migration.c

   1 /*
   2  * QEMU migration support
   3  *
   4  * Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "hw/hw.h"
  26 #include "sysemu.h"
  27 #include "block.h"
  28 #include "console.h"
  29 #include "qemu-timer.h"
  30 #include "migration.h"
  31 #include "qemu_socket.h"
  32 #include "qemu-kvm.h"
  33
  34 #include <sys/wait.h>
  35
  36 #define MIN_FINALIZE_SIZE       (200 << 10)
  37 #define MAX_ITERATIONS           30
  38 #define MAX_RAPID_WRITES          5
  39
  40 typedef struct MigrationState
  41 {
  42     int fd;
  43     int throttle_count;
  44     int bps;
  45     int updated_pages;
  46     int last_updated_pages;
  47     int iteration;
  48     int n_buffer; /* number of bytes in @buffer already sent */
  49     int l_buffer; /* number of bytes in @buffer to send */
  50     int throttled;
  51     int *has_error;
  52     char buffer[TARGET_PAGE_SIZE + 4 + 4];
  53     target_ulong addr;
  54     QEMUTimer *timer;
  55     void *opaque;
  56     int detach;
  57     int (*release)(void *opaque);
  58     int rapid_writes;
  59 } MigrationState;
  60
  61 static uint32_t max_throttle = (32 << 20);
  62 static MigrationState *current_migration;
  63 static int wait_for_message_timeout = 3000; /* 3 seconds */
  64 static int status; /* last migration status */
  65
  66 enum { /* migration status values */
  67     MIG_STAT_SUCCESS           = 0,
  68
  69     /* src error codes */
  70     MIG_STAT_INVALID_PARAMS    = 1,
  71     MIG_STAT_INVALID_ADDR      = 2,
  72     MIG_STAT_SOCKET_FAILED     = 3,
  73     MIG_STAT_SOCKOPT_FAILED    = 4,
  74     MIG_STAT_BIND_FAILED       = 5,
  75     MIG_STAT_LISTEN_FAILED     = 6,
  76     MIG_STAT_ACCEPT_FAILED     = 7,
  77     MIG_STAT_CONNECT_FAILED    = 8,
  78     MIG_STAT_WRITE_FAILED      = 9,
  79     MIG_STAT_READ_FAILED       = 10,
  80     MIG_STAT_CONNECTION_CLOSED = 11,
  81     MIG_STAT_SELECT_FAILED     = 12,
  82     MIG_STAT_SELECT_TIMEOUT    = 13,
  83     MIG_STAT_SELECT_FD_NOT_SET = 14,
  84
  85     MIG_STAT_SAVEVM_FAILED     = 15,
  86     MIG_STAT_NO_MEM            = 16,
  87
  88     MIG_STAT_MIGRATION_CANCEL  = 20,
  89
  90     /* kvm error codes (on src) */
  91     MIG_STAT_KVM_UPDATE_DIRTY_PAGES_LOG_FAILED  = 101,
  92     MIG_STAT_KVM_SET_DIRTY_TRACKING_FAILED      = 102,
  93     MIG_STAT_KVM_GET_PAGE_BITMAP                = 103,
  94
  95     /* dst error codes */
  96     MIG_STAT_DST_INVALID_PARAMS    = 200 + MIG_STAT_INVALID_PARAMS,
  97     MIG_STAT_DST_INVALID_ADDR      = 200 + MIG_STAT_INVALID_ADDR,
  98     MIG_STAT_DST_SOCKET_FAILED     = 200 + MIG_STAT_SOCKET_FAILED,
  99     MIG_STAT_DST_SOCKOPT_FAILED    = 200 + MIG_STAT_SOCKOPT_FAILED,
 100     MIG_STAT_DST_BIND_FAILED       = 200 + MIG_STAT_BIND_FAILED,
 101     MIG_STAT_DST_LISTEN_FAILED     = 200 + MIG_STAT_LISTEN_FAILED,
 102     MIG_STAT_DST_ACCEPT_FAILED     = 200 + MIG_STAT_ACCEPT_FAILED,
 103     MIG_STAT_DST_CONNECT_FAILED    = 200 + MIG_STAT_CONNECT_FAILED,
 104     MIG_STAT_DST_WRITE_FAILED      = 200 + MIG_STAT_WRITE_FAILED,
 105     MIG_STAT_DST_READ_FAILED       = 200 + MIG_STAT_READ_FAILED,
 106
 107     MIG_STAT_DST_CONNECTION_CLOSED = 200 + MIG_STAT_CONNECTION_CLOSED,
 108     MIG_STAT_DST_SELECT_FAILED     = 200 + MIG_STAT_SELECT_FAILED,
 109     MIG_STAT_DST_SELECT_TIMEOUT    = 200 + MIG_STAT_SELECT_TIMEOUT,
 110     MIG_STAT_DST_SELECT_FD_NOT_SET = 200 + MIG_STAT_SELECT_FD_NOT_SET,
 111
 112     MIG_STAT_DST_LOADVM_FAILED     = 200 + MIG_STAT_SAVEVM_FAILED,
 113     MIG_STAT_DST_NO_MEM            = 200 + MIG_STAT_NO_MEM,
 114
 115     MIG_STAT_DST_GET_PAGE_FAILED       = 230,
 116     MIG_STAT_DST_GET_PAGE_UNKNOWN_TYPE = 231,
 117     MIG_STAT_DST_MEM_SIZE_MISMATCH     = 232,
 118     MIG_STAT_DST_MEM_OUT_OF_RANGE      = 233,
 119 };
 120
 121 //#define MIGRATION_VERIFY
 122 #ifdef MIGRATION_VERIFY
 123 static int save_verify_memory(QEMUFile *f, void *opaque);
 124 static int load_verify_memory(QEMUFile *f, void *opaque, int version_id);
 125 #endif /* MIGRATION_VERIFY */
 126
 127 /* QEMUFile migration implementation */
 128
 129 static void migrate_put_buffer(void *opaque, const uint8_t *buf, int64_t pos, int size)
 130 {
 131     MigrationState *s = opaque;
 132     int offset = 0;
 133
 134     if (*s->has_error)
 135         return;
 136
 137     while (offset < size) {
 138         ssize_t len;
 139
 140         len = write(s->fd, buf + offset, size - offset);
 141         if (len == -1) {
 142             if (errno == EAGAIN || errno == EINTR)
 143                 continue;
 144             term_printf("migration: write failed (%s)\n", strerror(errno));
 145             *s->has_error = MIG_STAT_WRITE_FAILED;
 146             break;
 147         } else if (len == 0) {
 148             term_printf("migration: other side closed connection\n");
 149             *s->has_error = MIG_STAT_CONNECTION_CLOSED;
 150             break;
 151         }
 152
 153         offset += len;
 154     }
 155 }
 156
 157 static void migrate_close(void *opaque)
 158 {
 159     MigrationState *s = opaque;
 160
 161     if (s->release)
 162         s->release(s->opaque);
 163
 164     qemu_free(s);
 165     current_migration = NULL;
 166 }
 167
 168 /* Outgoing migration routines */
 169 static void migrate_finish(MigrationState *s)
 170 {
 171     QEMUFile *f;
 172     int ret = 0;
 173     int *has_error = s->has_error;
 174     int saved_vm_running = vm_running;
 175     int detach = s->detach;
 176
 177     fcntl(s->fd, F_SETFL, 0);
 178
 179     if (! *has_error) {
 180         vm_stop(0);
 181         f = qemu_fopen(s, migrate_put_buffer, NULL, migrate_close);
 182         do {
 183             qemu_aio_flush();
 184         } while (qemu_bh_poll());
 185         bdrv_flush_all();
 186
 187         if (kvm_enabled() && !*s->has_error && kvm_update_dirty_pages_log())
 188             *s->has_error = MIG_STAT_KVM_UPDATE_DIRTY_PAGES_LOG_FAILED;
 189
 190         qemu_put_be32(f, 1);
 191         ret = qemu_live_savevm_state(f);
 192 #ifdef MIGRATION_VERIFY
 193         save_verify_memory(f, NULL);
 194 #endif /* MIGRATION_VERIFY */
 195         qemu_fclose(f);
 196     } else {
 197             migrate_close(s);
 198     }
 199     status = *has_error;
 200     if (ret && !status)
 201         status = MIG_STAT_SAVEVM_FAILED;
 202     if (status) {
 203         term_printf("Migration failed! ret=%d error=%d\n", ret, *has_error);
 204         if (saved_vm_running)
 205             vm_start();
 206     }
 207     if (!detach)
 208         monitor_resume();
 209     qemu_free(has_error);
 210     cpu_physical_memory_set_dirty_tracking(0);
 211 }
 212
 213 static int migrate_write_buffer(MigrationState *s)
 214 {
 215     if (*s->has_error)
 216         return 0;
 217
 218     if (s->n_buffer < s->l_buffer) {
 219         ssize_t len;
 220     again:
 221         len = write(s->fd, s->buffer + s->n_buffer, s->l_buffer - s->n_buffer);
 222         if (len == -1) {
 223             if (errno == EINTR)
 224                 goto again;
 225             if (errno == EAGAIN)
 226                 return 1;
 227             *s->has_error = MIG_STAT_WRITE_FAILED;
 228             return 0;
 229         }
 230         if (len == 0) {
 231             *s->has_error = MIG_STAT_CONNECTION_CLOSED;
 232             return 0;
 233         }
 234
 235         s->throttle_count += len;
 236         s->n_buffer += len;
 237         if (s->n_buffer < s->l_buffer)
 238             goto again;
 239     }
 240
 241     if (s->throttle_count > max_throttle) {
 242         s->throttled = 1;
 243         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 244         return 1;
 245     }
 246
 247     return 0;
 248 }
 249
 250 static int migrate_check_convergence(MigrationState *s)
 251 {
 252     target_ulong addr;
 253     int dirty_count = 0;
 254
 255     if ((s->iteration >= MAX_ITERATIONS) ||
 256         (s->rapid_writes >= MAX_RAPID_WRITES) ) {
 257         return 1;
 258     }
 259
 260     for (addr = 0; addr < phys_ram_size; addr += TARGET_PAGE_SIZE) {
 261         if (kvm_enabled() && (addr>=0xa0000) && (addr<0xc0000)) /* do not access video-addresses */
 262             continue;
 263         if (cpu_physical_memory_get_dirty(addr, MIGRATION_DIRTY_FLAG))
 264             dirty_count++;
 265     }
 266
 267     return ((dirty_count * TARGET_PAGE_SIZE) < MIN_FINALIZE_SIZE);
 268 }
 269
 270 static int ram_page_is_homogeneous(uint32_t addr)
 271 {
 272     int i, n;
 273     uint32_t *p, v;
 274
 275     n = TARGET_PAGE_SIZE / sizeof(v);
 276     p = (uint32 *)(phys_ram_base + addr);
 277
 278     v = p[0];
 279     for (i=1; i<n; i++)
 280         if (p[i] != v)
 281             return 0;
 282     return 1;
 283 }
 284
 285 static void migrate_prepare_page(MigrationState *s)
 286 {
 287     uint32_t value;
 288     char type;
 289     const char *buff;
 290     int bufflen;
 291
 292     value = cpu_to_be32(s->addr);
 293     memcpy(s->buffer, &value, 4);
 294
 295     if (ram_page_is_homogeneous(s->addr)) {
 296         type = 1; /* keeping ram_get_page() values */
 297         bufflen = 4;
 298     }
 299     else {
 300         type = 0;
 301         bufflen = TARGET_PAGE_SIZE;
 302     }
 303
 304     buff = phys_ram_base + s->addr;
 305     s->buffer[4] = type;
 306     memcpy(s->buffer + 4 + 1, phys_ram_base + s->addr, bufflen);
 307     s->n_buffer = 0;
 308     s->l_buffer = 4 + 1 + bufflen;
 309 }
 310
 311 static void migrate_write(void *opaque)
 312 {
 313     MigrationState *s = opaque;
 314
 315     if (migrate_write_buffer(s))
 316         return;
 317
 318     if (kvm_enabled() && !*s->has_error && kvm_update_dirty_pages_log())
 319         *s->has_error = MIG_STAT_KVM_UPDATE_DIRTY_PAGES_LOG_FAILED;
 320
 321     if (migrate_check_convergence(s) || *s->has_error) {
 322         qemu_del_timer(s->timer);
 323         qemu_free_timer(s->timer);
 324         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 325         migrate_finish(s);
 326         return;
 327     }
 328
 329     while (s->addr < phys_ram_size) {
 330         if (kvm_enabled() && (s->addr>=0xa0000) && (s->addr<0xc0000)) /* do not access video-addresses */
 331             s->addr = 0xc0000;
 332
 333         if (cpu_physical_memory_get_dirty(s->addr, MIGRATION_DIRTY_FLAG)) {
 334             migrate_prepare_page(s);
 335             cpu_physical_memory_reset_dirty(s->addr, s->addr + TARGET_PAGE_SIZE, MIGRATION_DIRTY_FLAG);
 336
 337             s->addr += TARGET_PAGE_SIZE;
 338
 339             s->updated_pages++;
 340
 341             if (migrate_write_buffer(s))
 342                 return;
 343         } else
 344             s->addr += TARGET_PAGE_SIZE;
 345     }
 346
 347     if ((s->iteration) && (s->last_updated_pages <= s->updated_pages)) {
 348         s->rapid_writes++; /* "dirt-speed" is faster than transfer speed */
 349         if (max_throttle < (1 << 30))
 350             max_throttle *= 2; /* try harder */
 351     }
 352     s->last_updated_pages = s->updated_pages;
 353     s->updated_pages = 0;
 354     s->addr = 0;
 355     s->iteration++;
 356 }
 357
 358 static void migrate_reset_throttle(void *opaque)
 359 {
 360     MigrationState *s = opaque;
 361
 362     s->bps = s->throttle_count;
 363
 364     if (s->throttled) {
 365         s->throttled = 0;
 366         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_write, s);
 367     }
 368     s->throttle_count = 0;
 369     qemu_mod_timer(s->timer, qemu_get_clock(rt_clock) + 1000);
 370 }
 371
 372 static int write_whole_buffer(int fd, const void *buff, size_t size)
 373 {
 374     size_t offset = 0, len;
 375
 376     while (offset < size) {
 377         len = write(fd, buff + offset, size - offset);
 378         if (len == -1 && errno == EINTR)
 379             continue;
 380         if (len < 1)
 381             return -EIO;
 382
 383         offset += len;
 384     }
 385     return !(offset == size); /* returns 0 on success */
 386 }
 387
 388 static int bit_is_set(int bit, unsigned char *map)
 389 {
 390     return map[bit/8] & (1 << (bit%8));
 391 }
 392
 393 static int start_migration(MigrationState *s)
 394 {
 395     uint32_t value = cpu_to_be32(phys_ram_size);
 396     target_phys_addr_t addr;
 397     int r;
 398     unsigned char running = vm_running?2:1; /* 1 + vm_running */
 399     int n = 0;
 400     unsigned char *phys_ram_page_exist_bitmap = NULL;
 401
 402     if (kvm_enabled()) {
 403         n = BITMAP_SIZE(phys_ram_size);
 404         phys_ram_page_exist_bitmap = qemu_malloc(n);
 405         if (!phys_ram_page_exist_bitmap) {
 406             perror("failed to allocate page bitmap");
 407             r = MIG_STAT_NO_MEM;
 408             goto out;
 409         }
 410         if (kvm_get_phys_ram_page_bitmap(phys_ram_page_exist_bitmap)) {
 411             r = MIG_STAT_KVM_GET_PAGE_BITMAP;
 412             perror("kvm_get_mem_map failed");
 413             goto out;
 414         }
 415     }
 416
 417     r = MIG_STAT_WRITE_FAILED;
 418     if (write_whole_buffer(s->fd, &running, sizeof(running))) {
 419         perror("vm_running write failed");
 420         goto out;
 421     }
 422     if (write_whole_buffer(s->fd, &value, sizeof(value))) {
 423         perror("phys_ram_size write failed");
 424         goto out;
 425     }
 426
 427     if (kvm_enabled()) {
 428         value = cpu_to_be32(n);
 429         if (write_whole_buffer(s->fd, &value, sizeof(value))) {
 430             perror("phys_ram_size_bitmap size write failed");
 431             goto out;
 432         }
 433         if (write_whole_buffer(s->fd, phys_ram_page_exist_bitmap, n)) {
 434             perror("phys_ram_page_exist_bitmap write failed");
 435             goto out;
 436         }
 437     }
 438
 439     fcntl(s->fd, F_SETFL, O_NONBLOCK);
 440
 441     for (addr = 0; addr < phys_ram_size; addr += TARGET_PAGE_SIZE) {
 442         if (kvm_enabled() && !bit_is_set(addr>>TARGET_PAGE_BITS, phys_ram_page_exist_bitmap)) {
 443             cpu_physical_memory_reset_dirty(addr,
 444                                             addr + TARGET_PAGE_SIZE,
 445                                             MIGRATION_DIRTY_FLAG);
 446             continue;
 447         }
 448         if (!cpu_physical_memory_get_dirty(addr, MIGRATION_DIRTY_FLAG))
 449             cpu_physical_memory_set_dirty(addr);
 450     }
 451
 452     if (cpu_physical_memory_set_dirty_tracking(1)) {
 453         *s->has_error = MIG_STAT_KVM_SET_DIRTY_TRACKING_FAILED;
 454         r = MIG_STAT_KVM_SET_DIRTY_TRACKING_FAILED;
 455         goto out;
 456     }
 457
 458     s->addr = 0;
 459     s->iteration = 0;
 460     s->updated_pages = 0;
 461     s->last_updated_pages = 0;
 462     s->n_buffer = s->l_buffer = 0;
 463     s->rapid_writes = 0;
 464     s->timer = qemu_new_timer(rt_clock, migrate_reset_throttle, s);
 465
 466     qemu_mod_timer(s->timer, qemu_get_clock(rt_clock));
 467     qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_write, s);
 468
 469     r = 0;
 470
 471  out:
 472     if (kvm_enabled() && phys_ram_page_exist_bitmap)
 473         qemu_free(phys_ram_page_exist_bitmap);
 474     return r;
 475 }
 476
 477 static MigrationState *migration_init_fd(int detach, int fd)
 478 {
 479     MigrationState *s;
 480
 481     s = qemu_mallocz(sizeof(MigrationState));
 482     if (s == NULL) {
 483         term_printf("Allocation error\n");
 484         return NULL;
 485     }
 486
 487     s->fd = fd;
 488     s->has_error = qemu_mallocz(sizeof(int));
 489     if (s->has_error == NULL) {
 490         term_printf("malloc failed (for has_error)\n");
 491         return NULL;
 492     }
 493     s->detach = detach;
 494
 495     current_migration = s;
 496
 497     if (start_migration(s)) {
 498         term_printf("Could not start migration\n");
 499         migrate_finish(s);
 500         return NULL;
 501     }
 502
 503     if (!detach)
 504         monitor_suspend();
 505
 506     return s;
 507 }
 508
 509 typedef struct MigrationCmdState
 510 {
 511     int fd;
 512     pid_t pid;
 513 } MigrationCmdState;
 514
 515 static int cmd_release(void *opaque)
 516 {
 517     MigrationCmdState *c = opaque;
 518     int status, ret;
 519
 520     close(c->fd);
 521
 522 again:
 523     ret = waitpid(c->pid, &status, 0);
 524     if (ret == -1 && errno == EINTR)
 525         goto again;
 526
 527     if (ret == -1) {
 528         term_printf("migration: waitpid failed (%s)\n", strerror(errno));
 529         return -1;
 530     }
 531     /* FIXME: check and uncomment
 532      * if (WIFEXITED(status))
 533      *     status = WEXITSTATUS(status);
 534      */
 535     return status;
 536 }
 537
 538 static MigrationState *migration_init_cmd(int detach, const char *command, char **argv)
 539 {
 540     int fds[2];
 541     pid_t pid;
 542     int i;
 543     MigrationState *s;
 544
 545     if (pipe(fds) == -1) {
 546         term_printf("pipe() (%s)\n", strerror(errno));
 547         return NULL;
 548     }
 549
 550     pid = fork();
 551     if (pid == -1) {
 552         close(fds[0]);
 553         close(fds[1]);
 554         term_printf("fork error (%s)\n", strerror(errno));
 555         return NULL;
 556     }
 557     if (pid == 0) {
 558         close(fds[1]);
 559         dup2(fds[0], STDIN_FILENO);
 560         execvp(command, argv);
 561         exit(1);
 562     } else
 563         close(fds[0]);
 564
 565     for (i = 0; argv[i]; i++)
 566         qemu_free(argv[i]);
 567     qemu_free(argv);
 568
 569     s = migration_init_fd(detach, fds[1]);
 570     if (s) {
 571         MigrationCmdState *c = qemu_mallocz(sizeof(*c));
 572         c->pid = pid;
 573         c->fd = fds[1];
 574         s->release = cmd_release;
 575         s->opaque = c;
 576     }
 577
 578     return s;
 579 }
 580
 581 static MigrationState *migration_init_exec(int detach, const char *command)
 582 {
 583     char **argv = NULL;
 584
 585     argv = qemu_mallocz(sizeof(char *) * 4);
 586     argv[0] = strdup("sh");
 587     argv[1] = strdup("-c");
 588     argv[2] = strdup(command);
 589     argv[3] = NULL;
 590
 591     return migration_init_cmd(detach, "/bin/sh", argv);
 592 }
 593
 594 static MigrationState *migration_init_ssh(int detach, const char *host)
 595 {
 596     int qemu_argc, daemonize = 0, argc, i;
 597     char **qemu_argv, **argv;
 598     const char *incoming = NULL;
 599
 600     qemu_get_launch_info(&qemu_argc, &qemu_argv, &daemonize, &incoming);
 601
 602     argc = 3 + qemu_argc;
 603     if (!daemonize)
 604         argc++;
 605     if (!incoming)
 606         argc+=2;
 607
 608     argv = qemu_mallocz(sizeof(char *) * (argc + 1));
 609     argv[0] = strdup("ssh");
 610     argv[1] = strdup("-XC");
 611     argv[2] = strdup(host);
 612
 613     for (i = 0; i < qemu_argc; i++)
 614         argv[3 + i] = strdup(qemu_argv[i]);
 615
 616     if (!daemonize)
 617         argv[3 + i++] = strdup("-daemonize");
 618     if (!incoming) {
 619         argv[3 + i++] = strdup("-incoming");
 620         argv[3 + i++] = strdup("stdio");
 621     }
 622
 623     argv[3 + i] = NULL;
 624
 625     return migration_init_cmd(detach, "ssh", argv);
 626 }
 627
 628 /* (busy) wait timeout (miliseconds) for a message to arrive on fd. */
 629 /* returns 0 on success error_code otherwise (18 for timeout) */
 630 static int wait_for_message(const char *msg, int fd, int timeout)
 631 {
 632     fd_set rfds;
 633     struct timeval tv;
 634     int64_t now, expiration, delta; /* milliseconds */
 635     int n = 0;
 636
 637     now = qemu_get_clock(rt_clock);
 638     expiration = now + timeout;
 639     do {
 640         FD_ZERO(&rfds);
 641         FD_SET(fd, &rfds);
 642         tv.tv_sec = tv.tv_usec = 0;
 643         now = qemu_get_clock(rt_clock);
 644         delta = expiration - now;
 645         if (delta > 0)
 646             tv.tv_usec = delta * 1000;
 647         n = select(fd + 1, &rfds, NULL, NULL, &tv);
 648     } while ( (n == -1) && (errno == EINTR) );
 649
 650     switch (n) {
 651     case -1: /* error */
 652         fprintf(stderr, "%s FAILED: ", msg);
 653         perror("");
 654         return MIG_STAT_SELECT_FAILED;
 655     case 0: /* timeout */
 656         fprintf(stderr, "%s: timeout reached\n", msg);
 657         return MIG_STAT_SELECT_TIMEOUT;
 658     case 1:
 659         break;
 660     default:
 661         fprintf(stderr, "wait_for_message: %s: select returned  %d\n", msg, n);
 662     }
 663     if (!FD_ISSET(fd, &rfds)) {
 664         fprintf(stderr, "wait_for_message: %s: s->fd not set\n", msg);
 665         return MIG_STAT_SELECT_FD_NOT_SET;
 666     }
 667
 668     return 0;
 669 }
 670
 671 static int tcp_release(void *opaque)
 672 {
 673     MigrationState *s = opaque;
 674     uint8_t status = 0;
 675     ssize_t len = 0;
 676     int n;
 677
 678     if (*s->has_error)
 679         goto out;
 680
 681     n = wait_for_message("WAIT FOR ACK", s->fd, wait_for_message_timeout);
 682     if (n) {
 683         *s->has_error = n;
 684         goto out;
 685     }
 686
 687 wait_for_ack:
 688     len = read(s->fd, &status, 1);
 689     if (len == -1 && errno == EINTR)
 690         goto wait_for_ack;
 691     if (len != 1 || status != 0) {
 692         *s->has_error = MIG_STAT_READ_FAILED;
 693         fprintf(stderr, "migration: wait_for_ack: read error l=%zu s=%d(%s)\n",
 694                 len, status, strerror(errno));
 695         goto out;
 696     }
 697
 698 send_go:
 699     len = write(s->fd, &status, 1);
 700     if (len == -1 && errno == EINTR)
 701         goto send_go;
 702     if (len != 1) {
 703         fprintf(stderr, "migration: send_go: write error l=%zu (%s)\n",
 704                 len, strerror(errno));
 705         *s->has_error = MIG_STAT_WRITE_FAILED;
 706     }
 707
 708 out:
 709     close(s->fd);
 710
 711     return (len != 1 || status != 0);
 712 }
 713
 714 static MigrationState *migration_init_tcp(int detach, const char *host)
 715 {
 716     int fd;
 717     struct sockaddr_in addr;
 718     MigrationState *s;
 719
 720     fd = socket(PF_INET, SOCK_STREAM, 0);
 721     if (fd == -1) {
 722         term_printf("socket() failed %s\n", strerror(errno));
 723         status = MIG_STAT_SOCKET_FAILED;
 724         return NULL;
 725     }
 726
 727     addr.sin_family = AF_INET;
 728     if (parse_host_port(&addr, host) == -1) {
 729         term_printf("parse_host_port() FAILED for %s\n", host);
 730         close(fd);
 731         status = MIG_STAT_INVALID_ADDR;
 732         return NULL;
 733     }
 734
 735 again:
 736     if (connect(fd, (struct sockaddr *)&addr, sizeof(addr)) == -1) {
 737         if (errno == EINTR)
 738             goto again;
 739         term_printf("connect() failed %s\n", strerror(errno));
 740         close(fd);
 741         status = MIG_STAT_CONNECT_FAILED;
 742         return NULL;
 743     }
 744
 745     s = migration_init_fd(detach, fd);
 746     if (s) {
 747         s->opaque = s;
 748         s->release = tcp_release;
 749     }
 750     return s;
 751 }
 752
 753 /* Incoming migration */
 754
 755 static void migrate_incoming_homogeneous_page(uint32_t addr, uint32_t v)
 756 {
 757     int i, n;
 758     uint32_t *p;
 759
 760     n = TARGET_PAGE_SIZE / sizeof(v);
 761     p = (uint32 *)(phys_ram_base + addr);
 762
 763     for (i=0; i<n; i++)
 764         p[i] = v;
 765 }
 766
 767 static int migrate_incoming_page(QEMUFile *f, uint32_t addr)
 768 {
 769     int l, v, ret = 0;
 770
 771     switch (qemu_get_byte(f)) {
 772     case 0: /* the whole page */
 773         l = qemu_get_buffer(f, phys_ram_base + addr, TARGET_PAGE_SIZE);
 774         if (l != TARGET_PAGE_SIZE)
 775             ret = MIG_STAT_DST_GET_PAGE_FAILED;
 776         break;
 777     case 1: /* homogeneous page -- a single byte */
 778         l = qemu_get_buffer(f, (void*)&v, 4);
 779         if (l != 4)
 780             ret = MIG_STAT_DST_GET_PAGE_FAILED;
 781         migrate_incoming_homogeneous_page(addr, v);
 782         break;
 783     default:
 784         ret = MIG_STAT_DST_GET_PAGE_UNKNOWN_TYPE;
 785     }
 786
 787     return ret;
 788 }
 789
 790 static int migrate_incoming_fd(int fd)
 791 {
 792     int ret = 0;
 793     QEMUFile *f = qemu_fopen_fd(fd);
 794     uint32_t addr, size;
 795     extern void qemu_announce_self(void);
 796     unsigned char running;
 797
 798     running = qemu_get_byte(f);
 799     if ((running != 1) && (running != 2)) {
 800         fprintf(stderr, "migration: illegal running value %u, not (1 or 2)\n",
 801                 running);
 802         return MIG_STAT_DST_MEM_SIZE_MISMATCH;
 803     }
 804     autostart = running - 1;
 805
 806     size = qemu_get_be32(f);
 807     if (size != phys_ram_size) {
 808         fprintf(stderr, "migration: memory size mismatch: recv %u mine %u\n",
 809                 size, phys_ram_size);
 810         return MIG_STAT_DST_MEM_SIZE_MISMATCH;
 811     }
 812
 813     if (kvm_enabled()) {
 814         int n, m;
 815         unsigned char *phys_ram_page_exist_bitmap = NULL;
 816
 817         /* allocate memory bitmap */
 818         n = BITMAP_SIZE(phys_ram_size);
 819         m = qemu_get_be32(f);
 820         if (m > n) // allocate enough space
 821             n = m;
 822         phys_ram_page_exist_bitmap = qemu_malloc(n);
 823         if (!phys_ram_page_exist_bitmap) {
 824             perror("failed to allocate page bitmap");
 825             return MIG_STAT_NO_MEM;
 826         }
 827
 828         /* receive memory bitmap */
 829         qemu_get_buffer(f, phys_ram_page_exist_bitmap, m);
 830
 831         /* FIXME: free dellocated-at-src guest memory pages */
 832
 833         qemu_free(phys_ram_page_exist_bitmap);
 834     }
 835
 836     do {
 837         addr = qemu_get_be32(f);
 838         if (addr == 1)
 839             break;
 840         if (addr >= phys_ram_size)
 841             return MIG_STAT_DST_MEM_OUT_OF_RANGE;
 842         ret = migrate_incoming_page(f, addr);
 843         if (ret)
 844             return ret;
 845     } while (1);
 846
 847
 848     qemu_aio_flush();
 849     vm_stop(0);
 850     if (qemu_live_loadvm_state(f))
 851         ret = MIG_STAT_DST_LOADVM_FAILED;
 852 #ifdef MIGRATION_VERIFY
 853     if (ret==0) ret=load_verify_memory(f, NULL, 1);
 854 #endif /* MIGRATION_VERIFY */
 855     qemu_fclose(f);
 856
 857     return ret;
 858 }
 859
 860 static int migrate_incoming_tcp(const char *host)
 861 {
 862     struct sockaddr_in addr;
 863     socklen_t addrlen = sizeof(addr);
 864     int fd, sfd;
 865     ssize_t len;
 866     uint8_t status = 0;
 867     int reuse = 1;
 868     int rc;
 869
 870     addr.sin_family = AF_INET;
 871     if (parse_host_port(&addr, host) == -1) {
 872         fprintf(stderr, "parse_host_port() failed for %s\n", host);
 873         rc = MIG_STAT_DST_INVALID_ADDR;
 874         goto error;
 875     }
 876
 877     fd = socket(PF_INET, SOCK_STREAM, 0);
 878     if (fd == -1) {
 879         perror("socket failed");
 880         rc = MIG_STAT_DST_SOCKET_FAILED;
 881         goto error;
 882     }
 883
 884     if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &reuse, sizeof(reuse)) == -1) {
 885         perror("setsockopt() failed");
 886         rc = MIG_STAT_DST_SOCKOPT_FAILED;
 887         goto error_socket;
 888     }
 889
 890     if (bind(fd, (struct sockaddr *)&addr, sizeof(addr)) == -1) {
 891         perror("bind() failed");
 892         rc = MIG_STAT_DST_BIND_FAILED;
 893         goto error_socket;
 894     }
 895
 896     if (listen(fd, 1) == -1) {
 897         perror("listen() failed");
 898         rc = MIG_STAT_DST_LISTEN_FAILED;
 899         goto error_socket;
 900     }
 901
 902 again:
 903     sfd = accept(fd, (struct sockaddr *)&addr, &addrlen);
 904     if (sfd == -1) {
 905         if (errno == EINTR)
 906             goto again;
 907         perror("accept() failed");
 908         rc = MIG_STAT_DST_ACCEPT_FAILED;
 909         goto error_socket;
 910     }
 911
 912     rc = migrate_incoming_fd(sfd);
 913     if (rc != 0) {
 914         fprintf(stderr, "migrate_incoming_fd failed (rc=%d)\n", rc);
 915         goto error_accept;
 916     }
 917
 918 send_ack:
 919     len = write(sfd, &status, 1);
 920     if (len == -1 && errno == EAGAIN)
 921         goto send_ack;
 922     if (len != 1) {
 923         fprintf(stderr, "migration: send_ack: write error len=%zu (%s)\n",
 924                 len, strerror(errno));
 925         rc = MIG_STAT_DST_WRITE_FAILED;
 926         goto error_accept;
 927     }
 928
 929     rc = wait_for_message("WAIT FOR GO", sfd, wait_for_message_timeout);
 930     if (rc) {
 931         rc += 200;
 932         goto error_accept;
 933     }
 934
 935 wait_for_go:
 936     len = read(sfd, &status, 1);
 937     if (len == -1 && errno == EAGAIN)
 938         goto wait_for_go;
 939     if (len != 1) {
 940         rc = MIG_STAT_DST_READ_FAILED;
 941         fprintf(stderr, "migration: wait_for_go: read error len=%zu (%s)\n",
 942                 len, strerror(errno));
 943     }
 944
 945 error_accept:
 946     close(sfd);
 947 error_socket:
 948     close(fd);
 949 error:
 950     return rc;
 951 }
 952
 953 int migrate_incoming(const char *device)
 954 {
 955     const char *ptr;
 956     int ret = 0;
 957
 958     if (strcmp(device, "stdio") == 0)
 959         ret = migrate_incoming_fd(STDIN_FILENO);
 960     else if (strstart(device, "file://", &ptr)) {
 961         int fd = open(ptr, O_RDONLY);
 962         if (fd < 0) {
 963             ret = MIG_STAT_DST_INVALID_PARAMS;
 964         } else {
 965             ret = migrate_incoming_fd(fd);
 966             close(fd);
 967         }
 968     } else if (strstart(device, "tcp://", &ptr)) {
 969         char *host, *end;
 970         host = strdup(ptr);
 971         end = strchr(host, '/');
 972         if (end) *end = 0;
 973         ret = migrate_incoming_tcp(host);
 974         qemu_free(host);
 975     } else {
 976         errno = EINVAL;
 977         ret = MIG_STAT_DST_INVALID_PARAMS;
 978     }
 979
 980     return ret;
 981 }
 982
 983 /* Migration monitor command */
 984
 985 /* TODO:
 986    1) audit all error paths
 987 */
 988
 989 void do_migrate(int detach, const char *uri)
 990 {
 991     const char *ptr;
 992     MigrationState *s = current_migration;
 993
 994     if (s) {
 995         term_printf("Migration already active\n");
 996         return;
 997     }
 998
 999     status = MIG_STAT_INVALID_PARAMS;
1000     if (strstart(uri, "exec:", &ptr)) {
1001         char *command = urldecode(ptr);
1002         migration_init_exec(detach, command);
1003         free(command);
1004     } else if (strstart(uri, "ssh://", &ptr)) {
1005         char *host, *end;
1006
1007         host = strdup(ptr);
1008         end = strchr(host, '/');
1009         if (end) *end = 0;
1010         migration_init_ssh(detach, host);
1011         qemu_free(host);
1012     } else if (strstart(uri, "tcp://", &ptr)) {
1013         char *host, *end;
1014
1015         host = strdup(ptr);
1016         end = strchr(host, '/');
1017         if (end) *end = 0;
1018
1019         if (migration_init_tcp(detach, host) == NULL)
1020             term_printf("migration failed (migration_init_tcp for %s failed)\n", host);
1021         free(host);
1022     } else {
1023         term_printf("Unknown migration protocol '%s'\n", uri);
1024         return;
1025     }
1026 }
1027
1028 void do_migrate_set_speed(const char *value)
1029 {
1030     double d;
1031     char *ptr;
1032
1033     d = strtod(value, &ptr);
1034     switch (*ptr) {
1035     case 'G': case 'g':
1036         d *= 1024;
1037     case 'M': case 'm':
1038         d *= 1024;
1039     case 'K': case 'k':
1040         d *= 1024;
1041     default:
1042         break;
1043     }
1044
1045     max_throttle = (uint32_t)d;
1046 }
1047
1048 void do_info_migration(void)
1049 {
1050     MigrationState *s = current_migration;
1051
1052     if (s) {
1053         term_printf("Migration active\n");
1054         if (s->bps < (1 << 20))
1055             term_printf("Transfer rate %3.1f kb/s\n",
1056                         (double)s->bps / 1024);
1057         else
1058             term_printf("Transfer rate %3.1f mb/s\n",
1059                         (double)s->bps / (1024 * 1024));
1060         term_printf("Iteration %d\n", s->iteration);
1061         term_printf("Transferred %d/%d pages\n", s->updated_pages, phys_ram_size >> TARGET_PAGE_BITS);
1062         if (s->iteration)
1063             term_printf("Last iteration found %d dirty pages\n", s->last_updated_pages);
1064     } else {
1065         term_printf("Migration inactive\n");
1066         term_printf("last migration status is %d\n", status);
1067     }
1068     term_printf("Maximum migration speed is ");
1069     if (max_throttle < (1 << 20))
1070         term_printf("%3.1f kb/s\n", (double)max_throttle / 1024);
1071     else
1072         term_printf("%3.1f mb/s\n", (double)max_throttle / (1024 * 1024));
1073 }
1074
1075 void do_migrate_cancel(void)
1076 {
1077     MigrationState *s = current_migration;
1078
1079     if (s)
1080         *s->has_error = MIG_STAT_MIGRATION_CANCEL;
1081 }
1082
1083
1084
1085 #ifdef MIGRATION_VERIFY
1086 unsigned int calc_page_checksum(target_ulong addr)
1087 {
1088     unsigned int sum=0;
1089     unsigned int *p = (unsigned int *)(phys_ram_base + addr);
1090     unsigned int *q = p + (TARGET_PAGE_SIZE / sizeof(unsigned int));
1091
1092     for ( /*initialized already */ ; p<q ; p++)
1093         sum += *p;
1094     return sum;
1095 }
1096
1097
1098 static int save_verify_memory(QEMUFile *f, void *opaque)
1099 {
1100     unsigned int addr;
1101     unsigned int sum;
1102
1103     for (addr = 0; addr < phys_ram_size; addr += TARGET_PAGE_SIZE) {
1104         if (kvm_enabled() && (addr>=0xa0000) && (addr<0xc0000)) /* do not access video-addresses */
1105             continue;
1106         sum = calc_page_checksum(addr);
1107         qemu_put_be32(f, addr);
1108         qemu_put_be32(f, sum);
1109     }
1110     return 0;
1111 }
1112
1113 static int load_verify_memory(QEMUFile *f, void *opaque, int version_id)
1114 {
1115     unsigned int addr, raddr;
1116     unsigned int sum, rsum;
1117     int num_errors = 0;
1118
1119     for (addr = 0; addr < phys_ram_size; addr += TARGET_PAGE_SIZE) {
1120         if (kvm_enabled() && (addr>=0xa0000) && (addr<0xc0000)) /* do not access video-addresses */
1121             continue;
1122         sum = calc_page_checksum(addr);
1123         raddr = qemu_get_be32(f);
1124         rsum  = qemu_get_be32(f);
1125         if ((raddr != addr) || (rsum != sum)) {
1126             term_printf("checksum mismatch: src:0x%x 0x%x , dst:0x%x 0x%x\n",
1127                     raddr, rsum, addr, sum);
1128             num_errors++;
1129         }
1130     }
1131     printf("memory_verify: num_errors=%d\n", num_errors);
1132     term_printf("memory_verify: num_errors=%d\n", num_errors);
1133     return 0/* num_errors */;
1134 }
1135 #endif /* MIGRATION_VERIFY */