xref: /openbmc/qemu/dump/dump.c (revision 5242ef88)
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-common.h"
16 #include "qemu/cutils.h"
17 #include "elf.h"
18 #include "exec/hwaddr.h"
19 #include "monitor/monitor.h"
20 #include "sysemu/kvm.h"
21 #include "sysemu/dump.h"
22 #include "sysemu/memory_mapping.h"
23 #include "sysemu/runstate.h"
24 #include "sysemu/cpus.h"
25 #include "qapi/error.h"
26 #include "qapi/qapi-commands-dump.h"
27 #include "qapi/qapi-events-dump.h"
28 #include "qapi/qmp/qerror.h"
29 #include "qemu/error-report.h"
30 #include "qemu/main-loop.h"
31 #include "hw/misc/vmcoreinfo.h"
32 #include "migration/blocker.h"
33 
34 #ifdef TARGET_X86_64
35 #include "win_dump.h"
36 #endif
37 
38 #include <zlib.h>
39 #ifdef CONFIG_LZO
40 #include <lzo/lzo1x.h>
41 #endif
42 #ifdef CONFIG_SNAPPY
43 #include <snappy-c.h>
44 #endif
45 #ifndef ELF_MACHINE_UNAME
46 #define ELF_MACHINE_UNAME "Unknown"
47 #endif
48 
49 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
50 
51 static Error *dump_migration_blocker;
52 
53 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size)   \
54     ((DIV_ROUND_UP((hdr_size), 4) +                     \
55       DIV_ROUND_UP((name_size), 4) +                    \
56       DIV_ROUND_UP((desc_size), 4)) * 4)
57 
58 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
59 {
60     if (s->dump_info.d_endian == ELFDATA2LSB) {
61         val = cpu_to_le16(val);
62     } else {
63         val = cpu_to_be16(val);
64     }
65 
66     return val;
67 }
68 
69 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
70 {
71     if (s->dump_info.d_endian == ELFDATA2LSB) {
72         val = cpu_to_le32(val);
73     } else {
74         val = cpu_to_be32(val);
75     }
76 
77     return val;
78 }
79 
80 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
81 {
82     if (s->dump_info.d_endian == ELFDATA2LSB) {
83         val = cpu_to_le64(val);
84     } else {
85         val = cpu_to_be64(val);
86     }
87 
88     return val;
89 }
90 
91 static int dump_cleanup(DumpState *s)
92 {
93     guest_phys_blocks_free(&s->guest_phys_blocks);
94     memory_mapping_list_free(&s->list);
95     close(s->fd);
96     g_free(s->guest_note);
97     s->guest_note = NULL;
98     if (s->resume) {
99         if (s->detached) {
100             qemu_mutex_lock_iothread();
101         }
102         vm_start();
103         if (s->detached) {
104             qemu_mutex_unlock_iothread();
105         }
106     }
107     migrate_del_blocker(dump_migration_blocker);
108 
109     return 0;
110 }
111 
112 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
113 {
114     DumpState *s = opaque;
115     size_t written_size;
116 
117     written_size = qemu_write_full(s->fd, buf, size);
118     if (written_size != size) {
119         return -errno;
120     }
121 
122     return 0;
123 }
124 
125 static void write_elf64_header(DumpState *s, Error **errp)
126 {
127     Elf64_Ehdr elf_header;
128     int ret;
129 
130     memset(&elf_header, 0, sizeof(Elf64_Ehdr));
131     memcpy(&elf_header, ELFMAG, SELFMAG);
132     elf_header.e_ident[EI_CLASS] = ELFCLASS64;
133     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
134     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
135     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
136     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
137     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
138     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
139     elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
140     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
141     elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
142     if (s->have_section) {
143         uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
144 
145         elf_header.e_shoff = cpu_to_dump64(s, shoff);
146         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
147         elf_header.e_shnum = cpu_to_dump16(s, 1);
148     }
149 
150     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
151     if (ret < 0) {
152         error_setg_errno(errp, -ret, "dump: failed to write elf header");
153     }
154 }
155 
156 static void write_elf32_header(DumpState *s, Error **errp)
157 {
158     Elf32_Ehdr elf_header;
159     int ret;
160 
161     memset(&elf_header, 0, sizeof(Elf32_Ehdr));
162     memcpy(&elf_header, ELFMAG, SELFMAG);
163     elf_header.e_ident[EI_CLASS] = ELFCLASS32;
164     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
165     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
166     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
167     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
168     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
169     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
170     elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
171     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
172     elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
173     if (s->have_section) {
174         uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
175 
176         elf_header.e_shoff = cpu_to_dump32(s, shoff);
177         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
178         elf_header.e_shnum = cpu_to_dump16(s, 1);
179     }
180 
181     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
182     if (ret < 0) {
183         error_setg_errno(errp, -ret, "dump: failed to write elf header");
184     }
185 }
186 
187 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
188                              int phdr_index, hwaddr offset,
189                              hwaddr filesz, Error **errp)
190 {
191     Elf64_Phdr phdr;
192     int ret;
193 
194     memset(&phdr, 0, sizeof(Elf64_Phdr));
195     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
196     phdr.p_offset = cpu_to_dump64(s, offset);
197     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
198     phdr.p_filesz = cpu_to_dump64(s, filesz);
199     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
200     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
201 
202     assert(memory_mapping->length >= filesz);
203 
204     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
205     if (ret < 0) {
206         error_setg_errno(errp, -ret,
207                          "dump: failed to write program header table");
208     }
209 }
210 
211 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
212                              int phdr_index, hwaddr offset,
213                              hwaddr filesz, Error **errp)
214 {
215     Elf32_Phdr phdr;
216     int ret;
217 
218     memset(&phdr, 0, sizeof(Elf32_Phdr));
219     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
220     phdr.p_offset = cpu_to_dump32(s, offset);
221     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
222     phdr.p_filesz = cpu_to_dump32(s, filesz);
223     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
224     phdr.p_vaddr =
225         cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
226 
227     assert(memory_mapping->length >= filesz);
228 
229     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
230     if (ret < 0) {
231         error_setg_errno(errp, -ret,
232                          "dump: failed to write program header table");
233     }
234 }
235 
236 static void write_elf64_note(DumpState *s, Error **errp)
237 {
238     Elf64_Phdr phdr;
239     hwaddr begin = s->memory_offset - s->note_size;
240     int ret;
241 
242     memset(&phdr, 0, sizeof(Elf64_Phdr));
243     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
244     phdr.p_offset = cpu_to_dump64(s, begin);
245     phdr.p_paddr = 0;
246     phdr.p_filesz = cpu_to_dump64(s, s->note_size);
247     phdr.p_memsz = cpu_to_dump64(s, s->note_size);
248     phdr.p_vaddr = 0;
249 
250     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
251     if (ret < 0) {
252         error_setg_errno(errp, -ret,
253                          "dump: failed to write program header table");
254     }
255 }
256 
257 static inline int cpu_index(CPUState *cpu)
258 {
259     return cpu->cpu_index + 1;
260 }
261 
262 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
263                              Error **errp)
264 {
265     int ret;
266 
267     if (s->guest_note) {
268         ret = f(s->guest_note, s->guest_note_size, s);
269         if (ret < 0) {
270             error_setg(errp, "dump: failed to write guest note");
271         }
272     }
273 }
274 
275 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
276                               Error **errp)
277 {
278     CPUState *cpu;
279     int ret;
280     int id;
281 
282     CPU_FOREACH(cpu) {
283         id = cpu_index(cpu);
284         ret = cpu_write_elf64_note(f, cpu, id, s);
285         if (ret < 0) {
286             error_setg(errp, "dump: failed to write elf notes");
287             return;
288         }
289     }
290 
291     CPU_FOREACH(cpu) {
292         ret = cpu_write_elf64_qemunote(f, cpu, s);
293         if (ret < 0) {
294             error_setg(errp, "dump: failed to write CPU status");
295             return;
296         }
297     }
298 
299     write_guest_note(f, s, errp);
300 }
301 
302 static void write_elf32_note(DumpState *s, Error **errp)
303 {
304     hwaddr begin = s->memory_offset - s->note_size;
305     Elf32_Phdr phdr;
306     int ret;
307 
308     memset(&phdr, 0, sizeof(Elf32_Phdr));
309     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
310     phdr.p_offset = cpu_to_dump32(s, begin);
311     phdr.p_paddr = 0;
312     phdr.p_filesz = cpu_to_dump32(s, s->note_size);
313     phdr.p_memsz = cpu_to_dump32(s, s->note_size);
314     phdr.p_vaddr = 0;
315 
316     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
317     if (ret < 0) {
318         error_setg_errno(errp, -ret,
319                          "dump: failed to write program header table");
320     }
321 }
322 
323 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
324                               Error **errp)
325 {
326     CPUState *cpu;
327     int ret;
328     int id;
329 
330     CPU_FOREACH(cpu) {
331         id = cpu_index(cpu);
332         ret = cpu_write_elf32_note(f, cpu, id, s);
333         if (ret < 0) {
334             error_setg(errp, "dump: failed to write elf notes");
335             return;
336         }
337     }
338 
339     CPU_FOREACH(cpu) {
340         ret = cpu_write_elf32_qemunote(f, cpu, s);
341         if (ret < 0) {
342             error_setg(errp, "dump: failed to write CPU status");
343             return;
344         }
345     }
346 
347     write_guest_note(f, s, errp);
348 }
349 
350 static void write_elf_section(DumpState *s, int type, Error **errp)
351 {
352     Elf32_Shdr shdr32;
353     Elf64_Shdr shdr64;
354     int shdr_size;
355     void *shdr;
356     int ret;
357 
358     if (type == 0) {
359         shdr_size = sizeof(Elf32_Shdr);
360         memset(&shdr32, 0, shdr_size);
361         shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
362         shdr = &shdr32;
363     } else {
364         shdr_size = sizeof(Elf64_Shdr);
365         memset(&shdr64, 0, shdr_size);
366         shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
367         shdr = &shdr64;
368     }
369 
370     ret = fd_write_vmcore(shdr, shdr_size, s);
371     if (ret < 0) {
372         error_setg_errno(errp, -ret,
373                          "dump: failed to write section header table");
374     }
375 }
376 
377 static void write_data(DumpState *s, void *buf, int length, Error **errp)
378 {
379     int ret;
380 
381     ret = fd_write_vmcore(buf, length, s);
382     if (ret < 0) {
383         error_setg_errno(errp, -ret, "dump: failed to save memory");
384     } else {
385         s->written_size += length;
386     }
387 }
388 
389 /* write the memory to vmcore. 1 page per I/O. */
390 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
391                          int64_t size, Error **errp)
392 {
393     int64_t i;
394     Error *local_err = NULL;
395 
396     for (i = 0; i < size / s->dump_info.page_size; i++) {
397         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
398                    s->dump_info.page_size, &local_err);
399         if (local_err) {
400             error_propagate(errp, local_err);
401             return;
402         }
403     }
404 
405     if ((size % s->dump_info.page_size) != 0) {
406         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
407                    size % s->dump_info.page_size, &local_err);
408         if (local_err) {
409             error_propagate(errp, local_err);
410             return;
411         }
412     }
413 }
414 
415 /* get the memory's offset and size in the vmcore */
416 static void get_offset_range(hwaddr phys_addr,
417                              ram_addr_t mapping_length,
418                              DumpState *s,
419                              hwaddr *p_offset,
420                              hwaddr *p_filesz)
421 {
422     GuestPhysBlock *block;
423     hwaddr offset = s->memory_offset;
424     int64_t size_in_block, start;
425 
426     /* When the memory is not stored into vmcore, offset will be -1 */
427     *p_offset = -1;
428     *p_filesz = 0;
429 
430     if (s->has_filter) {
431         if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
432             return;
433         }
434     }
435 
436     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
437         if (s->has_filter) {
438             if (block->target_start >= s->begin + s->length ||
439                 block->target_end <= s->begin) {
440                 /* This block is out of the range */
441                 continue;
442             }
443 
444             if (s->begin <= block->target_start) {
445                 start = block->target_start;
446             } else {
447                 start = s->begin;
448             }
449 
450             size_in_block = block->target_end - start;
451             if (s->begin + s->length < block->target_end) {
452                 size_in_block -= block->target_end - (s->begin + s->length);
453             }
454         } else {
455             start = block->target_start;
456             size_in_block = block->target_end - block->target_start;
457         }
458 
459         if (phys_addr >= start && phys_addr < start + size_in_block) {
460             *p_offset = phys_addr - start + offset;
461 
462             /* The offset range mapped from the vmcore file must not spill over
463              * the GuestPhysBlock, clamp it. The rest of the mapping will be
464              * zero-filled in memory at load time; see
465              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
466              */
467             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
468                         mapping_length :
469                         size_in_block - (phys_addr - start);
470             return;
471         }
472 
473         offset += size_in_block;
474     }
475 }
476 
477 static void write_elf_loads(DumpState *s, Error **errp)
478 {
479     hwaddr offset, filesz;
480     MemoryMapping *memory_mapping;
481     uint32_t phdr_index = 1;
482     uint32_t max_index;
483     Error *local_err = NULL;
484 
485     if (s->have_section) {
486         max_index = s->sh_info;
487     } else {
488         max_index = s->phdr_num;
489     }
490 
491     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
492         get_offset_range(memory_mapping->phys_addr,
493                          memory_mapping->length,
494                          s, &offset, &filesz);
495         if (s->dump_info.d_class == ELFCLASS64) {
496             write_elf64_load(s, memory_mapping, phdr_index++, offset,
497                              filesz, &local_err);
498         } else {
499             write_elf32_load(s, memory_mapping, phdr_index++, offset,
500                              filesz, &local_err);
501         }
502 
503         if (local_err) {
504             error_propagate(errp, local_err);
505             return;
506         }
507 
508         if (phdr_index >= max_index) {
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     Error *local_err = NULL;
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 (s->dump_info.d_class == ELFCLASS64) {
545         write_elf64_header(s, &local_err);
546     } else {
547         write_elf32_header(s, &local_err);
548     }
549     if (local_err) {
550         error_propagate(errp, local_err);
551         return;
552     }
553 
554     if (s->dump_info.d_class == ELFCLASS64) {
555         /* write PT_NOTE to vmcore */
556         write_elf64_note(s, &local_err);
557         if (local_err) {
558             error_propagate(errp, local_err);
559             return;
560         }
561 
562         /* write all PT_LOAD to vmcore */
563         write_elf_loads(s, &local_err);
564         if (local_err) {
565             error_propagate(errp, local_err);
566             return;
567         }
568 
569         /* write section to vmcore */
570         if (s->have_section) {
571             write_elf_section(s, 1, &local_err);
572             if (local_err) {
573                 error_propagate(errp, local_err);
574                 return;
575             }
576         }
577 
578         /* write notes to vmcore */
579         write_elf64_notes(fd_write_vmcore, s, &local_err);
580         if (local_err) {
581             error_propagate(errp, local_err);
582             return;
583         }
584     } else {
585         /* write PT_NOTE to vmcore */
586         write_elf32_note(s, &local_err);
587         if (local_err) {
588             error_propagate(errp, local_err);
589             return;
590         }
591 
592         /* write all PT_LOAD to vmcore */
593         write_elf_loads(s, &local_err);
594         if (local_err) {
595             error_propagate(errp, local_err);
596             return;
597         }
598 
599         /* write section to vmcore */
600         if (s->have_section) {
601             write_elf_section(s, 0, &local_err);
602             if (local_err) {
603                 error_propagate(errp, local_err);
604                 return;
605             }
606         }
607 
608         /* write notes to vmcore */
609         write_elf32_notes(fd_write_vmcore, s, &local_err);
610         if (local_err) {
611             error_propagate(errp, local_err);
612             return;
613         }
614     }
615 }
616 
617 static int get_next_block(DumpState *s, GuestPhysBlock *block)
618 {
619     while (1) {
620         block = QTAILQ_NEXT(block, next);
621         if (!block) {
622             /* no more block */
623             return 1;
624         }
625 
626         s->start = 0;
627         s->next_block = block;
628         if (s->has_filter) {
629             if (block->target_start >= s->begin + s->length ||
630                 block->target_end <= s->begin) {
631                 /* This block is out of the range */
632                 continue;
633             }
634 
635             if (s->begin > block->target_start) {
636                 s->start = s->begin - block->target_start;
637             }
638         }
639 
640         return 0;
641     }
642 }
643 
644 /* write all memory to vmcore */
645 static void dump_iterate(DumpState *s, Error **errp)
646 {
647     GuestPhysBlock *block;
648     int64_t size;
649     Error *local_err = NULL;
650 
651     do {
652         block = s->next_block;
653 
654         size = block->target_end - block->target_start;
655         if (s->has_filter) {
656             size -= s->start;
657             if (s->begin + s->length < block->target_end) {
658                 size -= block->target_end - (s->begin + s->length);
659             }
660         }
661         write_memory(s, block, s->start, size, &local_err);
662         if (local_err) {
663             error_propagate(errp, local_err);
664             return;
665         }
666 
667     } while (!get_next_block(s, block));
668 }
669 
670 static void create_vmcore(DumpState *s, Error **errp)
671 {
672     Error *local_err = NULL;
673 
674     dump_begin(s, &local_err);
675     if (local_err) {
676         error_propagate(errp, local_err);
677         return;
678     }
679 
680     dump_iterate(s, errp);
681 }
682 
683 static int write_start_flat_header(int fd)
684 {
685     MakedumpfileHeader *mh;
686     int ret = 0;
687 
688     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
689     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
690 
691     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
692            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
693 
694     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
695     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
696 
697     size_t written_size;
698     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
699     if (written_size != MAX_SIZE_MDF_HEADER) {
700         ret = -1;
701     }
702 
703     g_free(mh);
704     return ret;
705 }
706 
707 static int write_end_flat_header(int fd)
708 {
709     MakedumpfileDataHeader mdh;
710 
711     mdh.offset = END_FLAG_FLAT_HEADER;
712     mdh.buf_size = END_FLAG_FLAT_HEADER;
713 
714     size_t written_size;
715     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
716     if (written_size != sizeof(mdh)) {
717         return -1;
718     }
719 
720     return 0;
721 }
722 
723 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
724 {
725     size_t written_size;
726     MakedumpfileDataHeader mdh;
727 
728     mdh.offset = cpu_to_be64(offset);
729     mdh.buf_size = cpu_to_be64(size);
730 
731     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
732     if (written_size != sizeof(mdh)) {
733         return -1;
734     }
735 
736     written_size = qemu_write_full(fd, buf, size);
737     if (written_size != size) {
738         return -1;
739     }
740 
741     return 0;
742 }
743 
744 static int buf_write_note(const void *buf, size_t size, void *opaque)
745 {
746     DumpState *s = opaque;
747 
748     /* note_buf is not enough */
749     if (s->note_buf_offset + size > s->note_size) {
750         return -1;
751     }
752 
753     memcpy(s->note_buf + s->note_buf_offset, buf, size);
754 
755     s->note_buf_offset += size;
756 
757     return 0;
758 }
759 
760 /*
761  * This function retrieves various sizes from an elf header.
762  *
763  * @note has to be a valid ELF note. The return sizes are unmodified
764  * (not padded or rounded up to be multiple of 4).
765  */
766 static void get_note_sizes(DumpState *s, const void *note,
767                            uint64_t *note_head_size,
768                            uint64_t *name_size,
769                            uint64_t *desc_size)
770 {
771     uint64_t note_head_sz;
772     uint64_t name_sz;
773     uint64_t desc_sz;
774 
775     if (s->dump_info.d_class == ELFCLASS64) {
776         const Elf64_Nhdr *hdr = note;
777         note_head_sz = sizeof(Elf64_Nhdr);
778         name_sz = tswap64(hdr->n_namesz);
779         desc_sz = tswap64(hdr->n_descsz);
780     } else {
781         const Elf32_Nhdr *hdr = note;
782         note_head_sz = sizeof(Elf32_Nhdr);
783         name_sz = tswap32(hdr->n_namesz);
784         desc_sz = tswap32(hdr->n_descsz);
785     }
786 
787     if (note_head_size) {
788         *note_head_size = note_head_sz;
789     }
790     if (name_size) {
791         *name_size = name_sz;
792     }
793     if (desc_size) {
794         *desc_size = desc_sz;
795     }
796 }
797 
798 static bool note_name_equal(DumpState *s,
799                             const uint8_t *note, const char *name)
800 {
801     int len = strlen(name) + 1;
802     uint64_t head_size, name_size;
803 
804     get_note_sizes(s, note, &head_size, &name_size, NULL);
805     head_size = ROUND_UP(head_size, 4);
806 
807     return name_size == len && memcmp(note + head_size, name, len) == 0;
808 }
809 
810 /* write common header, sub header and elf note to vmcore */
811 static void create_header32(DumpState *s, Error **errp)
812 {
813     DiskDumpHeader32 *dh = NULL;
814     KdumpSubHeader32 *kh = NULL;
815     size_t size;
816     uint32_t block_size;
817     uint32_t sub_hdr_size;
818     uint32_t bitmap_blocks;
819     uint32_t status = 0;
820     uint64_t offset_note;
821     Error *local_err = NULL;
822 
823     /* write common header, the version of kdump-compressed format is 6th */
824     size = sizeof(DiskDumpHeader32);
825     dh = g_malloc0(size);
826 
827     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
828     dh->header_version = cpu_to_dump32(s, 6);
829     block_size = s->dump_info.page_size;
830     dh->block_size = cpu_to_dump32(s, block_size);
831     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
832     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
833     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
834     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
835     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
836     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
837     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
838     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
839     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
840 
841     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
842         status |= DUMP_DH_COMPRESSED_ZLIB;
843     }
844 #ifdef CONFIG_LZO
845     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
846         status |= DUMP_DH_COMPRESSED_LZO;
847     }
848 #endif
849 #ifdef CONFIG_SNAPPY
850     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
851         status |= DUMP_DH_COMPRESSED_SNAPPY;
852     }
853 #endif
854     dh->status = cpu_to_dump32(s, status);
855 
856     if (write_buffer(s->fd, 0, dh, size) < 0) {
857         error_setg(errp, "dump: failed to write disk dump header");
858         goto out;
859     }
860 
861     /* write sub header */
862     size = sizeof(KdumpSubHeader32);
863     kh = g_malloc0(size);
864 
865     /* 64bit max_mapnr_64 */
866     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
867     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
868     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
869 
870     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
871     if (s->guest_note &&
872         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
873         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
874 
875         get_note_sizes(s, s->guest_note,
876                        &hsize, &name_size, &size_vmcoreinfo_desc);
877         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
878             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
879         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
880         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
881     }
882 
883     kh->offset_note = cpu_to_dump64(s, offset_note);
884     kh->note_size = cpu_to_dump32(s, s->note_size);
885 
886     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
887                      block_size, kh, size) < 0) {
888         error_setg(errp, "dump: failed to write kdump sub header");
889         goto out;
890     }
891 
892     /* write note */
893     s->note_buf = g_malloc0(s->note_size);
894     s->note_buf_offset = 0;
895 
896     /* use s->note_buf to store notes temporarily */
897     write_elf32_notes(buf_write_note, s, &local_err);
898     if (local_err) {
899         error_propagate(errp, local_err);
900         goto out;
901     }
902     if (write_buffer(s->fd, offset_note, s->note_buf,
903                      s->note_size) < 0) {
904         error_setg(errp, "dump: failed to write notes");
905         goto out;
906     }
907 
908     /* get offset of dump_bitmap */
909     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
910                              block_size;
911 
912     /* get offset of page */
913     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
914                      block_size;
915 
916 out:
917     g_free(dh);
918     g_free(kh);
919     g_free(s->note_buf);
920 }
921 
922 /* write common header, sub header and elf note to vmcore */
923 static void create_header64(DumpState *s, Error **errp)
924 {
925     DiskDumpHeader64 *dh = NULL;
926     KdumpSubHeader64 *kh = NULL;
927     size_t size;
928     uint32_t block_size;
929     uint32_t sub_hdr_size;
930     uint32_t bitmap_blocks;
931     uint32_t status = 0;
932     uint64_t offset_note;
933     Error *local_err = NULL;
934 
935     /* write common header, the version of kdump-compressed format is 6th */
936     size = sizeof(DiskDumpHeader64);
937     dh = g_malloc0(size);
938 
939     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
940     dh->header_version = cpu_to_dump32(s, 6);
941     block_size = s->dump_info.page_size;
942     dh->block_size = cpu_to_dump32(s, block_size);
943     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
944     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
945     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
946     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
947     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
948     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
949     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
950     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
951     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
952 
953     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
954         status |= DUMP_DH_COMPRESSED_ZLIB;
955     }
956 #ifdef CONFIG_LZO
957     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
958         status |= DUMP_DH_COMPRESSED_LZO;
959     }
960 #endif
961 #ifdef CONFIG_SNAPPY
962     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
963         status |= DUMP_DH_COMPRESSED_SNAPPY;
964     }
965 #endif
966     dh->status = cpu_to_dump32(s, status);
967 
968     if (write_buffer(s->fd, 0, dh, size) < 0) {
969         error_setg(errp, "dump: failed to write disk dump header");
970         goto out;
971     }
972 
973     /* write sub header */
974     size = sizeof(KdumpSubHeader64);
975     kh = g_malloc0(size);
976 
977     /* 64bit max_mapnr_64 */
978     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
979     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
980     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
981 
982     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
983     if (s->guest_note &&
984         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
985         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
986 
987         get_note_sizes(s, s->guest_note,
988                        &hsize, &name_size, &size_vmcoreinfo_desc);
989         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
990             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
991         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
992         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
993     }
994 
995     kh->offset_note = cpu_to_dump64(s, offset_note);
996     kh->note_size = cpu_to_dump64(s, s->note_size);
997 
998     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
999                      block_size, kh, size) < 0) {
1000         error_setg(errp, "dump: failed to write kdump sub header");
1001         goto out;
1002     }
1003 
1004     /* write note */
1005     s->note_buf = g_malloc0(s->note_size);
1006     s->note_buf_offset = 0;
1007 
1008     /* use s->note_buf to store notes temporarily */
1009     write_elf64_notes(buf_write_note, s, &local_err);
1010     if (local_err) {
1011         error_propagate(errp, local_err);
1012         goto out;
1013     }
1014 
1015     if (write_buffer(s->fd, offset_note, s->note_buf,
1016                      s->note_size) < 0) {
1017         error_setg(errp, "dump: failed to write notes");
1018         goto out;
1019     }
1020 
1021     /* get offset of dump_bitmap */
1022     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1023                              block_size;
1024 
1025     /* get offset of page */
1026     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1027                      block_size;
1028 
1029 out:
1030     g_free(dh);
1031     g_free(kh);
1032     g_free(s->note_buf);
1033 }
1034 
1035 static void write_dump_header(DumpState *s, Error **errp)
1036 {
1037     if (s->dump_info.d_class == ELFCLASS32) {
1038         create_header32(s, errp);
1039     } else {
1040         create_header64(s, errp);
1041     }
1042 }
1043 
1044 static size_t dump_bitmap_get_bufsize(DumpState *s)
1045 {
1046     return s->dump_info.page_size;
1047 }
1048 
1049 /*
1050  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1051  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1052  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1053  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1054  * vmcore, ie. synchronizing un-sync bit into vmcore.
1055  */
1056 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1057                            uint8_t *buf, DumpState *s)
1058 {
1059     off_t old_offset, new_offset;
1060     off_t offset_bitmap1, offset_bitmap2;
1061     uint32_t byte, bit;
1062     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1063     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1064 
1065     /* should not set the previous place */
1066     assert(last_pfn <= pfn);
1067 
1068     /*
1069      * if the bit needed to be set is not cached in buf, flush the data in buf
1070      * to vmcore firstly.
1071      * making new_offset be bigger than old_offset can also sync remained data
1072      * into vmcore.
1073      */
1074     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1075     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1076 
1077     while (old_offset < new_offset) {
1078         /* calculate the offset and write dump_bitmap */
1079         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1080         if (write_buffer(s->fd, offset_bitmap1, buf,
1081                          bitmap_bufsize) < 0) {
1082             return -1;
1083         }
1084 
1085         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1086         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1087                          old_offset;
1088         if (write_buffer(s->fd, offset_bitmap2, buf,
1089                          bitmap_bufsize) < 0) {
1090             return -1;
1091         }
1092 
1093         memset(buf, 0, bitmap_bufsize);
1094         old_offset += bitmap_bufsize;
1095     }
1096 
1097     /* get the exact place of the bit in the buf, and set it */
1098     byte = (pfn % bits_per_buf) / CHAR_BIT;
1099     bit = (pfn % bits_per_buf) % CHAR_BIT;
1100     if (value) {
1101         buf[byte] |= 1u << bit;
1102     } else {
1103         buf[byte] &= ~(1u << bit);
1104     }
1105 
1106     return 0;
1107 }
1108 
1109 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1110 {
1111     int target_page_shift = ctz32(s->dump_info.page_size);
1112 
1113     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1114 }
1115 
1116 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1117 {
1118     int target_page_shift = ctz32(s->dump_info.page_size);
1119 
1120     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1121 }
1122 
1123 /*
1124  * exam every page and return the page frame number and the address of the page.
1125  * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1126  * blocks, so block->target_start and block->target_end should be interal
1127  * multiples of the target page size.
1128  */
1129 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1130                           uint8_t **bufptr, DumpState *s)
1131 {
1132     GuestPhysBlock *block = *blockptr;
1133     hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1134     uint8_t *buf;
1135 
1136     /* block == NULL means the start of the iteration */
1137     if (!block) {
1138         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1139         *blockptr = block;
1140         assert((block->target_start & ~target_page_mask) == 0);
1141         assert((block->target_end & ~target_page_mask) == 0);
1142         *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1143         if (bufptr) {
1144             *bufptr = block->host_addr;
1145         }
1146         return true;
1147     }
1148 
1149     *pfnptr = *pfnptr + 1;
1150     addr = dump_pfn_to_paddr(s, *pfnptr);
1151 
1152     if ((addr >= block->target_start) &&
1153         (addr + s->dump_info.page_size <= block->target_end)) {
1154         buf = block->host_addr + (addr - block->target_start);
1155     } else {
1156         /* the next page is in the next block */
1157         block = QTAILQ_NEXT(block, next);
1158         *blockptr = block;
1159         if (!block) {
1160             return false;
1161         }
1162         assert((block->target_start & ~target_page_mask) == 0);
1163         assert((block->target_end & ~target_page_mask) == 0);
1164         *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1165         buf = block->host_addr;
1166     }
1167 
1168     if (bufptr) {
1169         *bufptr = buf;
1170     }
1171 
1172     return true;
1173 }
1174 
1175 static void write_dump_bitmap(DumpState *s, Error **errp)
1176 {
1177     int ret = 0;
1178     uint64_t last_pfn, pfn;
1179     void *dump_bitmap_buf;
1180     size_t num_dumpable;
1181     GuestPhysBlock *block_iter = NULL;
1182     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1183     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1184 
1185     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1186     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1187 
1188     num_dumpable = 0;
1189     last_pfn = 0;
1190 
1191     /*
1192      * exam memory page by page, and set the bit in dump_bitmap corresponded
1193      * to the existing page.
1194      */
1195     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1196         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1197         if (ret < 0) {
1198             error_setg(errp, "dump: failed to set dump_bitmap");
1199             goto out;
1200         }
1201 
1202         last_pfn = pfn;
1203         num_dumpable++;
1204     }
1205 
1206     /*
1207      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1208      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1209      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1210      */
1211     if (num_dumpable > 0) {
1212         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1213                               dump_bitmap_buf, s);
1214         if (ret < 0) {
1215             error_setg(errp, "dump: failed to sync dump_bitmap");
1216             goto out;
1217         }
1218     }
1219 
1220     /* number of dumpable pages that will be dumped later */
1221     s->num_dumpable = num_dumpable;
1222 
1223 out:
1224     g_free(dump_bitmap_buf);
1225 }
1226 
1227 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1228                                off_t offset)
1229 {
1230     data_cache->fd = s->fd;
1231     data_cache->data_size = 0;
1232     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1233     data_cache->buf = g_malloc0(data_cache->buf_size);
1234     data_cache->offset = offset;
1235 }
1236 
1237 static int write_cache(DataCache *dc, const void *buf, size_t size,
1238                        bool flag_sync)
1239 {
1240     /*
1241      * dc->buf_size should not be less than size, otherwise dc will never be
1242      * enough
1243      */
1244     assert(size <= dc->buf_size);
1245 
1246     /*
1247      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1248      * otherwise check if the space is enough for caching data in buf, if not,
1249      * write the data in dc->buf to dc->fd and reset dc->buf
1250      */
1251     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1252         (flag_sync && dc->data_size > 0)) {
1253         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1254             return -1;
1255         }
1256 
1257         dc->offset += dc->data_size;
1258         dc->data_size = 0;
1259     }
1260 
1261     if (!flag_sync) {
1262         memcpy(dc->buf + dc->data_size, buf, size);
1263         dc->data_size += size;
1264     }
1265 
1266     return 0;
1267 }
1268 
1269 static void free_data_cache(DataCache *data_cache)
1270 {
1271     g_free(data_cache->buf);
1272 }
1273 
1274 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1275 {
1276     switch (flag_compress) {
1277     case DUMP_DH_COMPRESSED_ZLIB:
1278         return compressBound(page_size);
1279 
1280     case DUMP_DH_COMPRESSED_LZO:
1281         /*
1282          * LZO will expand incompressible data by a little amount. Please check
1283          * the following URL to see the expansion calculation:
1284          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1285          */
1286         return page_size + page_size / 16 + 64 + 3;
1287 
1288 #ifdef CONFIG_SNAPPY
1289     case DUMP_DH_COMPRESSED_SNAPPY:
1290         return snappy_max_compressed_length(page_size);
1291 #endif
1292     }
1293     return 0;
1294 }
1295 
1296 static void write_dump_pages(DumpState *s, Error **errp)
1297 {
1298     int ret = 0;
1299     DataCache page_desc, page_data;
1300     size_t len_buf_out, size_out;
1301 #ifdef CONFIG_LZO
1302     lzo_bytep wrkmem = NULL;
1303 #endif
1304     uint8_t *buf_out = NULL;
1305     off_t offset_desc, offset_data;
1306     PageDescriptor pd, pd_zero;
1307     uint8_t *buf;
1308     GuestPhysBlock *block_iter = NULL;
1309     uint64_t pfn_iter;
1310 
1311     /* get offset of page_desc and page_data in dump file */
1312     offset_desc = s->offset_page;
1313     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1314 
1315     prepare_data_cache(&page_desc, s, offset_desc);
1316     prepare_data_cache(&page_data, s, offset_data);
1317 
1318     /* prepare buffer to store compressed data */
1319     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1320     assert(len_buf_out != 0);
1321 
1322 #ifdef CONFIG_LZO
1323     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1324 #endif
1325 
1326     buf_out = g_malloc(len_buf_out);
1327 
1328     /*
1329      * init zero page's page_desc and page_data, because every zero page
1330      * uses the same page_data
1331      */
1332     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1333     pd_zero.flags = cpu_to_dump32(s, 0);
1334     pd_zero.offset = cpu_to_dump64(s, offset_data);
1335     pd_zero.page_flags = cpu_to_dump64(s, 0);
1336     buf = g_malloc0(s->dump_info.page_size);
1337     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1338     g_free(buf);
1339     if (ret < 0) {
1340         error_setg(errp, "dump: failed to write page data (zero page)");
1341         goto out;
1342     }
1343 
1344     offset_data += s->dump_info.page_size;
1345 
1346     /*
1347      * dump memory to vmcore page by page. zero page will all be resided in the
1348      * first page of page section
1349      */
1350     while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1351         /* check zero page */
1352         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1353             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1354                               false);
1355             if (ret < 0) {
1356                 error_setg(errp, "dump: failed to write page desc");
1357                 goto out;
1358             }
1359         } else {
1360             /*
1361              * not zero page, then:
1362              * 1. compress the page
1363              * 2. write the compressed page into the cache of page_data
1364              * 3. get page desc of the compressed page and write it into the
1365              *    cache of page_desc
1366              *
1367              * only one compression format will be used here, for
1368              * s->flag_compress is set. But when compression fails to work,
1369              * we fall back to save in plaintext.
1370              */
1371              size_out = len_buf_out;
1372              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1373                     (compress2(buf_out, (uLongf *)&size_out, buf,
1374                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1375                     (size_out < s->dump_info.page_size)) {
1376                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
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 #ifdef CONFIG_LZO
1385             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1386                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1387                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1388                     (size_out < s->dump_info.page_size)) {
1389                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1390                 pd.size  = cpu_to_dump32(s, size_out);
1391 
1392                 ret = write_cache(&page_data, buf_out, size_out, false);
1393                 if (ret < 0) {
1394                     error_setg(errp, "dump: failed to write page data");
1395                     goto out;
1396                 }
1397 #endif
1398 #ifdef CONFIG_SNAPPY
1399             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1400                     (snappy_compress((char *)buf, s->dump_info.page_size,
1401                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1402                     (size_out < s->dump_info.page_size)) {
1403                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1404                 pd.size  = cpu_to_dump32(s, size_out);
1405 
1406                 ret = write_cache(&page_data, buf_out, size_out, false);
1407                 if (ret < 0) {
1408                     error_setg(errp, "dump: failed to write page data");
1409                     goto out;
1410                 }
1411 #endif
1412             } else {
1413                 /*
1414                  * fall back to save in plaintext, size_out should be
1415                  * assigned the target's page size
1416                  */
1417                 pd.flags = cpu_to_dump32(s, 0);
1418                 size_out = s->dump_info.page_size;
1419                 pd.size = cpu_to_dump32(s, size_out);
1420 
1421                 ret = write_cache(&page_data, buf,
1422                                   s->dump_info.page_size, false);
1423                 if (ret < 0) {
1424                     error_setg(errp, "dump: failed to write page data");
1425                     goto out;
1426                 }
1427             }
1428 
1429             /* get and write page desc here */
1430             pd.page_flags = cpu_to_dump64(s, 0);
1431             pd.offset = cpu_to_dump64(s, offset_data);
1432             offset_data += size_out;
1433 
1434             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1435             if (ret < 0) {
1436                 error_setg(errp, "dump: failed to write page desc");
1437                 goto out;
1438             }
1439         }
1440         s->written_size += s->dump_info.page_size;
1441     }
1442 
1443     ret = write_cache(&page_desc, NULL, 0, true);
1444     if (ret < 0) {
1445         error_setg(errp, "dump: failed to sync cache for page_desc");
1446         goto out;
1447     }
1448     ret = write_cache(&page_data, NULL, 0, true);
1449     if (ret < 0) {
1450         error_setg(errp, "dump: failed to sync cache for page_data");
1451         goto out;
1452     }
1453 
1454 out:
1455     free_data_cache(&page_desc);
1456     free_data_cache(&page_data);
1457 
1458 #ifdef CONFIG_LZO
1459     g_free(wrkmem);
1460 #endif
1461 
1462     g_free(buf_out);
1463 }
1464 
1465 static void create_kdump_vmcore(DumpState *s, Error **errp)
1466 {
1467     int ret;
1468     Error *local_err = NULL;
1469 
1470     /*
1471      * the kdump-compressed format is:
1472      *                                               File offset
1473      *  +------------------------------------------+ 0x0
1474      *  |    main header (struct disk_dump_header) |
1475      *  |------------------------------------------+ block 1
1476      *  |    sub header (struct kdump_sub_header)  |
1477      *  |------------------------------------------+ block 2
1478      *  |            1st-dump_bitmap               |
1479      *  |------------------------------------------+ block 2 + X blocks
1480      *  |            2nd-dump_bitmap               | (aligned by block)
1481      *  |------------------------------------------+ block 2 + 2 * X blocks
1482      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1483      *  |  page desc for pfn 1 (struct page_desc)  |
1484      *  |                    :                     |
1485      *  |------------------------------------------| (not aligned by block)
1486      *  |         page data (pfn 0)                |
1487      *  |         page data (pfn 1)                |
1488      *  |                    :                     |
1489      *  +------------------------------------------+
1490      */
1491 
1492     ret = write_start_flat_header(s->fd);
1493     if (ret < 0) {
1494         error_setg(errp, "dump: failed to write start flat header");
1495         return;
1496     }
1497 
1498     write_dump_header(s, &local_err);
1499     if (local_err) {
1500         error_propagate(errp, local_err);
1501         return;
1502     }
1503 
1504     write_dump_bitmap(s, &local_err);
1505     if (local_err) {
1506         error_propagate(errp, local_err);
1507         return;
1508     }
1509 
1510     write_dump_pages(s, &local_err);
1511     if (local_err) {
1512         error_propagate(errp, local_err);
1513         return;
1514     }
1515 
1516     ret = write_end_flat_header(s->fd);
1517     if (ret < 0) {
1518         error_setg(errp, "dump: failed to write end flat header");
1519         return;
1520     }
1521 }
1522 
1523 static ram_addr_t get_start_block(DumpState *s)
1524 {
1525     GuestPhysBlock *block;
1526 
1527     if (!s->has_filter) {
1528         s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1529         return 0;
1530     }
1531 
1532     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1533         if (block->target_start >= s->begin + s->length ||
1534             block->target_end <= s->begin) {
1535             /* This block is out of the range */
1536             continue;
1537         }
1538 
1539         s->next_block = block;
1540         if (s->begin > block->target_start) {
1541             s->start = s->begin - block->target_start;
1542         } else {
1543             s->start = 0;
1544         }
1545         return s->start;
1546     }
1547 
1548     return -1;
1549 }
1550 
1551 static void get_max_mapnr(DumpState *s)
1552 {
1553     GuestPhysBlock *last_block;
1554 
1555     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1556     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1557 }
1558 
1559 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1560 
1561 static void dump_state_prepare(DumpState *s)
1562 {
1563     /* zero the struct, setting status to active */
1564     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1565 }
1566 
1567 bool dump_in_progress(void)
1568 {
1569     DumpState *state = &dump_state_global;
1570     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1571 }
1572 
1573 /* calculate total size of memory to be dumped (taking filter into
1574  * acoount.) */
1575 static int64_t dump_calculate_size(DumpState *s)
1576 {
1577     GuestPhysBlock *block;
1578     int64_t size = 0, total = 0, left = 0, right = 0;
1579 
1580     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1581         if (s->has_filter) {
1582             /* calculate the overlapped region. */
1583             left = MAX(s->begin, block->target_start);
1584             right = MIN(s->begin + s->length, block->target_end);
1585             size = right - left;
1586             size = size > 0 ? size : 0;
1587         } else {
1588             /* count the whole region in */
1589             size = (block->target_end - block->target_start);
1590         }
1591         total += size;
1592     }
1593 
1594     return total;
1595 }
1596 
1597 static void vmcoreinfo_update_phys_base(DumpState *s)
1598 {
1599     uint64_t size, note_head_size, name_size, phys_base;
1600     char **lines;
1601     uint8_t *vmci;
1602     size_t i;
1603 
1604     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1605         return;
1606     }
1607 
1608     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1609     note_head_size = ROUND_UP(note_head_size, 4);
1610 
1611     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1612     *(vmci + size) = '\0';
1613 
1614     lines = g_strsplit((char *)vmci, "\n", -1);
1615     for (i = 0; lines[i]; i++) {
1616         const char *prefix = NULL;
1617 
1618         if (s->dump_info.d_machine == EM_X86_64) {
1619             prefix = "NUMBER(phys_base)=";
1620         } else if (s->dump_info.d_machine == EM_AARCH64) {
1621             prefix = "NUMBER(PHYS_OFFSET)=";
1622         }
1623 
1624         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1625             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1626                               &phys_base) < 0) {
1627                 warn_report("Failed to read %s", prefix);
1628             } else {
1629                 s->dump_info.phys_base = phys_base;
1630             }
1631             break;
1632         }
1633     }
1634 
1635     g_strfreev(lines);
1636 }
1637 
1638 static void dump_init(DumpState *s, int fd, bool has_format,
1639                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1640                       int64_t begin, int64_t length, Error **errp)
1641 {
1642     VMCoreInfoState *vmci = vmcoreinfo_find();
1643     CPUState *cpu;
1644     int nr_cpus;
1645     Error *err = NULL;
1646     int ret;
1647 
1648     s->has_format = has_format;
1649     s->format = format;
1650     s->written_size = 0;
1651 
1652     /* kdump-compressed is conflict with paging and filter */
1653     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1654         assert(!paging && !has_filter);
1655     }
1656 
1657     if (runstate_is_running()) {
1658         vm_stop(RUN_STATE_SAVE_VM);
1659         s->resume = true;
1660     } else {
1661         s->resume = false;
1662     }
1663 
1664     /* If we use KVM, we should synchronize the registers before we get dump
1665      * info or physmap info.
1666      */
1667     cpu_synchronize_all_states();
1668     nr_cpus = 0;
1669     CPU_FOREACH(cpu) {
1670         nr_cpus++;
1671     }
1672 
1673     s->fd = fd;
1674     s->has_filter = has_filter;
1675     s->begin = begin;
1676     s->length = length;
1677 
1678     memory_mapping_list_init(&s->list);
1679 
1680     guest_phys_blocks_init(&s->guest_phys_blocks);
1681     guest_phys_blocks_append(&s->guest_phys_blocks);
1682     s->total_size = dump_calculate_size(s);
1683 #ifdef DEBUG_DUMP_GUEST_MEMORY
1684     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1685 #endif
1686 
1687     /* it does not make sense to dump non-existent memory */
1688     if (!s->total_size) {
1689         error_setg(errp, "dump: no guest memory to dump");
1690         goto cleanup;
1691     }
1692 
1693     s->start = get_start_block(s);
1694     if (s->start == -1) {
1695         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1696         goto cleanup;
1697     }
1698 
1699     /* get dump info: endian, class and architecture.
1700      * If the target architecture is not supported, cpu_get_dump_info() will
1701      * return -1.
1702      */
1703     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1704     if (ret < 0) {
1705         error_setg(errp, QERR_UNSUPPORTED);
1706         goto cleanup;
1707     }
1708 
1709     if (!s->dump_info.page_size) {
1710         s->dump_info.page_size = TARGET_PAGE_SIZE;
1711     }
1712 
1713     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1714                                      s->dump_info.d_machine, nr_cpus);
1715     if (s->note_size < 0) {
1716         error_setg(errp, QERR_UNSUPPORTED);
1717         goto cleanup;
1718     }
1719 
1720     /*
1721      * The goal of this block is to (a) update the previously guessed
1722      * phys_base, (b) copy the guest note out of the guest.
1723      * Failure to do so is not fatal for dumping.
1724      */
1725     if (vmci) {
1726         uint64_t addr, note_head_size, name_size, desc_size;
1727         uint32_t size;
1728         uint16_t format;
1729 
1730         note_head_size = s->dump_info.d_class == ELFCLASS32 ?
1731             sizeof(Elf32_Nhdr) : sizeof(Elf64_Nhdr);
1732 
1733         format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1734         size = le32_to_cpu(vmci->vmcoreinfo.size);
1735         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1736         if (!vmci->has_vmcoreinfo) {
1737             warn_report("guest note is not present");
1738         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1739             warn_report("guest note size is invalid: %" PRIu32, size);
1740         } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1741             warn_report("guest note format is unsupported: %" PRIu16, format);
1742         } else {
1743             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1744             cpu_physical_memory_read(addr, s->guest_note, size);
1745 
1746             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1747             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1748                                                desc_size);
1749             if (name_size > MAX_GUEST_NOTE_SIZE ||
1750                 desc_size > MAX_GUEST_NOTE_SIZE ||
1751                 s->guest_note_size > size) {
1752                 warn_report("Invalid guest note header");
1753                 g_free(s->guest_note);
1754                 s->guest_note = NULL;
1755             } else {
1756                 vmcoreinfo_update_phys_base(s);
1757                 s->note_size += s->guest_note_size;
1758             }
1759         }
1760     }
1761 
1762     /* get memory mapping */
1763     if (paging) {
1764         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1765         if (err != NULL) {
1766             error_propagate(errp, err);
1767             goto cleanup;
1768         }
1769     } else {
1770         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1771     }
1772 
1773     s->nr_cpus = nr_cpus;
1774 
1775     get_max_mapnr(s);
1776 
1777     uint64_t tmp;
1778     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1779                        s->dump_info.page_size);
1780     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1781 
1782     /* init for kdump-compressed format */
1783     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1784         switch (format) {
1785         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1786             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1787             break;
1788 
1789         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1790 #ifdef CONFIG_LZO
1791             if (lzo_init() != LZO_E_OK) {
1792                 error_setg(errp, "failed to initialize the LZO library");
1793                 goto cleanup;
1794             }
1795 #endif
1796             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1797             break;
1798 
1799         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1800             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1801             break;
1802 
1803         default:
1804             s->flag_compress = 0;
1805         }
1806 
1807         return;
1808     }
1809 
1810     if (s->has_filter) {
1811         memory_mapping_filter(&s->list, s->begin, s->length);
1812     }
1813 
1814     /*
1815      * calculate phdr_num
1816      *
1817      * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1818      */
1819     s->phdr_num = 1; /* PT_NOTE */
1820     if (s->list.num < UINT16_MAX - 2) {
1821         s->phdr_num += s->list.num;
1822         s->have_section = false;
1823     } else {
1824         s->have_section = true;
1825         s->phdr_num = PN_XNUM;
1826         s->sh_info = 1; /* PT_NOTE */
1827 
1828         /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1829         if (s->list.num <= UINT32_MAX - 1) {
1830             s->sh_info += s->list.num;
1831         } else {
1832             s->sh_info = UINT32_MAX;
1833         }
1834     }
1835 
1836     if (s->dump_info.d_class == ELFCLASS64) {
1837         if (s->have_section) {
1838             s->memory_offset = sizeof(Elf64_Ehdr) +
1839                                sizeof(Elf64_Phdr) * s->sh_info +
1840                                sizeof(Elf64_Shdr) + s->note_size;
1841         } else {
1842             s->memory_offset = sizeof(Elf64_Ehdr) +
1843                                sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1844         }
1845     } else {
1846         if (s->have_section) {
1847             s->memory_offset = sizeof(Elf32_Ehdr) +
1848                                sizeof(Elf32_Phdr) * s->sh_info +
1849                                sizeof(Elf32_Shdr) + s->note_size;
1850         } else {
1851             s->memory_offset = sizeof(Elf32_Ehdr) +
1852                                sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1853         }
1854     }
1855 
1856     return;
1857 
1858 cleanup:
1859     dump_cleanup(s);
1860 }
1861 
1862 /* this operation might be time consuming. */
1863 static void dump_process(DumpState *s, Error **errp)
1864 {
1865     Error *local_err = NULL;
1866     DumpQueryResult *result = NULL;
1867 
1868     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1869 #ifdef TARGET_X86_64
1870         create_win_dump(s, &local_err);
1871 #endif
1872     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1873         create_kdump_vmcore(s, &local_err);
1874     } else {
1875         create_vmcore(s, &local_err);
1876     }
1877 
1878     /* make sure status is written after written_size updates */
1879     smp_wmb();
1880     qatomic_set(&s->status,
1881                (local_err ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1882 
1883     /* send DUMP_COMPLETED message (unconditionally) */
1884     result = qmp_query_dump(NULL);
1885     /* should never fail */
1886     assert(result);
1887     qapi_event_send_dump_completed(result, !!local_err, (local_err ?
1888                                    error_get_pretty(local_err) : NULL));
1889     qapi_free_DumpQueryResult(result);
1890 
1891     error_propagate(errp, local_err);
1892     dump_cleanup(s);
1893 }
1894 
1895 static void *dump_thread(void *data)
1896 {
1897     DumpState *s = (DumpState *)data;
1898     dump_process(s, NULL);
1899     return NULL;
1900 }
1901 
1902 DumpQueryResult *qmp_query_dump(Error **errp)
1903 {
1904     DumpQueryResult *result = g_new(DumpQueryResult, 1);
1905     DumpState *state = &dump_state_global;
1906     result->status = qatomic_read(&state->status);
1907     /* make sure we are reading status and written_size in order */
1908     smp_rmb();
1909     result->completed = state->written_size;
1910     result->total = state->total_size;
1911     return result;
1912 }
1913 
1914 void qmp_dump_guest_memory(bool paging, const char *file,
1915                            bool has_detach, bool detach,
1916                            bool has_begin, int64_t begin, bool has_length,
1917                            int64_t length, bool has_format,
1918                            DumpGuestMemoryFormat format, Error **errp)
1919 {
1920     const char *p;
1921     int fd = -1;
1922     DumpState *s;
1923     Error *local_err = NULL;
1924     bool detach_p = false;
1925 
1926     if (runstate_check(RUN_STATE_INMIGRATE)) {
1927         error_setg(errp, "Dump not allowed during incoming migration.");
1928         return;
1929     }
1930 
1931     /* if there is a dump in background, we should wait until the dump
1932      * finished */
1933     if (dump_in_progress()) {
1934         error_setg(errp, "There is a dump in process, please wait.");
1935         return;
1936     }
1937 
1938     /*
1939      * kdump-compressed format need the whole memory dumped, so paging or
1940      * filter is not supported here.
1941      */
1942     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1943         (paging || has_begin || has_length)) {
1944         error_setg(errp, "kdump-compressed format doesn't support paging or "
1945                          "filter");
1946         return;
1947     }
1948     if (has_begin && !has_length) {
1949         error_setg(errp, QERR_MISSING_PARAMETER, "length");
1950         return;
1951     }
1952     if (!has_begin && has_length) {
1953         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1954         return;
1955     }
1956     if (has_detach) {
1957         detach_p = detach;
1958     }
1959 
1960     /* check whether lzo/snappy is supported */
1961 #ifndef CONFIG_LZO
1962     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1963         error_setg(errp, "kdump-lzo is not available now");
1964         return;
1965     }
1966 #endif
1967 
1968 #ifndef CONFIG_SNAPPY
1969     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1970         error_setg(errp, "kdump-snappy is not available now");
1971         return;
1972     }
1973 #endif
1974 
1975 #ifndef TARGET_X86_64
1976     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1977         error_setg(errp, "Windows dump is only available for x86-64");
1978         return;
1979     }
1980 #endif
1981 
1982 #if !defined(WIN32)
1983     if (strstart(file, "fd:", &p)) {
1984         fd = monitor_get_fd(monitor_cur(), p, errp);
1985         if (fd == -1) {
1986             return;
1987         }
1988     }
1989 #endif
1990 
1991     if  (strstart(file, "file:", &p)) {
1992         fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1993         if (fd < 0) {
1994             error_setg_file_open(errp, errno, p);
1995             return;
1996         }
1997     }
1998 
1999     if (fd == -1) {
2000         error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
2001         return;
2002     }
2003 
2004     if (!dump_migration_blocker) {
2005         error_setg(&dump_migration_blocker,
2006                    "Live migration disabled: dump-guest-memory in progress");
2007     }
2008 
2009     /*
2010      * Allows even for -only-migratable, but forbid migration during the
2011      * process of dump guest memory.
2012      */
2013     if (migrate_add_blocker_internal(dump_migration_blocker, errp)) {
2014         /* Remember to release the fd before passing it over to dump state */
2015         close(fd);
2016         return;
2017     }
2018 
2019     s = &dump_state_global;
2020     dump_state_prepare(s);
2021 
2022     dump_init(s, fd, has_format, format, paging, has_begin,
2023               begin, length, &local_err);
2024     if (local_err) {
2025         error_propagate(errp, local_err);
2026         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2027         return;
2028     }
2029 
2030     if (detach_p) {
2031         /* detached dump */
2032         s->detached = true;
2033         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2034                            s, QEMU_THREAD_DETACHED);
2035     } else {
2036         /* sync dump */
2037         dump_process(s, errp);
2038     }
2039 }
2040 
2041 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2042 {
2043     DumpGuestMemoryCapability *cap =
2044                                   g_new0(DumpGuestMemoryCapability, 1);
2045     DumpGuestMemoryFormatList **tail = &cap->formats;
2046 
2047     /* elf is always available */
2048     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2049 
2050     /* kdump-zlib is always available */
2051     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2052 
2053     /* add new item if kdump-lzo is available */
2054 #ifdef CONFIG_LZO
2055     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2056 #endif
2057 
2058     /* add new item if kdump-snappy is available */
2059 #ifdef CONFIG_SNAPPY
2060     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2061 #endif
2062 
2063     /* Windows dump is available only if target is x86_64 */
2064 #ifdef TARGET_X86_64
2065     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2066 #endif
2067 
2068     return cap;
2069 }
2070