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     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     QSIMPLEQ_HEAD(, NBDRequest) requests;
 102     QTAILQ_ENTRY(NBDExport) next;
 103 };
 104
 105 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
 106
 107 struct NBDClient {
 108     int refcount;
 109     void (*close)(NBDClient *client);
 110
 111     NBDExport *exp;
 112     int sock;
 113
 114     Coroutine *recv_coroutine;
 115
 116     CoMutex send_lock;
 117     Coroutine *send_coroutine;
 118
 119     QTAILQ_ENTRY(NBDClient) next;
 120     int nb_requests;
 121     bool closing;
 122 };
 123
 124 /* That's all folks */
 125
 126 ssize_t nbd_wr_sync(int fd, void *buffer, size_t size, bool do_read)
 127 {
 128     size_t offset = 0;
 129     int err;
 130
 131     if (qemu_in_coroutine()) {
 132         if (do_read) {
 133             return qemu_co_recv(fd, buffer, size);
 134         } else {
 135             return qemu_co_send(fd, buffer, size);
 136         }
 137     }
 138
 139     while (offset < size) {
 140         ssize_t len;
 141
 142         if (do_read) {
 143             len = qemu_recv(fd, buffer + offset, size - offset, 0);
 144         } else {
 145             len = send(fd, buffer + offset, size - offset, 0);
 146         }
 147
 148         if (len < 0) {
 149             err = socket_error();
 150
 151             /* recoverable error */
 152             if (err == EINTR || (offset > 0 && err == EAGAIN)) {
 153                 continue;
 154             }
 155
 156             /* unrecoverable error */
 157             return -err;
 158         }
 159
 160         /* eof */
 161         if (len == 0) {
 162             break;
 163         }
 164
 165         offset += len;
 166     }
 167
 168     return offset;
 169 }
 170
 171 static ssize_t read_sync(int fd, void *buffer, size_t size)
 172 {
 173     /* Sockets are kept in blocking mode in the negotiation phase.  After
 174      * that, a non-readable socket simply means that another thread stole
 175      * our request/reply.  Synchronization is done with recv_coroutine, so
 176      * that this is coroutine-safe.
 177      */
 178     return nbd_wr_sync(fd, buffer, size, true);
 179 }
 180
 181 static ssize_t write_sync(int fd, void *buffer, size_t size)
 182 {
 183     int ret;
 184     do {
 185         /* For writes, we do expect the socket to be writable.  */
 186         ret = nbd_wr_sync(fd, buffer, size, false);
 187     } while (ret == -EAGAIN);
 188     return ret;
 189 }
 190
 191 static void combine_addr(char *buf, size_t len, const char* address,
 192                          uint16_t port)
 193 {
 194     /* If the address-part contains a colon, it's an IPv6 IP so needs [] */
 195     if (strstr(address, ":")) {
 196         snprintf(buf, len, "[%s]:%u", address, port);
 197     } else {
 198         snprintf(buf, len, "%s:%u", address, port);
 199     }
 200 }
 201
 202 int tcp_socket_outgoing(const char *address, uint16_t port)
 203 {
 204     char address_and_port[128];
 205     combine_addr(address_and_port, 128, address, port);
 206     return tcp_socket_outgoing_spec(address_and_port);
 207 }
 208
 209 int tcp_socket_outgoing_spec(const char *address_and_port)
 210 {
 211     Error *local_err = NULL;
 212     int fd = inet_connect(address_and_port, &local_err);
 213
 214     if (local_err != NULL) {
 215         qerror_report_err(local_err);
 216         error_free(local_err);
 217     }
 218     return fd;
 219 }
 220
 221 int tcp_socket_incoming(const char *address, uint16_t port)
 222 {
 223     char address_and_port[128];
 224     combine_addr(address_and_port, 128, address, port);
 225     return tcp_socket_incoming_spec(address_and_port);
 226 }
 227
 228 int tcp_socket_incoming_spec(const char *address_and_port)
 229 {
 230     Error *local_err = NULL;
 231     int fd = inet_listen(address_and_port, NULL, 0, SOCK_STREAM, 0, &local_err);
 232
 233     if (local_err != NULL) {
 234         qerror_report_err(local_err);
 235         error_free(local_err);
 236     }
 237     return fd;
 238 }
 239
 240 int unix_socket_incoming(const char *path)
 241 {
 242     Error *local_err = NULL;
 243     int fd = unix_listen(path, NULL, 0, &local_err);
 244
 245     if (local_err != NULL) {
 246         qerror_report_err(local_err);
 247         error_free(local_err);
 248     }
 249     return fd;
 250 }
 251
 252 int unix_socket_outgoing(const char *path)
 253 {
 254     Error *local_err = NULL;
 255     int fd = unix_connect(path, &local_err);
 256
 257     if (local_err != NULL) {
 258         qerror_report_err(local_err);
 259         error_free(local_err);
 260     }
 261     return fd;
 262 }
 263
 264 /* Basic flow for negotiation
 265
 266    Server         Client
 267    Negotiate
 268
 269    or
 270
 271    Server         Client
 272    Negotiate #1
 273                   Option
 274    Negotiate #2
 275
 276    ----
 277
 278    followed by
 279
 280    Server         Client
 281                   Request
 282    Response
 283                   Request
 284    Response
 285                   ...
 286    ...
 287                   Request (type == 2)
 288
 289 */
 290
 291 static int nbd_receive_options(NBDClient *client)
 292 {
 293     int csock = client->sock;
 294     char name[256];
 295     uint32_t tmp, length;
 296     uint64_t magic;
 297     int rc;
 298
 299     /* Client sends:
 300         [ 0 ..   3]   reserved (0)
 301         [ 4 ..  11]   NBD_OPTS_MAGIC
 302         [12 ..  15]   NBD_OPT_EXPORT_NAME
 303         [16 ..  19]   length
 304         [20 ..  xx]   export name (length bytes)
 305      */
 306
 307     rc = -EINVAL;
 308     if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 309         LOG("read failed");
 310         goto fail;
 311     }
 312     TRACE("Checking reserved");
 313     if (tmp != 0) {
 314         LOG("Bad reserved received");
 315         goto fail;
 316     }
 317
 318     if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 319         LOG("read failed");
 320         goto fail;
 321     }
 322     TRACE("Checking reserved");
 323     if (magic != be64_to_cpu(NBD_OPTS_MAGIC)) {
 324         LOG("Bad magic received");
 325         goto fail;
 326     }
 327
 328     if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 329         LOG("read failed");
 330         goto fail;
 331     }
 332     TRACE("Checking option");
 333     if (tmp != be32_to_cpu(NBD_OPT_EXPORT_NAME)) {
 334         LOG("Bad option received");
 335         goto fail;
 336     }
 337
 338     if (read_sync(csock, &length, sizeof(length)) != sizeof(length)) {
 339         LOG("read failed");
 340         goto fail;
 341     }
 342     TRACE("Checking length");
 343     length = be32_to_cpu(length);
 344     if (length > 255) {
 345         LOG("Bad length received");
 346         goto fail;
 347     }
 348     if (read_sync(csock, name, length) != length) {
 349         LOG("read failed");
 350         goto fail;
 351     }
 352     name[length] = '\0';
 353
 354     client->exp = nbd_export_find(name);
 355     if (!client->exp) {
 356         LOG("export not found");
 357         goto fail;
 358     }
 359
 360     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
 361     nbd_export_get(client->exp);
 362
 363     TRACE("Option negotiation succeeded.");
 364     rc = 0;
 365 fail:
 366     return rc;
 367 }
 368
 369 static int nbd_send_negotiate(NBDClient *client)
 370 {
 371     int csock = client->sock;
 372     char buf[8 + 8 + 8 + 128];
 373     int rc;
 374     const int myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
 375                          NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
 376
 377     /* Negotiation header without options:
 378         [ 0 ..   7]   passwd       ("NBDMAGIC")
 379         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
 380         [16 ..  23]   size
 381         [24 ..  25]   server flags (0)
 382         [24 ..  27]   export flags
 383         [28 .. 151]   reserved     (0)
 384
 385        Negotiation header with options, part 1:
 386         [ 0 ..   7]   passwd       ("NBDMAGIC")
 387         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
 388         [16 ..  17]   server flags (0)
 389
 390        part 2 (after options are sent):
 391         [18 ..  25]   size
 392         [26 ..  27]   export flags
 393         [28 .. 151]   reserved     (0)
 394      */
 395
 396     socket_set_block(csock);
 397     rc = -EINVAL;
 398
 399     TRACE("Beginning negotiation.");
 400     memcpy(buf, "NBDMAGIC", 8);
 401     if (client->exp) {
 402         assert ((client->exp->nbdflags & ~65535) == 0);
 403         cpu_to_be64w((uint64_t*)(buf + 8), NBD_CLIENT_MAGIC);
 404         cpu_to_be64w((uint64_t*)(buf + 16), client->exp->size);
 405         cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
 406     } else {
 407         cpu_to_be64w((uint64_t*)(buf + 8), NBD_OPTS_MAGIC);
 408     }
 409     memset(buf + 28, 0, 124);
 410
 411     if (client->exp) {
 412         if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
 413             LOG("write failed");
 414             goto fail;
 415         }
 416     } else {
 417         if (write_sync(csock, buf, 18) != 18) {
 418             LOG("write failed");
 419             goto fail;
 420         }
 421         rc = nbd_receive_options(client);
 422         if (rc < 0) {
 423             LOG("option negotiation failed");
 424             goto fail;
 425         }
 426
 427         assert ((client->exp->nbdflags & ~65535) == 0);
 428         cpu_to_be64w((uint64_t*)(buf + 18), client->exp->size);
 429         cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
 430         if (write_sync(csock, buf + 18, sizeof(buf) - 18) != sizeof(buf) - 18) {
 431             LOG("write failed");
 432             goto fail;
 433         }
 434     }
 435
 436     TRACE("Negotiation succeeded.");
 437     rc = 0;
 438 fail:
 439     socket_set_nonblock(csock);
 440     return rc;
 441 }
 442
 443 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
 444                           off_t *size, size_t *blocksize)
 445 {
 446     char buf[256];
 447     uint64_t magic, s;
 448     uint16_t tmp;
 449     int rc;
 450
 451     TRACE("Receiving negotiation.");
 452
 453     socket_set_block(csock);
 454     rc = -EINVAL;
 455
 456     if (read_sync(csock, buf, 8) != 8) {
 457         LOG("read failed");
 458         goto fail;
 459     }
 460
 461     buf[8] = '\0';
 462     if (strlen(buf) == 0) {
 463         LOG("server connection closed");
 464         goto fail;
 465     }
 466
 467     TRACE("Magic is %c%c%c%c%c%c%c%c",
 468           qemu_isprint(buf[0]) ? buf[0] : '.',
 469           qemu_isprint(buf[1]) ? buf[1] : '.',
 470           qemu_isprint(buf[2]) ? buf[2] : '.',
 471           qemu_isprint(buf[3]) ? buf[3] : '.',
 472           qemu_isprint(buf[4]) ? buf[4] : '.',
 473           qemu_isprint(buf[5]) ? buf[5] : '.',
 474           qemu_isprint(buf[6]) ? buf[6] : '.',
 475           qemu_isprint(buf[7]) ? buf[7] : '.');
 476
 477     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 478         LOG("Invalid magic received");
 479         goto fail;
 480     }
 481
 482     if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 483         LOG("read failed");
 484         goto fail;
 485     }
 486     magic = be64_to_cpu(magic);
 487     TRACE("Magic is 0x%" PRIx64, magic);
 488
 489     if (name) {
 490         uint32_t reserved = 0;
 491         uint32_t opt;
 492         uint32_t namesize;
 493
 494         TRACE("Checking magic (opts_magic)");
 495         if (magic != NBD_OPTS_MAGIC) {
 496             LOG("Bad magic received");
 497             goto fail;
 498         }
 499         if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 500             LOG("flags read failed");
 501             goto fail;
 502         }
 503         *flags = be16_to_cpu(tmp) << 16;
 504         /* reserved for future use */
 505         if (write_sync(csock, &reserved, sizeof(reserved)) !=
 506             sizeof(reserved)) {
 507             LOG("write failed (reserved)");
 508             goto fail;
 509         }
 510         /* write the export name */
 511         magic = cpu_to_be64(magic);
 512         if (write_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 513             LOG("write failed (magic)");
 514             goto fail;
 515         }
 516         opt = cpu_to_be32(NBD_OPT_EXPORT_NAME);
 517         if (write_sync(csock, &opt, sizeof(opt)) != sizeof(opt)) {
 518             LOG("write failed (opt)");
 519             goto fail;
 520         }
 521         namesize = cpu_to_be32(strlen(name));
 522         if (write_sync(csock, &namesize, sizeof(namesize)) !=
 523             sizeof(namesize)) {
 524             LOG("write failed (namesize)");
 525             goto fail;
 526         }
 527         if (write_sync(csock, (char*)name, strlen(name)) != strlen(name)) {
 528             LOG("write failed (name)");
 529             goto fail;
 530         }
 531     } else {
 532         TRACE("Checking magic (cli_magic)");
 533
 534         if (magic != NBD_CLIENT_MAGIC) {
 535             LOG("Bad magic received");
 536             goto fail;
 537         }
 538     }
 539
 540     if (read_sync(csock, &s, sizeof(s)) != sizeof(s)) {
 541         LOG("read failed");
 542         goto fail;
 543     }
 544     *size = be64_to_cpu(s);
 545     *blocksize = 1024;
 546     TRACE("Size is %" PRIu64, *size);
 547
 548     if (!name) {
 549         if (read_sync(csock, flags, sizeof(*flags)) != sizeof(*flags)) {
 550             LOG("read failed (flags)");
 551             goto fail;
 552         }
 553         *flags = be32_to_cpup(flags);
 554     } else {
 555         if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 556             LOG("read failed (tmp)");
 557             goto fail;
 558         }
 559         *flags |= be32_to_cpu(tmp);
 560     }
 561     if (read_sync(csock, &buf, 124) != 124) {
 562         LOG("read failed (buf)");
 563         goto fail;
 564     }
 565     rc = 0;
 566
 567 fail:
 568     socket_set_nonblock(csock);
 569     return rc;
 570 }
 571
 572 #ifdef __linux__
 573 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 574 {
 575     TRACE("Setting NBD socket");
 576
 577     if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
 578         int serrno = errno;
 579         LOG("Failed to set NBD socket");
 580         return -serrno;
 581     }
 582
 583     TRACE("Setting block size to %lu", (unsigned long)blocksize);
 584
 585     if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
 586         int serrno = errno;
 587         LOG("Failed setting NBD block size");
 588         return -serrno;
 589     }
 590
 591         TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
 592
 593     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
 594         int serrno = errno;
 595         LOG("Failed setting size (in blocks)");
 596         return -serrno;
 597     }
 598
 599     if (flags & NBD_FLAG_READ_ONLY) {
 600         int read_only = 1;
 601         TRACE("Setting readonly attribute");
 602
 603         if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 604             int serrno = errno;
 605             LOG("Failed setting read-only attribute");
 606             return -serrno;
 607         }
 608     }
 609
 610     if (ioctl(fd, NBD_SET_FLAGS, flags) < 0
 611         && errno != ENOTTY) {
 612         int serrno = errno;
 613         LOG("Failed setting flags");
 614         return -serrno;
 615     }
 616
 617     TRACE("Negotiation ended");
 618
 619     return 0;
 620 }
 621
 622 int nbd_disconnect(int fd)
 623 {
 624     ioctl(fd, NBD_CLEAR_QUE);
 625     ioctl(fd, NBD_DISCONNECT);
 626     ioctl(fd, NBD_CLEAR_SOCK);
 627     return 0;
 628 }
 629
 630 int nbd_client(int fd)
 631 {
 632     int ret;
 633     int serrno;
 634
 635     TRACE("Doing NBD loop");
 636
 637     ret = ioctl(fd, NBD_DO_IT);
 638     if (ret < 0 && errno == EPIPE) {
 639         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 640          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 641          * that case.
 642          */
 643         ret = 0;
 644     }
 645     serrno = errno;
 646
 647     TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
 648
 649     TRACE("Clearing NBD queue");
 650     ioctl(fd, NBD_CLEAR_QUE);
 651
 652     TRACE("Clearing NBD socket");
 653     ioctl(fd, NBD_CLEAR_SOCK);
 654
 655     errno = serrno;
 656     return ret;
 657 }
 658 #else
 659 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 660 {
 661     return -ENOTSUP;
 662 }
 663
 664 int nbd_disconnect(int fd)
 665 {
 666     return -ENOTSUP;
 667 }
 668
 669 int nbd_client(int fd)
 670 {
 671     return -ENOTSUP;
 672 }
 673 #endif
 674
 675 ssize_t nbd_send_request(int csock, struct nbd_request *request)
 676 {
 677     uint8_t buf[NBD_REQUEST_SIZE];
 678     ssize_t ret;
 679
 680     cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
 681     cpu_to_be32w((uint32_t*)(buf + 4), request->type);
 682     cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
 683     cpu_to_be64w((uint64_t*)(buf + 16), request->from);
 684     cpu_to_be32w((uint32_t*)(buf + 24), request->len);
 685
 686     TRACE("Sending request to client: "
 687           "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
 688           request->from, request->len, request->handle, request->type);
 689
 690     ret = write_sync(csock, buf, sizeof(buf));
 691     if (ret < 0) {
 692         return ret;
 693     }
 694
 695     if (ret != sizeof(buf)) {
 696         LOG("writing to socket failed");
 697         return -EINVAL;
 698     }
 699     return 0;
 700 }
 701
 702 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
 703 {
 704     uint8_t buf[NBD_REQUEST_SIZE];
 705     uint32_t magic;
 706     ssize_t ret;
 707
 708     ret = read_sync(csock, buf, sizeof(buf));
 709     if (ret < 0) {
 710         return ret;
 711     }
 712
 713     if (ret != sizeof(buf)) {
 714         LOG("read failed");
 715         return -EINVAL;
 716     }
 717
 718     /* Request
 719        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
 720        [ 4 ..  7]   type    (0 == READ, 1 == WRITE)
 721        [ 8 .. 15]   handle
 722        [16 .. 23]   from
 723        [24 .. 27]   len
 724      */
 725
 726     magic = be32_to_cpup((uint32_t*)buf);
 727     request->type  = be32_to_cpup((uint32_t*)(buf + 4));
 728     request->handle = be64_to_cpup((uint64_t*)(buf + 8));
 729     request->from  = be64_to_cpup((uint64_t*)(buf + 16));
 730     request->len   = be32_to_cpup((uint32_t*)(buf + 24));
 731
 732     TRACE("Got request: "
 733           "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
 734           magic, request->type, request->from, request->len);
 735
 736     if (magic != NBD_REQUEST_MAGIC) {
 737         LOG("invalid magic (got 0x%x)", magic);
 738         return -EINVAL;
 739     }
 740     return 0;
 741 }
 742
 743 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
 744 {
 745     uint8_t buf[NBD_REPLY_SIZE];
 746     uint32_t magic;
 747     ssize_t ret;
 748
 749     ret = read_sync(csock, buf, sizeof(buf));
 750     if (ret < 0) {
 751         return ret;
 752     }
 753
 754     if (ret != sizeof(buf)) {
 755         LOG("read failed");
 756         return -EINVAL;
 757     }
 758
 759     /* Reply
 760        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 761        [ 4 ..  7]    error   (0 == no error)
 762        [ 7 .. 15]    handle
 763      */
 764
 765     magic = be32_to_cpup((uint32_t*)buf);
 766     reply->error  = be32_to_cpup((uint32_t*)(buf + 4));
 767     reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
 768
 769     TRACE("Got reply: "
 770           "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
 771           magic, reply->error, reply->handle);
 772
 773     if (magic != NBD_REPLY_MAGIC) {
 774         LOG("invalid magic (got 0x%x)", magic);
 775         return -EINVAL;
 776     }
 777     return 0;
 778 }
 779
 780 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
 781 {
 782     uint8_t buf[NBD_REPLY_SIZE];
 783     ssize_t ret;
 784
 785     /* Reply
 786        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 787        [ 4 ..  7]    error   (0 == no error)
 788        [ 7 .. 15]    handle
 789      */
 790     cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
 791     cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
 792     cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
 793
 794     TRACE("Sending response to client");
 795
 796     ret = write_sync(csock, buf, sizeof(buf));
 797     if (ret < 0) {
 798         return ret;
 799     }
 800
 801     if (ret != sizeof(buf)) {
 802         LOG("writing to socket failed");
 803         return -EINVAL;
 804     }
 805     return 0;
 806 }
 807
 808 #define MAX_NBD_REQUESTS 16
 809
 810 void nbd_client_get(NBDClient *client)
 811 {
 812     client->refcount++;
 813 }
 814
 815 void nbd_client_put(NBDClient *client)
 816 {
 817     if (--client->refcount == 0) {
 818         /* The last reference should be dropped by client->close,
 819          * which is called by nbd_client_close.
 820          */
 821         assert(client->closing);
 822
 823         qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
 824         close(client->sock);
 825         client->sock = -1;
 826         if (client->exp) {
 827             QTAILQ_REMOVE(&client->exp->clients, client, next);
 828             nbd_export_put(client->exp);
 829         }
 830         g_free(client);
 831     }
 832 }
 833
 834 void nbd_client_close(NBDClient *client)
 835 {
 836     if (client->closing) {
 837         return;
 838     }
 839
 840     client->closing = true;
 841
 842     /* Force requests to finish.  They will drop their own references,
 843      * then we'll close the socket and free the NBDClient.
 844      */
 845     shutdown(client->sock, 2);
 846
 847     /* Also tell the client, so that they release their reference.  */
 848     if (client->close) {
 849         client->close(client);
 850     }
 851 }
 852
 853 static NBDRequest *nbd_request_get(NBDClient *client)
 854 {
 855     NBDRequest *req;
 856     NBDExport *exp = client->exp;
 857
 858     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
 859     client->nb_requests++;
 860
 861     if (QSIMPLEQ_EMPTY(&exp->requests)) {
 862         req = g_malloc0(sizeof(NBDRequest));
 863         req->data = qemu_blockalign(exp->bs, NBD_BUFFER_SIZE);
 864     } else {
 865         req = QSIMPLEQ_FIRST(&exp->requests);
 866         QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
 867     }
 868     nbd_client_get(client);
 869     req->client = client;
 870     return req;
 871 }
 872
 873 static void nbd_request_put(NBDRequest *req)
 874 {
 875     NBDClient *client = req->client;
 876     QSIMPLEQ_INSERT_HEAD(&client->exp->requests, req, entry);
 877     if (client->nb_requests-- == MAX_NBD_REQUESTS) {
 878         qemu_notify_event();
 879     }
 880     nbd_client_put(client);
 881 }
 882
 883 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
 884                           off_t size, uint32_t nbdflags,
 885                           void (*close)(NBDExport *))
 886 {
 887     NBDExport *exp = g_malloc0(sizeof(NBDExport));
 888     QSIMPLEQ_INIT(&exp->requests);
 889     exp->refcount = 1;
 890     QTAILQ_INIT(&exp->clients);
 891     exp->bs = bs;
 892     exp->dev_offset = dev_offset;
 893     exp->nbdflags = nbdflags;
 894     exp->size = size == -1 ? bdrv_getlength(bs) : size;
 895     exp->close = close;
 896     return exp;
 897 }
 898
 899 NBDExport *nbd_export_find(const char *name)
 900 {
 901     NBDExport *exp;
 902     QTAILQ_FOREACH(exp, &exports, next) {
 903         if (strcmp(name, exp->name) == 0) {
 904             return exp;
 905         }
 906     }
 907
 908     return NULL;
 909 }
 910
 911 void nbd_export_set_name(NBDExport *exp, const char *name)
 912 {
 913     if (exp->name == name) {
 914         return;
 915     }
 916
 917     nbd_export_get(exp);
 918     if (exp->name != NULL) {
 919         g_free(exp->name);
 920         exp->name = NULL;
 921         QTAILQ_REMOVE(&exports, exp, next);
 922         nbd_export_put(exp);
 923     }
 924     if (name != NULL) {
 925         nbd_export_get(exp);
 926         exp->name = g_strdup(name);
 927         QTAILQ_INSERT_TAIL(&exports, exp, next);
 928     }
 929     nbd_export_put(exp);
 930 }
 931
 932 void nbd_export_close(NBDExport *exp)
 933 {
 934     NBDClient *client, *next;
 935
 936     nbd_export_get(exp);
 937     QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) {
 938         nbd_client_close(client);
 939     }
 940     nbd_export_set_name(exp, NULL);
 941     nbd_export_put(exp);
 942 }
 943
 944 void nbd_export_get(NBDExport *exp)
 945 {
 946     assert(exp->refcount > 0);
 947     exp->refcount++;
 948 }
 949
 950 void nbd_export_put(NBDExport *exp)
 951 {
 952     assert(exp->refcount > 0);
 953     if (exp->refcount == 1) {
 954         nbd_export_close(exp);
 955     }
 956
 957     if (--exp->refcount == 0) {
 958         assert(exp->name == NULL);
 959
 960         if (exp->close) {
 961             exp->close(exp);
 962         }
 963
 964         while (!QSIMPLEQ_EMPTY(&exp->requests)) {
 965             NBDRequest *first = QSIMPLEQ_FIRST(&exp->requests);
 966             QSIMPLEQ_REMOVE_HEAD(&exp->requests, entry);
 967             qemu_vfree(first->data);
 968             g_free(first);
 969         }
 970
 971         g_free(exp);
 972     }
 973 }
 974
 975 BlockDriverState *nbd_export_get_blockdev(NBDExport *exp)
 976 {
 977     return exp->bs;
 978 }
 979
 980 void nbd_export_close_all(void)
 981 {
 982     NBDExport *exp, *next;
 983
 984     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
 985         nbd_export_close(exp);
 986     }
 987 }
 988
 989 static int nbd_can_read(void *opaque);
 990 static void nbd_read(void *opaque);
 991 static void nbd_restart_write(void *opaque);
 992
 993 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
 994                                  int len)
 995 {
 996     NBDClient *client = req->client;
 997     int csock = client->sock;
 998     ssize_t rc, ret;
 999
1000     qemu_co_mutex_lock(&client->send_lock);
1001     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
1002                          nbd_restart_write, client);
1003     client->send_coroutine = qemu_coroutine_self();
1004
1005     if (!len) {
1006         rc = nbd_send_reply(csock, reply);
1007     } else {
1008         socket_set_cork(csock, 1);
1009         rc = nbd_send_reply(csock, reply);
1010         if (rc >= 0) {
1011             ret = qemu_co_send(csock, req->data, len);
1012             if (ret != len) {
1013                 rc = -EIO;
1014             }
1015         }
1016         socket_set_cork(csock, 0);
1017     }
1018
1019     client->send_coroutine = NULL;
1020     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1021     qemu_co_mutex_unlock(&client->send_lock);
1022     return rc;
1023 }
1024
1025 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
1026 {
1027     NBDClient *client = req->client;
1028     int csock = client->sock;
1029     ssize_t rc;
1030
1031     client->recv_coroutine = qemu_coroutine_self();
1032     rc = nbd_receive_request(csock, request);
1033     if (rc < 0) {
1034         if (rc != -EAGAIN) {
1035             rc = -EIO;
1036         }
1037         goto out;
1038     }
1039
1040     if (request->len > NBD_BUFFER_SIZE) {
1041         LOG("len (%u) is larger than max len (%u)",
1042             request->len, NBD_BUFFER_SIZE);
1043         rc = -EINVAL;
1044         goto out;
1045     }
1046
1047     if ((request->from + request->len) < request->from) {
1048         LOG("integer overflow detected! "
1049             "you're probably being attacked");
1050         rc = -EINVAL;
1051         goto out;
1052     }
1053
1054     TRACE("Decoding type");
1055
1056     if ((request->type & NBD_CMD_MASK_COMMAND) == NBD_CMD_WRITE) {
1057         TRACE("Reading %u byte(s)", request->len);
1058
1059         if (qemu_co_recv(csock, req->data, request->len) != request->len) {
1060             LOG("reading from socket failed");
1061             rc = -EIO;
1062             goto out;
1063         }
1064     }
1065     rc = 0;
1066
1067 out:
1068     client->recv_coroutine = NULL;
1069     return rc;
1070 }
1071
1072 static void nbd_trip(void *opaque)
1073 {
1074     NBDClient *client = opaque;
1075     NBDExport *exp = client->exp;
1076     NBDRequest *req;
1077     struct nbd_request request;
1078     struct nbd_reply reply;
1079     ssize_t ret;
1080
1081     TRACE("Reading request.");
1082     if (client->closing) {
1083         return;
1084     }
1085
1086     req = nbd_request_get(client);
1087     ret = nbd_co_receive_request(req, &request);
1088     if (ret == -EAGAIN) {
1089         goto done;
1090     }
1091     if (ret == -EIO) {
1092         goto out;
1093     }
1094
1095     reply.handle = request.handle;
1096     reply.error = 0;
1097
1098     if (ret < 0) {
1099         reply.error = -ret;
1100         goto error_reply;
1101     }
1102
1103     if ((request.from + request.len) > exp->size) {
1104             LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
1105             ", Offset: %" PRIu64 "\n",
1106                     request.from, request.len,
1107                     (uint64_t)exp->size, (uint64_t)exp->dev_offset);
1108         LOG("requested operation past EOF--bad client?");
1109         goto invalid_request;
1110     }
1111
1112     switch (request.type & NBD_CMD_MASK_COMMAND) {
1113     case NBD_CMD_READ:
1114         TRACE("Request type is READ");
1115
1116         if (request.type & NBD_CMD_FLAG_FUA) {
1117             ret = bdrv_co_flush(exp->bs);
1118             if (ret < 0) {
1119                 LOG("flush failed");
1120                 reply.error = -ret;
1121                 goto error_reply;
1122             }
1123         }
1124
1125         ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
1126                         req->data, request.len / 512);
1127         if (ret < 0) {
1128             LOG("reading from file failed");
1129             reply.error = -ret;
1130             goto error_reply;
1131         }
1132
1133         TRACE("Read %u byte(s)", request.len);
1134         if (nbd_co_send_reply(req, &reply, request.len) < 0)
1135             goto out;
1136         break;
1137     case NBD_CMD_WRITE:
1138         TRACE("Request type is WRITE");
1139
1140         if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1141             TRACE("Server is read-only, return error");
1142             reply.error = EROFS;
1143             goto error_reply;
1144         }
1145
1146         TRACE("Writing to device");
1147
1148         ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
1149                          req->data, request.len / 512);
1150         if (ret < 0) {
1151             LOG("writing to file failed");
1152             reply.error = -ret;
1153             goto error_reply;
1154         }
1155
1156         if (request.type & NBD_CMD_FLAG_FUA) {
1157             ret = bdrv_co_flush(exp->bs);
1158             if (ret < 0) {
1159                 LOG("flush failed");
1160                 reply.error = -ret;
1161                 goto error_reply;
1162             }
1163         }
1164
1165         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1166             goto out;
1167         }
1168         break;
1169     case NBD_CMD_DISC:
1170         TRACE("Request type is DISCONNECT");
1171         errno = 0;
1172         goto out;
1173     case NBD_CMD_FLUSH:
1174         TRACE("Request type is FLUSH");
1175
1176         ret = bdrv_co_flush(exp->bs);
1177         if (ret < 0) {
1178             LOG("flush failed");
1179             reply.error = -ret;
1180         }
1181         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1182             goto out;
1183         }
1184         break;
1185     case NBD_CMD_TRIM:
1186         TRACE("Request type is TRIM");
1187         ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
1188                               request.len / 512);
1189         if (ret < 0) {
1190             LOG("discard failed");
1191             reply.error = -ret;
1192         }
1193         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1194             goto out;
1195         }
1196         break;
1197     default:
1198         LOG("invalid request type (%u) received", request.type);
1199     invalid_request:
1200         reply.error = -EINVAL;
1201     error_reply:
1202         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1203             goto out;
1204         }
1205         break;
1206     }
1207
1208     TRACE("Request/Reply complete");
1209
1210 done:
1211     nbd_request_put(req);
1212     return;
1213
1214 out:
1215     nbd_request_put(req);
1216     nbd_client_close(client);
1217 }
1218
1219 static int nbd_can_read(void *opaque)
1220 {
1221     NBDClient *client = opaque;
1222
1223     return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
1224 }
1225
1226 static void nbd_read(void *opaque)
1227 {
1228     NBDClient *client = opaque;
1229
1230     if (client->recv_coroutine) {
1231         qemu_coroutine_enter(client->recv_coroutine, NULL);
1232     } else {
1233         qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
1234     }
1235 }
1236
1237 static void nbd_restart_write(void *opaque)
1238 {
1239     NBDClient *client = opaque;
1240
1241     qemu_coroutine_enter(client->send_coroutine, NULL);
1242 }
1243
1244 NBDClient *nbd_client_new(NBDExport *exp, int csock,
1245                           void (*close)(NBDClient *))
1246 {
1247     NBDClient *client;
1248     client = g_malloc0(sizeof(NBDClient));
1249     client->refcount = 1;
1250     client->exp = exp;
1251     client->sock = csock;
1252     if (nbd_send_negotiate(client) < 0) {
1253         g_free(client);
1254         return NULL;
1255     }
1256     client->close = close;
1257     qemu_co_mutex_init(&client->send_lock);
1258     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1259
1260     if (exp) {
1261         QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1262         nbd_export_get(exp);
1263     }
1264     return client;
1265 }