404         TRACE("Setting readonly attribute");
 405
 406         if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 407             int serrno = errno;
 408             LOG("Failed setting read-only attribute");
 409             return -serrno;
 410         }
 411     }
 412
 413     if (ioctl(fd, NBD_SET_FLAGS, flags) < 0
 414         && errno != ENOTTY) {
 415         int serrno = errno;
 416         LOG("Failed setting flags");
 417         return -serrno;
 418     }
 419
 420     TRACE("Negotiation ended");
 421
 422     return 0;
 423 }
 424
 425 int nbd_disconnect(int fd)
 426 {
 427     ioctl(fd, NBD_CLEAR_QUE);
 428     ioctl(fd, NBD_DISCONNECT);
 429     ioctl(fd, NBD_CLEAR_SOCK);
 430     return 0;
 431 }
 432
 433 int nbd_client(int fd)
 434 {
 435     int ret;
 436     int serrno;
 437
 438     TRACE("Doing NBD loop");
 439
 440     ret = ioctl(fd, NBD_DO_IT);
 441     if (ret < 0 && errno == EPIPE) {
 442         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 443          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 444          * that case.
 445          */
 446         ret = 0;
 447     }
 448     serrno = errno;
 449
 450     TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
 451
 452     TRACE("Clearing NBD queue");
 453     ioctl(fd, NBD_CLEAR_QUE);
 454
 455     TRACE("Clearing NBD socket");
 456     ioctl(fd, NBD_CLEAR_SOCK);
 457
 458     errno = serrno;
 459     return ret;
 460 }
 461 #else
 462 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 463 {
 464     return -ENOTSUP;
 465 }
 466
 467 int nbd_disconnect(int fd)
 468 {
 469     return -ENOTSUP;
 470 }
 471
 472 int nbd_client(int fd)
 473 {
 474     return -ENOTSUP;
 475 }
 476 #endif
 477
 478 ssize_t nbd_send_request(int csock, struct nbd_request *request)
 479 {
 480     uint8_t buf[4 + 4 + 8 + 8 + 4];
 481     ssize_t ret;
 482
 483     cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
 484     cpu_to_be32w((uint32_t*)(buf + 4), request->type);
 485     cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
 486     cpu_to_be64w((uint64_t*)(buf + 16), request->from);
 487     cpu_to_be32w((uint32_t*)(buf + 24), request->len);
 488
 489     TRACE("Sending request to client: "
 490           "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
 491           request->from, request->len, request->handle, request->type);
 492
 493     ret = write_sync(csock, buf, sizeof(buf));
 494     if (ret < 0) {
 495         return ret;
 496     }
 497
 498     if (ret != sizeof(buf)) {
 499         LOG("writing to socket failed");
 500         return -EINVAL;
 501     }
 502     return 0;
 503 }
 504
 505 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
 506 {
 507     uint8_t buf[4 + 4 + 8 + 8 + 4];
 508     uint32_t magic;
 509     ssize_t ret;
 510
 511     ret = read_sync(csock, buf, sizeof(buf));
 512     if (ret < 0) {
 513         return ret;
 514     }
 515
 516     if (ret != sizeof(buf)) {
 517         LOG("read failed");
 518         return -EINVAL;
 519     }
 520
 521     /* Request
 522        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
 523        [ 4 ..  7]   type    (0 == READ, 1 == WRITE)
 524        [ 8 .. 15]   handle
 525        [16 .. 23]   from
 526        [24 .. 27]   len
 527      */
 528
 529     magic = be32_to_cpup((uint32_t*)buf);
 530     request->type  = be32_to_cpup((uint32_t*)(buf + 4));
 531     request->handle = be64_to_cpup((uint64_t*)(buf + 8));
 532     request->from  = be64_to_cpup((uint64_t*)(buf + 16));
 533     request->len   = be32_to_cpup((uint32_t*)(buf + 24));
 534
 535     TRACE("Got request: "
 536           "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
 537           magic, request->type, request->from, request->len);
 538
 539     if (magic != NBD_REQUEST_MAGIC) {
 540         LOG("invalid magic (got 0x%x)", magic);
 541         return -EINVAL;
 542     }
 543     return 0;
 544 }
 545
 546 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
 547 {
 548     uint8_t buf[NBD_REPLY_SIZE];
 549     uint32_t magic;
 550     ssize_t ret;
 551
 552     ret = read_sync(csock, buf, sizeof(buf));
 553     if (ret < 0) {
 554         return ret;
 555     }
 556
 557     if (ret != sizeof(buf)) {
 558         LOG("read failed");
 559         return -EINVAL;
 560     }
 561
 562     /* Reply
 563        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 564        [ 4 ..  7]    error   (0 == no error)
 565        [ 7 .. 15]    handle
 566      */
 567
 568     magic = be32_to_cpup((uint32_t*)buf);
 569     reply->error  = be32_to_cpup((uint32_t*)(buf + 4));
 570     reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
 571
 572     TRACE("Got reply: "
 573           "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
 574           magic, reply->error, reply->handle);
 575
 576     if (magic != NBD_REPLY_MAGIC) {
 577         LOG("invalid magic (got 0x%x)", magic);
 578         return -EINVAL;
 579     }
 580     return 0;
 581 }
 582
 583 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
 584 {
 585     uint8_t buf[4 + 4 + 8];
 586     ssize_t ret;
 587
 588     /* Reply
 589        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 590        [ 4 ..  7]    error   (0 == no error)
 591        [ 7 .. 15]    handle
 592      */
 593     cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
 594     cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
 595     cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
 596
 597     TRACE("Sending response to client");
 598
 599     ret = write_sync(csock, buf, sizeof(buf));
 600     if (ret < 0) {
 601         return ret;
 602     }
 603
 604     if (ret != sizeof(buf)) {
 605         LOG("writing to socket failed");
 606         return -EINVAL;
 607     }
 608     return 0;
 609 }
 610
 611 #define MAX_NBD_REQUESTS 16
 612
 613 typedef struct NBDRequest NBDRequest;
 614
 615 struct NBDRequest {
 616     QSIMPLEQ_ENTRY(NBDRequest) entry;
 617     NBDClient *client;
 618     uint8_t *data;
 619 };
 620
 621 struct NBDExport {
 622     BlockDriverState *bs;
 623     off_t dev_offset;
 624     off_t size;
 625     uint32_t nbdflags;
 626     QSIMPLEQ_HEAD(, NBDRequest) requests;
 627 };
 628
 629 struct NBDClient {
 630     int refcount;
 631     void (*close)(NBDClient *client);
 632
 633     NBDExport *exp;
 634     int sock;
 635
 636     Coroutine *recv_coroutine;
 637
 638     CoMutex send_lock;
 639     Coroutine *send_coroutine;
 640
 641     int nb_requests;
 642 };
 643
 644 static void nbd_client_get(NBDClient *client)
 645 {
 646     client->refcount++;
 647 }
 648
 649 static void nbd_client_put(NBDClient *client)
 650 {
 651     if (--client->refcount == 0) {
 652         g_free(client);
 653     }
 654 }
 655
 656 static void nbd_client_close(NBDClient *client)
 657 {
 658     qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
 659     close(client->sock);
 660     client->sock = -1;
 661     if (client->close) {
 662         client->close(client);
 663     }
 664     nbd_client_put(client);
 665 }
 666
 667 static NBDRequest *nbd_request_get(NBDClient *client)
 668 {
 669     NBDRequest *req;
 670     NBDExport *exp = client->exp;
 671
 672     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
 673     client->nb_requests++;
 674
 675     if (QSIMPLEQ_EMPTY(&exp->requests)) {
 676         req = g_malloc0(sizeof(NBDRequest));
 677         req->data = qemu_blockalign(exp->bs, NBD_BUFFER_SIZE);
 678     } else {
 679         req = QSIMPLEQ_FIRST(&exp->requests);
 680         QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
 681     }
 682     nbd_client_get(client);
 683     req->client = client;
 684     return req;
 685 }
 686
 687 static void nbd_request_put(NBDRequest *req)
 688 {
 689     NBDClient *client = req->client;
 690     QSIMPLEQ_INSERT_HEAD(&client->exp->requests, req, entry);
 691     if (client->nb_requests-- == MAX_NBD_REQUESTS) {
 692         qemu_notify_event();
 693     }
 694     nbd_client_put(client);
 695 }
 696
 697 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
 698                           off_t size, uint32_t nbdflags)
 699 {
 700     NBDExport *exp = g_malloc0(sizeof(NBDExport));
 701     QSIMPLEQ_INIT(&exp->requests);
 702     exp->bs = bs;
 703     exp->dev_offset = dev_offset;
 704     exp->nbdflags = nbdflags;
 705     exp->size = size == -1 ? bdrv_getlength(bs) : size;
 706     return exp;
 707 }
 708
 709 void nbd_export_close(NBDExport *exp)
 710 {
 711     while (!QSIMPLEQ_EMPTY(&exp->requests)) {
 712         NBDRequest *first = QSIMPLEQ_FIRST(&exp->requests);
 713         QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
 714         qemu_vfree(first->data);
 715         g_free(first);
 716     }
 717
 718     bdrv_close(exp->bs);
 719     g_free(exp);
 720 }
 721
 722 static int nbd_can_read(void *opaque);
 723 static void nbd_read(void *opaque);
 724 static void nbd_restart_write(void *opaque);
 725
 726 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
 727                                  int len)
 728 {
 729     NBDClient *client = req->client;
 730     int csock = client->sock;
 731     ssize_t rc, ret;
 732
 733     qemu_co_mutex_lock(&client->send_lock);
 734     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
 735                          nbd_restart_write, client);
 736     client->send_coroutine = qemu_coroutine_self();
 737
 738     if (!len) {
 739         rc = nbd_send_reply(csock, reply);
 740     } else {
 741         socket_set_cork(csock, 1);
 742         rc = nbd_send_reply(csock, reply);
 743         if (rc >= 0) {
 744             ret = qemu_co_send(csock, req->data, len);
 745             if (ret != len) {
 746                 rc = -EIO;
 747             }
 748         }
 749         socket_set_cork(csock, 0);
 750     }
 751
 752     client->send_coroutine = NULL;
 753     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
 754     qemu_co_mutex_unlock(&client->send_lock);
 755     return rc;
 756 }
 757
 758 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
 759 {
 760     NBDClient *client = req->client;
 761     int csock = client->sock;
 762     ssize_t rc;
 763
 764     client->recv_coroutine = qemu_coroutine_self();
 765     rc = nbd_receive_request(csock, request);
 766     if (rc < 0) {
 767         if (rc != -EAGAIN) {
 768             rc = -EIO;
 769         }
 770         goto out;
 771     }
 772
 773     if (request->len > NBD_BUFFER_SIZE) {
 774         LOG("len (%u) is larger than max len (%u)",
 775             request->len, NBD_BUFFER_SIZE);
 776         rc = -EINVAL;
 777         goto out;
 778     }
 779
 780     if ((request->from + request->len) < request->from) {
 781         LOG("integer overflow detected! "
 782             "you're probably being attacked");
 783         rc = -EINVAL;
 784         goto out;
 785     }
 786
 787     TRACE("Decoding type");
 788
 789     if ((request->type & NBD_CMD_MASK_COMMAND) == NBD_CMD_WRITE) {
 790         TRACE("Reading %u byte(s)", request->len);
 791
 792         if (qemu_co_recv(csock, req->data, request->len) != request->len) {
 793             LOG("reading from socket failed");
 794             rc = -EIO;
 795             goto out;
 796         }
 797     }
 798     rc = 0;
 799
 800 out:
 801     client->recv_coroutine = NULL;
 802     return rc;
 803 }
 804
 805 static void nbd_trip(void *opaque)
 806 {
 807     NBDClient *client = opaque;
 808     NBDRequest *req = nbd_request_get(client);
 809     NBDExport *exp = client->exp;
 810     struct nbd_request request;
 811     struct nbd_reply reply;
 812     ssize_t ret;
 813
 814     TRACE("Reading request.");
 815
 816     ret = nbd_co_receive_request(req, &request);
 817     if (ret == -EAGAIN) {
 818         goto done;
 819     }
 820     if (ret == -EIO) {
 821         goto out;
 822     }
 823
 824     reply.handle = request.handle;
 825     reply.error = 0;
 826
 827     if (ret < 0) {
 828         reply.error = -ret;
 829         goto error_reply;
 830     }
 831
 832     if ((request.from + request.len) > exp->size) {
 833             LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
 834             ", Offset: %" PRIu64 "\n",
 835                     request.from, request.len,
 836                     (uint64_t)exp->size, (uint64_t)exp->dev_offset);
 837         LOG("requested operation past EOF--bad client?");
 838         goto invalid_request;
 839     }
 840
 841     switch (request.type & NBD_CMD_MASK_COMMAND) {
 842     case NBD_CMD_READ:
 843         TRACE("Request type is READ");
 844
 845         if (request.type & NBD_CMD_FLAG_FUA) {
 846             ret = bdrv_co_flush(exp->bs);
 847             if (ret < 0) {
 848                 LOG("flush failed");
 849                 reply.error = -ret;
 850                 goto error_reply;
 851             }
 852         }
 853
 854         ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
 855                         req->data, request.len / 512);
 856         if (ret < 0) {
 857             LOG("reading from file failed");
 858             reply.error = -ret;
 859             goto error_reply;
 860         }
 861
 862         TRACE("Read %u byte(s)", request.len);
 863         if (nbd_co_send_reply(req, &reply, request.len) < 0)
 864             goto out;
 865         break;
 866     case NBD_CMD_WRITE:
 867         TRACE("Request type is WRITE");
 868
 869         if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
 870             TRACE("Server is read-only, return error");
 871             reply.error = EROFS;
 872             goto error_reply;
 873         }
 874
 875         TRACE("Writing to device");
 876
 877         ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
 878                          req->data, request.len / 512);
 879         if (ret < 0) {
 880             LOG("writing to file failed");
 881             reply.error = -ret;
 882             goto error_reply;
 883         }
 884
 885         if (request.type & NBD_CMD_FLAG_FUA) {
 886             ret = bdrv_co_flush(exp->bs);
 887             if (ret < 0) {
 888                 LOG("flush failed");
 889                 reply.error = -ret;
 890                 goto error_reply;
 891             }
 892         }
 893
 894         if (nbd_co_send_reply(req, &reply, 0) < 0) {
 895             goto out;
 896         }
 897         break;
 898     case NBD_CMD_DISC:
 899         TRACE("Request type is DISCONNECT");
 900         errno = 0;
 901         goto out;
 902     case NBD_CMD_FLUSH:
 903         TRACE("Request type is FLUSH");
 904
 905         ret = bdrv_co_flush(exp->bs);
 906         if (ret < 0) {
 907             LOG("flush failed");
 908             reply.error = -ret;
 909         }
 910         if (nbd_co_send_reply(req, &reply, 0) < 0) {
 911             goto out;
 912         }
 913         break;
 914     case NBD_CMD_TRIM:
 915         TRACE("Request type is TRIM");
 916         ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
 917                               request.len / 512);
 918         if (ret < 0) {
 919             LOG("discard failed");
 920             reply.error = -ret;
 921         }
 922         if (nbd_co_send_reply(req, &reply, 0) < 0) {
 923             goto out;
 924         }
 925         break;
 926     default:
 927         LOG("invalid request type (%u) received", request.type);
 928     invalid_request:
 929         reply.error = -EINVAL;
 930     error_reply:
 931         if (nbd_co_send_reply(req, &reply, 0) < 0) {
 932             goto out;
 933         }
 934         break;
 935     }
 936
 937     TRACE("Request/Reply complete");
 938
 939 done:
 940     nbd_request_put(req);
 941     return;
 942
 943 out:
 944     nbd_request_put(req);
 945     nbd_client_close(client);
 946 }
 947
 948 static int nbd_can_read(void *opaque)
 949 {
 950     NBDClient *client = opaque;
 951
 952     return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
 953 }
 954
 955 static void nbd_read(void *opaque)
 956 {
 957     NBDClient *client = opaque;
 958
 959     if (client->recv_coroutine) {
 960         qemu_coroutine_enter(client->recv_coroutine, NULL);
 961     } else {
 962         qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
 963     }
 964 }
 965
 966 static void nbd_restart_write(void *opaque)
 967 {
 968     NBDClient *client = opaque;
 969
 970     qemu_coroutine_enter(client->send_coroutine, NULL);
 971 }
 972
 973 NBDClient *nbd_client_new(NBDExport *exp, int csock,
 974                           void (*close)(NBDClient *))
 975 {
 976     NBDClient *client;
 977     if (nbd_send_negotiate(csock, exp->size, exp->nbdflags) < 0) {
 978         return NULL;
 979     }
 980     client = g_malloc0(sizeof(NBDClient));
 981     client->refcount = 1;
 982     client->exp = exp;
 983     client->sock = csock;
 984     client->close = close;
 985     qemu_co_mutex_init(&client->send_lock);
 986     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
 987     return client;
 988 }