xref: /openbmc/qemu/dump/dump.c (revision 05bbaa50)
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 static inline bool dump_is_64bit(DumpState *s)
58 {
59     return s->dump_info.d_class == ELFCLASS64;
60 }
61 
62 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
63 {
64     if (s->dump_info.d_endian == ELFDATA2LSB) {
65         val = cpu_to_le16(val);
66     } else {
67         val = cpu_to_be16(val);
68     }
69 
70     return val;
71 }
72 
73 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
74 {
75     if (s->dump_info.d_endian == ELFDATA2LSB) {
76         val = cpu_to_le32(val);
77     } else {
78         val = cpu_to_be32(val);
79     }
80 
81     return val;
82 }
83 
84 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
85 {
86     if (s->dump_info.d_endian == ELFDATA2LSB) {
87         val = cpu_to_le64(val);
88     } else {
89         val = cpu_to_be64(val);
90     }
91 
92     return val;
93 }
94 
95 static int dump_cleanup(DumpState *s)
96 {
97     guest_phys_blocks_free(&s->guest_phys_blocks);
98     memory_mapping_list_free(&s->list);
99     close(s->fd);
100     g_free(s->guest_note);
101     s->guest_note = NULL;
102     if (s->resume) {
103         if (s->detached) {
104             qemu_mutex_lock_iothread();
105         }
106         vm_start();
107         if (s->detached) {
108             qemu_mutex_unlock_iothread();
109         }
110     }
111     migrate_del_blocker(dump_migration_blocker);
112 
113     return 0;
114 }
115 
116 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
117 {
118     DumpState *s = opaque;
119     size_t written_size;
120 
121     written_size = qemu_write_full(s->fd, buf, size);
122     if (written_size != size) {
123         return -errno;
124     }
125 
126     return 0;
127 }
128 
129 static void write_elf64_header(DumpState *s, Error **errp)
130 {
131     /*
132      * phnum in the elf header is 16 bit, if we have more segments we
133      * set phnum to PN_XNUM and write the real number of segments to a
134      * special section.
135      */
136     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
137     Elf64_Ehdr elf_header;
138     int ret;
139 
140     memset(&elf_header, 0, sizeof(Elf64_Ehdr));
141     memcpy(&elf_header, ELFMAG, SELFMAG);
142     elf_header.e_ident[EI_CLASS] = ELFCLASS64;
143     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
144     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
145     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
146     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
147     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
148     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
149     elf_header.e_phoff = cpu_to_dump64(s, s->phdr_offset);
150     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
151     elf_header.e_phnum = cpu_to_dump16(s, phnum);
152     if (s->shdr_num) {
153         elf_header.e_shoff = cpu_to_dump64(s, s->shdr_offset);
154         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
155         elf_header.e_shnum = cpu_to_dump16(s, s->shdr_num);
156     }
157 
158     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
159     if (ret < 0) {
160         error_setg_errno(errp, -ret, "dump: failed to write elf header");
161     }
162 }
163 
164 static void write_elf32_header(DumpState *s, Error **errp)
165 {
166     /*
167      * phnum in the elf header is 16 bit, if we have more segments we
168      * set phnum to PN_XNUM and write the real number of segments to a
169      * special section.
170      */
171     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
172     Elf32_Ehdr elf_header;
173     int ret;
174 
175     memset(&elf_header, 0, sizeof(Elf32_Ehdr));
176     memcpy(&elf_header, ELFMAG, SELFMAG);
177     elf_header.e_ident[EI_CLASS] = ELFCLASS32;
178     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
179     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
180     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
181     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
182     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
183     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
184     elf_header.e_phoff = cpu_to_dump32(s, s->phdr_offset);
185     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
186     elf_header.e_phnum = cpu_to_dump16(s, phnum);
187     if (s->shdr_num) {
188         elf_header.e_shoff = cpu_to_dump32(s, s->shdr_offset);
189         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
190         elf_header.e_shnum = cpu_to_dump16(s, s->shdr_num);
191     }
192 
193     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
194     if (ret < 0) {
195         error_setg_errno(errp, -ret, "dump: failed to write elf header");
196     }
197 }
198 
199 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
200                              int phdr_index, hwaddr offset,
201                              hwaddr filesz, Error **errp)
202 {
203     Elf64_Phdr phdr;
204     int ret;
205 
206     memset(&phdr, 0, sizeof(Elf64_Phdr));
207     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
208     phdr.p_offset = cpu_to_dump64(s, offset);
209     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
210     phdr.p_filesz = cpu_to_dump64(s, filesz);
211     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
212     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
213 
214     assert(memory_mapping->length >= filesz);
215 
216     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
217     if (ret < 0) {
218         error_setg_errno(errp, -ret,
219                          "dump: failed to write program header table");
220     }
221 }
222 
223 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
224                              int phdr_index, hwaddr offset,
225                              hwaddr filesz, Error **errp)
226 {
227     Elf32_Phdr phdr;
228     int ret;
229 
230     memset(&phdr, 0, sizeof(Elf32_Phdr));
231     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
232     phdr.p_offset = cpu_to_dump32(s, offset);
233     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
234     phdr.p_filesz = cpu_to_dump32(s, filesz);
235     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
236     phdr.p_vaddr =
237         cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
238 
239     assert(memory_mapping->length >= filesz);
240 
241     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
242     if (ret < 0) {
243         error_setg_errno(errp, -ret,
244                          "dump: failed to write program header table");
245     }
246 }
247 
248 static void write_elf64_note(DumpState *s, Error **errp)
249 {
250     Elf64_Phdr phdr;
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, s->note_offset);
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     Elf32_Phdr phdr;
316     int ret;
317 
318     memset(&phdr, 0, sizeof(Elf32_Phdr));
319     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
320     phdr.p_offset = cpu_to_dump32(s, s->note_offset);
321     phdr.p_paddr = 0;
322     phdr.p_filesz = cpu_to_dump32(s, s->note_size);
323     phdr.p_memsz = cpu_to_dump32(s, s->note_size);
324     phdr.p_vaddr = 0;
325 
326     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
327     if (ret < 0) {
328         error_setg_errno(errp, -ret,
329                          "dump: failed to write program header table");
330     }
331 }
332 
333 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
334                               Error **errp)
335 {
336     CPUState *cpu;
337     int ret;
338     int id;
339 
340     CPU_FOREACH(cpu) {
341         id = cpu_index(cpu);
342         ret = cpu_write_elf32_note(f, cpu, id, s);
343         if (ret < 0) {
344             error_setg(errp, "dump: failed to write elf notes");
345             return;
346         }
347     }
348 
349     CPU_FOREACH(cpu) {
350         ret = cpu_write_elf32_qemunote(f, cpu, s);
351         if (ret < 0) {
352             error_setg(errp, "dump: failed to write CPU status");
353             return;
354         }
355     }
356 
357     write_guest_note(f, s, errp);
358 }
359 
360 static void write_elf_section(DumpState *s, int type, Error **errp)
361 {
362     Elf32_Shdr shdr32;
363     Elf64_Shdr shdr64;
364     int shdr_size;
365     void *shdr;
366     int ret;
367 
368     if (type == 0) {
369         shdr_size = sizeof(Elf32_Shdr);
370         memset(&shdr32, 0, shdr_size);
371         shdr32.sh_info = cpu_to_dump32(s, s->phdr_num);
372         shdr = &shdr32;
373     } else {
374         shdr_size = sizeof(Elf64_Shdr);
375         memset(&shdr64, 0, shdr_size);
376         shdr64.sh_info = cpu_to_dump32(s, s->phdr_num);
377         shdr = &shdr64;
378     }
379 
380     ret = fd_write_vmcore(shdr, shdr_size, s);
381     if (ret < 0) {
382         error_setg_errno(errp, -ret,
383                          "dump: failed to write section header table");
384     }
385 }
386 
387 static void write_data(DumpState *s, void *buf, int length, Error **errp)
388 {
389     int ret;
390 
391     ret = fd_write_vmcore(buf, length, s);
392     if (ret < 0) {
393         error_setg_errno(errp, -ret, "dump: failed to save memory");
394     } else {
395         s->written_size += length;
396     }
397 }
398 
399 /* write the memory to vmcore. 1 page per I/O. */
400 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
401                          int64_t size, Error **errp)
402 {
403     ERRP_GUARD();
404     int64_t i;
405 
406     for (i = 0; i < size / s->dump_info.page_size; i++) {
407         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
408                    s->dump_info.page_size, errp);
409         if (*errp) {
410             return;
411         }
412     }
413 
414     if ((size % s->dump_info.page_size) != 0) {
415         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
416                    size % s->dump_info.page_size, errp);
417         if (*errp) {
418             return;
419         }
420     }
421 }
422 
423 /* get the memory's offset and size in the vmcore */
424 static void get_offset_range(hwaddr phys_addr,
425                              ram_addr_t mapping_length,
426                              DumpState *s,
427                              hwaddr *p_offset,
428                              hwaddr *p_filesz)
429 {
430     GuestPhysBlock *block;
431     hwaddr offset = s->memory_offset;
432     int64_t size_in_block, start;
433 
434     /* When the memory is not stored into vmcore, offset will be -1 */
435     *p_offset = -1;
436     *p_filesz = 0;
437 
438     if (s->has_filter) {
439         if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
440             return;
441         }
442     }
443 
444     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
445         if (s->has_filter) {
446             if (block->target_start >= s->begin + s->length ||
447                 block->target_end <= s->begin) {
448                 /* This block is out of the range */
449                 continue;
450             }
451 
452             if (s->begin <= block->target_start) {
453                 start = block->target_start;
454             } else {
455                 start = s->begin;
456             }
457 
458             size_in_block = block->target_end - start;
459             if (s->begin + s->length < block->target_end) {
460                 size_in_block -= block->target_end - (s->begin + s->length);
461             }
462         } else {
463             start = block->target_start;
464             size_in_block = block->target_end - block->target_start;
465         }
466 
467         if (phys_addr >= start && phys_addr < start + size_in_block) {
468             *p_offset = phys_addr - start + offset;
469 
470             /* The offset range mapped from the vmcore file must not spill over
471              * the GuestPhysBlock, clamp it. The rest of the mapping will be
472              * zero-filled in memory at load time; see
473              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
474              */
475             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
476                         mapping_length :
477                         size_in_block - (phys_addr - start);
478             return;
479         }
480 
481         offset += size_in_block;
482     }
483 }
484 
485 static void write_elf_loads(DumpState *s, Error **errp)
486 {
487     ERRP_GUARD();
488     hwaddr offset, filesz;
489     MemoryMapping *memory_mapping;
490     uint32_t phdr_index = 1;
491 
492     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
493         get_offset_range(memory_mapping->phys_addr,
494                          memory_mapping->length,
495                          s, &offset, &filesz);
496         if (dump_is_64bit(s)) {
497             write_elf64_load(s, memory_mapping, phdr_index++, offset,
498                              filesz, errp);
499         } else {
500             write_elf32_load(s, memory_mapping, phdr_index++, offset,
501                              filesz, errp);
502         }
503 
504         if (*errp) {
505             return;
506         }
507 
508         if (phdr_index >= s->phdr_num) {
509             break;
510         }
511     }
512 }
513 
514 /* write elf header, PT_NOTE and elf note to vmcore. */
515 static void dump_begin(DumpState *s, Error **errp)
516 {
517     ERRP_GUARD();
518 
519     /*
520      * the vmcore's format is:
521      *   --------------
522      *   |  elf header |
523      *   --------------
524      *   |  PT_NOTE    |
525      *   --------------
526      *   |  PT_LOAD    |
527      *   --------------
528      *   |  ......     |
529      *   --------------
530      *   |  PT_LOAD    |
531      *   --------------
532      *   |  sec_hdr    |
533      *   --------------
534      *   |  elf note   |
535      *   --------------
536      *   |  memory     |
537      *   --------------
538      *
539      * we only know where the memory is saved after we write elf note into
540      * vmcore.
541      */
542 
543     /* write elf header to vmcore */
544     if (dump_is_64bit(s)) {
545         write_elf64_header(s, errp);
546     } else {
547         write_elf32_header(s, errp);
548     }
549     if (*errp) {
550         return;
551     }
552 
553     if (dump_is_64bit(s)) {
554         /* write PT_NOTE to vmcore */
555         write_elf64_note(s, errp);
556         if (*errp) {
557             return;
558         }
559 
560         /* write all PT_LOAD to vmcore */
561         write_elf_loads(s, errp);
562         if (*errp) {
563             return;
564         }
565 
566         /* write section to vmcore */
567         if (s->shdr_num) {
568             write_elf_section(s, 1, errp);
569             if (*errp) {
570                 return;
571             }
572         }
573 
574         /* write notes to vmcore */
575         write_elf64_notes(fd_write_vmcore, s, errp);
576         if (*errp) {
577             return;
578         }
579     } else {
580         /* write PT_NOTE to vmcore */
581         write_elf32_note(s, errp);
582         if (*errp) {
583             return;
584         }
585 
586         /* write all PT_LOAD to vmcore */
587         write_elf_loads(s, errp);
588         if (*errp) {
589             return;
590         }
591 
592         /* write section to vmcore */
593         if (s->shdr_num) {
594             write_elf_section(s, 0, errp);
595             if (*errp) {
596                 return;
597             }
598         }
599 
600         /* write notes to vmcore */
601         write_elf32_notes(fd_write_vmcore, s, errp);
602         if (*errp) {
603             return;
604         }
605     }
606 }
607 
608 static int get_next_block(DumpState *s, GuestPhysBlock *block)
609 {
610     while (1) {
611         block = QTAILQ_NEXT(block, next);
612         if (!block) {
613             /* no more block */
614             return 1;
615         }
616 
617         s->start = 0;
618         s->next_block = block;
619         if (s->has_filter) {
620             if (block->target_start >= s->begin + s->length ||
621                 block->target_end <= s->begin) {
622                 /* This block is out of the range */
623                 continue;
624             }
625 
626             if (s->begin > block->target_start) {
627                 s->start = s->begin - block->target_start;
628             }
629         }
630 
631         return 0;
632     }
633 }
634 
635 /* write all memory to vmcore */
636 static void dump_iterate(DumpState *s, Error **errp)
637 {
638     ERRP_GUARD();
639     GuestPhysBlock *block;
640     int64_t size;
641 
642     do {
643         block = s->next_block;
644 
645         size = block->target_end - block->target_start;
646         if (s->has_filter) {
647             size -= s->start;
648             if (s->begin + s->length < block->target_end) {
649                 size -= block->target_end - (s->begin + s->length);
650             }
651         }
652         write_memory(s, block, s->start, size, errp);
653         if (*errp) {
654             return;
655         }
656 
657     } while (!get_next_block(s, block));
658 }
659 
660 static void create_vmcore(DumpState *s, Error **errp)
661 {
662     ERRP_GUARD();
663 
664     dump_begin(s, errp);
665     if (*errp) {
666         return;
667     }
668 
669     dump_iterate(s, errp);
670 }
671 
672 static int write_start_flat_header(int fd)
673 {
674     MakedumpfileHeader *mh;
675     int ret = 0;
676 
677     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
678     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
679 
680     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
681            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
682 
683     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
684     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
685 
686     size_t written_size;
687     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
688     if (written_size != MAX_SIZE_MDF_HEADER) {
689         ret = -1;
690     }
691 
692     g_free(mh);
693     return ret;
694 }
695 
696 static int write_end_flat_header(int fd)
697 {
698     MakedumpfileDataHeader mdh;
699 
700     mdh.offset = END_FLAG_FLAT_HEADER;
701     mdh.buf_size = END_FLAG_FLAT_HEADER;
702 
703     size_t written_size;
704     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
705     if (written_size != sizeof(mdh)) {
706         return -1;
707     }
708 
709     return 0;
710 }
711 
712 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
713 {
714     size_t written_size;
715     MakedumpfileDataHeader mdh;
716 
717     mdh.offset = cpu_to_be64(offset);
718     mdh.buf_size = cpu_to_be64(size);
719 
720     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
721     if (written_size != sizeof(mdh)) {
722         return -1;
723     }
724 
725     written_size = qemu_write_full(fd, buf, size);
726     if (written_size != size) {
727         return -1;
728     }
729 
730     return 0;
731 }
732 
733 static int buf_write_note(const void *buf, size_t size, void *opaque)
734 {
735     DumpState *s = opaque;
736 
737     /* note_buf is not enough */
738     if (s->note_buf_offset + size > s->note_size) {
739         return -1;
740     }
741 
742     memcpy(s->note_buf + s->note_buf_offset, buf, size);
743 
744     s->note_buf_offset += size;
745 
746     return 0;
747 }
748 
749 /*
750  * This function retrieves various sizes from an elf header.
751  *
752  * @note has to be a valid ELF note. The return sizes are unmodified
753  * (not padded or rounded up to be multiple of 4).
754  */
755 static void get_note_sizes(DumpState *s, const void *note,
756                            uint64_t *note_head_size,
757                            uint64_t *name_size,
758                            uint64_t *desc_size)
759 {
760     uint64_t note_head_sz;
761     uint64_t name_sz;
762     uint64_t desc_sz;
763 
764     if (dump_is_64bit(s)) {
765         const Elf64_Nhdr *hdr = note;
766         note_head_sz = sizeof(Elf64_Nhdr);
767         name_sz = tswap64(hdr->n_namesz);
768         desc_sz = tswap64(hdr->n_descsz);
769     } else {
770         const Elf32_Nhdr *hdr = note;
771         note_head_sz = sizeof(Elf32_Nhdr);
772         name_sz = tswap32(hdr->n_namesz);
773         desc_sz = tswap32(hdr->n_descsz);
774     }
775 
776     if (note_head_size) {
777         *note_head_size = note_head_sz;
778     }
779     if (name_size) {
780         *name_size = name_sz;
781     }
782     if (desc_size) {
783         *desc_size = desc_sz;
784     }
785 }
786 
787 static bool note_name_equal(DumpState *s,
788                             const uint8_t *note, const char *name)
789 {
790     int len = strlen(name) + 1;
791     uint64_t head_size, name_size;
792 
793     get_note_sizes(s, note, &head_size, &name_size, NULL);
794     head_size = ROUND_UP(head_size, 4);
795 
796     return name_size == len && memcmp(note + head_size, name, len) == 0;
797 }
798 
799 /* write common header, sub header and elf note to vmcore */
800 static void create_header32(DumpState *s, Error **errp)
801 {
802     ERRP_GUARD();
803     DiskDumpHeader32 *dh = NULL;
804     KdumpSubHeader32 *kh = NULL;
805     size_t size;
806     uint32_t block_size;
807     uint32_t sub_hdr_size;
808     uint32_t bitmap_blocks;
809     uint32_t status = 0;
810     uint64_t offset_note;
811 
812     /* write common header, the version of kdump-compressed format is 6th */
813     size = sizeof(DiskDumpHeader32);
814     dh = g_malloc0(size);
815 
816     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
817     dh->header_version = cpu_to_dump32(s, 6);
818     block_size = s->dump_info.page_size;
819     dh->block_size = cpu_to_dump32(s, block_size);
820     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
821     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
822     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
823     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
824     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
825     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
826     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
827     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
828     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
829 
830     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
831         status |= DUMP_DH_COMPRESSED_ZLIB;
832     }
833 #ifdef CONFIG_LZO
834     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
835         status |= DUMP_DH_COMPRESSED_LZO;
836     }
837 #endif
838 #ifdef CONFIG_SNAPPY
839     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
840         status |= DUMP_DH_COMPRESSED_SNAPPY;
841     }
842 #endif
843     dh->status = cpu_to_dump32(s, status);
844 
845     if (write_buffer(s->fd, 0, dh, size) < 0) {
846         error_setg(errp, "dump: failed to write disk dump header");
847         goto out;
848     }
849 
850     /* write sub header */
851     size = sizeof(KdumpSubHeader32);
852     kh = g_malloc0(size);
853 
854     /* 64bit max_mapnr_64 */
855     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
856     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
857     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
858 
859     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
860     if (s->guest_note &&
861         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
862         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
863 
864         get_note_sizes(s, s->guest_note,
865                        &hsize, &name_size, &size_vmcoreinfo_desc);
866         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
867             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
868         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
869         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
870     }
871 
872     kh->offset_note = cpu_to_dump64(s, offset_note);
873     kh->note_size = cpu_to_dump32(s, s->note_size);
874 
875     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
876                      block_size, kh, size) < 0) {
877         error_setg(errp, "dump: failed to write kdump sub header");
878         goto out;
879     }
880 
881     /* write note */
882     s->note_buf = g_malloc0(s->note_size);
883     s->note_buf_offset = 0;
884 
885     /* use s->note_buf to store notes temporarily */
886     write_elf32_notes(buf_write_note, s, errp);
887     if (*errp) {
888         goto out;
889     }
890     if (write_buffer(s->fd, offset_note, s->note_buf,
891                      s->note_size) < 0) {
892         error_setg(errp, "dump: failed to write notes");
893         goto out;
894     }
895 
896     /* get offset of dump_bitmap */
897     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
898                              block_size;
899 
900     /* get offset of page */
901     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
902                      block_size;
903 
904 out:
905     g_free(dh);
906     g_free(kh);
907     g_free(s->note_buf);
908 }
909 
910 /* write common header, sub header and elf note to vmcore */
911 static void create_header64(DumpState *s, Error **errp)
912 {
913     ERRP_GUARD();
914     DiskDumpHeader64 *dh = NULL;
915     KdumpSubHeader64 *kh = NULL;
916     size_t size;
917     uint32_t block_size;
918     uint32_t sub_hdr_size;
919     uint32_t bitmap_blocks;
920     uint32_t status = 0;
921     uint64_t offset_note;
922 
923     /* write common header, the version of kdump-compressed format is 6th */
924     size = sizeof(DiskDumpHeader64);
925     dh = g_malloc0(size);
926 
927     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
928     dh->header_version = cpu_to_dump32(s, 6);
929     block_size = s->dump_info.page_size;
930     dh->block_size = cpu_to_dump32(s, block_size);
931     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
932     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
933     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
934     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
935     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
936     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
937     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
938     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
939     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
940 
941     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
942         status |= DUMP_DH_COMPRESSED_ZLIB;
943     }
944 #ifdef CONFIG_LZO
945     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
946         status |= DUMP_DH_COMPRESSED_LZO;
947     }
948 #endif
949 #ifdef CONFIG_SNAPPY
950     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
951         status |= DUMP_DH_COMPRESSED_SNAPPY;
952     }
953 #endif
954     dh->status = cpu_to_dump32(s, status);
955 
956     if (write_buffer(s->fd, 0, dh, size) < 0) {
957         error_setg(errp, "dump: failed to write disk dump header");
958         goto out;
959     }
960 
961     /* write sub header */
962     size = sizeof(KdumpSubHeader64);
963     kh = g_malloc0(size);
964 
965     /* 64bit max_mapnr_64 */
966     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
967     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
968     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
969 
970     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
971     if (s->guest_note &&
972         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
973         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
974 
975         get_note_sizes(s, s->guest_note,
976                        &hsize, &name_size, &size_vmcoreinfo_desc);
977         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
978             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
979         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
980         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
981     }
982 
983     kh->offset_note = cpu_to_dump64(s, offset_note);
984     kh->note_size = cpu_to_dump64(s, s->note_size);
985 
986     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
987                      block_size, kh, size) < 0) {
988         error_setg(errp, "dump: failed to write kdump sub header");
989         goto out;
990     }
991 
992     /* write note */
993     s->note_buf = g_malloc0(s->note_size);
994     s->note_buf_offset = 0;
995 
996     /* use s->note_buf to store notes temporarily */
997     write_elf64_notes(buf_write_note, s, errp);
998     if (*errp) {
999         goto out;
1000     }
1001 
1002     if (write_buffer(s->fd, offset_note, s->note_buf,
1003                      s->note_size) < 0) {
1004         error_setg(errp, "dump: failed to write notes");
1005         goto out;
1006     }
1007 
1008     /* get offset of dump_bitmap */
1009     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1010                              block_size;
1011 
1012     /* get offset of page */
1013     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1014                      block_size;
1015 
1016 out:
1017     g_free(dh);
1018     g_free(kh);
1019     g_free(s->note_buf);
1020 }
1021 
1022 static void write_dump_header(DumpState *s, Error **errp)
1023 {
1024     if (dump_is_64bit(s)) {
1025         create_header64(s, errp);
1026     } else {
1027         create_header32(s, errp);
1028     }
1029 }
1030 
1031 static size_t dump_bitmap_get_bufsize(DumpState *s)
1032 {
1033     return s->dump_info.page_size;
1034 }
1035 
1036 /*
1037  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1038  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1039  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1040  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1041  * vmcore, ie. synchronizing un-sync bit into vmcore.
1042  */
1043 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1044                            uint8_t *buf, DumpState *s)
1045 {
1046     off_t old_offset, new_offset;
1047     off_t offset_bitmap1, offset_bitmap2;
1048     uint32_t byte, bit;
1049     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1050     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1051 
1052     /* should not set the previous place */
1053     assert(last_pfn <= pfn);
1054 
1055     /*
1056      * if the bit needed to be set is not cached in buf, flush the data in buf
1057      * to vmcore firstly.
1058      * making new_offset be bigger than old_offset can also sync remained data
1059      * into vmcore.
1060      */
1061     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1062     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1063 
1064     while (old_offset < new_offset) {
1065         /* calculate the offset and write dump_bitmap */
1066         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1067         if (write_buffer(s->fd, offset_bitmap1, buf,
1068                          bitmap_bufsize) < 0) {
1069             return -1;
1070         }
1071 
1072         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1073         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1074                          old_offset;
1075         if (write_buffer(s->fd, offset_bitmap2, buf,
1076                          bitmap_bufsize) < 0) {
1077             return -1;
1078         }
1079 
1080         memset(buf, 0, bitmap_bufsize);
1081         old_offset += bitmap_bufsize;
1082     }
1083 
1084     /* get the exact place of the bit in the buf, and set it */
1085     byte = (pfn % bits_per_buf) / CHAR_BIT;
1086     bit = (pfn % bits_per_buf) % CHAR_BIT;
1087     if (value) {
1088         buf[byte] |= 1u << bit;
1089     } else {
1090         buf[byte] &= ~(1u << bit);
1091     }
1092 
1093     return 0;
1094 }
1095 
1096 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1097 {
1098     int target_page_shift = ctz32(s->dump_info.page_size);
1099 
1100     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1101 }
1102 
1103 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1104 {
1105     int target_page_shift = ctz32(s->dump_info.page_size);
1106 
1107     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1108 }
1109 
1110 /*
1111  * exam every page and return the page frame number and the address of the page.
1112  * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1113  * blocks, so block->target_start and block->target_end should be interal
1114  * multiples of the target page size.
1115  */
1116 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1117                           uint8_t **bufptr, DumpState *s)
1118 {
1119     GuestPhysBlock *block = *blockptr;
1120     hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1121     uint8_t *buf;
1122 
1123     /* block == NULL means the start of the iteration */
1124     if (!block) {
1125         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1126         *blockptr = block;
1127         assert((block->target_start & ~target_page_mask) == 0);
1128         assert((block->target_end & ~target_page_mask) == 0);
1129         *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1130         if (bufptr) {
1131             *bufptr = block->host_addr;
1132         }
1133         return true;
1134     }
1135 
1136     *pfnptr = *pfnptr + 1;
1137     addr = dump_pfn_to_paddr(s, *pfnptr);
1138 
1139     if ((addr >= block->target_start) &&
1140         (addr + s->dump_info.page_size <= block->target_end)) {
1141         buf = block->host_addr + (addr - block->target_start);
1142     } else {
1143         /* the next page is in the next block */
1144         block = QTAILQ_NEXT(block, next);
1145         *blockptr = block;
1146         if (!block) {
1147             return false;
1148         }
1149         assert((block->target_start & ~target_page_mask) == 0);
1150         assert((block->target_end & ~target_page_mask) == 0);
1151         *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1152         buf = block->host_addr;
1153     }
1154 
1155     if (bufptr) {
1156         *bufptr = buf;
1157     }
1158 
1159     return true;
1160 }
1161 
1162 static void write_dump_bitmap(DumpState *s, Error **errp)
1163 {
1164     int ret = 0;
1165     uint64_t last_pfn, pfn;
1166     void *dump_bitmap_buf;
1167     size_t num_dumpable;
1168     GuestPhysBlock *block_iter = NULL;
1169     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1170     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1171 
1172     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1173     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1174 
1175     num_dumpable = 0;
1176     last_pfn = 0;
1177 
1178     /*
1179      * exam memory page by page, and set the bit in dump_bitmap corresponded
1180      * to the existing page.
1181      */
1182     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1183         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1184         if (ret < 0) {
1185             error_setg(errp, "dump: failed to set dump_bitmap");
1186             goto out;
1187         }
1188 
1189         last_pfn = pfn;
1190         num_dumpable++;
1191     }
1192 
1193     /*
1194      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1195      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1196      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1197      */
1198     if (num_dumpable > 0) {
1199         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1200                               dump_bitmap_buf, s);
1201         if (ret < 0) {
1202             error_setg(errp, "dump: failed to sync dump_bitmap");
1203             goto out;
1204         }
1205     }
1206 
1207     /* number of dumpable pages that will be dumped later */
1208     s->num_dumpable = num_dumpable;
1209 
1210 out:
1211     g_free(dump_bitmap_buf);
1212 }
1213 
1214 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1215                                off_t offset)
1216 {
1217     data_cache->fd = s->fd;
1218     data_cache->data_size = 0;
1219     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1220     data_cache->buf = g_malloc0(data_cache->buf_size);
1221     data_cache->offset = offset;
1222 }
1223 
1224 static int write_cache(DataCache *dc, const void *buf, size_t size,
1225                        bool flag_sync)
1226 {
1227     /*
1228      * dc->buf_size should not be less than size, otherwise dc will never be
1229      * enough
1230      */
1231     assert(size <= dc->buf_size);
1232 
1233     /*
1234      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1235      * otherwise check if the space is enough for caching data in buf, if not,
1236      * write the data in dc->buf to dc->fd and reset dc->buf
1237      */
1238     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1239         (flag_sync && dc->data_size > 0)) {
1240         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1241             return -1;
1242         }
1243 
1244         dc->offset += dc->data_size;
1245         dc->data_size = 0;
1246     }
1247 
1248     if (!flag_sync) {
1249         memcpy(dc->buf + dc->data_size, buf, size);
1250         dc->data_size += size;
1251     }
1252 
1253     return 0;
1254 }
1255 
1256 static void free_data_cache(DataCache *data_cache)
1257 {
1258     g_free(data_cache->buf);
1259 }
1260 
1261 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1262 {
1263     switch (flag_compress) {
1264     case DUMP_DH_COMPRESSED_ZLIB:
1265         return compressBound(page_size);
1266 
1267     case DUMP_DH_COMPRESSED_LZO:
1268         /*
1269          * LZO will expand incompressible data by a little amount. Please check
1270          * the following URL to see the expansion calculation:
1271          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1272          */
1273         return page_size + page_size / 16 + 64 + 3;
1274 
1275 #ifdef CONFIG_SNAPPY
1276     case DUMP_DH_COMPRESSED_SNAPPY:
1277         return snappy_max_compressed_length(page_size);
1278 #endif
1279     }
1280     return 0;
1281 }
1282 
1283 static void write_dump_pages(DumpState *s, Error **errp)
1284 {
1285     int ret = 0;
1286     DataCache page_desc, page_data;
1287     size_t len_buf_out, size_out;
1288 #ifdef CONFIG_LZO
1289     lzo_bytep wrkmem = NULL;
1290 #endif
1291     uint8_t *buf_out = NULL;
1292     off_t offset_desc, offset_data;
1293     PageDescriptor pd, pd_zero;
1294     uint8_t *buf;
1295     GuestPhysBlock *block_iter = NULL;
1296     uint64_t pfn_iter;
1297 
1298     /* get offset of page_desc and page_data in dump file */
1299     offset_desc = s->offset_page;
1300     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1301 
1302     prepare_data_cache(&page_desc, s, offset_desc);
1303     prepare_data_cache(&page_data, s, offset_data);
1304 
1305     /* prepare buffer to store compressed data */
1306     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1307     assert(len_buf_out != 0);
1308 
1309 #ifdef CONFIG_LZO
1310     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1311 #endif
1312 
1313     buf_out = g_malloc(len_buf_out);
1314 
1315     /*
1316      * init zero page's page_desc and page_data, because every zero page
1317      * uses the same page_data
1318      */
1319     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1320     pd_zero.flags = cpu_to_dump32(s, 0);
1321     pd_zero.offset = cpu_to_dump64(s, offset_data);
1322     pd_zero.page_flags = cpu_to_dump64(s, 0);
1323     buf = g_malloc0(s->dump_info.page_size);
1324     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1325     g_free(buf);
1326     if (ret < 0) {
1327         error_setg(errp, "dump: failed to write page data (zero page)");
1328         goto out;
1329     }
1330 
1331     offset_data += s->dump_info.page_size;
1332 
1333     /*
1334      * dump memory to vmcore page by page. zero page will all be resided in the
1335      * first page of page section
1336      */
1337     while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1338         /* check zero page */
1339         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1340             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1341                               false);
1342             if (ret < 0) {
1343                 error_setg(errp, "dump: failed to write page desc");
1344                 goto out;
1345             }
1346         } else {
1347             /*
1348              * not zero page, then:
1349              * 1. compress the page
1350              * 2. write the compressed page into the cache of page_data
1351              * 3. get page desc of the compressed page and write it into the
1352              *    cache of page_desc
1353              *
1354              * only one compression format will be used here, for
1355              * s->flag_compress is set. But when compression fails to work,
1356              * we fall back to save in plaintext.
1357              */
1358              size_out = len_buf_out;
1359              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1360                     (compress2(buf_out, (uLongf *)&size_out, buf,
1361                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1362                     (size_out < s->dump_info.page_size)) {
1363                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1364                 pd.size  = cpu_to_dump32(s, size_out);
1365 
1366                 ret = write_cache(&page_data, buf_out, size_out, false);
1367                 if (ret < 0) {
1368                     error_setg(errp, "dump: failed to write page data");
1369                     goto out;
1370                 }
1371 #ifdef CONFIG_LZO
1372             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1373                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1374                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1375                     (size_out < s->dump_info.page_size)) {
1376                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1377                 pd.size  = cpu_to_dump32(s, size_out);
1378 
1379                 ret = write_cache(&page_data, buf_out, size_out, false);
1380                 if (ret < 0) {
1381                     error_setg(errp, "dump: failed to write page data");
1382                     goto out;
1383                 }
1384 #endif
1385 #ifdef CONFIG_SNAPPY
1386             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1387                     (snappy_compress((char *)buf, s->dump_info.page_size,
1388                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1389                     (size_out < s->dump_info.page_size)) {
1390                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1391                 pd.size  = cpu_to_dump32(s, size_out);
1392 
1393                 ret = write_cache(&page_data, buf_out, size_out, false);
1394                 if (ret < 0) {
1395                     error_setg(errp, "dump: failed to write page data");
1396                     goto out;
1397                 }
1398 #endif
1399             } else {
1400                 /*
1401                  * fall back to save in plaintext, size_out should be
1402                  * assigned the target's page size
1403                  */
1404                 pd.flags = cpu_to_dump32(s, 0);
1405                 size_out = s->dump_info.page_size;
1406                 pd.size = cpu_to_dump32(s, size_out);
1407 
1408                 ret = write_cache(&page_data, buf,
1409                                   s->dump_info.page_size, false);
1410                 if (ret < 0) {
1411                     error_setg(errp, "dump: failed to write page data");
1412                     goto out;
1413                 }
1414             }
1415 
1416             /* get and write page desc here */
1417             pd.page_flags = cpu_to_dump64(s, 0);
1418             pd.offset = cpu_to_dump64(s, offset_data);
1419             offset_data += size_out;
1420 
1421             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1422             if (ret < 0) {
1423                 error_setg(errp, "dump: failed to write page desc");
1424                 goto out;
1425             }
1426         }
1427         s->written_size += s->dump_info.page_size;
1428     }
1429 
1430     ret = write_cache(&page_desc, NULL, 0, true);
1431     if (ret < 0) {
1432         error_setg(errp, "dump: failed to sync cache for page_desc");
1433         goto out;
1434     }
1435     ret = write_cache(&page_data, NULL, 0, true);
1436     if (ret < 0) {
1437         error_setg(errp, "dump: failed to sync cache for page_data");
1438         goto out;
1439     }
1440 
1441 out:
1442     free_data_cache(&page_desc);
1443     free_data_cache(&page_data);
1444 
1445 #ifdef CONFIG_LZO
1446     g_free(wrkmem);
1447 #endif
1448 
1449     g_free(buf_out);
1450 }
1451 
1452 static void create_kdump_vmcore(DumpState *s, Error **errp)
1453 {
1454     ERRP_GUARD();
1455     int ret;
1456 
1457     /*
1458      * the kdump-compressed format is:
1459      *                                               File offset
1460      *  +------------------------------------------+ 0x0
1461      *  |    main header (struct disk_dump_header) |
1462      *  |------------------------------------------+ block 1
1463      *  |    sub header (struct kdump_sub_header)  |
1464      *  |------------------------------------------+ block 2
1465      *  |            1st-dump_bitmap               |
1466      *  |------------------------------------------+ block 2 + X blocks
1467      *  |            2nd-dump_bitmap               | (aligned by block)
1468      *  |------------------------------------------+ block 2 + 2 * X blocks
1469      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1470      *  |  page desc for pfn 1 (struct page_desc)  |
1471      *  |                    :                     |
1472      *  |------------------------------------------| (not aligned by block)
1473      *  |         page data (pfn 0)                |
1474      *  |         page data (pfn 1)                |
1475      *  |                    :                     |
1476      *  +------------------------------------------+
1477      */
1478 
1479     ret = write_start_flat_header(s->fd);
1480     if (ret < 0) {
1481         error_setg(errp, "dump: failed to write start flat header");
1482         return;
1483     }
1484 
1485     write_dump_header(s, errp);
1486     if (*errp) {
1487         return;
1488     }
1489 
1490     write_dump_bitmap(s, errp);
1491     if (*errp) {
1492         return;
1493     }
1494 
1495     write_dump_pages(s, errp);
1496     if (*errp) {
1497         return;
1498     }
1499 
1500     ret = write_end_flat_header(s->fd);
1501     if (ret < 0) {
1502         error_setg(errp, "dump: failed to write end flat header");
1503         return;
1504     }
1505 }
1506 
1507 static ram_addr_t get_start_block(DumpState *s)
1508 {
1509     GuestPhysBlock *block;
1510 
1511     if (!s->has_filter) {
1512         s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1513         return 0;
1514     }
1515 
1516     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1517         if (block->target_start >= s->begin + s->length ||
1518             block->target_end <= s->begin) {
1519             /* This block is out of the range */
1520             continue;
1521         }
1522 
1523         s->next_block = block;
1524         if (s->begin > block->target_start) {
1525             s->start = s->begin - block->target_start;
1526         } else {
1527             s->start = 0;
1528         }
1529         return s->start;
1530     }
1531 
1532     return -1;
1533 }
1534 
1535 static void get_max_mapnr(DumpState *s)
1536 {
1537     GuestPhysBlock *last_block;
1538 
1539     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1540     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1541 }
1542 
1543 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1544 
1545 static void dump_state_prepare(DumpState *s)
1546 {
1547     /* zero the struct, setting status to active */
1548     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1549 }
1550 
1551 bool qemu_system_dump_in_progress(void)
1552 {
1553     DumpState *state = &dump_state_global;
1554     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1555 }
1556 
1557 /* calculate total size of memory to be dumped (taking filter into
1558  * acoount.) */
1559 static int64_t dump_calculate_size(DumpState *s)
1560 {
1561     GuestPhysBlock *block;
1562     int64_t size = 0, total = 0, left = 0, right = 0;
1563 
1564     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1565         if (s->has_filter) {
1566             /* calculate the overlapped region. */
1567             left = MAX(s->begin, block->target_start);
1568             right = MIN(s->begin + s->length, block->target_end);
1569             size = right - left;
1570             size = size > 0 ? size : 0;
1571         } else {
1572             /* count the whole region in */
1573             size = (block->target_end - block->target_start);
1574         }
1575         total += size;
1576     }
1577 
1578     return total;
1579 }
1580 
1581 static void vmcoreinfo_update_phys_base(DumpState *s)
1582 {
1583     uint64_t size, note_head_size, name_size, phys_base;
1584     char **lines;
1585     uint8_t *vmci;
1586     size_t i;
1587 
1588     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1589         return;
1590     }
1591 
1592     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1593     note_head_size = ROUND_UP(note_head_size, 4);
1594 
1595     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1596     *(vmci + size) = '\0';
1597 
1598     lines = g_strsplit((char *)vmci, "\n", -1);
1599     for (i = 0; lines[i]; i++) {
1600         const char *prefix = NULL;
1601 
1602         if (s->dump_info.d_machine == EM_X86_64) {
1603             prefix = "NUMBER(phys_base)=";
1604         } else if (s->dump_info.d_machine == EM_AARCH64) {
1605             prefix = "NUMBER(PHYS_OFFSET)=";
1606         }
1607 
1608         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1609             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1610                               &phys_base) < 0) {
1611                 warn_report("Failed to read %s", prefix);
1612             } else {
1613                 s->dump_info.phys_base = phys_base;
1614             }
1615             break;
1616         }
1617     }
1618 
1619     g_strfreev(lines);
1620 }
1621 
1622 static void dump_init(DumpState *s, int fd, bool has_format,
1623                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1624                       int64_t begin, int64_t length, Error **errp)
1625 {
1626     ERRP_GUARD();
1627     VMCoreInfoState *vmci = vmcoreinfo_find();
1628     CPUState *cpu;
1629     int nr_cpus;
1630     int ret;
1631 
1632     s->has_format = has_format;
1633     s->format = format;
1634     s->written_size = 0;
1635 
1636     /* kdump-compressed is conflict with paging and filter */
1637     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1638         assert(!paging && !has_filter);
1639     }
1640 
1641     if (runstate_is_running()) {
1642         vm_stop(RUN_STATE_SAVE_VM);
1643         s->resume = true;
1644     } else {
1645         s->resume = false;
1646     }
1647 
1648     /* If we use KVM, we should synchronize the registers before we get dump
1649      * info or physmap info.
1650      */
1651     cpu_synchronize_all_states();
1652     nr_cpus = 0;
1653     CPU_FOREACH(cpu) {
1654         nr_cpus++;
1655     }
1656 
1657     s->fd = fd;
1658     s->has_filter = has_filter;
1659     s->begin = begin;
1660     s->length = length;
1661 
1662     memory_mapping_list_init(&s->list);
1663 
1664     guest_phys_blocks_init(&s->guest_phys_blocks);
1665     guest_phys_blocks_append(&s->guest_phys_blocks);
1666     s->total_size = dump_calculate_size(s);
1667 #ifdef DEBUG_DUMP_GUEST_MEMORY
1668     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1669 #endif
1670 
1671     /* it does not make sense to dump non-existent memory */
1672     if (!s->total_size) {
1673         error_setg(errp, "dump: no guest memory to dump");
1674         goto cleanup;
1675     }
1676 
1677     s->start = get_start_block(s);
1678     if (s->start == -1) {
1679         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1680         goto cleanup;
1681     }
1682 
1683     /* get dump info: endian, class and architecture.
1684      * If the target architecture is not supported, cpu_get_dump_info() will
1685      * return -1.
1686      */
1687     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1688     if (ret < 0) {
1689         error_setg(errp, QERR_UNSUPPORTED);
1690         goto cleanup;
1691     }
1692 
1693     if (!s->dump_info.page_size) {
1694         s->dump_info.page_size = TARGET_PAGE_SIZE;
1695     }
1696 
1697     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1698                                      s->dump_info.d_machine, nr_cpus);
1699     if (s->note_size < 0) {
1700         error_setg(errp, QERR_UNSUPPORTED);
1701         goto cleanup;
1702     }
1703 
1704     /*
1705      * The goal of this block is to (a) update the previously guessed
1706      * phys_base, (b) copy the guest note out of the guest.
1707      * Failure to do so is not fatal for dumping.
1708      */
1709     if (vmci) {
1710         uint64_t addr, note_head_size, name_size, desc_size;
1711         uint32_t size;
1712         uint16_t format;
1713 
1714         note_head_size = dump_is_64bit(s) ?
1715             sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr);
1716 
1717         format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1718         size = le32_to_cpu(vmci->vmcoreinfo.size);
1719         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1720         if (!vmci->has_vmcoreinfo) {
1721             warn_report("guest note is not present");
1722         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1723             warn_report("guest note size is invalid: %" PRIu32, size);
1724         } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1725             warn_report("guest note format is unsupported: %" PRIu16, format);
1726         } else {
1727             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1728             cpu_physical_memory_read(addr, s->guest_note, size);
1729 
1730             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1731             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1732                                                desc_size);
1733             if (name_size > MAX_GUEST_NOTE_SIZE ||
1734                 desc_size > MAX_GUEST_NOTE_SIZE ||
1735                 s->guest_note_size > size) {
1736                 warn_report("Invalid guest note header");
1737                 g_free(s->guest_note);
1738                 s->guest_note = NULL;
1739             } else {
1740                 vmcoreinfo_update_phys_base(s);
1741                 s->note_size += s->guest_note_size;
1742             }
1743         }
1744     }
1745 
1746     /* get memory mapping */
1747     if (paging) {
1748         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1749         if (*errp) {
1750             goto cleanup;
1751         }
1752     } else {
1753         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1754     }
1755 
1756     s->nr_cpus = nr_cpus;
1757 
1758     get_max_mapnr(s);
1759 
1760     uint64_t tmp;
1761     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1762                        s->dump_info.page_size);
1763     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1764 
1765     /* init for kdump-compressed format */
1766     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1767         switch (format) {
1768         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1769             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1770             break;
1771 
1772         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1773 #ifdef CONFIG_LZO
1774             if (lzo_init() != LZO_E_OK) {
1775                 error_setg(errp, "failed to initialize the LZO library");
1776                 goto cleanup;
1777             }
1778 #endif
1779             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1780             break;
1781 
1782         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1783             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1784             break;
1785 
1786         default:
1787             s->flag_compress = 0;
1788         }
1789 
1790         return;
1791     }
1792 
1793     if (s->has_filter) {
1794         memory_mapping_filter(&s->list, s->begin, s->length);
1795     }
1796 
1797     /*
1798      * calculate phdr_num
1799      *
1800      * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1801      */
1802     s->phdr_num = 1; /* PT_NOTE */
1803     if (s->list.num < UINT16_MAX - 2) {
1804         s->shdr_num = 0;
1805         s->phdr_num += s->list.num;
1806     } else {
1807         /* sh_info of section 0 holds the real number of phdrs */
1808         s->shdr_num = 1;
1809 
1810         /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1811         if (s->list.num <= UINT32_MAX - 1) {
1812             s->phdr_num += s->list.num;
1813         } else {
1814             s->phdr_num = UINT32_MAX;
1815         }
1816     }
1817 
1818     if (dump_is_64bit(s)) {
1819         s->phdr_offset = sizeof(Elf64_Ehdr);
1820         s->shdr_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
1821         s->note_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
1822         s->memory_offset = s->note_offset + s->note_size;
1823     } else {
1824 
1825         s->phdr_offset = sizeof(Elf32_Ehdr);
1826         s->shdr_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
1827         s->note_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
1828         s->memory_offset = s->note_offset + s->note_size;
1829     }
1830 
1831     return;
1832 
1833 cleanup:
1834     dump_cleanup(s);
1835 }
1836 
1837 /* this operation might be time consuming. */
1838 static void dump_process(DumpState *s, Error **errp)
1839 {
1840     ERRP_GUARD();
1841     DumpQueryResult *result = NULL;
1842 
1843     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1844 #ifdef TARGET_X86_64
1845         create_win_dump(s, errp);
1846 #endif
1847     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1848         create_kdump_vmcore(s, errp);
1849     } else {
1850         create_vmcore(s, errp);
1851     }
1852 
1853     /* make sure status is written after written_size updates */
1854     smp_wmb();
1855     qatomic_set(&s->status,
1856                (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1857 
1858     /* send DUMP_COMPLETED message (unconditionally) */
1859     result = qmp_query_dump(NULL);
1860     /* should never fail */
1861     assert(result);
1862     qapi_event_send_dump_completed(result, !!*errp, (*errp ?
1863                                                      error_get_pretty(*errp) : NULL));
1864     qapi_free_DumpQueryResult(result);
1865 
1866     dump_cleanup(s);
1867 }
1868 
1869 static void *dump_thread(void *data)
1870 {
1871     DumpState *s = (DumpState *)data;
1872     dump_process(s, NULL);
1873     return NULL;
1874 }
1875 
1876 DumpQueryResult *qmp_query_dump(Error **errp)
1877 {
1878     DumpQueryResult *result = g_new(DumpQueryResult, 1);
1879     DumpState *state = &dump_state_global;
1880     result->status = qatomic_read(&state->status);
1881     /* make sure we are reading status and written_size in order */
1882     smp_rmb();
1883     result->completed = state->written_size;
1884     result->total = state->total_size;
1885     return result;
1886 }
1887 
1888 void qmp_dump_guest_memory(bool paging, const char *file,
1889                            bool has_detach, bool detach,
1890                            bool has_begin, int64_t begin, bool has_length,
1891                            int64_t length, bool has_format,
1892                            DumpGuestMemoryFormat format, Error **errp)
1893 {
1894     ERRP_GUARD();
1895     const char *p;
1896     int fd = -1;
1897     DumpState *s;
1898     bool detach_p = false;
1899 
1900     if (runstate_check(RUN_STATE_INMIGRATE)) {
1901         error_setg(errp, "Dump not allowed during incoming migration.");
1902         return;
1903     }
1904 
1905     /* if there is a dump in background, we should wait until the dump
1906      * finished */
1907     if (qemu_system_dump_in_progress()) {
1908         error_setg(errp, "There is a dump in process, please wait.");
1909         return;
1910     }
1911 
1912     /*
1913      * kdump-compressed format need the whole memory dumped, so paging or
1914      * filter is not supported here.
1915      */
1916     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1917         (paging || has_begin || has_length)) {
1918         error_setg(errp, "kdump-compressed format doesn't support paging or "
1919                          "filter");
1920         return;
1921     }
1922     if (has_begin && !has_length) {
1923         error_setg(errp, QERR_MISSING_PARAMETER, "length");
1924         return;
1925     }
1926     if (!has_begin && has_length) {
1927         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1928         return;
1929     }
1930     if (has_detach) {
1931         detach_p = detach;
1932     }
1933 
1934     /* check whether lzo/snappy is supported */
1935 #ifndef CONFIG_LZO
1936     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1937         error_setg(errp, "kdump-lzo is not available now");
1938         return;
1939     }
1940 #endif
1941 
1942 #ifndef CONFIG_SNAPPY
1943     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1944         error_setg(errp, "kdump-snappy is not available now");
1945         return;
1946     }
1947 #endif
1948 
1949 #ifndef TARGET_X86_64
1950     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1951         error_setg(errp, "Windows dump is only available for x86-64");
1952         return;
1953     }
1954 #endif
1955 
1956 #if !defined(WIN32)
1957     if (strstart(file, "fd:", &p)) {
1958         fd = monitor_get_fd(monitor_cur(), p, errp);
1959         if (fd == -1) {
1960             return;
1961         }
1962     }
1963 #endif
1964 
1965     if  (strstart(file, "file:", &p)) {
1966         fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1967         if (fd < 0) {
1968             error_setg_file_open(errp, errno, p);
1969             return;
1970         }
1971     }
1972 
1973     if (fd == -1) {
1974         error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
1975         return;
1976     }
1977 
1978     if (!dump_migration_blocker) {
1979         error_setg(&dump_migration_blocker,
1980                    "Live migration disabled: dump-guest-memory in progress");
1981     }
1982 
1983     /*
1984      * Allows even for -only-migratable, but forbid migration during the
1985      * process of dump guest memory.
1986      */
1987     if (migrate_add_blocker_internal(dump_migration_blocker, errp)) {
1988         /* Remember to release the fd before passing it over to dump state */
1989         close(fd);
1990         return;
1991     }
1992 
1993     s = &dump_state_global;
1994     dump_state_prepare(s);
1995 
1996     dump_init(s, fd, has_format, format, paging, has_begin,
1997               begin, length, errp);
1998     if (*errp) {
1999         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2000         return;
2001     }
2002 
2003     if (detach_p) {
2004         /* detached dump */
2005         s->detached = true;
2006         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2007                            s, QEMU_THREAD_DETACHED);
2008     } else {
2009         /* sync dump */
2010         dump_process(s, errp);
2011     }
2012 }
2013 
2014 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2015 {
2016     DumpGuestMemoryCapability *cap =
2017                                   g_new0(DumpGuestMemoryCapability, 1);
2018     DumpGuestMemoryFormatList **tail = &cap->formats;
2019 
2020     /* elf is always available */
2021     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2022 
2023     /* kdump-zlib is always available */
2024     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2025 
2026     /* add new item if kdump-lzo is available */
2027 #ifdef CONFIG_LZO
2028     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2029 #endif
2030 
2031     /* add new item if kdump-snappy is available */
2032 #ifdef CONFIG_SNAPPY
2033     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2034 #endif
2035 
2036     /* Windows dump is available only if target is x86_64 */
2037 #ifdef TARGET_X86_64
2038     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2039 #endif
2040 
2041     return cap;
2042 }
2043