2 * SPDX-License-Identifier: BSD-2-Clause
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.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
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.
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
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include <sys/types.h>
40 #ifndef WITHOUT_CAPSICUM
41 #include <sys/capsicum.h>
44 #include <sys/socket.h>
49 #include <machine/atomic.h>
51 #ifndef WITHOUT_CAPSICUM
52 #include <capsicum_helpers.h>
66 #include <pthread_np.h>
69 #include <sys/ioctl.h>
71 #include <machine/vmm.h>
72 #ifndef WITHOUT_CAPSICUM
73 #include <machine/vmm_dev.h>
75 #include <machine/vmm_snapshot.h>
92 #include "smbiostbl.h"
95 #include "spinup_ap.h"
101 struct spinner_info {
102 const size_t *crtval;
107 extern int guest_ncpus;
109 static struct winsize winsize;
110 static sig_t old_winch_handler;
113 #define MB (1024UL * KB)
114 #define GB (1024UL * MB)
116 #define SNAPSHOT_CHUNK (4 * MB)
117 #define PROG_BUF_SZ (8192)
119 #define SNAPSHOT_BUFFER_SIZE (20 * MB)
121 #define JSON_KERNEL_ARR_KEY "kern_structs"
122 #define JSON_DEV_ARR_KEY "devices"
123 #define JSON_BASIC_METADATA_KEY "basic metadata"
124 #define JSON_SNAPSHOT_REQ_KEY "device"
125 #define JSON_SIZE_KEY "size"
126 #define JSON_FILE_OFFSET_KEY "file_offset"
128 #define JSON_NCPUS_KEY "ncpus"
129 #define JSON_VMNAME_KEY "vmname"
130 #define JSON_MEMSIZE_KEY "memsize"
131 #define JSON_MEMFLAGS_KEY "memflags"
135 __typeof__ (a) _a = (a); \
136 __typeof__ (b) _b = (b); \
140 static const struct vm_snapshot_kern_info snapshot_kern_structs[] = {
141 { "vhpet", STRUCT_VHPET },
143 { "vioapic", STRUCT_VIOAPIC },
144 { "vlapic", STRUCT_VLAPIC },
145 { "vmcx", STRUCT_VMCX },
146 { "vatpit", STRUCT_VATPIT },
147 { "vatpic", STRUCT_VATPIC },
148 { "vpmtmr", STRUCT_VPMTMR },
149 { "vrtc", STRUCT_VRTC },
152 static cpuset_t vcpus_active, vcpus_suspended;
153 static pthread_mutex_t vcpu_lock;
154 static pthread_cond_t vcpus_idle, vcpus_can_run;
155 static bool checkpoint_active;
158 * TODO: Harden this function and all of its callers since 'base_str' is a user
162 strcat_extension(const char *base_str, const char *ext)
165 size_t base_len, ext_len;
167 base_len = strnlen(base_str, NAME_MAX);
168 ext_len = strnlen(ext, NAME_MAX);
170 if (base_len + ext_len > NAME_MAX) {
171 fprintf(stderr, "Filename exceeds maximum length.\n");
175 res = malloc(base_len + ext_len + 1);
177 perror("Failed to allocate memory.");
181 memcpy(res, base_str, base_len);
182 memcpy(res + base_len, ext, ext_len);
183 res[base_len + ext_len] = 0;
189 destroy_restore_state(struct restore_state *rstate)
191 if (rstate == NULL) {
192 fprintf(stderr, "Attempting to destroy NULL restore struct.\n");
196 if (rstate->kdata_map != MAP_FAILED)
197 munmap(rstate->kdata_map, rstate->kdata_len);
199 if (rstate->kdata_fd > 0)
200 close(rstate->kdata_fd);
201 if (rstate->vmmem_fd > 0)
202 close(rstate->vmmem_fd);
204 if (rstate->meta_root_obj != NULL)
205 ucl_object_unref(rstate->meta_root_obj);
206 if (rstate->meta_parser != NULL)
207 ucl_parser_free(rstate->meta_parser);
211 load_vmmem_file(const char *filename, struct restore_state *rstate)
216 rstate->vmmem_fd = open(filename, O_RDONLY);
217 if (rstate->vmmem_fd < 0) {
218 perror("Failed to open restore file");
222 err = fstat(rstate->vmmem_fd, &sb);
224 perror("Failed to stat restore file");
228 if (sb.st_size == 0) {
229 fprintf(stderr, "Restore file is empty.\n");
233 rstate->vmmem_len = sb.st_size;
238 if (rstate->vmmem_fd > 0)
239 close(rstate->vmmem_fd);
244 load_kdata_file(const char *filename, struct restore_state *rstate)
249 rstate->kdata_fd = open(filename, O_RDONLY);
250 if (rstate->kdata_fd < 0) {
251 perror("Failed to open kernel data file");
255 err = fstat(rstate->kdata_fd, &sb);
257 perror("Failed to stat kernel data file");
261 if (sb.st_size == 0) {
262 fprintf(stderr, "Kernel data file is empty.\n");
266 rstate->kdata_len = sb.st_size;
267 rstate->kdata_map = mmap(NULL, rstate->kdata_len, PROT_READ,
268 MAP_SHARED, rstate->kdata_fd, 0);
269 if (rstate->kdata_map == MAP_FAILED) {
270 perror("Failed to map restore file");
277 if (rstate->kdata_fd > 0)
278 close(rstate->kdata_fd);
283 load_metadata_file(const char *filename, struct restore_state *rstate)
286 struct ucl_parser *parser;
289 parser = ucl_parser_new(UCL_PARSER_DEFAULT);
290 if (parser == NULL) {
291 fprintf(stderr, "Failed to initialize UCL parser.\n");
293 goto err_load_metadata;
296 err = ucl_parser_add_file(parser, filename);
298 fprintf(stderr, "Failed to parse metadata file: '%s'\n",
301 goto err_load_metadata;
304 obj = ucl_parser_get_object(parser);
306 fprintf(stderr, "Failed to parse object.\n");
308 goto err_load_metadata;
311 rstate->meta_parser = parser;
312 rstate->meta_root_obj = (ucl_object_t *)obj;
318 ucl_parser_free(parser);
323 load_restore_file(const char *filename, struct restore_state *rstate)
326 char *kdata_filename = NULL, *meta_filename = NULL;
328 assert(filename != NULL);
329 assert(rstate != NULL);
331 memset(rstate, 0, sizeof(*rstate));
332 rstate->kdata_map = MAP_FAILED;
334 err = load_vmmem_file(filename, rstate);
336 fprintf(stderr, "Failed to load guest RAM file.\n");
340 kdata_filename = strcat_extension(filename, ".kern");
341 if (kdata_filename == NULL) {
342 fprintf(stderr, "Failed to construct kernel data filename.\n");
346 err = load_kdata_file(kdata_filename, rstate);
348 fprintf(stderr, "Failed to load guest kernel data file.\n");
352 meta_filename = strcat_extension(filename, ".meta");
353 if (meta_filename == NULL) {
354 fprintf(stderr, "Failed to construct kernel metadata filename.\n");
358 err = load_metadata_file(meta_filename, rstate);
360 fprintf(stderr, "Failed to load guest metadata file.\n");
367 destroy_restore_state(rstate);
368 if (kdata_filename != NULL)
369 free(kdata_filename);
370 if (meta_filename != NULL)
375 #define JSON_GET_INT_OR_RETURN(key, obj, result_ptr, ret) \
377 const ucl_object_t *obj__; \
378 obj__ = ucl_object_lookup(obj, key); \
379 if (obj__ == NULL) { \
380 fprintf(stderr, "Missing key: '%s'", key); \
383 if (!ucl_object_toint_safe(obj__, result_ptr)) { \
384 fprintf(stderr, "Cannot convert '%s' value to int.", key); \
389 #define JSON_GET_STRING_OR_RETURN(key, obj, result_ptr, ret) \
391 const ucl_object_t *obj__; \
392 obj__ = ucl_object_lookup(obj, key); \
393 if (obj__ == NULL) { \
394 fprintf(stderr, "Missing key: '%s'", key); \
397 if (!ucl_object_tostring_safe(obj__, result_ptr)) { \
398 fprintf(stderr, "Cannot convert '%s' value to string.", key); \
404 lookup_check_dev(const char *dev_name, struct restore_state *rstate,
405 const ucl_object_t *obj, size_t *data_size)
407 const char *snapshot_req;
408 int64_t size, file_offset;
411 JSON_GET_STRING_OR_RETURN(JSON_SNAPSHOT_REQ_KEY, obj,
412 &snapshot_req, NULL);
413 assert(snapshot_req != NULL);
414 if (!strcmp(snapshot_req, dev_name)) {
415 JSON_GET_INT_OR_RETURN(JSON_SIZE_KEY, obj,
419 JSON_GET_INT_OR_RETURN(JSON_FILE_OFFSET_KEY, obj,
421 assert(file_offset >= 0);
422 assert((uint64_t)file_offset + size <= rstate->kdata_len);
424 *data_size = (size_t)size;
425 return ((uint8_t *)rstate->kdata_map + file_offset);
432 lookup_dev(const char *dev_name, const char *key, struct restore_state *rstate,
435 const ucl_object_t *devs = NULL, *obj = NULL;
436 ucl_object_iter_t it = NULL;
439 devs = ucl_object_lookup(rstate->meta_root_obj, key);
441 fprintf(stderr, "Failed to find '%s' object.\n",
446 if (ucl_object_type(devs) != UCL_ARRAY) {
447 fprintf(stderr, "Object '%s' is not an array.\n",
452 while ((obj = ucl_object_iterate(devs, &it, true)) != NULL) {
453 ret = lookup_check_dev(dev_name, rstate, obj, data_size);
461 static const ucl_object_t *
462 lookup_basic_metadata_object(struct restore_state *rstate)
464 const ucl_object_t *basic_meta_obj = NULL;
466 basic_meta_obj = ucl_object_lookup(rstate->meta_root_obj,
467 JSON_BASIC_METADATA_KEY);
468 if (basic_meta_obj == NULL) {
469 fprintf(stderr, "Failed to find '%s' object.\n",
470 JSON_BASIC_METADATA_KEY);
474 if (ucl_object_type(basic_meta_obj) != UCL_OBJECT) {
475 fprintf(stderr, "Object '%s' is not a JSON object.\n",
476 JSON_BASIC_METADATA_KEY);
480 return (basic_meta_obj);
484 lookup_vmname(struct restore_state *rstate)
487 const ucl_object_t *obj;
489 obj = lookup_basic_metadata_object(rstate);
493 JSON_GET_STRING_OR_RETURN(JSON_VMNAME_KEY, obj, &vmname, NULL);
498 lookup_memflags(struct restore_state *rstate)
501 const ucl_object_t *obj;
503 obj = lookup_basic_metadata_object(rstate);
507 JSON_GET_INT_OR_RETURN(JSON_MEMFLAGS_KEY, obj, &memflags, 0);
509 return ((int)memflags);
513 lookup_memsize(struct restore_state *rstate)
516 const ucl_object_t *obj;
518 obj = lookup_basic_metadata_object(rstate);
522 JSON_GET_INT_OR_RETURN(JSON_MEMSIZE_KEY, obj, &memsize, 0);
526 return ((size_t)memsize);
531 lookup_guest_ncpus(struct restore_state *rstate)
534 const ucl_object_t *obj;
536 obj = lookup_basic_metadata_object(rstate);
540 JSON_GET_INT_OR_RETURN(JSON_NCPUS_KEY, obj, &ncpus, 0);
545 winch_handler(int signal __unused)
548 ioctl(STDOUT_FILENO, TIOCGWINSZ, &winsize);
549 #endif /* TIOCGWINSZ */
553 print_progress(size_t crtval, const size_t maxval)
556 double crtval_gb, maxval_gb;
557 size_t i, win_width, prog_start, prog_done, prog_end;
560 static char prog_buf[PROG_BUF_SZ];
561 static const size_t len = sizeof(prog_buf);
564 static const char *div_str;
566 static char wip_bar[] = { '/', '-', '\\', '|' };
567 static int wip_idx = 0;
570 printf("[0B / 0B]\r\n");
577 if (maxval > 10 * GB) {
580 } else if (maxval > 10 * MB) {
588 crtval_gb = (double) crtval / div;
589 maxval_gb = (double) maxval / div;
591 rc = snprintf(prog_buf, len, "%.03lf", maxval_gb);
593 fprintf(stderr, "Maxval too big\n");
598 rc = snprintf(prog_buf, len, "\r[%*.03lf%s / %.03lf%s] |",
599 mval_len, crtval_gb, div_str, maxval_gb, div_str);
602 fprintf(stderr, "Buffer too small to print progress\n");
606 win_width = min(winsize.ws_col, len);
609 if (prog_start < (win_width - 2)) {
610 prog_end = win_width - prog_start - 2;
611 prog_done = prog_end * (crtval_gb / maxval_gb);
613 for (i = prog_start; i < prog_start + prog_done; i++)
616 if (crtval != maxval) {
617 prog_buf[i] = wip_bar[wip_idx];
618 wip_idx = (wip_idx + 1) % sizeof(wip_bar);
624 for (; i < win_width - 2; i++)
627 prog_buf[win_width - 2] = '|';
630 prog_buf[win_width - 1] = '\0';
631 write(STDOUT_FILENO, prog_buf, win_width);
637 snapshot_spinner_cb(void *arg)
640 size_t crtval, maxval, total;
641 struct spinner_info *si;
649 ts.tv_nsec = 50 * 1000 * 1000; /* 50 ms sleep time */
652 crtval = *si->crtval;
656 rc = print_progress(crtval, total);
658 fprintf(stderr, "Failed to parse progress\n");
662 nanosleep(&ts, NULL);
663 } while (crtval < maxval);
670 vm_snapshot_mem_part(const int snapfd, const size_t foff, void *src,
671 const size_t len, const size_t totalmem, const bool op_wr)
674 size_t part_done, todo, rem;
677 pthread_t spinner_th;
678 struct spinner_info *si;
680 if (lseek(snapfd, foff, SEEK_SET) < 0) {
681 perror("Failed to change file offset");
685 show_progress = false;
686 if (isatty(STDIN_FILENO) && (winsize.ws_col != 0))
687 show_progress = true;
693 si = &(struct spinner_info) {
694 .crtval = &part_done,
695 .maxval = foff + len,
699 rc = pthread_create(&spinner_th, 0, snapshot_spinner_cb, si);
701 perror("Unable to create spinner thread");
702 show_progress = false;
708 todo = min(SNAPSHOT_CHUNK, rem);
713 done = write(snapfd, src, todo);
715 done = read(snapfd, src, todo);
717 perror("Failed to write in file");
721 src = (uint8_t *)src + done;
727 rc = pthread_join(spinner_th, NULL);
729 perror("Unable to end spinner thread");
736 vm_snapshot_mem(struct vmctx *ctx, int snapfd, size_t memsz, const bool op_wr)
739 size_t lowmem, highmem, totalmem;
742 ret = vm_get_guestmem_from_ctx(ctx, &baseaddr, &lowmem, &highmem);
744 fprintf(stderr, "%s: unable to retrieve guest memory size\r\n",
748 totalmem = lowmem + highmem;
750 if ((op_wr == false) && (totalmem != memsz)) {
751 fprintf(stderr, "%s: mem size mismatch: %ld vs %ld\r\n",
752 __func__, totalmem, memsz);
758 ioctl(STDOUT_FILENO, TIOCGWINSZ, &winsize);
759 #endif /* TIOCGWINSZ */
760 old_winch_handler = signal(SIGWINCH, winch_handler);
762 ret = vm_snapshot_mem_part(snapfd, 0, baseaddr, lowmem,
765 fprintf(stderr, "%s: Could not %s lowmem\r\n",
766 __func__, op_wr ? "write" : "read");
774 ret = vm_snapshot_mem_part(snapfd, lowmem, baseaddr + 4*GB,
775 highmem, totalmem, op_wr);
777 fprintf(stderr, "%s: Could not %s highmem\r\n",
778 __func__, op_wr ? "write" : "read");
785 signal(SIGWINCH, old_winch_handler);
791 restore_vm_mem(struct vmctx *ctx, struct restore_state *rstate)
795 restored = vm_snapshot_mem(ctx, rstate->vmmem_fd, rstate->vmmem_len,
798 if (restored != rstate->vmmem_len)
805 vm_restore_kern_structs(struct vmctx *ctx, struct restore_state *rstate)
807 for (unsigned i = 0; i < nitems(snapshot_kern_structs); i++) {
808 const struct vm_snapshot_kern_info *info;
809 struct vm_snapshot_meta *meta;
813 info = &snapshot_kern_structs[i];
814 data = lookup_dev(info->struct_name, JSON_KERNEL_ARR_KEY, rstate, &size);
816 errx(EX_DATAERR, "Cannot find kern struct %s",
820 errx(EX_DATAERR, "data with zero size for %s",
823 meta = &(struct vm_snapshot_meta) {
824 .dev_name = info->struct_name,
825 .dev_req = info->req,
827 .buffer.buf_start = data,
828 .buffer.buf_size = size,
831 .buffer.buf_rem = size,
833 .op = VM_SNAPSHOT_RESTORE,
836 if (vm_snapshot_req(ctx, meta))
837 err(EX_DATAERR, "Failed to restore %s",
844 vm_restore_device(struct restore_state *rstate, vm_snapshot_dev_cb func,
845 const char *name, void *data)
850 struct vm_snapshot_meta *meta;
852 dev_ptr = lookup_dev(name, JSON_DEV_ARR_KEY, rstate, &dev_size);
854 if (dev_ptr == NULL) {
855 EPRINTLN("Failed to lookup dev: %s", name);
860 EPRINTLN("Restore device size is 0: %s", name);
864 meta = &(struct vm_snapshot_meta) {
868 .buffer.buf_start = dev_ptr,
869 .buffer.buf_size = dev_size,
871 .buffer.buf = dev_ptr,
872 .buffer.buf_rem = dev_size,
874 .op = VM_SNAPSHOT_RESTORE,
879 EPRINTLN("Failed to restore dev: %s %d", name, ret);
887 vm_restore_devices(struct restore_state *rstate)
890 struct pci_devinst *pdi = NULL;
892 while ((pdi = pci_next(pdi)) != NULL) {
893 ret = vm_restore_device(rstate, pci_snapshot, pdi->pi_name, pdi);
898 return (vm_restore_device(rstate, atkbdc_snapshot, "atkbdc", NULL));
902 vm_pause_devices(void)
905 struct pci_devinst *pdi = NULL;
907 while ((pdi = pci_next(pdi)) != NULL) {
908 ret = pci_pause(pdi);
910 EPRINTLN("Cannot pause dev %s: %d", pdi->pi_name, ret);
919 vm_resume_devices(void)
922 struct pci_devinst *pdi = NULL;
924 while ((pdi = pci_next(pdi)) != NULL) {
925 ret = pci_resume(pdi);
927 EPRINTLN("Cannot resume '%s': %d", pdi->pi_name, ret);
936 vm_save_kern_struct(struct vmctx *ctx, int data_fd, xo_handle_t *xop,
937 const char *array_key, struct vm_snapshot_meta *meta, off_t *offset)
943 ret = vm_snapshot_req(ctx, meta);
945 fprintf(stderr, "%s: Failed to snapshot struct %s\r\n",
946 __func__, meta->dev_name);
951 data_size = vm_get_snapshot_size(meta);
953 /* XXX-MJ no handling for short writes. */
954 write_cnt = write(data_fd, meta->buffer.buf_start, data_size);
955 if (write_cnt < 0 || (size_t)write_cnt != data_size) {
956 perror("Failed to write all snapshotted data.");
961 /* Write metadata. */
962 xo_open_instance_h(xop, array_key);
963 xo_emit_h(xop, "{:" JSON_SNAPSHOT_REQ_KEY "/%s}\n",
965 xo_emit_h(xop, "{:" JSON_SIZE_KEY "/%lu}\n", data_size);
966 xo_emit_h(xop, "{:" JSON_FILE_OFFSET_KEY "/%lu}\n", *offset);
967 xo_close_instance_h(xop, JSON_KERNEL_ARR_KEY);
969 *offset += data_size;
976 vm_save_kern_structs(struct vmctx *ctx, int data_fd, xo_handle_t *xop)
979 size_t buf_size, i, offset;
981 struct vm_snapshot_meta *meta;
985 buf_size = SNAPSHOT_BUFFER_SIZE;
987 buffer = malloc(SNAPSHOT_BUFFER_SIZE * sizeof(char));
988 if (buffer == NULL) {
990 perror("Failed to allocate memory for snapshot buffer");
991 goto err_vm_snapshot_kern_data;
994 meta = &(struct vm_snapshot_meta) {
995 .buffer.buf_start = buffer,
996 .buffer.buf_size = buf_size,
998 .op = VM_SNAPSHOT_SAVE,
1001 xo_open_list_h(xop, JSON_KERNEL_ARR_KEY);
1002 for (i = 0; i < nitems(snapshot_kern_structs); i++) {
1003 meta->dev_name = snapshot_kern_structs[i].struct_name;
1004 meta->dev_req = snapshot_kern_structs[i].req;
1006 memset(meta->buffer.buf_start, 0, meta->buffer.buf_size);
1007 meta->buffer.buf = meta->buffer.buf_start;
1008 meta->buffer.buf_rem = meta->buffer.buf_size;
1010 ret = vm_save_kern_struct(ctx, data_fd, xop,
1011 JSON_DEV_ARR_KEY, meta, &offset);
1014 goto err_vm_snapshot_kern_data;
1017 xo_close_list_h(xop, JSON_KERNEL_ARR_KEY);
1019 err_vm_snapshot_kern_data:
1026 vm_snapshot_basic_metadata(struct vmctx *ctx, xo_handle_t *xop, size_t memsz)
1029 xo_open_container_h(xop, JSON_BASIC_METADATA_KEY);
1030 xo_emit_h(xop, "{:" JSON_NCPUS_KEY "/%ld}\n", guest_ncpus);
1031 xo_emit_h(xop, "{:" JSON_VMNAME_KEY "/%s}\n", vm_get_name(ctx));
1032 xo_emit_h(xop, "{:" JSON_MEMSIZE_KEY "/%lu}\n", memsz);
1033 xo_emit_h(xop, "{:" JSON_MEMFLAGS_KEY "/%d}\n", vm_get_memflags(ctx));
1034 xo_close_container_h(xop, JSON_BASIC_METADATA_KEY);
1040 vm_snapshot_dev_write_data(int data_fd, xo_handle_t *xop, const char *array_key,
1041 struct vm_snapshot_meta *meta, off_t *offset)
1046 data_size = vm_get_snapshot_size(meta);
1048 /* XXX-MJ no handling for short writes. */
1049 ret = write(data_fd, meta->buffer.buf_start, data_size);
1050 if (ret < 0 || (size_t)ret != data_size) {
1051 perror("Failed to write all snapshotted data.");
1055 /* Write metadata. */
1056 xo_open_instance_h(xop, array_key);
1057 xo_emit_h(xop, "{:" JSON_SNAPSHOT_REQ_KEY "/%s}\n", meta->dev_name);
1058 xo_emit_h(xop, "{:" JSON_SIZE_KEY "/%lu}\n", data_size);
1059 xo_emit_h(xop, "{:" JSON_FILE_OFFSET_KEY "/%lu}\n", *offset);
1060 xo_close_instance_h(xop, array_key);
1062 *offset += data_size;
1068 vm_snapshot_device(vm_snapshot_dev_cb func, const char *dev_name,
1069 void *devdata, int data_fd, xo_handle_t *xop,
1070 struct vm_snapshot_meta *meta, off_t *offset)
1074 memset(meta->buffer.buf_start, 0, meta->buffer.buf_size);
1075 meta->buffer.buf = meta->buffer.buf_start;
1076 meta->buffer.buf_rem = meta->buffer.buf_size;
1077 meta->dev_name = dev_name;
1078 meta->dev_data = devdata;
1082 EPRINTLN("Failed to snapshot %s; ret=%d", dev_name, ret);
1086 ret = vm_snapshot_dev_write_data(data_fd, xop, JSON_DEV_ARR_KEY, meta,
1095 vm_snapshot_devices(int data_fd, xo_handle_t *xop)
1101 struct vm_snapshot_meta *meta;
1102 struct pci_devinst *pdi;
1104 buf_size = SNAPSHOT_BUFFER_SIZE;
1106 offset = lseek(data_fd, 0, SEEK_CUR);
1108 perror("Failed to get data file current offset.");
1112 buffer = malloc(buf_size);
1113 if (buffer == NULL) {
1114 perror("Failed to allocate memory for snapshot buffer");
1119 meta = &(struct vm_snapshot_meta) {
1120 .buffer.buf_start = buffer,
1121 .buffer.buf_size = buf_size,
1123 .op = VM_SNAPSHOT_SAVE,
1126 xo_open_list_h(xop, JSON_DEV_ARR_KEY);
1128 /* Save PCI devices */
1130 while ((pdi = pci_next(pdi)) != NULL) {
1131 ret = vm_snapshot_device(pci_snapshot, pdi->pi_name, pdi,
1132 data_fd, xop, meta, &offset);
1137 ret = vm_snapshot_device(atkbdc_snapshot, "atkbdc", NULL,
1138 data_fd, xop, meta, &offset);
1140 xo_close_list_h(xop, JSON_DEV_ARR_KEY);
1149 checkpoint_cpu_add(int vcpu)
1152 pthread_mutex_lock(&vcpu_lock);
1153 CPU_SET(vcpu, &vcpus_active);
1155 if (checkpoint_active) {
1156 CPU_SET(vcpu, &vcpus_suspended);
1157 while (checkpoint_active)
1158 pthread_cond_wait(&vcpus_can_run, &vcpu_lock);
1159 CPU_CLR(vcpu, &vcpus_suspended);
1161 pthread_mutex_unlock(&vcpu_lock);
1165 * When a vCPU is suspended for any reason, it calls
1166 * checkpoint_cpu_suspend(). This records that the vCPU is idle.
1167 * Before returning from suspension, checkpoint_cpu_resume() is
1168 * called. In suspend we note that the vCPU is idle. In resume we
1169 * pause the vCPU thread until the checkpoint is complete. The reason
1170 * for the two-step process is that vCPUs might already be stopped in
1171 * the debug server when a checkpoint is requested. This approach
1172 * allows us to account for and handle those vCPUs.
1175 checkpoint_cpu_suspend(int vcpu)
1178 pthread_mutex_lock(&vcpu_lock);
1179 CPU_SET(vcpu, &vcpus_suspended);
1180 if (checkpoint_active && CPU_CMP(&vcpus_active, &vcpus_suspended) == 0)
1181 pthread_cond_signal(&vcpus_idle);
1182 pthread_mutex_unlock(&vcpu_lock);
1186 checkpoint_cpu_resume(int vcpu)
1189 pthread_mutex_lock(&vcpu_lock);
1190 while (checkpoint_active)
1191 pthread_cond_wait(&vcpus_can_run, &vcpu_lock);
1192 CPU_CLR(vcpu, &vcpus_suspended);
1193 pthread_mutex_unlock(&vcpu_lock);
1197 vm_vcpu_pause(struct vmctx *ctx)
1200 pthread_mutex_lock(&vcpu_lock);
1201 checkpoint_active = true;
1202 vm_suspend_all_cpus(ctx);
1203 while (CPU_CMP(&vcpus_active, &vcpus_suspended) != 0)
1204 pthread_cond_wait(&vcpus_idle, &vcpu_lock);
1205 pthread_mutex_unlock(&vcpu_lock);
1209 vm_vcpu_resume(struct vmctx *ctx)
1212 pthread_mutex_lock(&vcpu_lock);
1213 checkpoint_active = false;
1214 pthread_mutex_unlock(&vcpu_lock);
1215 vm_resume_all_cpus(ctx);
1216 pthread_cond_broadcast(&vcpus_can_run);
1220 vm_checkpoint(struct vmctx *ctx, int fddir, const char *checkpoint_file,
1223 int fd_checkpoint = 0, kdata_fd = 0, fd_meta;
1227 xo_handle_t *xop = NULL;
1228 char *meta_filename = NULL;
1229 char *kdata_filename = NULL;
1230 FILE *meta_file = NULL;
1232 kdata_filename = strcat_extension(checkpoint_file, ".kern");
1233 if (kdata_filename == NULL) {
1234 fprintf(stderr, "Failed to construct kernel data filename.\n");
1238 kdata_fd = openat(fddir, kdata_filename, O_WRONLY | O_CREAT | O_TRUNC, 0700);
1240 perror("Failed to open kernel data snapshot file.");
1245 fd_checkpoint = openat(fddir, checkpoint_file, O_RDWR | O_CREAT | O_TRUNC, 0700);
1247 if (fd_checkpoint < 0) {
1248 perror("Failed to create checkpoint file");
1253 meta_filename = strcat_extension(checkpoint_file, ".meta");
1254 if (meta_filename == NULL) {
1255 fprintf(stderr, "Failed to construct vm metadata filename.\n");
1259 fd_meta = openat(fddir, meta_filename, O_WRONLY | O_CREAT | O_TRUNC, 0700);
1261 meta_file = fdopen(fd_meta, "w");
1262 if (meta_file == NULL) {
1263 perror("Failed to open vm metadata snapshot file.");
1268 xop = xo_create_to_file(meta_file, XO_STYLE_JSON, XOF_PRETTY);
1270 perror("Failed to get libxo handle on metadata file.");
1276 ret = vm_pause_devices();
1278 fprintf(stderr, "Could not pause devices\r\n");
1283 memsz = vm_snapshot_mem(ctx, fd_checkpoint, 0, true);
1285 perror("Could not write guest memory to file");
1290 ret = vm_snapshot_basic_metadata(ctx, xop, memsz);
1292 fprintf(stderr, "Failed to snapshot vm basic metadata.\n");
1297 ret = vm_save_kern_structs(ctx, kdata_fd, xop);
1299 fprintf(stderr, "Failed to snapshot vm kernel data.\n");
1304 ret = vm_snapshot_devices(kdata_fd, xop);
1306 fprintf(stderr, "Failed to snapshot device state.\n");
1319 ret = vm_resume_devices();
1321 fprintf(stderr, "Could not resume devices\r\n");
1322 vm_vcpu_resume(ctx);
1323 if (fd_checkpoint > 0)
1324 close(fd_checkpoint);
1325 if (meta_filename != NULL)
1326 free(meta_filename);
1327 if (kdata_filename != NULL)
1328 free(kdata_filename);
1331 if (meta_file != NULL)
1339 handle_message(struct vmctx *ctx, nvlist_t *nvl)
1342 struct ipc_command **ipc_cmd;
1344 if (!nvlist_exists_string(nvl, "cmd"))
1347 cmd = nvlist_get_string(nvl, "cmd");
1348 IPC_COMMAND_FOREACH(ipc_cmd, ipc_cmd_set) {
1349 if (strcmp(cmd, (*ipc_cmd)->name) == 0)
1350 return ((*ipc_cmd)->handler(ctx, nvl));
1353 return (EOPNOTSUPP);
1357 * Listen for commands from bhyvectl
1360 checkpoint_thread(void *param)
1363 struct checkpoint_thread_info *thread_info;
1366 pthread_set_name_np(pthread_self(), "checkpoint thread");
1367 thread_info = (struct checkpoint_thread_info *)param;
1369 while ((fd = accept(thread_info->socket_fd, NULL, NULL)) != -1) {
1370 nvl = nvlist_recv(fd, 0);
1372 handle_message(thread_info->ctx, nvl);
1374 EPRINTLN("nvlist_recv() failed: %s", strerror(errno));
1377 nvlist_destroy(nvl);
1384 vm_do_checkpoint(struct vmctx *ctx, const nvlist_t *nvl)
1388 if (!nvlist_exists_string(nvl, "filename") ||
1389 !nvlist_exists_bool(nvl, "suspend") ||
1390 !nvlist_exists_descriptor(nvl, "fddir"))
1393 error = vm_checkpoint(ctx,
1394 nvlist_get_descriptor(nvl, "fddir"),
1395 nvlist_get_string(nvl, "filename"),
1396 nvlist_get_bool(nvl, "suspend"));
1400 IPC_COMMAND(ipc_cmd_set, checkpoint, vm_do_checkpoint);
1407 err = pthread_mutex_init(&vcpu_lock, NULL);
1409 errc(1, err, "checkpoint mutex init");
1410 err = pthread_cond_init(&vcpus_idle, NULL);
1412 errc(1, err, "checkpoint cv init (vcpus_idle)");
1413 err = pthread_cond_init(&vcpus_can_run, NULL);
1415 errc(1, err, "checkpoint cv init (vcpus_can_run)");
1419 * Create the listening socket for IPC with bhyvectl
1422 init_checkpoint_thread(struct vmctx *ctx)
1424 struct checkpoint_thread_info *checkpoint_info = NULL;
1425 struct sockaddr_un addr;
1427 pthread_t checkpoint_pthread;
1429 #ifndef WITHOUT_CAPSICUM
1430 cap_rights_t rights;
1433 memset(&addr, 0, sizeof(addr));
1435 socket_fd = socket(PF_UNIX, SOCK_STREAM, 0);
1436 if (socket_fd < 0) {
1437 EPRINTLN("Socket creation failed: %s", strerror(errno));
1442 addr.sun_family = AF_UNIX;
1444 snprintf(addr.sun_path, sizeof(addr.sun_path), "%s%s",
1445 BHYVE_RUN_DIR, vm_get_name(ctx));
1446 addr.sun_len = SUN_LEN(&addr);
1447 unlink(addr.sun_path);
1449 if (bind(socket_fd, (struct sockaddr *)&addr, addr.sun_len) != 0) {
1450 EPRINTLN("Failed to bind socket \"%s\": %s\n",
1451 addr.sun_path, strerror(errno));
1456 if (listen(socket_fd, 10) < 0) {
1457 EPRINTLN("ipc socket listen: %s\n", strerror(errno));
1462 #ifndef WITHOUT_CAPSICUM
1463 cap_rights_init(&rights, CAP_ACCEPT, CAP_READ, CAP_RECV, CAP_WRITE,
1464 CAP_SEND, CAP_GETSOCKOPT);
1466 if (caph_rights_limit(socket_fd, &rights) == -1)
1467 errx(EX_OSERR, "Unable to apply rights for sandbox");
1469 checkpoint_info = calloc(1, sizeof(*checkpoint_info));
1470 checkpoint_info->ctx = ctx;
1471 checkpoint_info->socket_fd = socket_fd;
1473 err = pthread_create(&checkpoint_pthread, NULL, checkpoint_thread,
1480 free(checkpoint_info);
1483 unlink(addr.sun_path);
1489 vm_snapshot_buf_err(const char *bufname, const enum vm_snapshot_op op)
1493 if (op == VM_SNAPSHOT_SAVE)
1495 else if (op == VM_SNAPSHOT_RESTORE)
1500 fprintf(stderr, "%s: snapshot-%s failed for %s\r\n",
1501 __func__, __op, bufname);
1505 vm_snapshot_buf(void *data, size_t data_size, struct vm_snapshot_meta *meta)
1507 struct vm_snapshot_buffer *buffer;
1510 buffer = &meta->buffer;
1513 if (buffer->buf_rem < data_size) {
1514 fprintf(stderr, "%s: buffer too small\r\n", __func__);
1518 if (op == VM_SNAPSHOT_SAVE)
1519 memcpy(buffer->buf, data, data_size);
1520 else if (op == VM_SNAPSHOT_RESTORE)
1521 memcpy(data, buffer->buf, data_size);
1525 buffer->buf += data_size;
1526 buffer->buf_rem -= data_size;
1532 vm_get_snapshot_size(struct vm_snapshot_meta *meta)
1535 struct vm_snapshot_buffer *buffer;
1537 buffer = &meta->buffer;
1539 if (buffer->buf_size < buffer->buf_rem) {
1540 fprintf(stderr, "%s: Invalid buffer: size = %zu, rem = %zu\r\n",
1541 __func__, buffer->buf_size, buffer->buf_rem);
1544 length = buffer->buf_size - buffer->buf_rem;
1551 vm_snapshot_guest2host_addr(struct vmctx *ctx, void **addrp, size_t len,
1552 bool restore_null, struct vm_snapshot_meta *meta)
1557 if (meta->op == VM_SNAPSHOT_SAVE) {
1558 gaddr = paddr_host2guest(ctx, *addrp);
1559 if (gaddr == (vm_paddr_t) -1) {
1560 if (!restore_null ||
1561 (restore_null && (*addrp != NULL))) {
1567 SNAPSHOT_VAR_OR_LEAVE(gaddr, meta, ret, done);
1568 } else if (meta->op == VM_SNAPSHOT_RESTORE) {
1569 SNAPSHOT_VAR_OR_LEAVE(gaddr, meta, ret, done);
1570 if (gaddr == (vm_paddr_t) -1) {
1571 if (!restore_null) {
1577 *addrp = paddr_guest2host(ctx, gaddr, len);
1587 vm_snapshot_buf_cmp(void *data, size_t data_size, struct vm_snapshot_meta *meta)
1589 struct vm_snapshot_buffer *buffer;
1593 buffer = &meta->buffer;
1596 if (buffer->buf_rem < data_size) {
1597 fprintf(stderr, "%s: buffer too small\r\n", __func__);
1602 if (op == VM_SNAPSHOT_SAVE) {
1604 memcpy(buffer->buf, data, data_size);
1605 } else if (op == VM_SNAPSHOT_RESTORE) {
1606 ret = memcmp(data, buffer->buf, data_size);
1612 buffer->buf += data_size;
1613 buffer->buf_rem -= data_size;