1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2016 Flavius Anton 5 * Copyright (c) 2016 Mihai Tiganus 6 * Copyright (c) 2016-2019 Mihai Carabas 7 * Copyright (c) 2017-2019 Darius Mihai 8 * Copyright (c) 2017-2019 Elena Mihailescu 9 * Copyright (c) 2018-2019 Sergiu Weisz 10 * All rights reserved. 11 * The bhyve-snapshot feature was developed under sponsorships 12 * from Matthew Grooms. 13 * 14 * Redistribution and use in source and binary forms, with or without 15 * modification, are permitted provided that the following conditions 16 * are met: 17 * 1. Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 2. Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in the 21 * documentation and/or other materials provided with the distribution. 22 * 23 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include <sys/types.h> 40 #ifndef WITHOUT_CAPSICUM 41 #include <sys/capsicum.h> 42 #endif 43 #include <sys/mman.h> 44 #include <sys/socket.h> 45 #include <sys/stat.h> 46 #include <sys/time.h> 47 #include <sys/un.h> 48 49 #include <machine/atomic.h> 50 #include <machine/segments.h> 51 52 #ifndef WITHOUT_CAPSICUM 53 #include <capsicum_helpers.h> 54 #endif 55 #include <stdio.h> 56 #include <stdlib.h> 57 #include <string.h> 58 #include <err.h> 59 #include <errno.h> 60 #include <fcntl.h> 61 #include <libgen.h> 62 #include <signal.h> 63 #include <unistd.h> 64 #include <assert.h> 65 #include <errno.h> 66 #include <pthread.h> 67 #include <pthread_np.h> 68 #include <sysexits.h> 69 #include <stdbool.h> 70 #include <sys/ioctl.h> 71 72 #include <machine/vmm.h> 73 #ifndef WITHOUT_CAPSICUM 74 #include <machine/vmm_dev.h> 75 #endif 76 #include <machine/vmm_snapshot.h> 77 #include <vmmapi.h> 78 79 #include "bhyverun.h" 80 #include "acpi.h" 81 #include "atkbdc.h" 82 #include "debug.h" 83 #include "inout.h" 84 #include "ipc.h" 85 #include "fwctl.h" 86 #include "ioapic.h" 87 #include "mem.h" 88 #include "mevent.h" 89 #include "mptbl.h" 90 #include "pci_emul.h" 91 #include "pci_irq.h" 92 #include "pci_lpc.h" 93 #include "smbiostbl.h" 94 #include "snapshot.h" 95 #include "xmsr.h" 96 #include "spinup_ap.h" 97 #include "rtc.h" 98 99 #include <libxo/xo.h> 100 #include <ucl.h> 101 102 struct spinner_info { 103 const size_t *crtval; 104 const size_t maxval; 105 const size_t total; 106 }; 107 108 extern int guest_ncpus; 109 110 static struct winsize winsize; 111 static sig_t old_winch_handler; 112 113 #define KB (1024UL) 114 #define MB (1024UL * KB) 115 #define GB (1024UL * MB) 116 117 #define SNAPSHOT_CHUNK (4 * MB) 118 #define PROG_BUF_SZ (8192) 119 120 #define SNAPSHOT_BUFFER_SIZE (20 * MB) 121 122 #define JSON_STRUCT_ARR_KEY "structs" 123 #define JSON_DEV_ARR_KEY "devices" 124 #define JSON_BASIC_METADATA_KEY "basic metadata" 125 #define JSON_SNAPSHOT_REQ_KEY "snapshot_req" 126 #define JSON_SIZE_KEY "size" 127 #define JSON_FILE_OFFSET_KEY "file_offset" 128 129 #define JSON_NCPUS_KEY "ncpus" 130 #define JSON_VMNAME_KEY "vmname" 131 #define JSON_MEMSIZE_KEY "memsize" 132 #define JSON_MEMFLAGS_KEY "memflags" 133 134 #define min(a,b) \ 135 ({ \ 136 __typeof__ (a) _a = (a); \ 137 __typeof__ (b) _b = (b); \ 138 _a < _b ? _a : _b; \ 139 }) 140 141 static const struct vm_snapshot_dev_info snapshot_devs[] = { 142 { "atkbdc", atkbdc_snapshot, NULL, NULL }, 143 { "virtio-net", pci_snapshot, pci_pause, pci_resume }, 144 { "virtio-blk", pci_snapshot, pci_pause, pci_resume }, 145 { "virtio-rnd", pci_snapshot, NULL, NULL }, 146 { "lpc", pci_snapshot, NULL, NULL }, 147 { "fbuf", pci_snapshot, NULL, NULL }, 148 { "xhci", pci_snapshot, NULL, NULL }, 149 { "e1000", pci_snapshot, NULL, NULL }, 150 { "ahci", pci_snapshot, pci_pause, pci_resume }, 151 { "ahci-hd", pci_snapshot, pci_pause, pci_resume }, 152 { "ahci-cd", pci_snapshot, pci_pause, pci_resume }, 153 }; 154 155 static const struct vm_snapshot_kern_info snapshot_kern_structs[] = { 156 { "vhpet", STRUCT_VHPET }, 157 { "vm", STRUCT_VM }, 158 { "vmx", STRUCT_VMX }, 159 { "vioapic", STRUCT_VIOAPIC }, 160 { "vlapic", STRUCT_VLAPIC }, 161 { "vmcx", STRUCT_VMCX }, 162 { "vatpit", STRUCT_VATPIT }, 163 { "vatpic", STRUCT_VATPIC }, 164 { "vpmtmr", STRUCT_VPMTMR }, 165 { "vrtc", STRUCT_VRTC }, 166 }; 167 168 static cpuset_t vcpus_active, vcpus_suspended; 169 static pthread_mutex_t vcpu_lock; 170 static pthread_cond_t vcpus_idle, vcpus_can_run; 171 static bool checkpoint_active; 172 173 /* 174 * TODO: Harden this function and all of its callers since 'base_str' is a user 175 * provided string. 176 */ 177 static char * 178 strcat_extension(const char *base_str, const char *ext) 179 { 180 char *res; 181 size_t base_len, ext_len; 182 183 base_len = strnlen(base_str, NAME_MAX); 184 ext_len = strnlen(ext, NAME_MAX); 185 186 if (base_len + ext_len > NAME_MAX) { 187 fprintf(stderr, "Filename exceeds maximum length.\n"); 188 return (NULL); 189 } 190 191 res = malloc(base_len + ext_len + 1); 192 if (res == NULL) { 193 perror("Failed to allocate memory."); 194 return (NULL); 195 } 196 197 memcpy(res, base_str, base_len); 198 memcpy(res + base_len, ext, ext_len); 199 res[base_len + ext_len] = 0; 200 201 return (res); 202 } 203 204 void 205 destroy_restore_state(struct restore_state *rstate) 206 { 207 if (rstate == NULL) { 208 fprintf(stderr, "Attempting to destroy NULL restore struct.\n"); 209 return; 210 } 211 212 if (rstate->kdata_map != MAP_FAILED) 213 munmap(rstate->kdata_map, rstate->kdata_len); 214 215 if (rstate->kdata_fd > 0) 216 close(rstate->kdata_fd); 217 if (rstate->vmmem_fd > 0) 218 close(rstate->vmmem_fd); 219 220 if (rstate->meta_root_obj != NULL) 221 ucl_object_unref(rstate->meta_root_obj); 222 if (rstate->meta_parser != NULL) 223 ucl_parser_free(rstate->meta_parser); 224 } 225 226 static int 227 load_vmmem_file(const char *filename, struct restore_state *rstate) 228 { 229 struct stat sb; 230 int err; 231 232 rstate->vmmem_fd = open(filename, O_RDONLY); 233 if (rstate->vmmem_fd < 0) { 234 perror("Failed to open restore file"); 235 return (-1); 236 } 237 238 err = fstat(rstate->vmmem_fd, &sb); 239 if (err < 0) { 240 perror("Failed to stat restore file"); 241 goto err_load_vmmem; 242 } 243 244 if (sb.st_size == 0) { 245 fprintf(stderr, "Restore file is empty.\n"); 246 goto err_load_vmmem; 247 } 248 249 rstate->vmmem_len = sb.st_size; 250 251 return (0); 252 253 err_load_vmmem: 254 if (rstate->vmmem_fd > 0) 255 close(rstate->vmmem_fd); 256 return (-1); 257 } 258 259 static int 260 load_kdata_file(const char *filename, struct restore_state *rstate) 261 { 262 struct stat sb; 263 int err; 264 265 rstate->kdata_fd = open(filename, O_RDONLY); 266 if (rstate->kdata_fd < 0) { 267 perror("Failed to open kernel data file"); 268 return (-1); 269 } 270 271 err = fstat(rstate->kdata_fd, &sb); 272 if (err < 0) { 273 perror("Failed to stat kernel data file"); 274 goto err_load_kdata; 275 } 276 277 if (sb.st_size == 0) { 278 fprintf(stderr, "Kernel data file is empty.\n"); 279 goto err_load_kdata; 280 } 281 282 rstate->kdata_len = sb.st_size; 283 rstate->kdata_map = mmap(NULL, rstate->kdata_len, PROT_READ, 284 MAP_SHARED, rstate->kdata_fd, 0); 285 if (rstate->kdata_map == MAP_FAILED) { 286 perror("Failed to map restore file"); 287 goto err_load_kdata; 288 } 289 290 return (0); 291 292 err_load_kdata: 293 if (rstate->kdata_fd > 0) 294 close(rstate->kdata_fd); 295 return (-1); 296 } 297 298 static int 299 load_metadata_file(const char *filename, struct restore_state *rstate) 300 { 301 ucl_object_t *obj; 302 struct ucl_parser *parser; 303 int err; 304 305 parser = ucl_parser_new(UCL_PARSER_DEFAULT); 306 if (parser == NULL) { 307 fprintf(stderr, "Failed to initialize UCL parser.\n"); 308 err = -1; 309 goto err_load_metadata; 310 } 311 312 err = ucl_parser_add_file(parser, filename); 313 if (err == 0) { 314 fprintf(stderr, "Failed to parse metadata file: '%s'\n", 315 filename); 316 err = -1; 317 goto err_load_metadata; 318 } 319 320 obj = ucl_parser_get_object(parser); 321 if (obj == NULL) { 322 fprintf(stderr, "Failed to parse object.\n"); 323 err = -1; 324 goto err_load_metadata; 325 } 326 327 rstate->meta_parser = parser; 328 rstate->meta_root_obj = (ucl_object_t *)obj; 329 330 return (0); 331 332 err_load_metadata: 333 if (parser != NULL) 334 ucl_parser_free(parser); 335 return (err); 336 } 337 338 int 339 load_restore_file(const char *filename, struct restore_state *rstate) 340 { 341 int err = 0; 342 char *kdata_filename = NULL, *meta_filename = NULL; 343 344 assert(filename != NULL); 345 assert(rstate != NULL); 346 347 memset(rstate, 0, sizeof(*rstate)); 348 rstate->kdata_map = MAP_FAILED; 349 350 err = load_vmmem_file(filename, rstate); 351 if (err != 0) { 352 fprintf(stderr, "Failed to load guest RAM file.\n"); 353 goto err_restore; 354 } 355 356 kdata_filename = strcat_extension(filename, ".kern"); 357 if (kdata_filename == NULL) { 358 fprintf(stderr, "Failed to construct kernel data filename.\n"); 359 goto err_restore; 360 } 361 362 err = load_kdata_file(kdata_filename, rstate); 363 if (err != 0) { 364 fprintf(stderr, "Failed to load guest kernel data file.\n"); 365 goto err_restore; 366 } 367 368 meta_filename = strcat_extension(filename, ".meta"); 369 if (meta_filename == NULL) { 370 fprintf(stderr, "Failed to construct kernel metadata filename.\n"); 371 goto err_restore; 372 } 373 374 err = load_metadata_file(meta_filename, rstate); 375 if (err != 0) { 376 fprintf(stderr, "Failed to load guest metadata file.\n"); 377 goto err_restore; 378 } 379 380 return (0); 381 382 err_restore: 383 destroy_restore_state(rstate); 384 if (kdata_filename != NULL) 385 free(kdata_filename); 386 if (meta_filename != NULL) 387 free(meta_filename); 388 return (-1); 389 } 390 391 #define JSON_GET_INT_OR_RETURN(key, obj, result_ptr, ret) \ 392 do { \ 393 const ucl_object_t *obj__; \ 394 obj__ = ucl_object_lookup(obj, key); \ 395 if (obj__ == NULL) { \ 396 fprintf(stderr, "Missing key: '%s'", key); \ 397 return (ret); \ 398 } \ 399 if (!ucl_object_toint_safe(obj__, result_ptr)) { \ 400 fprintf(stderr, "Cannot convert '%s' value to int.", key); \ 401 return (ret); \ 402 } \ 403 } while(0) 404 405 #define JSON_GET_STRING_OR_RETURN(key, obj, result_ptr, ret) \ 406 do { \ 407 const ucl_object_t *obj__; \ 408 obj__ = ucl_object_lookup(obj, key); \ 409 if (obj__ == NULL) { \ 410 fprintf(stderr, "Missing key: '%s'", key); \ 411 return (ret); \ 412 } \ 413 if (!ucl_object_tostring_safe(obj__, result_ptr)) { \ 414 fprintf(stderr, "Cannot convert '%s' value to string.", key); \ 415 return (ret); \ 416 } \ 417 } while(0) 418 419 static void * 420 lookup_struct(enum snapshot_req struct_id, struct restore_state *rstate, 421 size_t *struct_size) 422 { 423 const ucl_object_t *structs = NULL, *obj = NULL; 424 ucl_object_iter_t it = NULL; 425 int64_t snapshot_req, size, file_offset; 426 427 structs = ucl_object_lookup(rstate->meta_root_obj, JSON_STRUCT_ARR_KEY); 428 if (structs == NULL) { 429 fprintf(stderr, "Failed to find '%s' object.\n", 430 JSON_STRUCT_ARR_KEY); 431 return (NULL); 432 } 433 434 if (ucl_object_type(structs) != UCL_ARRAY) { 435 fprintf(stderr, "Object '%s' is not an array.\n", 436 JSON_STRUCT_ARR_KEY); 437 return (NULL); 438 } 439 440 while ((obj = ucl_object_iterate(structs, &it, true)) != NULL) { 441 snapshot_req = -1; 442 JSON_GET_INT_OR_RETURN(JSON_SNAPSHOT_REQ_KEY, obj, 443 &snapshot_req, NULL); 444 assert(snapshot_req >= 0); 445 if ((enum snapshot_req) snapshot_req == struct_id) { 446 JSON_GET_INT_OR_RETURN(JSON_SIZE_KEY, obj, 447 &size, NULL); 448 assert(size >= 0); 449 450 JSON_GET_INT_OR_RETURN(JSON_FILE_OFFSET_KEY, obj, 451 &file_offset, NULL); 452 assert(file_offset >= 0); 453 assert((uint64_t)file_offset + size <= 454 rstate->kdata_len); 455 456 *struct_size = (size_t)size; 457 return ((uint8_t *)rstate->kdata_map + file_offset); 458 } 459 } 460 461 return (NULL); 462 } 463 464 static void * 465 lookup_check_dev(const char *dev_name, struct restore_state *rstate, 466 const ucl_object_t *obj, size_t *data_size) 467 { 468 const char *snapshot_req; 469 int64_t size, file_offset; 470 471 snapshot_req = NULL; 472 JSON_GET_STRING_OR_RETURN(JSON_SNAPSHOT_REQ_KEY, obj, 473 &snapshot_req, NULL); 474 assert(snapshot_req != NULL); 475 if (!strcmp(snapshot_req, dev_name)) { 476 JSON_GET_INT_OR_RETURN(JSON_SIZE_KEY, obj, 477 &size, NULL); 478 assert(size >= 0); 479 480 JSON_GET_INT_OR_RETURN(JSON_FILE_OFFSET_KEY, obj, 481 &file_offset, NULL); 482 assert(file_offset >= 0); 483 assert((uint64_t)file_offset + size <= rstate->kdata_len); 484 485 *data_size = (size_t)size; 486 return ((uint8_t *)rstate->kdata_map + file_offset); 487 } 488 489 return (NULL); 490 } 491 492 static void* 493 lookup_dev(const char *dev_name, struct restore_state *rstate, 494 size_t *data_size) 495 { 496 const ucl_object_t *devs = NULL, *obj = NULL; 497 ucl_object_iter_t it = NULL; 498 void *ret; 499 500 devs = ucl_object_lookup(rstate->meta_root_obj, JSON_DEV_ARR_KEY); 501 if (devs == NULL) { 502 fprintf(stderr, "Failed to find '%s' object.\n", 503 JSON_DEV_ARR_KEY); 504 return (NULL); 505 } 506 507 if (ucl_object_type(devs) != UCL_ARRAY) { 508 fprintf(stderr, "Object '%s' is not an array.\n", 509 JSON_DEV_ARR_KEY); 510 return (NULL); 511 } 512 513 while ((obj = ucl_object_iterate(devs, &it, true)) != NULL) { 514 ret = lookup_check_dev(dev_name, rstate, obj, data_size); 515 if (ret != NULL) 516 return (ret); 517 } 518 519 return (NULL); 520 } 521 522 static const ucl_object_t * 523 lookup_basic_metadata_object(struct restore_state *rstate) 524 { 525 const ucl_object_t *basic_meta_obj = NULL; 526 527 basic_meta_obj = ucl_object_lookup(rstate->meta_root_obj, 528 JSON_BASIC_METADATA_KEY); 529 if (basic_meta_obj == NULL) { 530 fprintf(stderr, "Failed to find '%s' object.\n", 531 JSON_BASIC_METADATA_KEY); 532 return (NULL); 533 } 534 535 if (ucl_object_type(basic_meta_obj) != UCL_OBJECT) { 536 fprintf(stderr, "Object '%s' is not a JSON object.\n", 537 JSON_BASIC_METADATA_KEY); 538 return (NULL); 539 } 540 541 return (basic_meta_obj); 542 } 543 544 const char * 545 lookup_vmname(struct restore_state *rstate) 546 { 547 const char *vmname; 548 const ucl_object_t *obj; 549 550 obj = lookup_basic_metadata_object(rstate); 551 if (obj == NULL) 552 return (NULL); 553 554 JSON_GET_STRING_OR_RETURN(JSON_VMNAME_KEY, obj, &vmname, NULL); 555 return (vmname); 556 } 557 558 int 559 lookup_memflags(struct restore_state *rstate) 560 { 561 int64_t memflags; 562 const ucl_object_t *obj; 563 564 obj = lookup_basic_metadata_object(rstate); 565 if (obj == NULL) 566 return (0); 567 568 JSON_GET_INT_OR_RETURN(JSON_MEMFLAGS_KEY, obj, &memflags, 0); 569 570 return ((int)memflags); 571 } 572 573 size_t 574 lookup_memsize(struct restore_state *rstate) 575 { 576 int64_t memsize; 577 const ucl_object_t *obj; 578 579 obj = lookup_basic_metadata_object(rstate); 580 if (obj == NULL) 581 return (0); 582 583 JSON_GET_INT_OR_RETURN(JSON_MEMSIZE_KEY, obj, &memsize, 0); 584 if (memsize < 0) 585 memsize = 0; 586 587 return ((size_t)memsize); 588 } 589 590 591 int 592 lookup_guest_ncpus(struct restore_state *rstate) 593 { 594 int64_t ncpus; 595 const ucl_object_t *obj; 596 597 obj = lookup_basic_metadata_object(rstate); 598 if (obj == NULL) 599 return (0); 600 601 JSON_GET_INT_OR_RETURN(JSON_NCPUS_KEY, obj, &ncpus, 0); 602 return ((int)ncpus); 603 } 604 605 static void 606 winch_handler(int signal __unused) 607 { 608 #ifdef TIOCGWINSZ 609 ioctl(STDOUT_FILENO, TIOCGWINSZ, &winsize); 610 #endif /* TIOCGWINSZ */ 611 } 612 613 static int 614 print_progress(size_t crtval, const size_t maxval) 615 { 616 size_t rc; 617 double crtval_gb, maxval_gb; 618 size_t i, win_width, prog_start, prog_done, prog_end; 619 int mval_len; 620 621 static char prog_buf[PROG_BUF_SZ]; 622 static const size_t len = sizeof(prog_buf); 623 624 static size_t div; 625 static const char *div_str; 626 627 static char wip_bar[] = { '/', '-', '\\', '|' }; 628 static int wip_idx = 0; 629 630 if (maxval == 0) { 631 printf("[0B / 0B]\r\n"); 632 return (0); 633 } 634 635 if (crtval > maxval) 636 crtval = maxval; 637 638 if (maxval > 10 * GB) { 639 div = GB; 640 div_str = "GiB"; 641 } else if (maxval > 10 * MB) { 642 div = MB; 643 div_str = "MiB"; 644 } else { 645 div = KB; 646 div_str = "KiB"; 647 } 648 649 crtval_gb = (double) crtval / div; 650 maxval_gb = (double) maxval / div; 651 652 rc = snprintf(prog_buf, len, "%.03lf", maxval_gb); 653 if (rc == len) { 654 fprintf(stderr, "Maxval too big\n"); 655 return (-1); 656 } 657 mval_len = rc; 658 659 rc = snprintf(prog_buf, len, "\r[%*.03lf%s / %.03lf%s] |", 660 mval_len, crtval_gb, div_str, maxval_gb, div_str); 661 662 if (rc == len) { 663 fprintf(stderr, "Buffer too small to print progress\n"); 664 return (-1); 665 } 666 667 win_width = min(winsize.ws_col, len); 668 prog_start = rc; 669 670 if (prog_start < (win_width - 2)) { 671 prog_end = win_width - prog_start - 2; 672 prog_done = prog_end * (crtval_gb / maxval_gb); 673 674 for (i = prog_start; i < prog_start + prog_done; i++) 675 prog_buf[i] = '#'; 676 677 if (crtval != maxval) { 678 prog_buf[i] = wip_bar[wip_idx]; 679 wip_idx = (wip_idx + 1) % sizeof(wip_bar); 680 i++; 681 } else { 682 prog_buf[i++] = '#'; 683 } 684 685 for (; i < win_width - 2; i++) 686 prog_buf[i] = '_'; 687 688 prog_buf[win_width - 2] = '|'; 689 } 690 691 prog_buf[win_width - 1] = '\0'; 692 write(STDOUT_FILENO, prog_buf, win_width); 693 694 return (0); 695 } 696 697 static void * 698 snapshot_spinner_cb(void *arg) 699 { 700 int rc; 701 size_t crtval, maxval, total; 702 struct spinner_info *si; 703 struct timespec ts; 704 705 si = arg; 706 if (si == NULL) 707 pthread_exit(NULL); 708 709 ts.tv_sec = 0; 710 ts.tv_nsec = 50 * 1000 * 1000; /* 50 ms sleep time */ 711 712 do { 713 crtval = *si->crtval; 714 maxval = si->maxval; 715 total = si->total; 716 717 rc = print_progress(crtval, total); 718 if (rc < 0) { 719 fprintf(stderr, "Failed to parse progress\n"); 720 break; 721 } 722 723 nanosleep(&ts, NULL); 724 } while (crtval < maxval); 725 726 pthread_exit(NULL); 727 return NULL; 728 } 729 730 static int 731 vm_snapshot_mem_part(const int snapfd, const size_t foff, void *src, 732 const size_t len, const size_t totalmem, const bool op_wr) 733 { 734 int rc; 735 size_t part_done, todo, rem; 736 ssize_t done; 737 bool show_progress; 738 pthread_t spinner_th; 739 struct spinner_info *si; 740 741 if (lseek(snapfd, foff, SEEK_SET) < 0) { 742 perror("Failed to change file offset"); 743 return (-1); 744 } 745 746 show_progress = false; 747 if (isatty(STDIN_FILENO) && (winsize.ws_col != 0)) 748 show_progress = true; 749 750 part_done = foff; 751 rem = len; 752 753 if (show_progress) { 754 si = &(struct spinner_info) { 755 .crtval = &part_done, 756 .maxval = foff + len, 757 .total = totalmem 758 }; 759 760 rc = pthread_create(&spinner_th, 0, snapshot_spinner_cb, si); 761 if (rc) { 762 perror("Unable to create spinner thread"); 763 show_progress = false; 764 } 765 } 766 767 while (rem > 0) { 768 if (show_progress) 769 todo = min(SNAPSHOT_CHUNK, rem); 770 else 771 todo = rem; 772 773 if (op_wr) 774 done = write(snapfd, src, todo); 775 else 776 done = read(snapfd, src, todo); 777 if (done < 0) { 778 perror("Failed to write in file"); 779 return (-1); 780 } 781 782 src = (uint8_t *)src + done; 783 part_done += done; 784 rem -= done; 785 } 786 787 if (show_progress) { 788 rc = pthread_join(spinner_th, NULL); 789 if (rc) 790 perror("Unable to end spinner thread"); 791 } 792 793 return (0); 794 } 795 796 static size_t 797 vm_snapshot_mem(struct vmctx *ctx, int snapfd, size_t memsz, const bool op_wr) 798 { 799 int ret; 800 size_t lowmem, highmem, totalmem; 801 char *baseaddr; 802 803 ret = vm_get_guestmem_from_ctx(ctx, &baseaddr, &lowmem, &highmem); 804 if (ret) { 805 fprintf(stderr, "%s: unable to retrieve guest memory size\r\n", 806 __func__); 807 return (0); 808 } 809 totalmem = lowmem + highmem; 810 811 if ((op_wr == false) && (totalmem != memsz)) { 812 fprintf(stderr, "%s: mem size mismatch: %ld vs %ld\r\n", 813 __func__, totalmem, memsz); 814 return (0); 815 } 816 817 winsize.ws_col = 80; 818 #ifdef TIOCGWINSZ 819 ioctl(STDOUT_FILENO, TIOCGWINSZ, &winsize); 820 #endif /* TIOCGWINSZ */ 821 old_winch_handler = signal(SIGWINCH, winch_handler); 822 823 ret = vm_snapshot_mem_part(snapfd, 0, baseaddr, lowmem, 824 totalmem, op_wr); 825 if (ret) { 826 fprintf(stderr, "%s: Could not %s lowmem\r\n", 827 __func__, op_wr ? "write" : "read"); 828 totalmem = 0; 829 goto done; 830 } 831 832 if (highmem == 0) 833 goto done; 834 835 ret = vm_snapshot_mem_part(snapfd, lowmem, baseaddr + 4*GB, 836 highmem, totalmem, op_wr); 837 if (ret) { 838 fprintf(stderr, "%s: Could not %s highmem\r\n", 839 __func__, op_wr ? "write" : "read"); 840 totalmem = 0; 841 goto done; 842 } 843 844 done: 845 printf("\r\n"); 846 signal(SIGWINCH, old_winch_handler); 847 848 return (totalmem); 849 } 850 851 int 852 restore_vm_mem(struct vmctx *ctx, struct restore_state *rstate) 853 { 854 size_t restored; 855 856 restored = vm_snapshot_mem(ctx, rstate->vmmem_fd, rstate->vmmem_len, 857 false); 858 859 if (restored != rstate->vmmem_len) 860 return (-1); 861 862 return (0); 863 } 864 865 static int 866 vm_restore_kern_struct(struct vmctx *ctx, struct restore_state *rstate, 867 const struct vm_snapshot_kern_info *info) 868 { 869 void *struct_ptr; 870 size_t struct_size; 871 int ret; 872 struct vm_snapshot_meta *meta; 873 874 struct_ptr = lookup_struct(info->req, rstate, &struct_size); 875 if (struct_ptr == NULL) { 876 fprintf(stderr, "%s: Failed to lookup struct %s\r\n", 877 __func__, info->struct_name); 878 ret = -1; 879 goto done; 880 } 881 882 if (struct_size == 0) { 883 fprintf(stderr, "%s: Kernel struct size was 0 for: %s\r\n", 884 __func__, info->struct_name); 885 ret = -1; 886 goto done; 887 } 888 889 meta = &(struct vm_snapshot_meta) { 890 .ctx = ctx, 891 .dev_name = info->struct_name, 892 .dev_req = info->req, 893 894 .buffer.buf_start = struct_ptr, 895 .buffer.buf_size = struct_size, 896 897 .buffer.buf = struct_ptr, 898 .buffer.buf_rem = struct_size, 899 900 .op = VM_SNAPSHOT_RESTORE, 901 }; 902 903 ret = vm_snapshot_req(meta); 904 if (ret != 0) { 905 fprintf(stderr, "%s: Failed to restore struct: %s\r\n", 906 __func__, info->struct_name); 907 goto done; 908 } 909 910 done: 911 return (ret); 912 } 913 914 int 915 vm_restore_kern_structs(struct vmctx *ctx, struct restore_state *rstate) 916 { 917 size_t i; 918 int ret; 919 920 for (i = 0; i < nitems(snapshot_kern_structs); i++) { 921 ret = vm_restore_kern_struct(ctx, rstate, 922 &snapshot_kern_structs[i]); 923 if (ret != 0) 924 return (ret); 925 } 926 927 return (0); 928 } 929 930 static int 931 vm_restore_user_dev(struct vmctx *ctx, struct restore_state *rstate, 932 const struct vm_snapshot_dev_info *info) 933 { 934 void *dev_ptr; 935 size_t dev_size; 936 int ret; 937 struct vm_snapshot_meta *meta; 938 939 dev_ptr = lookup_dev(info->dev_name, rstate, &dev_size); 940 if (dev_ptr == NULL) { 941 fprintf(stderr, "Failed to lookup dev: %s\r\n", info->dev_name); 942 fprintf(stderr, "Continuing the restore/migration process\r\n"); 943 return (0); 944 } 945 946 if (dev_size == 0) { 947 fprintf(stderr, "%s: Device size is 0. " 948 "Assuming %s is not used\r\n", 949 __func__, info->dev_name); 950 return (0); 951 } 952 953 meta = &(struct vm_snapshot_meta) { 954 .ctx = ctx, 955 .dev_name = info->dev_name, 956 957 .buffer.buf_start = dev_ptr, 958 .buffer.buf_size = dev_size, 959 960 .buffer.buf = dev_ptr, 961 .buffer.buf_rem = dev_size, 962 963 .op = VM_SNAPSHOT_RESTORE, 964 }; 965 966 ret = (*info->snapshot_cb)(meta); 967 if (ret != 0) { 968 fprintf(stderr, "Failed to restore dev: %s\r\n", 969 info->dev_name); 970 return (-1); 971 } 972 973 return (0); 974 } 975 976 977 int 978 vm_restore_user_devs(struct vmctx *ctx, struct restore_state *rstate) 979 { 980 size_t i; 981 int ret; 982 983 for (i = 0; i < nitems(snapshot_devs); i++) { 984 ret = vm_restore_user_dev(ctx, rstate, &snapshot_devs[i]); 985 if (ret != 0) 986 return (ret); 987 } 988 989 return 0; 990 } 991 992 int 993 vm_pause_user_devs(void) 994 { 995 const struct vm_snapshot_dev_info *info; 996 size_t i; 997 int ret; 998 999 for (i = 0; i < nitems(snapshot_devs); i++) { 1000 info = &snapshot_devs[i]; 1001 if (info->pause_cb == NULL) 1002 continue; 1003 1004 ret = info->pause_cb(info->dev_name); 1005 if (ret != 0) 1006 return (ret); 1007 } 1008 1009 return (0); 1010 } 1011 1012 int 1013 vm_resume_user_devs(void) 1014 { 1015 const struct vm_snapshot_dev_info *info; 1016 size_t i; 1017 int ret; 1018 1019 for (i = 0; i < nitems(snapshot_devs); i++) { 1020 info = &snapshot_devs[i]; 1021 if (info->resume_cb == NULL) 1022 continue; 1023 1024 ret = info->resume_cb(info->dev_name); 1025 if (ret != 0) 1026 return (ret); 1027 } 1028 1029 return (0); 1030 } 1031 1032 static int 1033 vm_snapshot_kern_struct(int data_fd, xo_handle_t *xop, const char *array_key, 1034 struct vm_snapshot_meta *meta, off_t *offset) 1035 { 1036 int ret; 1037 size_t data_size; 1038 ssize_t write_cnt; 1039 1040 ret = vm_snapshot_req(meta); 1041 if (ret != 0) { 1042 fprintf(stderr, "%s: Failed to snapshot struct %s\r\n", 1043 __func__, meta->dev_name); 1044 ret = -1; 1045 goto done; 1046 } 1047 1048 data_size = vm_get_snapshot_size(meta); 1049 1050 /* XXX-MJ no handling for short writes. */ 1051 write_cnt = write(data_fd, meta->buffer.buf_start, data_size); 1052 if (write_cnt < 0 || (size_t)write_cnt != data_size) { 1053 perror("Failed to write all snapshotted data."); 1054 ret = -1; 1055 goto done; 1056 } 1057 1058 /* Write metadata. */ 1059 xo_open_instance_h(xop, array_key); 1060 xo_emit_h(xop, "{:debug_name/%s}\n", meta->dev_name); 1061 xo_emit_h(xop, "{:" JSON_SNAPSHOT_REQ_KEY "/%d}\n", 1062 meta->dev_req); 1063 xo_emit_h(xop, "{:" JSON_SIZE_KEY "/%lu}\n", data_size); 1064 xo_emit_h(xop, "{:" JSON_FILE_OFFSET_KEY "/%lu}\n", *offset); 1065 xo_close_instance_h(xop, JSON_STRUCT_ARR_KEY); 1066 1067 *offset += data_size; 1068 1069 done: 1070 return (ret); 1071 } 1072 1073 static int 1074 vm_snapshot_kern_structs(struct vmctx *ctx, int data_fd, xo_handle_t *xop) 1075 { 1076 int ret, error; 1077 size_t buf_size, i, offset; 1078 char *buffer; 1079 struct vm_snapshot_meta *meta; 1080 1081 error = 0; 1082 offset = 0; 1083 buf_size = SNAPSHOT_BUFFER_SIZE; 1084 1085 buffer = malloc(SNAPSHOT_BUFFER_SIZE * sizeof(char)); 1086 if (buffer == NULL) { 1087 error = ENOMEM; 1088 perror("Failed to allocate memory for snapshot buffer"); 1089 goto err_vm_snapshot_kern_data; 1090 } 1091 1092 meta = &(struct vm_snapshot_meta) { 1093 .ctx = ctx, 1094 1095 .buffer.buf_start = buffer, 1096 .buffer.buf_size = buf_size, 1097 1098 .op = VM_SNAPSHOT_SAVE, 1099 }; 1100 1101 xo_open_list_h(xop, JSON_STRUCT_ARR_KEY); 1102 for (i = 0; i < nitems(snapshot_kern_structs); i++) { 1103 meta->dev_name = snapshot_kern_structs[i].struct_name; 1104 meta->dev_req = snapshot_kern_structs[i].req; 1105 1106 memset(meta->buffer.buf_start, 0, meta->buffer.buf_size); 1107 meta->buffer.buf = meta->buffer.buf_start; 1108 meta->buffer.buf_rem = meta->buffer.buf_size; 1109 1110 ret = vm_snapshot_kern_struct(data_fd, xop, JSON_DEV_ARR_KEY, 1111 meta, &offset); 1112 if (ret != 0) { 1113 error = -1; 1114 goto err_vm_snapshot_kern_data; 1115 } 1116 } 1117 xo_close_list_h(xop, JSON_STRUCT_ARR_KEY); 1118 1119 err_vm_snapshot_kern_data: 1120 if (buffer != NULL) 1121 free(buffer); 1122 return (error); 1123 } 1124 1125 static int 1126 vm_snapshot_basic_metadata(struct vmctx *ctx, xo_handle_t *xop, size_t memsz) 1127 { 1128 1129 xo_open_container_h(xop, JSON_BASIC_METADATA_KEY); 1130 xo_emit_h(xop, "{:" JSON_NCPUS_KEY "/%ld}\n", guest_ncpus); 1131 xo_emit_h(xop, "{:" JSON_VMNAME_KEY "/%s}\n", vm_get_name(ctx)); 1132 xo_emit_h(xop, "{:" JSON_MEMSIZE_KEY "/%lu}\n", memsz); 1133 xo_emit_h(xop, "{:" JSON_MEMFLAGS_KEY "/%d}\n", vm_get_memflags(ctx)); 1134 xo_close_container_h(xop, JSON_BASIC_METADATA_KEY); 1135 1136 return (0); 1137 } 1138 1139 static int 1140 vm_snapshot_dev_write_data(int data_fd, xo_handle_t *xop, const char *array_key, 1141 struct vm_snapshot_meta *meta, off_t *offset) 1142 { 1143 ssize_t ret; 1144 size_t data_size; 1145 1146 data_size = vm_get_snapshot_size(meta); 1147 1148 /* XXX-MJ no handling for short writes. */ 1149 ret = write(data_fd, meta->buffer.buf_start, data_size); 1150 if (ret < 0 || (size_t)ret != data_size) { 1151 perror("Failed to write all snapshotted data."); 1152 return (-1); 1153 } 1154 1155 /* Write metadata. */ 1156 xo_open_instance_h(xop, array_key); 1157 xo_emit_h(xop, "{:" JSON_SNAPSHOT_REQ_KEY "/%s}\n", meta->dev_name); 1158 xo_emit_h(xop, "{:" JSON_SIZE_KEY "/%lu}\n", data_size); 1159 xo_emit_h(xop, "{:" JSON_FILE_OFFSET_KEY "/%lu}\n", *offset); 1160 xo_close_instance_h(xop, array_key); 1161 1162 *offset += data_size; 1163 1164 return (0); 1165 } 1166 1167 static int 1168 vm_snapshot_user_dev(const struct vm_snapshot_dev_info *info, 1169 int data_fd, xo_handle_t *xop, 1170 struct vm_snapshot_meta *meta, off_t *offset) 1171 { 1172 int ret; 1173 1174 ret = (*info->snapshot_cb)(meta); 1175 if (ret != 0) { 1176 fprintf(stderr, "Failed to snapshot %s; ret=%d\r\n", 1177 meta->dev_name, ret); 1178 return (ret); 1179 } 1180 1181 ret = vm_snapshot_dev_write_data(data_fd, xop, JSON_DEV_ARR_KEY, meta, 1182 offset); 1183 if (ret != 0) 1184 return (ret); 1185 1186 return (0); 1187 } 1188 1189 static int 1190 vm_snapshot_user_devs(struct vmctx *ctx, int data_fd, xo_handle_t *xop) 1191 { 1192 int ret; 1193 off_t offset; 1194 void *buffer; 1195 size_t buf_size, i; 1196 struct vm_snapshot_meta *meta; 1197 1198 buf_size = SNAPSHOT_BUFFER_SIZE; 1199 1200 offset = lseek(data_fd, 0, SEEK_CUR); 1201 if (offset < 0) { 1202 perror("Failed to get data file current offset."); 1203 return (-1); 1204 } 1205 1206 buffer = malloc(buf_size); 1207 if (buffer == NULL) { 1208 perror("Failed to allocate memory for snapshot buffer"); 1209 ret = ENOSPC; 1210 goto snapshot_err; 1211 } 1212 1213 meta = &(struct vm_snapshot_meta) { 1214 .ctx = ctx, 1215 1216 .buffer.buf_start = buffer, 1217 .buffer.buf_size = buf_size, 1218 1219 .op = VM_SNAPSHOT_SAVE, 1220 }; 1221 1222 xo_open_list_h(xop, JSON_DEV_ARR_KEY); 1223 1224 /* Restore other devices that support this feature */ 1225 for (i = 0; i < nitems(snapshot_devs); i++) { 1226 meta->dev_name = snapshot_devs[i].dev_name; 1227 1228 memset(meta->buffer.buf_start, 0, meta->buffer.buf_size); 1229 meta->buffer.buf = meta->buffer.buf_start; 1230 meta->buffer.buf_rem = meta->buffer.buf_size; 1231 1232 ret = vm_snapshot_user_dev(&snapshot_devs[i], data_fd, xop, 1233 meta, &offset); 1234 if (ret != 0) 1235 goto snapshot_err; 1236 } 1237 1238 xo_close_list_h(xop, JSON_DEV_ARR_KEY); 1239 1240 snapshot_err: 1241 if (buffer != NULL) 1242 free(buffer); 1243 return (ret); 1244 } 1245 1246 void 1247 checkpoint_cpu_add(int vcpu) 1248 { 1249 1250 pthread_mutex_lock(&vcpu_lock); 1251 CPU_SET(vcpu, &vcpus_active); 1252 1253 if (checkpoint_active) { 1254 CPU_SET(vcpu, &vcpus_suspended); 1255 while (checkpoint_active) 1256 pthread_cond_wait(&vcpus_can_run, &vcpu_lock); 1257 CPU_CLR(vcpu, &vcpus_suspended); 1258 } 1259 pthread_mutex_unlock(&vcpu_lock); 1260 } 1261 1262 /* 1263 * When a vCPU is suspended for any reason, it calls 1264 * checkpoint_cpu_suspend(). This records that the vCPU is idle. 1265 * Before returning from suspension, checkpoint_cpu_resume() is 1266 * called. In suspend we note that the vCPU is idle. In resume we 1267 * pause the vCPU thread until the checkpoint is complete. The reason 1268 * for the two-step process is that vCPUs might already be stopped in 1269 * the debug server when a checkpoint is requested. This approach 1270 * allows us to account for and handle those vCPUs. 1271 */ 1272 void 1273 checkpoint_cpu_suspend(int vcpu) 1274 { 1275 1276 pthread_mutex_lock(&vcpu_lock); 1277 CPU_SET(vcpu, &vcpus_suspended); 1278 if (checkpoint_active && CPU_CMP(&vcpus_active, &vcpus_suspended) == 0) 1279 pthread_cond_signal(&vcpus_idle); 1280 pthread_mutex_unlock(&vcpu_lock); 1281 } 1282 1283 void 1284 checkpoint_cpu_resume(int vcpu) 1285 { 1286 1287 pthread_mutex_lock(&vcpu_lock); 1288 while (checkpoint_active) 1289 pthread_cond_wait(&vcpus_can_run, &vcpu_lock); 1290 CPU_CLR(vcpu, &vcpus_suspended); 1291 pthread_mutex_unlock(&vcpu_lock); 1292 } 1293 1294 static void 1295 vm_vcpu_pause(struct vmctx *ctx) 1296 { 1297 1298 pthread_mutex_lock(&vcpu_lock); 1299 checkpoint_active = true; 1300 vm_suspend_cpu(ctx, -1); 1301 while (CPU_CMP(&vcpus_active, &vcpus_suspended) != 0) 1302 pthread_cond_wait(&vcpus_idle, &vcpu_lock); 1303 pthread_mutex_unlock(&vcpu_lock); 1304 } 1305 1306 static void 1307 vm_vcpu_resume(struct vmctx *ctx) 1308 { 1309 1310 pthread_mutex_lock(&vcpu_lock); 1311 checkpoint_active = false; 1312 pthread_mutex_unlock(&vcpu_lock); 1313 vm_resume_cpu(ctx, -1); 1314 pthread_cond_broadcast(&vcpus_can_run); 1315 } 1316 1317 static int 1318 vm_checkpoint(struct vmctx *ctx, const char *checkpoint_file, bool stop_vm) 1319 { 1320 int fd_checkpoint = 0, kdata_fd = 0; 1321 int ret = 0; 1322 int error = 0; 1323 size_t memsz; 1324 xo_handle_t *xop = NULL; 1325 char *meta_filename = NULL; 1326 char *kdata_filename = NULL; 1327 FILE *meta_file = NULL; 1328 1329 kdata_filename = strcat_extension(checkpoint_file, ".kern"); 1330 if (kdata_filename == NULL) { 1331 fprintf(stderr, "Failed to construct kernel data filename.\n"); 1332 return (-1); 1333 } 1334 1335 kdata_fd = open(kdata_filename, O_WRONLY | O_CREAT | O_TRUNC, 0700); 1336 if (kdata_fd < 0) { 1337 perror("Failed to open kernel data snapshot file."); 1338 error = -1; 1339 goto done; 1340 } 1341 1342 fd_checkpoint = open(checkpoint_file, O_RDWR | O_CREAT | O_TRUNC, 0700); 1343 1344 if (fd_checkpoint < 0) { 1345 perror("Failed to create checkpoint file"); 1346 error = -1; 1347 goto done; 1348 } 1349 1350 meta_filename = strcat_extension(checkpoint_file, ".meta"); 1351 if (meta_filename == NULL) { 1352 fprintf(stderr, "Failed to construct vm metadata filename.\n"); 1353 goto done; 1354 } 1355 1356 meta_file = fopen(meta_filename, "w"); 1357 if (meta_file == NULL) { 1358 perror("Failed to open vm metadata snapshot file."); 1359 goto done; 1360 } 1361 1362 xop = xo_create_to_file(meta_file, XO_STYLE_JSON, XOF_PRETTY); 1363 if (xop == NULL) { 1364 perror("Failed to get libxo handle on metadata file."); 1365 goto done; 1366 } 1367 1368 vm_vcpu_pause(ctx); 1369 1370 ret = vm_pause_user_devs(); 1371 if (ret != 0) { 1372 fprintf(stderr, "Could not pause devices\r\n"); 1373 error = ret; 1374 goto done; 1375 } 1376 1377 memsz = vm_snapshot_mem(ctx, fd_checkpoint, 0, true); 1378 if (memsz == 0) { 1379 perror("Could not write guest memory to file"); 1380 error = -1; 1381 goto done; 1382 } 1383 1384 ret = vm_snapshot_basic_metadata(ctx, xop, memsz); 1385 if (ret != 0) { 1386 fprintf(stderr, "Failed to snapshot vm basic metadata.\n"); 1387 error = -1; 1388 goto done; 1389 } 1390 1391 1392 ret = vm_snapshot_kern_structs(ctx, kdata_fd, xop); 1393 if (ret != 0) { 1394 fprintf(stderr, "Failed to snapshot vm kernel data.\n"); 1395 error = -1; 1396 goto done; 1397 } 1398 1399 ret = vm_snapshot_user_devs(ctx, kdata_fd, xop); 1400 if (ret != 0) { 1401 fprintf(stderr, "Failed to snapshot device state.\n"); 1402 error = -1; 1403 goto done; 1404 } 1405 1406 xo_finish_h(xop); 1407 1408 if (stop_vm) { 1409 vm_destroy(ctx); 1410 exit(0); 1411 } 1412 1413 done: 1414 ret = vm_resume_user_devs(); 1415 if (ret != 0) 1416 fprintf(stderr, "Could not resume devices\r\n"); 1417 vm_vcpu_resume(ctx); 1418 if (fd_checkpoint > 0) 1419 close(fd_checkpoint); 1420 if (meta_filename != NULL) 1421 free(meta_filename); 1422 if (kdata_filename != NULL) 1423 free(kdata_filename); 1424 if (xop != NULL) 1425 xo_destroy(xop); 1426 if (meta_file != NULL) 1427 fclose(meta_file); 1428 if (kdata_fd > 0) 1429 close(kdata_fd); 1430 return (error); 1431 } 1432 1433 static int 1434 handle_message(struct vmctx *ctx, nvlist_t *nvl) 1435 { 1436 const char *cmd; 1437 struct ipc_command **ipc_cmd; 1438 1439 if (!nvlist_exists_string(nvl, "cmd")) 1440 return (EINVAL); 1441 1442 cmd = nvlist_get_string(nvl, "cmd"); 1443 IPC_COMMAND_FOREACH(ipc_cmd, ipc_cmd_set) { 1444 if (strcmp(cmd, (*ipc_cmd)->name) == 0) 1445 return ((*ipc_cmd)->handler(ctx, nvl)); 1446 } 1447 1448 return (EOPNOTSUPP); 1449 } 1450 1451 /* 1452 * Listen for commands from bhyvectl 1453 */ 1454 void * 1455 checkpoint_thread(void *param) 1456 { 1457 int fd; 1458 struct checkpoint_thread_info *thread_info; 1459 nvlist_t *nvl; 1460 1461 pthread_set_name_np(pthread_self(), "checkpoint thread"); 1462 thread_info = (struct checkpoint_thread_info *)param; 1463 1464 while ((fd = accept(thread_info->socket_fd, NULL, NULL)) != -1) { 1465 nvl = nvlist_recv(fd, 0); 1466 if (nvl != NULL) 1467 handle_message(thread_info->ctx, nvl); 1468 else 1469 EPRINTLN("nvlist_recv() failed: %s", strerror(errno)); 1470 1471 close(fd); 1472 nvlist_destroy(nvl); 1473 } 1474 1475 return (NULL); 1476 } 1477 1478 static int 1479 vm_do_checkpoint(struct vmctx *ctx, const nvlist_t *nvl) 1480 { 1481 int error; 1482 1483 if (!nvlist_exists_string(nvl, "filename") || 1484 !nvlist_exists_bool(nvl, "suspend")) 1485 error = EINVAL; 1486 else 1487 error = vm_checkpoint(ctx, nvlist_get_string(nvl, "filename"), 1488 nvlist_get_bool(nvl, "suspend")); 1489 1490 return (error); 1491 } 1492 IPC_COMMAND(ipc_cmd_set, checkpoint, vm_do_checkpoint); 1493 1494 void 1495 init_snapshot(void) 1496 { 1497 int err; 1498 1499 err = pthread_mutex_init(&vcpu_lock, NULL); 1500 if (err != 0) 1501 errc(1, err, "checkpoint mutex init"); 1502 err = pthread_cond_init(&vcpus_idle, NULL); 1503 if (err != 0) 1504 errc(1, err, "checkpoint cv init (vcpus_idle)"); 1505 err = pthread_cond_init(&vcpus_can_run, NULL); 1506 if (err != 0) 1507 errc(1, err, "checkpoint cv init (vcpus_can_run)"); 1508 } 1509 1510 /* 1511 * Create the listening socket for IPC with bhyvectl 1512 */ 1513 int 1514 init_checkpoint_thread(struct vmctx *ctx) 1515 { 1516 struct checkpoint_thread_info *checkpoint_info = NULL; 1517 struct sockaddr_un addr; 1518 int socket_fd; 1519 pthread_t checkpoint_pthread; 1520 int err; 1521 1522 memset(&addr, 0, sizeof(addr)); 1523 1524 socket_fd = socket(PF_UNIX, SOCK_STREAM, 0); 1525 if (socket_fd < 0) { 1526 EPRINTLN("Socket creation failed: %s", strerror(errno)); 1527 err = -1; 1528 goto fail; 1529 } 1530 1531 addr.sun_family = AF_UNIX; 1532 1533 snprintf(addr.sun_path, sizeof(addr.sun_path), "%s%s", 1534 BHYVE_RUN_DIR, vm_get_name(ctx)); 1535 addr.sun_len = SUN_LEN(&addr); 1536 unlink(addr.sun_path); 1537 1538 if (bind(socket_fd, (struct sockaddr *)&addr, addr.sun_len) != 0) { 1539 EPRINTLN("Failed to bind socket \"%s\": %s\n", 1540 addr.sun_path, strerror(errno)); 1541 err = -1; 1542 goto fail; 1543 } 1544 1545 if (listen(socket_fd, 10) < 0) { 1546 EPRINTLN("ipc socket listen: %s\n", strerror(errno)); 1547 err = errno; 1548 goto fail; 1549 } 1550 1551 checkpoint_info = calloc(1, sizeof(*checkpoint_info)); 1552 checkpoint_info->ctx = ctx; 1553 checkpoint_info->socket_fd = socket_fd; 1554 1555 err = pthread_create(&checkpoint_pthread, NULL, checkpoint_thread, 1556 checkpoint_info); 1557 if (err != 0) 1558 goto fail; 1559 1560 return (0); 1561 fail: 1562 free(checkpoint_info); 1563 if (socket_fd > 0) 1564 close(socket_fd); 1565 unlink(addr.sun_path); 1566 1567 return (err); 1568 } 1569 1570 void 1571 vm_snapshot_buf_err(const char *bufname, const enum vm_snapshot_op op) 1572 { 1573 const char *__op; 1574 1575 if (op == VM_SNAPSHOT_SAVE) 1576 __op = "save"; 1577 else if (op == VM_SNAPSHOT_RESTORE) 1578 __op = "restore"; 1579 else 1580 __op = "unknown"; 1581 1582 fprintf(stderr, "%s: snapshot-%s failed for %s\r\n", 1583 __func__, __op, bufname); 1584 } 1585 1586 int 1587 vm_snapshot_buf(void *data, size_t data_size, struct vm_snapshot_meta *meta) 1588 { 1589 struct vm_snapshot_buffer *buffer; 1590 int op; 1591 1592 buffer = &meta->buffer; 1593 op = meta->op; 1594 1595 if (buffer->buf_rem < data_size) { 1596 fprintf(stderr, "%s: buffer too small\r\n", __func__); 1597 return (E2BIG); 1598 } 1599 1600 if (op == VM_SNAPSHOT_SAVE) 1601 memcpy(buffer->buf, data, data_size); 1602 else if (op == VM_SNAPSHOT_RESTORE) 1603 memcpy(data, buffer->buf, data_size); 1604 else 1605 return (EINVAL); 1606 1607 buffer->buf += data_size; 1608 buffer->buf_rem -= data_size; 1609 1610 return (0); 1611 } 1612 1613 size_t 1614 vm_get_snapshot_size(struct vm_snapshot_meta *meta) 1615 { 1616 size_t length; 1617 struct vm_snapshot_buffer *buffer; 1618 1619 buffer = &meta->buffer; 1620 1621 if (buffer->buf_size < buffer->buf_rem) { 1622 fprintf(stderr, "%s: Invalid buffer: size = %zu, rem = %zu\r\n", 1623 __func__, buffer->buf_size, buffer->buf_rem); 1624 length = 0; 1625 } else { 1626 length = buffer->buf_size - buffer->buf_rem; 1627 } 1628 1629 return (length); 1630 } 1631 1632 int 1633 vm_snapshot_guest2host_addr(void **addrp, size_t len, bool restore_null, 1634 struct vm_snapshot_meta *meta) 1635 { 1636 int ret; 1637 vm_paddr_t gaddr; 1638 1639 if (meta->op == VM_SNAPSHOT_SAVE) { 1640 gaddr = paddr_host2guest(meta->ctx, *addrp); 1641 if (gaddr == (vm_paddr_t) -1) { 1642 if (!restore_null || 1643 (restore_null && (*addrp != NULL))) { 1644 ret = EFAULT; 1645 goto done; 1646 } 1647 } 1648 1649 SNAPSHOT_VAR_OR_LEAVE(gaddr, meta, ret, done); 1650 } else if (meta->op == VM_SNAPSHOT_RESTORE) { 1651 SNAPSHOT_VAR_OR_LEAVE(gaddr, meta, ret, done); 1652 if (gaddr == (vm_paddr_t) -1) { 1653 if (!restore_null) { 1654 ret = EFAULT; 1655 goto done; 1656 } 1657 } 1658 1659 *addrp = paddr_guest2host(meta->ctx, gaddr, len); 1660 } else { 1661 ret = EINVAL; 1662 } 1663 1664 done: 1665 return (ret); 1666 } 1667 1668 int 1669 vm_snapshot_buf_cmp(void *data, size_t data_size, struct vm_snapshot_meta *meta) 1670 { 1671 struct vm_snapshot_buffer *buffer; 1672 int op; 1673 int ret; 1674 1675 buffer = &meta->buffer; 1676 op = meta->op; 1677 1678 if (buffer->buf_rem < data_size) { 1679 fprintf(stderr, "%s: buffer too small\r\n", __func__); 1680 ret = E2BIG; 1681 goto done; 1682 } 1683 1684 if (op == VM_SNAPSHOT_SAVE) { 1685 ret = 0; 1686 memcpy(buffer->buf, data, data_size); 1687 } else if (op == VM_SNAPSHOT_RESTORE) { 1688 ret = memcmp(data, buffer->buf, data_size); 1689 } else { 1690 ret = EINVAL; 1691 goto done; 1692 } 1693 1694 buffer->buf += data_size; 1695 buffer->buf_rem -= data_size; 1696 1697 done: 1698 return (ret); 1699 } 1700