xref: /openbmc/qemu/dump/dump.c (revision 046bc416)
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     /*
127      * phnum in the elf header is 16 bit, if we have more segments we
128      * set phnum to PN_XNUM and write the real number of segments to a
129      * special section.
130      */
131     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
132     Elf64_Ehdr elf_header;
133     int ret;
134 
135     memset(&elf_header, 0, sizeof(Elf64_Ehdr));
136     memcpy(&elf_header, ELFMAG, SELFMAG);
137     elf_header.e_ident[EI_CLASS] = ELFCLASS64;
138     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
139     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
140     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
141     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
142     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
143     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
144     elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
145     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
146     elf_header.e_phnum = cpu_to_dump16(s, phnum);
147     if (s->have_section) {
148         uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->phdr_num;
149 
150         elf_header.e_shoff = cpu_to_dump64(s, shoff);
151         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
152         elf_header.e_shnum = cpu_to_dump16(s, 1);
153     }
154 
155     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
156     if (ret < 0) {
157         error_setg_errno(errp, -ret, "dump: failed to write elf header");
158     }
159 }
160 
161 static void write_elf32_header(DumpState *s, Error **errp)
162 {
163     /*
164      * phnum in the elf header is 16 bit, if we have more segments we
165      * set phnum to PN_XNUM and write the real number of segments to a
166      * special section.
167      */
168     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
169     Elf32_Ehdr elf_header;
170     int ret;
171 
172     memset(&elf_header, 0, sizeof(Elf32_Ehdr));
173     memcpy(&elf_header, ELFMAG, SELFMAG);
174     elf_header.e_ident[EI_CLASS] = ELFCLASS32;
175     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
176     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
177     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
178     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
179     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
180     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
181     elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
182     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
183     elf_header.e_phnum = cpu_to_dump16(s, phnum);
184     if (s->have_section) {
185         uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->phdr_num;
186 
187         elf_header.e_shoff = cpu_to_dump32(s, shoff);
188         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
189         elf_header.e_shnum = cpu_to_dump16(s, 1);
190     }
191 
192     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
193     if (ret < 0) {
194         error_setg_errno(errp, -ret, "dump: failed to write elf header");
195     }
196 }
197 
198 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
199                              int phdr_index, hwaddr offset,
200                              hwaddr filesz, Error **errp)
201 {
202     Elf64_Phdr phdr;
203     int ret;
204 
205     memset(&phdr, 0, sizeof(Elf64_Phdr));
206     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
207     phdr.p_offset = cpu_to_dump64(s, offset);
208     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
209     phdr.p_filesz = cpu_to_dump64(s, filesz);
210     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
211     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
212 
213     assert(memory_mapping->length >= filesz);
214 
215     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
216     if (ret < 0) {
217         error_setg_errno(errp, -ret,
218                          "dump: failed to write program header table");
219     }
220 }
221 
222 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
223                              int phdr_index, hwaddr offset,
224                              hwaddr filesz, Error **errp)
225 {
226     Elf32_Phdr phdr;
227     int ret;
228 
229     memset(&phdr, 0, sizeof(Elf32_Phdr));
230     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
231     phdr.p_offset = cpu_to_dump32(s, offset);
232     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
233     phdr.p_filesz = cpu_to_dump32(s, filesz);
234     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
235     phdr.p_vaddr =
236         cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
237 
238     assert(memory_mapping->length >= filesz);
239 
240     ret = fd_write_vmcore(&phdr, sizeof(Elf32_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_elf64_note(DumpState *s, Error **errp)
248 {
249     Elf64_Phdr phdr;
250     hwaddr begin = s->memory_offset - s->note_size;
251     int ret;
252 
253     memset(&phdr, 0, sizeof(Elf64_Phdr));
254     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
255     phdr.p_offset = cpu_to_dump64(s, begin);
256     phdr.p_paddr = 0;
257     phdr.p_filesz = cpu_to_dump64(s, s->note_size);
258     phdr.p_memsz = cpu_to_dump64(s, s->note_size);
259     phdr.p_vaddr = 0;
260 
261     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
262     if (ret < 0) {
263         error_setg_errno(errp, -ret,
264                          "dump: failed to write program header table");
265     }
266 }
267 
268 static inline int cpu_index(CPUState *cpu)
269 {
270     return cpu->cpu_index + 1;
271 }
272 
273 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
274                              Error **errp)
275 {
276     int ret;
277 
278     if (s->guest_note) {
279         ret = f(s->guest_note, s->guest_note_size, s);
280         if (ret < 0) {
281             error_setg(errp, "dump: failed to write guest note");
282         }
283     }
284 }
285 
286 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
287                               Error **errp)
288 {
289     CPUState *cpu;
290     int ret;
291     int id;
292 
293     CPU_FOREACH(cpu) {
294         id = cpu_index(cpu);
295         ret = cpu_write_elf64_note(f, cpu, id, s);
296         if (ret < 0) {
297             error_setg(errp, "dump: failed to write elf notes");
298             return;
299         }
300     }
301 
302     CPU_FOREACH(cpu) {
303         ret = cpu_write_elf64_qemunote(f, cpu, s);
304         if (ret < 0) {
305             error_setg(errp, "dump: failed to write CPU status");
306             return;
307         }
308     }
309 
310     write_guest_note(f, s, errp);
311 }
312 
313 static void write_elf32_note(DumpState *s, Error **errp)
314 {
315     hwaddr begin = s->memory_offset - s->note_size;
316     Elf32_Phdr phdr;
317     int ret;
318 
319     memset(&phdr, 0, sizeof(Elf32_Phdr));
320     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
321     phdr.p_offset = cpu_to_dump32(s, begin);
322     phdr.p_paddr = 0;
323     phdr.p_filesz = cpu_to_dump32(s, s->note_size);
324     phdr.p_memsz = cpu_to_dump32(s, s->note_size);
325     phdr.p_vaddr = 0;
326 
327     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
328     if (ret < 0) {
329         error_setg_errno(errp, -ret,
330                          "dump: failed to write program header table");
331     }
332 }
333 
334 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
335                               Error **errp)
336 {
337     CPUState *cpu;
338     int ret;
339     int id;
340 
341     CPU_FOREACH(cpu) {
342         id = cpu_index(cpu);
343         ret = cpu_write_elf32_note(f, cpu, id, s);
344         if (ret < 0) {
345             error_setg(errp, "dump: failed to write elf notes");
346             return;
347         }
348     }
349 
350     CPU_FOREACH(cpu) {
351         ret = cpu_write_elf32_qemunote(f, cpu, s);
352         if (ret < 0) {
353             error_setg(errp, "dump: failed to write CPU status");
354             return;
355         }
356     }
357 
358     write_guest_note(f, s, errp);
359 }
360 
361 static void write_elf_section(DumpState *s, int type, Error **errp)
362 {
363     Elf32_Shdr shdr32;
364     Elf64_Shdr shdr64;
365     int shdr_size;
366     void *shdr;
367     int ret;
368 
369     if (type == 0) {
370         shdr_size = sizeof(Elf32_Shdr);
371         memset(&shdr32, 0, shdr_size);
372         shdr32.sh_info = cpu_to_dump32(s, s->phdr_num);
373         shdr = &shdr32;
374     } else {
375         shdr_size = sizeof(Elf64_Shdr);
376         memset(&shdr64, 0, shdr_size);
377         shdr64.sh_info = cpu_to_dump32(s, s->phdr_num);
378         shdr = &shdr64;
379     }
380 
381     ret = fd_write_vmcore(shdr, shdr_size, s);
382     if (ret < 0) {
383         error_setg_errno(errp, -ret,
384                          "dump: failed to write section header table");
385     }
386 }
387 
388 static void write_data(DumpState *s, void *buf, int length, Error **errp)
389 {
390     int ret;
391 
392     ret = fd_write_vmcore(buf, length, s);
393     if (ret < 0) {
394         error_setg_errno(errp, -ret, "dump: failed to save memory");
395     } else {
396         s->written_size += length;
397     }
398 }
399 
400 /* write the memory to vmcore. 1 page per I/O. */
401 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
402                          int64_t size, Error **errp)
403 {
404     ERRP_GUARD();
405     int64_t i;
406 
407     for (i = 0; i < size / s->dump_info.page_size; i++) {
408         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
409                    s->dump_info.page_size, errp);
410         if (*errp) {
411             return;
412         }
413     }
414 
415     if ((size % s->dump_info.page_size) != 0) {
416         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
417                    size % s->dump_info.page_size, errp);
418         if (*errp) {
419             return;
420         }
421     }
422 }
423 
424 /* get the memory's offset and size in the vmcore */
425 static void get_offset_range(hwaddr phys_addr,
426                              ram_addr_t mapping_length,
427                              DumpState *s,
428                              hwaddr *p_offset,
429                              hwaddr *p_filesz)
430 {
431     GuestPhysBlock *block;
432     hwaddr offset = s->memory_offset;
433     int64_t size_in_block, start;
434 
435     /* When the memory is not stored into vmcore, offset will be -1 */
436     *p_offset = -1;
437     *p_filesz = 0;
438 
439     if (s->has_filter) {
440         if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
441             return;
442         }
443     }
444 
445     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
446         if (s->has_filter) {
447             if (block->target_start >= s->begin + s->length ||
448                 block->target_end <= s->begin) {
449                 /* This block is out of the range */
450                 continue;
451             }
452 
453             if (s->begin <= block->target_start) {
454                 start = block->target_start;
455             } else {
456                 start = s->begin;
457             }
458 
459             size_in_block = block->target_end - start;
460             if (s->begin + s->length < block->target_end) {
461                 size_in_block -= block->target_end - (s->begin + s->length);
462             }
463         } else {
464             start = block->target_start;
465             size_in_block = block->target_end - block->target_start;
466         }
467 
468         if (phys_addr >= start && phys_addr < start + size_in_block) {
469             *p_offset = phys_addr - start + offset;
470 
471             /* The offset range mapped from the vmcore file must not spill over
472              * the GuestPhysBlock, clamp it. The rest of the mapping will be
473              * zero-filled in memory at load time; see
474              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
475              */
476             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
477                         mapping_length :
478                         size_in_block - (phys_addr - start);
479             return;
480         }
481 
482         offset += size_in_block;
483     }
484 }
485 
486 static void write_elf_loads(DumpState *s, Error **errp)
487 {
488     ERRP_GUARD();
489     hwaddr offset, filesz;
490     MemoryMapping *memory_mapping;
491     uint32_t phdr_index = 1;
492 
493     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
494         get_offset_range(memory_mapping->phys_addr,
495                          memory_mapping->length,
496                          s, &offset, &filesz);
497         if (s->dump_info.d_class == ELFCLASS64) {
498             write_elf64_load(s, memory_mapping, phdr_index++, offset,
499                              filesz, errp);
500         } else {
501             write_elf32_load(s, memory_mapping, phdr_index++, offset,
502                              filesz, errp);
503         }
504 
505         if (*errp) {
506             return;
507         }
508 
509         if (phdr_index >= s->phdr_num) {
510             break;
511         }
512     }
513 }
514 
515 /* write elf header, PT_NOTE and elf note to vmcore. */
516 static void dump_begin(DumpState *s, Error **errp)
517 {
518     ERRP_GUARD();
519 
520     /*
521      * the vmcore's format is:
522      *   --------------
523      *   |  elf header |
524      *   --------------
525      *   |  PT_NOTE    |
526      *   --------------
527      *   |  PT_LOAD    |
528      *   --------------
529      *   |  ......     |
530      *   --------------
531      *   |  PT_LOAD    |
532      *   --------------
533      *   |  sec_hdr    |
534      *   --------------
535      *   |  elf note   |
536      *   --------------
537      *   |  memory     |
538      *   --------------
539      *
540      * we only know where the memory is saved after we write elf note into
541      * vmcore.
542      */
543 
544     /* write elf header to vmcore */
545     if (s->dump_info.d_class == ELFCLASS64) {
546         write_elf64_header(s, errp);
547     } else {
548         write_elf32_header(s, errp);
549     }
550     if (*errp) {
551         return;
552     }
553 
554     if (s->dump_info.d_class == ELFCLASS64) {
555         /* write PT_NOTE to vmcore */
556         write_elf64_note(s, errp);
557         if (*errp) {
558             return;
559         }
560 
561         /* write all PT_LOAD to vmcore */
562         write_elf_loads(s, errp);
563         if (*errp) {
564             return;
565         }
566 
567         /* write section to vmcore */
568         if (s->have_section) {
569             write_elf_section(s, 1, errp);
570             if (*errp) {
571                 return;
572             }
573         }
574 
575         /* write notes to vmcore */
576         write_elf64_notes(fd_write_vmcore, s, errp);
577         if (*errp) {
578             return;
579         }
580     } else {
581         /* write PT_NOTE to vmcore */
582         write_elf32_note(s, errp);
583         if (*errp) {
584             return;
585         }
586 
587         /* write all PT_LOAD to vmcore */
588         write_elf_loads(s, errp);
589         if (*errp) {
590             return;
591         }
592 
593         /* write section to vmcore */
594         if (s->have_section) {
595             write_elf_section(s, 0, errp);
596             if (*errp) {
597                 return;
598             }
599         }
600 
601         /* write notes to vmcore */
602         write_elf32_notes(fd_write_vmcore, s, errp);
603         if (*errp) {
604             return;
605         }
606     }
607 }
608 
609 static int get_next_block(DumpState *s, GuestPhysBlock *block)
610 {
611     while (1) {
612         block = QTAILQ_NEXT(block, next);
613         if (!block) {
614             /* no more block */
615             return 1;
616         }
617 
618         s->start = 0;
619         s->next_block = block;
620         if (s->has_filter) {
621             if (block->target_start >= s->begin + s->length ||
622                 block->target_end <= s->begin) {
623                 /* This block is out of the range */
624                 continue;
625             }
626 
627             if (s->begin > block->target_start) {
628                 s->start = s->begin - block->target_start;
629             }
630         }
631 
632         return 0;
633     }
634 }
635 
636 /* write all memory to vmcore */
637 static void dump_iterate(DumpState *s, Error **errp)
638 {
639     ERRP_GUARD();
640     GuestPhysBlock *block;
641     int64_t size;
642 
643     do {
644         block = s->next_block;
645 
646         size = block->target_end - block->target_start;
647         if (s->has_filter) {
648             size -= s->start;
649             if (s->begin + s->length < block->target_end) {
650                 size -= block->target_end - (s->begin + s->length);
651             }
652         }
653         write_memory(s, block, s->start, size, errp);
654         if (*errp) {
655             return;
656         }
657 
658     } while (!get_next_block(s, block));
659 }
660 
661 static void create_vmcore(DumpState *s, Error **errp)
662 {
663     ERRP_GUARD();
664 
665     dump_begin(s, errp);
666     if (*errp) {
667         return;
668     }
669 
670     dump_iterate(s, errp);
671 }
672 
673 static int write_start_flat_header(int fd)
674 {
675     MakedumpfileHeader *mh;
676     int ret = 0;
677 
678     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
679     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
680 
681     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
682            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
683 
684     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
685     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
686 
687     size_t written_size;
688     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
689     if (written_size != MAX_SIZE_MDF_HEADER) {
690         ret = -1;
691     }
692 
693     g_free(mh);
694     return ret;
695 }
696 
697 static int write_end_flat_header(int fd)
698 {
699     MakedumpfileDataHeader mdh;
700 
701     mdh.offset = END_FLAG_FLAT_HEADER;
702     mdh.buf_size = END_FLAG_FLAT_HEADER;
703 
704     size_t written_size;
705     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
706     if (written_size != sizeof(mdh)) {
707         return -1;
708     }
709 
710     return 0;
711 }
712 
713 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
714 {
715     size_t written_size;
716     MakedumpfileDataHeader mdh;
717 
718     mdh.offset = cpu_to_be64(offset);
719     mdh.buf_size = cpu_to_be64(size);
720 
721     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
722     if (written_size != sizeof(mdh)) {
723         return -1;
724     }
725 
726     written_size = qemu_write_full(fd, buf, size);
727     if (written_size != size) {
728         return -1;
729     }
730 
731     return 0;
732 }
733 
734 static int buf_write_note(const void *buf, size_t size, void *opaque)
735 {
736     DumpState *s = opaque;
737 
738     /* note_buf is not enough */
739     if (s->note_buf_offset + size > s->note_size) {
740         return -1;
741     }
742 
743     memcpy(s->note_buf + s->note_buf_offset, buf, size);
744 
745     s->note_buf_offset += size;
746 
747     return 0;
748 }
749 
750 /*
751  * This function retrieves various sizes from an elf header.
752  *
753  * @note has to be a valid ELF note. The return sizes are unmodified
754  * (not padded or rounded up to be multiple of 4).
755  */
756 static void get_note_sizes(DumpState *s, const void *note,
757                            uint64_t *note_head_size,
758                            uint64_t *name_size,
759                            uint64_t *desc_size)
760 {
761     uint64_t note_head_sz;
762     uint64_t name_sz;
763     uint64_t desc_sz;
764 
765     if (s->dump_info.d_class == ELFCLASS64) {
766         const Elf64_Nhdr *hdr = note;
767         note_head_sz = sizeof(Elf64_Nhdr);
768         name_sz = tswap64(hdr->n_namesz);
769         desc_sz = tswap64(hdr->n_descsz);
770     } else {
771         const Elf32_Nhdr *hdr = note;
772         note_head_sz = sizeof(Elf32_Nhdr);
773         name_sz = tswap32(hdr->n_namesz);
774         desc_sz = tswap32(hdr->n_descsz);
775     }
776 
777     if (note_head_size) {
778         *note_head_size = note_head_sz;
779     }
780     if (name_size) {
781         *name_size = name_sz;
782     }
783     if (desc_size) {
784         *desc_size = desc_sz;
785     }
786 }
787 
788 static bool note_name_equal(DumpState *s,
789                             const uint8_t *note, const char *name)
790 {
791     int len = strlen(name) + 1;
792     uint64_t head_size, name_size;
793 
794     get_note_sizes(s, note, &head_size, &name_size, NULL);
795     head_size = ROUND_UP(head_size, 4);
796 
797     return name_size == len && memcmp(note + head_size, name, len) == 0;
798 }
799 
800 /* write common header, sub header and elf note to vmcore */
801 static void create_header32(DumpState *s, Error **errp)
802 {
803     ERRP_GUARD();
804     DiskDumpHeader32 *dh = NULL;
805     KdumpSubHeader32 *kh = NULL;
806     size_t size;
807     uint32_t block_size;
808     uint32_t sub_hdr_size;
809     uint32_t bitmap_blocks;
810     uint32_t status = 0;
811     uint64_t offset_note;
812 
813     /* write common header, the version of kdump-compressed format is 6th */
814     size = sizeof(DiskDumpHeader32);
815     dh = g_malloc0(size);
816 
817     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
818     dh->header_version = cpu_to_dump32(s, 6);
819     block_size = s->dump_info.page_size;
820     dh->block_size = cpu_to_dump32(s, block_size);
821     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
822     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
823     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
824     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
825     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
826     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
827     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
828     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
829     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
830 
831     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
832         status |= DUMP_DH_COMPRESSED_ZLIB;
833     }
834 #ifdef CONFIG_LZO
835     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
836         status |= DUMP_DH_COMPRESSED_LZO;
837     }
838 #endif
839 #ifdef CONFIG_SNAPPY
840     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
841         status |= DUMP_DH_COMPRESSED_SNAPPY;
842     }
843 #endif
844     dh->status = cpu_to_dump32(s, status);
845 
846     if (write_buffer(s->fd, 0, dh, size) < 0) {
847         error_setg(errp, "dump: failed to write disk dump header");
848         goto out;
849     }
850 
851     /* write sub header */
852     size = sizeof(KdumpSubHeader32);
853     kh = g_malloc0(size);
854 
855     /* 64bit max_mapnr_64 */
856     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
857     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
858     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
859 
860     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
861     if (s->guest_note &&
862         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
863         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
864 
865         get_note_sizes(s, s->guest_note,
866                        &hsize, &name_size, &size_vmcoreinfo_desc);
867         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
868             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
869         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
870         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
871     }
872 
873     kh->offset_note = cpu_to_dump64(s, offset_note);
874     kh->note_size = cpu_to_dump32(s, s->note_size);
875 
876     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
877                      block_size, kh, size) < 0) {
878         error_setg(errp, "dump: failed to write kdump sub header");
879         goto out;
880     }
881 
882     /* write note */
883     s->note_buf = g_malloc0(s->note_size);
884     s->note_buf_offset = 0;
885 
886     /* use s->note_buf to store notes temporarily */
887     write_elf32_notes(buf_write_note, s, errp);
888     if (*errp) {
889         goto out;
890     }
891     if (write_buffer(s->fd, offset_note, s->note_buf,
892                      s->note_size) < 0) {
893         error_setg(errp, "dump: failed to write notes");
894         goto out;
895     }
896 
897     /* get offset of dump_bitmap */
898     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
899                              block_size;
900 
901     /* get offset of page */
902     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
903                      block_size;
904 
905 out:
906     g_free(dh);
907     g_free(kh);
908     g_free(s->note_buf);
909 }
910 
911 /* write common header, sub header and elf note to vmcore */
912 static void create_header64(DumpState *s, Error **errp)
913 {
914     ERRP_GUARD();
915     DiskDumpHeader64 *dh = NULL;
916     KdumpSubHeader64 *kh = NULL;
917     size_t size;
918     uint32_t block_size;
919     uint32_t sub_hdr_size;
920     uint32_t bitmap_blocks;
921     uint32_t status = 0;
922     uint64_t offset_note;
923 
924     /* write common header, the version of kdump-compressed format is 6th */
925     size = sizeof(DiskDumpHeader64);
926     dh = g_malloc0(size);
927 
928     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
929     dh->header_version = cpu_to_dump32(s, 6);
930     block_size = s->dump_info.page_size;
931     dh->block_size = cpu_to_dump32(s, block_size);
932     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
933     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
934     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
935     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
936     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
937     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
938     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
939     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
940     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
941 
942     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
943         status |= DUMP_DH_COMPRESSED_ZLIB;
944     }
945 #ifdef CONFIG_LZO
946     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
947         status |= DUMP_DH_COMPRESSED_LZO;
948     }
949 #endif
950 #ifdef CONFIG_SNAPPY
951     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
952         status |= DUMP_DH_COMPRESSED_SNAPPY;
953     }
954 #endif
955     dh->status = cpu_to_dump32(s, status);
956 
957     if (write_buffer(s->fd, 0, dh, size) < 0) {
958         error_setg(errp, "dump: failed to write disk dump header");
959         goto out;
960     }
961 
962     /* write sub header */
963     size = sizeof(KdumpSubHeader64);
964     kh = g_malloc0(size);
965 
966     /* 64bit max_mapnr_64 */
967     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
968     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
969     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
970 
971     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
972     if (s->guest_note &&
973         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
974         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
975 
976         get_note_sizes(s, s->guest_note,
977                        &hsize, &name_size, &size_vmcoreinfo_desc);
978         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
979             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
980         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
981         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
982     }
983 
984     kh->offset_note = cpu_to_dump64(s, offset_note);
985     kh->note_size = cpu_to_dump64(s, s->note_size);
986 
987     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
988                      block_size, kh, size) < 0) {
989         error_setg(errp, "dump: failed to write kdump sub header");
990         goto out;
991     }
992 
993     /* write note */
994     s->note_buf = g_malloc0(s->note_size);
995     s->note_buf_offset = 0;
996 
997     /* use s->note_buf to store notes temporarily */
998     write_elf64_notes(buf_write_note, s, errp);
999     if (*errp) {
1000         goto out;
1001     }
1002 
1003     if (write_buffer(s->fd, offset_note, s->note_buf,
1004                      s->note_size) < 0) {
1005         error_setg(errp, "dump: failed to write notes");
1006         goto out;
1007     }
1008 
1009     /* get offset of dump_bitmap */
1010     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1011                              block_size;
1012 
1013     /* get offset of page */
1014     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1015                      block_size;
1016 
1017 out:
1018     g_free(dh);
1019     g_free(kh);
1020     g_free(s->note_buf);
1021 }
1022 
1023 static void write_dump_header(DumpState *s, Error **errp)
1024 {
1025     if (s->dump_info.d_class == ELFCLASS32) {
1026         create_header32(s, errp);
1027     } else {
1028         create_header64(s, errp);
1029     }
1030 }
1031 
1032 static size_t dump_bitmap_get_bufsize(DumpState *s)
1033 {
1034     return s->dump_info.page_size;
1035 }
1036 
1037 /*
1038  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1039  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1040  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1041  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1042  * vmcore, ie. synchronizing un-sync bit into vmcore.
1043  */
1044 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1045                            uint8_t *buf, DumpState *s)
1046 {
1047     off_t old_offset, new_offset;
1048     off_t offset_bitmap1, offset_bitmap2;
1049     uint32_t byte, bit;
1050     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1051     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1052 
1053     /* should not set the previous place */
1054     assert(last_pfn <= pfn);
1055 
1056     /*
1057      * if the bit needed to be set is not cached in buf, flush the data in buf
1058      * to vmcore firstly.
1059      * making new_offset be bigger than old_offset can also sync remained data
1060      * into vmcore.
1061      */
1062     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1063     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1064 
1065     while (old_offset < new_offset) {
1066         /* calculate the offset and write dump_bitmap */
1067         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1068         if (write_buffer(s->fd, offset_bitmap1, buf,
1069                          bitmap_bufsize) < 0) {
1070             return -1;
1071         }
1072 
1073         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1074         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1075                          old_offset;
1076         if (write_buffer(s->fd, offset_bitmap2, buf,
1077                          bitmap_bufsize) < 0) {
1078             return -1;
1079         }
1080 
1081         memset(buf, 0, bitmap_bufsize);
1082         old_offset += bitmap_bufsize;
1083     }
1084 
1085     /* get the exact place of the bit in the buf, and set it */
1086     byte = (pfn % bits_per_buf) / CHAR_BIT;
1087     bit = (pfn % bits_per_buf) % CHAR_BIT;
1088     if (value) {
1089         buf[byte] |= 1u << bit;
1090     } else {
1091         buf[byte] &= ~(1u << bit);
1092     }
1093 
1094     return 0;
1095 }
1096 
1097 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1098 {
1099     int target_page_shift = ctz32(s->dump_info.page_size);
1100 
1101     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1102 }
1103 
1104 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1105 {
1106     int target_page_shift = ctz32(s->dump_info.page_size);
1107 
1108     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1109 }
1110 
1111 /*
1112  * exam every page and return the page frame number and the address of the page.
1113  * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1114  * blocks, so block->target_start and block->target_end should be interal
1115  * multiples of the target page size.
1116  */
1117 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1118                           uint8_t **bufptr, DumpState *s)
1119 {
1120     GuestPhysBlock *block = *blockptr;
1121     hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1122     uint8_t *buf;
1123 
1124     /* block == NULL means the start of the iteration */
1125     if (!block) {
1126         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1127         *blockptr = block;
1128         assert((block->target_start & ~target_page_mask) == 0);
1129         assert((block->target_end & ~target_page_mask) == 0);
1130         *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1131         if (bufptr) {
1132             *bufptr = block->host_addr;
1133         }
1134         return true;
1135     }
1136 
1137     *pfnptr = *pfnptr + 1;
1138     addr = dump_pfn_to_paddr(s, *pfnptr);
1139 
1140     if ((addr >= block->target_start) &&
1141         (addr + s->dump_info.page_size <= block->target_end)) {
1142         buf = block->host_addr + (addr - block->target_start);
1143     } else {
1144         /* the next page is in the next block */
1145         block = QTAILQ_NEXT(block, next);
1146         *blockptr = block;
1147         if (!block) {
1148             return false;
1149         }
1150         assert((block->target_start & ~target_page_mask) == 0);
1151         assert((block->target_end & ~target_page_mask) == 0);
1152         *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1153         buf = block->host_addr;
1154     }
1155 
1156     if (bufptr) {
1157         *bufptr = buf;
1158     }
1159 
1160     return true;
1161 }
1162 
1163 static void write_dump_bitmap(DumpState *s, Error **errp)
1164 {
1165     int ret = 0;
1166     uint64_t last_pfn, pfn;
1167     void *dump_bitmap_buf;
1168     size_t num_dumpable;
1169     GuestPhysBlock *block_iter = NULL;
1170     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1171     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1172 
1173     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1174     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1175 
1176     num_dumpable = 0;
1177     last_pfn = 0;
1178 
1179     /*
1180      * exam memory page by page, and set the bit in dump_bitmap corresponded
1181      * to the existing page.
1182      */
1183     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1184         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1185         if (ret < 0) {
1186             error_setg(errp, "dump: failed to set dump_bitmap");
1187             goto out;
1188         }
1189 
1190         last_pfn = pfn;
1191         num_dumpable++;
1192     }
1193 
1194     /*
1195      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1196      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1197      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1198      */
1199     if (num_dumpable > 0) {
1200         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1201                               dump_bitmap_buf, s);
1202         if (ret < 0) {
1203             error_setg(errp, "dump: failed to sync dump_bitmap");
1204             goto out;
1205         }
1206     }
1207 
1208     /* number of dumpable pages that will be dumped later */
1209     s->num_dumpable = num_dumpable;
1210 
1211 out:
1212     g_free(dump_bitmap_buf);
1213 }
1214 
1215 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1216                                off_t offset)
1217 {
1218     data_cache->fd = s->fd;
1219     data_cache->data_size = 0;
1220     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1221     data_cache->buf = g_malloc0(data_cache->buf_size);
1222     data_cache->offset = offset;
1223 }
1224 
1225 static int write_cache(DataCache *dc, const void *buf, size_t size,
1226                        bool flag_sync)
1227 {
1228     /*
1229      * dc->buf_size should not be less than size, otherwise dc will never be
1230      * enough
1231      */
1232     assert(size <= dc->buf_size);
1233 
1234     /*
1235      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1236      * otherwise check if the space is enough for caching data in buf, if not,
1237      * write the data in dc->buf to dc->fd and reset dc->buf
1238      */
1239     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1240         (flag_sync && dc->data_size > 0)) {
1241         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1242             return -1;
1243         }
1244 
1245         dc->offset += dc->data_size;
1246         dc->data_size = 0;
1247     }
1248 
1249     if (!flag_sync) {
1250         memcpy(dc->buf + dc->data_size, buf, size);
1251         dc->data_size += size;
1252     }
1253 
1254     return 0;
1255 }
1256 
1257 static void free_data_cache(DataCache *data_cache)
1258 {
1259     g_free(data_cache->buf);
1260 }
1261 
1262 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1263 {
1264     switch (flag_compress) {
1265     case DUMP_DH_COMPRESSED_ZLIB:
1266         return compressBound(page_size);
1267 
1268     case DUMP_DH_COMPRESSED_LZO:
1269         /*
1270          * LZO will expand incompressible data by a little amount. Please check
1271          * the following URL to see the expansion calculation:
1272          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1273          */
1274         return page_size + page_size / 16 + 64 + 3;
1275 
1276 #ifdef CONFIG_SNAPPY
1277     case DUMP_DH_COMPRESSED_SNAPPY:
1278         return snappy_max_compressed_length(page_size);
1279 #endif
1280     }
1281     return 0;
1282 }
1283 
1284 static void write_dump_pages(DumpState *s, Error **errp)
1285 {
1286     int ret = 0;
1287     DataCache page_desc, page_data;
1288     size_t len_buf_out, size_out;
1289 #ifdef CONFIG_LZO
1290     lzo_bytep wrkmem = NULL;
1291 #endif
1292     uint8_t *buf_out = NULL;
1293     off_t offset_desc, offset_data;
1294     PageDescriptor pd, pd_zero;
1295     uint8_t *buf;
1296     GuestPhysBlock *block_iter = NULL;
1297     uint64_t pfn_iter;
1298 
1299     /* get offset of page_desc and page_data in dump file */
1300     offset_desc = s->offset_page;
1301     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1302 
1303     prepare_data_cache(&page_desc, s, offset_desc);
1304     prepare_data_cache(&page_data, s, offset_data);
1305 
1306     /* prepare buffer to store compressed data */
1307     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1308     assert(len_buf_out != 0);
1309 
1310 #ifdef CONFIG_LZO
1311     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1312 #endif
1313 
1314     buf_out = g_malloc(len_buf_out);
1315 
1316     /*
1317      * init zero page's page_desc and page_data, because every zero page
1318      * uses the same page_data
1319      */
1320     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1321     pd_zero.flags = cpu_to_dump32(s, 0);
1322     pd_zero.offset = cpu_to_dump64(s, offset_data);
1323     pd_zero.page_flags = cpu_to_dump64(s, 0);
1324     buf = g_malloc0(s->dump_info.page_size);
1325     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1326     g_free(buf);
1327     if (ret < 0) {
1328         error_setg(errp, "dump: failed to write page data (zero page)");
1329         goto out;
1330     }
1331 
1332     offset_data += s->dump_info.page_size;
1333 
1334     /*
1335      * dump memory to vmcore page by page. zero page will all be resided in the
1336      * first page of page section
1337      */
1338     while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1339         /* check zero page */
1340         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1341             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1342                               false);
1343             if (ret < 0) {
1344                 error_setg(errp, "dump: failed to write page desc");
1345                 goto out;
1346             }
1347         } else {
1348             /*
1349              * not zero page, then:
1350              * 1. compress the page
1351              * 2. write the compressed page into the cache of page_data
1352              * 3. get page desc of the compressed page and write it into the
1353              *    cache of page_desc
1354              *
1355              * only one compression format will be used here, for
1356              * s->flag_compress is set. But when compression fails to work,
1357              * we fall back to save in plaintext.
1358              */
1359              size_out = len_buf_out;
1360              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1361                     (compress2(buf_out, (uLongf *)&size_out, buf,
1362                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1363                     (size_out < s->dump_info.page_size)) {
1364                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1365                 pd.size  = cpu_to_dump32(s, size_out);
1366 
1367                 ret = write_cache(&page_data, buf_out, size_out, false);
1368                 if (ret < 0) {
1369                     error_setg(errp, "dump: failed to write page data");
1370                     goto out;
1371                 }
1372 #ifdef CONFIG_LZO
1373             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1374                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1375                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1376                     (size_out < s->dump_info.page_size)) {
1377                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1378                 pd.size  = cpu_to_dump32(s, size_out);
1379 
1380                 ret = write_cache(&page_data, buf_out, size_out, false);
1381                 if (ret < 0) {
1382                     error_setg(errp, "dump: failed to write page data");
1383                     goto out;
1384                 }
1385 #endif
1386 #ifdef CONFIG_SNAPPY
1387             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1388                     (snappy_compress((char *)buf, s->dump_info.page_size,
1389                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1390                     (size_out < s->dump_info.page_size)) {
1391                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1392                 pd.size  = cpu_to_dump32(s, size_out);
1393 
1394                 ret = write_cache(&page_data, buf_out, size_out, false);
1395                 if (ret < 0) {
1396                     error_setg(errp, "dump: failed to write page data");
1397                     goto out;
1398                 }
1399 #endif
1400             } else {
1401                 /*
1402                  * fall back to save in plaintext, size_out should be
1403                  * assigned the target's page size
1404                  */
1405                 pd.flags = cpu_to_dump32(s, 0);
1406                 size_out = s->dump_info.page_size;
1407                 pd.size = cpu_to_dump32(s, size_out);
1408 
1409                 ret = write_cache(&page_data, buf,
1410                                   s->dump_info.page_size, false);
1411                 if (ret < 0) {
1412                     error_setg(errp, "dump: failed to write page data");
1413                     goto out;
1414                 }
1415             }
1416 
1417             /* get and write page desc here */
1418             pd.page_flags = cpu_to_dump64(s, 0);
1419             pd.offset = cpu_to_dump64(s, offset_data);
1420             offset_data += size_out;
1421 
1422             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1423             if (ret < 0) {
1424                 error_setg(errp, "dump: failed to write page desc");
1425                 goto out;
1426             }
1427         }
1428         s->written_size += s->dump_info.page_size;
1429     }
1430 
1431     ret = write_cache(&page_desc, NULL, 0, true);
1432     if (ret < 0) {
1433         error_setg(errp, "dump: failed to sync cache for page_desc");
1434         goto out;
1435     }
1436     ret = write_cache(&page_data, NULL, 0, true);
1437     if (ret < 0) {
1438         error_setg(errp, "dump: failed to sync cache for page_data");
1439         goto out;
1440     }
1441 
1442 out:
1443     free_data_cache(&page_desc);
1444     free_data_cache(&page_data);
1445 
1446 #ifdef CONFIG_LZO
1447     g_free(wrkmem);
1448 #endif
1449 
1450     g_free(buf_out);
1451 }
1452 
1453 static void create_kdump_vmcore(DumpState *s, Error **errp)
1454 {
1455     ERRP_GUARD();
1456     int ret;
1457 
1458     /*
1459      * the kdump-compressed format is:
1460      *                                               File offset
1461      *  +------------------------------------------+ 0x0
1462      *  |    main header (struct disk_dump_header) |
1463      *  |------------------------------------------+ block 1
1464      *  |    sub header (struct kdump_sub_header)  |
1465      *  |------------------------------------------+ block 2
1466      *  |            1st-dump_bitmap               |
1467      *  |------------------------------------------+ block 2 + X blocks
1468      *  |            2nd-dump_bitmap               | (aligned by block)
1469      *  |------------------------------------------+ block 2 + 2 * X blocks
1470      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1471      *  |  page desc for pfn 1 (struct page_desc)  |
1472      *  |                    :                     |
1473      *  |------------------------------------------| (not aligned by block)
1474      *  |         page data (pfn 0)                |
1475      *  |         page data (pfn 1)                |
1476      *  |                    :                     |
1477      *  +------------------------------------------+
1478      */
1479 
1480     ret = write_start_flat_header(s->fd);
1481     if (ret < 0) {
1482         error_setg(errp, "dump: failed to write start flat header");
1483         return;
1484     }
1485 
1486     write_dump_header(s, errp);
1487     if (*errp) {
1488         return;
1489     }
1490 
1491     write_dump_bitmap(s, errp);
1492     if (*errp) {
1493         return;
1494     }
1495 
1496     write_dump_pages(s, errp);
1497     if (*errp) {
1498         return;
1499     }
1500 
1501     ret = write_end_flat_header(s->fd);
1502     if (ret < 0) {
1503         error_setg(errp, "dump: failed to write end flat header");
1504         return;
1505     }
1506 }
1507 
1508 static ram_addr_t get_start_block(DumpState *s)
1509 {
1510     GuestPhysBlock *block;
1511 
1512     if (!s->has_filter) {
1513         s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1514         return 0;
1515     }
1516 
1517     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1518         if (block->target_start >= s->begin + s->length ||
1519             block->target_end <= s->begin) {
1520             /* This block is out of the range */
1521             continue;
1522         }
1523 
1524         s->next_block = block;
1525         if (s->begin > block->target_start) {
1526             s->start = s->begin - block->target_start;
1527         } else {
1528             s->start = 0;
1529         }
1530         return s->start;
1531     }
1532 
1533     return -1;
1534 }
1535 
1536 static void get_max_mapnr(DumpState *s)
1537 {
1538     GuestPhysBlock *last_block;
1539 
1540     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1541     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1542 }
1543 
1544 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1545 
1546 static void dump_state_prepare(DumpState *s)
1547 {
1548     /* zero the struct, setting status to active */
1549     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1550 }
1551 
1552 bool qemu_system_dump_in_progress(void)
1553 {
1554     DumpState *state = &dump_state_global;
1555     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1556 }
1557 
1558 /* calculate total size of memory to be dumped (taking filter into
1559  * acoount.) */
1560 static int64_t dump_calculate_size(DumpState *s)
1561 {
1562     GuestPhysBlock *block;
1563     int64_t size = 0, total = 0, left = 0, right = 0;
1564 
1565     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1566         if (s->has_filter) {
1567             /* calculate the overlapped region. */
1568             left = MAX(s->begin, block->target_start);
1569             right = MIN(s->begin + s->length, block->target_end);
1570             size = right - left;
1571             size = size > 0 ? size : 0;
1572         } else {
1573             /* count the whole region in */
1574             size = (block->target_end - block->target_start);
1575         }
1576         total += size;
1577     }
1578 
1579     return total;
1580 }
1581 
1582 static void vmcoreinfo_update_phys_base(DumpState *s)
1583 {
1584     uint64_t size, note_head_size, name_size, phys_base;
1585     char **lines;
1586     uint8_t *vmci;
1587     size_t i;
1588 
1589     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1590         return;
1591     }
1592 
1593     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1594     note_head_size = ROUND_UP(note_head_size, 4);
1595 
1596     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1597     *(vmci + size) = '\0';
1598 
1599     lines = g_strsplit((char *)vmci, "\n", -1);
1600     for (i = 0; lines[i]; i++) {
1601         const char *prefix = NULL;
1602 
1603         if (s->dump_info.d_machine == EM_X86_64) {
1604             prefix = "NUMBER(phys_base)=";
1605         } else if (s->dump_info.d_machine == EM_AARCH64) {
1606             prefix = "NUMBER(PHYS_OFFSET)=";
1607         }
1608 
1609         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1610             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1611                               &phys_base) < 0) {
1612                 warn_report("Failed to read %s", prefix);
1613             } else {
1614                 s->dump_info.phys_base = phys_base;
1615             }
1616             break;
1617         }
1618     }
1619 
1620     g_strfreev(lines);
1621 }
1622 
1623 static void dump_init(DumpState *s, int fd, bool has_format,
1624                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1625                       int64_t begin, int64_t length, Error **errp)
1626 {
1627     ERRP_GUARD();
1628     VMCoreInfoState *vmci = vmcoreinfo_find();
1629     CPUState *cpu;
1630     int nr_cpus;
1631     int ret;
1632 
1633     s->has_format = has_format;
1634     s->format = format;
1635     s->written_size = 0;
1636 
1637     /* kdump-compressed is conflict with paging and filter */
1638     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1639         assert(!paging && !has_filter);
1640     }
1641 
1642     if (runstate_is_running()) {
1643         vm_stop(RUN_STATE_SAVE_VM);
1644         s->resume = true;
1645     } else {
1646         s->resume = false;
1647     }
1648 
1649     /* If we use KVM, we should synchronize the registers before we get dump
1650      * info or physmap info.
1651      */
1652     cpu_synchronize_all_states();
1653     nr_cpus = 0;
1654     CPU_FOREACH(cpu) {
1655         nr_cpus++;
1656     }
1657 
1658     s->fd = fd;
1659     s->has_filter = has_filter;
1660     s->begin = begin;
1661     s->length = length;
1662 
1663     memory_mapping_list_init(&s->list);
1664 
1665     guest_phys_blocks_init(&s->guest_phys_blocks);
1666     guest_phys_blocks_append(&s->guest_phys_blocks);
1667     s->total_size = dump_calculate_size(s);
1668 #ifdef DEBUG_DUMP_GUEST_MEMORY
1669     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1670 #endif
1671 
1672     /* it does not make sense to dump non-existent memory */
1673     if (!s->total_size) {
1674         error_setg(errp, "dump: no guest memory to dump");
1675         goto cleanup;
1676     }
1677 
1678     s->start = get_start_block(s);
1679     if (s->start == -1) {
1680         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1681         goto cleanup;
1682     }
1683 
1684     /* get dump info: endian, class and architecture.
1685      * If the target architecture is not supported, cpu_get_dump_info() will
1686      * return -1.
1687      */
1688     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1689     if (ret < 0) {
1690         error_setg(errp, QERR_UNSUPPORTED);
1691         goto cleanup;
1692     }
1693 
1694     if (!s->dump_info.page_size) {
1695         s->dump_info.page_size = TARGET_PAGE_SIZE;
1696     }
1697 
1698     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1699                                      s->dump_info.d_machine, nr_cpus);
1700     if (s->note_size < 0) {
1701         error_setg(errp, QERR_UNSUPPORTED);
1702         goto cleanup;
1703     }
1704 
1705     /*
1706      * The goal of this block is to (a) update the previously guessed
1707      * phys_base, (b) copy the guest note out of the guest.
1708      * Failure to do so is not fatal for dumping.
1709      */
1710     if (vmci) {
1711         uint64_t addr, note_head_size, name_size, desc_size;
1712         uint32_t size;
1713         uint16_t format;
1714 
1715         note_head_size = s->dump_info.d_class == ELFCLASS32 ?
1716             sizeof(Elf32_Nhdr) : sizeof(Elf64_Nhdr);
1717 
1718         format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1719         size = le32_to_cpu(vmci->vmcoreinfo.size);
1720         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1721         if (!vmci->has_vmcoreinfo) {
1722             warn_report("guest note is not present");
1723         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1724             warn_report("guest note size is invalid: %" PRIu32, size);
1725         } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1726             warn_report("guest note format is unsupported: %" PRIu16, format);
1727         } else {
1728             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1729             cpu_physical_memory_read(addr, s->guest_note, size);
1730 
1731             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1732             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1733                                                desc_size);
1734             if (name_size > MAX_GUEST_NOTE_SIZE ||
1735                 desc_size > MAX_GUEST_NOTE_SIZE ||
1736                 s->guest_note_size > size) {
1737                 warn_report("Invalid guest note header");
1738                 g_free(s->guest_note);
1739                 s->guest_note = NULL;
1740             } else {
1741                 vmcoreinfo_update_phys_base(s);
1742                 s->note_size += s->guest_note_size;
1743             }
1744         }
1745     }
1746 
1747     /* get memory mapping */
1748     if (paging) {
1749         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1750         if (*errp) {
1751             goto cleanup;
1752         }
1753     } else {
1754         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1755     }
1756 
1757     s->nr_cpus = nr_cpus;
1758 
1759     get_max_mapnr(s);
1760 
1761     uint64_t tmp;
1762     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1763                        s->dump_info.page_size);
1764     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1765 
1766     /* init for kdump-compressed format */
1767     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1768         switch (format) {
1769         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1770             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1771             break;
1772 
1773         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1774 #ifdef CONFIG_LZO
1775             if (lzo_init() != LZO_E_OK) {
1776                 error_setg(errp, "failed to initialize the LZO library");
1777                 goto cleanup;
1778             }
1779 #endif
1780             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1781             break;
1782 
1783         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1784             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1785             break;
1786 
1787         default:
1788             s->flag_compress = 0;
1789         }
1790 
1791         return;
1792     }
1793 
1794     if (s->has_filter) {
1795         memory_mapping_filter(&s->list, s->begin, s->length);
1796     }
1797 
1798     /*
1799      * calculate phdr_num
1800      *
1801      * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1802      */
1803     s->phdr_num = 1; /* PT_NOTE */
1804     if (s->list.num < UINT16_MAX - 2) {
1805         s->phdr_num += s->list.num;
1806         s->have_section = false;
1807     } else {
1808         /* sh_info of section 0 holds the real number of phdrs */
1809         s->have_section = true;
1810 
1811         /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1812         if (s->list.num <= UINT32_MAX - 1) {
1813             s->phdr_num += s->list.num;
1814         } else {
1815             s->phdr_num = UINT32_MAX;
1816         }
1817     }
1818 
1819     if (s->dump_info.d_class == ELFCLASS64) {
1820         if (s->have_section) {
1821             s->memory_offset = sizeof(Elf64_Ehdr) +
1822                                sizeof(Elf64_Phdr) * s->phdr_num +
1823                                sizeof(Elf64_Shdr) + s->note_size;
1824         } else {
1825             s->memory_offset = sizeof(Elf64_Ehdr) +
1826                                sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1827         }
1828     } else {
1829         if (s->have_section) {
1830             s->memory_offset = sizeof(Elf32_Ehdr) +
1831                                sizeof(Elf32_Phdr) * s->phdr_num +
1832                                sizeof(Elf32_Shdr) + s->note_size;
1833         } else {
1834             s->memory_offset = sizeof(Elf32_Ehdr) +
1835                                sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1836         }
1837     }
1838 
1839     return;
1840 
1841 cleanup:
1842     dump_cleanup(s);
1843 }
1844 
1845 /* this operation might be time consuming. */
1846 static void dump_process(DumpState *s, Error **errp)
1847 {
1848     ERRP_GUARD();
1849     DumpQueryResult *result = NULL;
1850 
1851     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1852 #ifdef TARGET_X86_64
1853         create_win_dump(s, errp);
1854 #endif
1855     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1856         create_kdump_vmcore(s, errp);
1857     } else {
1858         create_vmcore(s, errp);
1859     }
1860 
1861     /* make sure status is written after written_size updates */
1862     smp_wmb();
1863     qatomic_set(&s->status,
1864                (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1865 
1866     /* send DUMP_COMPLETED message (unconditionally) */
1867     result = qmp_query_dump(NULL);
1868     /* should never fail */
1869     assert(result);
1870     qapi_event_send_dump_completed(result, !!*errp, (*errp ?
1871                                                      error_get_pretty(*errp) : NULL));
1872     qapi_free_DumpQueryResult(result);
1873 
1874     dump_cleanup(s);
1875 }
1876 
1877 static void *dump_thread(void *data)
1878 {
1879     DumpState *s = (DumpState *)data;
1880     dump_process(s, NULL);
1881     return NULL;
1882 }
1883 
1884 DumpQueryResult *qmp_query_dump(Error **errp)
1885 {
1886     DumpQueryResult *result = g_new(DumpQueryResult, 1);
1887     DumpState *state = &dump_state_global;
1888     result->status = qatomic_read(&state->status);
1889     /* make sure we are reading status and written_size in order */
1890     smp_rmb();
1891     result->completed = state->written_size;
1892     result->total = state->total_size;
1893     return result;
1894 }
1895 
1896 void qmp_dump_guest_memory(bool paging, const char *file,
1897                            bool has_detach, bool detach,
1898                            bool has_begin, int64_t begin, bool has_length,
1899                            int64_t length, bool has_format,
1900                            DumpGuestMemoryFormat format, Error **errp)
1901 {
1902     ERRP_GUARD();
1903     const char *p;
1904     int fd = -1;
1905     DumpState *s;
1906     bool detach_p = false;
1907 
1908     if (runstate_check(RUN_STATE_INMIGRATE)) {
1909         error_setg(errp, "Dump not allowed during incoming migration.");
1910         return;
1911     }
1912 
1913     /* if there is a dump in background, we should wait until the dump
1914      * finished */
1915     if (qemu_system_dump_in_progress()) {
1916         error_setg(errp, "There is a dump in process, please wait.");
1917         return;
1918     }
1919 
1920     /*
1921      * kdump-compressed format need the whole memory dumped, so paging or
1922      * filter is not supported here.
1923      */
1924     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1925         (paging || has_begin || has_length)) {
1926         error_setg(errp, "kdump-compressed format doesn't support paging or "
1927                          "filter");
1928         return;
1929     }
1930     if (has_begin && !has_length) {
1931         error_setg(errp, QERR_MISSING_PARAMETER, "length");
1932         return;
1933     }
1934     if (!has_begin && has_length) {
1935         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1936         return;
1937     }
1938     if (has_detach) {
1939         detach_p = detach;
1940     }
1941 
1942     /* check whether lzo/snappy is supported */
1943 #ifndef CONFIG_LZO
1944     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1945         error_setg(errp, "kdump-lzo is not available now");
1946         return;
1947     }
1948 #endif
1949 
1950 #ifndef CONFIG_SNAPPY
1951     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1952         error_setg(errp, "kdump-snappy is not available now");
1953         return;
1954     }
1955 #endif
1956 
1957 #ifndef TARGET_X86_64
1958     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1959         error_setg(errp, "Windows dump is only available for x86-64");
1960         return;
1961     }
1962 #endif
1963 
1964 #if !defined(WIN32)
1965     if (strstart(file, "fd:", &p)) {
1966         fd = monitor_get_fd(monitor_cur(), p, errp);
1967         if (fd == -1) {
1968             return;
1969         }
1970     }
1971 #endif
1972 
1973     if  (strstart(file, "file:", &p)) {
1974         fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1975         if (fd < 0) {
1976             error_setg_file_open(errp, errno, p);
1977             return;
1978         }
1979     }
1980 
1981     if (fd == -1) {
1982         error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
1983         return;
1984     }
1985 
1986     if (!dump_migration_blocker) {
1987         error_setg(&dump_migration_blocker,
1988                    "Live migration disabled: dump-guest-memory in progress");
1989     }
1990 
1991     /*
1992      * Allows even for -only-migratable, but forbid migration during the
1993      * process of dump guest memory.
1994      */
1995     if (migrate_add_blocker_internal(dump_migration_blocker, errp)) {
1996         /* Remember to release the fd before passing it over to dump state */
1997         close(fd);
1998         return;
1999     }
2000 
2001     s = &dump_state_global;
2002     dump_state_prepare(s);
2003 
2004     dump_init(s, fd, has_format, format, paging, has_begin,
2005               begin, length, errp);
2006     if (*errp) {
2007         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2008         return;
2009     }
2010 
2011     if (detach_p) {
2012         /* detached dump */
2013         s->detached = true;
2014         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2015                            s, QEMU_THREAD_DETACHED);
2016     } else {
2017         /* sync dump */
2018         dump_process(s, errp);
2019     }
2020 }
2021 
2022 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2023 {
2024     DumpGuestMemoryCapability *cap =
2025                                   g_new0(DumpGuestMemoryCapability, 1);
2026     DumpGuestMemoryFormatList **tail = &cap->formats;
2027 
2028     /* elf is always available */
2029     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2030 
2031     /* kdump-zlib is always available */
2032     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2033 
2034     /* add new item if kdump-lzo is available */
2035 #ifdef CONFIG_LZO
2036     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2037 #endif
2038 
2039     /* add new item if kdump-snappy is available */
2040 #ifdef CONFIG_SNAPPY
2041     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2042 #endif
2043 
2044     /* Windows dump is available only if target is x86_64 */
2045 #ifdef TARGET_X86_64
2046     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2047 #endif
2048 
2049     return cap;
2050 }
2051