qemu: avoid memory leak while remove disk
[qemu/kevin.git] / dump.c
blobef1d8025c9dc0e2fc155545d3b033ac7304df132
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);
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 = cpu_to_dump32(s, memory_mapping->virt_addr);
221 assert(memory_mapping->length >= filesz);
223 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
224 if (ret < 0) {
225 error_setg_errno(errp, -ret,
226 "dump: failed to write program header table");
230 static void write_elf64_note(DumpState *s, Error **errp)
232 Elf64_Phdr phdr;
233 hwaddr begin = s->memory_offset - s->note_size;
234 int ret;
236 memset(&phdr, 0, sizeof(Elf64_Phdr));
237 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
238 phdr.p_offset = cpu_to_dump64(s, begin);
239 phdr.p_paddr = 0;
240 phdr.p_filesz = cpu_to_dump64(s, s->note_size);
241 phdr.p_memsz = cpu_to_dump64(s, s->note_size);
242 phdr.p_vaddr = 0;
244 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
245 if (ret < 0) {
246 error_setg_errno(errp, -ret,
247 "dump: failed to write program header table");
251 static inline int cpu_index(CPUState *cpu)
253 return cpu->cpu_index + 1;
256 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
257 Error **errp)
259 int ret;
261 if (s->guest_note) {
262 ret = f(s->guest_note, s->guest_note_size, s);
263 if (ret < 0) {
264 error_setg(errp, "dump: failed to write guest note");
269 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
270 Error **errp)
272 CPUState *cpu;
273 int ret;
274 int id;
276 CPU_FOREACH(cpu) {
277 id = cpu_index(cpu);
278 ret = cpu_write_elf64_note(f, cpu, id, s);
279 if (ret < 0) {
280 error_setg(errp, "dump: failed to write elf notes");
281 return;
285 CPU_FOREACH(cpu) {
286 ret = cpu_write_elf64_qemunote(f, cpu, s);
287 if (ret < 0) {
288 error_setg(errp, "dump: failed to write CPU status");
289 return;
293 write_guest_note(f, s, errp);
296 static void write_elf32_note(DumpState *s, Error **errp)
298 hwaddr begin = s->memory_offset - s->note_size;
299 Elf32_Phdr phdr;
300 int ret;
302 memset(&phdr, 0, sizeof(Elf32_Phdr));
303 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
304 phdr.p_offset = cpu_to_dump32(s, begin);
305 phdr.p_paddr = 0;
306 phdr.p_filesz = cpu_to_dump32(s, s->note_size);
307 phdr.p_memsz = cpu_to_dump32(s, s->note_size);
308 phdr.p_vaddr = 0;
310 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
311 if (ret < 0) {
312 error_setg_errno(errp, -ret,
313 "dump: failed to write program header table");
317 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
318 Error **errp)
320 CPUState *cpu;
321 int ret;
322 int id;
324 CPU_FOREACH(cpu) {
325 id = cpu_index(cpu);
326 ret = cpu_write_elf32_note(f, cpu, id, s);
327 if (ret < 0) {
328 error_setg(errp, "dump: failed to write elf notes");
329 return;
333 CPU_FOREACH(cpu) {
334 ret = cpu_write_elf32_qemunote(f, cpu, s);
335 if (ret < 0) {
336 error_setg(errp, "dump: failed to write CPU status");
337 return;
341 write_guest_note(f, s, errp);
344 static void write_elf_section(DumpState *s, int type, Error **errp)
346 Elf32_Shdr shdr32;
347 Elf64_Shdr shdr64;
348 int shdr_size;
349 void *shdr;
350 int ret;
352 if (type == 0) {
353 shdr_size = sizeof(Elf32_Shdr);
354 memset(&shdr32, 0, shdr_size);
355 shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
356 shdr = &shdr32;
357 } else {
358 shdr_size = sizeof(Elf64_Shdr);
359 memset(&shdr64, 0, shdr_size);
360 shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
361 shdr = &shdr64;
364 ret = fd_write_vmcore(&shdr, shdr_size, s);
365 if (ret < 0) {
366 error_setg_errno(errp, -ret,
367 "dump: failed to write section header table");
371 static void write_data(DumpState *s, void *buf, int length, Error **errp)
373 int ret;
375 ret = fd_write_vmcore(buf, length, s);
376 if (ret < 0) {
377 error_setg_errno(errp, -ret, "dump: failed to save memory");
378 } else {
379 s->written_size += length;
383 /* write the memory to vmcore. 1 page per I/O. */
384 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
385 int64_t size, Error **errp)
387 int64_t i;
388 Error *local_err = NULL;
390 for (i = 0; i < size / s->dump_info.page_size; i++) {
391 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
392 s->dump_info.page_size, &local_err);
393 if (local_err) {
394 error_propagate(errp, local_err);
395 return;
399 if ((size % s->dump_info.page_size) != 0) {
400 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
401 size % s->dump_info.page_size, &local_err);
402 if (local_err) {
403 error_propagate(errp, local_err);
404 return;
409 /* get the memory's offset and size in the vmcore */
410 static void get_offset_range(hwaddr phys_addr,
411 ram_addr_t mapping_length,
412 DumpState *s,
413 hwaddr *p_offset,
414 hwaddr *p_filesz)
416 GuestPhysBlock *block;
417 hwaddr offset = s->memory_offset;
418 int64_t size_in_block, start;
420 /* When the memory is not stored into vmcore, offset will be -1 */
421 *p_offset = -1;
422 *p_filesz = 0;
424 if (s->has_filter) {
425 if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
426 return;
430 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
431 if (s->has_filter) {
432 if (block->target_start >= s->begin + s->length ||
433 block->target_end <= s->begin) {
434 /* This block is out of the range */
435 continue;
438 if (s->begin <= block->target_start) {
439 start = block->target_start;
440 } else {
441 start = s->begin;
444 size_in_block = block->target_end - start;
445 if (s->begin + s->length < block->target_end) {
446 size_in_block -= block->target_end - (s->begin + s->length);
448 } else {
449 start = block->target_start;
450 size_in_block = block->target_end - block->target_start;
453 if (phys_addr >= start && phys_addr < start + size_in_block) {
454 *p_offset = phys_addr - start + offset;
456 /* The offset range mapped from the vmcore file must not spill over
457 * the GuestPhysBlock, clamp it. The rest of the mapping will be
458 * zero-filled in memory at load time; see
459 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
461 *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
462 mapping_length :
463 size_in_block - (phys_addr - start);
464 return;
467 offset += size_in_block;
471 static void write_elf_loads(DumpState *s, Error **errp)
473 hwaddr offset, filesz;
474 MemoryMapping *memory_mapping;
475 uint32_t phdr_index = 1;
476 uint32_t max_index;
477 Error *local_err = NULL;
479 if (s->have_section) {
480 max_index = s->sh_info;
481 } else {
482 max_index = s->phdr_num;
485 QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
486 get_offset_range(memory_mapping->phys_addr,
487 memory_mapping->length,
488 s, &offset, &filesz);
489 if (s->dump_info.d_class == ELFCLASS64) {
490 write_elf64_load(s, memory_mapping, phdr_index++, offset,
491 filesz, &local_err);
492 } else {
493 write_elf32_load(s, memory_mapping, phdr_index++, offset,
494 filesz, &local_err);
497 if (local_err) {
498 error_propagate(errp, local_err);
499 return;
502 if (phdr_index >= max_index) {
503 break;
508 /* write elf header, PT_NOTE and elf note to vmcore. */
509 static void dump_begin(DumpState *s, Error **errp)
511 Error *local_err = NULL;
514 * the vmcore's format is:
515 * --------------
516 * | elf header |
517 * --------------
518 * | PT_NOTE |
519 * --------------
520 * | PT_LOAD |
521 * --------------
522 * | ...... |
523 * --------------
524 * | PT_LOAD |
525 * --------------
526 * | sec_hdr |
527 * --------------
528 * | elf note |
529 * --------------
530 * | memory |
531 * --------------
533 * we only know where the memory is saved after we write elf note into
534 * vmcore.
537 /* write elf header to vmcore */
538 if (s->dump_info.d_class == ELFCLASS64) {
539 write_elf64_header(s, &local_err);
540 } else {
541 write_elf32_header(s, &local_err);
543 if (local_err) {
544 error_propagate(errp, local_err);
545 return;
548 if (s->dump_info.d_class == ELFCLASS64) {
549 /* write PT_NOTE to vmcore */
550 write_elf64_note(s, &local_err);
551 if (local_err) {
552 error_propagate(errp, local_err);
553 return;
556 /* write all PT_LOAD to vmcore */
557 write_elf_loads(s, &local_err);
558 if (local_err) {
559 error_propagate(errp, local_err);
560 return;
563 /* write section to vmcore */
564 if (s->have_section) {
565 write_elf_section(s, 1, &local_err);
566 if (local_err) {
567 error_propagate(errp, local_err);
568 return;
572 /* write notes to vmcore */
573 write_elf64_notes(fd_write_vmcore, s, &local_err);
574 if (local_err) {
575 error_propagate(errp, local_err);
576 return;
578 } else {
579 /* write PT_NOTE to vmcore */
580 write_elf32_note(s, &local_err);
581 if (local_err) {
582 error_propagate(errp, local_err);
583 return;
586 /* write all PT_LOAD to vmcore */
587 write_elf_loads(s, &local_err);
588 if (local_err) {
589 error_propagate(errp, local_err);
590 return;
593 /* write section to vmcore */
594 if (s->have_section) {
595 write_elf_section(s, 0, &local_err);
596 if (local_err) {
597 error_propagate(errp, local_err);
598 return;
602 /* write notes to vmcore */
603 write_elf32_notes(fd_write_vmcore, s, &local_err);
604 if (local_err) {
605 error_propagate(errp, local_err);
606 return;
611 static int get_next_block(DumpState *s, GuestPhysBlock *block)
613 while (1) {
614 block = QTAILQ_NEXT(block, next);
615 if (!block) {
616 /* no more block */
617 return 1;
620 s->start = 0;
621 s->next_block = block;
622 if (s->has_filter) {
623 if (block->target_start >= s->begin + s->length ||
624 block->target_end <= s->begin) {
625 /* This block is out of the range */
626 continue;
629 if (s->begin > block->target_start) {
630 s->start = s->begin - block->target_start;
634 return 0;
638 /* write all memory to vmcore */
639 static void dump_iterate(DumpState *s, Error **errp)
641 GuestPhysBlock *block;
642 int64_t size;
643 Error *local_err = NULL;
645 do {
646 block = s->next_block;
648 size = block->target_end - block->target_start;
649 if (s->has_filter) {
650 size -= s->start;
651 if (s->begin + s->length < block->target_end) {
652 size -= block->target_end - (s->begin + s->length);
655 write_memory(s, block, s->start, size, &local_err);
656 if (local_err) {
657 error_propagate(errp, local_err);
658 return;
661 } while (!get_next_block(s, block));
664 static void create_vmcore(DumpState *s, Error **errp)
666 Error *local_err = NULL;
668 dump_begin(s, &local_err);
669 if (local_err) {
670 error_propagate(errp, local_err);
671 return;
674 dump_iterate(s, errp);
677 static int write_start_flat_header(int fd)
679 MakedumpfileHeader *mh;
680 int ret = 0;
682 QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
683 mh = g_malloc0(MAX_SIZE_MDF_HEADER);
685 memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
686 MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
688 mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
689 mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
691 size_t written_size;
692 written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
693 if (written_size != MAX_SIZE_MDF_HEADER) {
694 ret = -1;
697 g_free(mh);
698 return ret;
701 static int write_end_flat_header(int fd)
703 MakedumpfileDataHeader mdh;
705 mdh.offset = END_FLAG_FLAT_HEADER;
706 mdh.buf_size = END_FLAG_FLAT_HEADER;
708 size_t written_size;
709 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
710 if (written_size != sizeof(mdh)) {
711 return -1;
714 return 0;
717 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
719 size_t written_size;
720 MakedumpfileDataHeader mdh;
722 mdh.offset = cpu_to_be64(offset);
723 mdh.buf_size = cpu_to_be64(size);
725 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
726 if (written_size != sizeof(mdh)) {
727 return -1;
730 written_size = qemu_write_full(fd, buf, size);
731 if (written_size != size) {
732 return -1;
735 return 0;
738 static int buf_write_note(const void *buf, size_t size, void *opaque)
740 DumpState *s = opaque;
742 /* note_buf is not enough */
743 if (s->note_buf_offset + size > s->note_size) {
744 return -1;
747 memcpy(s->note_buf + s->note_buf_offset, buf, size);
749 s->note_buf_offset += size;
751 return 0;
755 * This function retrieves various sizes from an elf header.
757 * @note has to be a valid ELF note. The return sizes are unmodified
758 * (not padded or rounded up to be multiple of 4).
760 static void get_note_sizes(DumpState *s, const void *note,
761 uint64_t *note_head_size,
762 uint64_t *name_size,
763 uint64_t *desc_size)
765 uint64_t note_head_sz;
766 uint64_t name_sz;
767 uint64_t desc_sz;
769 if (s->dump_info.d_class == ELFCLASS64) {
770 const Elf64_Nhdr *hdr = note;
771 note_head_sz = sizeof(Elf64_Nhdr);
772 name_sz = tswap64(hdr->n_namesz);
773 desc_sz = tswap64(hdr->n_descsz);
774 } else {
775 const Elf32_Nhdr *hdr = note;
776 note_head_sz = sizeof(Elf32_Nhdr);
777 name_sz = tswap32(hdr->n_namesz);
778 desc_sz = tswap32(hdr->n_descsz);
781 if (note_head_size) {
782 *note_head_size = note_head_sz;
784 if (name_size) {
785 *name_size = name_sz;
787 if (desc_size) {
788 *desc_size = desc_sz;
792 static bool note_name_equal(DumpState *s,
793 const uint8_t *note, const char *name)
795 int len = strlen(name) + 1;
796 uint64_t head_size, name_size;
798 get_note_sizes(s, note, &head_size, &name_size, NULL);
799 head_size = ROUND_UP(head_size, 4);
801 return name_size == len && memcmp(note + head_size, name, len) == 0;
804 /* write common header, sub header and elf note to vmcore */
805 static void create_header32(DumpState *s, Error **errp)
807 DiskDumpHeader32 *dh = NULL;
808 KdumpSubHeader32 *kh = NULL;
809 size_t size;
810 uint32_t block_size;
811 uint32_t sub_hdr_size;
812 uint32_t bitmap_blocks;
813 uint32_t status = 0;
814 uint64_t offset_note;
815 Error *local_err = NULL;
817 /* write common header, the version of kdump-compressed format is 6th */
818 size = sizeof(DiskDumpHeader32);
819 dh = g_malloc0(size);
821 memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
822 dh->header_version = cpu_to_dump32(s, 6);
823 block_size = s->dump_info.page_size;
824 dh->block_size = cpu_to_dump32(s, block_size);
825 sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
826 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
827 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
828 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
829 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
830 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
831 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
832 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
833 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
835 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
836 status |= DUMP_DH_COMPRESSED_ZLIB;
838 #ifdef CONFIG_LZO
839 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
840 status |= DUMP_DH_COMPRESSED_LZO;
842 #endif
843 #ifdef CONFIG_SNAPPY
844 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
845 status |= DUMP_DH_COMPRESSED_SNAPPY;
847 #endif
848 dh->status = cpu_to_dump32(s, status);
850 if (write_buffer(s->fd, 0, dh, size) < 0) {
851 error_setg(errp, "dump: failed to write disk dump header");
852 goto out;
855 /* write sub header */
856 size = sizeof(KdumpSubHeader32);
857 kh = g_malloc0(size);
859 /* 64bit max_mapnr_64 */
860 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
861 kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
862 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
864 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
865 if (s->guest_note &&
866 note_name_equal(s, s->guest_note, "VMCOREINFO")) {
867 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
869 get_note_sizes(s, s->guest_note,
870 &hsize, &name_size, &size_vmcoreinfo_desc);
871 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
872 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
873 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
874 kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
877 kh->offset_note = cpu_to_dump64(s, offset_note);
878 kh->note_size = cpu_to_dump32(s, s->note_size);
880 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
881 block_size, kh, size) < 0) {
882 error_setg(errp, "dump: failed to write kdump sub header");
883 goto out;
886 /* write note */
887 s->note_buf = g_malloc0(s->note_size);
888 s->note_buf_offset = 0;
890 /* use s->note_buf to store notes temporarily */
891 write_elf32_notes(buf_write_note, s, &local_err);
892 if (local_err) {
893 error_propagate(errp, local_err);
894 goto out;
896 if (write_buffer(s->fd, offset_note, s->note_buf,
897 s->note_size) < 0) {
898 error_setg(errp, "dump: failed to write notes");
899 goto out;
902 /* get offset of dump_bitmap */
903 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
904 block_size;
906 /* get offset of page */
907 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
908 block_size;
910 out:
911 g_free(dh);
912 g_free(kh);
913 g_free(s->note_buf);
916 /* write common header, sub header and elf note to vmcore */
917 static void create_header64(DumpState *s, Error **errp)
919 DiskDumpHeader64 *dh = NULL;
920 KdumpSubHeader64 *kh = NULL;
921 size_t size;
922 uint32_t block_size;
923 uint32_t sub_hdr_size;
924 uint32_t bitmap_blocks;
925 uint32_t status = 0;
926 uint64_t offset_note;
927 Error *local_err = NULL;
929 /* write common header, the version of kdump-compressed format is 6th */
930 size = sizeof(DiskDumpHeader64);
931 dh = g_malloc0(size);
933 memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
934 dh->header_version = cpu_to_dump32(s, 6);
935 block_size = s->dump_info.page_size;
936 dh->block_size = cpu_to_dump32(s, block_size);
937 sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
938 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
939 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
940 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
941 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
942 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
943 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
944 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
945 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
947 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
948 status |= DUMP_DH_COMPRESSED_ZLIB;
950 #ifdef CONFIG_LZO
951 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
952 status |= DUMP_DH_COMPRESSED_LZO;
954 #endif
955 #ifdef CONFIG_SNAPPY
956 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
957 status |= DUMP_DH_COMPRESSED_SNAPPY;
959 #endif
960 dh->status = cpu_to_dump32(s, status);
962 if (write_buffer(s->fd, 0, dh, size) < 0) {
963 error_setg(errp, "dump: failed to write disk dump header");
964 goto out;
967 /* write sub header */
968 size = sizeof(KdumpSubHeader64);
969 kh = g_malloc0(size);
971 /* 64bit max_mapnr_64 */
972 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
973 kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
974 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
976 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
977 if (s->guest_note &&
978 note_name_equal(s, s->guest_note, "VMCOREINFO")) {
979 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
981 get_note_sizes(s, s->guest_note,
982 &hsize, &name_size, &size_vmcoreinfo_desc);
983 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
984 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
985 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
986 kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
989 kh->offset_note = cpu_to_dump64(s, offset_note);
990 kh->note_size = cpu_to_dump64(s, s->note_size);
992 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
993 block_size, kh, size) < 0) {
994 error_setg(errp, "dump: failed to write kdump sub header");
995 goto out;
998 /* write note */
999 s->note_buf = g_malloc0(s->note_size);
1000 s->note_buf_offset = 0;
1002 /* use s->note_buf to store notes temporarily */
1003 write_elf64_notes(buf_write_note, s, &local_err);
1004 if (local_err) {
1005 error_propagate(errp, local_err);
1006 goto out;
1009 if (write_buffer(s->fd, offset_note, s->note_buf,
1010 s->note_size) < 0) {
1011 error_setg(errp, "dump: failed to write notes");
1012 goto out;
1015 /* get offset of dump_bitmap */
1016 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1017 block_size;
1019 /* get offset of page */
1020 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1021 block_size;
1023 out:
1024 g_free(dh);
1025 g_free(kh);
1026 g_free(s->note_buf);
1029 static void write_dump_header(DumpState *s, Error **errp)
1031 Error *local_err = NULL;
1033 if (s->dump_info.d_class == ELFCLASS32) {
1034 create_header32(s, &local_err);
1035 } else {
1036 create_header64(s, &local_err);
1038 error_propagate(errp, local_err);
1041 static size_t dump_bitmap_get_bufsize(DumpState *s)
1043 return s->dump_info.page_size;
1047 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1048 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1049 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1050 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1051 * vmcore, ie. synchronizing un-sync bit into vmcore.
1053 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1054 uint8_t *buf, DumpState *s)
1056 off_t old_offset, new_offset;
1057 off_t offset_bitmap1, offset_bitmap2;
1058 uint32_t byte, bit;
1059 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1060 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1062 /* should not set the previous place */
1063 assert(last_pfn <= pfn);
1066 * if the bit needed to be set is not cached in buf, flush the data in buf
1067 * to vmcore firstly.
1068 * making new_offset be bigger than old_offset can also sync remained data
1069 * into vmcore.
1071 old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1072 new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1074 while (old_offset < new_offset) {
1075 /* calculate the offset and write dump_bitmap */
1076 offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1077 if (write_buffer(s->fd, offset_bitmap1, buf,
1078 bitmap_bufsize) < 0) {
1079 return -1;
1082 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1083 offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1084 old_offset;
1085 if (write_buffer(s->fd, offset_bitmap2, buf,
1086 bitmap_bufsize) < 0) {
1087 return -1;
1090 memset(buf, 0, bitmap_bufsize);
1091 old_offset += bitmap_bufsize;
1094 /* get the exact place of the bit in the buf, and set it */
1095 byte = (pfn % bits_per_buf) / CHAR_BIT;
1096 bit = (pfn % bits_per_buf) % CHAR_BIT;
1097 if (value) {
1098 buf[byte] |= 1u << bit;
1099 } else {
1100 buf[byte] &= ~(1u << bit);
1103 return 0;
1106 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1108 int target_page_shift = ctz32(s->dump_info.page_size);
1110 return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1113 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1115 int target_page_shift = ctz32(s->dump_info.page_size);
1117 return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1121 * exam every page and return the page frame number and the address of the page.
1122 * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1123 * blocks, so block->target_start and block->target_end should be interal
1124 * multiples of the target page size.
1126 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1127 uint8_t **bufptr, DumpState *s)
1129 GuestPhysBlock *block = *blockptr;
1130 hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1131 uint8_t *buf;
1133 /* block == NULL means the start of the iteration */
1134 if (!block) {
1135 block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1136 *blockptr = block;
1137 assert((block->target_start & ~target_page_mask) == 0);
1138 assert((block->target_end & ~target_page_mask) == 0);
1139 *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1140 if (bufptr) {
1141 *bufptr = block->host_addr;
1143 return true;
1146 *pfnptr = *pfnptr + 1;
1147 addr = dump_pfn_to_paddr(s, *pfnptr);
1149 if ((addr >= block->target_start) &&
1150 (addr + s->dump_info.page_size <= block->target_end)) {
1151 buf = block->host_addr + (addr - block->target_start);
1152 } else {
1153 /* the next page is in the next block */
1154 block = QTAILQ_NEXT(block, next);
1155 *blockptr = block;
1156 if (!block) {
1157 return false;
1159 assert((block->target_start & ~target_page_mask) == 0);
1160 assert((block->target_end & ~target_page_mask) == 0);
1161 *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1162 buf = block->host_addr;
1165 if (bufptr) {
1166 *bufptr = buf;
1169 return true;
1172 static void write_dump_bitmap(DumpState *s, Error **errp)
1174 int ret = 0;
1175 uint64_t last_pfn, pfn;
1176 void *dump_bitmap_buf;
1177 size_t num_dumpable;
1178 GuestPhysBlock *block_iter = NULL;
1179 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1180 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1182 /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1183 dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1185 num_dumpable = 0;
1186 last_pfn = 0;
1189 * exam memory page by page, and set the bit in dump_bitmap corresponded
1190 * to the existing page.
1192 while (get_next_page(&block_iter, &pfn, NULL, s)) {
1193 ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1194 if (ret < 0) {
1195 error_setg(errp, "dump: failed to set dump_bitmap");
1196 goto out;
1199 last_pfn = pfn;
1200 num_dumpable++;
1204 * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1205 * set the remaining bits from last_pfn to the end of the bitmap buffer to
1206 * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1208 if (num_dumpable > 0) {
1209 ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1210 dump_bitmap_buf, s);
1211 if (ret < 0) {
1212 error_setg(errp, "dump: failed to sync dump_bitmap");
1213 goto out;
1217 /* number of dumpable pages that will be dumped later */
1218 s->num_dumpable = num_dumpable;
1220 out:
1221 g_free(dump_bitmap_buf);
1224 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1225 off_t offset)
1227 data_cache->fd = s->fd;
1228 data_cache->data_size = 0;
1229 data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1230 data_cache->buf = g_malloc0(data_cache->buf_size);
1231 data_cache->offset = offset;
1234 static int write_cache(DataCache *dc, const void *buf, size_t size,
1235 bool flag_sync)
1238 * dc->buf_size should not be less than size, otherwise dc will never be
1239 * enough
1241 assert(size <= dc->buf_size);
1244 * if flag_sync is set, synchronize data in dc->buf into vmcore.
1245 * otherwise check if the space is enough for caching data in buf, if not,
1246 * write the data in dc->buf to dc->fd and reset dc->buf
1248 if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1249 (flag_sync && dc->data_size > 0)) {
1250 if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1251 return -1;
1254 dc->offset += dc->data_size;
1255 dc->data_size = 0;
1258 if (!flag_sync) {
1259 memcpy(dc->buf + dc->data_size, buf, size);
1260 dc->data_size += size;
1263 return 0;
1266 static void free_data_cache(DataCache *data_cache)
1268 g_free(data_cache->buf);
1271 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1273 switch (flag_compress) {
1274 case DUMP_DH_COMPRESSED_ZLIB:
1275 return compressBound(page_size);
1277 case DUMP_DH_COMPRESSED_LZO:
1279 * LZO will expand incompressible data by a little amount. Please check
1280 * the following URL to see the expansion calculation:
1281 * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1283 return page_size + page_size / 16 + 64 + 3;
1285 #ifdef CONFIG_SNAPPY
1286 case DUMP_DH_COMPRESSED_SNAPPY:
1287 return snappy_max_compressed_length(page_size);
1288 #endif
1290 return 0;
1294 * check if the page is all 0
1296 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1298 return buffer_is_zero(buf, page_size);
1301 static void write_dump_pages(DumpState *s, Error **errp)
1303 int ret = 0;
1304 DataCache page_desc, page_data;
1305 size_t len_buf_out, size_out;
1306 #ifdef CONFIG_LZO
1307 lzo_bytep wrkmem = NULL;
1308 #endif
1309 uint8_t *buf_out = NULL;
1310 off_t offset_desc, offset_data;
1311 PageDescriptor pd, pd_zero;
1312 uint8_t *buf;
1313 GuestPhysBlock *block_iter = NULL;
1314 uint64_t pfn_iter;
1316 /* get offset of page_desc and page_data in dump file */
1317 offset_desc = s->offset_page;
1318 offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1320 prepare_data_cache(&page_desc, s, offset_desc);
1321 prepare_data_cache(&page_data, s, offset_data);
1323 /* prepare buffer to store compressed data */
1324 len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1325 assert(len_buf_out != 0);
1327 #ifdef CONFIG_LZO
1328 wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1329 #endif
1331 buf_out = g_malloc(len_buf_out);
1334 * init zero page's page_desc and page_data, because every zero page
1335 * uses the same page_data
1337 pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1338 pd_zero.flags = cpu_to_dump32(s, 0);
1339 pd_zero.offset = cpu_to_dump64(s, offset_data);
1340 pd_zero.page_flags = cpu_to_dump64(s, 0);
1341 buf = g_malloc0(s->dump_info.page_size);
1342 ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1343 g_free(buf);
1344 if (ret < 0) {
1345 error_setg(errp, "dump: failed to write page data (zero page)");
1346 goto out;
1349 offset_data += s->dump_info.page_size;
1352 * dump memory to vmcore page by page. zero page will all be resided in the
1353 * first page of page section
1355 while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1356 /* check zero page */
1357 if (is_zero_page(buf, s->dump_info.page_size)) {
1358 ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1359 false);
1360 if (ret < 0) {
1361 error_setg(errp, "dump: failed to write page desc");
1362 goto out;
1364 } else {
1366 * not zero page, then:
1367 * 1. compress the page
1368 * 2. write the compressed page into the cache of page_data
1369 * 3. get page desc of the compressed page and write it into the
1370 * cache of page_desc
1372 * only one compression format will be used here, for
1373 * s->flag_compress is set. But when compression fails to work,
1374 * we fall back to save in plaintext.
1376 size_out = len_buf_out;
1377 if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1378 (compress2(buf_out, (uLongf *)&size_out, buf,
1379 s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1380 (size_out < s->dump_info.page_size)) {
1381 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1382 pd.size = cpu_to_dump32(s, size_out);
1384 ret = write_cache(&page_data, buf_out, size_out, false);
1385 if (ret < 0) {
1386 error_setg(errp, "dump: failed to write page data");
1387 goto out;
1389 #ifdef CONFIG_LZO
1390 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1391 (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1392 (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1393 (size_out < s->dump_info.page_size)) {
1394 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1395 pd.size = cpu_to_dump32(s, size_out);
1397 ret = write_cache(&page_data, buf_out, size_out, false);
1398 if (ret < 0) {
1399 error_setg(errp, "dump: failed to write page data");
1400 goto out;
1402 #endif
1403 #ifdef CONFIG_SNAPPY
1404 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1405 (snappy_compress((char *)buf, s->dump_info.page_size,
1406 (char *)buf_out, &size_out) == SNAPPY_OK) &&
1407 (size_out < s->dump_info.page_size)) {
1408 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1409 pd.size = cpu_to_dump32(s, size_out);
1411 ret = write_cache(&page_data, buf_out, size_out, false);
1412 if (ret < 0) {
1413 error_setg(errp, "dump: failed to write page data");
1414 goto out;
1416 #endif
1417 } else {
1419 * fall back to save in plaintext, size_out should be
1420 * assigned the target's page size
1422 pd.flags = cpu_to_dump32(s, 0);
1423 size_out = s->dump_info.page_size;
1424 pd.size = cpu_to_dump32(s, size_out);
1426 ret = write_cache(&page_data, buf,
1427 s->dump_info.page_size, false);
1428 if (ret < 0) {
1429 error_setg(errp, "dump: failed to write page data");
1430 goto out;
1434 /* get and write page desc here */
1435 pd.page_flags = cpu_to_dump64(s, 0);
1436 pd.offset = cpu_to_dump64(s, offset_data);
1437 offset_data += size_out;
1439 ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1440 if (ret < 0) {
1441 error_setg(errp, "dump: failed to write page desc");
1442 goto out;
1445 s->written_size += s->dump_info.page_size;
1448 ret = write_cache(&page_desc, NULL, 0, true);
1449 if (ret < 0) {
1450 error_setg(errp, "dump: failed to sync cache for page_desc");
1451 goto out;
1453 ret = write_cache(&page_data, NULL, 0, true);
1454 if (ret < 0) {
1455 error_setg(errp, "dump: failed to sync cache for page_data");
1456 goto out;
1459 out:
1460 free_data_cache(&page_desc);
1461 free_data_cache(&page_data);
1463 #ifdef CONFIG_LZO
1464 g_free(wrkmem);
1465 #endif
1467 g_free(buf_out);
1470 static void create_kdump_vmcore(DumpState *s, Error **errp)
1472 int ret;
1473 Error *local_err = NULL;
1476 * the kdump-compressed format is:
1477 * File offset
1478 * +------------------------------------------+ 0x0
1479 * | main header (struct disk_dump_header) |
1480 * |------------------------------------------+ block 1
1481 * | sub header (struct kdump_sub_header) |
1482 * |------------------------------------------+ block 2
1483 * | 1st-dump_bitmap |
1484 * |------------------------------------------+ block 2 + X blocks
1485 * | 2nd-dump_bitmap | (aligned by block)
1486 * |------------------------------------------+ block 2 + 2 * X blocks
1487 * | page desc for pfn 0 (struct page_desc) | (aligned by block)
1488 * | page desc for pfn 1 (struct page_desc) |
1489 * | : |
1490 * |------------------------------------------| (not aligned by block)
1491 * | page data (pfn 0) |
1492 * | page data (pfn 1) |
1493 * | : |
1494 * +------------------------------------------+
1497 ret = write_start_flat_header(s->fd);
1498 if (ret < 0) {
1499 error_setg(errp, "dump: failed to write start flat header");
1500 return;
1503 write_dump_header(s, &local_err);
1504 if (local_err) {
1505 error_propagate(errp, local_err);
1506 return;
1509 write_dump_bitmap(s, &local_err);
1510 if (local_err) {
1511 error_propagate(errp, local_err);
1512 return;
1515 write_dump_pages(s, &local_err);
1516 if (local_err) {
1517 error_propagate(errp, local_err);
1518 return;
1521 ret = write_end_flat_header(s->fd);
1522 if (ret < 0) {
1523 error_setg(errp, "dump: failed to write end flat header");
1524 return;
1528 static ram_addr_t get_start_block(DumpState *s)
1530 GuestPhysBlock *block;
1532 if (!s->has_filter) {
1533 s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1534 return 0;
1537 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1538 if (block->target_start >= s->begin + s->length ||
1539 block->target_end <= s->begin) {
1540 /* This block is out of the range */
1541 continue;
1544 s->next_block = block;
1545 if (s->begin > block->target_start) {
1546 s->start = s->begin - block->target_start;
1547 } else {
1548 s->start = 0;
1550 return s->start;
1553 return -1;
1556 static void get_max_mapnr(DumpState *s)
1558 GuestPhysBlock *last_block;
1560 last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1561 s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1564 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1566 static void dump_state_prepare(DumpState *s)
1568 /* zero the struct, setting status to active */
1569 *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1572 bool dump_in_progress(void)
1574 DumpState *state = &dump_state_global;
1575 return (atomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1578 /* calculate total size of memory to be dumped (taking filter into
1579 * acoount.) */
1580 static int64_t dump_calculate_size(DumpState *s)
1582 GuestPhysBlock *block;
1583 int64_t size = 0, total = 0, left = 0, right = 0;
1585 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1586 if (s->has_filter) {
1587 /* calculate the overlapped region. */
1588 left = MAX(s->begin, block->target_start);
1589 right = MIN(s->begin + s->length, block->target_end);
1590 size = right - left;
1591 size = size > 0 ? size : 0;
1592 } else {
1593 /* count the whole region in */
1594 size = (block->target_end - block->target_start);
1596 total += size;
1599 return total;
1602 static void vmcoreinfo_update_phys_base(DumpState *s)
1604 uint64_t size, note_head_size, name_size, phys_base;
1605 char **lines;
1606 uint8_t *vmci;
1607 size_t i;
1609 if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1610 return;
1613 get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1614 note_head_size = ROUND_UP(note_head_size, 4);
1616 vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1617 *(vmci + size) = '\0';
1619 lines = g_strsplit((char *)vmci, "\n", -1);
1620 for (i = 0; lines[i]; i++) {
1621 const char *prefix = NULL;
1623 if (s->dump_info.d_machine == EM_X86_64) {
1624 prefix = "NUMBER(phys_base)=";
1625 } else if (s->dump_info.d_machine == EM_AARCH64) {
1626 prefix = "NUMBER(PHYS_OFFSET)=";
1629 if (prefix && g_str_has_prefix(lines[i], prefix)) {
1630 if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1631 &phys_base) < 0) {
1632 warn_report("Failed to read %s", prefix);
1633 } else {
1634 s->dump_info.phys_base = phys_base;
1636 break;
1640 g_strfreev(lines);
1643 static void dump_init(DumpState *s, int fd, bool has_format,
1644 DumpGuestMemoryFormat format, bool paging, bool has_filter,
1645 int64_t begin, int64_t length, Error **errp)
1647 VMCoreInfoState *vmci = vmcoreinfo_find();
1648 CPUState *cpu;
1649 int nr_cpus;
1650 Error *err = NULL;
1651 int ret;
1653 s->has_format = has_format;
1654 s->format = format;
1655 s->written_size = 0;
1657 /* kdump-compressed is conflict with paging and filter */
1658 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1659 assert(!paging && !has_filter);
1662 if (runstate_is_running()) {
1663 vm_stop(RUN_STATE_SAVE_VM);
1664 s->resume = true;
1665 } else {
1666 s->resume = false;
1669 /* If we use KVM, we should synchronize the registers before we get dump
1670 * info or physmap info.
1672 cpu_synchronize_all_states();
1673 nr_cpus = 0;
1674 CPU_FOREACH(cpu) {
1675 nr_cpus++;
1678 s->fd = fd;
1679 s->has_filter = has_filter;
1680 s->begin = begin;
1681 s->length = length;
1683 memory_mapping_list_init(&s->list);
1685 guest_phys_blocks_init(&s->guest_phys_blocks);
1686 guest_phys_blocks_append(&s->guest_phys_blocks);
1687 s->total_size = dump_calculate_size(s);
1688 #ifdef DEBUG_DUMP_GUEST_MEMORY
1689 fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1690 #endif
1692 /* it does not make sense to dump non-existent memory */
1693 if (!s->total_size) {
1694 error_setg(errp, "dump: no guest memory to dump");
1695 goto cleanup;
1698 s->start = get_start_block(s);
1699 if (s->start == -1) {
1700 error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1701 goto cleanup;
1704 /* get dump info: endian, class and architecture.
1705 * If the target architecture is not supported, cpu_get_dump_info() will
1706 * return -1.
1708 ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1709 if (ret < 0) {
1710 error_setg(errp, QERR_UNSUPPORTED);
1711 goto cleanup;
1714 if (!s->dump_info.page_size) {
1715 s->dump_info.page_size = TARGET_PAGE_SIZE;
1718 s->note_size = cpu_get_note_size(s->dump_info.d_class,
1719 s->dump_info.d_machine, nr_cpus);
1720 if (s->note_size < 0) {
1721 error_setg(errp, QERR_UNSUPPORTED);
1722 goto cleanup;
1726 * The goal of this block is to (a) update the previously guessed
1727 * phys_base, (b) copy the guest note out of the guest.
1728 * Failure to do so is not fatal for dumping.
1730 if (vmci) {
1731 uint64_t addr, note_head_size, name_size, desc_size;
1732 uint32_t size;
1733 uint16_t format;
1735 note_head_size = s->dump_info.d_class == ELFCLASS32 ?
1736 sizeof(Elf32_Nhdr) : sizeof(Elf64_Nhdr);
1738 format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1739 size = le32_to_cpu(vmci->vmcoreinfo.size);
1740 addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1741 if (!vmci->has_vmcoreinfo) {
1742 warn_report("guest note is not present");
1743 } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1744 warn_report("guest note size is invalid: %" PRIu32, size);
1745 } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1746 warn_report("guest note format is unsupported: %" PRIu16, format);
1747 } else {
1748 s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1749 cpu_physical_memory_read(addr, s->guest_note, size);
1751 get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1752 s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1753 desc_size);
1754 if (name_size > MAX_GUEST_NOTE_SIZE ||
1755 desc_size > MAX_GUEST_NOTE_SIZE ||
1756 s->guest_note_size > size) {
1757 warn_report("Invalid guest note header");
1758 g_free(s->guest_note);
1759 s->guest_note = NULL;
1760 } else {
1761 vmcoreinfo_update_phys_base(s);
1762 s->note_size += s->guest_note_size;
1767 /* get memory mapping */
1768 if (paging) {
1769 qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1770 if (err != NULL) {
1771 error_propagate(errp, err);
1772 goto cleanup;
1774 } else {
1775 qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1778 s->nr_cpus = nr_cpus;
1780 get_max_mapnr(s);
1782 uint64_t tmp;
1783 tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1784 s->dump_info.page_size);
1785 s->len_dump_bitmap = tmp * s->dump_info.page_size;
1787 /* init for kdump-compressed format */
1788 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1789 switch (format) {
1790 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1791 s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1792 break;
1794 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1795 #ifdef CONFIG_LZO
1796 if (lzo_init() != LZO_E_OK) {
1797 error_setg(errp, "failed to initialize the LZO library");
1798 goto cleanup;
1800 #endif
1801 s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1802 break;
1804 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1805 s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1806 break;
1808 default:
1809 s->flag_compress = 0;
1812 return;
1815 if (s->has_filter) {
1816 memory_mapping_filter(&s->list, s->begin, s->length);
1820 * calculate phdr_num
1822 * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1824 s->phdr_num = 1; /* PT_NOTE */
1825 if (s->list.num < UINT16_MAX - 2) {
1826 s->phdr_num += s->list.num;
1827 s->have_section = false;
1828 } else {
1829 s->have_section = true;
1830 s->phdr_num = PN_XNUM;
1831 s->sh_info = 1; /* PT_NOTE */
1833 /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1834 if (s->list.num <= UINT32_MAX - 1) {
1835 s->sh_info += s->list.num;
1836 } else {
1837 s->sh_info = UINT32_MAX;
1841 if (s->dump_info.d_class == ELFCLASS64) {
1842 if (s->have_section) {
1843 s->memory_offset = sizeof(Elf64_Ehdr) +
1844 sizeof(Elf64_Phdr) * s->sh_info +
1845 sizeof(Elf64_Shdr) + s->note_size;
1846 } else {
1847 s->memory_offset = sizeof(Elf64_Ehdr) +
1848 sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1850 } else {
1851 if (s->have_section) {
1852 s->memory_offset = sizeof(Elf32_Ehdr) +
1853 sizeof(Elf32_Phdr) * s->sh_info +
1854 sizeof(Elf32_Shdr) + s->note_size;
1855 } else {
1856 s->memory_offset = sizeof(Elf32_Ehdr) +
1857 sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1861 return;
1863 cleanup:
1864 dump_cleanup(s);
1867 /* this operation might be time consuming. */
1868 static void dump_process(DumpState *s, Error **errp)
1870 Error *local_err = NULL;
1871 DumpQueryResult *result = NULL;
1873 if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1874 #ifdef TARGET_X86_64
1875 create_win_dump(s, &local_err);
1876 #endif
1877 } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1878 create_kdump_vmcore(s, &local_err);
1879 } else {
1880 create_vmcore(s, &local_err);
1883 /* make sure status is written after written_size updates */
1884 smp_wmb();
1885 atomic_set(&s->status,
1886 (local_err ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1888 /* send DUMP_COMPLETED message (unconditionally) */
1889 result = qmp_query_dump(NULL);
1890 /* should never fail */
1891 assert(result);
1892 qapi_event_send_dump_completed(result, !!local_err, (local_err ? \
1893 error_get_pretty(local_err) : NULL));
1894 qapi_free_DumpQueryResult(result);
1896 error_propagate(errp, local_err);
1897 dump_cleanup(s);
1900 static void *dump_thread(void *data)
1902 DumpState *s = (DumpState *)data;
1903 dump_process(s, NULL);
1904 return NULL;
1907 DumpQueryResult *qmp_query_dump(Error **errp)
1909 DumpQueryResult *result = g_new(DumpQueryResult, 1);
1910 DumpState *state = &dump_state_global;
1911 result->status = atomic_read(&state->status);
1912 /* make sure we are reading status and written_size in order */
1913 smp_rmb();
1914 result->completed = state->written_size;
1915 result->total = state->total_size;
1916 return result;
1919 void qmp_dump_guest_memory(bool paging, const char *file,
1920 bool has_detach, bool detach,
1921 bool has_begin, int64_t begin, bool has_length,
1922 int64_t length, bool has_format,
1923 DumpGuestMemoryFormat format, Error **errp)
1925 const char *p;
1926 int fd = -1;
1927 DumpState *s;
1928 Error *local_err = NULL;
1929 bool detach_p = false;
1931 if (runstate_check(RUN_STATE_INMIGRATE)) {
1932 error_setg(errp, "Dump not allowed during incoming migration.");
1933 return;
1936 /* if there is a dump in background, we should wait until the dump
1937 * finished */
1938 if (dump_in_progress()) {
1939 error_setg(errp, "There is a dump in process, please wait.");
1940 return;
1944 * kdump-compressed format need the whole memory dumped, so paging or
1945 * filter is not supported here.
1947 if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1948 (paging || has_begin || has_length)) {
1949 error_setg(errp, "kdump-compressed format doesn't support paging or "
1950 "filter");
1951 return;
1953 if (has_begin && !has_length) {
1954 error_setg(errp, QERR_MISSING_PARAMETER, "length");
1955 return;
1957 if (!has_begin && has_length) {
1958 error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1959 return;
1961 if (has_detach) {
1962 detach_p = detach;
1965 /* check whether lzo/snappy is supported */
1966 #ifndef CONFIG_LZO
1967 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1968 error_setg(errp, "kdump-lzo is not available now");
1969 return;
1971 #endif
1973 #ifndef CONFIG_SNAPPY
1974 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1975 error_setg(errp, "kdump-snappy is not available now");
1976 return;
1978 #endif
1980 #ifndef TARGET_X86_64
1981 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1982 error_setg(errp, "Windows dump is only available for x86-64");
1983 return;
1985 #endif
1987 #if !defined(WIN32)
1988 if (strstart(file, "fd:", &p)) {
1989 fd = monitor_get_fd(cur_mon, p, errp);
1990 if (fd == -1) {
1991 return;
1994 #endif
1996 if (strstart(file, "file:", &p)) {
1997 fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1998 if (fd < 0) {
1999 error_setg_file_open(errp, errno, p);
2000 return;
2004 if (fd == -1) {
2005 error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
2006 return;
2009 s = &dump_state_global;
2010 dump_state_prepare(s);
2012 dump_init(s, fd, has_format, format, paging, has_begin,
2013 begin, length, &local_err);
2014 if (local_err) {
2015 error_propagate(errp, local_err);
2016 atomic_set(&s->status, DUMP_STATUS_FAILED);
2017 return;
2020 if (detach_p) {
2021 /* detached dump */
2022 s->detached = true;
2023 qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2024 s, QEMU_THREAD_DETACHED);
2025 } else {
2026 /* sync dump */
2027 dump_process(s, errp);
2031 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2033 DumpGuestMemoryFormatList *item;
2034 DumpGuestMemoryCapability *cap =
2035 g_malloc0(sizeof(DumpGuestMemoryCapability));
2037 /* elf is always available */
2038 item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2039 cap->formats = item;
2040 item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
2042 /* kdump-zlib is always available */
2043 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2044 item = item->next;
2045 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
2047 /* add new item if kdump-lzo is available */
2048 #ifdef CONFIG_LZO
2049 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2050 item = item->next;
2051 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
2052 #endif
2054 /* add new item if kdump-snappy is available */
2055 #ifdef CONFIG_SNAPPY
2056 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2057 item = item->next;
2058 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
2059 #endif
2061 /* Windows dump is available only if target is x86_64 */
2062 #ifdef TARGET_X86_64
2063 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2064 item = item->next;
2065 item->value = DUMP_GUEST_MEMORY_FORMAT_WIN_DMP;
2066 #endif
2068 return cap;