hw/net: fix mcf_fec driver receiver
[qemu.git] / dump.c
1 /*
2 * QEMU dump
3 *
4 * Copyright Fujitsu, Corp. 2011, 2012
5 *
6 * Authors:
7 * Wen Congyang <wency@cn.fujitsu.com>
8 *
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.
11 *
12 */
13
14 #include "qemu-common.h"
15 #include "elf.h"
16 #include "cpu.h"
17 #include "exec/cpu-all.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/qmp/qerror.h"
27 #include "qmp-commands.h"
28
29 #include <zlib.h>
30 #ifdef CONFIG_LZO
31 #include <lzo/lzo1x.h>
32 #endif
33 #ifdef CONFIG_SNAPPY
34 #include <snappy-c.h>
35 #endif
36 #ifndef ELF_MACHINE_UNAME
37 #define ELF_MACHINE_UNAME "Unknown"
38 #endif
39
40 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
41 {
42 if (s->dump_info.d_endian == ELFDATA2LSB) {
43 val = cpu_to_le16(val);
44 } else {
45 val = cpu_to_be16(val);
46 }
47
48 return val;
49 }
50
51 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
52 {
53 if (s->dump_info.d_endian == ELFDATA2LSB) {
54 val = cpu_to_le32(val);
55 } else {
56 val = cpu_to_be32(val);
57 }
58
59 return val;
60 }
61
62 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
63 {
64 if (s->dump_info.d_endian == ELFDATA2LSB) {
65 val = cpu_to_le64(val);
66 } else {
67 val = cpu_to_be64(val);
68 }
69
70 return val;
71 }
72
73 static int dump_cleanup(DumpState *s)
74 {
75 guest_phys_blocks_free(&s->guest_phys_blocks);
76 memory_mapping_list_free(&s->list);
77 close(s->fd);
78 if (s->resume) {
79 vm_start();
80 }
81
82 return 0;
83 }
84
85 static void dump_error(DumpState *s, const char *reason, Error **errp)
86 {
87 dump_cleanup(s);
88 error_setg(errp, "%s", reason);
89 }
90
91 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
92 {
93 DumpState *s = opaque;
94 size_t written_size;
95
96 written_size = qemu_write_full(s->fd, buf, size);
97 if (written_size != size) {
98 return -1;
99 }
100
101 return 0;
102 }
103
104 static void write_elf64_header(DumpState *s, Error **errp)
105 {
106 Elf64_Ehdr elf_header;
107 int ret;
108
109 memset(&elf_header, 0, sizeof(Elf64_Ehdr));
110 memcpy(&elf_header, ELFMAG, SELFMAG);
111 elf_header.e_ident[EI_CLASS] = ELFCLASS64;
112 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
113 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
114 elf_header.e_type = cpu_to_dump16(s, ET_CORE);
115 elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
116 elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
117 elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
118 elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
119 elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
120 elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
121 if (s->have_section) {
122 uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
123
124 elf_header.e_shoff = cpu_to_dump64(s, shoff);
125 elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
126 elf_header.e_shnum = cpu_to_dump16(s, 1);
127 }
128
129 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
130 if (ret < 0) {
131 dump_error(s, "dump: failed to write elf header", errp);
132 }
133 }
134
135 static void write_elf32_header(DumpState *s, Error **errp)
136 {
137 Elf32_Ehdr elf_header;
138 int ret;
139
140 memset(&elf_header, 0, sizeof(Elf32_Ehdr));
141 memcpy(&elf_header, ELFMAG, SELFMAG);
142 elf_header.e_ident[EI_CLASS] = ELFCLASS32;
143 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
144 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
145 elf_header.e_type = cpu_to_dump16(s, ET_CORE);
146 elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
147 elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
148 elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
149 elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
150 elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
151 elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
152 if (s->have_section) {
153 uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
154
155 elf_header.e_shoff = cpu_to_dump32(s, shoff);
156 elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
157 elf_header.e_shnum = cpu_to_dump16(s, 1);
158 }
159
160 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
161 if (ret < 0) {
162 dump_error(s, "dump: failed to write elf header", errp);
163 }
164 }
165
166 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
167 int phdr_index, hwaddr offset,
168 hwaddr filesz, Error **errp)
169 {
170 Elf64_Phdr phdr;
171 int ret;
172
173 memset(&phdr, 0, sizeof(Elf64_Phdr));
174 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
175 phdr.p_offset = cpu_to_dump64(s, offset);
176 phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
177 phdr.p_filesz = cpu_to_dump64(s, filesz);
178 phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
179 phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr);
180
181 assert(memory_mapping->length >= filesz);
182
183 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
184 if (ret < 0) {
185 dump_error(s, "dump: failed to write program header table", errp);
186 }
187 }
188
189 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
190 int phdr_index, hwaddr offset,
191 hwaddr filesz, Error **errp)
192 {
193 Elf32_Phdr phdr;
194 int ret;
195
196 memset(&phdr, 0, sizeof(Elf32_Phdr));
197 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
198 phdr.p_offset = cpu_to_dump32(s, offset);
199 phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
200 phdr.p_filesz = cpu_to_dump32(s, filesz);
201 phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
202 phdr.p_vaddr = cpu_to_dump32(s, memory_mapping->virt_addr);
203
204 assert(memory_mapping->length >= filesz);
205
206 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
207 if (ret < 0) {
208 dump_error(s, "dump: failed to write program header table", errp);
209 }
210 }
211
212 static void write_elf64_note(DumpState *s, Error **errp)
213 {
214 Elf64_Phdr phdr;
215 hwaddr begin = s->memory_offset - s->note_size;
216 int ret;
217
218 memset(&phdr, 0, sizeof(Elf64_Phdr));
219 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
220 phdr.p_offset = cpu_to_dump64(s, begin);
221 phdr.p_paddr = 0;
222 phdr.p_filesz = cpu_to_dump64(s, s->note_size);
223 phdr.p_memsz = cpu_to_dump64(s, s->note_size);
224 phdr.p_vaddr = 0;
225
226 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
227 if (ret < 0) {
228 dump_error(s, "dump: failed to write program header table", errp);
229 }
230 }
231
232 static inline int cpu_index(CPUState *cpu)
233 {
234 return cpu->cpu_index + 1;
235 }
236
237 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
238 Error **errp)
239 {
240 CPUState *cpu;
241 int ret;
242 int id;
243
244 CPU_FOREACH(cpu) {
245 id = cpu_index(cpu);
246 ret = cpu_write_elf64_note(f, cpu, id, s);
247 if (ret < 0) {
248 dump_error(s, "dump: failed to write elf notes", errp);
249 return;
250 }
251 }
252
253 CPU_FOREACH(cpu) {
254 ret = cpu_write_elf64_qemunote(f, cpu, s);
255 if (ret < 0) {
256 dump_error(s, "dump: failed to write CPU status", errp);
257 return;
258 }
259 }
260 }
261
262 static void write_elf32_note(DumpState *s, Error **errp)
263 {
264 hwaddr begin = s->memory_offset - s->note_size;
265 Elf32_Phdr phdr;
266 int ret;
267
268 memset(&phdr, 0, sizeof(Elf32_Phdr));
269 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
270 phdr.p_offset = cpu_to_dump32(s, begin);
271 phdr.p_paddr = 0;
272 phdr.p_filesz = cpu_to_dump32(s, s->note_size);
273 phdr.p_memsz = cpu_to_dump32(s, s->note_size);
274 phdr.p_vaddr = 0;
275
276 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
277 if (ret < 0) {
278 dump_error(s, "dump: failed to write program header table", errp);
279 }
280 }
281
282 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
283 Error **errp)
284 {
285 CPUState *cpu;
286 int ret;
287 int id;
288
289 CPU_FOREACH(cpu) {
290 id = cpu_index(cpu);
291 ret = cpu_write_elf32_note(f, cpu, id, s);
292 if (ret < 0) {
293 dump_error(s, "dump: failed to write elf notes", errp);
294 return;
295 }
296 }
297
298 CPU_FOREACH(cpu) {
299 ret = cpu_write_elf32_qemunote(f, cpu, s);
300 if (ret < 0) {
301 dump_error(s, "dump: failed to write CPU status", errp);
302 return;
303 }
304 }
305 }
306
307 static void write_elf_section(DumpState *s, int type, Error **errp)
308 {
309 Elf32_Shdr shdr32;
310 Elf64_Shdr shdr64;
311 int shdr_size;
312 void *shdr;
313 int ret;
314
315 if (type == 0) {
316 shdr_size = sizeof(Elf32_Shdr);
317 memset(&shdr32, 0, shdr_size);
318 shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
319 shdr = &shdr32;
320 } else {
321 shdr_size = sizeof(Elf64_Shdr);
322 memset(&shdr64, 0, shdr_size);
323 shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
324 shdr = &shdr64;
325 }
326
327 ret = fd_write_vmcore(&shdr, shdr_size, s);
328 if (ret < 0) {
329 dump_error(s, "dump: failed to write section header table", errp);
330 }
331 }
332
333 static void write_data(DumpState *s, void *buf, int length, Error **errp)
334 {
335 int ret;
336
337 ret = fd_write_vmcore(buf, length, s);
338 if (ret < 0) {
339 dump_error(s, "dump: failed to save memory", errp);
340 }
341 }
342
343 /* write the memory to vmcore. 1 page per I/O. */
344 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
345 int64_t size, Error **errp)
346 {
347 int64_t i;
348 Error *local_err = NULL;
349
350 for (i = 0; i < size / TARGET_PAGE_SIZE; i++) {
351 write_data(s, block->host_addr + start + i * TARGET_PAGE_SIZE,
352 TARGET_PAGE_SIZE, &local_err);
353 if (local_err) {
354 error_propagate(errp, local_err);
355 return;
356 }
357 }
358
359 if ((size % TARGET_PAGE_SIZE) != 0) {
360 write_data(s, block->host_addr + start + i * TARGET_PAGE_SIZE,
361 size % TARGET_PAGE_SIZE, &local_err);
362 if (local_err) {
363 error_propagate(errp, local_err);
364 return;
365 }
366 }
367 }
368
369 /* get the memory's offset and size in the vmcore */
370 static void get_offset_range(hwaddr phys_addr,
371 ram_addr_t mapping_length,
372 DumpState *s,
373 hwaddr *p_offset,
374 hwaddr *p_filesz)
375 {
376 GuestPhysBlock *block;
377 hwaddr offset = s->memory_offset;
378 int64_t size_in_block, start;
379
380 /* When the memory is not stored into vmcore, offset will be -1 */
381 *p_offset = -1;
382 *p_filesz = 0;
383
384 if (s->has_filter) {
385 if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
386 return;
387 }
388 }
389
390 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
391 if (s->has_filter) {
392 if (block->target_start >= s->begin + s->length ||
393 block->target_end <= s->begin) {
394 /* This block is out of the range */
395 continue;
396 }
397
398 if (s->begin <= block->target_start) {
399 start = block->target_start;
400 } else {
401 start = s->begin;
402 }
403
404 size_in_block = block->target_end - start;
405 if (s->begin + s->length < block->target_end) {
406 size_in_block -= block->target_end - (s->begin + s->length);
407 }
408 } else {
409 start = block->target_start;
410 size_in_block = block->target_end - block->target_start;
411 }
412
413 if (phys_addr >= start && phys_addr < start + size_in_block) {
414 *p_offset = phys_addr - start + offset;
415
416 /* The offset range mapped from the vmcore file must not spill over
417 * the GuestPhysBlock, clamp it. The rest of the mapping will be
418 * zero-filled in memory at load time; see
419 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
420 */
421 *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
422 mapping_length :
423 size_in_block - (phys_addr - start);
424 return;
425 }
426
427 offset += size_in_block;
428 }
429 }
430
431 static void write_elf_loads(DumpState *s, Error **errp)
432 {
433 hwaddr offset, filesz;
434 MemoryMapping *memory_mapping;
435 uint32_t phdr_index = 1;
436 uint32_t max_index;
437 Error *local_err = NULL;
438
439 if (s->have_section) {
440 max_index = s->sh_info;
441 } else {
442 max_index = s->phdr_num;
443 }
444
445 QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
446 get_offset_range(memory_mapping->phys_addr,
447 memory_mapping->length,
448 s, &offset, &filesz);
449 if (s->dump_info.d_class == ELFCLASS64) {
450 write_elf64_load(s, memory_mapping, phdr_index++, offset,
451 filesz, &local_err);
452 } else {
453 write_elf32_load(s, memory_mapping, phdr_index++, offset,
454 filesz, &local_err);
455 }
456
457 if (local_err) {
458 error_propagate(errp, local_err);
459 return;
460 }
461
462 if (phdr_index >= max_index) {
463 break;
464 }
465 }
466 }
467
468 /* write elf header, PT_NOTE and elf note to vmcore. */
469 static void dump_begin(DumpState *s, Error **errp)
470 {
471 Error *local_err = NULL;
472
473 /*
474 * the vmcore's format is:
475 * --------------
476 * | elf header |
477 * --------------
478 * | PT_NOTE |
479 * --------------
480 * | PT_LOAD |
481 * --------------
482 * | ...... |
483 * --------------
484 * | PT_LOAD |
485 * --------------
486 * | sec_hdr |
487 * --------------
488 * | elf note |
489 * --------------
490 * | memory |
491 * --------------
492 *
493 * we only know where the memory is saved after we write elf note into
494 * vmcore.
495 */
496
497 /* write elf header to vmcore */
498 if (s->dump_info.d_class == ELFCLASS64) {
499 write_elf64_header(s, &local_err);
500 } else {
501 write_elf32_header(s, &local_err);
502 }
503 if (local_err) {
504 error_propagate(errp, local_err);
505 return;
506 }
507
508 if (s->dump_info.d_class == ELFCLASS64) {
509 /* write PT_NOTE to vmcore */
510 write_elf64_note(s, &local_err);
511 if (local_err) {
512 error_propagate(errp, local_err);
513 return;
514 }
515
516 /* write all PT_LOAD to vmcore */
517 write_elf_loads(s, &local_err);
518 if (local_err) {
519 error_propagate(errp, local_err);
520 return;
521 }
522
523 /* write section to vmcore */
524 if (s->have_section) {
525 write_elf_section(s, 1, &local_err);
526 if (local_err) {
527 error_propagate(errp, local_err);
528 return;
529 }
530 }
531
532 /* write notes to vmcore */
533 write_elf64_notes(fd_write_vmcore, s, &local_err);
534 if (local_err) {
535 error_propagate(errp, local_err);
536 return;
537 }
538 } else {
539 /* write PT_NOTE to vmcore */
540 write_elf32_note(s, &local_err);
541 if (local_err) {
542 error_propagate(errp, local_err);
543 return;
544 }
545
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;
551 }
552
553 /* write section to vmcore */
554 if (s->have_section) {
555 write_elf_section(s, 0, &local_err);
556 if (local_err) {
557 error_propagate(errp, local_err);
558 return;
559 }
560 }
561
562 /* write notes to vmcore */
563 write_elf32_notes(fd_write_vmcore, s, &local_err);
564 if (local_err) {
565 error_propagate(errp, local_err);
566 return;
567 }
568 }
569 }
570
571 static void dump_completed(DumpState *s)
572 {
573 dump_cleanup(s);
574 }
575
576 static int get_next_block(DumpState *s, GuestPhysBlock *block)
577 {
578 while (1) {
579 block = QTAILQ_NEXT(block, next);
580 if (!block) {
581 /* no more block */
582 return 1;
583 }
584
585 s->start = 0;
586 s->next_block = block;
587 if (s->has_filter) {
588 if (block->target_start >= s->begin + s->length ||
589 block->target_end <= s->begin) {
590 /* This block is out of the range */
591 continue;
592 }
593
594 if (s->begin > block->target_start) {
595 s->start = s->begin - block->target_start;
596 }
597 }
598
599 return 0;
600 }
601 }
602
603 /* write all memory to vmcore */
604 static void dump_iterate(DumpState *s, Error **errp)
605 {
606 GuestPhysBlock *block;
607 int64_t size;
608 Error *local_err = NULL;
609
610 do {
611 block = s->next_block;
612
613 size = block->target_end - block->target_start;
614 if (s->has_filter) {
615 size -= s->start;
616 if (s->begin + s->length < block->target_end) {
617 size -= block->target_end - (s->begin + s->length);
618 }
619 }
620 write_memory(s, block, s->start, size, &local_err);
621 if (local_err) {
622 error_propagate(errp, local_err);
623 return;
624 }
625
626 } while (!get_next_block(s, block));
627
628 dump_completed(s);
629 }
630
631 static void create_vmcore(DumpState *s, Error **errp)
632 {
633 Error *local_err = NULL;
634
635 dump_begin(s, &local_err);
636 if (local_err) {
637 error_propagate(errp, local_err);
638 return;
639 }
640
641 dump_iterate(s, errp);
642 }
643
644 static int write_start_flat_header(int fd)
645 {
646 MakedumpfileHeader *mh;
647 int ret = 0;
648
649 QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
650 mh = g_malloc0(MAX_SIZE_MDF_HEADER);
651
652 memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
653 MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
654
655 mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
656 mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
657
658 size_t written_size;
659 written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
660 if (written_size != MAX_SIZE_MDF_HEADER) {
661 ret = -1;
662 }
663
664 g_free(mh);
665 return ret;
666 }
667
668 static int write_end_flat_header(int fd)
669 {
670 MakedumpfileDataHeader mdh;
671
672 mdh.offset = END_FLAG_FLAT_HEADER;
673 mdh.buf_size = END_FLAG_FLAT_HEADER;
674
675 size_t written_size;
676 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
677 if (written_size != sizeof(mdh)) {
678 return -1;
679 }
680
681 return 0;
682 }
683
684 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
685 {
686 size_t written_size;
687 MakedumpfileDataHeader mdh;
688
689 mdh.offset = cpu_to_be64(offset);
690 mdh.buf_size = cpu_to_be64(size);
691
692 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
693 if (written_size != sizeof(mdh)) {
694 return -1;
695 }
696
697 written_size = qemu_write_full(fd, buf, size);
698 if (written_size != size) {
699 return -1;
700 }
701
702 return 0;
703 }
704
705 static int buf_write_note(const void *buf, size_t size, void *opaque)
706 {
707 DumpState *s = opaque;
708
709 /* note_buf is not enough */
710 if (s->note_buf_offset + size > s->note_size) {
711 return -1;
712 }
713
714 memcpy(s->note_buf + s->note_buf_offset, buf, size);
715
716 s->note_buf_offset += size;
717
718 return 0;
719 }
720
721 /* write common header, sub header and elf note to vmcore */
722 static void create_header32(DumpState *s, Error **errp)
723 {
724 DiskDumpHeader32 *dh = NULL;
725 KdumpSubHeader32 *kh = NULL;
726 size_t size;
727 uint32_t block_size;
728 uint32_t sub_hdr_size;
729 uint32_t bitmap_blocks;
730 uint32_t status = 0;
731 uint64_t offset_note;
732 Error *local_err = NULL;
733
734 /* write common header, the version of kdump-compressed format is 6th */
735 size = sizeof(DiskDumpHeader32);
736 dh = g_malloc0(size);
737
738 strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
739 dh->header_version = cpu_to_dump32(s, 6);
740 block_size = TARGET_PAGE_SIZE;
741 dh->block_size = cpu_to_dump32(s, block_size);
742 sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
743 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
744 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
745 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
746 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
747 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
748 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
749 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
750 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
751
752 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
753 status |= DUMP_DH_COMPRESSED_ZLIB;
754 }
755 #ifdef CONFIG_LZO
756 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
757 status |= DUMP_DH_COMPRESSED_LZO;
758 }
759 #endif
760 #ifdef CONFIG_SNAPPY
761 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
762 status |= DUMP_DH_COMPRESSED_SNAPPY;
763 }
764 #endif
765 dh->status = cpu_to_dump32(s, status);
766
767 if (write_buffer(s->fd, 0, dh, size) < 0) {
768 dump_error(s, "dump: failed to write disk dump header", errp);
769 goto out;
770 }
771
772 /* write sub header */
773 size = sizeof(KdumpSubHeader32);
774 kh = g_malloc0(size);
775
776 /* 64bit max_mapnr_64 */
777 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
778 kh->phys_base = cpu_to_dump32(s, PHYS_BASE);
779 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
780
781 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
782 kh->offset_note = cpu_to_dump64(s, offset_note);
783 kh->note_size = cpu_to_dump32(s, s->note_size);
784
785 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
786 block_size, kh, size) < 0) {
787 dump_error(s, "dump: failed to write kdump sub header", errp);
788 goto out;
789 }
790
791 /* write note */
792 s->note_buf = g_malloc0(s->note_size);
793 s->note_buf_offset = 0;
794
795 /* use s->note_buf to store notes temporarily */
796 write_elf32_notes(buf_write_note, s, &local_err);
797 if (local_err) {
798 error_propagate(errp, local_err);
799 goto out;
800 }
801 if (write_buffer(s->fd, offset_note, s->note_buf,
802 s->note_size) < 0) {
803 dump_error(s, "dump: failed to write notes", errp);
804 goto out;
805 }
806
807 /* get offset of dump_bitmap */
808 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
809 block_size;
810
811 /* get offset of page */
812 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
813 block_size;
814
815 out:
816 g_free(dh);
817 g_free(kh);
818 g_free(s->note_buf);
819 }
820
821 /* write common header, sub header and elf note to vmcore */
822 static void create_header64(DumpState *s, Error **errp)
823 {
824 DiskDumpHeader64 *dh = NULL;
825 KdumpSubHeader64 *kh = NULL;
826 size_t size;
827 uint32_t block_size;
828 uint32_t sub_hdr_size;
829 uint32_t bitmap_blocks;
830 uint32_t status = 0;
831 uint64_t offset_note;
832 Error *local_err = NULL;
833
834 /* write common header, the version of kdump-compressed format is 6th */
835 size = sizeof(DiskDumpHeader64);
836 dh = g_malloc0(size);
837
838 strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
839 dh->header_version = cpu_to_dump32(s, 6);
840 block_size = TARGET_PAGE_SIZE;
841 dh->block_size = cpu_to_dump32(s, block_size);
842 sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
843 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
844 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
845 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
846 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
847 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
848 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
849 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
850 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
851
852 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
853 status |= DUMP_DH_COMPRESSED_ZLIB;
854 }
855 #ifdef CONFIG_LZO
856 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
857 status |= DUMP_DH_COMPRESSED_LZO;
858 }
859 #endif
860 #ifdef CONFIG_SNAPPY
861 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
862 status |= DUMP_DH_COMPRESSED_SNAPPY;
863 }
864 #endif
865 dh->status = cpu_to_dump32(s, status);
866
867 if (write_buffer(s->fd, 0, dh, size) < 0) {
868 dump_error(s, "dump: failed to write disk dump header", errp);
869 goto out;
870 }
871
872 /* write sub header */
873 size = sizeof(KdumpSubHeader64);
874 kh = g_malloc0(size);
875
876 /* 64bit max_mapnr_64 */
877 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
878 kh->phys_base = cpu_to_dump64(s, PHYS_BASE);
879 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
880
881 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
882 kh->offset_note = cpu_to_dump64(s, offset_note);
883 kh->note_size = cpu_to_dump64(s, s->note_size);
884
885 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
886 block_size, kh, size) < 0) {
887 dump_error(s, "dump: failed to write kdump sub header", errp);
888 goto out;
889 }
890
891 /* write note */
892 s->note_buf = g_malloc0(s->note_size);
893 s->note_buf_offset = 0;
894
895 /* use s->note_buf to store notes temporarily */
896 write_elf64_notes(buf_write_note, s, &local_err);
897 if (local_err) {
898 error_propagate(errp, local_err);
899 goto out;
900 }
901
902 if (write_buffer(s->fd, offset_note, s->note_buf,
903 s->note_size) < 0) {
904 dump_error(s, "dump: failed to write notes", errp);
905 goto out;
906 }
907
908 /* get offset of dump_bitmap */
909 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
910 block_size;
911
912 /* get offset of page */
913 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
914 block_size;
915
916 out:
917 g_free(dh);
918 g_free(kh);
919 g_free(s->note_buf);
920 }
921
922 static void write_dump_header(DumpState *s, Error **errp)
923 {
924 Error *local_err = NULL;
925
926 if (s->dump_info.d_class == ELFCLASS32) {
927 create_header32(s, &local_err);
928 } else {
929 create_header64(s, &local_err);
930 }
931 if (local_err) {
932 error_propagate(errp, local_err);
933 }
934 }
935
936 /*
937 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
938 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
939 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
940 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
941 * vmcore, ie. synchronizing un-sync bit into vmcore.
942 */
943 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
944 uint8_t *buf, DumpState *s)
945 {
946 off_t old_offset, new_offset;
947 off_t offset_bitmap1, offset_bitmap2;
948 uint32_t byte, bit;
949
950 /* should not set the previous place */
951 assert(last_pfn <= pfn);
952
953 /*
954 * if the bit needed to be set is not cached in buf, flush the data in buf
955 * to vmcore firstly.
956 * making new_offset be bigger than old_offset can also sync remained data
957 * into vmcore.
958 */
959 old_offset = BUFSIZE_BITMAP * (last_pfn / PFN_BUFBITMAP);
960 new_offset = BUFSIZE_BITMAP * (pfn / PFN_BUFBITMAP);
961
962 while (old_offset < new_offset) {
963 /* calculate the offset and write dump_bitmap */
964 offset_bitmap1 = s->offset_dump_bitmap + old_offset;
965 if (write_buffer(s->fd, offset_bitmap1, buf,
966 BUFSIZE_BITMAP) < 0) {
967 return -1;
968 }
969
970 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
971 offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
972 old_offset;
973 if (write_buffer(s->fd, offset_bitmap2, buf,
974 BUFSIZE_BITMAP) < 0) {
975 return -1;
976 }
977
978 memset(buf, 0, BUFSIZE_BITMAP);
979 old_offset += BUFSIZE_BITMAP;
980 }
981
982 /* get the exact place of the bit in the buf, and set it */
983 byte = (pfn % PFN_BUFBITMAP) / CHAR_BIT;
984 bit = (pfn % PFN_BUFBITMAP) % CHAR_BIT;
985 if (value) {
986 buf[byte] |= 1u << bit;
987 } else {
988 buf[byte] &= ~(1u << bit);
989 }
990
991 return 0;
992 }
993
994 /*
995 * exam every page and return the page frame number and the address of the page.
996 * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
997 * blocks, so block->target_start and block->target_end should be interal
998 * multiples of the target page size.
999 */
1000 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1001 uint8_t **bufptr, DumpState *s)
1002 {
1003 GuestPhysBlock *block = *blockptr;
1004 hwaddr addr;
1005 uint8_t *buf;
1006
1007 /* block == NULL means the start of the iteration */
1008 if (!block) {
1009 block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1010 *blockptr = block;
1011 assert((block->target_start & ~TARGET_PAGE_MASK) == 0);
1012 assert((block->target_end & ~TARGET_PAGE_MASK) == 0);
1013 *pfnptr = paddr_to_pfn(block->target_start);
1014 if (bufptr) {
1015 *bufptr = block->host_addr;
1016 }
1017 return true;
1018 }
1019
1020 *pfnptr = *pfnptr + 1;
1021 addr = pfn_to_paddr(*pfnptr);
1022
1023 if ((addr >= block->target_start) &&
1024 (addr + TARGET_PAGE_SIZE <= block->target_end)) {
1025 buf = block->host_addr + (addr - block->target_start);
1026 } else {
1027 /* the next page is in the next block */
1028 block = QTAILQ_NEXT(block, next);
1029 *blockptr = block;
1030 if (!block) {
1031 return false;
1032 }
1033 assert((block->target_start & ~TARGET_PAGE_MASK) == 0);
1034 assert((block->target_end & ~TARGET_PAGE_MASK) == 0);
1035 *pfnptr = paddr_to_pfn(block->target_start);
1036 buf = block->host_addr;
1037 }
1038
1039 if (bufptr) {
1040 *bufptr = buf;
1041 }
1042
1043 return true;
1044 }
1045
1046 static void write_dump_bitmap(DumpState *s, Error **errp)
1047 {
1048 int ret = 0;
1049 uint64_t last_pfn, pfn;
1050 void *dump_bitmap_buf;
1051 size_t num_dumpable;
1052 GuestPhysBlock *block_iter = NULL;
1053
1054 /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1055 dump_bitmap_buf = g_malloc0(BUFSIZE_BITMAP);
1056
1057 num_dumpable = 0;
1058 last_pfn = 0;
1059
1060 /*
1061 * exam memory page by page, and set the bit in dump_bitmap corresponded
1062 * to the existing page.
1063 */
1064 while (get_next_page(&block_iter, &pfn, NULL, s)) {
1065 ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1066 if (ret < 0) {
1067 dump_error(s, "dump: failed to set dump_bitmap", errp);
1068 goto out;
1069 }
1070
1071 last_pfn = pfn;
1072 num_dumpable++;
1073 }
1074
1075 /*
1076 * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1077 * set last_pfn + PFN_BUFBITMAP to 0 and those set but un-sync bit will be
1078 * synchronized into vmcore.
1079 */
1080 if (num_dumpable > 0) {
1081 ret = set_dump_bitmap(last_pfn, last_pfn + PFN_BUFBITMAP, false,
1082 dump_bitmap_buf, s);
1083 if (ret < 0) {
1084 dump_error(s, "dump: failed to sync dump_bitmap", errp);
1085 goto out;
1086 }
1087 }
1088
1089 /* number of dumpable pages that will be dumped later */
1090 s->num_dumpable = num_dumpable;
1091
1092 out:
1093 g_free(dump_bitmap_buf);
1094 }
1095
1096 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1097 off_t offset)
1098 {
1099 data_cache->fd = s->fd;
1100 data_cache->data_size = 0;
1101 data_cache->buf_size = BUFSIZE_DATA_CACHE;
1102 data_cache->buf = g_malloc0(BUFSIZE_DATA_CACHE);
1103 data_cache->offset = offset;
1104 }
1105
1106 static int write_cache(DataCache *dc, const void *buf, size_t size,
1107 bool flag_sync)
1108 {
1109 /*
1110 * dc->buf_size should not be less than size, otherwise dc will never be
1111 * enough
1112 */
1113 assert(size <= dc->buf_size);
1114
1115 /*
1116 * if flag_sync is set, synchronize data in dc->buf into vmcore.
1117 * otherwise check if the space is enough for caching data in buf, if not,
1118 * write the data in dc->buf to dc->fd and reset dc->buf
1119 */
1120 if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1121 (flag_sync && dc->data_size > 0)) {
1122 if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1123 return -1;
1124 }
1125
1126 dc->offset += dc->data_size;
1127 dc->data_size = 0;
1128 }
1129
1130 if (!flag_sync) {
1131 memcpy(dc->buf + dc->data_size, buf, size);
1132 dc->data_size += size;
1133 }
1134
1135 return 0;
1136 }
1137
1138 static void free_data_cache(DataCache *data_cache)
1139 {
1140 g_free(data_cache->buf);
1141 }
1142
1143 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1144 {
1145 switch (flag_compress) {
1146 case DUMP_DH_COMPRESSED_ZLIB:
1147 return compressBound(page_size);
1148
1149 case DUMP_DH_COMPRESSED_LZO:
1150 /*
1151 * LZO will expand incompressible data by a little amount. Please check
1152 * the following URL to see the expansion calculation:
1153 * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1154 */
1155 return page_size + page_size / 16 + 64 + 3;
1156
1157 #ifdef CONFIG_SNAPPY
1158 case DUMP_DH_COMPRESSED_SNAPPY:
1159 return snappy_max_compressed_length(page_size);
1160 #endif
1161 }
1162 return 0;
1163 }
1164
1165 /*
1166 * check if the page is all 0
1167 */
1168 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1169 {
1170 return buffer_is_zero(buf, page_size);
1171 }
1172
1173 static void write_dump_pages(DumpState *s, Error **errp)
1174 {
1175 int ret = 0;
1176 DataCache page_desc, page_data;
1177 size_t len_buf_out, size_out;
1178 #ifdef CONFIG_LZO
1179 lzo_bytep wrkmem = NULL;
1180 #endif
1181 uint8_t *buf_out = NULL;
1182 off_t offset_desc, offset_data;
1183 PageDescriptor pd, pd_zero;
1184 uint8_t *buf;
1185 GuestPhysBlock *block_iter = NULL;
1186 uint64_t pfn_iter;
1187
1188 /* get offset of page_desc and page_data in dump file */
1189 offset_desc = s->offset_page;
1190 offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1191
1192 prepare_data_cache(&page_desc, s, offset_desc);
1193 prepare_data_cache(&page_data, s, offset_data);
1194
1195 /* prepare buffer to store compressed data */
1196 len_buf_out = get_len_buf_out(TARGET_PAGE_SIZE, s->flag_compress);
1197 assert(len_buf_out != 0);
1198
1199 #ifdef CONFIG_LZO
1200 wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1201 #endif
1202
1203 buf_out = g_malloc(len_buf_out);
1204
1205 /*
1206 * init zero page's page_desc and page_data, because every zero page
1207 * uses the same page_data
1208 */
1209 pd_zero.size = cpu_to_dump32(s, TARGET_PAGE_SIZE);
1210 pd_zero.flags = cpu_to_dump32(s, 0);
1211 pd_zero.offset = cpu_to_dump64(s, offset_data);
1212 pd_zero.page_flags = cpu_to_dump64(s, 0);
1213 buf = g_malloc0(TARGET_PAGE_SIZE);
1214 ret = write_cache(&page_data, buf, TARGET_PAGE_SIZE, false);
1215 g_free(buf);
1216 if (ret < 0) {
1217 dump_error(s, "dump: failed to write page data (zero page)", errp);
1218 goto out;
1219 }
1220
1221 offset_data += TARGET_PAGE_SIZE;
1222
1223 /*
1224 * dump memory to vmcore page by page. zero page will all be resided in the
1225 * first page of page section
1226 */
1227 while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1228 /* check zero page */
1229 if (is_zero_page(buf, TARGET_PAGE_SIZE)) {
1230 ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1231 false);
1232 if (ret < 0) {
1233 dump_error(s, "dump: failed to write page desc", errp);
1234 goto out;
1235 }
1236 } else {
1237 /*
1238 * not zero page, then:
1239 * 1. compress the page
1240 * 2. write the compressed page into the cache of page_data
1241 * 3. get page desc of the compressed page and write it into the
1242 * cache of page_desc
1243 *
1244 * only one compression format will be used here, for
1245 * s->flag_compress is set. But when compression fails to work,
1246 * we fall back to save in plaintext.
1247 */
1248 size_out = len_buf_out;
1249 if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1250 (compress2(buf_out, (uLongf *)&size_out, buf,
1251 TARGET_PAGE_SIZE, Z_BEST_SPEED) == Z_OK) &&
1252 (size_out < TARGET_PAGE_SIZE)) {
1253 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1254 pd.size = cpu_to_dump32(s, size_out);
1255
1256 ret = write_cache(&page_data, buf_out, size_out, false);
1257 if (ret < 0) {
1258 dump_error(s, "dump: failed to write page data", errp);
1259 goto out;
1260 }
1261 #ifdef CONFIG_LZO
1262 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1263 (lzo1x_1_compress(buf, TARGET_PAGE_SIZE, buf_out,
1264 (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1265 (size_out < TARGET_PAGE_SIZE)) {
1266 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1267 pd.size = cpu_to_dump32(s, size_out);
1268
1269 ret = write_cache(&page_data, buf_out, size_out, false);
1270 if (ret < 0) {
1271 dump_error(s, "dump: failed to write page data", errp);
1272 goto out;
1273 }
1274 #endif
1275 #ifdef CONFIG_SNAPPY
1276 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1277 (snappy_compress((char *)buf, TARGET_PAGE_SIZE,
1278 (char *)buf_out, &size_out) == SNAPPY_OK) &&
1279 (size_out < TARGET_PAGE_SIZE)) {
1280 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1281 pd.size = cpu_to_dump32(s, size_out);
1282
1283 ret = write_cache(&page_data, buf_out, size_out, false);
1284 if (ret < 0) {
1285 dump_error(s, "dump: failed to write page data", errp);
1286 goto out;
1287 }
1288 #endif
1289 } else {
1290 /*
1291 * fall back to save in plaintext, size_out should be
1292 * assigned TARGET_PAGE_SIZE
1293 */
1294 pd.flags = cpu_to_dump32(s, 0);
1295 size_out = TARGET_PAGE_SIZE;
1296 pd.size = cpu_to_dump32(s, size_out);
1297
1298 ret = write_cache(&page_data, buf, TARGET_PAGE_SIZE, false);
1299 if (ret < 0) {
1300 dump_error(s, "dump: failed to write page data", errp);
1301 goto out;
1302 }
1303 }
1304
1305 /* get and write page desc here */
1306 pd.page_flags = cpu_to_dump64(s, 0);
1307 pd.offset = cpu_to_dump64(s, offset_data);
1308 offset_data += size_out;
1309
1310 ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1311 if (ret < 0) {
1312 dump_error(s, "dump: failed to write page desc", errp);
1313 goto out;
1314 }
1315 }
1316 }
1317
1318 ret = write_cache(&page_desc, NULL, 0, true);
1319 if (ret < 0) {
1320 dump_error(s, "dump: failed to sync cache for page_desc", errp);
1321 goto out;
1322 }
1323 ret = write_cache(&page_data, NULL, 0, true);
1324 if (ret < 0) {
1325 dump_error(s, "dump: failed to sync cache for page_data", errp);
1326 goto out;
1327 }
1328
1329 out:
1330 free_data_cache(&page_desc);
1331 free_data_cache(&page_data);
1332
1333 #ifdef CONFIG_LZO
1334 g_free(wrkmem);
1335 #endif
1336
1337 g_free(buf_out);
1338 }
1339
1340 static void create_kdump_vmcore(DumpState *s, Error **errp)
1341 {
1342 int ret;
1343 Error *local_err = NULL;
1344
1345 /*
1346 * the kdump-compressed format is:
1347 * File offset
1348 * +------------------------------------------+ 0x0
1349 * | main header (struct disk_dump_header) |
1350 * |------------------------------------------+ block 1
1351 * | sub header (struct kdump_sub_header) |
1352 * |------------------------------------------+ block 2
1353 * | 1st-dump_bitmap |
1354 * |------------------------------------------+ block 2 + X blocks
1355 * | 2nd-dump_bitmap | (aligned by block)
1356 * |------------------------------------------+ block 2 + 2 * X blocks
1357 * | page desc for pfn 0 (struct page_desc) | (aligned by block)
1358 * | page desc for pfn 1 (struct page_desc) |
1359 * | : |
1360 * |------------------------------------------| (not aligned by block)
1361 * | page data (pfn 0) |
1362 * | page data (pfn 1) |
1363 * | : |
1364 * +------------------------------------------+
1365 */
1366
1367 ret = write_start_flat_header(s->fd);
1368 if (ret < 0) {
1369 dump_error(s, "dump: failed to write start flat header", errp);
1370 return;
1371 }
1372
1373 write_dump_header(s, &local_err);
1374 if (local_err) {
1375 error_propagate(errp, local_err);
1376 return;
1377 }
1378
1379 write_dump_bitmap(s, &local_err);
1380 if (local_err) {
1381 error_propagate(errp, local_err);
1382 return;
1383 }
1384
1385 write_dump_pages(s, &local_err);
1386 if (local_err) {
1387 error_propagate(errp, local_err);
1388 return;
1389 }
1390
1391 ret = write_end_flat_header(s->fd);
1392 if (ret < 0) {
1393 dump_error(s, "dump: failed to write end flat header", errp);
1394 return;
1395 }
1396
1397 dump_completed(s);
1398 }
1399
1400 static ram_addr_t get_start_block(DumpState *s)
1401 {
1402 GuestPhysBlock *block;
1403
1404 if (!s->has_filter) {
1405 s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1406 return 0;
1407 }
1408
1409 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1410 if (block->target_start >= s->begin + s->length ||
1411 block->target_end <= s->begin) {
1412 /* This block is out of the range */
1413 continue;
1414 }
1415
1416 s->next_block = block;
1417 if (s->begin > block->target_start) {
1418 s->start = s->begin - block->target_start;
1419 } else {
1420 s->start = 0;
1421 }
1422 return s->start;
1423 }
1424
1425 return -1;
1426 }
1427
1428 static void get_max_mapnr(DumpState *s)
1429 {
1430 GuestPhysBlock *last_block;
1431
1432 last_block = QTAILQ_LAST(&s->guest_phys_blocks.head, GuestPhysBlockHead);
1433 s->max_mapnr = paddr_to_pfn(last_block->target_end);
1434 }
1435
1436 static void dump_init(DumpState *s, int fd, bool has_format,
1437 DumpGuestMemoryFormat format, bool paging, bool has_filter,
1438 int64_t begin, int64_t length, Error **errp)
1439 {
1440 CPUState *cpu;
1441 int nr_cpus;
1442 Error *err = NULL;
1443 int ret;
1444
1445 /* kdump-compressed is conflict with paging and filter */
1446 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1447 assert(!paging && !has_filter);
1448 }
1449
1450 if (runstate_is_running()) {
1451 vm_stop(RUN_STATE_SAVE_VM);
1452 s->resume = true;
1453 } else {
1454 s->resume = false;
1455 }
1456
1457 /* If we use KVM, we should synchronize the registers before we get dump
1458 * info or physmap info.
1459 */
1460 cpu_synchronize_all_states();
1461 nr_cpus = 0;
1462 CPU_FOREACH(cpu) {
1463 nr_cpus++;
1464 }
1465
1466 s->fd = fd;
1467 s->has_filter = has_filter;
1468 s->begin = begin;
1469 s->length = length;
1470
1471 memory_mapping_list_init(&s->list);
1472
1473 guest_phys_blocks_init(&s->guest_phys_blocks);
1474 guest_phys_blocks_append(&s->guest_phys_blocks);
1475
1476 s->start = get_start_block(s);
1477 if (s->start == -1) {
1478 error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1479 goto cleanup;
1480 }
1481
1482 /* get dump info: endian, class and architecture.
1483 * If the target architecture is not supported, cpu_get_dump_info() will
1484 * return -1.
1485 */
1486 ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1487 if (ret < 0) {
1488 error_setg(errp, QERR_UNSUPPORTED);
1489 goto cleanup;
1490 }
1491
1492 s->note_size = cpu_get_note_size(s->dump_info.d_class,
1493 s->dump_info.d_machine, nr_cpus);
1494 if (s->note_size < 0) {
1495 error_setg(errp, QERR_UNSUPPORTED);
1496 goto cleanup;
1497 }
1498
1499 /* get memory mapping */
1500 if (paging) {
1501 qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1502 if (err != NULL) {
1503 error_propagate(errp, err);
1504 goto cleanup;
1505 }
1506 } else {
1507 qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1508 }
1509
1510 s->nr_cpus = nr_cpus;
1511
1512 get_max_mapnr(s);
1513
1514 uint64_t tmp;
1515 tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT), TARGET_PAGE_SIZE);
1516 s->len_dump_bitmap = tmp * TARGET_PAGE_SIZE;
1517
1518 /* init for kdump-compressed format */
1519 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1520 switch (format) {
1521 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1522 s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1523 break;
1524
1525 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1526 #ifdef CONFIG_LZO
1527 if (lzo_init() != LZO_E_OK) {
1528 error_setg(errp, "failed to initialize the LZO library");
1529 goto cleanup;
1530 }
1531 #endif
1532 s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1533 break;
1534
1535 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1536 s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1537 break;
1538
1539 default:
1540 s->flag_compress = 0;
1541 }
1542
1543 return;
1544 }
1545
1546 if (s->has_filter) {
1547 memory_mapping_filter(&s->list, s->begin, s->length);
1548 }
1549
1550 /*
1551 * calculate phdr_num
1552 *
1553 * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1554 */
1555 s->phdr_num = 1; /* PT_NOTE */
1556 if (s->list.num < UINT16_MAX - 2) {
1557 s->phdr_num += s->list.num;
1558 s->have_section = false;
1559 } else {
1560 s->have_section = true;
1561 s->phdr_num = PN_XNUM;
1562 s->sh_info = 1; /* PT_NOTE */
1563
1564 /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1565 if (s->list.num <= UINT32_MAX - 1) {
1566 s->sh_info += s->list.num;
1567 } else {
1568 s->sh_info = UINT32_MAX;
1569 }
1570 }
1571
1572 if (s->dump_info.d_class == ELFCLASS64) {
1573 if (s->have_section) {
1574 s->memory_offset = sizeof(Elf64_Ehdr) +
1575 sizeof(Elf64_Phdr) * s->sh_info +
1576 sizeof(Elf64_Shdr) + s->note_size;
1577 } else {
1578 s->memory_offset = sizeof(Elf64_Ehdr) +
1579 sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1580 }
1581 } else {
1582 if (s->have_section) {
1583 s->memory_offset = sizeof(Elf32_Ehdr) +
1584 sizeof(Elf32_Phdr) * s->sh_info +
1585 sizeof(Elf32_Shdr) + s->note_size;
1586 } else {
1587 s->memory_offset = sizeof(Elf32_Ehdr) +
1588 sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1589 }
1590 }
1591
1592 return;
1593
1594 cleanup:
1595 dump_cleanup(s);
1596 }
1597
1598 void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
1599 int64_t begin, bool has_length,
1600 int64_t length, bool has_format,
1601 DumpGuestMemoryFormat format, Error **errp)
1602 {
1603 const char *p;
1604 int fd = -1;
1605 DumpState *s;
1606 Error *local_err = NULL;
1607
1608 /*
1609 * kdump-compressed format need the whole memory dumped, so paging or
1610 * filter is not supported here.
1611 */
1612 if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1613 (paging || has_begin || has_length)) {
1614 error_setg(errp, "kdump-compressed format doesn't support paging or "
1615 "filter");
1616 return;
1617 }
1618 if (has_begin && !has_length) {
1619 error_setg(errp, QERR_MISSING_PARAMETER, "length");
1620 return;
1621 }
1622 if (!has_begin && has_length) {
1623 error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1624 return;
1625 }
1626
1627 /* check whether lzo/snappy is supported */
1628 #ifndef CONFIG_LZO
1629 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1630 error_setg(errp, "kdump-lzo is not available now");
1631 return;
1632 }
1633 #endif
1634
1635 #ifndef CONFIG_SNAPPY
1636 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1637 error_setg(errp, "kdump-snappy is not available now");
1638 return;
1639 }
1640 #endif
1641
1642 #if !defined(WIN32)
1643 if (strstart(file, "fd:", &p)) {
1644 fd = monitor_get_fd(cur_mon, p, errp);
1645 if (fd == -1) {
1646 return;
1647 }
1648 }
1649 #endif
1650
1651 if (strstart(file, "file:", &p)) {
1652 fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1653 if (fd < 0) {
1654 error_setg_file_open(errp, errno, p);
1655 return;
1656 }
1657 }
1658
1659 if (fd == -1) {
1660 error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
1661 return;
1662 }
1663
1664 s = g_malloc0(sizeof(DumpState));
1665
1666 dump_init(s, fd, has_format, format, paging, has_begin,
1667 begin, length, &local_err);
1668 if (local_err) {
1669 g_free(s);
1670 error_propagate(errp, local_err);
1671 return;
1672 }
1673
1674 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1675 create_kdump_vmcore(s, errp);
1676 } else {
1677 create_vmcore(s, errp);
1678 }
1679
1680 g_free(s);
1681 }
1682
1683 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
1684 {
1685 DumpGuestMemoryFormatList *item;
1686 DumpGuestMemoryCapability *cap =
1687 g_malloc0(sizeof(DumpGuestMemoryCapability));
1688
1689 /* elf is always available */
1690 item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1691 cap->formats = item;
1692 item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
1693
1694 /* kdump-zlib is always available */
1695 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1696 item = item->next;
1697 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
1698
1699 /* add new item if kdump-lzo is available */
1700 #ifdef CONFIG_LZO
1701 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1702 item = item->next;
1703 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
1704 #endif
1705
1706 /* add new item if kdump-snappy is available */
1707 #ifdef CONFIG_SNAPPY
1708 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1709 item = item->next;
1710 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
1711 #endif
1712
1713 return cap;
1714 }