Merge remote-tracking branch 'remotes/thibault/tags/samuel-thibault' into staging
[qemu/ar7.git] / dump.c
blob107a67165a645dcd96dfcf64dfd6d398915014b1
1 /*
2 * QEMU dump
4 * Copyright Fujitsu, Corp. 2011, 2012
6 * Authors:
7 * Wen Congyang <wency@cn.fujitsu.com>
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
14 #include "qemu/osdep.h"
15 #include "qemu/cutils.h"
16 #include "elf.h"
17 #include "cpu.h"
18 #include "exec/hwaddr.h"
19 #include "monitor/monitor.h"
20 #include "sysemu/kvm.h"
21 #include "sysemu/dump.h"
22 #include "sysemu/sysemu.h"
23 #include "sysemu/memory_mapping.h"
24 #include "sysemu/cpus.h"
25 #include "qapi/error.h"
26 #include "qapi/qapi-commands-misc.h"
27 #include "qapi/qapi-events-misc.h"
28 #include "qapi/qmp/qerror.h"
29 #include "qemu/error-report.h"
30 #include "hw/misc/vmcoreinfo.h"
32 #ifdef TARGET_X86_64
33 #include "win_dump.h"
34 #endif
36 #include <zlib.h>
37 #ifdef CONFIG_LZO
38 #include <lzo/lzo1x.h>
39 #endif
40 #ifdef CONFIG_SNAPPY
41 #include <snappy-c.h>
42 #endif
43 #ifndef ELF_MACHINE_UNAME
44 #define ELF_MACHINE_UNAME "Unknown"
45 #endif
47 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
49 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size) \
50 ((DIV_ROUND_UP((hdr_size), 4) + \
51 DIV_ROUND_UP((name_size), 4) + \
52 DIV_ROUND_UP((desc_size), 4)) * 4)
54 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
56 if (s->dump_info.d_endian == ELFDATA2LSB) {
57 val = cpu_to_le16(val);
58 } else {
59 val = cpu_to_be16(val);
62 return val;
65 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
67 if (s->dump_info.d_endian == ELFDATA2LSB) {
68 val = cpu_to_le32(val);
69 } else {
70 val = cpu_to_be32(val);
73 return val;
76 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
78 if (s->dump_info.d_endian == ELFDATA2LSB) {
79 val = cpu_to_le64(val);
80 } else {
81 val = cpu_to_be64(val);
84 return val;
87 static int dump_cleanup(DumpState *s)
89 guest_phys_blocks_free(&s->guest_phys_blocks);
90 memory_mapping_list_free(&s->list);
91 close(s->fd);
92 g_free(s->guest_note);
93 s->guest_note = NULL;
94 if (s->resume) {
95 if (s->detached) {
96 qemu_mutex_lock_iothread();
98 vm_start();
99 if (s->detached) {
100 qemu_mutex_unlock_iothread();
104 return 0;
107 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
109 DumpState *s = opaque;
110 size_t written_size;
112 written_size = qemu_write_full(s->fd, buf, size);
113 if (written_size != size) {
114 return -errno;
117 return 0;
120 static void write_elf64_header(DumpState *s, Error **errp)
122 Elf64_Ehdr elf_header;
123 int ret;
125 memset(&elf_header, 0, sizeof(Elf64_Ehdr));
126 memcpy(&elf_header, ELFMAG, SELFMAG);
127 elf_header.e_ident[EI_CLASS] = ELFCLASS64;
128 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
129 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
130 elf_header.e_type = cpu_to_dump16(s, ET_CORE);
131 elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
132 elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
133 elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
134 elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
135 elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
136 elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
137 if (s->have_section) {
138 uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
140 elf_header.e_shoff = cpu_to_dump64(s, shoff);
141 elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
142 elf_header.e_shnum = cpu_to_dump16(s, 1);
145 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
146 if (ret < 0) {
147 error_setg_errno(errp, -ret, "dump: failed to write elf header");
151 static void write_elf32_header(DumpState *s, Error **errp)
153 Elf32_Ehdr elf_header;
154 int ret;
156 memset(&elf_header, 0, sizeof(Elf32_Ehdr));
157 memcpy(&elf_header, ELFMAG, SELFMAG);
158 elf_header.e_ident[EI_CLASS] = ELFCLASS32;
159 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
160 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
161 elf_header.e_type = cpu_to_dump16(s, ET_CORE);
162 elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
163 elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
164 elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
165 elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
166 elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
167 elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
168 if (s->have_section) {
169 uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
171 elf_header.e_shoff = cpu_to_dump32(s, shoff);
172 elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
173 elf_header.e_shnum = cpu_to_dump16(s, 1);
176 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
177 if (ret < 0) {
178 error_setg_errno(errp, -ret, "dump: failed to write elf header");
182 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
183 int phdr_index, hwaddr offset,
184 hwaddr filesz, Error **errp)
186 Elf64_Phdr phdr;
187 int ret;
189 memset(&phdr, 0, sizeof(Elf64_Phdr));
190 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
191 phdr.p_offset = cpu_to_dump64(s, offset);
192 phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
193 phdr.p_filesz = cpu_to_dump64(s, filesz);
194 phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
195 phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
197 assert(memory_mapping->length >= filesz);
199 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
200 if (ret < 0) {
201 error_setg_errno(errp, -ret,
202 "dump: failed to write program header table");
206 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
207 int phdr_index, hwaddr offset,
208 hwaddr filesz, Error **errp)
210 Elf32_Phdr phdr;
211 int ret;
213 memset(&phdr, 0, sizeof(Elf32_Phdr));
214 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
215 phdr.p_offset = cpu_to_dump32(s, offset);
216 phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
217 phdr.p_filesz = cpu_to_dump32(s, filesz);
218 phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
219 phdr.p_vaddr =
220 cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
222 assert(memory_mapping->length >= filesz);
224 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
225 if (ret < 0) {
226 error_setg_errno(errp, -ret,
227 "dump: failed to write program header table");
231 static void write_elf64_note(DumpState *s, Error **errp)
233 Elf64_Phdr phdr;
234 hwaddr begin = s->memory_offset - s->note_size;
235 int ret;
237 memset(&phdr, 0, sizeof(Elf64_Phdr));
238 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
239 phdr.p_offset = cpu_to_dump64(s, begin);
240 phdr.p_paddr = 0;
241 phdr.p_filesz = cpu_to_dump64(s, s->note_size);
242 phdr.p_memsz = cpu_to_dump64(s, s->note_size);
243 phdr.p_vaddr = 0;
245 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
246 if (ret < 0) {
247 error_setg_errno(errp, -ret,
248 "dump: failed to write program header table");
252 static inline int cpu_index(CPUState *cpu)
254 return cpu->cpu_index + 1;
257 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
258 Error **errp)
260 int ret;
262 if (s->guest_note) {
263 ret = f(s->guest_note, s->guest_note_size, s);
264 if (ret < 0) {
265 error_setg(errp, "dump: failed to write guest note");
270 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
271 Error **errp)
273 CPUState *cpu;
274 int ret;
275 int id;
277 CPU_FOREACH(cpu) {
278 id = cpu_index(cpu);
279 ret = cpu_write_elf64_note(f, cpu, id, s);
280 if (ret < 0) {
281 error_setg(errp, "dump: failed to write elf notes");
282 return;
286 CPU_FOREACH(cpu) {
287 ret = cpu_write_elf64_qemunote(f, cpu, s);
288 if (ret < 0) {
289 error_setg(errp, "dump: failed to write CPU status");
290 return;
294 write_guest_note(f, s, errp);
297 static void write_elf32_note(DumpState *s, Error **errp)
299 hwaddr begin = s->memory_offset - s->note_size;
300 Elf32_Phdr phdr;
301 int ret;
303 memset(&phdr, 0, sizeof(Elf32_Phdr));
304 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
305 phdr.p_offset = cpu_to_dump32(s, begin);
306 phdr.p_paddr = 0;
307 phdr.p_filesz = cpu_to_dump32(s, s->note_size);
308 phdr.p_memsz = cpu_to_dump32(s, s->note_size);
309 phdr.p_vaddr = 0;
311 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
312 if (ret < 0) {
313 error_setg_errno(errp, -ret,
314 "dump: failed to write program header table");
318 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
319 Error **errp)
321 CPUState *cpu;
322 int ret;
323 int id;
325 CPU_FOREACH(cpu) {
326 id = cpu_index(cpu);
327 ret = cpu_write_elf32_note(f, cpu, id, s);
328 if (ret < 0) {
329 error_setg(errp, "dump: failed to write elf notes");
330 return;
334 CPU_FOREACH(cpu) {
335 ret = cpu_write_elf32_qemunote(f, cpu, s);
336 if (ret < 0) {
337 error_setg(errp, "dump: failed to write CPU status");
338 return;
342 write_guest_note(f, s, errp);
345 static void write_elf_section(DumpState *s, int type, Error **errp)
347 Elf32_Shdr shdr32;
348 Elf64_Shdr shdr64;
349 int shdr_size;
350 void *shdr;
351 int ret;
353 if (type == 0) {
354 shdr_size = sizeof(Elf32_Shdr);
355 memset(&shdr32, 0, shdr_size);
356 shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
357 shdr = &shdr32;
358 } else {
359 shdr_size = sizeof(Elf64_Shdr);
360 memset(&shdr64, 0, shdr_size);
361 shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
362 shdr = &shdr64;
365 ret = fd_write_vmcore(&shdr, shdr_size, s);
366 if (ret < 0) {
367 error_setg_errno(errp, -ret,
368 "dump: failed to write section header table");
372 static void write_data(DumpState *s, void *buf, int length, Error **errp)
374 int ret;
376 ret = fd_write_vmcore(buf, length, s);
377 if (ret < 0) {
378 error_setg_errno(errp, -ret, "dump: failed to save memory");
379 } else {
380 s->written_size += length;
384 /* write the memory to vmcore. 1 page per I/O. */
385 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
386 int64_t size, Error **errp)
388 int64_t i;
389 Error *local_err = NULL;
391 for (i = 0; i < size / s->dump_info.page_size; i++) {
392 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
393 s->dump_info.page_size, &local_err);
394 if (local_err) {
395 error_propagate(errp, local_err);
396 return;
400 if ((size % s->dump_info.page_size) != 0) {
401 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
402 size % s->dump_info.page_size, &local_err);
403 if (local_err) {
404 error_propagate(errp, local_err);
405 return;
410 /* get the memory's offset and size in the vmcore */
411 static void get_offset_range(hwaddr phys_addr,
412 ram_addr_t mapping_length,
413 DumpState *s,
414 hwaddr *p_offset,
415 hwaddr *p_filesz)
417 GuestPhysBlock *block;
418 hwaddr offset = s->memory_offset;
419 int64_t size_in_block, start;
421 /* When the memory is not stored into vmcore, offset will be -1 */
422 *p_offset = -1;
423 *p_filesz = 0;
425 if (s->has_filter) {
426 if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
427 return;
431 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
432 if (s->has_filter) {
433 if (block->target_start >= s->begin + s->length ||
434 block->target_end <= s->begin) {
435 /* This block is out of the range */
436 continue;
439 if (s->begin <= block->target_start) {
440 start = block->target_start;
441 } else {
442 start = s->begin;
445 size_in_block = block->target_end - start;
446 if (s->begin + s->length < block->target_end) {
447 size_in_block -= block->target_end - (s->begin + s->length);
449 } else {
450 start = block->target_start;
451 size_in_block = block->target_end - block->target_start;
454 if (phys_addr >= start && phys_addr < start + size_in_block) {
455 *p_offset = phys_addr - start + offset;
457 /* The offset range mapped from the vmcore file must not spill over
458 * the GuestPhysBlock, clamp it. The rest of the mapping will be
459 * zero-filled in memory at load time; see
460 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
462 *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
463 mapping_length :
464 size_in_block - (phys_addr - start);
465 return;
468 offset += size_in_block;
472 static void write_elf_loads(DumpState *s, Error **errp)
474 hwaddr offset, filesz;
475 MemoryMapping *memory_mapping;
476 uint32_t phdr_index = 1;
477 uint32_t max_index;
478 Error *local_err = NULL;
480 if (s->have_section) {
481 max_index = s->sh_info;
482 } else {
483 max_index = s->phdr_num;
486 QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
487 get_offset_range(memory_mapping->phys_addr,
488 memory_mapping->length,
489 s, &offset, &filesz);
490 if (s->dump_info.d_class == ELFCLASS64) {
491 write_elf64_load(s, memory_mapping, phdr_index++, offset,
492 filesz, &local_err);
493 } else {
494 write_elf32_load(s, memory_mapping, phdr_index++, offset,
495 filesz, &local_err);
498 if (local_err) {
499 error_propagate(errp, local_err);
500 return;
503 if (phdr_index >= max_index) {
504 break;
509 /* write elf header, PT_NOTE and elf note to vmcore. */
510 static void dump_begin(DumpState *s, Error **errp)
512 Error *local_err = NULL;
515 * the vmcore's format is:
516 * --------------
517 * | elf header |
518 * --------------
519 * | PT_NOTE |
520 * --------------
521 * | PT_LOAD |
522 * --------------
523 * | ...... |
524 * --------------
525 * | PT_LOAD |
526 * --------------
527 * | sec_hdr |
528 * --------------
529 * | elf note |
530 * --------------
531 * | memory |
532 * --------------
534 * we only know where the memory is saved after we write elf note into
535 * vmcore.
538 /* write elf header to vmcore */
539 if (s->dump_info.d_class == ELFCLASS64) {
540 write_elf64_header(s, &local_err);
541 } else {
542 write_elf32_header(s, &local_err);
544 if (local_err) {
545 error_propagate(errp, local_err);
546 return;
549 if (s->dump_info.d_class == ELFCLASS64) {
550 /* write PT_NOTE to vmcore */
551 write_elf64_note(s, &local_err);
552 if (local_err) {
553 error_propagate(errp, local_err);
554 return;
557 /* write all PT_LOAD to vmcore */
558 write_elf_loads(s, &local_err);
559 if (local_err) {
560 error_propagate(errp, local_err);
561 return;
564 /* write section to vmcore */
565 if (s->have_section) {
566 write_elf_section(s, 1, &local_err);
567 if (local_err) {
568 error_propagate(errp, local_err);
569 return;
573 /* write notes to vmcore */
574 write_elf64_notes(fd_write_vmcore, s, &local_err);
575 if (local_err) {
576 error_propagate(errp, local_err);
577 return;
579 } else {
580 /* write PT_NOTE to vmcore */
581 write_elf32_note(s, &local_err);
582 if (local_err) {
583 error_propagate(errp, local_err);
584 return;
587 /* write all PT_LOAD to vmcore */
588 write_elf_loads(s, &local_err);
589 if (local_err) {
590 error_propagate(errp, local_err);
591 return;
594 /* write section to vmcore */
595 if (s->have_section) {
596 write_elf_section(s, 0, &local_err);
597 if (local_err) {
598 error_propagate(errp, local_err);
599 return;
603 /* write notes to vmcore */
604 write_elf32_notes(fd_write_vmcore, s, &local_err);
605 if (local_err) {
606 error_propagate(errp, local_err);
607 return;
612 static int get_next_block(DumpState *s, GuestPhysBlock *block)
614 while (1) {
615 block = QTAILQ_NEXT(block, next);
616 if (!block) {
617 /* no more block */
618 return 1;
621 s->start = 0;
622 s->next_block = block;
623 if (s->has_filter) {
624 if (block->target_start >= s->begin + s->length ||
625 block->target_end <= s->begin) {
626 /* This block is out of the range */
627 continue;
630 if (s->begin > block->target_start) {
631 s->start = s->begin - block->target_start;
635 return 0;
639 /* write all memory to vmcore */
640 static void dump_iterate(DumpState *s, Error **errp)
642 GuestPhysBlock *block;
643 int64_t size;
644 Error *local_err = NULL;
646 do {
647 block = s->next_block;
649 size = block->target_end - block->target_start;
650 if (s->has_filter) {
651 size -= s->start;
652 if (s->begin + s->length < block->target_end) {
653 size -= block->target_end - (s->begin + s->length);
656 write_memory(s, block, s->start, size, &local_err);
657 if (local_err) {
658 error_propagate(errp, local_err);
659 return;
662 } while (!get_next_block(s, block));
665 static void create_vmcore(DumpState *s, Error **errp)
667 Error *local_err = NULL;
669 dump_begin(s, &local_err);
670 if (local_err) {
671 error_propagate(errp, local_err);
672 return;
675 dump_iterate(s, errp);
678 static int write_start_flat_header(int fd)
680 MakedumpfileHeader *mh;
681 int ret = 0;
683 QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
684 mh = g_malloc0(MAX_SIZE_MDF_HEADER);
686 memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
687 MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
689 mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
690 mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
692 size_t written_size;
693 written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
694 if (written_size != MAX_SIZE_MDF_HEADER) {
695 ret = -1;
698 g_free(mh);
699 return ret;
702 static int write_end_flat_header(int fd)
704 MakedumpfileDataHeader mdh;
706 mdh.offset = END_FLAG_FLAT_HEADER;
707 mdh.buf_size = END_FLAG_FLAT_HEADER;
709 size_t written_size;
710 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
711 if (written_size != sizeof(mdh)) {
712 return -1;
715 return 0;
718 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
720 size_t written_size;
721 MakedumpfileDataHeader mdh;
723 mdh.offset = cpu_to_be64(offset);
724 mdh.buf_size = cpu_to_be64(size);
726 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
727 if (written_size != sizeof(mdh)) {
728 return -1;
731 written_size = qemu_write_full(fd, buf, size);
732 if (written_size != size) {
733 return -1;
736 return 0;
739 static int buf_write_note(const void *buf, size_t size, void *opaque)
741 DumpState *s = opaque;
743 /* note_buf is not enough */
744 if (s->note_buf_offset + size > s->note_size) {
745 return -1;
748 memcpy(s->note_buf + s->note_buf_offset, buf, size);
750 s->note_buf_offset += size;
752 return 0;
756 * This function retrieves various sizes from an elf header.
758 * @note has to be a valid ELF note. The return sizes are unmodified
759 * (not padded or rounded up to be multiple of 4).
761 static void get_note_sizes(DumpState *s, const void *note,
762 uint64_t *note_head_size,
763 uint64_t *name_size,
764 uint64_t *desc_size)
766 uint64_t note_head_sz;
767 uint64_t name_sz;
768 uint64_t desc_sz;
770 if (s->dump_info.d_class == ELFCLASS64) {
771 const Elf64_Nhdr *hdr = note;
772 note_head_sz = sizeof(Elf64_Nhdr);
773 name_sz = tswap64(hdr->n_namesz);
774 desc_sz = tswap64(hdr->n_descsz);
775 } else {
776 const Elf32_Nhdr *hdr = note;
777 note_head_sz = sizeof(Elf32_Nhdr);
778 name_sz = tswap32(hdr->n_namesz);
779 desc_sz = tswap32(hdr->n_descsz);
782 if (note_head_size) {
783 *note_head_size = note_head_sz;
785 if (name_size) {
786 *name_size = name_sz;
788 if (desc_size) {
789 *desc_size = desc_sz;
793 static bool note_name_equal(DumpState *s,
794 const uint8_t *note, const char *name)
796 int len = strlen(name) + 1;
797 uint64_t head_size, name_size;
799 get_note_sizes(s, note, &head_size, &name_size, NULL);
800 head_size = ROUND_UP(head_size, 4);
802 return name_size == len && memcmp(note + head_size, name, len) == 0;
805 /* write common header, sub header and elf note to vmcore */
806 static void create_header32(DumpState *s, Error **errp)
808 DiskDumpHeader32 *dh = NULL;
809 KdumpSubHeader32 *kh = NULL;
810 size_t size;
811 uint32_t block_size;
812 uint32_t sub_hdr_size;
813 uint32_t bitmap_blocks;
814 uint32_t status = 0;
815 uint64_t offset_note;
816 Error *local_err = NULL;
818 /* write common header, the version of kdump-compressed format is 6th */
819 size = sizeof(DiskDumpHeader32);
820 dh = g_malloc0(size);
822 memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
823 dh->header_version = cpu_to_dump32(s, 6);
824 block_size = s->dump_info.page_size;
825 dh->block_size = cpu_to_dump32(s, block_size);
826 sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
827 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
828 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
829 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
830 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
831 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
832 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
833 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
834 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
836 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
837 status |= DUMP_DH_COMPRESSED_ZLIB;
839 #ifdef CONFIG_LZO
840 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
841 status |= DUMP_DH_COMPRESSED_LZO;
843 #endif
844 #ifdef CONFIG_SNAPPY
845 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
846 status |= DUMP_DH_COMPRESSED_SNAPPY;
848 #endif
849 dh->status = cpu_to_dump32(s, status);
851 if (write_buffer(s->fd, 0, dh, size) < 0) {
852 error_setg(errp, "dump: failed to write disk dump header");
853 goto out;
856 /* write sub header */
857 size = sizeof(KdumpSubHeader32);
858 kh = g_malloc0(size);
860 /* 64bit max_mapnr_64 */
861 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
862 kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
863 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
865 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
866 if (s->guest_note &&
867 note_name_equal(s, s->guest_note, "VMCOREINFO")) {
868 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
870 get_note_sizes(s, s->guest_note,
871 &hsize, &name_size, &size_vmcoreinfo_desc);
872 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
873 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
874 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
875 kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
878 kh->offset_note = cpu_to_dump64(s, offset_note);
879 kh->note_size = cpu_to_dump32(s, s->note_size);
881 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
882 block_size, kh, size) < 0) {
883 error_setg(errp, "dump: failed to write kdump sub header");
884 goto out;
887 /* write note */
888 s->note_buf = g_malloc0(s->note_size);
889 s->note_buf_offset = 0;
891 /* use s->note_buf to store notes temporarily */
892 write_elf32_notes(buf_write_note, s, &local_err);
893 if (local_err) {
894 error_propagate(errp, local_err);
895 goto out;
897 if (write_buffer(s->fd, offset_note, s->note_buf,
898 s->note_size) < 0) {
899 error_setg(errp, "dump: failed to write notes");
900 goto out;
903 /* get offset of dump_bitmap */
904 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
905 block_size;
907 /* get offset of page */
908 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
909 block_size;
911 out:
912 g_free(dh);
913 g_free(kh);
914 g_free(s->note_buf);
917 /* write common header, sub header and elf note to vmcore */
918 static void create_header64(DumpState *s, Error **errp)
920 DiskDumpHeader64 *dh = NULL;
921 KdumpSubHeader64 *kh = NULL;
922 size_t size;
923 uint32_t block_size;
924 uint32_t sub_hdr_size;
925 uint32_t bitmap_blocks;
926 uint32_t status = 0;
927 uint64_t offset_note;
928 Error *local_err = NULL;
930 /* write common header, the version of kdump-compressed format is 6th */
931 size = sizeof(DiskDumpHeader64);
932 dh = g_malloc0(size);
934 memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
935 dh->header_version = cpu_to_dump32(s, 6);
936 block_size = s->dump_info.page_size;
937 dh->block_size = cpu_to_dump32(s, block_size);
938 sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
939 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
940 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
941 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
942 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
943 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
944 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
945 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
946 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
948 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
949 status |= DUMP_DH_COMPRESSED_ZLIB;
951 #ifdef CONFIG_LZO
952 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
953 status |= DUMP_DH_COMPRESSED_LZO;
955 #endif
956 #ifdef CONFIG_SNAPPY
957 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
958 status |= DUMP_DH_COMPRESSED_SNAPPY;
960 #endif
961 dh->status = cpu_to_dump32(s, status);
963 if (write_buffer(s->fd, 0, dh, size) < 0) {
964 error_setg(errp, "dump: failed to write disk dump header");
965 goto out;
968 /* write sub header */
969 size = sizeof(KdumpSubHeader64);
970 kh = g_malloc0(size);
972 /* 64bit max_mapnr_64 */
973 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
974 kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
975 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
977 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
978 if (s->guest_note &&
979 note_name_equal(s, s->guest_note, "VMCOREINFO")) {
980 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
982 get_note_sizes(s, s->guest_note,
983 &hsize, &name_size, &size_vmcoreinfo_desc);
984 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
985 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
986 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
987 kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
990 kh->offset_note = cpu_to_dump64(s, offset_note);
991 kh->note_size = cpu_to_dump64(s, s->note_size);
993 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
994 block_size, kh, size) < 0) {
995 error_setg(errp, "dump: failed to write kdump sub header");
996 goto out;
999 /* write note */
1000 s->note_buf = g_malloc0(s->note_size);
1001 s->note_buf_offset = 0;
1003 /* use s->note_buf to store notes temporarily */
1004 write_elf64_notes(buf_write_note, s, &local_err);
1005 if (local_err) {
1006 error_propagate(errp, local_err);
1007 goto out;
1010 if (write_buffer(s->fd, offset_note, s->note_buf,
1011 s->note_size) < 0) {
1012 error_setg(errp, "dump: failed to write notes");
1013 goto out;
1016 /* get offset of dump_bitmap */
1017 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1018 block_size;
1020 /* get offset of page */
1021 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1022 block_size;
1024 out:
1025 g_free(dh);
1026 g_free(kh);
1027 g_free(s->note_buf);
1030 static void write_dump_header(DumpState *s, Error **errp)
1032 Error *local_err = NULL;
1034 if (s->dump_info.d_class == ELFCLASS32) {
1035 create_header32(s, &local_err);
1036 } else {
1037 create_header64(s, &local_err);
1039 error_propagate(errp, local_err);
1042 static size_t dump_bitmap_get_bufsize(DumpState *s)
1044 return s->dump_info.page_size;
1048 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1049 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1050 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1051 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1052 * vmcore, ie. synchronizing un-sync bit into vmcore.
1054 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1055 uint8_t *buf, DumpState *s)
1057 off_t old_offset, new_offset;
1058 off_t offset_bitmap1, offset_bitmap2;
1059 uint32_t byte, bit;
1060 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1061 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1063 /* should not set the previous place */
1064 assert(last_pfn <= pfn);
1067 * if the bit needed to be set is not cached in buf, flush the data in buf
1068 * to vmcore firstly.
1069 * making new_offset be bigger than old_offset can also sync remained data
1070 * into vmcore.
1072 old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1073 new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1075 while (old_offset < new_offset) {
1076 /* calculate the offset and write dump_bitmap */
1077 offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1078 if (write_buffer(s->fd, offset_bitmap1, buf,
1079 bitmap_bufsize) < 0) {
1080 return -1;
1083 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1084 offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1085 old_offset;
1086 if (write_buffer(s->fd, offset_bitmap2, buf,
1087 bitmap_bufsize) < 0) {
1088 return -1;
1091 memset(buf, 0, bitmap_bufsize);
1092 old_offset += bitmap_bufsize;
1095 /* get the exact place of the bit in the buf, and set it */
1096 byte = (pfn % bits_per_buf) / CHAR_BIT;
1097 bit = (pfn % bits_per_buf) % CHAR_BIT;
1098 if (value) {
1099 buf[byte] |= 1u << bit;
1100 } else {
1101 buf[byte] &= ~(1u << bit);
1104 return 0;
1107 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1109 int target_page_shift = ctz32(s->dump_info.page_size);
1111 return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1114 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1116 int target_page_shift = ctz32(s->dump_info.page_size);
1118 return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1122 * exam every page and return the page frame number and the address of the page.
1123 * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1124 * blocks, so block->target_start and block->target_end should be interal
1125 * multiples of the target page size.
1127 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1128 uint8_t **bufptr, DumpState *s)
1130 GuestPhysBlock *block = *blockptr;
1131 hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1132 uint8_t *buf;
1134 /* block == NULL means the start of the iteration */
1135 if (!block) {
1136 block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1137 *blockptr = block;
1138 assert((block->target_start & ~target_page_mask) == 0);
1139 assert((block->target_end & ~target_page_mask) == 0);
1140 *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1141 if (bufptr) {
1142 *bufptr = block->host_addr;
1144 return true;
1147 *pfnptr = *pfnptr + 1;
1148 addr = dump_pfn_to_paddr(s, *pfnptr);
1150 if ((addr >= block->target_start) &&
1151 (addr + s->dump_info.page_size <= block->target_end)) {
1152 buf = block->host_addr + (addr - block->target_start);
1153 } else {
1154 /* the next page is in the next block */
1155 block = QTAILQ_NEXT(block, next);
1156 *blockptr = block;
1157 if (!block) {
1158 return false;
1160 assert((block->target_start & ~target_page_mask) == 0);
1161 assert((block->target_end & ~target_page_mask) == 0);
1162 *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1163 buf = block->host_addr;
1166 if (bufptr) {
1167 *bufptr = buf;
1170 return true;
1173 static void write_dump_bitmap(DumpState *s, Error **errp)
1175 int ret = 0;
1176 uint64_t last_pfn, pfn;
1177 void *dump_bitmap_buf;
1178 size_t num_dumpable;
1179 GuestPhysBlock *block_iter = NULL;
1180 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1181 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1183 /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1184 dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1186 num_dumpable = 0;
1187 last_pfn = 0;
1190 * exam memory page by page, and set the bit in dump_bitmap corresponded
1191 * to the existing page.
1193 while (get_next_page(&block_iter, &pfn, NULL, s)) {
1194 ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1195 if (ret < 0) {
1196 error_setg(errp, "dump: failed to set dump_bitmap");
1197 goto out;
1200 last_pfn = pfn;
1201 num_dumpable++;
1205 * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1206 * set the remaining bits from last_pfn to the end of the bitmap buffer to
1207 * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1209 if (num_dumpable > 0) {
1210 ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1211 dump_bitmap_buf, s);
1212 if (ret < 0) {
1213 error_setg(errp, "dump: failed to sync dump_bitmap");
1214 goto out;
1218 /* number of dumpable pages that will be dumped later */
1219 s->num_dumpable = num_dumpable;
1221 out:
1222 g_free(dump_bitmap_buf);
1225 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1226 off_t offset)
1228 data_cache->fd = s->fd;
1229 data_cache->data_size = 0;
1230 data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1231 data_cache->buf = g_malloc0(data_cache->buf_size);
1232 data_cache->offset = offset;
1235 static int write_cache(DataCache *dc, const void *buf, size_t size,
1236 bool flag_sync)
1239 * dc->buf_size should not be less than size, otherwise dc will never be
1240 * enough
1242 assert(size <= dc->buf_size);
1245 * if flag_sync is set, synchronize data in dc->buf into vmcore.
1246 * otherwise check if the space is enough for caching data in buf, if not,
1247 * write the data in dc->buf to dc->fd and reset dc->buf
1249 if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1250 (flag_sync && dc->data_size > 0)) {
1251 if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1252 return -1;
1255 dc->offset += dc->data_size;
1256 dc->data_size = 0;
1259 if (!flag_sync) {
1260 memcpy(dc->buf + dc->data_size, buf, size);
1261 dc->data_size += size;
1264 return 0;
1267 static void free_data_cache(DataCache *data_cache)
1269 g_free(data_cache->buf);
1272 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1274 switch (flag_compress) {
1275 case DUMP_DH_COMPRESSED_ZLIB:
1276 return compressBound(page_size);
1278 case DUMP_DH_COMPRESSED_LZO:
1280 * LZO will expand incompressible data by a little amount. Please check
1281 * the following URL to see the expansion calculation:
1282 * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1284 return page_size + page_size / 16 + 64 + 3;
1286 #ifdef CONFIG_SNAPPY
1287 case DUMP_DH_COMPRESSED_SNAPPY:
1288 return snappy_max_compressed_length(page_size);
1289 #endif
1291 return 0;
1295 * check if the page is all 0
1297 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1299 return buffer_is_zero(buf, page_size);
1302 static void write_dump_pages(DumpState *s, Error **errp)
1304 int ret = 0;
1305 DataCache page_desc, page_data;
1306 size_t len_buf_out, size_out;
1307 #ifdef CONFIG_LZO
1308 lzo_bytep wrkmem = NULL;
1309 #endif
1310 uint8_t *buf_out = NULL;
1311 off_t offset_desc, offset_data;
1312 PageDescriptor pd, pd_zero;
1313 uint8_t *buf;
1314 GuestPhysBlock *block_iter = NULL;
1315 uint64_t pfn_iter;
1317 /* get offset of page_desc and page_data in dump file */
1318 offset_desc = s->offset_page;
1319 offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1321 prepare_data_cache(&page_desc, s, offset_desc);
1322 prepare_data_cache(&page_data, s, offset_data);
1324 /* prepare buffer to store compressed data */
1325 len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1326 assert(len_buf_out != 0);
1328 #ifdef CONFIG_LZO
1329 wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1330 #endif
1332 buf_out = g_malloc(len_buf_out);
1335 * init zero page's page_desc and page_data, because every zero page
1336 * uses the same page_data
1338 pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1339 pd_zero.flags = cpu_to_dump32(s, 0);
1340 pd_zero.offset = cpu_to_dump64(s, offset_data);
1341 pd_zero.page_flags = cpu_to_dump64(s, 0);
1342 buf = g_malloc0(s->dump_info.page_size);
1343 ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1344 g_free(buf);
1345 if (ret < 0) {
1346 error_setg(errp, "dump: failed to write page data (zero page)");
1347 goto out;
1350 offset_data += s->dump_info.page_size;
1353 * dump memory to vmcore page by page. zero page will all be resided in the
1354 * first page of page section
1356 while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1357 /* check zero page */
1358 if (is_zero_page(buf, s->dump_info.page_size)) {
1359 ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1360 false);
1361 if (ret < 0) {
1362 error_setg(errp, "dump: failed to write page desc");
1363 goto out;
1365 } else {
1367 * not zero page, then:
1368 * 1. compress the page
1369 * 2. write the compressed page into the cache of page_data
1370 * 3. get page desc of the compressed page and write it into the
1371 * cache of page_desc
1373 * only one compression format will be used here, for
1374 * s->flag_compress is set. But when compression fails to work,
1375 * we fall back to save in plaintext.
1377 size_out = len_buf_out;
1378 if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1379 (compress2(buf_out, (uLongf *)&size_out, buf,
1380 s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1381 (size_out < s->dump_info.page_size)) {
1382 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1383 pd.size = cpu_to_dump32(s, size_out);
1385 ret = write_cache(&page_data, buf_out, size_out, false);
1386 if (ret < 0) {
1387 error_setg(errp, "dump: failed to write page data");
1388 goto out;
1390 #ifdef CONFIG_LZO
1391 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1392 (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1393 (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1394 (size_out < s->dump_info.page_size)) {
1395 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1396 pd.size = cpu_to_dump32(s, size_out);
1398 ret = write_cache(&page_data, buf_out, size_out, false);
1399 if (ret < 0) {
1400 error_setg(errp, "dump: failed to write page data");
1401 goto out;
1403 #endif
1404 #ifdef CONFIG_SNAPPY
1405 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1406 (snappy_compress((char *)buf, s->dump_info.page_size,
1407 (char *)buf_out, &size_out) == SNAPPY_OK) &&
1408 (size_out < s->dump_info.page_size)) {
1409 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1410 pd.size = cpu_to_dump32(s, size_out);
1412 ret = write_cache(&page_data, buf_out, size_out, false);
1413 if (ret < 0) {
1414 error_setg(errp, "dump: failed to write page data");
1415 goto out;
1417 #endif
1418 } else {
1420 * fall back to save in plaintext, size_out should be
1421 * assigned the target's page size
1423 pd.flags = cpu_to_dump32(s, 0);
1424 size_out = s->dump_info.page_size;
1425 pd.size = cpu_to_dump32(s, size_out);
1427 ret = write_cache(&page_data, buf,
1428 s->dump_info.page_size, false);
1429 if (ret < 0) {
1430 error_setg(errp, "dump: failed to write page data");
1431 goto out;
1435 /* get and write page desc here */
1436 pd.page_flags = cpu_to_dump64(s, 0);
1437 pd.offset = cpu_to_dump64(s, offset_data);
1438 offset_data += size_out;
1440 ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1441 if (ret < 0) {
1442 error_setg(errp, "dump: failed to write page desc");
1443 goto out;
1446 s->written_size += s->dump_info.page_size;
1449 ret = write_cache(&page_desc, NULL, 0, true);
1450 if (ret < 0) {
1451 error_setg(errp, "dump: failed to sync cache for page_desc");
1452 goto out;
1454 ret = write_cache(&page_data, NULL, 0, true);
1455 if (ret < 0) {
1456 error_setg(errp, "dump: failed to sync cache for page_data");
1457 goto out;
1460 out:
1461 free_data_cache(&page_desc);
1462 free_data_cache(&page_data);
1464 #ifdef CONFIG_LZO
1465 g_free(wrkmem);
1466 #endif
1468 g_free(buf_out);
1471 static void create_kdump_vmcore(DumpState *s, Error **errp)
1473 int ret;
1474 Error *local_err = NULL;
1477 * the kdump-compressed format is:
1478 * File offset
1479 * +------------------------------------------+ 0x0
1480 * | main header (struct disk_dump_header) |
1481 * |------------------------------------------+ block 1
1482 * | sub header (struct kdump_sub_header) |
1483 * |------------------------------------------+ block 2
1484 * | 1st-dump_bitmap |
1485 * |------------------------------------------+ block 2 + X blocks
1486 * | 2nd-dump_bitmap | (aligned by block)
1487 * |------------------------------------------+ block 2 + 2 * X blocks
1488 * | page desc for pfn 0 (struct page_desc) | (aligned by block)
1489 * | page desc for pfn 1 (struct page_desc) |
1490 * | : |
1491 * |------------------------------------------| (not aligned by block)
1492 * | page data (pfn 0) |
1493 * | page data (pfn 1) |
1494 * | : |
1495 * +------------------------------------------+
1498 ret = write_start_flat_header(s->fd);
1499 if (ret < 0) {
1500 error_setg(errp, "dump: failed to write start flat header");
1501 return;
1504 write_dump_header(s, &local_err);
1505 if (local_err) {
1506 error_propagate(errp, local_err);
1507 return;
1510 write_dump_bitmap(s, &local_err);
1511 if (local_err) {
1512 error_propagate(errp, local_err);
1513 return;
1516 write_dump_pages(s, &local_err);
1517 if (local_err) {
1518 error_propagate(errp, local_err);
1519 return;
1522 ret = write_end_flat_header(s->fd);
1523 if (ret < 0) {
1524 error_setg(errp, "dump: failed to write end flat header");
1525 return;
1529 static ram_addr_t get_start_block(DumpState *s)
1531 GuestPhysBlock *block;
1533 if (!s->has_filter) {
1534 s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1535 return 0;
1538 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1539 if (block->target_start >= s->begin + s->length ||
1540 block->target_end <= s->begin) {
1541 /* This block is out of the range */
1542 continue;
1545 s->next_block = block;
1546 if (s->begin > block->target_start) {
1547 s->start = s->begin - block->target_start;
1548 } else {
1549 s->start = 0;
1551 return s->start;
1554 return -1;
1557 static void get_max_mapnr(DumpState *s)
1559 GuestPhysBlock *last_block;
1561 last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1562 s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1565 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1567 static void dump_state_prepare(DumpState *s)
1569 /* zero the struct, setting status to active */
1570 *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1573 bool dump_in_progress(void)
1575 DumpState *state = &dump_state_global;
1576 return (atomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1579 /* calculate total size of memory to be dumped (taking filter into
1580 * acoount.) */
1581 static int64_t dump_calculate_size(DumpState *s)
1583 GuestPhysBlock *block;
1584 int64_t size = 0, total = 0, left = 0, right = 0;
1586 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1587 if (s->has_filter) {
1588 /* calculate the overlapped region. */
1589 left = MAX(s->begin, block->target_start);
1590 right = MIN(s->begin + s->length, block->target_end);
1591 size = right - left;
1592 size = size > 0 ? size : 0;
1593 } else {
1594 /* count the whole region in */
1595 size = (block->target_end - block->target_start);
1597 total += size;
1600 return total;
1603 static void vmcoreinfo_update_phys_base(DumpState *s)
1605 uint64_t size, note_head_size, name_size, phys_base;
1606 char **lines;
1607 uint8_t *vmci;
1608 size_t i;
1610 if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1611 return;
1614 get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1615 note_head_size = ROUND_UP(note_head_size, 4);
1617 vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1618 *(vmci + size) = '\0';
1620 lines = g_strsplit((char *)vmci, "\n", -1);
1621 for (i = 0; lines[i]; i++) {
1622 const char *prefix = NULL;
1624 if (s->dump_info.d_machine == EM_X86_64) {
1625 prefix = "NUMBER(phys_base)=";
1626 } else if (s->dump_info.d_machine == EM_AARCH64) {
1627 prefix = "NUMBER(PHYS_OFFSET)=";
1630 if (prefix && g_str_has_prefix(lines[i], prefix)) {
1631 if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1632 &phys_base) < 0) {
1633 warn_report("Failed to read %s", prefix);
1634 } else {
1635 s->dump_info.phys_base = phys_base;
1637 break;
1641 g_strfreev(lines);
1644 static void dump_init(DumpState *s, int fd, bool has_format,
1645 DumpGuestMemoryFormat format, bool paging, bool has_filter,
1646 int64_t begin, int64_t length, Error **errp)
1648 VMCoreInfoState *vmci = vmcoreinfo_find();
1649 CPUState *cpu;
1650 int nr_cpus;
1651 Error *err = NULL;
1652 int ret;
1654 s->has_format = has_format;
1655 s->format = format;
1656 s->written_size = 0;
1658 /* kdump-compressed is conflict with paging and filter */
1659 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1660 assert(!paging && !has_filter);
1663 if (runstate_is_running()) {
1664 vm_stop(RUN_STATE_SAVE_VM);
1665 s->resume = true;
1666 } else {
1667 s->resume = false;
1670 /* If we use KVM, we should synchronize the registers before we get dump
1671 * info or physmap info.
1673 cpu_synchronize_all_states();
1674 nr_cpus = 0;
1675 CPU_FOREACH(cpu) {
1676 nr_cpus++;
1679 s->fd = fd;
1680 s->has_filter = has_filter;
1681 s->begin = begin;
1682 s->length = length;
1684 memory_mapping_list_init(&s->list);
1686 guest_phys_blocks_init(&s->guest_phys_blocks);
1687 guest_phys_blocks_append(&s->guest_phys_blocks);
1688 s->total_size = dump_calculate_size(s);
1689 #ifdef DEBUG_DUMP_GUEST_MEMORY
1690 fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1691 #endif
1693 /* it does not make sense to dump non-existent memory */
1694 if (!s->total_size) {
1695 error_setg(errp, "dump: no guest memory to dump");
1696 goto cleanup;
1699 s->start = get_start_block(s);
1700 if (s->start == -1) {
1701 error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1702 goto cleanup;
1705 /* get dump info: endian, class and architecture.
1706 * If the target architecture is not supported, cpu_get_dump_info() will
1707 * return -1.
1709 ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1710 if (ret < 0) {
1711 error_setg(errp, QERR_UNSUPPORTED);
1712 goto cleanup;
1715 if (!s->dump_info.page_size) {
1716 s->dump_info.page_size = TARGET_PAGE_SIZE;
1719 s->note_size = cpu_get_note_size(s->dump_info.d_class,
1720 s->dump_info.d_machine, nr_cpus);
1721 if (s->note_size < 0) {
1722 error_setg(errp, QERR_UNSUPPORTED);
1723 goto cleanup;
1727 * The goal of this block is to (a) update the previously guessed
1728 * phys_base, (b) copy the guest note out of the guest.
1729 * Failure to do so is not fatal for dumping.
1731 if (vmci) {
1732 uint64_t addr, note_head_size, name_size, desc_size;
1733 uint32_t size;
1734 uint16_t format;
1736 note_head_size = s->dump_info.d_class == ELFCLASS32 ?
1737 sizeof(Elf32_Nhdr) : sizeof(Elf64_Nhdr);
1739 format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1740 size = le32_to_cpu(vmci->vmcoreinfo.size);
1741 addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1742 if (!vmci->has_vmcoreinfo) {
1743 warn_report("guest note is not present");
1744 } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1745 warn_report("guest note size is invalid: %" PRIu32, size);
1746 } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1747 warn_report("guest note format is unsupported: %" PRIu16, format);
1748 } else {
1749 s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1750 cpu_physical_memory_read(addr, s->guest_note, size);
1752 get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1753 s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1754 desc_size);
1755 if (name_size > MAX_GUEST_NOTE_SIZE ||
1756 desc_size > MAX_GUEST_NOTE_SIZE ||
1757 s->guest_note_size > size) {
1758 warn_report("Invalid guest note header");
1759 g_free(s->guest_note);
1760 s->guest_note = NULL;
1761 } else {
1762 vmcoreinfo_update_phys_base(s);
1763 s->note_size += s->guest_note_size;
1768 /* get memory mapping */
1769 if (paging) {
1770 qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1771 if (err != NULL) {
1772 error_propagate(errp, err);
1773 goto cleanup;
1775 } else {
1776 qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1779 s->nr_cpus = nr_cpus;
1781 get_max_mapnr(s);
1783 uint64_t tmp;
1784 tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1785 s->dump_info.page_size);
1786 s->len_dump_bitmap = tmp * s->dump_info.page_size;
1788 /* init for kdump-compressed format */
1789 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1790 switch (format) {
1791 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1792 s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1793 break;
1795 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1796 #ifdef CONFIG_LZO
1797 if (lzo_init() != LZO_E_OK) {
1798 error_setg(errp, "failed to initialize the LZO library");
1799 goto cleanup;
1801 #endif
1802 s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1803 break;
1805 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1806 s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1807 break;
1809 default:
1810 s->flag_compress = 0;
1813 return;
1816 if (s->has_filter) {
1817 memory_mapping_filter(&s->list, s->begin, s->length);
1821 * calculate phdr_num
1823 * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1825 s->phdr_num = 1; /* PT_NOTE */
1826 if (s->list.num < UINT16_MAX - 2) {
1827 s->phdr_num += s->list.num;
1828 s->have_section = false;
1829 } else {
1830 s->have_section = true;
1831 s->phdr_num = PN_XNUM;
1832 s->sh_info = 1; /* PT_NOTE */
1834 /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1835 if (s->list.num <= UINT32_MAX - 1) {
1836 s->sh_info += s->list.num;
1837 } else {
1838 s->sh_info = UINT32_MAX;
1842 if (s->dump_info.d_class == ELFCLASS64) {
1843 if (s->have_section) {
1844 s->memory_offset = sizeof(Elf64_Ehdr) +
1845 sizeof(Elf64_Phdr) * s->sh_info +
1846 sizeof(Elf64_Shdr) + s->note_size;
1847 } else {
1848 s->memory_offset = sizeof(Elf64_Ehdr) +
1849 sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1851 } else {
1852 if (s->have_section) {
1853 s->memory_offset = sizeof(Elf32_Ehdr) +
1854 sizeof(Elf32_Phdr) * s->sh_info +
1855 sizeof(Elf32_Shdr) + s->note_size;
1856 } else {
1857 s->memory_offset = sizeof(Elf32_Ehdr) +
1858 sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1862 return;
1864 cleanup:
1865 dump_cleanup(s);
1868 /* this operation might be time consuming. */
1869 static void dump_process(DumpState *s, Error **errp)
1871 Error *local_err = NULL;
1872 DumpQueryResult *result = NULL;
1874 if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1875 #ifdef TARGET_X86_64
1876 create_win_dump(s, &local_err);
1877 #endif
1878 } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1879 create_kdump_vmcore(s, &local_err);
1880 } else {
1881 create_vmcore(s, &local_err);
1884 /* make sure status is written after written_size updates */
1885 smp_wmb();
1886 atomic_set(&s->status,
1887 (local_err ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1889 /* send DUMP_COMPLETED message (unconditionally) */
1890 result = qmp_query_dump(NULL);
1891 /* should never fail */
1892 assert(result);
1893 qapi_event_send_dump_completed(result, !!local_err, (local_err ? \
1894 error_get_pretty(local_err) : NULL));
1895 qapi_free_DumpQueryResult(result);
1897 error_propagate(errp, local_err);
1898 dump_cleanup(s);
1901 static void *dump_thread(void *data)
1903 DumpState *s = (DumpState *)data;
1904 dump_process(s, NULL);
1905 return NULL;
1908 DumpQueryResult *qmp_query_dump(Error **errp)
1910 DumpQueryResult *result = g_new(DumpQueryResult, 1);
1911 DumpState *state = &dump_state_global;
1912 result->status = atomic_read(&state->status);
1913 /* make sure we are reading status and written_size in order */
1914 smp_rmb();
1915 result->completed = state->written_size;
1916 result->total = state->total_size;
1917 return result;
1920 void qmp_dump_guest_memory(bool paging, const char *file,
1921 bool has_detach, bool detach,
1922 bool has_begin, int64_t begin, bool has_length,
1923 int64_t length, bool has_format,
1924 DumpGuestMemoryFormat format, Error **errp)
1926 const char *p;
1927 int fd = -1;
1928 DumpState *s;
1929 Error *local_err = NULL;
1930 bool detach_p = false;
1932 if (runstate_check(RUN_STATE_INMIGRATE)) {
1933 error_setg(errp, "Dump not allowed during incoming migration.");
1934 return;
1937 /* if there is a dump in background, we should wait until the dump
1938 * finished */
1939 if (dump_in_progress()) {
1940 error_setg(errp, "There is a dump in process, please wait.");
1941 return;
1945 * kdump-compressed format need the whole memory dumped, so paging or
1946 * filter is not supported here.
1948 if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1949 (paging || has_begin || has_length)) {
1950 error_setg(errp, "kdump-compressed format doesn't support paging or "
1951 "filter");
1952 return;
1954 if (has_begin && !has_length) {
1955 error_setg(errp, QERR_MISSING_PARAMETER, "length");
1956 return;
1958 if (!has_begin && has_length) {
1959 error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1960 return;
1962 if (has_detach) {
1963 detach_p = detach;
1966 /* check whether lzo/snappy is supported */
1967 #ifndef CONFIG_LZO
1968 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1969 error_setg(errp, "kdump-lzo is not available now");
1970 return;
1972 #endif
1974 #ifndef CONFIG_SNAPPY
1975 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1976 error_setg(errp, "kdump-snappy is not available now");
1977 return;
1979 #endif
1981 #ifndef TARGET_X86_64
1982 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1983 error_setg(errp, "Windows dump is only available for x86-64");
1984 return;
1986 #endif
1988 #if !defined(WIN32)
1989 if (strstart(file, "fd:", &p)) {
1990 fd = monitor_get_fd(cur_mon, p, errp);
1991 if (fd == -1) {
1992 return;
1995 #endif
1997 if (strstart(file, "file:", &p)) {
1998 fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1999 if (fd < 0) {
2000 error_setg_file_open(errp, errno, p);
2001 return;
2005 if (fd == -1) {
2006 error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
2007 return;
2010 s = &dump_state_global;
2011 dump_state_prepare(s);
2013 dump_init(s, fd, has_format, format, paging, has_begin,
2014 begin, length, &local_err);
2015 if (local_err) {
2016 error_propagate(errp, local_err);
2017 atomic_set(&s->status, DUMP_STATUS_FAILED);
2018 return;
2021 if (detach_p) {
2022 /* detached dump */
2023 s->detached = true;
2024 qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2025 s, QEMU_THREAD_DETACHED);
2026 } else {
2027 /* sync dump */
2028 dump_process(s, errp);
2032 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2034 DumpGuestMemoryFormatList *item;
2035 DumpGuestMemoryCapability *cap =
2036 g_malloc0(sizeof(DumpGuestMemoryCapability));
2038 /* elf is always available */
2039 item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2040 cap->formats = item;
2041 item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
2043 /* kdump-zlib is always available */
2044 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2045 item = item->next;
2046 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
2048 /* add new item if kdump-lzo is available */
2049 #ifdef CONFIG_LZO
2050 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2051 item = item->next;
2052 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
2053 #endif
2055 /* add new item if kdump-snappy is available */
2056 #ifdef CONFIG_SNAPPY
2057 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2058 item = item->next;
2059 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
2060 #endif
2062 /* Windows dump is available only if target is x86_64 */
2063 #ifdef TARGET_X86_64
2064 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2065 item = item->next;
2066 item->value = DUMP_GUEST_MEMORY_FORMAT_WIN_DMP;
2067 #endif
2069 return cap;