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_REQUEST_SIZE        (4 + 4 + 8 + 8 + 4)
  61 #define NBD_REPLY_SIZE          (4 + 4 + 8)
  62 #define NBD_REQUEST_MAGIC       0x25609513
  63 #define NBD_REPLY_MAGIC         0x67446698
  64 #define NBD_OPTS_MAGIC          0x49484156454F5054LL
  65 #define NBD_CLIENT_MAGIC        0x0000420281861253LL
  66
  67 #define NBD_SET_SOCK            _IO(0xab, 0)
  68 #define NBD_SET_BLKSIZE         _IO(0xab, 1)
  69 #define NBD_SET_SIZE            _IO(0xab, 2)
  70 #define NBD_DO_IT               _IO(0xab, 3)
  71 #define NBD_CLEAR_SOCK          _IO(0xab, 4)
  72 #define NBD_CLEAR_QUE           _IO(0xab, 5)
  73 #define NBD_PRINT_DEBUG         _IO(0xab, 6)
  74 #define NBD_SET_SIZE_BLOCKS     _IO(0xab, 7)
  75 #define NBD_DISCONNECT          _IO(0xab, 8)
  76 #define NBD_SET_TIMEOUT         _IO(0xab, 9)
  77 #define NBD_SET_FLAGS           _IO(0xab, 10)
  78
  79 #define NBD_OPT_EXPORT_NAME     (1 << 0)
  80
  81 /* Definitions for opaque data types */
  82
  83 typedef struct NBDRequest NBDRequest;
  84
  85 struct NBDRequest {
  86     QSIMPLEQ_ENTRY(NBDRequest) entry;
  87     NBDClient *client;
  88     uint8_t *data;
  89 };
  90
  91 struct NBDExport {
  92     BlockDriverState *bs;
  93     off_t dev_offset;
  94     off_t size;
  95     uint32_t nbdflags;
  96     QSIMPLEQ_HEAD(, NBDRequest) requests;
  97 };
  98
  99 struct NBDClient {
 100     int refcount;
 101     void (*close)(NBDClient *client);
 102
 103     NBDExport *exp;
 104     int sock;
 105
 106     Coroutine *recv_coroutine;
 107
 108     CoMutex send_lock;
 109     Coroutine *send_coroutine;
 110
 111     int nb_requests;
 112 };
 113
 114 /* That's all folks */
 115
 116 ssize_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
 117 {
 118     size_t offset = 0;
 119     int err;
 120
 121     if (qemu_in_coroutine()) {
 122         if (do_read) {
 123             return qemu_co_recv(fd, buffer, size);
 124         } else {
 125             return qemu_co_send(fd, buffer, size);
 126         }
 127     }
 128
 129     while (offset < size) {
 130         ssize_t len;
 131
 132         if (do_read) {
 133             len = qemu_recv(fd, buffer + offset, size - offset, 0);
 134         } else {
 135             len = send(fd, buffer + offset, size - offset, 0);
 136         }
 137
 138         if (len < 0) {
 139             err = socket_error();
 140
 141             /* recoverable error */
 142             if (err == EINTR || (offset > 0 && err == EAGAIN)) {
 143                 continue;
 144             }
 145
 146             /* unrecoverable error */
 147             return -err;
 148         }
 149
 150         /* eof */
 151         if (len == 0) {
 152             break;
 153         }
 154
 155         offset += len;
 156     }
 157
 158     return offset;
 159 }
 160
 161 static ssize_t read_sync(int fd, void *buffer, size_t size)
 162 {
 163     /* Sockets are kept in blocking mode in the negotiation phase.  After
 164      * that, a non-readable socket simply means that another thread stole
 165      * our request/reply.  Synchronization is done with recv_coroutine, so
 166      * that this is coroutine-safe.
 167      */
 168     return nbd_wr_sync(fd, buffer, size, true);
 169 }
 170
 171 static ssize_t write_sync(int fd, void *buffer, size_t size)
 172 {
 173     int ret;
 174     do {
 175         /* For writes, we do expect the socket to be writable.  */
 176         ret = nbd_wr_sync(fd, buffer, size, false);
 177     } while (ret == -EAGAIN);
 178     return ret;
 179 }
 180
 181 static void combine_addr(char *buf, size_t len, const char* address,
 182                          uint16_t port)
 183 {
 184     /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
 185     if (strstr(address, ":")) {
 186         snprintf(buf, len, "[%s]:%u", address, port);
 187     } else {
 188         snprintf(buf, len, "%s:%u", address, port);
 189     }
 190 }
 191
 192 int tcp_socket_outgoing(const char *address, uint16_t port)
 193 {
 194     char address_and_port[128];
 195     combine_addr(address_and_port, 128, address, port);
 196     return tcp_socket_outgoing_spec(address_and_port);
 197 }
 198
 199 int tcp_socket_outgoing_spec(const char *address_and_port)
 200 {
 201     return inet_connect(address_and_port, true, NULL, NULL);
 202 }
 203
 204 int tcp_socket_incoming(const char *address, uint16_t port)
 205 {
 206     char address_and_port[128];
 207     combine_addr(address_and_port, 128, address, port);
 208     return tcp_socket_incoming_spec(address_and_port);
 209 }
 210
 211 int tcp_socket_incoming_spec(const char *address_and_port)
 212 {
 213     char *ostr  = NULL;
 214     int olen = 0;
 215     return inet_listen(address_and_port, ostr, olen, SOCK_STREAM, 0, NULL);
 216 }
 217
 218 int unix_socket_incoming(const char *path)
 219 {
 220     char *ostr = NULL;
 221     int olen = 0;
 222
 223     return unix_listen(path, ostr, olen);
 224 }
 225
 226 int unix_socket_outgoing(const char *path)
 227 {
 228     return unix_connect(path);
 229 }
 230
 231 /* Basic flow
 232
 233    Server         Client
 234
 235    Negotiate
 236                   Request
 237    Response
 238                   Request
 239    Response
 240                   ...
 241    ...
 242                   Request (type == 2)
 243 */
 244
 245 static int nbd_send_negotiate(NBDClient *client)
 246 {
 247     int csock = client->sock;
 248     char buf[8 + 8 + 8 + 128];
 249     int rc;
 250
 251     /* Negotiate
 252         [ 0 ..   7]   passwd   ("NBDMAGIC")
 253         [ 8 ..  15]   magic    (NBD_CLIENT_MAGIC)
 254         [16 ..  23]   size
 255         [24 ..  27]   flags
 256         [28 .. 151]   reserved (0)
 257      */
 258
 259     socket_set_block(csock);
 260     rc = -EINVAL;
 261
 262     TRACE("Beginning negotiation.");
 263     memcpy(buf, "NBDMAGIC", 8);
 264     cpu_to_be64w((uint64_t*)(buf + 8), NBD_CLIENT_MAGIC);
 265     cpu_to_be64w((uint64_t*)(buf + 16), client->exp->size);
 266     cpu_to_be32w((uint32_t*)(buf + 24),
 267                  client->exp->nbdflags | NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
 268                  NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
 269     memset(buf + 28, 0, 124);
 270
 271     if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
 272         LOG("write failed");
 273         goto fail;
 274     }
 275
 276     TRACE("Negotiation succeeded.");
 277     rc = 0;
 278 fail:
 279     socket_set_nonblock(csock);
 280     return rc;
 281 }
 282
 283 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
 284                           off_t *size, size_t *blocksize)
 285 {
 286     char buf[256];
 287     uint64_t magic, s;
 288     uint16_t tmp;
 289     int rc;
 290
 291     TRACE("Receiving negotiation.");
 292
 293     socket_set_block(csock);
 294     rc = -EINVAL;
 295
 296     if (read_sync(csock, buf, 8) != 8) {
 297         LOG("read failed");
 298         goto fail;
 299     }
 300
 301     buf[8] = '\0';
 302     if (strlen(buf) == 0) {
 303         LOG("server connection closed");
 304         goto fail;
 305     }
 306
 307     TRACE("Magic is %c%c%c%c%c%c%c%c",
 308           qemu_isprint(buf[0]) ? buf[0] : '.',
 309           qemu_isprint(buf[1]) ? buf[1] : '.',
 310           qemu_isprint(buf[2]) ? buf[2] : '.',
 311           qemu_isprint(buf[3]) ? buf[3] : '.',
 312           qemu_isprint(buf[4]) ? buf[4] : '.',
 313           qemu_isprint(buf[5]) ? buf[5] : '.',
 314           qemu_isprint(buf[6]) ? buf[6] : '.',
 315           qemu_isprint(buf[7]) ? buf[7] : '.');
 316
 317     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 318         LOG("Invalid magic received");
 319         goto fail;
 320     }
 321
 322     if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 323         LOG("read failed");
 324         goto fail;
 325     }
 326     magic = be64_to_cpu(magic);
 327     TRACE("Magic is 0x%" PRIx64, magic);
 328
 329     if (name) {
 330         uint32_t reserved = 0;
 331         uint32_t opt;
 332         uint32_t namesize;
 333
 334         TRACE("Checking magic (opts_magic)");
 335         if (magic != NBD_OPTS_MAGIC) {
 336             LOG("Bad magic received");
 337             goto fail;
 338         }
 339         if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 340             LOG("flags read failed");
 341             goto fail;
 342         }
 343         *flags = be16_to_cpu(tmp) << 16;
 344         /* reserved for future use */
 345         if (write_sync(csock, &reserved, sizeof(reserved)) !=
 346             sizeof(reserved)) {
 347             LOG("write failed (reserved)");
 348             goto fail;
 349         }
 350         /* write the export name */
 351         magic = cpu_to_be64(magic);
 352         if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 353             LOG("write failed (magic)");
 354             goto fail;
 355         }
 356         opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
 357         if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
 358             LOG("write failed (opt)");
 359             goto fail;
 360         }
 361         namesize = cpu_to_be32(strlen(name));
 362         if (write_sync(csock, &namesize, sizeof(namesize)) !=
 363             sizeof(namesize)) {
 364             LOG("write failed (namesize)");
 365             goto fail;
 366         }
 367         if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
 368             LOG("write failed (name)");
 369             goto fail;
 370         }
 371     } else {
 372         TRACE("Checking magic (cli_magic)");
 373
 374         if (magic != NBD_CLIENT_MAGIC) {
 375             LOG("Bad magic received");
 376             goto fail;
 377         }
 378     }
 379
 380     if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
 381         LOG("read failed");
 382         goto fail;
 383     }
 384     *size = be64_to_cpu(s);
 385     *blocksize = 1024;
 386     TRACE("Size is %" PRIu64, *size);
 387
 388     if (!name) {
 389         if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
 390             LOG("read failed (flags)");
 391             goto fail;
 392         }
 393         *flags = be32_to_cpup(flags);
 394     } else {
 395         if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 396             LOG("read failed (tmp)");
 397             goto fail;
 398         }
 399         *flags |= be32_to_cpu(tmp);
 400     }
 401     if (read_sync(csock, &buf, 124) != 124) {
 402         LOG("read failed (buf)");
 403         goto fail;
 404     }
 405     rc = 0;
 406
 407 fail:
 408     socket_set_nonblock(csock);
 409     return rc;
 410 }
 411
 412 #ifdef __linux__
 413 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 414 {
 415     TRACE("Setting NBD socket");
 416
 417     if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
 418         int serrno = errno;
 419         LOG("Failed to set NBD socket");
 420         return -serrno;
 421     }
 422
 423     TRACE("Setting block size to %lu", (unsigned long)blocksize);
 424
 425     if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
 426         int serrno = errno;
 427         LOG("Failed setting NBD block size");
 428         return -serrno;
 429     }
 430
 431         TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
 432
 433     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
 434         int serrno = errno;
 435         LOG("Failed setting size (in blocks)");
 436         return -serrno;
 437     }
 438
 439     if (flags & NBD_FLAG_READ_ONLY) {
 440         int read_only = 1;
 441         TRACE("Setting readonly attribute");
 442
 443         if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 444             int serrno = errno;
 445             LOG("Failed setting read-only attribute");
 446             return -serrno;
 447         }
 448     }
 449
 450     if (ioctl(fd, NBD_SET_FLAGS, flags) < 0
 451         && errno != ENOTTY) {
 452         int serrno = errno;
 453         LOG("Failed setting flags");
 454         return -serrno;
 455     }
 456
 457     TRACE("Negotiation ended");
 458
 459     return 0;
 460 }
 461
 462 int nbd_disconnect(int fd)
 463 {
 464     ioctl(fd, NBD_CLEAR_QUE);
 465     ioctl(fd, NBD_DISCONNECT);
 466     ioctl(fd, NBD_CLEAR_SOCK);
 467     return 0;
 468 }
 469
 470 int nbd_client(int fd)
 471 {
 472     int ret;
 473     int serrno;
 474
 475     TRACE("Doing NBD loop");
 476
 477     ret = ioctl(fd, NBD_DO_IT);
 478     if (ret < 0 && errno == EPIPE) {
 479         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 480          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 481          * that case.
 482          */
 483         ret = 0;
 484     }
 485     serrno = errno;
 486
 487     TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
 488
 489     TRACE("Clearing NBD queue");
 490     ioctl(fd, NBD_CLEAR_QUE);
 491
 492     TRACE("Clearing NBD socket");
 493     ioctl(fd, NBD_CLEAR_SOCK);
 494
 495     errno = serrno;
 496     return ret;
 497 }
 498 #else
 499 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 500 {
 501     return -ENOTSUP;
 502 }
 503
 504 int nbd_disconnect(int fd)
 505 {
 506     return -ENOTSUP;
 507 }
 508
 509 int nbd_client(int fd)
 510 {
 511     return -ENOTSUP;
 512 }
 513 #endif
 514
 515 ssize_t nbd_send_request(int csock, struct nbd_request *request)
 516 {
 517     uint8_t buf[NBD_REQUEST_SIZE];
 518     ssize_t ret;
 519
 520     cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
 521     cpu_to_be32w((uint32_t*)(buf + 4), request->type);
 522     cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
 523     cpu_to_be64w((uint64_t*)(buf + 16), request->from);
 524     cpu_to_be32w((uint32_t*)(buf + 24), request->len);
 525
 526     TRACE("Sending request to client: "
 527           "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
 528           request->from, request->len, request->handle, request->type);
 529
 530     ret = write_sync(csock, buf, sizeof(buf));
 531     if (ret < 0) {
 532         return ret;
 533     }
 534
 535     if (ret != sizeof(buf)) {
 536         LOG("writing to socket failed");
 537         return -EINVAL;
 538     }
 539     return 0;
 540 }
 541
 542 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
 543 {
 544     uint8_t buf[NBD_REQUEST_SIZE];
 545     uint32_t magic;
 546     ssize_t ret;
 547
 548     ret = read_sync(csock, buf, sizeof(buf));
 549     if (ret < 0) {
 550         return ret;
 551     }
 552
 553     if (ret != sizeof(buf)) {
 554         LOG("read failed");
 555         return -EINVAL;
 556     }
 557
 558     /* Request
 559        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
 560        [ 4 ..  7]   type    (0 == READ, 1 == WRITE)
 561        [ 8 .. 15]   handle
 562        [16 .. 23]   from
 563        [24 .. 27]   len
 564      */
 565
 566     magic = be32_to_cpup((uint32_t*)buf);
 567     request->type  = be32_to_cpup((uint32_t*)(buf + 4));
 568     request->handle = be64_to_cpup((uint64_t*)(buf + 8));
 569     request->from  = be64_to_cpup((uint64_t*)(buf + 16));
 570     request->len   = be32_to_cpup((uint32_t*)(buf + 24));
 571
 572     TRACE("Got request: "
 573           "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
 574           magic, request->type, request->from, request->len);
 575
 576     if (magic != NBD_REQUEST_MAGIC) {
 577         LOG("invalid magic (got 0x%x)", magic);
 578         return -EINVAL;
 579     }
 580     return 0;
 581 }
 582
 583 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
 584 {
 585     uint8_t buf[NBD_REPLY_SIZE];
 586     uint32_t magic;
 587     ssize_t ret;
 588
 589     ret = read_sync(csock, buf, sizeof(buf));
 590     if (ret < 0) {
 591         return ret;
 592     }
 593
 594     if (ret != sizeof(buf)) {
 595         LOG("read failed");
 596         return -EINVAL;
 597     }
 598
 599     /* Reply
 600        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 601        [ 4 ..  7]    error   (0 == no error)
 602        [ 7 .. 15]    handle
 603      */
 604
 605     magic = be32_to_cpup((uint32_t*)buf);
 606     reply->error  = be32_to_cpup((uint32_t*)(buf + 4));
 607     reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
 608
 609     TRACE("Got reply: "
 610           "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
 611           magic, reply->error, reply->handle);
 612
 613     if (magic != NBD_REPLY_MAGIC) {
 614         LOG("invalid magic (got 0x%x)", magic);
 615         return -EINVAL;
 616     }
 617     return 0;
 618 }
 619
 620 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
 621 {
 622     uint8_t buf[NBD_REPLY_SIZE];
 623     ssize_t ret;
 624
 625     /* Reply
 626        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 627        [ 4 ..  7]    error   (0 == no error)
 628        [ 7 .. 15]    handle
 629      */
 630     cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
 631     cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
 632     cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
 633
 634     TRACE("Sending response to client");
 635
 636     ret = write_sync(csock, buf, sizeof(buf));
 637     if (ret < 0) {
 638         return ret;
 639     }
 640
 641     if (ret != sizeof(buf)) {
 642         LOG("writing to socket failed");
 643         return -EINVAL;
 644     }
 645     return 0;
 646 }
 647
 648 #define MAX_NBD_REQUESTS 16
 649
 650 static void nbd_client_get(NBDClient *client)
 651 {
 652     client->refcount++;
 653 }
 654
 655 static void nbd_client_put(NBDClient *client)
 656 {
 657     if (--client->refcount == 0) {
 658         g_free(client);
 659     }
 660 }
 661
 662 static void nbd_client_close(NBDClient *client)
 663 {
 664     qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
 665     close(client->sock);
 666     client->sock = -1;
 667     if (client->close) {
 668         client->close(client);
 669     }
 670     nbd_client_put(client);
 671 }
 672
 673 static NBDRequest *nbd_request_get(NBDClient *client)
 674 {
 675     NBDRequest *req;
 676     NBDExport *exp = client->exp;
 677
 678     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
 679     client->nb_requests++;
 680
 681     if (QSIMPLEQ_EMPTY(&exp->requests)) {
 682         req = g_malloc0(sizeof(NBDRequest));
 683         req->data = qemu_blockalign(exp->bs, NBD_BUFFER_SIZE);
 684     } else {
 685         req = QSIMPLEQ_FIRST(&exp->requests);
 686         QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
 687     }
 688     nbd_client_get(client);
 689     req->client = client;
 690     return req;
 691 }
 692
 693 static void nbd_request_put(NBDRequest *req)
 694 {
 695     NBDClient *client = req->client;
 696     QSIMPLEQ_INSERT_HEAD(&client->exp->requests, req, entry);
 697     if (client->nb_requests-- == MAX_NBD_REQUESTS) {
 698         qemu_notify_event();
 699     }
 700     nbd_client_put(client);
 701 }
 702
 703 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
 704                           off_t size, uint32_t nbdflags)
 705 {
 706     NBDExport *exp = g_malloc0(sizeof(NBDExport));
 707     QSIMPLEQ_INIT(&exp->requests);
 708     exp->bs = bs;
 709     exp->dev_offset = dev_offset;
 710     exp->nbdflags = nbdflags;
 711     exp->size = size == -1 ? bdrv_getlength(bs) : size;
 712     return exp;
 713 }
 714
 715 void nbd_export_close(NBDExport *exp)
 716 {
 717     while (!QSIMPLEQ_EMPTY(&exp->requests)) {
 718         NBDRequest *first = QSIMPLEQ_FIRST(&exp->requests);
 719         QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
 720         qemu_vfree(first->data);
 721         g_free(first);
 722     }
 723
 724     bdrv_close(exp->bs);
 725     g_free(exp);
 726 }
 727
 728 static int nbd_can_read(void *opaque);
 729 static void nbd_read(void *opaque);
 730 static void nbd_restart_write(void *opaque);
 731
 732 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
 733                                  int len)
 734 {
 735     NBDClient *client = req->client;
 736     int csock = client->sock;
 737     ssize_t rc, ret;
 738
 739     qemu_co_mutex_lock(&client->send_lock);
 740     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
 741                          nbd_restart_write, client);
 742     client->send_coroutine = qemu_coroutine_self();
 743
 744     if (!len) {
 745         rc = nbd_send_reply(csock, reply);
 746     } else {
 747         socket_set_cork(csock, 1);
 748         rc = nbd_send_reply(csock, reply);
 749         if (rc >= 0) {
 750             ret = qemu_co_send(csock, req->data, len);
 751             if (ret != len) {
 752                 rc = -EIO;
 753             }
 754         }
 755         socket_set_cork(csock, 0);
 756     }
 757
 758     client->send_coroutine = NULL;
 759     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
 760     qemu_co_mutex_unlock(&client->send_lock);
 761     return rc;
 762 }
 763
 764 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
 765 {
 766     NBDClient *client = req->client;
 767     int csock = client->sock;
 768     ssize_t rc;
 769
 770     client->recv_coroutine = qemu_coroutine_self();
 771     rc = nbd_receive_request(csock, request);
 772     if (rc < 0) {
 773         if (rc != -EAGAIN) {
 774             rc = -EIO;
 775         }
 776         goto out;
 777     }
 778
 779     if (request->len > NBD_BUFFER_SIZE) {
 780         LOG("len (%u) is larger than max len (%u)",
 781             request->len, NBD_BUFFER_SIZE);
 782         rc = -EINVAL;
 783         goto out;
 784     }
 785
 786     if ((request->from + request->len) < request->from) {
 787         LOG("integer overflow detected! "
 788             "you're probably being attacked");
 789         rc = -EINVAL;
 790         goto out;
 791     }
 792
 793     TRACE("Decoding type");
 794
 795     if ((request->type & NBD_CMD_MASK_COMMAND) == NBD_CMD_WRITE) {
 796         TRACE("Reading %u byte(s)", request->len);
 797
 798         if (qemu_co_recv(csock, req->data, request->len) != request->len) {
 799             LOG("reading from socket failed");
 800             rc = -EIO;
 801             goto out;
 802         }
 803     }
 804     rc = 0;
 805
 806 out:
 807     client->recv_coroutine = NULL;
 808     return rc;
 809 }
 810
 811 static void nbd_trip(void *opaque)
 812 {
 813     NBDClient *client = opaque;
 814     NBDRequest *req = nbd_request_get(client);
 815     NBDExport *exp = client->exp;
 816     struct nbd_request request;
 817     struct nbd_reply reply;
 818     ssize_t ret;
 819
 820     TRACE("Reading request.");
 821
 822     ret = nbd_co_receive_request(req, &request);
 823     if (ret == -EAGAIN) {
 824         goto done;
 825     }
 826     if (ret == -EIO) {
 827         goto out;
 828     }
 829
 830     reply.handle = request.handle;
 831     reply.error = 0;
 832
 833     if (ret < 0) {
 834         reply.error = -ret;
 835         goto error_reply;
 836     }
 837
 838     if ((request.from + request.len) > exp->size) {
 839             LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
 840             ", Offset: %" PRIu64 "\n",
 841                     request.from, request.len,
 842                     (uint64_t)exp->size, (uint64_t)exp->dev_offset);
 843         LOG("requested operation past EOF--bad client?");
 844         goto invalid_request;
 845     }
 846
 847     switch (request.type & NBD_CMD_MASK_COMMAND) {
 848     case NBD_CMD_READ:
 849         TRACE("Request type is READ");
 850
 851         if (request.type & NBD_CMD_FLAG_FUA) {
 852             ret = bdrv_co_flush(exp->bs);
 853             if (ret < 0) {
 854                 LOG("flush failed");
 855                 reply.error = -ret;
 856                 goto error_reply;
 857             }
 858         }
 859
 860         ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
 861                         req->data, request.len / 512);
 862         if (ret < 0) {
 863             LOG("reading from file failed");
 864             reply.error = -ret;
 865             goto error_reply;
 866         }
 867
 868         TRACE("Read %u byte(s)", request.len);
 869         if (nbd_co_send_reply(req, &reply, request.len) < 0)
 870             goto out;
 871         break;
 872     case NBD_CMD_WRITE:
 873         TRACE("Request type is WRITE");
 874
 875         if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
 876             TRACE("Server is read-only, return error");
 877             reply.error = EROFS;
 878             goto error_reply;
 879         }
 880
 881         TRACE("Writing to device");
 882
 883         ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
 884                          req->data, request.len / 512);
 885         if (ret < 0) {
 886             LOG("writing to file failed");
 887             reply.error = -ret;
 888             goto error_reply;
 889         }
 890
 891         if (request.type & NBD_CMD_FLAG_FUA) {
 892             ret = bdrv_co_flush(exp->bs);
 893             if (ret < 0) {
 894                 LOG("flush failed");
 895                 reply.error = -ret;
 896                 goto error_reply;
 897             }
 898         }
 899
 900         if (nbd_co_send_reply(req, &reply, 0) < 0) {
 901             goto out;
 902         }
 903         break;
 904     case NBD_CMD_DISC:
 905         TRACE("Request type is DISCONNECT");
 906         errno = 0;
 907         goto out;
 908     case NBD_CMD_FLUSH:
 909         TRACE("Request type is FLUSH");
 910
 911         ret = bdrv_co_flush(exp->bs);
 912         if (ret < 0) {
 913             LOG("flush failed");
 914             reply.error = -ret;
 915         }
 916         if (nbd_co_send_reply(req, &reply, 0) < 0) {
 917             goto out;
 918         }
 919         break;
 920     case NBD_CMD_TRIM:
 921         TRACE("Request type is TRIM");
 922         ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
 923                               request.len / 512);
 924         if (ret < 0) {
 925             LOG("discard failed");
 926             reply.error = -ret;
 927         }
 928         if (nbd_co_send_reply(req, &reply, 0) < 0) {
 929             goto out;
 930         }
 931         break;
 932     default:
 933         LOG("invalid request type (%u) received", request.type);
 934     invalid_request:
 935         reply.error = -EINVAL;
 936     error_reply:
 937         if (nbd_co_send_reply(req, &reply, 0) < 0) {
 938             goto out;
 939         }
 940         break;
 941     }
 942
 943     TRACE("Request/Reply complete");
 944
 945 done:
 946     nbd_request_put(req);
 947     return;
 948
 949 out:
 950     nbd_request_put(req);
 951     nbd_client_close(client);
 952 }
 953
 954 static int nbd_can_read(void *opaque)
 955 {
 956     NBDClient *client = opaque;
 957
 958     return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
 959 }
 960
 961 static void nbd_read(void *opaque)
 962 {
 963     NBDClient *client = opaque;
 964
 965     if (client->recv_coroutine) {
 966         qemu_coroutine_enter(client->recv_coroutine, NULL);
 967     } else {
 968         qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
 969     }
 970 }
 971
 972 static void nbd_restart_write(void *opaque)
 973 {
 974     NBDClient *client = opaque;
 975
 976     qemu_coroutine_enter(client->send_coroutine, NULL);
 977 }
 978
 979 NBDClient *nbd_client_new(NBDExport *exp, int csock,
 980                           void (*close)(NBDClient *))
 981 {
 982     NBDClient *client;
 983     client = g_malloc0(sizeof(NBDClient));
 984     client->refcount = 1;
 985     client->exp = exp;
 986     client->sock = csock;
 987     if (nbd_send_negotiate(client) < 0) {
 988         g_free(client);
 989         return NULL;
 990     }
 991     client->close = close;
 992     qemu_co_mutex_init(&client->send_lock);
 993     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
 994     return client;
 995 }