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 #include "qemu/main-loop.h"
  42
  43 //#define DEBUG_NBD
  44
  45 #ifdef DEBUG_NBD
  46 #define TRACE(msg, ...) do { \
  47     LOG(msg, ## __VA_ARGS__); \
  48 } while(0)
  49 #else
  50 #define TRACE(msg, ...) \
  51     do { } while (0)
  52 #endif
  53
  54 #define LOG(msg, ...) do { \
  55     fprintf(stderr, "%s:%s():L%d: " msg "\n", \
  56             __FILE__, __FUNCTION__, __LINE__, ## __VA_ARGS__); \
  57 } while(0)
  58
  59 /* This is all part of the "official" NBD API */
  60
  61 #define NBD_REQUEST_SIZE        (4 + 4 + 8 + 8 + 4)
  62 #define NBD_REPLY_SIZE          (4 + 4 + 8)
  63 #define NBD_REQUEST_MAGIC       0x25609513
  64 #define NBD_REPLY_MAGIC         0x67446698
  65 #define NBD_OPTS_MAGIC          0x49484156454F5054LL
  66 #define NBD_CLIENT_MAGIC        0x0000420281861253LL
  67
  68 #define NBD_SET_SOCK            _IO(0xab, 0)
  69 #define NBD_SET_BLKSIZE         _IO(0xab, 1)
  70 #define NBD_SET_SIZE            _IO(0xab, 2)
  71 #define NBD_DO_IT               _IO(0xab, 3)
  72 #define NBD_CLEAR_SOCK          _IO(0xab, 4)
  73 #define NBD_CLEAR_QUE           _IO(0xab, 5)
  74 #define NBD_PRINT_DEBUG         _IO(0xab, 6)
  75 #define NBD_SET_SIZE_BLOCKS     _IO(0xab, 7)
  76 #define NBD_DISCONNECT          _IO(0xab, 8)
  77 #define NBD_SET_TIMEOUT         _IO(0xab, 9)
  78 #define NBD_SET_FLAGS           _IO(0xab, 10)
  79
  80 #define NBD_OPT_EXPORT_NAME     (1 << 0)
  81
  82 /* Definitions for opaque data types */
  83
  84 typedef struct NBDRequest NBDRequest;
  85
  86 struct NBDRequest {
  87     QSIMPLEQ_ENTRY(NBDRequest) entry;
  88     NBDClient *client;
  89     uint8_t *data;
  90 };
  91
  92 struct NBDExport {
  93     int refcount;
  94     void (*close)(NBDExport *exp);
  95
  96     BlockDriverState *bs;
  97     char *name;
  98     off_t dev_offset;
  99     off_t size;
 100     uint32_t nbdflags;
 101     QTAILQ_HEAD(, NBDClient) clients;
 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_opts(QemuOpts *opts)
 203 {
 204     Error *local_err = NULL;
 205     int fd = inet_connect_opts(opts, &local_err, NULL, NULL);
 206     if (local_err != NULL) {
 207         qerror_report_err(local_err);
 208         error_free(local_err);
 209     }
 210
 211     return fd;
 212 }
 213
 214 int tcp_socket_incoming(const char *address, uint16_t port)
 215 {
 216     char address_and_port[128];
 217     combine_addr(address_and_port, 128, address, port);
 218     return tcp_socket_incoming_spec(address_and_port);
 219 }
 220
 221 int tcp_socket_incoming_spec(const char *address_and_port)
 222 {
 223     Error *local_err = NULL;
 224     int fd = inet_listen(address_and_port, NULL, 0, SOCK_STREAM, 0, &local_err);
 225
 226     if (local_err != NULL) {
 227         qerror_report_err(local_err);
 228         error_free(local_err);
 229     }
 230     return fd;
 231 }
 232
 233 int unix_socket_incoming(const char *path)
 234 {
 235     Error *local_err = NULL;
 236     int fd = unix_listen(path, NULL, 0, &local_err);
 237
 238     if (local_err != NULL) {
 239         qerror_report_err(local_err);
 240         error_free(local_err);
 241     }
 242     return fd;
 243 }
 244
 245 int unix_socket_outgoing(const char *path)
 246 {
 247     Error *local_err = NULL;
 248     int fd = unix_connect(path, &local_err);
 249
 250     if (local_err != NULL) {
 251         qerror_report_err(local_err);
 252         error_free(local_err);
 253     }
 254     return fd;
 255 }
 256
 257 /* Basic flow for negotiation
 258
 259    Server         Client
 260    Negotiate
 261
 262    or
 263
 264    Server         Client
 265    Negotiate #1
 266                   Option
 267    Negotiate #2
 268
 269    ----
 270
 271    followed by
 272
 273    Server         Client
 274                   Request
 275    Response
 276                   Request
 277    Response
 278                   ...
 279    ...
 280                   Request (type == 2)
 281
 282 */
 283
 284 static int nbd_receive_options(NBDClient *client)
 285 {
 286     int csock = client->sock;
 287     char name[256];
 288     uint32_t tmp, length;
 289     uint64_t magic;
 290     int rc;
 291
 292     /* Client sends:
 293         [ 0 ..   3]   reserved (0)
 294         [ 4 ..  11]   NBD_OPTS_MAGIC
 295         [12 ..  15]   NBD_OPT_EXPORT_NAME
 296         [16 ..  19]   length
 297         [20 ..  xx]   export name (length bytes)
 298      */
 299
 300     rc = -EINVAL;
 301     if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 302         LOG("read failed");
 303         goto fail;
 304     }
 305     TRACE("Checking reserved");
 306     if (tmp != 0) {
 307         LOG("Bad reserved received");
 308         goto fail;
 309     }
 310
 311     if (read_sync(csock, &magic, sizeof(magic)) != sizeof(magic)) {
 312         LOG("read failed");
 313         goto fail;
 314     }
 315     TRACE("Checking reserved");
 316     if (magic != be64_to_cpu(NBD_OPTS_MAGIC)) {
 317         LOG("Bad magic received");
 318         goto fail;
 319     }
 320
 321     if (read_sync(csock, &tmp, sizeof(tmp)) != sizeof(tmp)) {
 322         LOG("read failed");
 323         goto fail;
 324     }
 325     TRACE("Checking option");
 326     if (tmp != be32_to_cpu(NBD_OPT_EXPORT_NAME)) {
 327         LOG("Bad option received");
 328         goto fail;
 329     }
 330
 331     if (read_sync(csock, &length, sizeof(length)) != sizeof(length)) {
 332         LOG("read failed");
 333         goto fail;
 334     }
 335     TRACE("Checking length");
 336     length = be32_to_cpu(length);
 337     if (length > 255) {
 338         LOG("Bad length received");
 339         goto fail;
 340     }
 341     if (read_sync(csock, name, length) != length) {
 342         LOG("read failed");
 343         goto fail;
 344     }
 345     name[length] = '\0';
 346
 347     client->exp = nbd_export_find(name);
 348     if (!client->exp) {
 349         LOG("export not found");
 350         goto fail;
 351     }
 352
 353     QTAILQ_INSERT_TAIL(&client->exp->clients, client, next);
 354     nbd_export_get(client->exp);
 355
 356     TRACE("Option negotiation succeeded.");
 357     rc = 0;
 358 fail:
 359     return rc;
 360 }
 361
 362 static int nbd_send_negotiate(NBDClient *client)
 363 {
 364     int csock = client->sock;
 365     char buf[8 + 8 + 8 + 128];
 366     int rc;
 367     const int myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM |
 368                          NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA);
 369
 370     /* Negotiation header without options:
 371         [ 0 ..   7]   passwd       ("NBDMAGIC")
 372         [ 8 ..  15]   magic        (NBD_CLIENT_MAGIC)
 373         [16 ..  23]   size
 374         [24 ..  25]   server flags (0)
 375         [24 ..  27]   export flags
 376         [28 .. 151]   reserved     (0)
 377
 378        Negotiation header with options, part 1:
 379         [ 0 ..   7]   passwd       ("NBDMAGIC")
 380         [ 8 ..  15]   magic        (NBD_OPTS_MAGIC)
 381         [16 ..  17]   server flags (0)
 382
 383        part 2 (after options are sent):
 384         [18 ..  25]   size
 385         [26 ..  27]   export flags
 386         [28 .. 151]   reserved     (0)
 387      */
 388
 389     qemu_set_block(csock);
 390     rc = -EINVAL;
 391
 392     TRACE("Beginning negotiation.");
 393     memset(buf, 0, sizeof(buf));
 394     memcpy(buf, "NBDMAGIC", 8);
 395     if (client->exp) {
 396         assert ((client->exp->nbdflags & ~65535) == 0);
 397         cpu_to_be64w((uint64_t*)(buf + 8), NBD_CLIENT_MAGIC);
 398         cpu_to_be64w((uint64_t*)(buf + 16), client->exp->size);
 399         cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
 400     } else {
 401         cpu_to_be64w((uint64_t*)(buf + 8), NBD_OPTS_MAGIC);
 402     }
 403
 404     if (client->exp) {
 405         if (write_sync(csock, buf, sizeof(buf)) != sizeof(buf)) {
 406             LOG("write failed");
 407             goto fail;
 408         }
 409     } else {
 410         if (write_sync(csock, buf, 18) != 18) {
 411             LOG("write failed");
 412             goto fail;
 413         }
 414         rc = nbd_receive_options(client);
 415         if (rc < 0) {
 416             LOG("option negotiation failed");
 417             goto fail;
 418         }
 419
 420         assert ((client->exp->nbdflags & ~65535) == 0);
 421         cpu_to_be64w((uint64_t*)(buf + 18), client->exp->size);
 422         cpu_to_be16w((uint16_t*)(buf + 26), client->exp->nbdflags | myflags);
 423         if (write_sync(csock, buf + 18, sizeof(buf) - 18) != sizeof(buf) - 18) {
 424             LOG("write failed");
 425             goto fail;
 426         }
 427     }
 428
 429     TRACE("Negotiation succeeded.");
 430     rc = 0;
 431 fail:
 432     qemu_set_nonblock(csock);
 433     return rc;
 434 }
 435
 436 int nbd_receive_negotiate(int csock, const char *name, uint32_t *flags,
 437                           off_t *size, size_t *blocksize)
 438 {
 439     char buf[256];
 440     uint64_t magic, s;
 441     uint16_t tmp;
 442     int rc;
 443
 444     TRACE("Receiving negotiation.");
 445
 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     return rc;
 561 }
 562
 563 #ifdef __linux__
 564 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 565 {
 566     TRACE("Setting NBD socket");
 567
 568     if (ioctl(fd, NBD_SET_SOCK, csock) < 0) {
 569         int serrno = errno;
 570         LOG("Failed to set NBD socket");
 571         return -serrno;
 572     }
 573
 574     TRACE("Setting block size to %lu", (unsigned long)blocksize);
 575
 576     if (ioctl(fd, NBD_SET_BLKSIZE, blocksize) < 0) {
 577         int serrno = errno;
 578         LOG("Failed setting NBD block size");
 579         return -serrno;
 580     }
 581
 582         TRACE("Setting size to %zd block(s)", (size_t)(size / blocksize));
 583
 584     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, size / blocksize) < 0) {
 585         int serrno = errno;
 586         LOG("Failed setting size (in blocks)");
 587         return -serrno;
 588     }
 589
 590     if (ioctl(fd, NBD_SET_FLAGS, flags) < 0) {
 591         if (errno == ENOTTY) {
 592             int read_only = (flags & NBD_FLAG_READ_ONLY) != 0;
 593             TRACE("Setting readonly attribute");
 594
 595             if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 596                 int serrno = errno;
 597                 LOG("Failed setting read-only attribute");
 598                 return -serrno;
 599             }
 600         } else {
 601             int serrno = errno;
 602             LOG("Failed setting flags");
 603             return -serrno;
 604         }
 605     }
 606
 607     TRACE("Negotiation ended");
 608
 609     return 0;
 610 }
 611
 612 int nbd_disconnect(int fd)
 613 {
 614     ioctl(fd, NBD_CLEAR_QUE);
 615     ioctl(fd, NBD_DISCONNECT);
 616     ioctl(fd, NBD_CLEAR_SOCK);
 617     return 0;
 618 }
 619
 620 int nbd_client(int fd)
 621 {
 622     int ret;
 623     int serrno;
 624
 625     TRACE("Doing NBD loop");
 626
 627     ret = ioctl(fd, NBD_DO_IT);
 628     if (ret < 0 && errno == EPIPE) {
 629         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 630          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 631          * that case.
 632          */
 633         ret = 0;
 634     }
 635     serrno = errno;
 636
 637     TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
 638
 639     TRACE("Clearing NBD queue");
 640     ioctl(fd, NBD_CLEAR_QUE);
 641
 642     TRACE("Clearing NBD socket");
 643     ioctl(fd, NBD_CLEAR_SOCK);
 644
 645     errno = serrno;
 646     return ret;
 647 }
 648 #else
 649 int nbd_init(int fd, int csock, uint32_t flags, off_t size, size_t blocksize)
 650 {
 651     return -ENOTSUP;
 652 }
 653
 654 int nbd_disconnect(int fd)
 655 {
 656     return -ENOTSUP;
 657 }
 658
 659 int nbd_client(int fd)
 660 {
 661     return -ENOTSUP;
 662 }
 663 #endif
 664
 665 ssize_t nbd_send_request(int csock, struct nbd_request *request)
 666 {
 667     uint8_t buf[NBD_REQUEST_SIZE];
 668     ssize_t ret;
 669
 670     cpu_to_be32w((uint32_t*)buf, NBD_REQUEST_MAGIC);
 671     cpu_to_be32w((uint32_t*)(buf + 4), request->type);
 672     cpu_to_be64w((uint64_t*)(buf + 8), request->handle);
 673     cpu_to_be64w((uint64_t*)(buf + 16), request->from);
 674     cpu_to_be32w((uint32_t*)(buf + 24), request->len);
 675
 676     TRACE("Sending request to client: "
 677           "{ .from = %" PRIu64", .len = %u, .handle = %" PRIu64", .type=%i}",
 678           request->from, request->len, request->handle, request->type);
 679
 680     ret = write_sync(csock, buf, sizeof(buf));
 681     if (ret < 0) {
 682         return ret;
 683     }
 684
 685     if (ret != sizeof(buf)) {
 686         LOG("writing to socket failed");
 687         return -EINVAL;
 688     }
 689     return 0;
 690 }
 691
 692 static ssize_t nbd_receive_request(int csock, struct nbd_request *request)
 693 {
 694     uint8_t buf[NBD_REQUEST_SIZE];
 695     uint32_t magic;
 696     ssize_t ret;
 697
 698     ret = read_sync(csock, buf, sizeof(buf));
 699     if (ret < 0) {
 700         return ret;
 701     }
 702
 703     if (ret != sizeof(buf)) {
 704         LOG("read failed");
 705         return -EINVAL;
 706     }
 707
 708     /* Request
 709        [ 0 ..  3]   magic   (NBD_REQUEST_MAGIC)
 710        [ 4 ..  7]   type    (0 == READ, 1 == WRITE)
 711        [ 8 .. 15]   handle
 712        [16 .. 23]   from
 713        [24 .. 27]   len
 714      */
 715
 716     magic = be32_to_cpup((uint32_t*)buf);
 717     request->type  = be32_to_cpup((uint32_t*)(buf + 4));
 718     request->handle = be64_to_cpup((uint64_t*)(buf + 8));
 719     request->from  = be64_to_cpup((uint64_t*)(buf + 16));
 720     request->len   = be32_to_cpup((uint32_t*)(buf + 24));
 721
 722     TRACE("Got request: "
 723           "{ magic = 0x%x, .type = %d, from = %" PRIu64" , len = %u }",
 724           magic, request->type, request->from, request->len);
 725
 726     if (magic != NBD_REQUEST_MAGIC) {
 727         LOG("invalid magic (got 0x%x)", magic);
 728         return -EINVAL;
 729     }
 730     return 0;
 731 }
 732
 733 ssize_t nbd_receive_reply(int csock, struct nbd_reply *reply)
 734 {
 735     uint8_t buf[NBD_REPLY_SIZE];
 736     uint32_t magic;
 737     ssize_t ret;
 738
 739     ret = read_sync(csock, buf, sizeof(buf));
 740     if (ret < 0) {
 741         return ret;
 742     }
 743
 744     if (ret != sizeof(buf)) {
 745         LOG("read failed");
 746         return -EINVAL;
 747     }
 748
 749     /* Reply
 750        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 751        [ 4 ..  7]    error   (0 == no error)
 752        [ 7 .. 15]    handle
 753      */
 754
 755     magic = be32_to_cpup((uint32_t*)buf);
 756     reply->error  = be32_to_cpup((uint32_t*)(buf + 4));
 757     reply->handle = be64_to_cpup((uint64_t*)(buf + 8));
 758
 759     TRACE("Got reply: "
 760           "{ magic = 0x%x, .error = %d, handle = %" PRIu64" }",
 761           magic, reply->error, reply->handle);
 762
 763     if (magic != NBD_REPLY_MAGIC) {
 764         LOG("invalid magic (got 0x%x)", magic);
 765         return -EINVAL;
 766     }
 767     return 0;
 768 }
 769
 770 static ssize_t nbd_send_reply(int csock, struct nbd_reply *reply)
 771 {
 772     uint8_t buf[NBD_REPLY_SIZE];
 773     ssize_t ret;
 774
 775     /* Reply
 776        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 777        [ 4 ..  7]    error   (0 == no error)
 778        [ 7 .. 15]    handle
 779      */
 780     cpu_to_be32w((uint32_t*)buf, NBD_REPLY_MAGIC);
 781     cpu_to_be32w((uint32_t*)(buf + 4), reply->error);
 782     cpu_to_be64w((uint64_t*)(buf + 8), reply->handle);
 783
 784     TRACE("Sending response to client");
 785
 786     ret = write_sync(csock, buf, sizeof(buf));
 787     if (ret < 0) {
 788         return ret;
 789     }
 790
 791     if (ret != sizeof(buf)) {
 792         LOG("writing to socket failed");
 793         return -EINVAL;
 794     }
 795     return 0;
 796 }
 797
 798 #define MAX_NBD_REQUESTS 16
 799
 800 void nbd_client_get(NBDClient *client)
 801 {
 802     client->refcount++;
 803 }
 804
 805 void nbd_client_put(NBDClient *client)
 806 {
 807     if (--client->refcount == 0) {
 808         /* The last reference should be dropped by client->close,
 809          * which is called by nbd_client_close.
 810          */
 811         assert(client->closing);
 812
 813         qemu_set_fd_handler2(client->sock, NULL, NULL, NULL, NULL);
 814         close(client->sock);
 815         client->sock = -1;
 816         if (client->exp) {
 817             QTAILQ_REMOVE(&client->exp->clients, client, next);
 818             nbd_export_put(client->exp);
 819         }
 820         g_free(client);
 821     }
 822 }
 823
 824 void nbd_client_close(NBDClient *client)
 825 {
 826     if (client->closing) {
 827         return;
 828     }
 829
 830     client->closing = true;
 831
 832     /* Force requests to finish.  They will drop their own references,
 833      * then we'll close the socket and free the NBDClient.
 834      */
 835     shutdown(client->sock, 2);
 836
 837     /* Also tell the client, so that they release their reference.  */
 838     if (client->close) {
 839         client->close(client);
 840     }
 841 }
 842
 843 static NBDRequest *nbd_request_get(NBDClient *client)
 844 {
 845     NBDRequest *req;
 846
 847     assert(client->nb_requests <= MAX_NBD_REQUESTS - 1);
 848     client->nb_requests++;
 849
 850     req = g_slice_new0(NBDRequest);
 851     nbd_client_get(client);
 852     req->client = client;
 853     return req;
 854 }
 855
 856 static void nbd_request_put(NBDRequest *req)
 857 {
 858     NBDClient *client = req->client;
 859
 860     if (req->data) {
 861         qemu_vfree(req->data);
 862     }
 863     g_slice_free(NBDRequest, req);
 864
 865     if (client->nb_requests-- == MAX_NBD_REQUESTS) {
 866         qemu_notify_event();
 867     }
 868     nbd_client_put(client);
 869 }
 870
 871 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset,
 872                           off_t size, uint32_t nbdflags,
 873                           void (*close)(NBDExport *))
 874 {
 875     NBDExport *exp = g_malloc0(sizeof(NBDExport));
 876     exp->refcount = 1;
 877     QTAILQ_INIT(&exp->clients);
 878     exp->bs = bs;
 879     exp->dev_offset = dev_offset;
 880     exp->nbdflags = nbdflags;
 881     exp->size = size == -1 ? bdrv_getlength(bs) : size;
 882     exp->close = close;
 883     bdrv_ref(bs);
 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     if (exp->bs) {
 931         bdrv_unref(exp->bs);
 932         exp->bs = NULL;
 933     }
 934 }
 935
 936 void nbd_export_get(NBDExport *exp)
 937 {
 938     assert(exp->refcount > 0);
 939     exp->refcount++;
 940 }
 941
 942 void nbd_export_put(NBDExport *exp)
 943 {
 944     assert(exp->refcount > 0);
 945     if (exp->refcount == 1) {
 946         nbd_export_close(exp);
 947     }
 948
 949     if (--exp->refcount == 0) {
 950         assert(exp->name == NULL);
 951
 952         if (exp->close) {
 953             exp->close(exp);
 954         }
 955
 956         g_free(exp);
 957     }
 958 }
 959
 960 BlockDriverState *nbd_export_get_blockdev(NBDExport *exp)
 961 {
 962     return exp->bs;
 963 }
 964
 965 void nbd_export_close_all(void)
 966 {
 967     NBDExport *exp, *next;
 968
 969     QTAILQ_FOREACH_SAFE(exp, &exports, next, next) {
 970         nbd_export_close(exp);
 971     }
 972 }
 973
 974 static int nbd_can_read(void *opaque);
 975 static void nbd_read(void *opaque);
 976 static void nbd_restart_write(void *opaque);
 977
 978 static ssize_t nbd_co_send_reply(NBDRequest *req, struct nbd_reply *reply,
 979                                  int len)
 980 {
 981     NBDClient *client = req->client;
 982     int csock = client->sock;
 983     ssize_t rc, ret;
 984
 985     qemu_co_mutex_lock(&client->send_lock);
 986     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read,
 987                          nbd_restart_write, client);
 988     client->send_coroutine = qemu_coroutine_self();
 989
 990     if (!len) {
 991         rc = nbd_send_reply(csock, reply);
 992     } else {
 993         socket_set_cork(csock, 1);
 994         rc = nbd_send_reply(csock, reply);
 995         if (rc >= 0) {
 996             ret = qemu_co_send(csock, req->data, len);
 997             if (ret != len) {
 998                 rc = -EIO;
 999             }
1000         }
1001         socket_set_cork(csock, 0);
1002     }
1003
1004     client->send_coroutine = NULL;
1005     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1006     qemu_co_mutex_unlock(&client->send_lock);
1007     return rc;
1008 }
1009
1010 static ssize_t nbd_co_receive_request(NBDRequest *req, struct nbd_request *request)
1011 {
1012     NBDClient *client = req->client;
1013     int csock = client->sock;
1014     uint32_t command;
1015     ssize_t rc;
1016
1017     client->recv_coroutine = qemu_coroutine_self();
1018     rc = nbd_receive_request(csock, request);
1019     if (rc < 0) {
1020         if (rc != -EAGAIN) {
1021             rc = -EIO;
1022         }
1023         goto out;
1024     }
1025
1026     if (request->len > NBD_MAX_BUFFER_SIZE) {
1027         LOG("len (%u) is larger than max len (%u)",
1028             request->len, NBD_MAX_BUFFER_SIZE);
1029         rc = -EINVAL;
1030         goto out;
1031     }
1032
1033     if ((request->from + request->len) < request->from) {
1034         LOG("integer overflow detected! "
1035             "you're probably being attacked");
1036         rc = -EINVAL;
1037         goto out;
1038     }
1039
1040     TRACE("Decoding type");
1041
1042     command = request->type & NBD_CMD_MASK_COMMAND;
1043     if (command == NBD_CMD_READ || command == NBD_CMD_WRITE) {
1044         req->data = qemu_blockalign(client->exp->bs, request->len);
1045     }
1046     if (command == NBD_CMD_WRITE) {
1047         TRACE("Reading %u byte(s)", request->len);
1048
1049         if (qemu_co_recv(csock, req->data, request->len) != request->len) {
1050             LOG("reading from socket failed");
1051             rc = -EIO;
1052             goto out;
1053         }
1054     }
1055     rc = 0;
1056
1057 out:
1058     client->recv_coroutine = NULL;
1059     return rc;
1060 }
1061
1062 static void nbd_trip(void *opaque)
1063 {
1064     NBDClient *client = opaque;
1065     NBDExport *exp = client->exp;
1066     NBDRequest *req;
1067     struct nbd_request request;
1068     struct nbd_reply reply;
1069     ssize_t ret;
1070
1071     TRACE("Reading request.");
1072     if (client->closing) {
1073         return;
1074     }
1075
1076     req = nbd_request_get(client);
1077     ret = nbd_co_receive_request(req, &request);
1078     if (ret == -EAGAIN) {
1079         goto done;
1080     }
1081     if (ret == -EIO) {
1082         goto out;
1083     }
1084
1085     reply.handle = request.handle;
1086     reply.error = 0;
1087
1088     if (ret < 0) {
1089         reply.error = -ret;
1090         goto error_reply;
1091     }
1092
1093     if ((request.from + request.len) > exp->size) {
1094             LOG("From: %" PRIu64 ", Len: %u, Size: %" PRIu64
1095             ", Offset: %" PRIu64 "\n",
1096                     request.from, request.len,
1097                     (uint64_t)exp->size, (uint64_t)exp->dev_offset);
1098         LOG("requested operation past EOF--bad client?");
1099         goto invalid_request;
1100     }
1101
1102     switch (request.type & NBD_CMD_MASK_COMMAND) {
1103     case NBD_CMD_READ:
1104         TRACE("Request type is READ");
1105
1106         if (request.type & NBD_CMD_FLAG_FUA) {
1107             ret = bdrv_co_flush(exp->bs);
1108             if (ret < 0) {
1109                 LOG("flush failed");
1110                 reply.error = -ret;
1111                 goto error_reply;
1112             }
1113         }
1114
1115         ret = bdrv_read(exp->bs, (request.from + exp->dev_offset) / 512,
1116                         req->data, request.len / 512);
1117         if (ret < 0) {
1118             LOG("reading from file failed");
1119             reply.error = -ret;
1120             goto error_reply;
1121         }
1122
1123         TRACE("Read %u byte(s)", request.len);
1124         if (nbd_co_send_reply(req, &reply, request.len) < 0)
1125             goto out;
1126         break;
1127     case NBD_CMD_WRITE:
1128         TRACE("Request type is WRITE");
1129
1130         if (exp->nbdflags & NBD_FLAG_READ_ONLY) {
1131             TRACE("Server is read-only, return error");
1132             reply.error = EROFS;
1133             goto error_reply;
1134         }
1135
1136         TRACE("Writing to device");
1137
1138         ret = bdrv_write(exp->bs, (request.from + exp->dev_offset) / 512,
1139                          req->data, request.len / 512);
1140         if (ret < 0) {
1141             LOG("writing to file failed");
1142             reply.error = -ret;
1143             goto error_reply;
1144         }
1145
1146         if (request.type & NBD_CMD_FLAG_FUA) {
1147             ret = bdrv_co_flush(exp->bs);
1148             if (ret < 0) {
1149                 LOG("flush failed");
1150                 reply.error = -ret;
1151                 goto error_reply;
1152             }
1153         }
1154
1155         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1156             goto out;
1157         }
1158         break;
1159     case NBD_CMD_DISC:
1160         TRACE("Request type is DISCONNECT");
1161         errno = 0;
1162         goto out;
1163     case NBD_CMD_FLUSH:
1164         TRACE("Request type is FLUSH");
1165
1166         ret = bdrv_co_flush(exp->bs);
1167         if (ret < 0) {
1168             LOG("flush failed");
1169             reply.error = -ret;
1170         }
1171         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1172             goto out;
1173         }
1174         break;
1175     case NBD_CMD_TRIM:
1176         TRACE("Request type is TRIM");
1177         ret = bdrv_co_discard(exp->bs, (request.from + exp->dev_offset) / 512,
1178                               request.len / 512);
1179         if (ret < 0) {
1180             LOG("discard failed");
1181             reply.error = -ret;
1182         }
1183         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1184             goto out;
1185         }
1186         break;
1187     default:
1188         LOG("invalid request type (%u) received", request.type);
1189     invalid_request:
1190         reply.error = -EINVAL;
1191     error_reply:
1192         if (nbd_co_send_reply(req, &reply, 0) < 0) {
1193             goto out;
1194         }
1195         break;
1196     }
1197
1198     TRACE("Request/Reply complete");
1199
1200 done:
1201     nbd_request_put(req);
1202     return;
1203
1204 out:
1205     nbd_request_put(req);
1206     nbd_client_close(client);
1207 }
1208
1209 static int nbd_can_read(void *opaque)
1210 {
1211     NBDClient *client = opaque;
1212
1213     return client->recv_coroutine || client->nb_requests < MAX_NBD_REQUESTS;
1214 }
1215
1216 static void nbd_read(void *opaque)
1217 {
1218     NBDClient *client = opaque;
1219
1220     if (client->recv_coroutine) {
1221         qemu_coroutine_enter(client->recv_coroutine, NULL);
1222     } else {
1223         qemu_coroutine_enter(qemu_coroutine_create(nbd_trip), client);
1224     }
1225 }
1226
1227 static void nbd_restart_write(void *opaque)
1228 {
1229     NBDClient *client = opaque;
1230
1231     qemu_coroutine_enter(client->send_coroutine, NULL);
1232 }
1233
1234 NBDClient *nbd_client_new(NBDExport *exp, int csock,
1235                           void (*close)(NBDClient *))
1236 {
1237     NBDClient *client;
1238     client = g_malloc0(sizeof(NBDClient));
1239     client->refcount = 1;
1240     client->exp = exp;
1241     client->sock = csock;
1242     if (nbd_send_negotiate(client) < 0) {
1243         g_free(client);
1244         return NULL;
1245     }
1246     client->close = close;
1247     qemu_co_mutex_init(&client->send_lock);
1248     qemu_set_fd_handler2(csock, nbd_can_read, nbd_read, NULL, client);
1249
1250     if (exp) {
1251         QTAILQ_INSERT_TAIL(&exp->clients, client, next);
1252         nbd_export_get(exp);
1253     }
1254     return client;
1255 }