master
c 2,249 lines 66.3 KB
Raw
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 "qemu/bswap.h"
18 #include "exec/cpu-common.h"
19 #include "exec/target_page.h"
20 #include "monitor/monitor.h"
21 #include "system/dump.h"
22 #include "system/runstate.h"
23 #include "system/cpus.h"
24 #include "system/physmem.h"
25 #include "qapi/error.h"
26 #include "qapi/qapi-commands-dump.h"
27 #include "qapi/qapi-events-dump.h"
28 #include "qapi/qmp/qerror.h"
29 #include "qemu/error-report.h"
30 #include "qemu/main-loop.h"
31 #include "hw/misc/vmcoreinfo.h"
32 #include "migration/blocker.h"
33 #include "migration/misc.h"
34 #include "hw/core/cpu.h"
35 #include "win_dump.h"
36 #include "qemu/range.h"
37
38 #include <zlib.h>
39 #ifdef CONFIG_LZO
40 #include <lzo/lzo1x.h>
41 #endif
42 #ifdef CONFIG_SNAPPY
43 #include <snappy-c.h>
44 #endif
45 #ifndef ELF_MACHINE_UNAME
46 #define ELF_MACHINE_UNAME "Unknown"
47 #endif
48
49 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
50
51 static Error *dump_migration_blocker;
52
53 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size) \
54 ((DIV_ROUND_UP((hdr_size), 4) + \
55 DIV_ROUND_UP((name_size), 4) + \
56 DIV_ROUND_UP((desc_size), 4)) * 4)
57
58 static inline bool dump_is_64bit(DumpState *s)
59 {
60 return s->dump_info.d_class == ELFCLASS64;
61 }
62
63 static inline bool dump_has_filter(DumpState *s)
64 {
65 return s->filter_area_length > 0;
66 }
67
68 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
69 {
70 if (s->dump_info.d_endian == ELFDATA2LSB) {
71 val = cpu_to_le16(val);
72 } else {
73 val = cpu_to_be16(val);
74 }
75
76 return val;
77 }
78
79 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
80 {
81 if (s->dump_info.d_endian == ELFDATA2LSB) {
82 val = cpu_to_le32(val);
83 } else {
84 val = cpu_to_be32(val);
85 }
86
87 return val;
88 }
89
90 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
91 {
92 if (s->dump_info.d_endian == ELFDATA2LSB) {
93 val = cpu_to_le64(val);
94 } else {
95 val = cpu_to_be64(val);
96 }
97
98 return val;
99 }
100
101 static int dump_cleanup(DumpState *s)
102 {
103 if (s->dump_info.arch_cleanup_fn) {
104 s->dump_info.arch_cleanup_fn(s);
105 }
106
107 guest_phys_blocks_free(&s->guest_phys_blocks);
108 memory_mapping_list_free(&s->list);
109 if (s->fd != -1) {
110 close(s->fd);
111 }
112 s->fd = -1;
113 g_free(s->guest_note);
114 g_clear_pointer(&s->string_table_buf, g_array_unref);
115 s->guest_note = NULL;
116 if (s->resume) {
117 if (s->detached) {
118 bql_lock();
119 }
120 vm_start();
121 if (s->detached) {
122 bql_unlock();
123 }
124 }
125 migrate_del_blocker(&dump_migration_blocker);
126
127 return 0;
128 }
129
130 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
131 {
132 DumpState *s = opaque;
133 size_t written_size;
134
135 written_size = qemu_write_full(s->fd, buf, size);
136 if (written_size != size) {
137 return -errno;
138 }
139
140 return 0;
141 }
142
143 static void prepare_elf64_header(DumpState *s, Elf64_Ehdr *elf_header)
144 {
145 /*
146 * phnum in the elf header is 16 bit, if we have more segments we
147 * set phnum to PN_XNUM and write the real number of segments to a
148 * special section.
149 */
150 uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
151
152 memset(elf_header, 0, sizeof(Elf64_Ehdr));
153 memcpy(elf_header, ELFMAG, SELFMAG);
154 elf_header->e_ident[EI_CLASS] = ELFCLASS64;
155 elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
156 elf_header->e_ident[EI_VERSION] = EV_CURRENT;
157 elf_header->e_type = cpu_to_dump16(s, ET_CORE);
158 elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
159 elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
160 elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
161 elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset);
162 elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
163 elf_header->e_phnum = cpu_to_dump16(s, phnum);
164 elf_header->e_shoff = cpu_to_dump64(s, s->shdr_offset);
165 elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
166 elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
167 elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
168 }
169
170 static void prepare_elf32_header(DumpState *s, Elf32_Ehdr *elf_header)
171 {
172 /*
173 * phnum in the elf header is 16 bit, if we have more segments we
174 * set phnum to PN_XNUM and write the real number of segments to a
175 * special section.
176 */
177 uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
178
179 memset(elf_header, 0, sizeof(Elf32_Ehdr));
180 memcpy(elf_header, ELFMAG, SELFMAG);
181 elf_header->e_ident[EI_CLASS] = ELFCLASS32;
182 elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
183 elf_header->e_ident[EI_VERSION] = EV_CURRENT;
184 elf_header->e_type = cpu_to_dump16(s, ET_CORE);
185 elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
186 elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
187 elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
188 elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset);
189 elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
190 elf_header->e_phnum = cpu_to_dump16(s, phnum);
191 elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset);
192 elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
193 elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
194 elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
195 }
196
197 static void write_elf_header(DumpState *s, Error **errp)
198 {
199 Elf32_Ehdr elf32_header;
200 Elf64_Ehdr elf64_header;
201 size_t header_size;
202 void *header_ptr;
203 int ret;
204
205 /* The NULL header and the shstrtab are always defined */
206 assert(s->shdr_num >= 2);
207 if (dump_is_64bit(s)) {
208 prepare_elf64_header(s, &elf64_header);
209 header_size = sizeof(elf64_header);
210 header_ptr = &elf64_header;
211 } else {
212 prepare_elf32_header(s, &elf32_header);
213 header_size = sizeof(elf32_header);
214 header_ptr = &elf32_header;
215 }
216
217 ret = fd_write_vmcore(header_ptr, header_size, s);
218 if (ret < 0) {
219 error_setg_errno(errp, -ret, "dump: failed to write elf header");
220 }
221 }
222
223 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
224 int phdr_index, hwaddr offset,
225 hwaddr filesz, Error **errp)
226 {
227 Elf64_Phdr phdr;
228 int ret;
229
230 memset(&phdr, 0, sizeof(Elf64_Phdr));
231 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
232 phdr.p_offset = cpu_to_dump64(s, offset);
233 phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
234 phdr.p_filesz = cpu_to_dump64(s, filesz);
235 phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
236 phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
237
238 assert(memory_mapping->length >= filesz);
239
240 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
241 if (ret < 0) {
242 error_setg_errno(errp, -ret,
243 "dump: failed to write program header table");
244 }
245 }
246
247 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
248 int phdr_index, hwaddr offset,
249 hwaddr filesz, Error **errp)
250 {
251 Elf32_Phdr phdr;
252 int ret;
253
254 memset(&phdr, 0, sizeof(Elf32_Phdr));
255 phdr.p_type = cpu_to_dump32(s, PT_LOAD);
256 phdr.p_offset = cpu_to_dump32(s, offset);
257 phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
258 phdr.p_filesz = cpu_to_dump32(s, filesz);
259 phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
260 phdr.p_vaddr =
261 cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
262
263 assert(memory_mapping->length >= filesz);
264
265 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
266 if (ret < 0) {
267 error_setg_errno(errp, -ret,
268 "dump: failed to write program header table");
269 }
270 }
271
272 static void prepare_elf64_phdr_note(DumpState *s, Elf64_Phdr *phdr)
273 {
274 memset(phdr, 0, sizeof(*phdr));
275 phdr->p_type = cpu_to_dump32(s, PT_NOTE);
276 phdr->p_offset = cpu_to_dump64(s, s->note_offset);
277 phdr->p_paddr = 0;
278 phdr->p_filesz = cpu_to_dump64(s, s->note_size);
279 phdr->p_memsz = cpu_to_dump64(s, s->note_size);
280 phdr->p_vaddr = 0;
281 }
282
283 static inline int cpu_index(CPUState *cpu)
284 {
285 return cpu->cpu_index + 1;
286 }
287
288 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
289 Error **errp)
290 {
291 int ret;
292
293 if (s->guest_note) {
294 ret = f(s->guest_note, s->guest_note_size, s);
295 if (ret < 0) {
296 error_setg(errp, "dump: failed to write guest note");
297 }
298 }
299 }
300
301 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
302 Error **errp)
303 {
304 CPUState *cpu;
305 int ret;
306 int id;
307
308 CPU_FOREACH(cpu) {
309 id = cpu_index(cpu);
310 ret = cpu_write_elf64_note(f, cpu, id, s);
311 if (ret < 0) {
312 error_setg(errp, "dump: failed to write elf notes");
313 return;
314 }
315 }
316
317 CPU_FOREACH(cpu) {
318 ret = cpu_write_elf64_qemunote(f, cpu, s);
319 if (ret < 0) {
320 error_setg(errp, "dump: failed to write CPU status");
321 return;
322 }
323 }
324
325 write_guest_note(f, s, errp);
326 }
327
328 static void prepare_elf32_phdr_note(DumpState *s, Elf32_Phdr *phdr)
329 {
330 memset(phdr, 0, sizeof(*phdr));
331 phdr->p_type = cpu_to_dump32(s, PT_NOTE);
332 phdr->p_offset = cpu_to_dump32(s, s->note_offset);
333 phdr->p_paddr = 0;
334 phdr->p_filesz = cpu_to_dump32(s, s->note_size);
335 phdr->p_memsz = cpu_to_dump32(s, s->note_size);
336 phdr->p_vaddr = 0;
337 }
338
339 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
340 Error **errp)
341 {
342 CPUState *cpu;
343 int ret;
344 int id;
345
346 CPU_FOREACH(cpu) {
347 id = cpu_index(cpu);
348 ret = cpu_write_elf32_note(f, cpu, id, s);
349 if (ret < 0) {
350 error_setg(errp, "dump: failed to write elf notes");
351 return;
352 }
353 }
354
355 CPU_FOREACH(cpu) {
356 ret = cpu_write_elf32_qemunote(f, cpu, s);
357 if (ret < 0) {
358 error_setg(errp, "dump: failed to write CPU status");
359 return;
360 }
361 }
362
363 write_guest_note(f, s, errp);
364 }
365
366 static void write_elf_phdr_note(DumpState *s, Error **errp)
367 {
368 Elf32_Phdr phdr32;
369 Elf64_Phdr phdr64;
370 void *phdr;
371 size_t size;
372 int ret;
373
374 if (dump_is_64bit(s)) {
375 prepare_elf64_phdr_note(s, &phdr64);
376 size = sizeof(phdr64);
377 phdr = &phdr64;
378 } else {
379 prepare_elf32_phdr_note(s, &phdr32);
380 size = sizeof(phdr32);
381 phdr = &phdr32;
382 }
383
384 ret = fd_write_vmcore(phdr, size, s);
385 if (ret < 0) {
386 error_setg_errno(errp, -ret,
387 "dump: failed to write program header table");
388 }
389 }
390
391 static void prepare_elf_section_hdr_zero(DumpState *s)
392 {
393 if (dump_is_64bit(s)) {
394 Elf64_Shdr *shdr64 = s->elf_section_hdrs;
395
396 shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
397 } else {
398 Elf32_Shdr *shdr32 = s->elf_section_hdrs;
399
400 shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
401 }
402 }
403
404 static void prepare_elf_section_hdr_string(DumpState *s, void *buff)
405 {
406 uint64_t index = s->string_table_buf->len;
407 const char strtab[] = ".shstrtab";
408 Elf32_Shdr shdr32 = {};
409 Elf64_Shdr shdr64 = {};
410 int shdr_size;
411 void *shdr;
412
413 g_array_append_vals(s->string_table_buf, strtab, sizeof(strtab));
414 if (dump_is_64bit(s)) {
415 shdr_size = sizeof(Elf64_Shdr);
416 shdr64.sh_type = SHT_STRTAB;
417 shdr64.sh_offset = s->section_offset + s->elf_section_data_size;
418 shdr64.sh_name = index;
419 shdr64.sh_size = s->string_table_buf->len;
420 shdr = &shdr64;
421 } else {
422 shdr_size = sizeof(Elf32_Shdr);
423 shdr32.sh_type = SHT_STRTAB;
424 shdr32.sh_offset = s->section_offset + s->elf_section_data_size;
425 shdr32.sh_name = index;
426 shdr32.sh_size = s->string_table_buf->len;
427 shdr = &shdr32;
428 }
429 memcpy(buff, shdr, shdr_size);
430 }
431
432 static bool prepare_elf_section_hdrs(DumpState *s, Error **errp)
433 {
434 size_t len, sizeof_shdr;
435 void *buff_hdr;
436
437 /*
438 * Section ordering:
439 * - HDR zero
440 * - Arch section hdrs
441 * - String table hdr
442 */
443 sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
444 len = sizeof_shdr * s->shdr_num;
445 s->elf_section_hdrs = g_malloc0(len);
446 buff_hdr = s->elf_section_hdrs;
447
448 /*
449 * The first section header is ALWAYS a special initial section
450 * header.
451 *
452 * The header should be 0 with one exception being that if
453 * phdr_num is PN_XNUM then the sh_info field contains the real
454 * number of segment entries.
455 *
456 * As we zero allocate the buffer we will only need to modify
457 * sh_info for the PN_XNUM case.
458 */
459 if (s->phdr_num >= PN_XNUM) {
460 prepare_elf_section_hdr_zero(s);
461 }
462 buff_hdr += sizeof_shdr;
463
464 /* Add architecture defined section headers */
465 if (s->dump_info.arch_sections_write_hdr_fn
466 && s->shdr_num > 2) {
467 buff_hdr += s->dump_info.arch_sections_write_hdr_fn(s, buff_hdr);
468
469 if (s->shdr_num >= SHN_LORESERVE) {
470 error_setg_errno(errp, EINVAL,
471 "dump: too many architecture defined sections");
472 return false;
473 }
474 }
475
476 /*
477 * String table is the last section since strings are added via
478 * arch_sections_write_hdr().
479 */
480 prepare_elf_section_hdr_string(s, buff_hdr);
481 return true;
482 }
483
484 static void write_elf_section_headers(DumpState *s, Error **errp)
485 {
486 size_t sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
487 int ret;
488
489 if (!prepare_elf_section_hdrs(s, errp)) {
490 return;
491 }
492
493 ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
494 if (ret < 0) {
495 error_setg_errno(errp, -ret, "dump: failed to write section headers");
496 }
497
498 g_free(s->elf_section_hdrs);
499 }
500
501 static void write_elf_sections(DumpState *s, Error **errp)
502 {
503 int ret;
504
505 if (s->elf_section_data_size) {
506 /* Write architecture section data */
507 ret = fd_write_vmcore(s->elf_section_data,
508 s->elf_section_data_size, s);
509 if (ret < 0) {
510 error_setg_errno(errp, -ret,
511 "dump: failed to write architecture section data");
512 return;
513 }
514 }
515
516 /* Write string table */
517 ret = fd_write_vmcore(s->string_table_buf->data,
518 s->string_table_buf->len, s);
519 if (ret < 0) {
520 error_setg_errno(errp, -ret, "dump: failed to write string table data");
521 }
522 }
523
524 static void write_data(DumpState *s, void *buf, int length, Error **errp)
525 {
526 int ret;
527
528 ret = fd_write_vmcore(buf, length, s);
529 if (ret < 0) {
530 error_setg_errno(errp, -ret, "dump: failed to save memory");
531 } else {
532 s->written_size += length;
533 }
534 }
535
536 /* write the memory to vmcore. 1 page per I/O. */
537 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
538 int64_t size, Error **errp)
539 {
540 ERRP_GUARD();
541 int64_t i;
542
543 for (i = 0; i < size / s->dump_info.page_size; i++) {
544 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
545 s->dump_info.page_size, errp);
546 if (*errp) {
547 return;
548 }
549 }
550
551 if ((size % s->dump_info.page_size) != 0) {
552 write_data(s, block->host_addr + start + i * s->dump_info.page_size,
553 size % s->dump_info.page_size, errp);
554 if (*errp) {
555 return;
556 }
557 }
558 }
559
560 /* get the memory's offset and size in the vmcore */
561 static void get_offset_range(hwaddr phys_addr,
562 ram_addr_t mapping_length,
563 DumpState *s,
564 hwaddr *p_offset,
565 hwaddr *p_filesz)
566 {
567 GuestPhysBlock *block;
568 hwaddr offset = s->memory_offset;
569 int64_t size_in_block, start;
570
571 /* When the memory is not stored into vmcore, offset will be -1 */
572 *p_offset = -1;
573 *p_filesz = 0;
574
575 if (dump_has_filter(s)) {
576 if (phys_addr < s->filter_area_begin ||
577 phys_addr >= s->filter_area_begin + s->filter_area_length) {
578 return;
579 }
580 }
581
582 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
583 if (dump_has_filter(s)) {
584 if (!ranges_overlap(block->target_start,
585 block->target_end - block->target_start,
586 s->filter_area_begin,
587 s->filter_area_length)) {
588 /* This block is out of the range */
589 continue;
590 }
591
592 if (s->filter_area_begin <= block->target_start) {
593 start = block->target_start;
594 } else {
595 start = s->filter_area_begin;
596 }
597
598 size_in_block = block->target_end - start;
599 if (s->filter_area_begin + s->filter_area_length < block->target_end) {
600 size_in_block -= block->target_end - (s->filter_area_begin + s->filter_area_length);
601 }
602 } else {
603 start = block->target_start;
604 size_in_block = block->target_end - block->target_start;
605 }
606
607 if (phys_addr >= start && phys_addr < start + size_in_block) {
608 *p_offset = phys_addr - start + offset;
609
610 /* The offset range mapped from the vmcore file must not spill over
611 * the GuestPhysBlock, clamp it. The rest of the mapping will be
612 * zero-filled in memory at load time; see
613 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
614 */
615 *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
616 mapping_length :
617 size_in_block - (phys_addr - start);
618 return;
619 }
620
621 offset += size_in_block;
622 }
623 }
624
625 static void write_elf_phdr_loads(DumpState *s, Error **errp)
626 {
627 ERRP_GUARD();
628 hwaddr offset, filesz;
629 MemoryMapping *memory_mapping;
630 uint32_t phdr_index = 1;
631
632 QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
633 get_offset_range(memory_mapping->phys_addr,
634 memory_mapping->length,
635 s, &offset, &filesz);
636 if (dump_is_64bit(s)) {
637 write_elf64_load(s, memory_mapping, phdr_index++, offset,
638 filesz, errp);
639 } else {
640 write_elf32_load(s, memory_mapping, phdr_index++, offset,
641 filesz, errp);
642 }
643
644 if (*errp) {
645 return;
646 }
647
648 if (phdr_index >= s->phdr_num) {
649 break;
650 }
651 }
652 }
653
654 static void write_elf_notes(DumpState *s, Error **errp)
655 {
656 if (dump_is_64bit(s)) {
657 write_elf64_notes(fd_write_vmcore, s, errp);
658 } else {
659 write_elf32_notes(fd_write_vmcore, s, errp);
660 }
661 }
662
663 /* write elf header, PT_NOTE and elf note to vmcore. */
664 static void dump_begin(DumpState *s, Error **errp)
665 {
666 ERRP_GUARD();
667
668 /*
669 * the vmcore's format is:
670 * --------------
671 * | elf header |
672 * --------------
673 * | sctn_hdr |
674 * --------------
675 * | PT_NOTE |
676 * --------------
677 * | PT_LOAD |
678 * --------------
679 * | ...... |
680 * --------------
681 * | PT_LOAD |
682 * --------------
683 * | elf note |
684 * --------------
685 * | memory |
686 * --------------
687 *
688 * we only know where the memory is saved after we write elf note into
689 * vmcore.
690 */
691
692 /* write elf header to vmcore */
693 write_elf_header(s, errp);
694 if (*errp) {
695 return;
696 }
697
698 /* write section headers to vmcore */
699 write_elf_section_headers(s, errp);
700 if (*errp) {
701 return;
702 }
703
704 /* write PT_NOTE to vmcore */
705 write_elf_phdr_note(s, errp);
706 if (*errp) {
707 return;
708 }
709
710 /* write all PT_LOADs to vmcore */
711 write_elf_phdr_loads(s, errp);
712 if (*errp) {
713 return;
714 }
715
716 /* write notes to vmcore */
717 write_elf_notes(s, errp);
718 }
719
720 int64_t dump_filtered_memblock_size(GuestPhysBlock *block,
721 int64_t filter_area_start,
722 int64_t filter_area_length)
723 {
724 int64_t size, left, right;
725
726 /* No filter, return full size */
727 if (!filter_area_length) {
728 return block->target_end - block->target_start;
729 }
730
731 /* calculate the overlapped region. */
732 left = MAX(filter_area_start, block->target_start);
733 right = MIN(filter_area_start + filter_area_length, block->target_end);
734 size = right - left;
735 size = size > 0 ? size : 0;
736
737 return size;
738 }
739
740 int64_t dump_filtered_memblock_start(GuestPhysBlock *block,
741 int64_t filter_area_start,
742 int64_t filter_area_length)
743 {
744 if (filter_area_length) {
745 /* return -1 if the block is not within filter area */
746 if (!ranges_overlap(block->target_start,
747 block->target_end - block->target_start,
748 filter_area_start, filter_area_length)) {
749 return -1;
750 }
751
752 if (filter_area_start > block->target_start) {
753 return filter_area_start - block->target_start;
754 }
755 }
756
757 return 0;
758 }
759
760 /* write all memory to vmcore */
761 static void dump_iterate(DumpState *s, Error **errp)
762 {
763 ERRP_GUARD();
764 GuestPhysBlock *block;
765 int64_t memblock_size, memblock_start;
766
767 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
768 memblock_start = dump_filtered_memblock_start(block, s->filter_area_begin, s->filter_area_length);
769 if (memblock_start == -1) {
770 continue;
771 }
772
773 memblock_size = dump_filtered_memblock_size(block, s->filter_area_begin, s->filter_area_length);
774
775 /* Write the memory to file */
776 write_memory(s, block, memblock_start, memblock_size, errp);
777 if (*errp) {
778 return;
779 }
780 }
781 }
782
783 static void dump_end(DumpState *s, Error **errp)
784 {
785 int rc;
786
787 if (s->elf_section_data_size) {
788 s->elf_section_data = g_malloc0(s->elf_section_data_size);
789 }
790
791 /* Adds the architecture defined section data to s->elf_section_data */
792 if (s->dump_info.arch_sections_write_fn &&
793 s->elf_section_data_size) {
794 rc = s->dump_info.arch_sections_write_fn(s, s->elf_section_data);
795 if (rc) {
796 error_setg_errno(errp, rc,
797 "dump: failed to get arch section data");
798 g_free(s->elf_section_data);
799 return;
800 }
801 }
802
803 /* write sections to vmcore */
804 write_elf_sections(s, errp);
805 }
806
807 static void create_vmcore(DumpState *s, Error **errp)
808 {
809 ERRP_GUARD();
810
811 dump_begin(s, errp);
812 if (*errp) {
813 return;
814 }
815
816 /* Iterate over memory and dump it to file */
817 dump_iterate(s, errp);
818 if (*errp) {
819 return;
820 }
821
822 /* Write the section data */
823 dump_end(s, errp);
824 }
825
826 static int write_start_flat_header(DumpState *s)
827 {
828 MakedumpfileHeader *mh;
829 int ret = 0;
830
831 if (s->kdump_raw) {
832 return 0;
833 }
834
835 QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
836 mh = g_malloc0(MAX_SIZE_MDF_HEADER);
837
838 memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
839 MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
840
841 mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
842 mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
843
844 size_t written_size;
845 written_size = qemu_write_full(s->fd, mh, MAX_SIZE_MDF_HEADER);
846 if (written_size != MAX_SIZE_MDF_HEADER) {
847 ret = -1;
848 }
849
850 g_free(mh);
851 return ret;
852 }
853
854 static int write_end_flat_header(DumpState *s)
855 {
856 MakedumpfileDataHeader mdh;
857
858 if (s->kdump_raw) {
859 return 0;
860 }
861
862 mdh.offset = END_FLAG_FLAT_HEADER;
863 mdh.buf_size = END_FLAG_FLAT_HEADER;
864
865 size_t written_size;
866 written_size = qemu_write_full(s->fd, &mdh, sizeof(mdh));
867 if (written_size != sizeof(mdh)) {
868 return -1;
869 }
870
871 return 0;
872 }
873
874 static int write_buffer(DumpState *s, off_t offset, const void *buf, size_t size)
875 {
876 size_t written_size;
877 MakedumpfileDataHeader mdh;
878 off_t seek_loc;
879
880 if (s->kdump_raw) {
881 seek_loc = lseek(s->fd, offset, SEEK_SET);
882 if (seek_loc == (off_t) -1) {
883 return -1;
884 }
885 } else {
886 mdh.offset = cpu_to_be64(offset);
887 mdh.buf_size = cpu_to_be64(size);
888
889 written_size = qemu_write_full(s->fd, &mdh, sizeof(mdh));
890 if (written_size != sizeof(mdh)) {
891 return -1;
892 }
893 }
894
895 written_size = qemu_write_full(s->fd, buf, size);
896 if (written_size != size) {
897 return -1;
898 }
899
900 return 0;
901 }
902
903 static int buf_write_note(const void *buf, size_t size, void *opaque)
904 {
905 DumpState *s = opaque;
906
907 /* note_buf is not enough */
908 if (s->note_buf_offset + size > s->note_size) {
909 return -1;
910 }
911
912 memcpy(s->note_buf + s->note_buf_offset, buf, size);
913
914 s->note_buf_offset += size;
915
916 return 0;
917 }
918
919 /*
920 * This function retrieves various sizes from an elf header.
921 *
922 * @note has to be a valid ELF note. The return sizes are unmodified
923 * (not padded or rounded up to be multiple of 4).
924 */
925 static void get_note_sizes(DumpState *s, const void *note,
926 uint64_t *note_head_size,
927 uint64_t *name_size,
928 uint64_t *desc_size)
929 {
930 uint64_t note_head_sz;
931 uint64_t name_sz;
932 uint64_t desc_sz;
933
934 if (dump_is_64bit(s)) {
935 const Elf64_Nhdr *hdr = note;
936 note_head_sz = sizeof(Elf64_Nhdr);
937 name_sz = cpu_to_dump64(s, hdr->n_namesz);
938 desc_sz = cpu_to_dump64(s, hdr->n_descsz);
939 } else {
940 const Elf32_Nhdr *hdr = note;
941 note_head_sz = sizeof(Elf32_Nhdr);
942 name_sz = cpu_to_dump32(s, hdr->n_namesz);
943 desc_sz = cpu_to_dump32(s, hdr->n_descsz);
944 }
945
946 if (note_head_size) {
947 *note_head_size = note_head_sz;
948 }
949 if (name_size) {
950 *name_size = name_sz;
951 }
952 if (desc_size) {
953 *desc_size = desc_sz;
954 }
955 }
956
957 static bool note_name_equal(DumpState *s,
958 const uint8_t *note, const char *name)
959 {
960 int len = strlen(name) + 1;
961 uint64_t head_size, name_size;
962
963 get_note_sizes(s, note, &head_size, &name_size, NULL);
964 head_size = ROUND_UP(head_size, 4);
965
966 return name_size == len && memcmp(note + head_size, name, len) == 0;
967 }
968
969 /* write common header, sub header and elf note to vmcore */
970 static void create_header32(DumpState *s, Error **errp)
971 {
972 ERRP_GUARD();
973 DiskDumpHeader32 *dh = NULL;
974 KdumpSubHeader32 *kh = NULL;
975 size_t size;
976 uint32_t block_size;
977 uint32_t sub_hdr_size;
978 uint32_t bitmap_blocks;
979 uint32_t status = 0;
980 uint64_t offset_note;
981
982 /* write common header, the version of kdump-compressed format is 6th */
983 size = sizeof(DiskDumpHeader32);
984 dh = g_malloc0(size);
985
986 memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
987 dh->header_version = cpu_to_dump32(s, 6);
988 block_size = s->dump_info.page_size;
989 dh->block_size = cpu_to_dump32(s, block_size);
990 sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
991 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
992 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
993 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
994 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
995 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
996 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
997 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
998 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
999
1000 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
1001 status |= DUMP_DH_COMPRESSED_ZLIB;
1002 }
1003 #ifdef CONFIG_LZO
1004 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
1005 status |= DUMP_DH_COMPRESSED_LZO;
1006 }
1007 #endif
1008 #ifdef CONFIG_SNAPPY
1009 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1010 status |= DUMP_DH_COMPRESSED_SNAPPY;
1011 }
1012 #endif
1013 dh->status = cpu_to_dump32(s, status);
1014
1015 if (write_buffer(s, 0, dh, size) < 0) {
1016 error_setg(errp, "dump: failed to write disk dump header");
1017 goto out;
1018 }
1019
1020 /* write sub header */
1021 size = sizeof(KdumpSubHeader32);
1022 kh = g_malloc0(size);
1023
1024 /* 64bit max_mapnr_64 */
1025 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1026 kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
1027 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1028
1029 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1030 if (s->guest_note &&
1031 note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1032 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1033
1034 get_note_sizes(s, s->guest_note,
1035 &hsize, &name_size, &size_vmcoreinfo_desc);
1036 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1037 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1038 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1039 kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
1040 }
1041
1042 kh->offset_note = cpu_to_dump64(s, offset_note);
1043 kh->note_size = cpu_to_dump32(s, s->note_size);
1044
1045 if (write_buffer(s, DISKDUMP_HEADER_BLOCKS *
1046 block_size, kh, size) < 0) {
1047 error_setg(errp, "dump: failed to write kdump sub header");
1048 goto out;
1049 }
1050
1051 /* write note */
1052 s->note_buf = g_malloc0(s->note_size);
1053 s->note_buf_offset = 0;
1054
1055 /* use s->note_buf to store notes temporarily */
1056 write_elf32_notes(buf_write_note, s, errp);
1057 if (*errp) {
1058 goto out;
1059 }
1060 if (write_buffer(s, offset_note, s->note_buf,
1061 s->note_size) < 0) {
1062 error_setg(errp, "dump: failed to write notes");
1063 goto out;
1064 }
1065
1066 /* get offset of dump_bitmap */
1067 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1068 block_size;
1069
1070 /* get offset of page */
1071 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1072 block_size;
1073
1074 out:
1075 g_free(dh);
1076 g_free(kh);
1077 g_free(s->note_buf);
1078 }
1079
1080 /* write common header, sub header and elf note to vmcore */
1081 static void create_header64(DumpState *s, Error **errp)
1082 {
1083 ERRP_GUARD();
1084 DiskDumpHeader64 *dh = NULL;
1085 KdumpSubHeader64 *kh = NULL;
1086 size_t size;
1087 uint32_t block_size;
1088 uint32_t sub_hdr_size;
1089 uint32_t bitmap_blocks;
1090 uint32_t status = 0;
1091 uint64_t offset_note;
1092
1093 /* write common header, the version of kdump-compressed format is 6th */
1094 size = sizeof(DiskDumpHeader64);
1095 dh = g_malloc0(size);
1096
1097 memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
1098 dh->header_version = cpu_to_dump32(s, 6);
1099 block_size = s->dump_info.page_size;
1100 dh->block_size = cpu_to_dump32(s, block_size);
1101 sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
1102 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
1103 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
1104 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
1105 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
1106 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
1107 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
1108 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
1109 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
1110
1111 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
1112 status |= DUMP_DH_COMPRESSED_ZLIB;
1113 }
1114 #ifdef CONFIG_LZO
1115 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
1116 status |= DUMP_DH_COMPRESSED_LZO;
1117 }
1118 #endif
1119 #ifdef CONFIG_SNAPPY
1120 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1121 status |= DUMP_DH_COMPRESSED_SNAPPY;
1122 }
1123 #endif
1124 dh->status = cpu_to_dump32(s, status);
1125
1126 if (write_buffer(s, 0, dh, size) < 0) {
1127 error_setg(errp, "dump: failed to write disk dump header");
1128 goto out;
1129 }
1130
1131 /* write sub header */
1132 size = sizeof(KdumpSubHeader64);
1133 kh = g_malloc0(size);
1134
1135 /* 64bit max_mapnr_64 */
1136 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1137 kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
1138 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1139
1140 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1141 if (s->guest_note &&
1142 note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1143 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1144
1145 get_note_sizes(s, s->guest_note,
1146 &hsize, &name_size, &size_vmcoreinfo_desc);
1147 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1148 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1149 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1150 kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
1151 }
1152
1153 kh->offset_note = cpu_to_dump64(s, offset_note);
1154 kh->note_size = cpu_to_dump64(s, s->note_size);
1155
1156 if (write_buffer(s, DISKDUMP_HEADER_BLOCKS *
1157 block_size, kh, size) < 0) {
1158 error_setg(errp, "dump: failed to write kdump sub header");
1159 goto out;
1160 }
1161
1162 /* write note */
1163 s->note_buf = g_malloc0(s->note_size);
1164 s->note_buf_offset = 0;
1165
1166 /* use s->note_buf to store notes temporarily */
1167 write_elf64_notes(buf_write_note, s, errp);
1168 if (*errp) {
1169 goto out;
1170 }
1171
1172 if (write_buffer(s, offset_note, s->note_buf,
1173 s->note_size) < 0) {
1174 error_setg(errp, "dump: failed to write notes");
1175 goto out;
1176 }
1177
1178 /* get offset of dump_bitmap */
1179 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1180 block_size;
1181
1182 /* get offset of page */
1183 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1184 block_size;
1185
1186 out:
1187 g_free(dh);
1188 g_free(kh);
1189 g_free(s->note_buf);
1190 }
1191
1192 static void write_dump_header(DumpState *s, Error **errp)
1193 {
1194 if (dump_is_64bit(s)) {
1195 create_header64(s, errp);
1196 } else {
1197 create_header32(s, errp);
1198 }
1199 }
1200
1201 static size_t dump_bitmap_get_bufsize(DumpState *s)
1202 {
1203 return s->dump_info.page_size;
1204 }
1205
1206 /*
1207 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1208 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1209 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1210 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1211 * vmcore, ie. synchronizing un-sync bit into vmcore.
1212 */
1213 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1214 uint8_t *buf, DumpState *s)
1215 {
1216 off_t old_offset, new_offset;
1217 off_t offset_bitmap1, offset_bitmap2;
1218 uint32_t byte, bit;
1219 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1220 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1221
1222 /* should not set the previous place */
1223 assert(last_pfn <= pfn);
1224
1225 /*
1226 * if the bit needed to be set is not cached in buf, flush the data in buf
1227 * to vmcore firstly.
1228 * making new_offset be bigger than old_offset can also sync remained data
1229 * into vmcore.
1230 */
1231 old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1232 new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1233
1234 while (old_offset < new_offset) {
1235 /* calculate the offset and write dump_bitmap */
1236 offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1237 if (write_buffer(s, offset_bitmap1, buf,
1238 bitmap_bufsize) < 0) {
1239 return -1;
1240 }
1241
1242 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1243 offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1244 old_offset;
1245 if (write_buffer(s, offset_bitmap2, buf,
1246 bitmap_bufsize) < 0) {
1247 return -1;
1248 }
1249
1250 memset(buf, 0, bitmap_bufsize);
1251 old_offset += bitmap_bufsize;
1252 }
1253
1254 /* get the exact place of the bit in the buf, and set it */
1255 byte = (pfn % bits_per_buf) / CHAR_BIT;
1256 bit = (pfn % bits_per_buf) % CHAR_BIT;
1257 if (value) {
1258 buf[byte] |= 1u << bit;
1259 } else {
1260 buf[byte] &= ~(1u << bit);
1261 }
1262
1263 return 0;
1264 }
1265
1266 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1267 {
1268 int target_page_shift = ctz32(s->dump_info.page_size);
1269
1270 return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1271 }
1272
1273 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1274 {
1275 int target_page_shift = ctz32(s->dump_info.page_size);
1276
1277 return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1278 }
1279
1280 /*
1281 * Return the page frame number and the page content in *bufptr. bufptr can be
1282 * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated
1283 * memory. This is not necessarily the memory returned.
1284 */
1285 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1286 uint8_t **bufptr, DumpState *s)
1287 {
1288 GuestPhysBlock *block = *blockptr;
1289 uint32_t page_size = s->dump_info.page_size;
1290 uint8_t *buf = NULL, *hbuf;
1291 hwaddr addr;
1292
1293 /* block == NULL means the start of the iteration */
1294 if (!block) {
1295 block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1296 assert(block);
1297 *blockptr = block;
1298 addr = block->target_start;
1299 *pfnptr = dump_paddr_to_pfn(s, addr);
1300 } else {
1301 *pfnptr += 1;
1302 addr = dump_pfn_to_paddr(s, *pfnptr);
1303 }
1304
1305 while (1) {
1306 if (addr >= block->target_start && addr < block->target_end) {
1307 size_t n = MIN(block->target_end - addr, page_size - addr % page_size);
1308 hbuf = block->host_addr + (addr - block->target_start);
1309 if (!buf) {
1310 if (n == page_size) {
1311 /* this is a whole target page, go for it */
1312 assert(addr % page_size == 0);
1313 buf = hbuf;
1314 break;
1315 } else if (bufptr) {
1316 assert(*bufptr);
1317 buf = *bufptr;
1318 memset(buf, 0, page_size);
1319 } else {
1320 return true;
1321 }
1322 }
1323
1324 memcpy(buf + addr % page_size, hbuf, n);
1325 addr += n;
1326 if (addr % page_size == 0 || addr >= block->target_end) {
1327 /* we filled up the page or the current block is finished */
1328 break;
1329 }
1330 } else {
1331 /* the next page is in the next block */
1332 *blockptr = block = QTAILQ_NEXT(block, next);
1333 if (!block) {
1334 break;
1335 }
1336
1337 addr = block->target_start;
1338 /* are we still in the same page? */
1339 if (dump_paddr_to_pfn(s, addr) != *pfnptr) {
1340 if (buf) {
1341 /* no, but we already filled something earlier, return it */
1342 break;
1343 } else {
1344 /* else continue from there */
1345 *pfnptr = dump_paddr_to_pfn(s, addr);
1346 }
1347 }
1348 }
1349 }
1350
1351 if (bufptr) {
1352 *bufptr = buf;
1353 }
1354
1355 return buf != NULL;
1356 }
1357
1358 static void write_dump_bitmap(DumpState *s, Error **errp)
1359 {
1360 int ret = 0;
1361 uint64_t last_pfn, pfn;
1362 void *dump_bitmap_buf;
1363 size_t num_dumpable;
1364 GuestPhysBlock *block_iter = NULL;
1365 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1366 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1367
1368 /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1369 dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1370
1371 num_dumpable = 0;
1372 last_pfn = 0;
1373
1374 /*
1375 * exam memory page by page, and set the bit in dump_bitmap corresponded
1376 * to the existing page.
1377 */
1378 while (get_next_page(&block_iter, &pfn, NULL, s)) {
1379 ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1380 if (ret < 0) {
1381 error_setg(errp, "dump: failed to set dump_bitmap");
1382 goto out;
1383 }
1384
1385 last_pfn = pfn;
1386 num_dumpable++;
1387 }
1388
1389 /*
1390 * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1391 * set the remaining bits from last_pfn to the end of the bitmap buffer to
1392 * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1393 */
1394 if (num_dumpable > 0) {
1395 ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1396 dump_bitmap_buf, s);
1397 if (ret < 0) {
1398 error_setg(errp, "dump: failed to sync dump_bitmap");
1399 goto out;
1400 }
1401 }
1402
1403 /* number of dumpable pages that will be dumped later */
1404 s->num_dumpable = num_dumpable;
1405
1406 out:
1407 g_free(dump_bitmap_buf);
1408 }
1409
1410 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1411 off_t offset)
1412 {
1413 data_cache->state = s;
1414 data_cache->data_size = 0;
1415 data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1416 data_cache->buf = g_malloc0(data_cache->buf_size);
1417 data_cache->offset = offset;
1418 }
1419
1420 static int write_cache(DataCache *dc, const void *buf, size_t size,
1421 bool flag_sync)
1422 {
1423 /*
1424 * dc->buf_size should not be less than size, otherwise dc will never be
1425 * enough
1426 */
1427 assert(size <= dc->buf_size);
1428
1429 /*
1430 * if flag_sync is set, synchronize data in dc->buf into vmcore.
1431 * otherwise check if the space is enough for caching data in buf, if not,
1432 * write the data in dc->buf to dc->state->fd and reset dc->buf
1433 */
1434 if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1435 (flag_sync && dc->data_size > 0)) {
1436 if (write_buffer(dc->state, dc->offset, dc->buf, dc->data_size) < 0) {
1437 return -1;
1438 }
1439
1440 dc->offset += dc->data_size;
1441 dc->data_size = 0;
1442 }
1443
1444 if (!flag_sync) {
1445 memcpy(dc->buf + dc->data_size, buf, size);
1446 dc->data_size += size;
1447 }
1448
1449 return 0;
1450 }
1451
1452 static void free_data_cache(DataCache *data_cache)
1453 {
1454 g_free(data_cache->buf);
1455 }
1456
1457 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1458 {
1459 switch (flag_compress) {
1460 case DUMP_DH_COMPRESSED_ZLIB:
1461 return compressBound(page_size);
1462
1463 case DUMP_DH_COMPRESSED_LZO:
1464 /*
1465 * LZO will expand incompressible data by a little amount. Please check
1466 * the following URL to see the expansion calculation:
1467 * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1468 */
1469 return page_size + page_size / 16 + 64 + 3;
1470
1471 #ifdef CONFIG_SNAPPY
1472 case DUMP_DH_COMPRESSED_SNAPPY:
1473 return snappy_max_compressed_length(page_size);
1474 #endif
1475 }
1476 return 0;
1477 }
1478
1479 static void write_dump_pages(DumpState *s, Error **errp)
1480 {
1481 int ret = 0;
1482 DataCache page_desc, page_data;
1483 size_t len_buf_out, size_out;
1484 #ifdef CONFIG_LZO
1485 lzo_bytep wrkmem = NULL;
1486 #endif
1487 uint8_t *buf_out = NULL;
1488 off_t offset_desc, offset_data;
1489 PageDescriptor pd, pd_zero;
1490 uint8_t *buf;
1491 GuestPhysBlock *block_iter = NULL;
1492 uint64_t pfn_iter;
1493 g_autofree uint8_t *page = NULL;
1494
1495 /* get offset of page_desc and page_data in dump file */
1496 offset_desc = s->offset_page;
1497 offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1498
1499 prepare_data_cache(&page_desc, s, offset_desc);
1500 prepare_data_cache(&page_data, s, offset_data);
1501
1502 /* prepare buffer to store compressed data */
1503 len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1504 assert(len_buf_out != 0);
1505
1506 #ifdef CONFIG_LZO
1507 wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1508 #endif
1509
1510 buf_out = g_malloc(len_buf_out);
1511
1512 /*
1513 * init zero page's page_desc and page_data, because every zero page
1514 * uses the same page_data
1515 */
1516 pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1517 pd_zero.flags = cpu_to_dump32(s, 0);
1518 pd_zero.offset = cpu_to_dump64(s, offset_data);
1519 pd_zero.page_flags = cpu_to_dump64(s, 0);
1520 buf = g_malloc0(s->dump_info.page_size);
1521 ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1522 g_free(buf);
1523 if (ret < 0) {
1524 error_setg(errp, "dump: failed to write page data (zero page)");
1525 goto out;
1526 }
1527
1528 offset_data += s->dump_info.page_size;
1529 page = g_malloc(s->dump_info.page_size);
1530
1531 /*
1532 * dump memory to vmcore page by page. zero page will all be resided in the
1533 * first page of page section
1534 */
1535 for (buf = page; get_next_page(&block_iter, &pfn_iter, &buf, s); buf = page) {
1536 /* check zero page */
1537 if (buffer_is_zero(buf, s->dump_info.page_size)) {
1538 ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1539 false);
1540 if (ret < 0) {
1541 error_setg(errp, "dump: failed to write page desc");
1542 goto out;
1543 }
1544 } else {
1545 /*
1546 * not zero page, then:
1547 * 1. compress the page
1548 * 2. write the compressed page into the cache of page_data
1549 * 3. get page desc of the compressed page and write it into the
1550 * cache of page_desc
1551 *
1552 * only one compression format will be used here, for
1553 * s->flag_compress is set. But when compression fails to work,
1554 * we fall back to save in plaintext.
1555 */
1556 size_out = len_buf_out;
1557 if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1558 (compress2(buf_out, (uLongf *)&size_out, buf,
1559 s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1560 (size_out < s->dump_info.page_size)) {
1561 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1562 pd.size = cpu_to_dump32(s, size_out);
1563
1564 ret = write_cache(&page_data, buf_out, size_out, false);
1565 if (ret < 0) {
1566 error_setg(errp, "dump: failed to write page data");
1567 goto out;
1568 }
1569 #ifdef CONFIG_LZO
1570 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1571 (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1572 (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1573 (size_out < s->dump_info.page_size)) {
1574 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1575 pd.size = cpu_to_dump32(s, size_out);
1576
1577 ret = write_cache(&page_data, buf_out, size_out, false);
1578 if (ret < 0) {
1579 error_setg(errp, "dump: failed to write page data");
1580 goto out;
1581 }
1582 #endif
1583 #ifdef CONFIG_SNAPPY
1584 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1585 (snappy_compress((char *)buf, s->dump_info.page_size,
1586 (char *)buf_out, &size_out) == SNAPPY_OK) &&
1587 (size_out < s->dump_info.page_size)) {
1588 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1589 pd.size = cpu_to_dump32(s, size_out);
1590
1591 ret = write_cache(&page_data, buf_out, size_out, false);
1592 if (ret < 0) {
1593 error_setg(errp, "dump: failed to write page data");
1594 goto out;
1595 }
1596 #endif
1597 } else {
1598 /*
1599 * fall back to save in plaintext, size_out should be
1600 * assigned the target's page size
1601 */
1602 pd.flags = cpu_to_dump32(s, 0);
1603 size_out = s->dump_info.page_size;
1604 pd.size = cpu_to_dump32(s, size_out);
1605
1606 ret = write_cache(&page_data, buf,
1607 s->dump_info.page_size, false);
1608 if (ret < 0) {
1609 error_setg(errp, "dump: failed to write page data");
1610 goto out;
1611 }
1612 }
1613
1614 /* get and write page desc here */
1615 pd.page_flags = cpu_to_dump64(s, 0);
1616 pd.offset = cpu_to_dump64(s, offset_data);
1617 offset_data += size_out;
1618
1619 ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1620 if (ret < 0) {
1621 error_setg(errp, "dump: failed to write page desc");
1622 goto out;
1623 }
1624 }
1625 s->written_size += s->dump_info.page_size;
1626 }
1627
1628 ret = write_cache(&page_desc, NULL, 0, true);
1629 if (ret < 0) {
1630 error_setg(errp, "dump: failed to sync cache for page_desc");
1631 goto out;
1632 }
1633 ret = write_cache(&page_data, NULL, 0, true);
1634 if (ret < 0) {
1635 error_setg(errp, "dump: failed to sync cache for page_data");
1636 goto out;
1637 }
1638
1639 out:
1640 free_data_cache(&page_desc);
1641 free_data_cache(&page_data);
1642
1643 #ifdef CONFIG_LZO
1644 g_free(wrkmem);
1645 #endif
1646
1647 g_free(buf_out);
1648 }
1649
1650 static void create_kdump_vmcore(DumpState *s, Error **errp)
1651 {
1652 ERRP_GUARD();
1653 int ret;
1654
1655 /*
1656 * the kdump-compressed format is:
1657 * File offset
1658 * +------------------------------------------+ 0x0
1659 * | main header (struct disk_dump_header) |
1660 * |------------------------------------------+ block 1
1661 * | sub header (struct kdump_sub_header) |
1662 * |------------------------------------------+ block 2
1663 * | 1st-dump_bitmap |
1664 * |------------------------------------------+ block 2 + X blocks
1665 * | 2nd-dump_bitmap | (aligned by block)
1666 * |------------------------------------------+ block 2 + 2 * X blocks
1667 * | page desc for pfn 0 (struct page_desc) | (aligned by block)
1668 * | page desc for pfn 1 (struct page_desc) |
1669 * | : |
1670 * |------------------------------------------| (not aligned by block)
1671 * | page data (pfn 0) |
1672 * | page data (pfn 1) |
1673 * | : |
1674 * +------------------------------------------+
1675 */
1676
1677 ret = write_start_flat_header(s);
1678 if (ret < 0) {
1679 error_setg(errp, "dump: failed to write start flat header");
1680 return;
1681 }
1682
1683 write_dump_header(s, errp);
1684 if (*errp) {
1685 return;
1686 }
1687
1688 write_dump_bitmap(s, errp);
1689 if (*errp) {
1690 return;
1691 }
1692
1693 write_dump_pages(s, errp);
1694 if (*errp) {
1695 return;
1696 }
1697
1698 ret = write_end_flat_header(s);
1699 if (ret < 0) {
1700 error_setg(errp, "dump: failed to write end flat header");
1701 return;
1702 }
1703 }
1704
1705 static void get_max_mapnr(DumpState *s)
1706 {
1707 GuestPhysBlock *last_block;
1708
1709 last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1710 s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1711 }
1712
1713 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1714
1715 static void dump_state_prepare(DumpState *s)
1716 {
1717 /* zero the struct, setting status to active and fd to -1 */
1718 *s = (DumpState) { .fd = -1, .status = DUMP_STATUS_ACTIVE };
1719 }
1720
1721 bool qemu_system_dump_in_progress(void)
1722 {
1723 DumpState *state = &dump_state_global;
1724 return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1725 }
1726
1727 /*
1728 * calculate total size of memory to be dumped (taking filter into
1729 * account.)
1730 */
1731 static int64_t dump_calculate_size(DumpState *s)
1732 {
1733 GuestPhysBlock *block;
1734 int64_t total = 0;
1735
1736 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1737 total += dump_filtered_memblock_size(block,
1738 s->filter_area_begin,
1739 s->filter_area_length);
1740 }
1741
1742 return total;
1743 }
1744
1745 static void vmcoreinfo_update_phys_base(DumpState *s)
1746 {
1747 uint64_t size, note_head_size, name_size, phys_base;
1748 char **lines;
1749 uint8_t *vmci;
1750 size_t i;
1751
1752 if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1753 return;
1754 }
1755
1756 get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1757 note_head_size = ROUND_UP(note_head_size, 4);
1758
1759 vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1760 *(vmci + size) = '\0';
1761
1762 lines = g_strsplit((char *)vmci, "\n", -1);
1763 for (i = 0; lines[i]; i++) {
1764 const char *prefix = NULL;
1765
1766 if (s->dump_info.d_machine == EM_X86_64) {
1767 prefix = "NUMBER(phys_base)=";
1768 } else if (s->dump_info.d_machine == EM_AARCH64) {
1769 prefix = "NUMBER(PHYS_OFFSET)=";
1770 }
1771
1772 if (prefix && g_str_has_prefix(lines[i], prefix)) {
1773 if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1774 &phys_base) < 0) {
1775 warn_report("failed to parse %s in VMCOREINFO: '%s'",
1776 prefix, lines[i] + strlen(prefix));
1777 } else {
1778 s->dump_info.phys_base = phys_base;
1779 }
1780 break;
1781 }
1782 }
1783
1784 g_strfreev(lines);
1785 }
1786
1787 static void dump_init(DumpState *s, int fd, bool has_format,
1788 DumpGuestMemoryFormat format, bool paging, bool has_filter,
1789 int64_t begin, int64_t length, bool kdump_raw,
1790 Error **errp)
1791 {
1792 ERRP_GUARD();
1793 VMCoreInfoState *vmci = vmcoreinfo_find();
1794 CPUState *cpu;
1795 int nr_cpus;
1796 int ret;
1797
1798 s->has_format = has_format;
1799 s->format = format;
1800 s->written_size = 0;
1801 s->kdump_raw = kdump_raw;
1802
1803 /* kdump-compressed is conflict with paging and filter */
1804 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1805 assert(!paging && !has_filter);
1806 }
1807
1808 if (runstate_is_running()) {
1809 vm_stop(RUN_STATE_SAVE_VM);
1810 s->resume = true;
1811 } else {
1812 s->resume = false;
1813 }
1814
1815 /* If we use KVM, we should synchronize the registers before we get dump
1816 * info or physmap info.
1817 */
1818 cpu_synchronize_all_states();
1819 nr_cpus = 0;
1820 CPU_FOREACH(cpu) {
1821 nr_cpus++;
1822 }
1823
1824 s->fd = fd;
1825 if (has_filter && !length) {
1826 error_setg(errp, "parameter 'length' expects a non-zero size");
1827 goto cleanup;
1828 }
1829 s->filter_area_begin = begin;
1830 s->filter_area_length = length;
1831
1832 /* First index is 0, it's the special null name */
1833 s->string_table_buf = g_array_new(FALSE, TRUE, 1);
1834 /*
1835 * Allocate the null name, due to the clearing option set to true
1836 * it will be 0.
1837 */
1838 g_array_set_size(s->string_table_buf, 1);
1839
1840 memory_mapping_list_init(&s->list);
1841
1842 guest_phys_blocks_init(&s->guest_phys_blocks);
1843 guest_phys_blocks_append(&s->guest_phys_blocks);
1844 s->total_size = dump_calculate_size(s);
1845 #ifdef DEBUG_DUMP_GUEST_MEMORY
1846 fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1847 #endif
1848
1849 /* it does not make sense to dump non-existent memory */
1850 if (!s->total_size) {
1851 error_setg(errp, "dump: no guest memory to dump");
1852 goto cleanup;
1853 }
1854
1855 /* get dump info: endian, class and architecture.
1856 * If the target architecture is not supported, cpu_get_dump_info() will
1857 * return -1.
1858 */
1859 ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1860 if (ret < 0) {
1861 error_setg(errp,
1862 "dumping guest memory is not supported on this target");
1863 goto cleanup;
1864 }
1865
1866 if (!s->dump_info.page_size) {
1867 s->dump_info.page_size = qemu_target_page_size();
1868 }
1869
1870 s->note_size = cpu_get_note_size(s->dump_info.d_class,
1871 s->dump_info.d_machine, nr_cpus);
1872 assert(s->note_size >= 0);
1873
1874 /*
1875 * The goal of this block is to (a) update the previously guessed
1876 * phys_base, (b) copy the guest note out of the guest.
1877 * Failure to do so is not fatal for dumping.
1878 */
1879 if (vmci) {
1880 uint64_t addr, note_head_size, name_size, desc_size;
1881 uint32_t size;
1882 uint16_t guest_format;
1883
1884 note_head_size = dump_is_64bit(s) ?
1885 sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr);
1886
1887 guest_format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1888 size = le32_to_cpu(vmci->vmcoreinfo.size);
1889 addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1890 if (!vmci->has_vmcoreinfo) {
1891 warn_report("guest note is not present");
1892 } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1893 warn_report("guest note size is invalid: %" PRIu32, size);
1894 } else if (guest_format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1895 warn_report("guest note format is unsupported: %" PRIu16, guest_format);
1896 } else {
1897 s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1898 physical_memory_read(addr, s->guest_note, size);
1899
1900 get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1901 s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1902 desc_size);
1903 if (name_size > MAX_GUEST_NOTE_SIZE ||
1904 desc_size > MAX_GUEST_NOTE_SIZE ||
1905 s->guest_note_size > size) {
1906 warn_report("Invalid guest note header");
1907 g_free(s->guest_note);
1908 s->guest_note = NULL;
1909 } else {
1910 vmcoreinfo_update_phys_base(s);
1911 s->note_size += s->guest_note_size;
1912 }
1913 }
1914 }
1915
1916 /* get memory mapping */
1917 if (paging) {
1918 qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1919 if (*errp) {
1920 goto cleanup;
1921 }
1922 } else {
1923 qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1924 }
1925
1926 s->nr_cpus = nr_cpus;
1927
1928 get_max_mapnr(s);
1929
1930 uint64_t tmp;
1931 tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1932 s->dump_info.page_size);
1933 s->len_dump_bitmap = tmp * s->dump_info.page_size;
1934
1935 /* init for kdump-compressed format */
1936 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1937 switch (format) {
1938 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1939 s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1940 break;
1941
1942 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1943 #ifdef CONFIG_LZO
1944 if (lzo_init() != LZO_E_OK) {
1945 error_setg(errp, "failed to initialize the LZO library");
1946 goto cleanup;
1947 }
1948 #endif
1949 s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1950 break;
1951
1952 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1953 s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1954 break;
1955
1956 default:
1957 s->flag_compress = 0;
1958 }
1959
1960 return;
1961 }
1962
1963 if (dump_has_filter(s)) {
1964 memory_mapping_filter(&s->list, s->filter_area_begin, s->filter_area_length);
1965 }
1966
1967 /*
1968 * The first section header is always a special one in which most
1969 * fields are 0. The section header string table is also always
1970 * set.
1971 */
1972 s->shdr_num = 2;
1973
1974 /*
1975 * Adds the number of architecture sections to shdr_num and sets
1976 * elf_section_data_size so we know the offsets and sizes of all
1977 * parts.
1978 */
1979 if (s->dump_info.arch_sections_add_fn) {
1980 s->dump_info.arch_sections_add_fn(s);
1981 }
1982
1983 /*
1984 * calculate shdr_num so we know the offsets and sizes of all
1985 * parts.
1986 * Calculate phdr_num
1987 *
1988 * The absolute maximum amount of phdrs is UINT32_MAX - 1 as
1989 * sh_info is 32 bit. There's special handling once we go over
1990 * UINT16_MAX - 1 but that is handled in the ehdr and section
1991 * code.
1992 */
1993 s->phdr_num = 1; /* Reserve PT_NOTE */
1994 if (s->list.num <= UINT32_MAX - 1) {
1995 s->phdr_num += s->list.num;
1996 } else {
1997 s->phdr_num = UINT32_MAX;
1998 }
1999
2000 /*
2001 * Now that the number of section and program headers is known we
2002 * can calculate the offsets of the headers and data.
2003 */
2004 if (dump_is_64bit(s)) {
2005 s->shdr_offset = sizeof(Elf64_Ehdr);
2006 s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
2007 s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
2008 } else {
2009 s->shdr_offset = sizeof(Elf32_Ehdr);
2010 s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
2011 s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
2012 }
2013 s->memory_offset = s->note_offset + s->note_size;
2014 s->section_offset = s->memory_offset + s->total_size;
2015
2016 return;
2017
2018 cleanup:
2019 dump_cleanup(s);
2020 }
2021
2022 /* this operation might be time consuming. */
2023 static void dump_process(DumpState *s, Error **errp)
2024 {
2025 ERRP_GUARD();
2026 DumpQueryResult *result = NULL;
2027
2028 if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2029 create_win_dump(s, errp);
2030 } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
2031 create_kdump_vmcore(s, errp);
2032 } else {
2033 create_vmcore(s, errp);
2034 }
2035
2036 /* make sure status is written after written_size updates */
2037 smp_wmb();
2038 qatomic_set(&s->status,
2039 (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
2040
2041 /* send DUMP_COMPLETED message (unconditionally) */
2042 result = qmp_query_dump(NULL);
2043 /* should never fail */
2044 assert(result);
2045 qapi_event_send_dump_completed(result,
2046 *errp ? error_get_pretty(*errp) : NULL);
2047 qapi_free_DumpQueryResult(result);
2048
2049 dump_cleanup(s);
2050 }
2051
2052 static void *dump_thread(void *data)
2053 {
2054 DumpState *s = (DumpState *)data;
2055 dump_process(s, NULL);
2056 return NULL;
2057 }
2058
2059 DumpQueryResult *qmp_query_dump(Error **errp)
2060 {
2061 DumpQueryResult *result = g_new(DumpQueryResult, 1);
2062 DumpState *state = &dump_state_global;
2063 result->status = qatomic_read(&state->status);
2064 /* make sure we are reading status and written_size in order */
2065 smp_rmb();
2066 result->completed = state->written_size;
2067 result->total = state->total_size;
2068 return result;
2069 }
2070
2071 void qmp_dump_guest_memory(bool paging, const char *protocol,
2072 bool has_detach, bool detach,
2073 bool has_begin, int64_t begin,
2074 bool has_length, int64_t length,
2075 bool has_format, DumpGuestMemoryFormat format,
2076 Error **errp)
2077 {
2078 ERRP_GUARD();
2079 const char *p;
2080 int fd;
2081 DumpState *s;
2082 bool detach_p = false;
2083 bool kdump_raw = false;
2084
2085 if (migration_guest_ram_loading()) {
2086 error_setg(errp, "Dump not allowed during migration.");
2087 return;
2088 }
2089
2090 /* if there is a dump in background, we should wait until the dump
2091 * finished */
2092 if (qemu_system_dump_in_progress()) {
2093 error_setg(errp, "There is a dump in process, please wait.");
2094 return;
2095 }
2096
2097 /*
2098 * externally, we represent kdump-raw-* as separate formats, but internally
2099 * they are handled the same, except for the "raw" flag
2100 */
2101 if (has_format) {
2102 switch (format) {
2103 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_ZLIB:
2104 format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
2105 kdump_raw = true;
2106 break;
2107 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_LZO:
2108 format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
2109 kdump_raw = true;
2110 break;
2111 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_SNAPPY:
2112 format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
2113 kdump_raw = true;
2114 break;
2115 default:
2116 break;
2117 }
2118 }
2119
2120 /*
2121 * kdump-compressed format need the whole memory dumped, so paging or
2122 * filter is not supported here.
2123 */
2124 if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
2125 (paging || has_begin || has_length)) {
2126 error_setg(errp, "kdump-compressed format doesn't support paging or "
2127 "filter");
2128 return;
2129 }
2130 if (has_begin && !has_length) {
2131 error_setg(errp, QERR_MISSING_PARAMETER, "length");
2132 return;
2133 }
2134 if (!has_begin && has_length) {
2135 error_setg(errp, QERR_MISSING_PARAMETER, "begin");
2136 return;
2137 }
2138 if (has_detach) {
2139 detach_p = detach;
2140 }
2141
2142 /* check whether lzo/snappy is supported */
2143 #ifndef CONFIG_LZO
2144 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
2145 error_setg(errp, "kdump-lzo is not available now");
2146 return;
2147 }
2148 #endif
2149
2150 #ifndef CONFIG_SNAPPY
2151 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
2152 error_setg(errp, "kdump-snappy is not available now");
2153 return;
2154 }
2155 #endif
2156
2157 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
2158 && !win_dump_available(errp)) {
2159 return;
2160 }
2161
2162 if (strstart(protocol, "fd:", &p)) {
2163 fd = monitor_get_fd(monitor_cur(), p, errp);
2164 if (fd == -1) {
2165 return;
2166 }
2167 } else if (strstart(protocol, "file:", &p)) {
2168 fd = qemu_create(p, O_WRONLY | O_TRUNC | O_BINARY, S_IRUSR, errp);
2169 if (fd < 0) {
2170 return;
2171 }
2172 } else {
2173 error_setg(errp,
2174 "parameter 'protocol' must start with 'file:' or 'fd:'");
2175 return;
2176 }
2177 if (kdump_raw && lseek(fd, 0, SEEK_CUR) == (off_t) -1) {
2178 close(fd);
2179 error_setg(errp, "kdump-raw formats require a seekable file");
2180 return;
2181 }
2182
2183 if (!dump_migration_blocker) {
2184 error_setg(&dump_migration_blocker,
2185 "Live migration disabled: dump-guest-memory in progress");
2186 }
2187
2188 /*
2189 * Allows even for -only-migratable, but forbid migration during the
2190 * process of dump guest memory.
2191 */
2192 if (migrate_add_blocker_internal(&dump_migration_blocker, errp)) {
2193 /* Remember to release the fd before passing it over to dump state */
2194 close(fd);
2195 return;
2196 }
2197
2198 s = &dump_state_global;
2199 dump_state_prepare(s);
2200
2201 dump_init(s, fd, has_format, format, paging, has_begin,
2202 begin, length, kdump_raw, errp);
2203 if (*errp) {
2204 qatomic_set(&s->status, DUMP_STATUS_FAILED);
2205 return;
2206 }
2207
2208 if (detach_p) {
2209 /* detached dump */
2210 s->detached = true;
2211 qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2212 s, QEMU_THREAD_DETACHED);
2213 } else {
2214 /* sync dump */
2215 dump_process(s, errp);
2216 }
2217 }
2218
2219 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2220 {
2221 DumpGuestMemoryCapability *cap =
2222 g_new0(DumpGuestMemoryCapability, 1);
2223 DumpGuestMemoryFormatList **tail = &cap->formats;
2224
2225 /* elf is always available */
2226 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2227
2228 /* kdump-zlib is always available */
2229 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2230 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_ZLIB);
2231
2232 /* add new item if kdump-lzo is available */
2233 #ifdef CONFIG_LZO
2234 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2235 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_LZO);
2236 #endif
2237
2238 /* add new item if kdump-snappy is available */
2239 #ifdef CONFIG_SNAPPY
2240 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2241 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_SNAPPY);
2242 #endif
2243
2244 if (win_dump_available(NULL)) {
2245 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2246 }
2247
2248 return cap;
2249 }