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 "block/nbd.h"
  20 #include "block/block.h"
  21
  22 #include "block/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/sockets.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     int refcount;
  93     void (*close)(NBDExport *exp);
  94
  95     BlockDriverState *bs;
  96     char *name;
  97     off_t dev_offset;
  98     off_t size;
  99     uint32_t nbdflags;
 100     QTAILQ_HEAD(, NBDClient) clients;
 101     QTAILQ_ENTRY(NBDExport) next;
 102 };
 103
 104 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
 105
 106 struct NBDClient {
 107     int refcount;
 108     void (*close)(NBDClient *client);
 109
 110     NBDExport *exp;
 111     int sock;
 112
 113     Coroutine *recv_coroutine;
 114
 115     CoMutex send_lock;
 116     Coroutine *send_coroutine;
 117
 118     QTAILQ_ENTRY(NBDClient) next;
 119     int nb_requests;
 120     bool closing;
 121 };
 122
 123 /* That's all folks */
 124
 125 ssize_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
 126 {
 127     size_t offset = 0;
 128     int err;
 129
 130     if (qemu_in_coroutine()) {
 131         if (do_read) {
 132             return qemu_co_recv(fd, buffer, size);
 133         } else {
 134             return qemu_co_send(fd, buffer, size);
 135         }
 136     }
 137
 138     while (offset < size) {
 139         ssize_t len;
 140
 141         if (do_read) {
 142             len = qemu_recv(fd, buffer + offset, size - offset, 0);
 143         } else {
 144             len = send(fd, buffer + offset, size - offset, 0);
 145         }
 146
 147         if (len < 0) {
 148             err = socket_error();
 149
 150             /* recoverable error */
 151             if (err == EINTR || (offset > 0 && err == EAGAIN)) {
 152                 continue;
 153             }
 154
 155             /* unrecoverable error */
 156             return -err;
 157         }
 158
 159         /* eof */
 160         if (len == 0) {
 161             break;
 162         }
 163
 164         offset += len;
 165     }
 166
 167     return offset;
 168 }
 169
 170 static ssize_t read_sync(int fd, void *buffer, size_t size)
 171 {
 172     /* Sockets are kept in blocking mode in the negotiation phase.  After
 173      * that, a non-readable socket simply means that another thread stole
 174      * our request/reply.  Synchronization is done with recv_coroutine, so
 175      * that this is coroutine-safe.
 176      */
 177     return nbd_wr_sync(fd, buffer, size, true);
 178 }
 179
 180 static ssize_t write_sync(int fd, void *buffer, size_t size)
 181 {
 182     int ret;
 183     do {
 184         /* For writes, we do expect the socket to be writable.  */
 185         ret = nbd_wr_sync(fd, buffer, size, false);
 186     } while (ret == -EAGAIN);
 187     return ret;
 188 }
 189
 190 static void combine_addr(char *buf, size_t len, const char* address,
 191                          uint16_t port)
 192 {
 193     /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
 194     if (strstr(address, ":")) {
 195         snprintf(buf, len, "[%s]:%u", address, port);
 196     } else {
 197         snprintf(buf, len, "%s:%u", address, port);
 198     }
 199 }
 200
 201 int tcp_socket_outgoing_opts(QemuOpts *opts)
 202 {
 203     Error *local_err = NULL;
 204     int fd = inet_connect_opts(opts, &local_err, NULL, NULL);
 205     if (local_err != NULL) {
 206         qerror_report_err(local_err);
 207         error_free(local_err);
 208     }
 209
 210     return fd;
 211 }
 212
 213 int tcp_socket_incoming(const char *address, uint16_t port)
 214 {
 215     char address_and_port[128];
 216     combine_addr(address_and_port, 128, address, port);
 217     return tcp_socket_incoming_spec(address_and_port);
 218 }
 219
 220 int tcp_socket_incoming_spec(const char *address_and_port)
 221 {
 222     Error *local_err = NULL;
 223     int fd = inet_listen(address_and_port, NULL, 0, SOCK_STREAM, 0, &local_err);
 224
 225     if (local_err != NULL) {
 226         qerror_report_err(local_err);
 227         error_free(local_err);
 228     }
 229     return fd;
 230 }
 231
 232 int unix_socket_incoming(const char *path)
 233 {
 234     Error *local_err = NULL;
 235     int fd = unix_listen(path, NULL, 0, &local_err);
 236
 237     if (local_err != NULL) {
 238         qerror_report_err(local_err);
 239         error_free(local_err);
 240     }
 241     return fd;
 242 }
 243
 244 int unix_socket_outgoing(const char *path)
 245 {
 246     Error *local_err = NULL;
 247     int fd = unix_connect(path, &local_err);
 248
 249     if (local_err != NULL) {
 250         qerror_report_err(local_err);
 251         error_free(local_err);
 252     }
 253     return fd;
 254 }
 255
 256 /* Basic flow for negotiation
 257
 258    Server         Client
 259    Negotiate
 260
 261    or
 262
 263    Server         Client
 264    Negotiate #1
 265                   Option
 266    Negotiate #2
 267
 268    ----
 269
 270    followed by
 271
 272    Server         Client
 273                   Request
 274    Response
 275                   Request
 276    Response
 277                   ...
 278    ...
 279                   Request (type == 2)
 280
 281 */
 282
 283 static int nbd_receive_options(NBDClient *client)
 284 {
 285     int csock = client->sock;
 286     char name[256];
 287     uint32_t tmp, length;
 288     uint64_t magic;
 289     int rc;
 290
 291     /* Client sends:
 292         [ 0 ..   3]   reserved (0)
 293         [ 4 ..  11]   NBD_OPTS_MAGIC
 294         [12 ..  15]   NBD_OPT_EXPORT_NAME
 295         [16 ..  19]   length
 296         [20 ..  xx]   export name (length bytes)
 297      */
 298
 299     rc = -EINVAL;
 300     if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 301         LOG("read failed");
 302         goto fail;
 303     }
 304     TRACE("Checking reserved");
 305     if (tmp != 0) {
 306         LOG("Bad reserved received");
 307         goto fail;
 308     }
 309
 310     if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 311         LOG("read failed");
 312         goto fail;
 313     }
 314     TRACE("Checking reserved");
 315     if (magic != be64_to_cpu(NBD_OPTS_MAGIC)) {
 316         LOG("Bad magic received");
 317         goto fail;
 318     }
 319
 320     if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 321         LOG("read failed");
 322         goto fail;
 323     }
 324     TRACE("Checking option");
 325     if (tmp != be32_to_cpu(NBD_OPT_EXPORT_NAME)) {
 326         LOG("Bad option received");
 327         goto fail;
 328     }
 329
 330     if (read_sync(csock, &length, sizeof(length)) != sizeof(length)) {
 331         LOG("read failed");
 332         goto fail;
 333     }
 334     TRACE("Checking length");
 335     length = be32_to_cpu(length);
 336     if (length > 255) {
 337         LOG("Bad length received");
 338         goto fail;
 339     }
 340     if (read_sync(csock, name, length) != length) {
 341         LOG("read failed");
 342         goto fail;
 343     }
 344     name[length] = '\0';
 345
 346     client->exp = nbd_export_find(name);
 347     if (!client->exp) {
 348         LOG("export not found");
 349         goto fail;
 350     }
 351
 352     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
 353     nbd_export_get(client->exp);
 354
 355     TRACE("Option negotiation succeeded.");
 356     rc = 0;
 357 fail:
 358     return rc;
 359 }
 360
 361 static int nbd_send_negotiate(NBDClient *client)
 362 {
 363     int csock = client->sock;
 364     char buf[8 + 8 + 8 + 128];
 365     int rc;
 366     const int myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
 367                          NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
 368
 369     /* Negotiation header without options:
 370         [ 0 ..   7]   passwd       ("NBDMAGIC")
 371         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
 372         [16 ..  23]   size
 373         [24 ..  25]   server flags (0)
 374         [24 ..  27]   export flags
 375         [28 .. 151]   reserved     (0)
 376
 377        Negotiation header with options, part 1:
 378         [ 0 ..   7]   passwd       ("NBDMAGIC")
 379         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
 380         [16 ..  17]   server flags (0)
 381
 382        part 2 (after options are sent):
 383         [18 ..  25]   size
 384         [26 ..  27]   export flags
 385         [28 .. 151]   reserved     (0)
 386      */
 387
 388     qemu_set_block(csock);
 389     rc = -EINVAL;
 390
 391     TRACE("Beginning negotiation.");
 392     memset(buf, 0, sizeof(buf));
 393     memcpy(buf, "NBDMAGIC", 8);
 394     if (client->exp) {
 395         assert ((client->exp->nbdflags & ~65535) == 0);
 396         cpu_to_be64w((uint64_t*)(buf + 8), NBD_CLIENT_MAGIC);
 397         cpu_to_be64w((uint64_t*)(buf + 16), client->exp->size);
 398         cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
 399     } else {
 400         cpu_to_be64w((uint64_t*)(buf + 8), NBD_OPTS_MAGIC);
 401     }
 402
 403     if (client->exp) {
 404         if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
 405             LOG("write failed");
 406             goto fail;
 407         }
 408     } else {
 409         if (write_sync(csock, buf, 18) != 18) {
 410             LOG("write failed");
 411             goto fail;
 412         }
 413         rc = nbd_receive_options(client);
 414         if (rc < 0) {
 415             LOG("option negotiation failed");
 416             goto fail;
 417         }
 418
 419         assert ((client->exp->nbdflags & ~65535) == 0);
 420         cpu_to_be64w((uint64_t*)(buf + 18), client->exp->size);
 421         cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
 422         if (write_sync(csock, buf + 18, sizeof(buf) - 18) != sizeof(buf) - 18) {
 423             LOG("write failed");
 424             goto fail;
 425         }
 426     }
 427
 428     TRACE("Negotiation succeeded.");
 429     rc = 0;
 430 fail:
 431     qemu_set_nonblock(csock);
 432     return rc;
 433 }
 434
 435 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
 436                           off_t *size, size_t *blocksize)
 437 {
 438     char buf[256];
 439     uint64_t magic, s;
 440     uint16_t tmp;
 441     int rc;
 442
 443     TRACE("Receiving negotiation.");
 444
 445     qemu_set_block(csock);
 446     rc = -EINVAL;
 447
 448     if (read_sync(csock, buf, 8) != 8) {
 449         LOG("read failed");
 450         goto fail;
 451     }
 452
 453     buf[8] = '\0';
 454     if (strlen(buf) == 0) {
 455         LOG("server connection closed");
 456         goto fail;
 457     }
 458
 459     TRACE("Magic is %c%c%c%c%c%c%c%c",
 460           qemu_isprint(buf[0]) ? buf[0] : '.',
 461           qemu_isprint(buf[1]) ? buf[1] : '.',
 462           qemu_isprint(buf[2]) ? buf[2] : '.',
 463           qemu_isprint(buf[3]) ? buf[3] : '.',
 464           qemu_isprint(buf[4]) ? buf[4] : '.',
 465           qemu_isprint(buf[5]) ? buf[5] : '.',
 466           qemu_isprint(buf[6]) ? buf[6] : '.',
 467           qemu_isprint(buf[7]) ? buf[7] : '.');
 468
 469     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 470         LOG("Invalid magic received");
 471         goto fail;
 472     }
 473
 474     if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 475         LOG("read failed");
 476         goto fail;
 477     }
 478     magic = be64_to_cpu(magic);
 479     TRACE("Magic is 0x%" PRIx64, magic);
 480
 481     if (name) {
 482         uint32_t reserved = 0;
 483         uint32_t opt;
 484         uint32_t namesize;
 485
 486         TRACE("Checking magic (opts_magic)");
 487         if (magic != NBD_OPTS_MAGIC) {
 488             LOG("Bad magic received");
 489             goto fail;
 490         }
 491         if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 492             LOG("flags read failed");
 493             goto fail;
 494         }
 495         *flags = be16_to_cpu(tmp) << 16;
 496         /* reserved for future use */
 497         if (write_sync(csock, &reserved, sizeof(reserved)) !=
 498             sizeof(reserved)) {
 499             LOG("write failed (reserved)");
 500             goto fail;
 501         }
 502         /* write the export name */
 503         magic = cpu_to_be64(magic);
 504         if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 505             LOG("write failed (magic)");
 506             goto fail;
 507         }
 508         opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
 509         if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
 510             LOG("write failed (opt)");
 511             goto fail;
 512         }
 513         namesize = cpu_to_be32(strlen(name));
 514         if (write_sync(csock, &namesize, sizeof(namesize)) !=
 515             sizeof(namesize)) {
 516             LOG("write failed (namesize)");
 517             goto fail;
 518         }
 519         if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
 520             LOG("write failed (name)");
 521             goto fail;
 522         }
 523     } else {
 524         TRACE("Checking magic (cli_magic)");
 525
 526         if (magic != NBD_CLIENT_MAGIC) {
 527             LOG("Bad magic received");
 528             goto fail;
 529         }
 530     }
 531
 532     if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
 533         LOG("read failed");
 534         goto fail;
 535     }
 536     *size = be64_to_cpu(s);
 537     *blocksize = 1024;
 538     TRACE("Size is %" PRIu64, *size);
 539
 540     if (!name) {
 541         if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
 542             LOG("read failed (flags)");
 543             goto fail;
 544         }
 545         *flags = be32_to_cpup(flags);
 546     } else {
 547         if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 548             LOG("read failed (tmp)");
 549             goto fail;
 550         }
 551         *flags |= be32_to_cpu(tmp);
 552     }
 553     if (read_sync(csock, &buf, 124) != 124) {
 554         LOG("read failed (buf)");
 555         goto fail;
 556     }
 557     rc = 0;
 558
 559 fail:
 560     qemu_set_nonblock(csock);
 561     return rc;
 562 }
 563
 564 #ifdef __linux__
 565 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 566 {
 567     TRACE("Setting NBD socket");
 568
 569     if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
 570         int serrno = errno;
 571         LOG("Failed to set NBD socket");
 572         return -serrno;
 573     }
 574
 575     TRACE("Setting block size to %lu", (unsigned long)blocksize);
 576
 577     if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
 578         int serrno = errno;
 579         LOG("Failed setting NBD block size");
 580         return -serrno;
 581     }
 582
 583         TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
 584
 585     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
 586         int serrno = errno;
 587         LOG("Failed setting size (in blocks)");
 588         return -serrno;
 589     }
 590
 591     if (ioctl(fd, NBD_SET_FLAGS, flags) < 0) {
 592         if (errno == ENOTTY) {
 593             int read_only = (flags & NBD_FLAG_READ_ONLY) != 0;
 594             TRACE("Setting readonly attribute");
 595
 596             if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 597                 int serrno = errno;
 598                 LOG("Failed setting read-only attribute");
 599                 return -serrno;
 600             }
 601         } else {
 602             int serrno = errno;
 603             LOG("Failed setting flags");
 604             return -serrno;
 605         }
 606     }
 607
 608     TRACE("Negotiation ended");
 609
 610     return 0;
 611 }
 612
 613 int nbd_disconnect(int fd)
 614 {
 615     ioctl(fd, NBD_CLEAR_QUE);
 616     ioctl(fd, NBD_DISCONNECT);
 617     ioctl(fd, NBD_CLEAR_SOCK);
 618     return 0;
 619 }
 620
 621 int nbd_client(int fd)
 622 {
 623     int ret;
 624     int serrno;
 625
 626     TRACE("Doing NBD loop");
 627
 628     ret = ioctl(fd, NBD_DO_IT);
 629     if (ret < 0 && errno == EPIPE) {
 630         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 631          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 632          * that case.
 633          */
 634         ret = 0;
 635     }
 636     serrno = errno;
 637
 638     TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
 639
 640     TRACE("Clearing NBD queue");
 641     ioctl(fd, NBD_CLEAR_QUE);
 642
 643     TRACE("Clearing NBD socket");
 644     ioctl(fd, NBD_CLEAR_SOCK);
 645
 646     errno = serrno;
 647     return ret;
 648 }
 649 #else
 650 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 651 {
 652     return -ENOTSUP;
 653 }
 654
 655 int nbd_disconnect(int fd)
 656 {
 657     return -ENOTSUP;
 658 }
 659
 660 int nbd_client(int fd)
 661 {
 662     return -ENOTSUP;
 663 }
 664 #endif
 665
 666 ssize_t nbd_send_request(int csock, struct nbd_request *request)
 667 {
 668     uint8_t buf[NBD_REQUEST_SIZE];
 669     ssize_t ret;
 670
 671     cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
 672     cpu_to_be32w((uint32_t*)(buf + 4), request->type);
 673     cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
 674     cpu_to_be64w((uint64_t*)(buf + 16), request->from);
 675     cpu_to_be32w((uint32_t*)(buf + 24), request->len);
 676
 677     TRACE("Sending request to client: "
 678           "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
 679           request->from, request->len, request->handle, request->type);
 680
 681     ret = write_sync(csock, buf, sizeof(buf));
 682     if (ret < 0) {
 683         return ret;
 684     }
 685
 686     if (ret != sizeof(buf)) {
 687         LOG("writing to socket failed");
 688         return -EINVAL;
 689     }
 690     return 0;
 691 }
 692
 693 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
 694 {
 695     uint8_t buf[NBD_REQUEST_SIZE];
 696     uint32_t magic;
 697     ssize_t ret;
 698
 699     ret = read_sync(csock, buf, sizeof(buf));
 700     if (ret < 0) {
 701         return ret;
 702     }
 703
 704     if (ret != sizeof(buf)) {
 705         LOG("read failed");
 706         return -EINVAL;
 707     }
 708
 709     /* Request
 710        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
 711        [ 4 ..  7]   type    (0 == READ, 1 == WRITE)
 712        [ 8 .. 15]   handle
 713        [16 .. 23]   from
 714        [24 .. 27]   len
 715      */
 716
 717     magic = be32_to_cpup((uint32_t*)buf);
 718     request->type  = be32_to_cpup((uint32_t*)(buf + 4));
 719     request->handle = be64_to_cpup((uint64_t*)(buf + 8));
 720     request->from  = be64_to_cpup((uint64_t*)(buf + 16));
 721     request->len   = be32_to_cpup((uint32_t*)(buf + 24));
 722
 723     TRACE("Got request: "
 724           "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
 725           magic, request->type, request->from, request->len);
 726
 727     if (magic != NBD_REQUEST_MAGIC) {
 728         LOG("invalid magic (got 0x%x)", magic);
 729         return -EINVAL;
 730     }
 731     return 0;
 732 }
 733
 734 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
 735 {
 736     uint8_t buf[NBD_REPLY_SIZE];
 737     uint32_t magic;
 738     ssize_t ret;
 739
 740     ret = read_sync(csock, buf, sizeof(buf));
 741     if (ret < 0) {
 742         return ret;
 743     }
 744
 745     if (ret != sizeof(buf)) {
 746         LOG("read failed");
 747         return -EINVAL;
 748     }
 749
 750     /* Reply
 751        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 752        [ 4 ..  7]    error   (0 == no error)
 753        [ 7 .. 15]    handle
 754      */
 755
 756     magic = be32_to_cpup((uint32_t*)buf);
 757     reply->error  = be32_to_cpup((uint32_t*)(buf + 4));
 758     reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
 759
 760     TRACE("Got reply: "
 761           "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
 762           magic, reply->error, reply->handle);
 763
 764     if (magic != NBD_REPLY_MAGIC) {
 765         LOG("invalid magic (got 0x%x)", magic);
 766         return -EINVAL;
 767     }
 768     return 0;
 769 }
 770
 771 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
 772 {
 773     uint8_t buf[NBD_REPLY_SIZE];
 774     ssize_t ret;
 775
 776     /* Reply
 777        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 778        [ 4 ..  7]    error   (0 == no error)
 779        [ 7 .. 15]    handle
 780      */
 781     cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
 782     cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
 783     cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
 784
 785     TRACE("Sending response to client");
 786
 787     ret = write_sync(csock, buf, sizeof(buf));
 788     if (ret < 0) {
 789         return ret;
 790     }
 791
 792     if (ret != sizeof(buf)) {
 793         LOG("writing to socket failed");
 794         return -EINVAL;
 795     }
 796     return 0;
 797 }
 798
 799 #define MAX_NBD_REQUESTS 16
 800
 801 void nbd_client_get(NBDClient *client)
 802 {
 803     client->refcount++;
 804 }
 805
 806 void nbd_client_put(NBDClient *client)
 807 {
 808     if (--client->refcount == 0) {
 809         /* The last reference should be dropped by client->close,
 810          * which is called by nbd_client_close.
 811          */
 812         assert(client->closing);
 813
 814         qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
 815         close(client->sock);
 816         client->sock = -1;
 817         if (client->exp) {
 818             QTAILQ_REMOVE(&client->exp->clients, client, next);
 819             nbd_export_put(client->exp);
 820         }
 821         g_free(client);
 822     }
 823 }
 824
 825 void nbd_client_close(NBDClient *client)
 826 {
 827     if (client->closing) {
 828         return;
 829     }
 830
 831     client->closing = true;
 832
 833     /* Force requests to finish.  They will drop their own references,
 834      * then we'll close the socket and free the NBDClient.
 835      */
 836     shutdown(client->sock, 2);
 837
 838     /* Also tell the client, so that they release their reference.  */
 839     if (client->close) {
 840         client->close(client);
 841     }
 842 }
 843
 844 static NBDRequest *nbd_request_get(NBDClient *client)
 845 {
 846     NBDRequest *req;
 847
 848     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
 849     client->nb_requests++;
 850
 851     req = g_slice_new0(NBDRequest);
 852     nbd_client_get(client);
 853     req->client = client;
 854     return req;
 855 }
 856
 857 static void nbd_request_put(NBDRequest *req)
 858 {
 859     NBDClient *client = req->client;
 860
 861     if (req->data) {
 862         qemu_vfree(req->data);
 863     }
 864     g_slice_free(NBDRequest, req);
 865
 866     if (client->nb_requests-- == MAX_NBD_REQUESTS) {
 867         qemu_notify_event();
 868     }
 869     nbd_client_put(client);
 870 }
 871
 872 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
 873                           off_t size, uint32_t nbdflags,
 874                           void (*close)(NBDExport *))
 875 {
 876     NBDExport *exp = g_malloc0(sizeof(NBDExport));
 877     exp->refcount = 1;
 878     QTAILQ_INIT(&exp->clients);
 879     exp->bs = bs;
 880     exp->dev_offset = dev_offset;
 881     exp->nbdflags = nbdflags;
 882     exp->size = size == -1 ? bdrv_getlength(bs) : size;
 883     exp->close = close;
 884     return exp;
 885 }
 886
 887 NBDExport *nbd_export_find(const char *name)
 888 {
 889     NBDExport *exp;
 890     QTAILQ_FOREACH(exp, &exports, next) {
 891         if (strcmp(name, exp->name) == 0) {
 892             return exp;
 893         }
 894     }
 895
 896     return NULL;
 897 }
 898
 899 void nbd_export_set_name(NBDExport *exp, const char *name)
 900 {
 901     if (exp->name == name) {
 902         return;
 903     }
 904
 905     nbd_export_get(exp);
 906     if (exp->name != NULL) {
 907         g_free(exp->name);
 908         exp->name = NULL;
 909         QTAILQ_REMOVE(&exports, exp, next);
 910         nbd_export_put(exp);
 911     }
 912     if (name != NULL) {
 913         nbd_export_get(exp);
 914         exp->name = g_strdup(name);
 915         QTAILQ_INSERT_TAIL(&exports, exp, next);
 916     }
 917     nbd_export_put(exp);
 918 }
 919
 920 void nbd_export_close(NBDExport *exp)
 921 {
 922     NBDClient *client, *next;
 923
 924     nbd_export_get(exp);
 925     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
 926         nbd_client_close(client);
 927     }
 928     nbd_export_set_name(exp, NULL);
 929     nbd_export_put(exp);
 930 }
 931
 932 void nbd_export_get(NBDExport *exp)
 933 {
 934     assert(exp->refcount > 0);
 935     exp->refcount++;
 936 }
 937
 938 void nbd_export_put(NBDExport *exp)
 939 {
 940     assert(exp->refcount > 0);
 941     if (exp->refcount == 1) {
 942         nbd_export_close(exp);
 943     }
 944
 945     if (--exp->refcount == 0) {
 946         assert(exp->name == NULL);
 947
 948         if (exp->close) {
 949             exp->close(exp);
 950         }
 951
 952         g_free(exp);
 953     }
 954 }
 955
 956 BlockDriverState *nbd_export_get_blockdev(NBDExport *exp)
 957 {
 958     return exp->bs;
 959 }
 960
 961 void nbd_export_close_all(void)
 962 {
 963     NBDExport *exp, *next;
 964
 965     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
 966         nbd_export_close(exp);
 967     }
 968 }
 969
 970 static int nbd_can_read(void *opaque);
 971 static void nbd_read(void *opaque);
 972 static void nbd_restart_write(void *opaque);
 973
 974 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
 975                                  int len)
 976 {
 977     NBDClient *client = req->client;
 978     int csock = client->sock;
 979     ssize_t rc, ret;
 980
 981     qemu_co_mutex_lock(&client->send_lock);
 982     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
 983                          nbd_restart_write, client);
 984     client->send_coroutine = qemu_coroutine_self();
 985
 986     if (!len) {
 987         rc = nbd_send_reply(csock, reply);
 988     } else {
 989         socket_set_cork(csock, 1);
 990         rc = nbd_send_reply(csock, reply);
 991         if (rc >= 0) {
 992             ret = qemu_co_send(csock, req->data, len);
 993             if (ret != len) {
 994                 rc = -EIO;
 995             }
 996         }
 997         socket_set_cork(csock, 0);
 998     }
 999
1000     client->send_coroutine = NULL;
1001     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1002     qemu_co_mutex_unlock(&client->send_lock);
1003     return rc;
1004 }
1005
1006 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
1007 {
1008     NBDClient *client = req->client;
1009     int csock = client->sock;
1010     uint32_t command;
1011     ssize_t rc;
1012
1013     client->recv_coroutine = qemu_coroutine_self();
1014     rc = nbd_receive_request(csock, request);
1015     if (rc < 0) {
1016         if (rc != -EAGAIN) {
1017             rc = -EIO;
1018         }
1019         goto out;
1020     }
1021
1022     if (request->len > NBD_MAX_BUFFER_SIZE) {
1023         LOG("len (%u) is larger than max len (%u)",
1024             request->len, NBD_MAX_BUFFER_SIZE);
1025         rc = -EINVAL;
1026         goto out;
1027     }
1028
1029     if ((request->from + request->len) < request->from) {
1030         LOG("integer overflow detected! "
1031             "you're probably being attacked");
1032         rc = -EINVAL;
1033         goto out;
1034     }
1035
1036     TRACE("Decoding type");
1037
1038     command = request->type & NBD_CMD_MASK_COMMAND;
1039     if (command == NBD_CMD_READ || command == NBD_CMD_WRITE) {
1040         req->data = qemu_blockalign(client->exp->bs, request->len);
1041     }
1042     if (command == NBD_CMD_WRITE) {
1043         TRACE("Reading %u byte(s)", request->len);
1044
1045         if (qemu_co_recv(csock, req->data, request->len) != request->len) {
1046             LOG("reading from socket failed");
1047             rc = -EIO;
1048             goto out;
1049         }
1050     }
1051     rc = 0;
1052
1053 out:
1054     client->recv_coroutine = NULL;
1055     return rc;
1056 }
1057
1058 static void nbd_trip(void *opaque)
1059 {
1060     NBDClient *client = opaque;
1061     NBDExport *exp = client->exp;
1062     NBDRequest *req;
1063     struct nbd_request request;
1064     struct nbd_reply reply;
1065     ssize_t ret;
1066
1067     TRACE("Reading request.");
1068     if (client->closing) {
1069         return;
1070     }
1071
1072     req = nbd_request_get(client);
1073     ret = nbd_co_receive_request(req, &request);
1074     if (ret == -EAGAIN) {
1075         goto done;
1076     }
1077     if (ret == -EIO) {
1078         goto out;
1079     }
1080
1081     reply.handle = request.handle;
1082     reply.error = 0;
1083
1084     if (ret < 0) {
1085         reply.error = -ret;
1086         goto error_reply;
1087     }
1088
1089     if ((request.from + request.len) > exp->size) {
1090             LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
1091             ", Offset: %" PRIu64 "\n",
1092                     request.from, request.len,
1093                     (uint64_t)exp->size, (uint64_t)exp->dev_offset);
1094         LOG("requested operation past EOF--bad client?");
1095         goto invalid_request;
1096     }
1097
1098     switch (request.type & NBD_CMD_MASK_COMMAND) {
1099     case NBD_CMD_READ:
1100         TRACE("Request type is READ");
1101
1102         if (request.type & NBD_CMD_FLAG_FUA) {
1103             ret = bdrv_co_flush(exp->bs);
1104             if (ret < 0) {
1105                 LOG("flush failed");
1106                 reply.error = -ret;
1107                 goto error_reply;
1108             }
1109         }
1110
1111         ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
1112                         req->data, request.len / 512);
1113         if (ret < 0) {
1114             LOG("reading from file failed");
1115             reply.error = -ret;
1116             goto error_reply;
1117         }
1118
1119         TRACE("Read %u byte(s)", request.len);
1120         if (nbd_co_send_reply(req, &reply, request.len) < 0)
1121             goto out;
1122         break;
1123     case NBD_CMD_WRITE:
1124         TRACE("Request type is WRITE");
1125
1126         if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1127             TRACE("Server is read-only, return error");
1128             reply.error = EROFS;
1129             goto error_reply;
1130         }
1131
1132         TRACE("Writing to device");
1133
1134         ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
1135                          req->data, request.len / 512);
1136         if (ret < 0) {
1137             LOG("writing to file failed");
1138             reply.error = -ret;
1139             goto error_reply;
1140         }
1141
1142         if (request.type & NBD_CMD_FLAG_FUA) {
1143             ret = bdrv_co_flush(exp->bs);
1144             if (ret < 0) {
1145                 LOG("flush failed");
1146                 reply.error = -ret;
1147                 goto error_reply;
1148             }
1149         }
1150
1151         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1152             goto out;
1153         }
1154         break;
1155     case NBD_CMD_DISC:
1156         TRACE("Request type is DISCONNECT");
1157         errno = 0;
1158         goto out;
1159     case NBD_CMD_FLUSH:
1160         TRACE("Request type is FLUSH");
1161
1162         ret = bdrv_co_flush(exp->bs);
1163         if (ret < 0) {
1164             LOG("flush failed");
1165             reply.error = -ret;
1166         }
1167         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1168             goto out;
1169         }
1170         break;
1171     case NBD_CMD_TRIM:
1172         TRACE("Request type is TRIM");
1173         ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
1174                               request.len / 512);
1175         if (ret < 0) {
1176             LOG("discard failed");
1177             reply.error = -ret;
1178         }
1179         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1180             goto out;
1181         }
1182         break;
1183     default:
1184         LOG("invalid request type (%u) received", request.type);
1185     invalid_request:
1186         reply.error = -EINVAL;
1187     error_reply:
1188         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1189             goto out;
1190         }
1191         break;
1192     }
1193
1194     TRACE("Request/Reply complete");
1195
1196 done:
1197     nbd_request_put(req);
1198     return;
1199
1200 out:
1201     nbd_request_put(req);
1202     nbd_client_close(client);
1203 }
1204
1205 static int nbd_can_read(void *opaque)
1206 {
1207     NBDClient *client = opaque;
1208
1209     return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
1210 }
1211
1212 static void nbd_read(void *opaque)
1213 {
1214     NBDClient *client = opaque;
1215
1216     if (client->recv_coroutine) {
1217         qemu_coroutine_enter(client->recv_coroutine, NULL);
1218     } else {
1219         qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
1220     }
1221 }
1222
1223 static void nbd_restart_write(void *opaque)
1224 {
1225     NBDClient *client = opaque;
1226
1227     qemu_coroutine_enter(client->send_coroutine, NULL);
1228 }
1229
1230 NBDClient *nbd_client_new(NBDExport *exp, int csock,
1231                           void (*close)(NBDClient *))
1232 {
1233     NBDClient *client;
1234     client = g_malloc0(sizeof(NBDClient));
1235     client->refcount = 1;
1236     client->exp = exp;
1237     client->sock = csock;
1238     if (nbd_send_negotiate(client) < 0) {
1239         g_free(client);
1240         return NULL;
1241     }
1242     client->close = close;
1243     qemu_co_mutex_init(&client->send_lock);
1244     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1245
1246     if (exp) {
1247         QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1248         nbd_export_get(exp);
1249     }
1250     return client;
1251 }