tests: Consolidate accelerators declaration
[qemu/ar7.git] / dump.c
blobe9dfed060a5cc477d146f2324747e1188cf08133
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/qmp/qerror.h"
26 #include "qmp-commands.h"
27 #include "qapi-event.h"
28 #include "qemu/error-report.h"
29 #include "hw/misc/vmcoreinfo.h"
31 #include <zlib.h>
32 #ifdef CONFIG_LZO
33 #include <lzo/lzo1x.h>
34 #endif
35 #ifdef CONFIG_SNAPPY
36 #include <snappy-c.h>
37 #endif
38 #ifndef ELF_MACHINE_UNAME
39 #define ELF_MACHINE_UNAME "Unknown"
40 #endif
42 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
44 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size) \
45 ((DIV_ROUND_UP((hdr_size), 4) + \
46 DIV_ROUND_UP((name_size), 4) + \
47 DIV_ROUND_UP((desc_size), 4)) * 4)
49 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
51 if (s->dump_info.d_endian == ELFDATA2LSB) {
52 val = cpu_to_le16(val);
53 } else {
54 val = cpu_to_be16(val);
57 return val;
60 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
62 if (s->dump_info.d_endian == ELFDATA2LSB) {
63 val = cpu_to_le32(val);
64 } else {
65 val = cpu_to_be32(val);
68 return val;
71 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
73 if (s->dump_info.d_endian == ELFDATA2LSB) {
74 val = cpu_to_le64(val);
75 } else {
76 val = cpu_to_be64(val);
79 return val;
82 static int dump_cleanup(DumpState *s)
84 guest_phys_blocks_free(&s->guest_phys_blocks);
85 memory_mapping_list_free(&s->list);
86 close(s->fd);
87 g_free(s->guest_note);
88 s->guest_note = NULL;
89 if (s->resume) {
90 if (s->detached) {
91 qemu_mutex_lock_iothread();
93 vm_start();
94 if (s->detached) {
95 qemu_mutex_unlock_iothread();
99 return 0;
102 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
104 DumpState *s = opaque;
105 size_t written_size;
107 written_size = qemu_write_full(s->fd, buf, size);
108 if (written_size != size) {
109 return -1;
112 return 0;
115 static void write_elf64_header(DumpState *s, Error **errp)
117 Elf64_Ehdr elf_header;
118 int ret;
120 memset(&elf_header, 0, sizeof(Elf64_Ehdr));
121 memcpy(&elf_header, ELFMAG, SELFMAG);
122 elf_header.e_ident[EI_CLASS] = ELFCLASS64;
123 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
124 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
125 elf_header.e_type = cpu_to_dump16(s, ET_CORE);
126 elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
127 elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
128 elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
129 elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
130 elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
131 elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
132 if (s->have_section) {
133 uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
135 elf_header.e_shoff = cpu_to_dump64(s, shoff);
136 elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
137 elf_header.e_shnum = cpu_to_dump16(s, 1);
140 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
141 if (ret < 0) {
142 error_setg(errp, "dump: failed to write elf header");
146 static void write_elf32_header(DumpState *s, Error **errp)
148 Elf32_Ehdr elf_header;
149 int ret;
151 memset(&elf_header, 0, sizeof(Elf32_Ehdr));
152 memcpy(&elf_header, ELFMAG, SELFMAG);
153 elf_header.e_ident[EI_CLASS] = ELFCLASS32;
154 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
155 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
156 elf_header.e_type = cpu_to_dump16(s, ET_CORE);
157 elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
158 elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
159 elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
160 elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
161 elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
162 elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
163 if (s->have_section) {
164 uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
166 elf_header.e_shoff = cpu_to_dump32(s, shoff);
167 elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
168 elf_header.e_shnum = cpu_to_dump16(s, 1);
171 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
172 if (ret < 0) {
173 error_setg(errp, "dump: failed to write elf header");
177 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
178 int phdr_index, hwaddr offset,
179 hwaddr filesz, Error **errp)
181 Elf64_Phdr phdr;
182 int ret;
184 memset(&phdr, 0, sizeof(Elf64_Phdr));
185 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
186 phdr.p_offset = cpu_to_dump64(s, offset);
187 phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
188 phdr.p_filesz = cpu_to_dump64(s, filesz);
189 phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
190 phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr);
192 assert(memory_mapping->length >= filesz);
194 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
195 if (ret < 0) {
196 error_setg(errp, "dump: failed to write program header table");
200 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
201 int phdr_index, hwaddr offset,
202 hwaddr filesz, Error **errp)
204 Elf32_Phdr phdr;
205 int ret;
207 memset(&phdr, 0, sizeof(Elf32_Phdr));
208 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
209 phdr.p_offset = cpu_to_dump32(s, offset);
210 phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
211 phdr.p_filesz = cpu_to_dump32(s, filesz);
212 phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
213 phdr.p_vaddr = cpu_to_dump32(s, memory_mapping->virt_addr);
215 assert(memory_mapping->length >= filesz);
217 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
218 if (ret < 0) {
219 error_setg(errp, "dump: failed to write program header table");
223 static void write_elf64_note(DumpState *s, Error **errp)
225 Elf64_Phdr phdr;
226 hwaddr begin = s->memory_offset - s->note_size;
227 int ret;
229 memset(&phdr, 0, sizeof(Elf64_Phdr));
230 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
231 phdr.p_offset = cpu_to_dump64(s, begin);
232 phdr.p_paddr = 0;
233 phdr.p_filesz = cpu_to_dump64(s, s->note_size);
234 phdr.p_memsz = cpu_to_dump64(s, s->note_size);
235 phdr.p_vaddr = 0;
237 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
238 if (ret < 0) {
239 error_setg(errp, "dump: failed to write program header table");
243 static inline int cpu_index(CPUState *cpu)
245 return cpu->cpu_index + 1;
248 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
249 Error **errp)
251 int ret;
253 if (s->guest_note) {
254 ret = f(s->guest_note, s->guest_note_size, s);
255 if (ret < 0) {
256 error_setg(errp, "dump: failed to write guest note");
261 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
262 Error **errp)
264 CPUState *cpu;
265 int ret;
266 int id;
268 CPU_FOREACH(cpu) {
269 id = cpu_index(cpu);
270 ret = cpu_write_elf64_note(f, cpu, id, s);
271 if (ret < 0) {
272 error_setg(errp, "dump: failed to write elf notes");
273 return;
277 CPU_FOREACH(cpu) {
278 ret = cpu_write_elf64_qemunote(f, cpu, s);
279 if (ret < 0) {
280 error_setg(errp, "dump: failed to write CPU status");
281 return;
285 write_guest_note(f, s, errp);
288 static void write_elf32_note(DumpState *s, Error **errp)
290 hwaddr begin = s->memory_offset - s->note_size;
291 Elf32_Phdr phdr;
292 int ret;
294 memset(&phdr, 0, sizeof(Elf32_Phdr));
295 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
296 phdr.p_offset = cpu_to_dump32(s, begin);
297 phdr.p_paddr = 0;
298 phdr.p_filesz = cpu_to_dump32(s, s->note_size);
299 phdr.p_memsz = cpu_to_dump32(s, s->note_size);
300 phdr.p_vaddr = 0;
302 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
303 if (ret < 0) {
304 error_setg(errp, "dump: failed to write program header table");
308 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
309 Error **errp)
311 CPUState *cpu;
312 int ret;
313 int id;
315 CPU_FOREACH(cpu) {
316 id = cpu_index(cpu);
317 ret = cpu_write_elf32_note(f, cpu, id, s);
318 if (ret < 0) {
319 error_setg(errp, "dump: failed to write elf notes");
320 return;
324 CPU_FOREACH(cpu) {
325 ret = cpu_write_elf32_qemunote(f, cpu, s);
326 if (ret < 0) {
327 error_setg(errp, "dump: failed to write CPU status");
328 return;
332 write_guest_note(f, s, errp);
335 static void write_elf_section(DumpState *s, int type, Error **errp)
337 Elf32_Shdr shdr32;
338 Elf64_Shdr shdr64;
339 int shdr_size;
340 void *shdr;
341 int ret;
343 if (type == 0) {
344 shdr_size = sizeof(Elf32_Shdr);
345 memset(&shdr32, 0, shdr_size);
346 shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
347 shdr = &shdr32;
348 } else {
349 shdr_size = sizeof(Elf64_Shdr);
350 memset(&shdr64, 0, shdr_size);
351 shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
352 shdr = &shdr64;
355 ret = fd_write_vmcore(&shdr, shdr_size, s);
356 if (ret < 0) {
357 error_setg(errp, "dump: failed to write section header table");
361 static void write_data(DumpState *s, void *buf, int length, Error **errp)
363 int ret;
365 ret = fd_write_vmcore(buf, length, s);
366 if (ret < 0) {
367 error_setg(errp, "dump: failed to save memory");
368 } else {
369 s->written_size += length;
373 /* write the memory to vmcore. 1 page per I/O. */
374 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
375 int64_t size, Error **errp)
377 int64_t i;
378 Error *local_err = NULL;
380 for (i = 0; i < size / s->dump_info.page_size; i++) {
381 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
382 s->dump_info.page_size, &local_err);
383 if (local_err) {
384 error_propagate(errp, local_err);
385 return;
389 if ((size % s->dump_info.page_size) != 0) {
390 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
391 size % s->dump_info.page_size, &local_err);
392 if (local_err) {
393 error_propagate(errp, local_err);
394 return;
399 /* get the memory's offset and size in the vmcore */
400 static void get_offset_range(hwaddr phys_addr,
401 ram_addr_t mapping_length,
402 DumpState *s,
403 hwaddr *p_offset,
404 hwaddr *p_filesz)
406 GuestPhysBlock *block;
407 hwaddr offset = s->memory_offset;
408 int64_t size_in_block, start;
410 /* When the memory is not stored into vmcore, offset will be -1 */
411 *p_offset = -1;
412 *p_filesz = 0;
414 if (s->has_filter) {
415 if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
416 return;
420 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
421 if (s->has_filter) {
422 if (block->target_start >= s->begin + s->length ||
423 block->target_end <= s->begin) {
424 /* This block is out of the range */
425 continue;
428 if (s->begin <= block->target_start) {
429 start = block->target_start;
430 } else {
431 start = s->begin;
434 size_in_block = block->target_end - start;
435 if (s->begin + s->length < block->target_end) {
436 size_in_block -= block->target_end - (s->begin + s->length);
438 } else {
439 start = block->target_start;
440 size_in_block = block->target_end - block->target_start;
443 if (phys_addr >= start && phys_addr < start + size_in_block) {
444 *p_offset = phys_addr - start + offset;
446 /* The offset range mapped from the vmcore file must not spill over
447 * the GuestPhysBlock, clamp it. The rest of the mapping will be
448 * zero-filled in memory at load time; see
449 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
451 *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
452 mapping_length :
453 size_in_block - (phys_addr - start);
454 return;
457 offset += size_in_block;
461 static void write_elf_loads(DumpState *s, Error **errp)
463 hwaddr offset, filesz;
464 MemoryMapping *memory_mapping;
465 uint32_t phdr_index = 1;
466 uint32_t max_index;
467 Error *local_err = NULL;
469 if (s->have_section) {
470 max_index = s->sh_info;
471 } else {
472 max_index = s->phdr_num;
475 QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
476 get_offset_range(memory_mapping->phys_addr,
477 memory_mapping->length,
478 s, &offset, &filesz);
479 if (s->dump_info.d_class == ELFCLASS64) {
480 write_elf64_load(s, memory_mapping, phdr_index++, offset,
481 filesz, &local_err);
482 } else {
483 write_elf32_load(s, memory_mapping, phdr_index++, offset,
484 filesz, &local_err);
487 if (local_err) {
488 error_propagate(errp, local_err);
489 return;
492 if (phdr_index >= max_index) {
493 break;
498 /* write elf header, PT_NOTE and elf note to vmcore. */
499 static void dump_begin(DumpState *s, Error **errp)
501 Error *local_err = NULL;
504 * the vmcore's format is:
505 * --------------
506 * | elf header |
507 * --------------
508 * | PT_NOTE |
509 * --------------
510 * | PT_LOAD |
511 * --------------
512 * | ...... |
513 * --------------
514 * | PT_LOAD |
515 * --------------
516 * | sec_hdr |
517 * --------------
518 * | elf note |
519 * --------------
520 * | memory |
521 * --------------
523 * we only know where the memory is saved after we write elf note into
524 * vmcore.
527 /* write elf header to vmcore */
528 if (s->dump_info.d_class == ELFCLASS64) {
529 write_elf64_header(s, &local_err);
530 } else {
531 write_elf32_header(s, &local_err);
533 if (local_err) {
534 error_propagate(errp, local_err);
535 return;
538 if (s->dump_info.d_class == ELFCLASS64) {
539 /* write PT_NOTE to vmcore */
540 write_elf64_note(s, &local_err);
541 if (local_err) {
542 error_propagate(errp, local_err);
543 return;
546 /* write all PT_LOAD to vmcore */
547 write_elf_loads(s, &local_err);
548 if (local_err) {
549 error_propagate(errp, local_err);
550 return;
553 /* write section to vmcore */
554 if (s->have_section) {
555 write_elf_section(s, 1, &local_err);
556 if (local_err) {
557 error_propagate(errp, local_err);
558 return;
562 /* write notes to vmcore */
563 write_elf64_notes(fd_write_vmcore, s, &local_err);
564 if (local_err) {
565 error_propagate(errp, local_err);
566 return;
568 } else {
569 /* write PT_NOTE to vmcore */
570 write_elf32_note(s, &local_err);
571 if (local_err) {
572 error_propagate(errp, local_err);
573 return;
576 /* write all PT_LOAD to vmcore */
577 write_elf_loads(s, &local_err);
578 if (local_err) {
579 error_propagate(errp, local_err);
580 return;
583 /* write section to vmcore */
584 if (s->have_section) {
585 write_elf_section(s, 0, &local_err);
586 if (local_err) {
587 error_propagate(errp, local_err);
588 return;
592 /* write notes to vmcore */
593 write_elf32_notes(fd_write_vmcore, s, &local_err);
594 if (local_err) {
595 error_propagate(errp, local_err);
596 return;
601 static int get_next_block(DumpState *s, GuestPhysBlock *block)
603 while (1) {
604 block = QTAILQ_NEXT(block, next);
605 if (!block) {
606 /* no more block */
607 return 1;
610 s->start = 0;
611 s->next_block = block;
612 if (s->has_filter) {
613 if (block->target_start >= s->begin + s->length ||
614 block->target_end <= s->begin) {
615 /* This block is out of the range */
616 continue;
619 if (s->begin > block->target_start) {
620 s->start = s->begin - block->target_start;
624 return 0;
628 /* write all memory to vmcore */
629 static void dump_iterate(DumpState *s, Error **errp)
631 GuestPhysBlock *block;
632 int64_t size;
633 Error *local_err = NULL;
635 do {
636 block = s->next_block;
638 size = block->target_end - block->target_start;
639 if (s->has_filter) {
640 size -= s->start;
641 if (s->begin + s->length < block->target_end) {
642 size -= block->target_end - (s->begin + s->length);
645 write_memory(s, block, s->start, size, &local_err);
646 if (local_err) {
647 error_propagate(errp, local_err);
648 return;
651 } while (!get_next_block(s, block));
654 static void create_vmcore(DumpState *s, Error **errp)
656 Error *local_err = NULL;
658 dump_begin(s, &local_err);
659 if (local_err) {
660 error_propagate(errp, local_err);
661 return;
664 dump_iterate(s, errp);
667 static int write_start_flat_header(int fd)
669 MakedumpfileHeader *mh;
670 int ret = 0;
672 QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
673 mh = g_malloc0(MAX_SIZE_MDF_HEADER);
675 memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
676 MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
678 mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
679 mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
681 size_t written_size;
682 written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
683 if (written_size != MAX_SIZE_MDF_HEADER) {
684 ret = -1;
687 g_free(mh);
688 return ret;
691 static int write_end_flat_header(int fd)
693 MakedumpfileDataHeader mdh;
695 mdh.offset = END_FLAG_FLAT_HEADER;
696 mdh.buf_size = END_FLAG_FLAT_HEADER;
698 size_t written_size;
699 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
700 if (written_size != sizeof(mdh)) {
701 return -1;
704 return 0;
707 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
709 size_t written_size;
710 MakedumpfileDataHeader mdh;
712 mdh.offset = cpu_to_be64(offset);
713 mdh.buf_size = cpu_to_be64(size);
715 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
716 if (written_size != sizeof(mdh)) {
717 return -1;
720 written_size = qemu_write_full(fd, buf, size);
721 if (written_size != size) {
722 return -1;
725 return 0;
728 static int buf_write_note(const void *buf, size_t size, void *opaque)
730 DumpState *s = opaque;
732 /* note_buf is not enough */
733 if (s->note_buf_offset + size > s->note_size) {
734 return -1;
737 memcpy(s->note_buf + s->note_buf_offset, buf, size);
739 s->note_buf_offset += size;
741 return 0;
745 * This function retrieves various sizes from an elf header.
747 * @note has to be a valid ELF note. The return sizes are unmodified
748 * (not padded or rounded up to be multiple of 4).
750 static void get_note_sizes(DumpState *s, const void *note,
751 uint64_t *note_head_size,
752 uint64_t *name_size,
753 uint64_t *desc_size)
755 uint64_t note_head_sz;
756 uint64_t name_sz;
757 uint64_t desc_sz;
759 if (s->dump_info.d_class == ELFCLASS64) {
760 const Elf64_Nhdr *hdr = note;
761 note_head_sz = sizeof(Elf64_Nhdr);
762 name_sz = tswap64(hdr->n_namesz);
763 desc_sz = tswap64(hdr->n_descsz);
764 } else {
765 const Elf32_Nhdr *hdr = note;
766 note_head_sz = sizeof(Elf32_Nhdr);
767 name_sz = tswap32(hdr->n_namesz);
768 desc_sz = tswap32(hdr->n_descsz);
771 if (note_head_size) {
772 *note_head_size = note_head_sz;
774 if (name_size) {
775 *name_size = name_sz;
777 if (desc_size) {
778 *desc_size = desc_sz;
782 static bool note_name_equal(DumpState *s,
783 const uint8_t *note, const char *name)
785 int len = strlen(name) + 1;
786 uint64_t head_size, name_size;
788 get_note_sizes(s, note, &head_size, &name_size, NULL);
789 head_size = ROUND_UP(head_size, 4);
791 return name_size == len && memcmp(note + head_size, name, len) == 0;
794 /* write common header, sub header and elf note to vmcore */
795 static void create_header32(DumpState *s, Error **errp)
797 DiskDumpHeader32 *dh = NULL;
798 KdumpSubHeader32 *kh = NULL;
799 size_t size;
800 uint32_t block_size;
801 uint32_t sub_hdr_size;
802 uint32_t bitmap_blocks;
803 uint32_t status = 0;
804 uint64_t offset_note;
805 Error *local_err = NULL;
807 /* write common header, the version of kdump-compressed format is 6th */
808 size = sizeof(DiskDumpHeader32);
809 dh = g_malloc0(size);
811 strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
812 dh->header_version = cpu_to_dump32(s, 6);
813 block_size = s->dump_info.page_size;
814 dh->block_size = cpu_to_dump32(s, block_size);
815 sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
816 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
817 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
818 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
819 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
820 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
821 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
822 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
823 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
825 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
826 status |= DUMP_DH_COMPRESSED_ZLIB;
828 #ifdef CONFIG_LZO
829 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
830 status |= DUMP_DH_COMPRESSED_LZO;
832 #endif
833 #ifdef CONFIG_SNAPPY
834 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
835 status |= DUMP_DH_COMPRESSED_SNAPPY;
837 #endif
838 dh->status = cpu_to_dump32(s, status);
840 if (write_buffer(s->fd, 0, dh, size) < 0) {
841 error_setg(errp, "dump: failed to write disk dump header");
842 goto out;
845 /* write sub header */
846 size = sizeof(KdumpSubHeader32);
847 kh = g_malloc0(size);
849 /* 64bit max_mapnr_64 */
850 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
851 kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
852 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
854 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
855 if (s->guest_note &&
856 note_name_equal(s, s->guest_note, "VMCOREINFO")) {
857 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
859 get_note_sizes(s, s->guest_note,
860 &hsize, &name_size, &size_vmcoreinfo_desc);
861 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
862 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
863 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
864 kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
867 kh->offset_note = cpu_to_dump64(s, offset_note);
868 kh->note_size = cpu_to_dump32(s, s->note_size);
870 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
871 block_size, kh, size) < 0) {
872 error_setg(errp, "dump: failed to write kdump sub header");
873 goto out;
876 /* write note */
877 s->note_buf = g_malloc0(s->note_size);
878 s->note_buf_offset = 0;
880 /* use s->note_buf to store notes temporarily */
881 write_elf32_notes(buf_write_note, s, &local_err);
882 if (local_err) {
883 error_propagate(errp, local_err);
884 goto out;
886 if (write_buffer(s->fd, offset_note, s->note_buf,
887 s->note_size) < 0) {
888 error_setg(errp, "dump: failed to write notes");
889 goto out;
892 /* get offset of dump_bitmap */
893 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
894 block_size;
896 /* get offset of page */
897 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
898 block_size;
900 out:
901 g_free(dh);
902 g_free(kh);
903 g_free(s->note_buf);
906 /* write common header, sub header and elf note to vmcore */
907 static void create_header64(DumpState *s, Error **errp)
909 DiskDumpHeader64 *dh = NULL;
910 KdumpSubHeader64 *kh = NULL;
911 size_t size;
912 uint32_t block_size;
913 uint32_t sub_hdr_size;
914 uint32_t bitmap_blocks;
915 uint32_t status = 0;
916 uint64_t offset_note;
917 Error *local_err = NULL;
919 /* write common header, the version of kdump-compressed format is 6th */
920 size = sizeof(DiskDumpHeader64);
921 dh = g_malloc0(size);
923 strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
924 dh->header_version = cpu_to_dump32(s, 6);
925 block_size = s->dump_info.page_size;
926 dh->block_size = cpu_to_dump32(s, block_size);
927 sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
928 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
929 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
930 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
931 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
932 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
933 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
934 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
935 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
937 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
938 status |= DUMP_DH_COMPRESSED_ZLIB;
940 #ifdef CONFIG_LZO
941 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
942 status |= DUMP_DH_COMPRESSED_LZO;
944 #endif
945 #ifdef CONFIG_SNAPPY
946 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
947 status |= DUMP_DH_COMPRESSED_SNAPPY;
949 #endif
950 dh->status = cpu_to_dump32(s, status);
952 if (write_buffer(s->fd, 0, dh, size) < 0) {
953 error_setg(errp, "dump: failed to write disk dump header");
954 goto out;
957 /* write sub header */
958 size = sizeof(KdumpSubHeader64);
959 kh = g_malloc0(size);
961 /* 64bit max_mapnr_64 */
962 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
963 kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
964 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
966 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
967 if (s->guest_note &&
968 note_name_equal(s, s->guest_note, "VMCOREINFO")) {
969 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
971 get_note_sizes(s, s->guest_note,
972 &hsize, &name_size, &size_vmcoreinfo_desc);
973 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
974 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
975 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
976 kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
979 kh->offset_note = cpu_to_dump64(s, offset_note);
980 kh->note_size = cpu_to_dump64(s, s->note_size);
982 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
983 block_size, kh, size) < 0) {
984 error_setg(errp, "dump: failed to write kdump sub header");
985 goto out;
988 /* write note */
989 s->note_buf = g_malloc0(s->note_size);
990 s->note_buf_offset = 0;
992 /* use s->note_buf to store notes temporarily */
993 write_elf64_notes(buf_write_note, s, &local_err);
994 if (local_err) {
995 error_propagate(errp, local_err);
996 goto out;
999 if (write_buffer(s->fd, offset_note, s->note_buf,
1000 s->note_size) < 0) {
1001 error_setg(errp, "dump: failed to write notes");
1002 goto out;
1005 /* get offset of dump_bitmap */
1006 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1007 block_size;
1009 /* get offset of page */
1010 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1011 block_size;
1013 out:
1014 g_free(dh);
1015 g_free(kh);
1016 g_free(s->note_buf);
1019 static void write_dump_header(DumpState *s, Error **errp)
1021 Error *local_err = NULL;
1023 if (s->dump_info.d_class == ELFCLASS32) {
1024 create_header32(s, &local_err);
1025 } else {
1026 create_header64(s, &local_err);
1028 error_propagate(errp, local_err);
1031 static size_t dump_bitmap_get_bufsize(DumpState *s)
1033 return s->dump_info.page_size;
1037 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1038 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1039 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1040 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1041 * vmcore, ie. synchronizing un-sync bit into vmcore.
1043 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1044 uint8_t *buf, DumpState *s)
1046 off_t old_offset, new_offset;
1047 off_t offset_bitmap1, offset_bitmap2;
1048 uint32_t byte, bit;
1049 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1050 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1052 /* should not set the previous place */
1053 assert(last_pfn <= pfn);
1056 * if the bit needed to be set is not cached in buf, flush the data in buf
1057 * to vmcore firstly.
1058 * making new_offset be bigger than old_offset can also sync remained data
1059 * into vmcore.
1061 old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1062 new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1064 while (old_offset < new_offset) {
1065 /* calculate the offset and write dump_bitmap */
1066 offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1067 if (write_buffer(s->fd, offset_bitmap1, buf,
1068 bitmap_bufsize) < 0) {
1069 return -1;
1072 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1073 offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1074 old_offset;
1075 if (write_buffer(s->fd, offset_bitmap2, buf,
1076 bitmap_bufsize) < 0) {
1077 return -1;
1080 memset(buf, 0, bitmap_bufsize);
1081 old_offset += bitmap_bufsize;
1084 /* get the exact place of the bit in the buf, and set it */
1085 byte = (pfn % bits_per_buf) / CHAR_BIT;
1086 bit = (pfn % bits_per_buf) % CHAR_BIT;
1087 if (value) {
1088 buf[byte] |= 1u << bit;
1089 } else {
1090 buf[byte] &= ~(1u << bit);
1093 return 0;
1096 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1098 int target_page_shift = ctz32(s->dump_info.page_size);
1100 return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1103 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1105 int target_page_shift = ctz32(s->dump_info.page_size);
1107 return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1111 * exam every page and return the page frame number and the address of the page.
1112 * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1113 * blocks, so block->target_start and block->target_end should be interal
1114 * multiples of the target page size.
1116 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1117 uint8_t **bufptr, DumpState *s)
1119 GuestPhysBlock *block = *blockptr;
1120 hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1121 uint8_t *buf;
1123 /* block == NULL means the start of the iteration */
1124 if (!block) {
1125 block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1126 *blockptr = block;
1127 assert((block->target_start & ~target_page_mask) == 0);
1128 assert((block->target_end & ~target_page_mask) == 0);
1129 *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1130 if (bufptr) {
1131 *bufptr = block->host_addr;
1133 return true;
1136 *pfnptr = *pfnptr + 1;
1137 addr = dump_pfn_to_paddr(s, *pfnptr);
1139 if ((addr >= block->target_start) &&
1140 (addr + s->dump_info.page_size <= block->target_end)) {
1141 buf = block->host_addr + (addr - block->target_start);
1142 } else {
1143 /* the next page is in the next block */
1144 block = QTAILQ_NEXT(block, next);
1145 *blockptr = block;
1146 if (!block) {
1147 return false;
1149 assert((block->target_start & ~target_page_mask) == 0);
1150 assert((block->target_end & ~target_page_mask) == 0);
1151 *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1152 buf = block->host_addr;
1155 if (bufptr) {
1156 *bufptr = buf;
1159 return true;
1162 static void write_dump_bitmap(DumpState *s, Error **errp)
1164 int ret = 0;
1165 uint64_t last_pfn, pfn;
1166 void *dump_bitmap_buf;
1167 size_t num_dumpable;
1168 GuestPhysBlock *block_iter = NULL;
1169 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1170 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1172 /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1173 dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1175 num_dumpable = 0;
1176 last_pfn = 0;
1179 * exam memory page by page, and set the bit in dump_bitmap corresponded
1180 * to the existing page.
1182 while (get_next_page(&block_iter, &pfn, NULL, s)) {
1183 ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1184 if (ret < 0) {
1185 error_setg(errp, "dump: failed to set dump_bitmap");
1186 goto out;
1189 last_pfn = pfn;
1190 num_dumpable++;
1194 * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1195 * set the remaining bits from last_pfn to the end of the bitmap buffer to
1196 * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1198 if (num_dumpable > 0) {
1199 ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1200 dump_bitmap_buf, s);
1201 if (ret < 0) {
1202 error_setg(errp, "dump: failed to sync dump_bitmap");
1203 goto out;
1207 /* number of dumpable pages that will be dumped later */
1208 s->num_dumpable = num_dumpable;
1210 out:
1211 g_free(dump_bitmap_buf);
1214 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1215 off_t offset)
1217 data_cache->fd = s->fd;
1218 data_cache->data_size = 0;
1219 data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1220 data_cache->buf = g_malloc0(data_cache->buf_size);
1221 data_cache->offset = offset;
1224 static int write_cache(DataCache *dc, const void *buf, size_t size,
1225 bool flag_sync)
1228 * dc->buf_size should not be less than size, otherwise dc will never be
1229 * enough
1231 assert(size <= dc->buf_size);
1234 * if flag_sync is set, synchronize data in dc->buf into vmcore.
1235 * otherwise check if the space is enough for caching data in buf, if not,
1236 * write the data in dc->buf to dc->fd and reset dc->buf
1238 if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1239 (flag_sync && dc->data_size > 0)) {
1240 if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1241 return -1;
1244 dc->offset += dc->data_size;
1245 dc->data_size = 0;
1248 if (!flag_sync) {
1249 memcpy(dc->buf + dc->data_size, buf, size);
1250 dc->data_size += size;
1253 return 0;
1256 static void free_data_cache(DataCache *data_cache)
1258 g_free(data_cache->buf);
1261 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1263 switch (flag_compress) {
1264 case DUMP_DH_COMPRESSED_ZLIB:
1265 return compressBound(page_size);
1267 case DUMP_DH_COMPRESSED_LZO:
1269 * LZO will expand incompressible data by a little amount. Please check
1270 * the following URL to see the expansion calculation:
1271 * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1273 return page_size + page_size / 16 + 64 + 3;
1275 #ifdef CONFIG_SNAPPY
1276 case DUMP_DH_COMPRESSED_SNAPPY:
1277 return snappy_max_compressed_length(page_size);
1278 #endif
1280 return 0;
1284 * check if the page is all 0
1286 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1288 return buffer_is_zero(buf, page_size);
1291 static void write_dump_pages(DumpState *s, Error **errp)
1293 int ret = 0;
1294 DataCache page_desc, page_data;
1295 size_t len_buf_out, size_out;
1296 #ifdef CONFIG_LZO
1297 lzo_bytep wrkmem = NULL;
1298 #endif
1299 uint8_t *buf_out = NULL;
1300 off_t offset_desc, offset_data;
1301 PageDescriptor pd, pd_zero;
1302 uint8_t *buf;
1303 GuestPhysBlock *block_iter = NULL;
1304 uint64_t pfn_iter;
1306 /* get offset of page_desc and page_data in dump file */
1307 offset_desc = s->offset_page;
1308 offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1310 prepare_data_cache(&page_desc, s, offset_desc);
1311 prepare_data_cache(&page_data, s, offset_data);
1313 /* prepare buffer to store compressed data */
1314 len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1315 assert(len_buf_out != 0);
1317 #ifdef CONFIG_LZO
1318 wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1319 #endif
1321 buf_out = g_malloc(len_buf_out);
1324 * init zero page's page_desc and page_data, because every zero page
1325 * uses the same page_data
1327 pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1328 pd_zero.flags = cpu_to_dump32(s, 0);
1329 pd_zero.offset = cpu_to_dump64(s, offset_data);
1330 pd_zero.page_flags = cpu_to_dump64(s, 0);
1331 buf = g_malloc0(s->dump_info.page_size);
1332 ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1333 g_free(buf);
1334 if (ret < 0) {
1335 error_setg(errp, "dump: failed to write page data (zero page)");
1336 goto out;
1339 offset_data += s->dump_info.page_size;
1342 * dump memory to vmcore page by page. zero page will all be resided in the
1343 * first page of page section
1345 while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1346 /* check zero page */
1347 if (is_zero_page(buf, s->dump_info.page_size)) {
1348 ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1349 false);
1350 if (ret < 0) {
1351 error_setg(errp, "dump: failed to write page desc");
1352 goto out;
1354 } else {
1356 * not zero page, then:
1357 * 1. compress the page
1358 * 2. write the compressed page into the cache of page_data
1359 * 3. get page desc of the compressed page and write it into the
1360 * cache of page_desc
1362 * only one compression format will be used here, for
1363 * s->flag_compress is set. But when compression fails to work,
1364 * we fall back to save in plaintext.
1366 size_out = len_buf_out;
1367 if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1368 (compress2(buf_out, (uLongf *)&size_out, buf,
1369 s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1370 (size_out < s->dump_info.page_size)) {
1371 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1372 pd.size = cpu_to_dump32(s, size_out);
1374 ret = write_cache(&page_data, buf_out, size_out, false);
1375 if (ret < 0) {
1376 error_setg(errp, "dump: failed to write page data");
1377 goto out;
1379 #ifdef CONFIG_LZO
1380 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1381 (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1382 (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1383 (size_out < s->dump_info.page_size)) {
1384 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1385 pd.size = cpu_to_dump32(s, size_out);
1387 ret = write_cache(&page_data, buf_out, size_out, false);
1388 if (ret < 0) {
1389 error_setg(errp, "dump: failed to write page data");
1390 goto out;
1392 #endif
1393 #ifdef CONFIG_SNAPPY
1394 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1395 (snappy_compress((char *)buf, s->dump_info.page_size,
1396 (char *)buf_out, &size_out) == SNAPPY_OK) &&
1397 (size_out < s->dump_info.page_size)) {
1398 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1399 pd.size = cpu_to_dump32(s, size_out);
1401 ret = write_cache(&page_data, buf_out, size_out, false);
1402 if (ret < 0) {
1403 error_setg(errp, "dump: failed to write page data");
1404 goto out;
1406 #endif
1407 } else {
1409 * fall back to save in plaintext, size_out should be
1410 * assigned the target's page size
1412 pd.flags = cpu_to_dump32(s, 0);
1413 size_out = s->dump_info.page_size;
1414 pd.size = cpu_to_dump32(s, size_out);
1416 ret = write_cache(&page_data, buf,
1417 s->dump_info.page_size, false);
1418 if (ret < 0) {
1419 error_setg(errp, "dump: failed to write page data");
1420 goto out;
1424 /* get and write page desc here */
1425 pd.page_flags = cpu_to_dump64(s, 0);
1426 pd.offset = cpu_to_dump64(s, offset_data);
1427 offset_data += size_out;
1429 ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1430 if (ret < 0) {
1431 error_setg(errp, "dump: failed to write page desc");
1432 goto out;
1435 s->written_size += s->dump_info.page_size;
1438 ret = write_cache(&page_desc, NULL, 0, true);
1439 if (ret < 0) {
1440 error_setg(errp, "dump: failed to sync cache for page_desc");
1441 goto out;
1443 ret = write_cache(&page_data, NULL, 0, true);
1444 if (ret < 0) {
1445 error_setg(errp, "dump: failed to sync cache for page_data");
1446 goto out;
1449 out:
1450 free_data_cache(&page_desc);
1451 free_data_cache(&page_data);
1453 #ifdef CONFIG_LZO
1454 g_free(wrkmem);
1455 #endif
1457 g_free(buf_out);
1460 static void create_kdump_vmcore(DumpState *s, Error **errp)
1462 int ret;
1463 Error *local_err = NULL;
1466 * the kdump-compressed format is:
1467 * File offset
1468 * +------------------------------------------+ 0x0
1469 * | main header (struct disk_dump_header) |
1470 * |------------------------------------------+ block 1
1471 * | sub header (struct kdump_sub_header) |
1472 * |------------------------------------------+ block 2
1473 * | 1st-dump_bitmap |
1474 * |------------------------------------------+ block 2 + X blocks
1475 * | 2nd-dump_bitmap | (aligned by block)
1476 * |------------------------------------------+ block 2 + 2 * X blocks
1477 * | page desc for pfn 0 (struct page_desc) | (aligned by block)
1478 * | page desc for pfn 1 (struct page_desc) |
1479 * | : |
1480 * |------------------------------------------| (not aligned by block)
1481 * | page data (pfn 0) |
1482 * | page data (pfn 1) |
1483 * | : |
1484 * +------------------------------------------+
1487 ret = write_start_flat_header(s->fd);
1488 if (ret < 0) {
1489 error_setg(errp, "dump: failed to write start flat header");
1490 return;
1493 write_dump_header(s, &local_err);
1494 if (local_err) {
1495 error_propagate(errp, local_err);
1496 return;
1499 write_dump_bitmap(s, &local_err);
1500 if (local_err) {
1501 error_propagate(errp, local_err);
1502 return;
1505 write_dump_pages(s, &local_err);
1506 if (local_err) {
1507 error_propagate(errp, local_err);
1508 return;
1511 ret = write_end_flat_header(s->fd);
1512 if (ret < 0) {
1513 error_setg(errp, "dump: failed to write end flat header");
1514 return;
1518 static ram_addr_t get_start_block(DumpState *s)
1520 GuestPhysBlock *block;
1522 if (!s->has_filter) {
1523 s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1524 return 0;
1527 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1528 if (block->target_start >= s->begin + s->length ||
1529 block->target_end <= s->begin) {
1530 /* This block is out of the range */
1531 continue;
1534 s->next_block = block;
1535 if (s->begin > block->target_start) {
1536 s->start = s->begin - block->target_start;
1537 } else {
1538 s->start = 0;
1540 return s->start;
1543 return -1;
1546 static void get_max_mapnr(DumpState *s)
1548 GuestPhysBlock *last_block;
1550 last_block = QTAILQ_LAST(&s->guest_phys_blocks.head, GuestPhysBlockHead);
1551 s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1554 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1556 static void dump_state_prepare(DumpState *s)
1558 /* zero the struct, setting status to active */
1559 *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1562 bool dump_in_progress(void)
1564 DumpState *state = &dump_state_global;
1565 return (atomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1568 /* calculate total size of memory to be dumped (taking filter into
1569 * acoount.) */
1570 static int64_t dump_calculate_size(DumpState *s)
1572 GuestPhysBlock *block;
1573 int64_t size = 0, total = 0, left = 0, right = 0;
1575 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1576 if (s->has_filter) {
1577 /* calculate the overlapped region. */
1578 left = MAX(s->begin, block->target_start);
1579 right = MIN(s->begin + s->length, block->target_end);
1580 size = right - left;
1581 size = size > 0 ? size : 0;
1582 } else {
1583 /* count the whole region in */
1584 size = (block->target_end - block->target_start);
1586 total += size;
1589 return total;
1592 static void vmcoreinfo_update_phys_base(DumpState *s)
1594 uint64_t size, note_head_size, name_size, phys_base;
1595 char **lines;
1596 uint8_t *vmci;
1597 size_t i;
1599 if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1600 return;
1603 get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1604 note_head_size = ROUND_UP(note_head_size, 4);
1606 vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1607 *(vmci + size) = '\0';
1609 lines = g_strsplit((char *)vmci, "\n", -1);
1610 for (i = 0; lines[i]; i++) {
1611 if (g_str_has_prefix(lines[i], "NUMBER(phys_base)=")) {
1612 if (qemu_strtou64(lines[i] + 18, NULL, 16,
1613 &phys_base) < 0) {
1614 warn_report("Failed to read NUMBER(phys_base)=");
1615 } else {
1616 s->dump_info.phys_base = phys_base;
1618 break;
1622 g_strfreev(lines);
1625 static void dump_init(DumpState *s, int fd, bool has_format,
1626 DumpGuestMemoryFormat format, bool paging, bool has_filter,
1627 int64_t begin, int64_t length, Error **errp)
1629 VMCoreInfoState *vmci = vmcoreinfo_find();
1630 CPUState *cpu;
1631 int nr_cpus;
1632 Error *err = NULL;
1633 int ret;
1635 s->has_format = has_format;
1636 s->format = format;
1637 s->written_size = 0;
1639 /* kdump-compressed is conflict with paging and filter */
1640 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1641 assert(!paging && !has_filter);
1644 if (runstate_is_running()) {
1645 vm_stop(RUN_STATE_SAVE_VM);
1646 s->resume = true;
1647 } else {
1648 s->resume = false;
1651 /* If we use KVM, we should synchronize the registers before we get dump
1652 * info or physmap info.
1654 cpu_synchronize_all_states();
1655 nr_cpus = 0;
1656 CPU_FOREACH(cpu) {
1657 nr_cpus++;
1660 s->fd = fd;
1661 s->has_filter = has_filter;
1662 s->begin = begin;
1663 s->length = length;
1665 memory_mapping_list_init(&s->list);
1667 guest_phys_blocks_init(&s->guest_phys_blocks);
1668 guest_phys_blocks_append(&s->guest_phys_blocks);
1669 s->total_size = dump_calculate_size(s);
1670 #ifdef DEBUG_DUMP_GUEST_MEMORY
1671 fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1672 #endif
1674 /* it does not make sense to dump non-existent memory */
1675 if (!s->total_size) {
1676 error_setg(errp, "dump: no guest memory to dump");
1677 goto cleanup;
1680 s->start = get_start_block(s);
1681 if (s->start == -1) {
1682 error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1683 goto cleanup;
1686 /* get dump info: endian, class and architecture.
1687 * If the target architecture is not supported, cpu_get_dump_info() will
1688 * return -1.
1690 ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1691 if (ret < 0) {
1692 error_setg(errp, QERR_UNSUPPORTED);
1693 goto cleanup;
1696 if (!s->dump_info.page_size) {
1697 s->dump_info.page_size = TARGET_PAGE_SIZE;
1700 s->note_size = cpu_get_note_size(s->dump_info.d_class,
1701 s->dump_info.d_machine, nr_cpus);
1702 if (s->note_size < 0) {
1703 error_setg(errp, QERR_UNSUPPORTED);
1704 goto cleanup;
1708 * The goal of this block is to (a) update the previously guessed
1709 * phys_base, (b) copy the guest note out of the guest.
1710 * Failure to do so is not fatal for dumping.
1712 if (vmci) {
1713 uint64_t addr, note_head_size, name_size, desc_size;
1714 uint32_t size;
1715 uint16_t format;
1717 note_head_size = s->dump_info.d_class == ELFCLASS32 ?
1718 sizeof(Elf32_Nhdr) : sizeof(Elf64_Nhdr);
1720 format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1721 size = le32_to_cpu(vmci->vmcoreinfo.size);
1722 addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1723 if (!vmci->has_vmcoreinfo) {
1724 warn_report("guest note is not present");
1725 } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1726 warn_report("guest note size is invalid: %" PRIu32, size);
1727 } else if (format != VMCOREINFO_FORMAT_ELF) {
1728 warn_report("guest note format is unsupported: %" PRIu16, format);
1729 } else {
1730 s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1731 cpu_physical_memory_read(addr, s->guest_note, size);
1733 get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1734 s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1735 desc_size);
1736 if (name_size > MAX_GUEST_NOTE_SIZE ||
1737 desc_size > MAX_GUEST_NOTE_SIZE ||
1738 s->guest_note_size > size) {
1739 warn_report("Invalid guest note header");
1740 g_free(s->guest_note);
1741 s->guest_note = NULL;
1742 } else {
1743 vmcoreinfo_update_phys_base(s);
1744 s->note_size += s->guest_note_size;
1749 /* get memory mapping */
1750 if (paging) {
1751 qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1752 if (err != NULL) {
1753 error_propagate(errp, err);
1754 goto cleanup;
1756 } else {
1757 qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1760 s->nr_cpus = nr_cpus;
1762 get_max_mapnr(s);
1764 uint64_t tmp;
1765 tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1766 s->dump_info.page_size);
1767 s->len_dump_bitmap = tmp * s->dump_info.page_size;
1769 /* init for kdump-compressed format */
1770 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1771 switch (format) {
1772 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1773 s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1774 break;
1776 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1777 #ifdef CONFIG_LZO
1778 if (lzo_init() != LZO_E_OK) {
1779 error_setg(errp, "failed to initialize the LZO library");
1780 goto cleanup;
1782 #endif
1783 s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1784 break;
1786 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1787 s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1788 break;
1790 default:
1791 s->flag_compress = 0;
1794 return;
1797 if (s->has_filter) {
1798 memory_mapping_filter(&s->list, s->begin, s->length);
1802 * calculate phdr_num
1804 * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1806 s->phdr_num = 1; /* PT_NOTE */
1807 if (s->list.num < UINT16_MAX - 2) {
1808 s->phdr_num += s->list.num;
1809 s->have_section = false;
1810 } else {
1811 s->have_section = true;
1812 s->phdr_num = PN_XNUM;
1813 s->sh_info = 1; /* PT_NOTE */
1815 /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1816 if (s->list.num <= UINT32_MAX - 1) {
1817 s->sh_info += s->list.num;
1818 } else {
1819 s->sh_info = UINT32_MAX;
1823 if (s->dump_info.d_class == ELFCLASS64) {
1824 if (s->have_section) {
1825 s->memory_offset = sizeof(Elf64_Ehdr) +
1826 sizeof(Elf64_Phdr) * s->sh_info +
1827 sizeof(Elf64_Shdr) + s->note_size;
1828 } else {
1829 s->memory_offset = sizeof(Elf64_Ehdr) +
1830 sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1832 } else {
1833 if (s->have_section) {
1834 s->memory_offset = sizeof(Elf32_Ehdr) +
1835 sizeof(Elf32_Phdr) * s->sh_info +
1836 sizeof(Elf32_Shdr) + s->note_size;
1837 } else {
1838 s->memory_offset = sizeof(Elf32_Ehdr) +
1839 sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1843 return;
1845 cleanup:
1846 dump_cleanup(s);
1849 /* this operation might be time consuming. */
1850 static void dump_process(DumpState *s, Error **errp)
1852 Error *local_err = NULL;
1853 DumpQueryResult *result = NULL;
1855 if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1856 create_kdump_vmcore(s, &local_err);
1857 } else {
1858 create_vmcore(s, &local_err);
1861 /* make sure status is written after written_size updates */
1862 smp_wmb();
1863 atomic_set(&s->status,
1864 (local_err ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1866 /* send DUMP_COMPLETED message (unconditionally) */
1867 result = qmp_query_dump(NULL);
1868 /* should never fail */
1869 assert(result);
1870 qapi_event_send_dump_completed(result, !!local_err, (local_err ? \
1871 error_get_pretty(local_err) : NULL),
1872 &error_abort);
1873 qapi_free_DumpQueryResult(result);
1875 error_propagate(errp, local_err);
1876 dump_cleanup(s);
1879 static void *dump_thread(void *data)
1881 DumpState *s = (DumpState *)data;
1882 dump_process(s, NULL);
1883 return NULL;
1886 DumpQueryResult *qmp_query_dump(Error **errp)
1888 DumpQueryResult *result = g_new(DumpQueryResult, 1);
1889 DumpState *state = &dump_state_global;
1890 result->status = atomic_read(&state->status);
1891 /* make sure we are reading status and written_size in order */
1892 smp_rmb();
1893 result->completed = state->written_size;
1894 result->total = state->total_size;
1895 return result;
1898 void qmp_dump_guest_memory(bool paging, const char *file,
1899 bool has_detach, bool detach,
1900 bool has_begin, int64_t begin, bool has_length,
1901 int64_t length, bool has_format,
1902 DumpGuestMemoryFormat format, Error **errp)
1904 const char *p;
1905 int fd = -1;
1906 DumpState *s;
1907 Error *local_err = NULL;
1908 bool detach_p = false;
1910 if (runstate_check(RUN_STATE_INMIGRATE)) {
1911 error_setg(errp, "Dump not allowed during incoming migration.");
1912 return;
1915 /* if there is a dump in background, we should wait until the dump
1916 * finished */
1917 if (dump_in_progress()) {
1918 error_setg(errp, "There is a dump in process, please wait.");
1919 return;
1923 * kdump-compressed format need the whole memory dumped, so paging or
1924 * filter is not supported here.
1926 if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1927 (paging || has_begin || has_length)) {
1928 error_setg(errp, "kdump-compressed format doesn't support paging or "
1929 "filter");
1930 return;
1932 if (has_begin && !has_length) {
1933 error_setg(errp, QERR_MISSING_PARAMETER, "length");
1934 return;
1936 if (!has_begin && has_length) {
1937 error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1938 return;
1940 if (has_detach) {
1941 detach_p = detach;
1944 /* check whether lzo/snappy is supported */
1945 #ifndef CONFIG_LZO
1946 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1947 error_setg(errp, "kdump-lzo is not available now");
1948 return;
1950 #endif
1952 #ifndef CONFIG_SNAPPY
1953 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1954 error_setg(errp, "kdump-snappy is not available now");
1955 return;
1957 #endif
1959 #if !defined(WIN32)
1960 if (strstart(file, "fd:", &p)) {
1961 fd = monitor_get_fd(cur_mon, p, errp);
1962 if (fd == -1) {
1963 return;
1966 #endif
1968 if (strstart(file, "file:", &p)) {
1969 fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1970 if (fd < 0) {
1971 error_setg_file_open(errp, errno, p);
1972 return;
1976 if (fd == -1) {
1977 error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
1978 return;
1981 s = &dump_state_global;
1982 dump_state_prepare(s);
1984 dump_init(s, fd, has_format, format, paging, has_begin,
1985 begin, length, &local_err);
1986 if (local_err) {
1987 error_propagate(errp, local_err);
1988 atomic_set(&s->status, DUMP_STATUS_FAILED);
1989 return;
1992 if (detach_p) {
1993 /* detached dump */
1994 s->detached = true;
1995 qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
1996 s, QEMU_THREAD_DETACHED);
1997 } else {
1998 /* sync dump */
1999 dump_process(s, errp);
2003 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2005 DumpGuestMemoryFormatList *item;
2006 DumpGuestMemoryCapability *cap =
2007 g_malloc0(sizeof(DumpGuestMemoryCapability));
2009 /* elf is always available */
2010 item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2011 cap->formats = item;
2012 item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
2014 /* kdump-zlib is always available */
2015 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2016 item = item->next;
2017 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
2019 /* add new item if kdump-lzo is available */
2020 #ifdef CONFIG_LZO
2021 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2022 item = item->next;
2023 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
2024 #endif
2026 /* add new item if kdump-snappy is available */
2027 #ifdef CONFIG_SNAPPY
2028 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
2029 item = item->next;
2030 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
2031 #endif
2033 return cap;