docker: Add EXTRA_CONFIGURE_OPTS
[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/osdep.h"
15 #include "qemu/cutils.h"
16 #include "elf.h"
17 #include "cpu.h"
18 #include "exec/cpu-all.h"
19 #include "exec/hwaddr.h"
20 #include "monitor/monitor.h"
21 #include "sysemu/kvm.h"
22 #include "sysemu/dump.h"
23 #include "sysemu/sysemu.h"
24 #include "sysemu/memory_mapping.h"
25 #include "sysemu/cpus.h"
26 #include "qapi/qmp/qerror.h"
27 #include "qmp-commands.h"
28 #include "qapi-event.h"
29
30 #include <zlib.h>
31 #ifdef CONFIG_LZO
32 #include <lzo/lzo1x.h>
33 #endif
34 #ifdef CONFIG_SNAPPY
35 #include <snappy-c.h>
36 #endif
37 #ifndef ELF_MACHINE_UNAME
38 #define ELF_MACHINE_UNAME "Unknown"
39 #endif
40
41 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
42 {
43 if (s->dump_info.d_endian == ELFDATA2LSB) {
44 val = cpu_to_le16(val);
45 } else {
46 val = cpu_to_be16(val);
47 }
48
49 return val;
50 }
51
52 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
53 {
54 if (s->dump_info.d_endian == ELFDATA2LSB) {
55 val = cpu_to_le32(val);
56 } else {
57 val = cpu_to_be32(val);
58 }
59
60 return val;
61 }
62
63 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
64 {
65 if (s->dump_info.d_endian == ELFDATA2LSB) {
66 val = cpu_to_le64(val);
67 } else {
68 val = cpu_to_be64(val);
69 }
70
71 return val;
72 }
73
74 static int dump_cleanup(DumpState *s)
75 {
76 guest_phys_blocks_free(&s->guest_phys_blocks);
77 memory_mapping_list_free(&s->list);
78 close(s->fd);
79 if (s->resume) {
80 vm_start();
81 }
82
83 return 0;
84 }
85
86 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
87 {
88 DumpState *s = opaque;
89 size_t written_size;
90
91 written_size = qemu_write_full(s->fd, buf, size);
92 if (written_size != size) {
93 return -1;
94 }
95
96 return 0;
97 }
98
99 static void write_elf64_header(DumpState *s, Error **errp)
100 {
101 Elf64_Ehdr elf_header;
102 int ret;
103
104 memset(&elf_header, 0, sizeof(Elf64_Ehdr));
105 memcpy(&elf_header, ELFMAG, SELFMAG);
106 elf_header.e_ident[EI_CLASS] = ELFCLASS64;
107 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
108 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
109 elf_header.e_type = cpu_to_dump16(s, ET_CORE);
110 elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
111 elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
112 elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
113 elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
114 elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
115 elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
116 if (s->have_section) {
117 uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
118
119 elf_header.e_shoff = cpu_to_dump64(s, shoff);
120 elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
121 elf_header.e_shnum = cpu_to_dump16(s, 1);
122 }
123
124 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
125 if (ret < 0) {
126 error_setg(errp, "dump: failed to write elf header");
127 }
128 }
129
130 static void write_elf32_header(DumpState *s, Error **errp)
131 {
132 Elf32_Ehdr elf_header;
133 int ret;
134
135 memset(&elf_header, 0, sizeof(Elf32_Ehdr));
136 memcpy(&elf_header, ELFMAG, SELFMAG);
137 elf_header.e_ident[EI_CLASS] = ELFCLASS32;
138 elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
139 elf_header.e_ident[EI_VERSION] = EV_CURRENT;
140 elf_header.e_type = cpu_to_dump16(s, ET_CORE);
141 elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
142 elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
143 elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
144 elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
145 elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
146 elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
147 if (s->have_section) {
148 uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
149
150 elf_header.e_shoff = cpu_to_dump32(s, shoff);
151 elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
152 elf_header.e_shnum = cpu_to_dump16(s, 1);
153 }
154
155 ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
156 if (ret < 0) {
157 error_setg(errp, "dump: failed to write elf header");
158 }
159 }
160
161 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
162 int phdr_index, hwaddr offset,
163 hwaddr filesz, Error **errp)
164 {
165 Elf64_Phdr phdr;
166 int ret;
167
168 memset(&phdr, 0, sizeof(Elf64_Phdr));
169 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
170 phdr.p_offset = cpu_to_dump64(s, offset);
171 phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
172 phdr.p_filesz = cpu_to_dump64(s, filesz);
173 phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
174 phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr);
175
176 assert(memory_mapping->length >= filesz);
177
178 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
179 if (ret < 0) {
180 error_setg(errp, "dump: failed to write program header table");
181 }
182 }
183
184 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
185 int phdr_index, hwaddr offset,
186 hwaddr filesz, Error **errp)
187 {
188 Elf32_Phdr phdr;
189 int ret;
190
191 memset(&phdr, 0, sizeof(Elf32_Phdr));
192 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
193 phdr.p_offset = cpu_to_dump32(s, offset);
194 phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
195 phdr.p_filesz = cpu_to_dump32(s, filesz);
196 phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
197 phdr.p_vaddr = cpu_to_dump32(s, memory_mapping->virt_addr);
198
199 assert(memory_mapping->length >= filesz);
200
201 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
202 if (ret < 0) {
203 error_setg(errp, "dump: failed to write program header table");
204 }
205 }
206
207 static void write_elf64_note(DumpState *s, Error **errp)
208 {
209 Elf64_Phdr phdr;
210 hwaddr begin = s->memory_offset - s->note_size;
211 int ret;
212
213 memset(&phdr, 0, sizeof(Elf64_Phdr));
214 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
215 phdr.p_offset = cpu_to_dump64(s, begin);
216 phdr.p_paddr = 0;
217 phdr.p_filesz = cpu_to_dump64(s, s->note_size);
218 phdr.p_memsz = cpu_to_dump64(s, s->note_size);
219 phdr.p_vaddr = 0;
220
221 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
222 if (ret < 0) {
223 error_setg(errp, "dump: failed to write program header table");
224 }
225 }
226
227 static inline int cpu_index(CPUState *cpu)
228 {
229 return cpu->cpu_index + 1;
230 }
231
232 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
233 Error **errp)
234 {
235 CPUState *cpu;
236 int ret;
237 int id;
238
239 CPU_FOREACH(cpu) {
240 id = cpu_index(cpu);
241 ret = cpu_write_elf64_note(f, cpu, id, s);
242 if (ret < 0) {
243 error_setg(errp, "dump: failed to write elf notes");
244 return;
245 }
246 }
247
248 CPU_FOREACH(cpu) {
249 ret = cpu_write_elf64_qemunote(f, cpu, s);
250 if (ret < 0) {
251 error_setg(errp, "dump: failed to write CPU status");
252 return;
253 }
254 }
255 }
256
257 static void write_elf32_note(DumpState *s, Error **errp)
258 {
259 hwaddr begin = s->memory_offset - s->note_size;
260 Elf32_Phdr phdr;
261 int ret;
262
263 memset(&phdr, 0, sizeof(Elf32_Phdr));
264 phdr.p_type = cpu_to_dump32(s, PT_NOTE);
265 phdr.p_offset = cpu_to_dump32(s, begin);
266 phdr.p_paddr = 0;
267 phdr.p_filesz = cpu_to_dump32(s, s->note_size);
268 phdr.p_memsz = cpu_to_dump32(s, s->note_size);
269 phdr.p_vaddr = 0;
270
271 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
272 if (ret < 0) {
273 error_setg(errp, "dump: failed to write program header table");
274 }
275 }
276
277 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
278 Error **errp)
279 {
280 CPUState *cpu;
281 int ret;
282 int id;
283
284 CPU_FOREACH(cpu) {
285 id = cpu_index(cpu);
286 ret = cpu_write_elf32_note(f, cpu, id, s);
287 if (ret < 0) {
288 error_setg(errp, "dump: failed to write elf notes");
289 return;
290 }
291 }
292
293 CPU_FOREACH(cpu) {
294 ret = cpu_write_elf32_qemunote(f, cpu, s);
295 if (ret < 0) {
296 error_setg(errp, "dump: failed to write CPU status");
297 return;
298 }
299 }
300 }
301
302 static void write_elf_section(DumpState *s, int type, Error **errp)
303 {
304 Elf32_Shdr shdr32;
305 Elf64_Shdr shdr64;
306 int shdr_size;
307 void *shdr;
308 int ret;
309
310 if (type == 0) {
311 shdr_size = sizeof(Elf32_Shdr);
312 memset(&shdr32, 0, shdr_size);
313 shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
314 shdr = &shdr32;
315 } else {
316 shdr_size = sizeof(Elf64_Shdr);
317 memset(&shdr64, 0, shdr_size);
318 shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
319 shdr = &shdr64;
320 }
321
322 ret = fd_write_vmcore(&shdr, shdr_size, s);
323 if (ret < 0) {
324 error_setg(errp, "dump: failed to write section header table");
325 }
326 }
327
328 static void write_data(DumpState *s, void *buf, int length, Error **errp)
329 {
330 int ret;
331
332 ret = fd_write_vmcore(buf, length, s);
333 if (ret < 0) {
334 error_setg(errp, "dump: failed to save memory");
335 } else {
336 s->written_size += length;
337 }
338 }
339
340 /* write the memory to vmcore. 1 page per I/O. */
341 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
342 int64_t size, Error **errp)
343 {
344 int64_t i;
345 Error *local_err = NULL;
346
347 for (i = 0; i < size / s->dump_info.page_size; i++) {
348 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
349 s->dump_info.page_size, &local_err);
350 if (local_err) {
351 error_propagate(errp, local_err);
352 return;
353 }
354 }
355
356 if ((size % s->dump_info.page_size) != 0) {
357 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
358 size % s->dump_info.page_size, &local_err);
359 if (local_err) {
360 error_propagate(errp, local_err);
361 return;
362 }
363 }
364 }
365
366 /* get the memory's offset and size in the vmcore */
367 static void get_offset_range(hwaddr phys_addr,
368 ram_addr_t mapping_length,
369 DumpState *s,
370 hwaddr *p_offset,
371 hwaddr *p_filesz)
372 {
373 GuestPhysBlock *block;
374 hwaddr offset = s->memory_offset;
375 int64_t size_in_block, start;
376
377 /* When the memory is not stored into vmcore, offset will be -1 */
378 *p_offset = -1;
379 *p_filesz = 0;
380
381 if (s->has_filter) {
382 if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
383 return;
384 }
385 }
386
387 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
388 if (s->has_filter) {
389 if (block->target_start >= s->begin + s->length ||
390 block->target_end <= s->begin) {
391 /* This block is out of the range */
392 continue;
393 }
394
395 if (s->begin <= block->target_start) {
396 start = block->target_start;
397 } else {
398 start = s->begin;
399 }
400
401 size_in_block = block->target_end - start;
402 if (s->begin + s->length < block->target_end) {
403 size_in_block -= block->target_end - (s->begin + s->length);
404 }
405 } else {
406 start = block->target_start;
407 size_in_block = block->target_end - block->target_start;
408 }
409
410 if (phys_addr >= start && phys_addr < start + size_in_block) {
411 *p_offset = phys_addr - start + offset;
412
413 /* The offset range mapped from the vmcore file must not spill over
414 * the GuestPhysBlock, clamp it. The rest of the mapping will be
415 * zero-filled in memory at load time; see
416 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
417 */
418 *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
419 mapping_length :
420 size_in_block - (phys_addr - start);
421 return;
422 }
423
424 offset += size_in_block;
425 }
426 }
427
428 static void write_elf_loads(DumpState *s, Error **errp)
429 {
430 hwaddr offset, filesz;
431 MemoryMapping *memory_mapping;
432 uint32_t phdr_index = 1;
433 uint32_t max_index;
434 Error *local_err = NULL;
435
436 if (s->have_section) {
437 max_index = s->sh_info;
438 } else {
439 max_index = s->phdr_num;
440 }
441
442 QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
443 get_offset_range(memory_mapping->phys_addr,
444 memory_mapping->length,
445 s, &offset, &filesz);
446 if (s->dump_info.d_class == ELFCLASS64) {
447 write_elf64_load(s, memory_mapping, phdr_index++, offset,
448 filesz, &local_err);
449 } else {
450 write_elf32_load(s, memory_mapping, phdr_index++, offset,
451 filesz, &local_err);
452 }
453
454 if (local_err) {
455 error_propagate(errp, local_err);
456 return;
457 }
458
459 if (phdr_index >= max_index) {
460 break;
461 }
462 }
463 }
464
465 /* write elf header, PT_NOTE and elf note to vmcore. */
466 static void dump_begin(DumpState *s, Error **errp)
467 {
468 Error *local_err = NULL;
469
470 /*
471 * the vmcore's format is:
472 * --------------
473 * | elf header |
474 * --------------
475 * | PT_NOTE |
476 * --------------
477 * | PT_LOAD |
478 * --------------
479 * | ...... |
480 * --------------
481 * | PT_LOAD |
482 * --------------
483 * | sec_hdr |
484 * --------------
485 * | elf note |
486 * --------------
487 * | memory |
488 * --------------
489 *
490 * we only know where the memory is saved after we write elf note into
491 * vmcore.
492 */
493
494 /* write elf header to vmcore */
495 if (s->dump_info.d_class == ELFCLASS64) {
496 write_elf64_header(s, &local_err);
497 } else {
498 write_elf32_header(s, &local_err);
499 }
500 if (local_err) {
501 error_propagate(errp, local_err);
502 return;
503 }
504
505 if (s->dump_info.d_class == ELFCLASS64) {
506 /* write PT_NOTE to vmcore */
507 write_elf64_note(s, &local_err);
508 if (local_err) {
509 error_propagate(errp, local_err);
510 return;
511 }
512
513 /* write all PT_LOAD to vmcore */
514 write_elf_loads(s, &local_err);
515 if (local_err) {
516 error_propagate(errp, local_err);
517 return;
518 }
519
520 /* write section to vmcore */
521 if (s->have_section) {
522 write_elf_section(s, 1, &local_err);
523 if (local_err) {
524 error_propagate(errp, local_err);
525 return;
526 }
527 }
528
529 /* write notes to vmcore */
530 write_elf64_notes(fd_write_vmcore, s, &local_err);
531 if (local_err) {
532 error_propagate(errp, local_err);
533 return;
534 }
535 } else {
536 /* write PT_NOTE to vmcore */
537 write_elf32_note(s, &local_err);
538 if (local_err) {
539 error_propagate(errp, local_err);
540 return;
541 }
542
543 /* write all PT_LOAD to vmcore */
544 write_elf_loads(s, &local_err);
545 if (local_err) {
546 error_propagate(errp, local_err);
547 return;
548 }
549
550 /* write section to vmcore */
551 if (s->have_section) {
552 write_elf_section(s, 0, &local_err);
553 if (local_err) {
554 error_propagate(errp, local_err);
555 return;
556 }
557 }
558
559 /* write notes to vmcore */
560 write_elf32_notes(fd_write_vmcore, s, &local_err);
561 if (local_err) {
562 error_propagate(errp, local_err);
563 return;
564 }
565 }
566 }
567
568 static int get_next_block(DumpState *s, GuestPhysBlock *block)
569 {
570 while (1) {
571 block = QTAILQ_NEXT(block, next);
572 if (!block) {
573 /* no more block */
574 return 1;
575 }
576
577 s->start = 0;
578 s->next_block = block;
579 if (s->has_filter) {
580 if (block->target_start >= s->begin + s->length ||
581 block->target_end <= s->begin) {
582 /* This block is out of the range */
583 continue;
584 }
585
586 if (s->begin > block->target_start) {
587 s->start = s->begin - block->target_start;
588 }
589 }
590
591 return 0;
592 }
593 }
594
595 /* write all memory to vmcore */
596 static void dump_iterate(DumpState *s, Error **errp)
597 {
598 GuestPhysBlock *block;
599 int64_t size;
600 Error *local_err = NULL;
601
602 do {
603 block = s->next_block;
604
605 size = block->target_end - block->target_start;
606 if (s->has_filter) {
607 size -= s->start;
608 if (s->begin + s->length < block->target_end) {
609 size -= block->target_end - (s->begin + s->length);
610 }
611 }
612 write_memory(s, block, s->start, size, &local_err);
613 if (local_err) {
614 error_propagate(errp, local_err);
615 return;
616 }
617
618 } while (!get_next_block(s, block));
619 }
620
621 static void create_vmcore(DumpState *s, Error **errp)
622 {
623 Error *local_err = NULL;
624
625 dump_begin(s, &local_err);
626 if (local_err) {
627 error_propagate(errp, local_err);
628 return;
629 }
630
631 dump_iterate(s, errp);
632 }
633
634 static int write_start_flat_header(int fd)
635 {
636 MakedumpfileHeader *mh;
637 int ret = 0;
638
639 QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
640 mh = g_malloc0(MAX_SIZE_MDF_HEADER);
641
642 memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
643 MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
644
645 mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
646 mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
647
648 size_t written_size;
649 written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
650 if (written_size != MAX_SIZE_MDF_HEADER) {
651 ret = -1;
652 }
653
654 g_free(mh);
655 return ret;
656 }
657
658 static int write_end_flat_header(int fd)
659 {
660 MakedumpfileDataHeader mdh;
661
662 mdh.offset = END_FLAG_FLAT_HEADER;
663 mdh.buf_size = END_FLAG_FLAT_HEADER;
664
665 size_t written_size;
666 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
667 if (written_size != sizeof(mdh)) {
668 return -1;
669 }
670
671 return 0;
672 }
673
674 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
675 {
676 size_t written_size;
677 MakedumpfileDataHeader mdh;
678
679 mdh.offset = cpu_to_be64(offset);
680 mdh.buf_size = cpu_to_be64(size);
681
682 written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
683 if (written_size != sizeof(mdh)) {
684 return -1;
685 }
686
687 written_size = qemu_write_full(fd, buf, size);
688 if (written_size != size) {
689 return -1;
690 }
691
692 return 0;
693 }
694
695 static int buf_write_note(const void *buf, size_t size, void *opaque)
696 {
697 DumpState *s = opaque;
698
699 /* note_buf is not enough */
700 if (s->note_buf_offset + size > s->note_size) {
701 return -1;
702 }
703
704 memcpy(s->note_buf + s->note_buf_offset, buf, size);
705
706 s->note_buf_offset += size;
707
708 return 0;
709 }
710
711 /* write common header, sub header and elf note to vmcore */
712 static void create_header32(DumpState *s, Error **errp)
713 {
714 DiskDumpHeader32 *dh = NULL;
715 KdumpSubHeader32 *kh = NULL;
716 size_t size;
717 uint32_t block_size;
718 uint32_t sub_hdr_size;
719 uint32_t bitmap_blocks;
720 uint32_t status = 0;
721 uint64_t offset_note;
722 Error *local_err = NULL;
723
724 /* write common header, the version of kdump-compressed format is 6th */
725 size = sizeof(DiskDumpHeader32);
726 dh = g_malloc0(size);
727
728 strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
729 dh->header_version = cpu_to_dump32(s, 6);
730 block_size = s->dump_info.page_size;
731 dh->block_size = cpu_to_dump32(s, block_size);
732 sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
733 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
734 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
735 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
736 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
737 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
738 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
739 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
740 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
741
742 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
743 status |= DUMP_DH_COMPRESSED_ZLIB;
744 }
745 #ifdef CONFIG_LZO
746 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
747 status |= DUMP_DH_COMPRESSED_LZO;
748 }
749 #endif
750 #ifdef CONFIG_SNAPPY
751 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
752 status |= DUMP_DH_COMPRESSED_SNAPPY;
753 }
754 #endif
755 dh->status = cpu_to_dump32(s, status);
756
757 if (write_buffer(s->fd, 0, dh, size) < 0) {
758 error_setg(errp, "dump: failed to write disk dump header");
759 goto out;
760 }
761
762 /* write sub header */
763 size = sizeof(KdumpSubHeader32);
764 kh = g_malloc0(size);
765
766 /* 64bit max_mapnr_64 */
767 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
768 kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
769 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
770
771 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
772 kh->offset_note = cpu_to_dump64(s, offset_note);
773 kh->note_size = cpu_to_dump32(s, s->note_size);
774
775 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
776 block_size, kh, size) < 0) {
777 error_setg(errp, "dump: failed to write kdump sub header");
778 goto out;
779 }
780
781 /* write note */
782 s->note_buf = g_malloc0(s->note_size);
783 s->note_buf_offset = 0;
784
785 /* use s->note_buf to store notes temporarily */
786 write_elf32_notes(buf_write_note, s, &local_err);
787 if (local_err) {
788 error_propagate(errp, local_err);
789 goto out;
790 }
791 if (write_buffer(s->fd, offset_note, s->note_buf,
792 s->note_size) < 0) {
793 error_setg(errp, "dump: failed to write notes");
794 goto out;
795 }
796
797 /* get offset of dump_bitmap */
798 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
799 block_size;
800
801 /* get offset of page */
802 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
803 block_size;
804
805 out:
806 g_free(dh);
807 g_free(kh);
808 g_free(s->note_buf);
809 }
810
811 /* write common header, sub header and elf note to vmcore */
812 static void create_header64(DumpState *s, Error **errp)
813 {
814 DiskDumpHeader64 *dh = NULL;
815 KdumpSubHeader64 *kh = NULL;
816 size_t size;
817 uint32_t block_size;
818 uint32_t sub_hdr_size;
819 uint32_t bitmap_blocks;
820 uint32_t status = 0;
821 uint64_t offset_note;
822 Error *local_err = NULL;
823
824 /* write common header, the version of kdump-compressed format is 6th */
825 size = sizeof(DiskDumpHeader64);
826 dh = g_malloc0(size);
827
828 strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
829 dh->header_version = cpu_to_dump32(s, 6);
830 block_size = s->dump_info.page_size;
831 dh->block_size = cpu_to_dump32(s, block_size);
832 sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
833 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
834 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
835 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
836 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
837 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
838 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
839 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
840 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
841
842 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
843 status |= DUMP_DH_COMPRESSED_ZLIB;
844 }
845 #ifdef CONFIG_LZO
846 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
847 status |= DUMP_DH_COMPRESSED_LZO;
848 }
849 #endif
850 #ifdef CONFIG_SNAPPY
851 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
852 status |= DUMP_DH_COMPRESSED_SNAPPY;
853 }
854 #endif
855 dh->status = cpu_to_dump32(s, status);
856
857 if (write_buffer(s->fd, 0, dh, size) < 0) {
858 error_setg(errp, "dump: failed to write disk dump header");
859 goto out;
860 }
861
862 /* write sub header */
863 size = sizeof(KdumpSubHeader64);
864 kh = g_malloc0(size);
865
866 /* 64bit max_mapnr_64 */
867 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
868 kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
869 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
870
871 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
872 kh->offset_note = cpu_to_dump64(s, offset_note);
873 kh->note_size = cpu_to_dump64(s, s->note_size);
874
875 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
876 block_size, kh, size) < 0) {
877 error_setg(errp, "dump: failed to write kdump sub header");
878 goto out;
879 }
880
881 /* write note */
882 s->note_buf = g_malloc0(s->note_size);
883 s->note_buf_offset = 0;
884
885 /* use s->note_buf to store notes temporarily */
886 write_elf64_notes(buf_write_note, s, &local_err);
887 if (local_err) {
888 error_propagate(errp, local_err);
889 goto out;
890 }
891
892 if (write_buffer(s->fd, offset_note, s->note_buf,
893 s->note_size) < 0) {
894 error_setg(errp, "dump: failed to write notes");
895 goto out;
896 }
897
898 /* get offset of dump_bitmap */
899 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
900 block_size;
901
902 /* get offset of page */
903 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
904 block_size;
905
906 out:
907 g_free(dh);
908 g_free(kh);
909 g_free(s->note_buf);
910 }
911
912 static void write_dump_header(DumpState *s, Error **errp)
913 {
914 Error *local_err = NULL;
915
916 if (s->dump_info.d_class == ELFCLASS32) {
917 create_header32(s, &local_err);
918 } else {
919 create_header64(s, &local_err);
920 }
921 if (local_err) {
922 error_propagate(errp, local_err);
923 }
924 }
925
926 static size_t dump_bitmap_get_bufsize(DumpState *s)
927 {
928 return s->dump_info.page_size;
929 }
930
931 /*
932 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
933 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
934 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
935 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
936 * vmcore, ie. synchronizing un-sync bit into vmcore.
937 */
938 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
939 uint8_t *buf, DumpState *s)
940 {
941 off_t old_offset, new_offset;
942 off_t offset_bitmap1, offset_bitmap2;
943 uint32_t byte, bit;
944 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
945 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
946
947 /* should not set the previous place */
948 assert(last_pfn <= pfn);
949
950 /*
951 * if the bit needed to be set is not cached in buf, flush the data in buf
952 * to vmcore firstly.
953 * making new_offset be bigger than old_offset can also sync remained data
954 * into vmcore.
955 */
956 old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
957 new_offset = bitmap_bufsize * (pfn / bits_per_buf);
958
959 while (old_offset < new_offset) {
960 /* calculate the offset and write dump_bitmap */
961 offset_bitmap1 = s->offset_dump_bitmap + old_offset;
962 if (write_buffer(s->fd, offset_bitmap1, buf,
963 bitmap_bufsize) < 0) {
964 return -1;
965 }
966
967 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
968 offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
969 old_offset;
970 if (write_buffer(s->fd, offset_bitmap2, buf,
971 bitmap_bufsize) < 0) {
972 return -1;
973 }
974
975 memset(buf, 0, bitmap_bufsize);
976 old_offset += bitmap_bufsize;
977 }
978
979 /* get the exact place of the bit in the buf, and set it */
980 byte = (pfn % bits_per_buf) / CHAR_BIT;
981 bit = (pfn % bits_per_buf) % CHAR_BIT;
982 if (value) {
983 buf[byte] |= 1u << bit;
984 } else {
985 buf[byte] &= ~(1u << bit);
986 }
987
988 return 0;
989 }
990
991 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
992 {
993 int target_page_shift = ctz32(s->dump_info.page_size);
994
995 return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
996 }
997
998 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
999 {
1000 int target_page_shift = ctz32(s->dump_info.page_size);
1001
1002 return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1003 }
1004
1005 /*
1006 * exam every page and return the page frame number and the address of the page.
1007 * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1008 * blocks, so block->target_start and block->target_end should be interal
1009 * multiples of the target page size.
1010 */
1011 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1012 uint8_t **bufptr, DumpState *s)
1013 {
1014 GuestPhysBlock *block = *blockptr;
1015 hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1016 uint8_t *buf;
1017
1018 /* block == NULL means the start of the iteration */
1019 if (!block) {
1020 block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1021 *blockptr = block;
1022 assert((block->target_start & ~target_page_mask) == 0);
1023 assert((block->target_end & ~target_page_mask) == 0);
1024 *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1025 if (bufptr) {
1026 *bufptr = block->host_addr;
1027 }
1028 return true;
1029 }
1030
1031 *pfnptr = *pfnptr + 1;
1032 addr = dump_pfn_to_paddr(s, *pfnptr);
1033
1034 if ((addr >= block->target_start) &&
1035 (addr + s->dump_info.page_size <= block->target_end)) {
1036 buf = block->host_addr + (addr - block->target_start);
1037 } else {
1038 /* the next page is in the next block */
1039 block = QTAILQ_NEXT(block, next);
1040 *blockptr = block;
1041 if (!block) {
1042 return false;
1043 }
1044 assert((block->target_start & ~target_page_mask) == 0);
1045 assert((block->target_end & ~target_page_mask) == 0);
1046 *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1047 buf = block->host_addr;
1048 }
1049
1050 if (bufptr) {
1051 *bufptr = buf;
1052 }
1053
1054 return true;
1055 }
1056
1057 static void write_dump_bitmap(DumpState *s, Error **errp)
1058 {
1059 int ret = 0;
1060 uint64_t last_pfn, pfn;
1061 void *dump_bitmap_buf;
1062 size_t num_dumpable;
1063 GuestPhysBlock *block_iter = NULL;
1064 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1065 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1066
1067 /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1068 dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1069
1070 num_dumpable = 0;
1071 last_pfn = 0;
1072
1073 /*
1074 * exam memory page by page, and set the bit in dump_bitmap corresponded
1075 * to the existing page.
1076 */
1077 while (get_next_page(&block_iter, &pfn, NULL, s)) {
1078 ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1079 if (ret < 0) {
1080 error_setg(errp, "dump: failed to set dump_bitmap");
1081 goto out;
1082 }
1083
1084 last_pfn = pfn;
1085 num_dumpable++;
1086 }
1087
1088 /*
1089 * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1090 * set the remaining bits from last_pfn to the end of the bitmap buffer to
1091 * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1092 */
1093 if (num_dumpable > 0) {
1094 ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1095 dump_bitmap_buf, s);
1096 if (ret < 0) {
1097 error_setg(errp, "dump: failed to sync dump_bitmap");
1098 goto out;
1099 }
1100 }
1101
1102 /* number of dumpable pages that will be dumped later */
1103 s->num_dumpable = num_dumpable;
1104
1105 out:
1106 g_free(dump_bitmap_buf);
1107 }
1108
1109 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1110 off_t offset)
1111 {
1112 data_cache->fd = s->fd;
1113 data_cache->data_size = 0;
1114 data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1115 data_cache->buf = g_malloc0(data_cache->buf_size);
1116 data_cache->offset = offset;
1117 }
1118
1119 static int write_cache(DataCache *dc, const void *buf, size_t size,
1120 bool flag_sync)
1121 {
1122 /*
1123 * dc->buf_size should not be less than size, otherwise dc will never be
1124 * enough
1125 */
1126 assert(size <= dc->buf_size);
1127
1128 /*
1129 * if flag_sync is set, synchronize data in dc->buf into vmcore.
1130 * otherwise check if the space is enough for caching data in buf, if not,
1131 * write the data in dc->buf to dc->fd and reset dc->buf
1132 */
1133 if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1134 (flag_sync && dc->data_size > 0)) {
1135 if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1136 return -1;
1137 }
1138
1139 dc->offset += dc->data_size;
1140 dc->data_size = 0;
1141 }
1142
1143 if (!flag_sync) {
1144 memcpy(dc->buf + dc->data_size, buf, size);
1145 dc->data_size += size;
1146 }
1147
1148 return 0;
1149 }
1150
1151 static void free_data_cache(DataCache *data_cache)
1152 {
1153 g_free(data_cache->buf);
1154 }
1155
1156 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1157 {
1158 switch (flag_compress) {
1159 case DUMP_DH_COMPRESSED_ZLIB:
1160 return compressBound(page_size);
1161
1162 case DUMP_DH_COMPRESSED_LZO:
1163 /*
1164 * LZO will expand incompressible data by a little amount. Please check
1165 * the following URL to see the expansion calculation:
1166 * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1167 */
1168 return page_size + page_size / 16 + 64 + 3;
1169
1170 #ifdef CONFIG_SNAPPY
1171 case DUMP_DH_COMPRESSED_SNAPPY:
1172 return snappy_max_compressed_length(page_size);
1173 #endif
1174 }
1175 return 0;
1176 }
1177
1178 /*
1179 * check if the page is all 0
1180 */
1181 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1182 {
1183 return buffer_is_zero(buf, page_size);
1184 }
1185
1186 static void write_dump_pages(DumpState *s, Error **errp)
1187 {
1188 int ret = 0;
1189 DataCache page_desc, page_data;
1190 size_t len_buf_out, size_out;
1191 #ifdef CONFIG_LZO
1192 lzo_bytep wrkmem = NULL;
1193 #endif
1194 uint8_t *buf_out = NULL;
1195 off_t offset_desc, offset_data;
1196 PageDescriptor pd, pd_zero;
1197 uint8_t *buf;
1198 GuestPhysBlock *block_iter = NULL;
1199 uint64_t pfn_iter;
1200
1201 /* get offset of page_desc and page_data in dump file */
1202 offset_desc = s->offset_page;
1203 offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1204
1205 prepare_data_cache(&page_desc, s, offset_desc);
1206 prepare_data_cache(&page_data, s, offset_data);
1207
1208 /* prepare buffer to store compressed data */
1209 len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1210 assert(len_buf_out != 0);
1211
1212 #ifdef CONFIG_LZO
1213 wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1214 #endif
1215
1216 buf_out = g_malloc(len_buf_out);
1217
1218 /*
1219 * init zero page's page_desc and page_data, because every zero page
1220 * uses the same page_data
1221 */
1222 pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1223 pd_zero.flags = cpu_to_dump32(s, 0);
1224 pd_zero.offset = cpu_to_dump64(s, offset_data);
1225 pd_zero.page_flags = cpu_to_dump64(s, 0);
1226 buf = g_malloc0(s->dump_info.page_size);
1227 ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1228 g_free(buf);
1229 if (ret < 0) {
1230 error_setg(errp, "dump: failed to write page data (zero page)");
1231 goto out;
1232 }
1233
1234 offset_data += s->dump_info.page_size;
1235
1236 /*
1237 * dump memory to vmcore page by page. zero page will all be resided in the
1238 * first page of page section
1239 */
1240 while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1241 /* check zero page */
1242 if (is_zero_page(buf, s->dump_info.page_size)) {
1243 ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1244 false);
1245 if (ret < 0) {
1246 error_setg(errp, "dump: failed to write page desc");
1247 goto out;
1248 }
1249 } else {
1250 /*
1251 * not zero page, then:
1252 * 1. compress the page
1253 * 2. write the compressed page into the cache of page_data
1254 * 3. get page desc of the compressed page and write it into the
1255 * cache of page_desc
1256 *
1257 * only one compression format will be used here, for
1258 * s->flag_compress is set. But when compression fails to work,
1259 * we fall back to save in plaintext.
1260 */
1261 size_out = len_buf_out;
1262 if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1263 (compress2(buf_out, (uLongf *)&size_out, buf,
1264 s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1265 (size_out < s->dump_info.page_size)) {
1266 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
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 error_setg(errp, "dump: failed to write page data");
1272 goto out;
1273 }
1274 #ifdef CONFIG_LZO
1275 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1276 (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1277 (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1278 (size_out < s->dump_info.page_size)) {
1279 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1280 pd.size = cpu_to_dump32(s, size_out);
1281
1282 ret = write_cache(&page_data, buf_out, size_out, false);
1283 if (ret < 0) {
1284 error_setg(errp, "dump: failed to write page data");
1285 goto out;
1286 }
1287 #endif
1288 #ifdef CONFIG_SNAPPY
1289 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1290 (snappy_compress((char *)buf, s->dump_info.page_size,
1291 (char *)buf_out, &size_out) == SNAPPY_OK) &&
1292 (size_out < s->dump_info.page_size)) {
1293 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1294 pd.size = cpu_to_dump32(s, size_out);
1295
1296 ret = write_cache(&page_data, buf_out, size_out, false);
1297 if (ret < 0) {
1298 error_setg(errp, "dump: failed to write page data");
1299 goto out;
1300 }
1301 #endif
1302 } else {
1303 /*
1304 * fall back to save in plaintext, size_out should be
1305 * assigned the target's page size
1306 */
1307 pd.flags = cpu_to_dump32(s, 0);
1308 size_out = s->dump_info.page_size;
1309 pd.size = cpu_to_dump32(s, size_out);
1310
1311 ret = write_cache(&page_data, buf,
1312 s->dump_info.page_size, false);
1313 if (ret < 0) {
1314 error_setg(errp, "dump: failed to write page data");
1315 goto out;
1316 }
1317 }
1318
1319 /* get and write page desc here */
1320 pd.page_flags = cpu_to_dump64(s, 0);
1321 pd.offset = cpu_to_dump64(s, offset_data);
1322 offset_data += size_out;
1323
1324 ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1325 if (ret < 0) {
1326 error_setg(errp, "dump: failed to write page desc");
1327 goto out;
1328 }
1329 }
1330 s->written_size += s->dump_info.page_size;
1331 }
1332
1333 ret = write_cache(&page_desc, NULL, 0, true);
1334 if (ret < 0) {
1335 error_setg(errp, "dump: failed to sync cache for page_desc");
1336 goto out;
1337 }
1338 ret = write_cache(&page_data, NULL, 0, true);
1339 if (ret < 0) {
1340 error_setg(errp, "dump: failed to sync cache for page_data");
1341 goto out;
1342 }
1343
1344 out:
1345 free_data_cache(&page_desc);
1346 free_data_cache(&page_data);
1347
1348 #ifdef CONFIG_LZO
1349 g_free(wrkmem);
1350 #endif
1351
1352 g_free(buf_out);
1353 }
1354
1355 static void create_kdump_vmcore(DumpState *s, Error **errp)
1356 {
1357 int ret;
1358 Error *local_err = NULL;
1359
1360 /*
1361 * the kdump-compressed format is:
1362 * File offset
1363 * +------------------------------------------+ 0x0
1364 * | main header (struct disk_dump_header) |
1365 * |------------------------------------------+ block 1
1366 * | sub header (struct kdump_sub_header) |
1367 * |------------------------------------------+ block 2
1368 * | 1st-dump_bitmap |
1369 * |------------------------------------------+ block 2 + X blocks
1370 * | 2nd-dump_bitmap | (aligned by block)
1371 * |------------------------------------------+ block 2 + 2 * X blocks
1372 * | page desc for pfn 0 (struct page_desc) | (aligned by block)
1373 * | page desc for pfn 1 (struct page_desc) |
1374 * | : |
1375 * |------------------------------------------| (not aligned by block)
1376 * | page data (pfn 0) |
1377 * | page data (pfn 1) |
1378 * | : |
1379 * +------------------------------------------+
1380 */
1381
1382 ret = write_start_flat_header(s->fd);
1383 if (ret < 0) {
1384 error_setg(errp, "dump: failed to write start flat header");
1385 return;
1386 }
1387
1388 write_dump_header(s, &local_err);
1389 if (local_err) {
1390 error_propagate(errp, local_err);
1391 return;
1392 }
1393
1394 write_dump_bitmap(s, &local_err);
1395 if (local_err) {
1396 error_propagate(errp, local_err);
1397 return;
1398 }
1399
1400 write_dump_pages(s, &local_err);
1401 if (local_err) {
1402 error_propagate(errp, local_err);
1403 return;
1404 }
1405
1406 ret = write_end_flat_header(s->fd);
1407 if (ret < 0) {
1408 error_setg(errp, "dump: failed to write end flat header");
1409 return;
1410 }
1411 }
1412
1413 static ram_addr_t get_start_block(DumpState *s)
1414 {
1415 GuestPhysBlock *block;
1416
1417 if (!s->has_filter) {
1418 s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1419 return 0;
1420 }
1421
1422 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1423 if (block->target_start >= s->begin + s->length ||
1424 block->target_end <= s->begin) {
1425 /* This block is out of the range */
1426 continue;
1427 }
1428
1429 s->next_block = block;
1430 if (s->begin > block->target_start) {
1431 s->start = s->begin - block->target_start;
1432 } else {
1433 s->start = 0;
1434 }
1435 return s->start;
1436 }
1437
1438 return -1;
1439 }
1440
1441 static void get_max_mapnr(DumpState *s)
1442 {
1443 GuestPhysBlock *last_block;
1444
1445 last_block = QTAILQ_LAST(&s->guest_phys_blocks.head, GuestPhysBlockHead);
1446 s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1447 }
1448
1449 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1450
1451 static void dump_state_prepare(DumpState *s)
1452 {
1453 /* zero the struct, setting status to active */
1454 *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1455 }
1456
1457 bool dump_in_progress(void)
1458 {
1459 DumpState *state = &dump_state_global;
1460 return (atomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1461 }
1462
1463 /* calculate total size of memory to be dumped (taking filter into
1464 * acoount.) */
1465 static int64_t dump_calculate_size(DumpState *s)
1466 {
1467 GuestPhysBlock *block;
1468 int64_t size = 0, total = 0, left = 0, right = 0;
1469
1470 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1471 if (s->has_filter) {
1472 /* calculate the overlapped region. */
1473 left = MAX(s->begin, block->target_start);
1474 right = MIN(s->begin + s->length, block->target_end);
1475 size = right - left;
1476 size = size > 0 ? size : 0;
1477 } else {
1478 /* count the whole region in */
1479 size = (block->target_end - block->target_start);
1480 }
1481 total += size;
1482 }
1483
1484 return total;
1485 }
1486
1487 static void dump_init(DumpState *s, int fd, bool has_format,
1488 DumpGuestMemoryFormat format, bool paging, bool has_filter,
1489 int64_t begin, int64_t length, Error **errp)
1490 {
1491 CPUState *cpu;
1492 int nr_cpus;
1493 Error *err = NULL;
1494 int ret;
1495
1496 s->has_format = has_format;
1497 s->format = format;
1498 s->written_size = 0;
1499
1500 /* kdump-compressed is conflict with paging and filter */
1501 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1502 assert(!paging && !has_filter);
1503 }
1504
1505 if (runstate_is_running()) {
1506 vm_stop(RUN_STATE_SAVE_VM);
1507 s->resume = true;
1508 } else {
1509 s->resume = false;
1510 }
1511
1512 /* If we use KVM, we should synchronize the registers before we get dump
1513 * info or physmap info.
1514 */
1515 cpu_synchronize_all_states();
1516 nr_cpus = 0;
1517 CPU_FOREACH(cpu) {
1518 nr_cpus++;
1519 }
1520
1521 s->fd = fd;
1522 s->has_filter = has_filter;
1523 s->begin = begin;
1524 s->length = length;
1525
1526 memory_mapping_list_init(&s->list);
1527
1528 guest_phys_blocks_init(&s->guest_phys_blocks);
1529 guest_phys_blocks_append(&s->guest_phys_blocks);
1530 s->total_size = dump_calculate_size(s);
1531 #ifdef DEBUG_DUMP_GUEST_MEMORY
1532 fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1533 #endif
1534
1535 s->start = get_start_block(s);
1536 if (s->start == -1) {
1537 error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1538 goto cleanup;
1539 }
1540
1541 /* get dump info: endian, class and architecture.
1542 * If the target architecture is not supported, cpu_get_dump_info() will
1543 * return -1.
1544 */
1545 ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1546 if (ret < 0) {
1547 error_setg(errp, QERR_UNSUPPORTED);
1548 goto cleanup;
1549 }
1550
1551 if (!s->dump_info.page_size) {
1552 s->dump_info.page_size = TARGET_PAGE_SIZE;
1553 }
1554
1555 s->note_size = cpu_get_note_size(s->dump_info.d_class,
1556 s->dump_info.d_machine, nr_cpus);
1557 if (s->note_size < 0) {
1558 error_setg(errp, QERR_UNSUPPORTED);
1559 goto cleanup;
1560 }
1561
1562 /* get memory mapping */
1563 if (paging) {
1564 qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1565 if (err != NULL) {
1566 error_propagate(errp, err);
1567 goto cleanup;
1568 }
1569 } else {
1570 qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1571 }
1572
1573 s->nr_cpus = nr_cpus;
1574
1575 get_max_mapnr(s);
1576
1577 uint64_t tmp;
1578 tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1579 s->dump_info.page_size);
1580 s->len_dump_bitmap = tmp * s->dump_info.page_size;
1581
1582 /* init for kdump-compressed format */
1583 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1584 switch (format) {
1585 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1586 s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1587 break;
1588
1589 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1590 #ifdef CONFIG_LZO
1591 if (lzo_init() != LZO_E_OK) {
1592 error_setg(errp, "failed to initialize the LZO library");
1593 goto cleanup;
1594 }
1595 #endif
1596 s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1597 break;
1598
1599 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1600 s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1601 break;
1602
1603 default:
1604 s->flag_compress = 0;
1605 }
1606
1607 return;
1608 }
1609
1610 if (s->has_filter) {
1611 memory_mapping_filter(&s->list, s->begin, s->length);
1612 }
1613
1614 /*
1615 * calculate phdr_num
1616 *
1617 * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1618 */
1619 s->phdr_num = 1; /* PT_NOTE */
1620 if (s->list.num < UINT16_MAX - 2) {
1621 s->phdr_num += s->list.num;
1622 s->have_section = false;
1623 } else {
1624 s->have_section = true;
1625 s->phdr_num = PN_XNUM;
1626 s->sh_info = 1; /* PT_NOTE */
1627
1628 /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1629 if (s->list.num <= UINT32_MAX - 1) {
1630 s->sh_info += s->list.num;
1631 } else {
1632 s->sh_info = UINT32_MAX;
1633 }
1634 }
1635
1636 if (s->dump_info.d_class == ELFCLASS64) {
1637 if (s->have_section) {
1638 s->memory_offset = sizeof(Elf64_Ehdr) +
1639 sizeof(Elf64_Phdr) * s->sh_info +
1640 sizeof(Elf64_Shdr) + s->note_size;
1641 } else {
1642 s->memory_offset = sizeof(Elf64_Ehdr) +
1643 sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1644 }
1645 } else {
1646 if (s->have_section) {
1647 s->memory_offset = sizeof(Elf32_Ehdr) +
1648 sizeof(Elf32_Phdr) * s->sh_info +
1649 sizeof(Elf32_Shdr) + s->note_size;
1650 } else {
1651 s->memory_offset = sizeof(Elf32_Ehdr) +
1652 sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1653 }
1654 }
1655
1656 return;
1657
1658 cleanup:
1659 dump_cleanup(s);
1660 }
1661
1662 /* this operation might be time consuming. */
1663 static void dump_process(DumpState *s, Error **errp)
1664 {
1665 Error *local_err = NULL;
1666 DumpQueryResult *result = NULL;
1667
1668 if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1669 create_kdump_vmcore(s, &local_err);
1670 } else {
1671 create_vmcore(s, &local_err);
1672 }
1673
1674 /* make sure status is written after written_size updates */
1675 smp_wmb();
1676 atomic_set(&s->status,
1677 (local_err ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1678
1679 /* send DUMP_COMPLETED message (unconditionally) */
1680 result = qmp_query_dump(NULL);
1681 /* should never fail */
1682 assert(result);
1683 qapi_event_send_dump_completed(result, !!local_err, (local_err ? \
1684 error_get_pretty(local_err) : NULL),
1685 &error_abort);
1686 qapi_free_DumpQueryResult(result);
1687
1688 error_propagate(errp, local_err);
1689 dump_cleanup(s);
1690 }
1691
1692 static void *dump_thread(void *data)
1693 {
1694 Error *err = NULL;
1695 DumpState *s = (DumpState *)data;
1696 dump_process(s, &err);
1697 error_free(err);
1698 return NULL;
1699 }
1700
1701 DumpQueryResult *qmp_query_dump(Error **errp)
1702 {
1703 DumpQueryResult *result = g_new(DumpQueryResult, 1);
1704 DumpState *state = &dump_state_global;
1705 result->status = atomic_read(&state->status);
1706 /* make sure we are reading status and written_size in order */
1707 smp_rmb();
1708 result->completed = state->written_size;
1709 result->total = state->total_size;
1710 return result;
1711 }
1712
1713 void qmp_dump_guest_memory(bool paging, const char *file,
1714 bool has_detach, bool detach,
1715 bool has_begin, int64_t begin, bool has_length,
1716 int64_t length, bool has_format,
1717 DumpGuestMemoryFormat format, Error **errp)
1718 {
1719 const char *p;
1720 int fd = -1;
1721 DumpState *s;
1722 Error *local_err = NULL;
1723 bool detach_p = false;
1724
1725 if (runstate_check(RUN_STATE_INMIGRATE)) {
1726 error_setg(errp, "Dump not allowed during incoming migration.");
1727 return;
1728 }
1729
1730 /* if there is a dump in background, we should wait until the dump
1731 * finished */
1732 if (dump_in_progress()) {
1733 error_setg(errp, "There is a dump in process, please wait.");
1734 return;
1735 }
1736
1737 /*
1738 * kdump-compressed format need the whole memory dumped, so paging or
1739 * filter is not supported here.
1740 */
1741 if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1742 (paging || has_begin || has_length)) {
1743 error_setg(errp, "kdump-compressed format doesn't support paging or "
1744 "filter");
1745 return;
1746 }
1747 if (has_begin && !has_length) {
1748 error_setg(errp, QERR_MISSING_PARAMETER, "length");
1749 return;
1750 }
1751 if (!has_begin && has_length) {
1752 error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1753 return;
1754 }
1755 if (has_detach) {
1756 detach_p = detach;
1757 }
1758
1759 /* check whether lzo/snappy is supported */
1760 #ifndef CONFIG_LZO
1761 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1762 error_setg(errp, "kdump-lzo is not available now");
1763 return;
1764 }
1765 #endif
1766
1767 #ifndef CONFIG_SNAPPY
1768 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1769 error_setg(errp, "kdump-snappy is not available now");
1770 return;
1771 }
1772 #endif
1773
1774 #if !defined(WIN32)
1775 if (strstart(file, "fd:", &p)) {
1776 fd = monitor_get_fd(cur_mon, p, errp);
1777 if (fd == -1) {
1778 return;
1779 }
1780 }
1781 #endif
1782
1783 if (strstart(file, "file:", &p)) {
1784 fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1785 if (fd < 0) {
1786 error_setg_file_open(errp, errno, p);
1787 return;
1788 }
1789 }
1790
1791 if (fd == -1) {
1792 error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
1793 return;
1794 }
1795
1796 s = &dump_state_global;
1797 dump_state_prepare(s);
1798
1799 dump_init(s, fd, has_format, format, paging, has_begin,
1800 begin, length, &local_err);
1801 if (local_err) {
1802 error_propagate(errp, local_err);
1803 atomic_set(&s->status, DUMP_STATUS_FAILED);
1804 return;
1805 }
1806
1807 if (detach_p) {
1808 /* detached dump */
1809 qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
1810 s, QEMU_THREAD_DETACHED);
1811 } else {
1812 /* sync dump */
1813 dump_process(s, errp);
1814 }
1815 }
1816
1817 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
1818 {
1819 DumpGuestMemoryFormatList *item;
1820 DumpGuestMemoryCapability *cap =
1821 g_malloc0(sizeof(DumpGuestMemoryCapability));
1822
1823 /* elf is always available */
1824 item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1825 cap->formats = item;
1826 item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
1827
1828 /* kdump-zlib is always available */
1829 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1830 item = item->next;
1831 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
1832
1833 /* add new item if kdump-lzo is available */
1834 #ifdef CONFIG_LZO
1835 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1836 item = item->next;
1837 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
1838 #endif
1839
1840 /* add new item if kdump-snappy is available */
1841 #ifdef CONFIG_SNAPPY
1842 item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1843 item = item->next;
1844 item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
1845 #endif
1846
1847 return cap;
1848 }