xref: /openbmc/qemu/dump/dump.c (revision 86a518bb)
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 "exec/hwaddr.h"
18 #include "monitor/monitor.h"
19 #include "sysemu/kvm.h"
20 #include "sysemu/dump.h"
21 #include "sysemu/memory_mapping.h"
22 #include "sysemu/runstate.h"
23 #include "sysemu/cpus.h"
24 #include "qapi/error.h"
25 #include "qapi/qapi-commands-dump.h"
26 #include "qapi/qapi-events-dump.h"
27 #include "qapi/qmp/qerror.h"
28 #include "qemu/error-report.h"
29 #include "qemu/main-loop.h"
30 #include "hw/misc/vmcoreinfo.h"
31 #include "migration/blocker.h"
32 
33 #ifdef TARGET_X86_64
34 #include "win_dump.h"
35 #endif
36 
37 #include <zlib.h>
38 #ifdef CONFIG_LZO
39 #include <lzo/lzo1x.h>
40 #endif
41 #ifdef CONFIG_SNAPPY
42 #include <snappy-c.h>
43 #endif
44 #ifndef ELF_MACHINE_UNAME
45 #define ELF_MACHINE_UNAME "Unknown"
46 #endif
47 
48 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
49 
50 static Error *dump_migration_blocker;
51 
52 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size)   \
53     ((DIV_ROUND_UP((hdr_size), 4) +                     \
54       DIV_ROUND_UP((name_size), 4) +                    \
55       DIV_ROUND_UP((desc_size), 4)) * 4)
56 
57 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
58 {
59     if (s->dump_info.d_endian == ELFDATA2LSB) {
60         val = cpu_to_le16(val);
61     } else {
62         val = cpu_to_be16(val);
63     }
64 
65     return val;
66 }
67 
68 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
69 {
70     if (s->dump_info.d_endian == ELFDATA2LSB) {
71         val = cpu_to_le32(val);
72     } else {
73         val = cpu_to_be32(val);
74     }
75 
76     return val;
77 }
78 
79 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
80 {
81     if (s->dump_info.d_endian == ELFDATA2LSB) {
82         val = cpu_to_le64(val);
83     } else {
84         val = cpu_to_be64(val);
85     }
86 
87     return val;
88 }
89 
90 static int dump_cleanup(DumpState *s)
91 {
92     guest_phys_blocks_free(&s->guest_phys_blocks);
93     memory_mapping_list_free(&s->list);
94     close(s->fd);
95     g_free(s->guest_note);
96     s->guest_note = NULL;
97     if (s->resume) {
98         if (s->detached) {
99             qemu_mutex_lock_iothread();
100         }
101         vm_start();
102         if (s->detached) {
103             qemu_mutex_unlock_iothread();
104         }
105     }
106     migrate_del_blocker(dump_migration_blocker);
107 
108     return 0;
109 }
110 
111 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
112 {
113     DumpState *s = opaque;
114     size_t written_size;
115 
116     written_size = qemu_write_full(s->fd, buf, size);
117     if (written_size != size) {
118         return -errno;
119     }
120 
121     return 0;
122 }
123 
124 static void write_elf64_header(DumpState *s, Error **errp)
125 {
126     Elf64_Ehdr elf_header;
127     int ret;
128 
129     memset(&elf_header, 0, sizeof(Elf64_Ehdr));
130     memcpy(&elf_header, ELFMAG, SELFMAG);
131     elf_header.e_ident[EI_CLASS] = ELFCLASS64;
132     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
133     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
134     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
135     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
136     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
137     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
138     elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
139     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
140     elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
141     if (s->have_section) {
142         uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
143 
144         elf_header.e_shoff = cpu_to_dump64(s, shoff);
145         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
146         elf_header.e_shnum = cpu_to_dump16(s, 1);
147     }
148 
149     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
150     if (ret < 0) {
151         error_setg_errno(errp, -ret, "dump: failed to write elf header");
152     }
153 }
154 
155 static void write_elf32_header(DumpState *s, Error **errp)
156 {
157     Elf32_Ehdr elf_header;
158     int ret;
159 
160     memset(&elf_header, 0, sizeof(Elf32_Ehdr));
161     memcpy(&elf_header, ELFMAG, SELFMAG);
162     elf_header.e_ident[EI_CLASS] = ELFCLASS32;
163     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
164     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
165     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
166     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
167     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
168     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
169     elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
170     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
171     elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
172     if (s->have_section) {
173         uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
174 
175         elf_header.e_shoff = cpu_to_dump32(s, shoff);
176         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
177         elf_header.e_shnum = cpu_to_dump16(s, 1);
178     }
179 
180     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
181     if (ret < 0) {
182         error_setg_errno(errp, -ret, "dump: failed to write elf header");
183     }
184 }
185 
186 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
187                              int phdr_index, hwaddr offset,
188                              hwaddr filesz, Error **errp)
189 {
190     Elf64_Phdr phdr;
191     int ret;
192 
193     memset(&phdr, 0, sizeof(Elf64_Phdr));
194     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
195     phdr.p_offset = cpu_to_dump64(s, offset);
196     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
197     phdr.p_filesz = cpu_to_dump64(s, filesz);
198     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
199     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
200 
201     assert(memory_mapping->length >= filesz);
202 
203     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
204     if (ret < 0) {
205         error_setg_errno(errp, -ret,
206                          "dump: failed to write program header table");
207     }
208 }
209 
210 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
211                              int phdr_index, hwaddr offset,
212                              hwaddr filesz, Error **errp)
213 {
214     Elf32_Phdr phdr;
215     int ret;
216 
217     memset(&phdr, 0, sizeof(Elf32_Phdr));
218     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
219     phdr.p_offset = cpu_to_dump32(s, offset);
220     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
221     phdr.p_filesz = cpu_to_dump32(s, filesz);
222     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
223     phdr.p_vaddr =
224         cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
225 
226     assert(memory_mapping->length >= filesz);
227 
228     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
229     if (ret < 0) {
230         error_setg_errno(errp, -ret,
231                          "dump: failed to write program header table");
232     }
233 }
234 
235 static void write_elf64_note(DumpState *s, Error **errp)
236 {
237     Elf64_Phdr phdr;
238     hwaddr begin = s->memory_offset - s->note_size;
239     int ret;
240 
241     memset(&phdr, 0, sizeof(Elf64_Phdr));
242     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
243     phdr.p_offset = cpu_to_dump64(s, begin);
244     phdr.p_paddr = 0;
245     phdr.p_filesz = cpu_to_dump64(s, s->note_size);
246     phdr.p_memsz = cpu_to_dump64(s, s->note_size);
247     phdr.p_vaddr = 0;
248 
249     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
250     if (ret < 0) {
251         error_setg_errno(errp, -ret,
252                          "dump: failed to write program header table");
253     }
254 }
255 
256 static inline int cpu_index(CPUState *cpu)
257 {
258     return cpu->cpu_index + 1;
259 }
260 
261 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
262                              Error **errp)
263 {
264     int ret;
265 
266     if (s->guest_note) {
267         ret = f(s->guest_note, s->guest_note_size, s);
268         if (ret < 0) {
269             error_setg(errp, "dump: failed to write guest note");
270         }
271     }
272 }
273 
274 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
275                               Error **errp)
276 {
277     CPUState *cpu;
278     int ret;
279     int id;
280 
281     CPU_FOREACH(cpu) {
282         id = cpu_index(cpu);
283         ret = cpu_write_elf64_note(f, cpu, id, s);
284         if (ret < 0) {
285             error_setg(errp, "dump: failed to write elf notes");
286             return;
287         }
288     }
289 
290     CPU_FOREACH(cpu) {
291         ret = cpu_write_elf64_qemunote(f, cpu, s);
292         if (ret < 0) {
293             error_setg(errp, "dump: failed to write CPU status");
294             return;
295         }
296     }
297 
298     write_guest_note(f, s, errp);
299 }
300 
301 static void write_elf32_note(DumpState *s, Error **errp)
302 {
303     hwaddr begin = s->memory_offset - s->note_size;
304     Elf32_Phdr phdr;
305     int ret;
306 
307     memset(&phdr, 0, sizeof(Elf32_Phdr));
308     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
309     phdr.p_offset = cpu_to_dump32(s, begin);
310     phdr.p_paddr = 0;
311     phdr.p_filesz = cpu_to_dump32(s, s->note_size);
312     phdr.p_memsz = cpu_to_dump32(s, s->note_size);
313     phdr.p_vaddr = 0;
314 
315     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
316     if (ret < 0) {
317         error_setg_errno(errp, -ret,
318                          "dump: failed to write program header table");
319     }
320 }
321 
322 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
323                               Error **errp)
324 {
325     CPUState *cpu;
326     int ret;
327     int id;
328 
329     CPU_FOREACH(cpu) {
330         id = cpu_index(cpu);
331         ret = cpu_write_elf32_note(f, cpu, id, s);
332         if (ret < 0) {
333             error_setg(errp, "dump: failed to write elf notes");
334             return;
335         }
336     }
337 
338     CPU_FOREACH(cpu) {
339         ret = cpu_write_elf32_qemunote(f, cpu, s);
340         if (ret < 0) {
341             error_setg(errp, "dump: failed to write CPU status");
342             return;
343         }
344     }
345 
346     write_guest_note(f, s, errp);
347 }
348 
349 static void write_elf_section(DumpState *s, int type, Error **errp)
350 {
351     Elf32_Shdr shdr32;
352     Elf64_Shdr shdr64;
353     int shdr_size;
354     void *shdr;
355     int ret;
356 
357     if (type == 0) {
358         shdr_size = sizeof(Elf32_Shdr);
359         memset(&shdr32, 0, shdr_size);
360         shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
361         shdr = &shdr32;
362     } else {
363         shdr_size = sizeof(Elf64_Shdr);
364         memset(&shdr64, 0, shdr_size);
365         shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
366         shdr = &shdr64;
367     }
368 
369     ret = fd_write_vmcore(shdr, shdr_size, s);
370     if (ret < 0) {
371         error_setg_errno(errp, -ret,
372                          "dump: failed to write section header table");
373     }
374 }
375 
376 static void write_data(DumpState *s, void *buf, int length, Error **errp)
377 {
378     int ret;
379 
380     ret = fd_write_vmcore(buf, length, s);
381     if (ret < 0) {
382         error_setg_errno(errp, -ret, "dump: failed to save memory");
383     } else {
384         s->written_size += length;
385     }
386 }
387 
388 /* write the memory to vmcore. 1 page per I/O. */
389 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
390                          int64_t size, Error **errp)
391 {
392     ERRP_GUARD();
393     int64_t i;
394 
395     for (i = 0; i < size / s->dump_info.page_size; i++) {
396         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
397                    s->dump_info.page_size, errp);
398         if (*errp) {
399             return;
400         }
401     }
402 
403     if ((size % s->dump_info.page_size) != 0) {
404         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
405                    size % s->dump_info.page_size, errp);
406         if (*errp) {
407             return;
408         }
409     }
410 }
411 
412 /* get the memory's offset and size in the vmcore */
413 static void get_offset_range(hwaddr phys_addr,
414                              ram_addr_t mapping_length,
415                              DumpState *s,
416                              hwaddr *p_offset,
417                              hwaddr *p_filesz)
418 {
419     GuestPhysBlock *block;
420     hwaddr offset = s->memory_offset;
421     int64_t size_in_block, start;
422 
423     /* When the memory is not stored into vmcore, offset will be -1 */
424     *p_offset = -1;
425     *p_filesz = 0;
426 
427     if (s->has_filter) {
428         if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
429             return;
430         }
431     }
432 
433     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
434         if (s->has_filter) {
435             if (block->target_start >= s->begin + s->length ||
436                 block->target_end <= s->begin) {
437                 /* This block is out of the range */
438                 continue;
439             }
440 
441             if (s->begin <= block->target_start) {
442                 start = block->target_start;
443             } else {
444                 start = s->begin;
445             }
446 
447             size_in_block = block->target_end - start;
448             if (s->begin + s->length < block->target_end) {
449                 size_in_block -= block->target_end - (s->begin + s->length);
450             }
451         } else {
452             start = block->target_start;
453             size_in_block = block->target_end - block->target_start;
454         }
455 
456         if (phys_addr >= start && phys_addr < start + size_in_block) {
457             *p_offset = phys_addr - start + offset;
458 
459             /* The offset range mapped from the vmcore file must not spill over
460              * the GuestPhysBlock, clamp it. The rest of the mapping will be
461              * zero-filled in memory at load time; see
462              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
463              */
464             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
465                         mapping_length :
466                         size_in_block - (phys_addr - start);
467             return;
468         }
469 
470         offset += size_in_block;
471     }
472 }
473 
474 static void write_elf_loads(DumpState *s, Error **errp)
475 {
476     ERRP_GUARD();
477     hwaddr offset, filesz;
478     MemoryMapping *memory_mapping;
479     uint32_t phdr_index = 1;
480     uint32_t max_index;
481 
482     if (s->have_section) {
483         max_index = s->sh_info;
484     } else {
485         max_index = s->phdr_num;
486     }
487 
488     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
489         get_offset_range(memory_mapping->phys_addr,
490                          memory_mapping->length,
491                          s, &offset, &filesz);
492         if (s->dump_info.d_class == ELFCLASS64) {
493             write_elf64_load(s, memory_mapping, phdr_index++, offset,
494                              filesz, errp);
495         } else {
496             write_elf32_load(s, memory_mapping, phdr_index++, offset,
497                              filesz, errp);
498         }
499 
500         if (*errp) {
501             return;
502         }
503 
504         if (phdr_index >= max_index) {
505             break;
506         }
507     }
508 }
509 
510 /* write elf header, PT_NOTE and elf note to vmcore. */
511 static void dump_begin(DumpState *s, Error **errp)
512 {
513     ERRP_GUARD();
514 
515     /*
516      * the vmcore's format is:
517      *   --------------
518      *   |  elf header |
519      *   --------------
520      *   |  PT_NOTE    |
521      *   --------------
522      *   |  PT_LOAD    |
523      *   --------------
524      *   |  ......     |
525      *   --------------
526      *   |  PT_LOAD    |
527      *   --------------
528      *   |  sec_hdr    |
529      *   --------------
530      *   |  elf note   |
531      *   --------------
532      *   |  memory     |
533      *   --------------
534      *
535      * we only know where the memory is saved after we write elf note into
536      * vmcore.
537      */
538 
539     /* write elf header to vmcore */
540     if (s->dump_info.d_class == ELFCLASS64) {
541         write_elf64_header(s, errp);
542     } else {
543         write_elf32_header(s, errp);
544     }
545     if (*errp) {
546         return;
547     }
548 
549     if (s->dump_info.d_class == ELFCLASS64) {
550         /* write PT_NOTE to vmcore */
551         write_elf64_note(s, errp);
552         if (*errp) {
553             return;
554         }
555 
556         /* write all PT_LOAD to vmcore */
557         write_elf_loads(s, errp);
558         if (*errp) {
559             return;
560         }
561 
562         /* write section to vmcore */
563         if (s->have_section) {
564             write_elf_section(s, 1, errp);
565             if (*errp) {
566                 return;
567             }
568         }
569 
570         /* write notes to vmcore */
571         write_elf64_notes(fd_write_vmcore, s, errp);
572         if (*errp) {
573             return;
574         }
575     } else {
576         /* write PT_NOTE to vmcore */
577         write_elf32_note(s, errp);
578         if (*errp) {
579             return;
580         }
581 
582         /* write all PT_LOAD to vmcore */
583         write_elf_loads(s, errp);
584         if (*errp) {
585             return;
586         }
587 
588         /* write section to vmcore */
589         if (s->have_section) {
590             write_elf_section(s, 0, errp);
591             if (*errp) {
592                 return;
593             }
594         }
595 
596         /* write notes to vmcore */
597         write_elf32_notes(fd_write_vmcore, s, errp);
598         if (*errp) {
599             return;
600         }
601     }
602 }
603 
604 static int get_next_block(DumpState *s, GuestPhysBlock *block)
605 {
606     while (1) {
607         block = QTAILQ_NEXT(block, next);
608         if (!block) {
609             /* no more block */
610             return 1;
611         }
612 
613         s->start = 0;
614         s->next_block = block;
615         if (s->has_filter) {
616             if (block->target_start >= s->begin + s->length ||
617                 block->target_end <= s->begin) {
618                 /* This block is out of the range */
619                 continue;
620             }
621 
622             if (s->begin > block->target_start) {
623                 s->start = s->begin - block->target_start;
624             }
625         }
626 
627         return 0;
628     }
629 }
630 
631 /* write all memory to vmcore */
632 static void dump_iterate(DumpState *s, Error **errp)
633 {
634     ERRP_GUARD();
635     GuestPhysBlock *block;
636     int64_t size;
637 
638     do {
639         block = s->next_block;
640 
641         size = block->target_end - block->target_start;
642         if (s->has_filter) {
643             size -= s->start;
644             if (s->begin + s->length < block->target_end) {
645                 size -= block->target_end - (s->begin + s->length);
646             }
647         }
648         write_memory(s, block, s->start, size, errp);
649         if (*errp) {
650             return;
651         }
652 
653     } while (!get_next_block(s, block));
654 }
655 
656 static void create_vmcore(DumpState *s, Error **errp)
657 {
658     ERRP_GUARD();
659 
660     dump_begin(s, errp);
661     if (*errp) {
662         return;
663     }
664 
665     dump_iterate(s, errp);
666 }
667 
668 static int write_start_flat_header(int fd)
669 {
670     MakedumpfileHeader *mh;
671     int ret = 0;
672 
673     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
674     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
675 
676     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
677            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
678 
679     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
680     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
681 
682     size_t written_size;
683     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
684     if (written_size != MAX_SIZE_MDF_HEADER) {
685         ret = -1;
686     }
687 
688     g_free(mh);
689     return ret;
690 }
691 
692 static int write_end_flat_header(int fd)
693 {
694     MakedumpfileDataHeader mdh;
695 
696     mdh.offset = END_FLAG_FLAT_HEADER;
697     mdh.buf_size = END_FLAG_FLAT_HEADER;
698 
699     size_t written_size;
700     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
701     if (written_size != sizeof(mdh)) {
702         return -1;
703     }
704 
705     return 0;
706 }
707 
708 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
709 {
710     size_t written_size;
711     MakedumpfileDataHeader mdh;
712 
713     mdh.offset = cpu_to_be64(offset);
714     mdh.buf_size = cpu_to_be64(size);
715 
716     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
717     if (written_size != sizeof(mdh)) {
718         return -1;
719     }
720 
721     written_size = qemu_write_full(fd, buf, size);
722     if (written_size != size) {
723         return -1;
724     }
725 
726     return 0;
727 }
728 
729 static int buf_write_note(const void *buf, size_t size, void *opaque)
730 {
731     DumpState *s = opaque;
732 
733     /* note_buf is not enough */
734     if (s->note_buf_offset + size > s->note_size) {
735         return -1;
736     }
737 
738     memcpy(s->note_buf + s->note_buf_offset, buf, size);
739 
740     s->note_buf_offset += size;
741 
742     return 0;
743 }
744 
745 /*
746  * This function retrieves various sizes from an elf header.
747  *
748  * @note has to be a valid ELF note. The return sizes are unmodified
749  * (not padded or rounded up to be multiple of 4).
750  */
751 static void get_note_sizes(DumpState *s, const void *note,
752                            uint64_t *note_head_size,
753                            uint64_t *name_size,
754                            uint64_t *desc_size)
755 {
756     uint64_t note_head_sz;
757     uint64_t name_sz;
758     uint64_t desc_sz;
759 
760     if (s->dump_info.d_class == ELFCLASS64) {
761         const Elf64_Nhdr *hdr = note;
762         note_head_sz = sizeof(Elf64_Nhdr);
763         name_sz = tswap64(hdr->n_namesz);
764         desc_sz = tswap64(hdr->n_descsz);
765     } else {
766         const Elf32_Nhdr *hdr = note;
767         note_head_sz = sizeof(Elf32_Nhdr);
768         name_sz = tswap32(hdr->n_namesz);
769         desc_sz = tswap32(hdr->n_descsz);
770     }
771 
772     if (note_head_size) {
773         *note_head_size = note_head_sz;
774     }
775     if (name_size) {
776         *name_size = name_sz;
777     }
778     if (desc_size) {
779         *desc_size = desc_sz;
780     }
781 }
782 
783 static bool note_name_equal(DumpState *s,
784                             const uint8_t *note, const char *name)
785 {
786     int len = strlen(name) + 1;
787     uint64_t head_size, name_size;
788 
789     get_note_sizes(s, note, &head_size, &name_size, NULL);
790     head_size = ROUND_UP(head_size, 4);
791 
792     return name_size == len && memcmp(note + head_size, name, len) == 0;
793 }
794 
795 /* write common header, sub header and elf note to vmcore */
796 static void create_header32(DumpState *s, Error **errp)
797 {
798     ERRP_GUARD();
799     DiskDumpHeader32 *dh = NULL;
800     KdumpSubHeader32 *kh = NULL;
801     size_t size;
802     uint32_t block_size;
803     uint32_t sub_hdr_size;
804     uint32_t bitmap_blocks;
805     uint32_t status = 0;
806     uint64_t offset_note;
807 
808     /* write common header, the version of kdump-compressed format is 6th */
809     size = sizeof(DiskDumpHeader32);
810     dh = g_malloc0(size);
811 
812     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
813     dh->header_version = cpu_to_dump32(s, 6);
814     block_size = s->dump_info.page_size;
815     dh->block_size = cpu_to_dump32(s, block_size);
816     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
817     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
818     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
819     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
820     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
821     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
822     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
823     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
824     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
825 
826     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
827         status |= DUMP_DH_COMPRESSED_ZLIB;
828     }
829 #ifdef CONFIG_LZO
830     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
831         status |= DUMP_DH_COMPRESSED_LZO;
832     }
833 #endif
834 #ifdef CONFIG_SNAPPY
835     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
836         status |= DUMP_DH_COMPRESSED_SNAPPY;
837     }
838 #endif
839     dh->status = cpu_to_dump32(s, status);
840 
841     if (write_buffer(s->fd, 0, dh, size) < 0) {
842         error_setg(errp, "dump: failed to write disk dump header");
843         goto out;
844     }
845 
846     /* write sub header */
847     size = sizeof(KdumpSubHeader32);
848     kh = g_malloc0(size);
849 
850     /* 64bit max_mapnr_64 */
851     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
852     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
853     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
854 
855     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
856     if (s->guest_note &&
857         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
858         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
859 
860         get_note_sizes(s, s->guest_note,
861                        &hsize, &name_size, &size_vmcoreinfo_desc);
862         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
863             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
864         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
865         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
866     }
867 
868     kh->offset_note = cpu_to_dump64(s, offset_note);
869     kh->note_size = cpu_to_dump32(s, s->note_size);
870 
871     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
872                      block_size, kh, size) < 0) {
873         error_setg(errp, "dump: failed to write kdump sub header");
874         goto out;
875     }
876 
877     /* write note */
878     s->note_buf = g_malloc0(s->note_size);
879     s->note_buf_offset = 0;
880 
881     /* use s->note_buf to store notes temporarily */
882     write_elf32_notes(buf_write_note, s, errp);
883     if (*errp) {
884         goto out;
885     }
886     if (write_buffer(s->fd, offset_note, s->note_buf,
887                      s->note_size) < 0) {
888         error_setg(errp, "dump: failed to write notes");
889         goto out;
890     }
891 
892     /* get offset of dump_bitmap */
893     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
894                              block_size;
895 
896     /* get offset of page */
897     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
898                      block_size;
899 
900 out:
901     g_free(dh);
902     g_free(kh);
903     g_free(s->note_buf);
904 }
905 
906 /* write common header, sub header and elf note to vmcore */
907 static void create_header64(DumpState *s, Error **errp)
908 {
909     ERRP_GUARD();
910     DiskDumpHeader64 *dh = NULL;
911     KdumpSubHeader64 *kh = NULL;
912     size_t size;
913     uint32_t block_size;
914     uint32_t sub_hdr_size;
915     uint32_t bitmap_blocks;
916     uint32_t status = 0;
917     uint64_t offset_note;
918 
919     /* write common header, the version of kdump-compressed format is 6th */
920     size = sizeof(DiskDumpHeader64);
921     dh = g_malloc0(size);
922 
923     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
924     dh->header_version = cpu_to_dump32(s, 6);
925     block_size = s->dump_info.page_size;
926     dh->block_size = cpu_to_dump32(s, block_size);
927     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
928     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
929     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
930     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
931     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
932     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
933     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
934     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
935     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
936 
937     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
938         status |= DUMP_DH_COMPRESSED_ZLIB;
939     }
940 #ifdef CONFIG_LZO
941     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
942         status |= DUMP_DH_COMPRESSED_LZO;
943     }
944 #endif
945 #ifdef CONFIG_SNAPPY
946     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
947         status |= DUMP_DH_COMPRESSED_SNAPPY;
948     }
949 #endif
950     dh->status = cpu_to_dump32(s, status);
951 
952     if (write_buffer(s->fd, 0, dh, size) < 0) {
953         error_setg(errp, "dump: failed to write disk dump header");
954         goto out;
955     }
956 
957     /* write sub header */
958     size = sizeof(KdumpSubHeader64);
959     kh = g_malloc0(size);
960 
961     /* 64bit max_mapnr_64 */
962     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
963     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
964     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
965 
966     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
967     if (s->guest_note &&
968         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
969         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
970 
971         get_note_sizes(s, s->guest_note,
972                        &hsize, &name_size, &size_vmcoreinfo_desc);
973         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
974             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
975         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
976         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
977     }
978 
979     kh->offset_note = cpu_to_dump64(s, offset_note);
980     kh->note_size = cpu_to_dump64(s, s->note_size);
981 
982     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
983                      block_size, kh, size) < 0) {
984         error_setg(errp, "dump: failed to write kdump sub header");
985         goto out;
986     }
987 
988     /* write note */
989     s->note_buf = g_malloc0(s->note_size);
990     s->note_buf_offset = 0;
991 
992     /* use s->note_buf to store notes temporarily */
993     write_elf64_notes(buf_write_note, s, errp);
994     if (*errp) {
995         goto out;
996     }
997 
998     if (write_buffer(s->fd, offset_note, s->note_buf,
999                      s->note_size) < 0) {
1000         error_setg(errp, "dump: failed to write notes");
1001         goto out;
1002     }
1003 
1004     /* get offset of dump_bitmap */
1005     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1006                              block_size;
1007 
1008     /* get offset of page */
1009     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1010                      block_size;
1011 
1012 out:
1013     g_free(dh);
1014     g_free(kh);
1015     g_free(s->note_buf);
1016 }
1017 
1018 static void write_dump_header(DumpState *s, Error **errp)
1019 {
1020     if (s->dump_info.d_class == ELFCLASS32) {
1021         create_header32(s, errp);
1022     } else {
1023         create_header64(s, errp);
1024     }
1025 }
1026 
1027 static size_t dump_bitmap_get_bufsize(DumpState *s)
1028 {
1029     return s->dump_info.page_size;
1030 }
1031 
1032 /*
1033  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1034  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1035  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1036  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1037  * vmcore, ie. synchronizing un-sync bit into vmcore.
1038  */
1039 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1040                            uint8_t *buf, DumpState *s)
1041 {
1042     off_t old_offset, new_offset;
1043     off_t offset_bitmap1, offset_bitmap2;
1044     uint32_t byte, bit;
1045     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1046     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1047 
1048     /* should not set the previous place */
1049     assert(last_pfn <= pfn);
1050 
1051     /*
1052      * if the bit needed to be set is not cached in buf, flush the data in buf
1053      * to vmcore firstly.
1054      * making new_offset be bigger than old_offset can also sync remained data
1055      * into vmcore.
1056      */
1057     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1058     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1059 
1060     while (old_offset < new_offset) {
1061         /* calculate the offset and write dump_bitmap */
1062         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1063         if (write_buffer(s->fd, offset_bitmap1, buf,
1064                          bitmap_bufsize) < 0) {
1065             return -1;
1066         }
1067 
1068         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1069         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1070                          old_offset;
1071         if (write_buffer(s->fd, offset_bitmap2, buf,
1072                          bitmap_bufsize) < 0) {
1073             return -1;
1074         }
1075 
1076         memset(buf, 0, bitmap_bufsize);
1077         old_offset += bitmap_bufsize;
1078     }
1079 
1080     /* get the exact place of the bit in the buf, and set it */
1081     byte = (pfn % bits_per_buf) / CHAR_BIT;
1082     bit = (pfn % bits_per_buf) % CHAR_BIT;
1083     if (value) {
1084         buf[byte] |= 1u << bit;
1085     } else {
1086         buf[byte] &= ~(1u << bit);
1087     }
1088 
1089     return 0;
1090 }
1091 
1092 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1093 {
1094     int target_page_shift = ctz32(s->dump_info.page_size);
1095 
1096     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1097 }
1098 
1099 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1100 {
1101     int target_page_shift = ctz32(s->dump_info.page_size);
1102 
1103     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1104 }
1105 
1106 /*
1107  * exam every page and return the page frame number and the address of the page.
1108  * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1109  * blocks, so block->target_start and block->target_end should be interal
1110  * multiples of the target page size.
1111  */
1112 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1113                           uint8_t **bufptr, DumpState *s)
1114 {
1115     GuestPhysBlock *block = *blockptr;
1116     hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1117     uint8_t *buf;
1118 
1119     /* block == NULL means the start of the iteration */
1120     if (!block) {
1121         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1122         *blockptr = block;
1123         assert((block->target_start & ~target_page_mask) == 0);
1124         assert((block->target_end & ~target_page_mask) == 0);
1125         *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1126         if (bufptr) {
1127             *bufptr = block->host_addr;
1128         }
1129         return true;
1130     }
1131 
1132     *pfnptr = *pfnptr + 1;
1133     addr = dump_pfn_to_paddr(s, *pfnptr);
1134 
1135     if ((addr >= block->target_start) &&
1136         (addr + s->dump_info.page_size <= block->target_end)) {
1137         buf = block->host_addr + (addr - block->target_start);
1138     } else {
1139         /* the next page is in the next block */
1140         block = QTAILQ_NEXT(block, next);
1141         *blockptr = block;
1142         if (!block) {
1143             return false;
1144         }
1145         assert((block->target_start & ~target_page_mask) == 0);
1146         assert((block->target_end & ~target_page_mask) == 0);
1147         *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1148         buf = block->host_addr;
1149     }
1150 
1151     if (bufptr) {
1152         *bufptr = buf;
1153     }
1154 
1155     return true;
1156 }
1157 
1158 static void write_dump_bitmap(DumpState *s, Error **errp)
1159 {
1160     int ret = 0;
1161     uint64_t last_pfn, pfn;
1162     void *dump_bitmap_buf;
1163     size_t num_dumpable;
1164     GuestPhysBlock *block_iter = NULL;
1165     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1166     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1167 
1168     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1169     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1170 
1171     num_dumpable = 0;
1172     last_pfn = 0;
1173 
1174     /*
1175      * exam memory page by page, and set the bit in dump_bitmap corresponded
1176      * to the existing page.
1177      */
1178     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1179         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1180         if (ret < 0) {
1181             error_setg(errp, "dump: failed to set dump_bitmap");
1182             goto out;
1183         }
1184 
1185         last_pfn = pfn;
1186         num_dumpable++;
1187     }
1188 
1189     /*
1190      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1191      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1192      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1193      */
1194     if (num_dumpable > 0) {
1195         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1196                               dump_bitmap_buf, s);
1197         if (ret < 0) {
1198             error_setg(errp, "dump: failed to sync dump_bitmap");
1199             goto out;
1200         }
1201     }
1202 
1203     /* number of dumpable pages that will be dumped later */
1204     s->num_dumpable = num_dumpable;
1205 
1206 out:
1207     g_free(dump_bitmap_buf);
1208 }
1209 
1210 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1211                                off_t offset)
1212 {
1213     data_cache->fd = s->fd;
1214     data_cache->data_size = 0;
1215     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1216     data_cache->buf = g_malloc0(data_cache->buf_size);
1217     data_cache->offset = offset;
1218 }
1219 
1220 static int write_cache(DataCache *dc, const void *buf, size_t size,
1221                        bool flag_sync)
1222 {
1223     /*
1224      * dc->buf_size should not be less than size, otherwise dc will never be
1225      * enough
1226      */
1227     assert(size <= dc->buf_size);
1228 
1229     /*
1230      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1231      * otherwise check if the space is enough for caching data in buf, if not,
1232      * write the data in dc->buf to dc->fd and reset dc->buf
1233      */
1234     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1235         (flag_sync && dc->data_size > 0)) {
1236         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1237             return -1;
1238         }
1239 
1240         dc->offset += dc->data_size;
1241         dc->data_size = 0;
1242     }
1243 
1244     if (!flag_sync) {
1245         memcpy(dc->buf + dc->data_size, buf, size);
1246         dc->data_size += size;
1247     }
1248 
1249     return 0;
1250 }
1251 
1252 static void free_data_cache(DataCache *data_cache)
1253 {
1254     g_free(data_cache->buf);
1255 }
1256 
1257 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1258 {
1259     switch (flag_compress) {
1260     case DUMP_DH_COMPRESSED_ZLIB:
1261         return compressBound(page_size);
1262 
1263     case DUMP_DH_COMPRESSED_LZO:
1264         /*
1265          * LZO will expand incompressible data by a little amount. Please check
1266          * the following URL to see the expansion calculation:
1267          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1268          */
1269         return page_size + page_size / 16 + 64 + 3;
1270 
1271 #ifdef CONFIG_SNAPPY
1272     case DUMP_DH_COMPRESSED_SNAPPY:
1273         return snappy_max_compressed_length(page_size);
1274 #endif
1275     }
1276     return 0;
1277 }
1278 
1279 static void write_dump_pages(DumpState *s, Error **errp)
1280 {
1281     int ret = 0;
1282     DataCache page_desc, page_data;
1283     size_t len_buf_out, size_out;
1284 #ifdef CONFIG_LZO
1285     lzo_bytep wrkmem = NULL;
1286 #endif
1287     uint8_t *buf_out = NULL;
1288     off_t offset_desc, offset_data;
1289     PageDescriptor pd, pd_zero;
1290     uint8_t *buf;
1291     GuestPhysBlock *block_iter = NULL;
1292     uint64_t pfn_iter;
1293 
1294     /* get offset of page_desc and page_data in dump file */
1295     offset_desc = s->offset_page;
1296     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1297 
1298     prepare_data_cache(&page_desc, s, offset_desc);
1299     prepare_data_cache(&page_data, s, offset_data);
1300 
1301     /* prepare buffer to store compressed data */
1302     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1303     assert(len_buf_out != 0);
1304 
1305 #ifdef CONFIG_LZO
1306     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1307 #endif
1308 
1309     buf_out = g_malloc(len_buf_out);
1310 
1311     /*
1312      * init zero page's page_desc and page_data, because every zero page
1313      * uses the same page_data
1314      */
1315     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1316     pd_zero.flags = cpu_to_dump32(s, 0);
1317     pd_zero.offset = cpu_to_dump64(s, offset_data);
1318     pd_zero.page_flags = cpu_to_dump64(s, 0);
1319     buf = g_malloc0(s->dump_info.page_size);
1320     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1321     g_free(buf);
1322     if (ret < 0) {
1323         error_setg(errp, "dump: failed to write page data (zero page)");
1324         goto out;
1325     }
1326 
1327     offset_data += s->dump_info.page_size;
1328 
1329     /*
1330      * dump memory to vmcore page by page. zero page will all be resided in the
1331      * first page of page section
1332      */
1333     while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1334         /* check zero page */
1335         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1336             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1337                               false);
1338             if (ret < 0) {
1339                 error_setg(errp, "dump: failed to write page desc");
1340                 goto out;
1341             }
1342         } else {
1343             /*
1344              * not zero page, then:
1345              * 1. compress the page
1346              * 2. write the compressed page into the cache of page_data
1347              * 3. get page desc of the compressed page and write it into the
1348              *    cache of page_desc
1349              *
1350              * only one compression format will be used here, for
1351              * s->flag_compress is set. But when compression fails to work,
1352              * we fall back to save in plaintext.
1353              */
1354              size_out = len_buf_out;
1355              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1356                     (compress2(buf_out, (uLongf *)&size_out, buf,
1357                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1358                     (size_out < s->dump_info.page_size)) {
1359                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1360                 pd.size  = cpu_to_dump32(s, size_out);
1361 
1362                 ret = write_cache(&page_data, buf_out, size_out, false);
1363                 if (ret < 0) {
1364                     error_setg(errp, "dump: failed to write page data");
1365                     goto out;
1366                 }
1367 #ifdef CONFIG_LZO
1368             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1369                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1370                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1371                     (size_out < s->dump_info.page_size)) {
1372                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1373                 pd.size  = cpu_to_dump32(s, size_out);
1374 
1375                 ret = write_cache(&page_data, buf_out, size_out, false);
1376                 if (ret < 0) {
1377                     error_setg(errp, "dump: failed to write page data");
1378                     goto out;
1379                 }
1380 #endif
1381 #ifdef CONFIG_SNAPPY
1382             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1383                     (snappy_compress((char *)buf, s->dump_info.page_size,
1384                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1385                     (size_out < s->dump_info.page_size)) {
1386                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1387                 pd.size  = cpu_to_dump32(s, size_out);
1388 
1389                 ret = write_cache(&page_data, buf_out, size_out, false);
1390                 if (ret < 0) {
1391                     error_setg(errp, "dump: failed to write page data");
1392                     goto out;
1393                 }
1394 #endif
1395             } else {
1396                 /*
1397                  * fall back to save in plaintext, size_out should be
1398                  * assigned the target's page size
1399                  */
1400                 pd.flags = cpu_to_dump32(s, 0);
1401                 size_out = s->dump_info.page_size;
1402                 pd.size = cpu_to_dump32(s, size_out);
1403 
1404                 ret = write_cache(&page_data, buf,
1405                                   s->dump_info.page_size, false);
1406                 if (ret < 0) {
1407                     error_setg(errp, "dump: failed to write page data");
1408                     goto out;
1409                 }
1410             }
1411 
1412             /* get and write page desc here */
1413             pd.page_flags = cpu_to_dump64(s, 0);
1414             pd.offset = cpu_to_dump64(s, offset_data);
1415             offset_data += size_out;
1416 
1417             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1418             if (ret < 0) {
1419                 error_setg(errp, "dump: failed to write page desc");
1420                 goto out;
1421             }
1422         }
1423         s->written_size += s->dump_info.page_size;
1424     }
1425 
1426     ret = write_cache(&page_desc, NULL, 0, true);
1427     if (ret < 0) {
1428         error_setg(errp, "dump: failed to sync cache for page_desc");
1429         goto out;
1430     }
1431     ret = write_cache(&page_data, NULL, 0, true);
1432     if (ret < 0) {
1433         error_setg(errp, "dump: failed to sync cache for page_data");
1434         goto out;
1435     }
1436 
1437 out:
1438     free_data_cache(&page_desc);
1439     free_data_cache(&page_data);
1440 
1441 #ifdef CONFIG_LZO
1442     g_free(wrkmem);
1443 #endif
1444 
1445     g_free(buf_out);
1446 }
1447 
1448 static void create_kdump_vmcore(DumpState *s, Error **errp)
1449 {
1450     ERRP_GUARD();
1451     int ret;
1452 
1453     /*
1454      * the kdump-compressed format is:
1455      *                                               File offset
1456      *  +------------------------------------------+ 0x0
1457      *  |    main header (struct disk_dump_header) |
1458      *  |------------------------------------------+ block 1
1459      *  |    sub header (struct kdump_sub_header)  |
1460      *  |------------------------------------------+ block 2
1461      *  |            1st-dump_bitmap               |
1462      *  |------------------------------------------+ block 2 + X blocks
1463      *  |            2nd-dump_bitmap               | (aligned by block)
1464      *  |------------------------------------------+ block 2 + 2 * X blocks
1465      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1466      *  |  page desc for pfn 1 (struct page_desc)  |
1467      *  |                    :                     |
1468      *  |------------------------------------------| (not aligned by block)
1469      *  |         page data (pfn 0)                |
1470      *  |         page data (pfn 1)                |
1471      *  |                    :                     |
1472      *  +------------------------------------------+
1473      */
1474 
1475     ret = write_start_flat_header(s->fd);
1476     if (ret < 0) {
1477         error_setg(errp, "dump: failed to write start flat header");
1478         return;
1479     }
1480 
1481     write_dump_header(s, errp);
1482     if (*errp) {
1483         return;
1484     }
1485 
1486     write_dump_bitmap(s, errp);
1487     if (*errp) {
1488         return;
1489     }
1490 
1491     write_dump_pages(s, errp);
1492     if (*errp) {
1493         return;
1494     }
1495 
1496     ret = write_end_flat_header(s->fd);
1497     if (ret < 0) {
1498         error_setg(errp, "dump: failed to write end flat header");
1499         return;
1500     }
1501 }
1502 
1503 static ram_addr_t get_start_block(DumpState *s)
1504 {
1505     GuestPhysBlock *block;
1506 
1507     if (!s->has_filter) {
1508         s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1509         return 0;
1510     }
1511 
1512     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1513         if (block->target_start >= s->begin + s->length ||
1514             block->target_end <= s->begin) {
1515             /* This block is out of the range */
1516             continue;
1517         }
1518 
1519         s->next_block = block;
1520         if (s->begin > block->target_start) {
1521             s->start = s->begin - block->target_start;
1522         } else {
1523             s->start = 0;
1524         }
1525         return s->start;
1526     }
1527 
1528     return -1;
1529 }
1530 
1531 static void get_max_mapnr(DumpState *s)
1532 {
1533     GuestPhysBlock *last_block;
1534 
1535     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1536     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1537 }
1538 
1539 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1540 
1541 static void dump_state_prepare(DumpState *s)
1542 {
1543     /* zero the struct, setting status to active */
1544     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1545 }
1546 
1547 bool qemu_system_dump_in_progress(void)
1548 {
1549     DumpState *state = &dump_state_global;
1550     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1551 }
1552 
1553 /* calculate total size of memory to be dumped (taking filter into
1554  * acoount.) */
1555 static int64_t dump_calculate_size(DumpState *s)
1556 {
1557     GuestPhysBlock *block;
1558     int64_t size = 0, total = 0, left = 0, right = 0;
1559 
1560     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1561         if (s->has_filter) {
1562             /* calculate the overlapped region. */
1563             left = MAX(s->begin, block->target_start);
1564             right = MIN(s->begin + s->length, block->target_end);
1565             size = right - left;
1566             size = size > 0 ? size : 0;
1567         } else {
1568             /* count the whole region in */
1569             size = (block->target_end - block->target_start);
1570         }
1571         total += size;
1572     }
1573 
1574     return total;
1575 }
1576 
1577 static void vmcoreinfo_update_phys_base(DumpState *s)
1578 {
1579     uint64_t size, note_head_size, name_size, phys_base;
1580     char **lines;
1581     uint8_t *vmci;
1582     size_t i;
1583 
1584     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1585         return;
1586     }
1587 
1588     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1589     note_head_size = ROUND_UP(note_head_size, 4);
1590 
1591     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1592     *(vmci + size) = '\0';
1593 
1594     lines = g_strsplit((char *)vmci, "\n", -1);
1595     for (i = 0; lines[i]; i++) {
1596         const char *prefix = NULL;
1597 
1598         if (s->dump_info.d_machine == EM_X86_64) {
1599             prefix = "NUMBER(phys_base)=";
1600         } else if (s->dump_info.d_machine == EM_AARCH64) {
1601             prefix = "NUMBER(PHYS_OFFSET)=";
1602         }
1603 
1604         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1605             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1606                               &phys_base) < 0) {
1607                 warn_report("Failed to read %s", prefix);
1608             } else {
1609                 s->dump_info.phys_base = phys_base;
1610             }
1611             break;
1612         }
1613     }
1614 
1615     g_strfreev(lines);
1616 }
1617 
1618 static void dump_init(DumpState *s, int fd, bool has_format,
1619                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1620                       int64_t begin, int64_t length, Error **errp)
1621 {
1622     ERRP_GUARD();
1623     VMCoreInfoState *vmci = vmcoreinfo_find();
1624     CPUState *cpu;
1625     int nr_cpus;
1626     int ret;
1627 
1628     s->has_format = has_format;
1629     s->format = format;
1630     s->written_size = 0;
1631 
1632     /* kdump-compressed is conflict with paging and filter */
1633     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1634         assert(!paging && !has_filter);
1635     }
1636 
1637     if (runstate_is_running()) {
1638         vm_stop(RUN_STATE_SAVE_VM);
1639         s->resume = true;
1640     } else {
1641         s->resume = false;
1642     }
1643 
1644     /* If we use KVM, we should synchronize the registers before we get dump
1645      * info or physmap info.
1646      */
1647     cpu_synchronize_all_states();
1648     nr_cpus = 0;
1649     CPU_FOREACH(cpu) {
1650         nr_cpus++;
1651     }
1652 
1653     s->fd = fd;
1654     s->has_filter = has_filter;
1655     s->begin = begin;
1656     s->length = length;
1657 
1658     memory_mapping_list_init(&s->list);
1659 
1660     guest_phys_blocks_init(&s->guest_phys_blocks);
1661     guest_phys_blocks_append(&s->guest_phys_blocks);
1662     s->total_size = dump_calculate_size(s);
1663 #ifdef DEBUG_DUMP_GUEST_MEMORY
1664     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1665 #endif
1666 
1667     /* it does not make sense to dump non-existent memory */
1668     if (!s->total_size) {
1669         error_setg(errp, "dump: no guest memory to dump");
1670         goto cleanup;
1671     }
1672 
1673     s->start = get_start_block(s);
1674     if (s->start == -1) {
1675         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1676         goto cleanup;
1677     }
1678 
1679     /* get dump info: endian, class and architecture.
1680      * If the target architecture is not supported, cpu_get_dump_info() will
1681      * return -1.
1682      */
1683     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1684     if (ret < 0) {
1685         error_setg(errp, QERR_UNSUPPORTED);
1686         goto cleanup;
1687     }
1688 
1689     if (!s->dump_info.page_size) {
1690         s->dump_info.page_size = TARGET_PAGE_SIZE;
1691     }
1692 
1693     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1694                                      s->dump_info.d_machine, nr_cpus);
1695     if (s->note_size < 0) {
1696         error_setg(errp, QERR_UNSUPPORTED);
1697         goto cleanup;
1698     }
1699 
1700     /*
1701      * The goal of this block is to (a) update the previously guessed
1702      * phys_base, (b) copy the guest note out of the guest.
1703      * Failure to do so is not fatal for dumping.
1704      */
1705     if (vmci) {
1706         uint64_t addr, note_head_size, name_size, desc_size;
1707         uint32_t size;
1708         uint16_t format;
1709 
1710         note_head_size = s->dump_info.d_class == ELFCLASS32 ?
1711             sizeof(Elf32_Nhdr) : sizeof(Elf64_Nhdr);
1712 
1713         format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1714         size = le32_to_cpu(vmci->vmcoreinfo.size);
1715         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1716         if (!vmci->has_vmcoreinfo) {
1717             warn_report("guest note is not present");
1718         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1719             warn_report("guest note size is invalid: %" PRIu32, size);
1720         } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1721             warn_report("guest note format is unsupported: %" PRIu16, format);
1722         } else {
1723             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1724             cpu_physical_memory_read(addr, s->guest_note, size);
1725 
1726             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1727             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1728                                                desc_size);
1729             if (name_size > MAX_GUEST_NOTE_SIZE ||
1730                 desc_size > MAX_GUEST_NOTE_SIZE ||
1731                 s->guest_note_size > size) {
1732                 warn_report("Invalid guest note header");
1733                 g_free(s->guest_note);
1734                 s->guest_note = NULL;
1735             } else {
1736                 vmcoreinfo_update_phys_base(s);
1737                 s->note_size += s->guest_note_size;
1738             }
1739         }
1740     }
1741 
1742     /* get memory mapping */
1743     if (paging) {
1744         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1745         if (*errp) {
1746             goto cleanup;
1747         }
1748     } else {
1749         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1750     }
1751 
1752     s->nr_cpus = nr_cpus;
1753 
1754     get_max_mapnr(s);
1755 
1756     uint64_t tmp;
1757     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1758                        s->dump_info.page_size);
1759     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1760 
1761     /* init for kdump-compressed format */
1762     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1763         switch (format) {
1764         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1765             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1766             break;
1767 
1768         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1769 #ifdef CONFIG_LZO
1770             if (lzo_init() != LZO_E_OK) {
1771                 error_setg(errp, "failed to initialize the LZO library");
1772                 goto cleanup;
1773             }
1774 #endif
1775             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1776             break;
1777 
1778         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1779             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1780             break;
1781 
1782         default:
1783             s->flag_compress = 0;
1784         }
1785 
1786         return;
1787     }
1788 
1789     if (s->has_filter) {
1790         memory_mapping_filter(&s->list, s->begin, s->length);
1791     }
1792 
1793     /*
1794      * calculate phdr_num
1795      *
1796      * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1797      */
1798     s->phdr_num = 1; /* PT_NOTE */
1799     if (s->list.num < UINT16_MAX - 2) {
1800         s->phdr_num += s->list.num;
1801         s->have_section = false;
1802     } else {
1803         s->have_section = true;
1804         s->phdr_num = PN_XNUM;
1805         s->sh_info = 1; /* PT_NOTE */
1806 
1807         /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1808         if (s->list.num <= UINT32_MAX - 1) {
1809             s->sh_info += s->list.num;
1810         } else {
1811             s->sh_info = UINT32_MAX;
1812         }
1813     }
1814 
1815     if (s->dump_info.d_class == ELFCLASS64) {
1816         if (s->have_section) {
1817             s->memory_offset = sizeof(Elf64_Ehdr) +
1818                                sizeof(Elf64_Phdr) * s->sh_info +
1819                                sizeof(Elf64_Shdr) + s->note_size;
1820         } else {
1821             s->memory_offset = sizeof(Elf64_Ehdr) +
1822                                sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1823         }
1824     } else {
1825         if (s->have_section) {
1826             s->memory_offset = sizeof(Elf32_Ehdr) +
1827                                sizeof(Elf32_Phdr) * s->sh_info +
1828                                sizeof(Elf32_Shdr) + s->note_size;
1829         } else {
1830             s->memory_offset = sizeof(Elf32_Ehdr) +
1831                                sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1832         }
1833     }
1834 
1835     return;
1836 
1837 cleanup:
1838     dump_cleanup(s);
1839 }
1840 
1841 /* this operation might be time consuming. */
1842 static void dump_process(DumpState *s, Error **errp)
1843 {
1844     ERRP_GUARD();
1845     DumpQueryResult *result = NULL;
1846 
1847     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1848 #ifdef TARGET_X86_64
1849         create_win_dump(s, errp);
1850 #endif
1851     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1852         create_kdump_vmcore(s, errp);
1853     } else {
1854         create_vmcore(s, errp);
1855     }
1856 
1857     /* make sure status is written after written_size updates */
1858     smp_wmb();
1859     qatomic_set(&s->status,
1860                (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1861 
1862     /* send DUMP_COMPLETED message (unconditionally) */
1863     result = qmp_query_dump(NULL);
1864     /* should never fail */
1865     assert(result);
1866     qapi_event_send_dump_completed(result, !!*errp, (*errp ?
1867                                                      error_get_pretty(*errp) : NULL));
1868     qapi_free_DumpQueryResult(result);
1869 
1870     dump_cleanup(s);
1871 }
1872 
1873 static void *dump_thread(void *data)
1874 {
1875     DumpState *s = (DumpState *)data;
1876     dump_process(s, NULL);
1877     return NULL;
1878 }
1879 
1880 DumpQueryResult *qmp_query_dump(Error **errp)
1881 {
1882     DumpQueryResult *result = g_new(DumpQueryResult, 1);
1883     DumpState *state = &dump_state_global;
1884     result->status = qatomic_read(&state->status);
1885     /* make sure we are reading status and written_size in order */
1886     smp_rmb();
1887     result->completed = state->written_size;
1888     result->total = state->total_size;
1889     return result;
1890 }
1891 
1892 void qmp_dump_guest_memory(bool paging, const char *file,
1893                            bool has_detach, bool detach,
1894                            bool has_begin, int64_t begin, bool has_length,
1895                            int64_t length, bool has_format,
1896                            DumpGuestMemoryFormat format, Error **errp)
1897 {
1898     ERRP_GUARD();
1899     const char *p;
1900     int fd = -1;
1901     DumpState *s;
1902     bool detach_p = false;
1903 
1904     if (runstate_check(RUN_STATE_INMIGRATE)) {
1905         error_setg(errp, "Dump not allowed during incoming migration.");
1906         return;
1907     }
1908 
1909     /* if there is a dump in background, we should wait until the dump
1910      * finished */
1911     if (qemu_system_dump_in_progress()) {
1912         error_setg(errp, "There is a dump in process, please wait.");
1913         return;
1914     }
1915 
1916     /*
1917      * kdump-compressed format need the whole memory dumped, so paging or
1918      * filter is not supported here.
1919      */
1920     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1921         (paging || has_begin || has_length)) {
1922         error_setg(errp, "kdump-compressed format doesn't support paging or "
1923                          "filter");
1924         return;
1925     }
1926     if (has_begin && !has_length) {
1927         error_setg(errp, QERR_MISSING_PARAMETER, "length");
1928         return;
1929     }
1930     if (!has_begin && has_length) {
1931         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1932         return;
1933     }
1934     if (has_detach) {
1935         detach_p = detach;
1936     }
1937 
1938     /* check whether lzo/snappy is supported */
1939 #ifndef CONFIG_LZO
1940     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1941         error_setg(errp, "kdump-lzo is not available now");
1942         return;
1943     }
1944 #endif
1945 
1946 #ifndef CONFIG_SNAPPY
1947     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1948         error_setg(errp, "kdump-snappy is not available now");
1949         return;
1950     }
1951 #endif
1952 
1953 #ifndef TARGET_X86_64
1954     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1955         error_setg(errp, "Windows dump is only available for x86-64");
1956         return;
1957     }
1958 #endif
1959 
1960 #if !defined(WIN32)
1961     if (strstart(file, "fd:", &p)) {
1962         fd = monitor_get_fd(monitor_cur(), p, errp);
1963         if (fd == -1) {
1964             return;
1965         }
1966     }
1967 #endif
1968 
1969     if  (strstart(file, "file:", &p)) {
1970         fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1971         if (fd < 0) {
1972             error_setg_file_open(errp, errno, p);
1973             return;
1974         }
1975     }
1976 
1977     if (fd == -1) {
1978         error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
1979         return;
1980     }
1981 
1982     if (!dump_migration_blocker) {
1983         error_setg(&dump_migration_blocker,
1984                    "Live migration disabled: dump-guest-memory in progress");
1985     }
1986 
1987     /*
1988      * Allows even for -only-migratable, but forbid migration during the
1989      * process of dump guest memory.
1990      */
1991     if (migrate_add_blocker_internal(dump_migration_blocker, errp)) {
1992         /* Remember to release the fd before passing it over to dump state */
1993         close(fd);
1994         return;
1995     }
1996 
1997     s = &dump_state_global;
1998     dump_state_prepare(s);
1999 
2000     dump_init(s, fd, has_format, format, paging, has_begin,
2001               begin, length, errp);
2002     if (*errp) {
2003         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2004         return;
2005     }
2006 
2007     if (detach_p) {
2008         /* detached dump */
2009         s->detached = true;
2010         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2011                            s, QEMU_THREAD_DETACHED);
2012     } else {
2013         /* sync dump */
2014         dump_process(s, errp);
2015     }
2016 }
2017 
2018 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2019 {
2020     DumpGuestMemoryCapability *cap =
2021                                   g_new0(DumpGuestMemoryCapability, 1);
2022     DumpGuestMemoryFormatList **tail = &cap->formats;
2023 
2024     /* elf is always available */
2025     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2026 
2027     /* kdump-zlib is always available */
2028     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2029 
2030     /* add new item if kdump-lzo is available */
2031 #ifdef CONFIG_LZO
2032     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2033 #endif
2034 
2035     /* add new item if kdump-snappy is available */
2036 #ifdef CONFIG_SNAPPY
2037     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2038 #endif
2039 
2040     /* Windows dump is available only if target is x86_64 */
2041 #ifdef TARGET_X86_64
2042     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2043 #endif
2044 
2045     return cap;
2046 }
2047