xref: /openbmc/qemu/dump/dump.c (revision 3e3df0d8)
1 /*
2  * QEMU dump
3  *
4  * Copyright Fujitsu, Corp. 2011, 2012
5  *
6  * Authors:
7  *     Wen Congyang <wency@cn.fujitsu.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
10  * See the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qemu/cutils.h"
16 #include "elf.h"
17 #include "qemu/bswap.h"
18 #include "exec/target_page.h"
19 #include "monitor/monitor.h"
20 #include "sysemu/dump.h"
21 #include "sysemu/runstate.h"
22 #include "sysemu/cpus.h"
23 #include "qapi/error.h"
24 #include "qapi/qapi-commands-dump.h"
25 #include "qapi/qapi-events-dump.h"
26 #include "qapi/qmp/qerror.h"
27 #include "qemu/error-report.h"
28 #include "qemu/main-loop.h"
29 #include "hw/misc/vmcoreinfo.h"
30 #include "migration/blocker.h"
31 #include "hw/core/cpu.h"
32 #include "win_dump.h"
33 
34 #include <zlib.h>
35 #ifdef CONFIG_LZO
36 #include <lzo/lzo1x.h>
37 #endif
38 #ifdef CONFIG_SNAPPY
39 #include <snappy-c.h>
40 #endif
41 #ifndef ELF_MACHINE_UNAME
42 #define ELF_MACHINE_UNAME "Unknown"
43 #endif
44 
45 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
46 
47 static Error *dump_migration_blocker;
48 
49 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size)   \
50     ((DIV_ROUND_UP((hdr_size), 4) +                     \
51       DIV_ROUND_UP((name_size), 4) +                    \
52       DIV_ROUND_UP((desc_size), 4)) * 4)
53 
54 static inline bool dump_is_64bit(DumpState *s)
55 {
56     return s->dump_info.d_class == ELFCLASS64;
57 }
58 
59 static inline bool dump_has_filter(DumpState *s)
60 {
61     return s->filter_area_length > 0;
62 }
63 
64 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
65 {
66     if (s->dump_info.d_endian == ELFDATA2LSB) {
67         val = cpu_to_le16(val);
68     } else {
69         val = cpu_to_be16(val);
70     }
71 
72     return val;
73 }
74 
75 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
76 {
77     if (s->dump_info.d_endian == ELFDATA2LSB) {
78         val = cpu_to_le32(val);
79     } else {
80         val = cpu_to_be32(val);
81     }
82 
83     return val;
84 }
85 
86 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
87 {
88     if (s->dump_info.d_endian == ELFDATA2LSB) {
89         val = cpu_to_le64(val);
90     } else {
91         val = cpu_to_be64(val);
92     }
93 
94     return val;
95 }
96 
97 static int dump_cleanup(DumpState *s)
98 {
99     guest_phys_blocks_free(&s->guest_phys_blocks);
100     memory_mapping_list_free(&s->list);
101     close(s->fd);
102     g_free(s->guest_note);
103     g_array_unref(s->string_table_buf);
104     s->guest_note = NULL;
105     if (s->resume) {
106         if (s->detached) {
107             qemu_mutex_lock_iothread();
108         }
109         vm_start();
110         if (s->detached) {
111             qemu_mutex_unlock_iothread();
112         }
113     }
114     migrate_del_blocker(dump_migration_blocker);
115 
116     return 0;
117 }
118 
119 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
120 {
121     DumpState *s = opaque;
122     size_t written_size;
123 
124     written_size = qemu_write_full(s->fd, buf, size);
125     if (written_size != size) {
126         return -errno;
127     }
128 
129     return 0;
130 }
131 
132 static void prepare_elf64_header(DumpState *s, Elf64_Ehdr *elf_header)
133 {
134     /*
135      * phnum in the elf header is 16 bit, if we have more segments we
136      * set phnum to PN_XNUM and write the real number of segments to a
137      * special section.
138      */
139     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
140 
141     memset(elf_header, 0, sizeof(Elf64_Ehdr));
142     memcpy(elf_header, ELFMAG, SELFMAG);
143     elf_header->e_ident[EI_CLASS] = ELFCLASS64;
144     elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
145     elf_header->e_ident[EI_VERSION] = EV_CURRENT;
146     elf_header->e_type = cpu_to_dump16(s, ET_CORE);
147     elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
148     elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
149     elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
150     elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset);
151     elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
152     elf_header->e_phnum = cpu_to_dump16(s, phnum);
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     elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
157 }
158 
159 static void prepare_elf32_header(DumpState *s, Elf32_Ehdr *elf_header)
160 {
161     /*
162      * phnum in the elf header is 16 bit, if we have more segments we
163      * set phnum to PN_XNUM and write the real number of segments to a
164      * special section.
165      */
166     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
167 
168     memset(elf_header, 0, sizeof(Elf32_Ehdr));
169     memcpy(elf_header, ELFMAG, SELFMAG);
170     elf_header->e_ident[EI_CLASS] = ELFCLASS32;
171     elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
172     elf_header->e_ident[EI_VERSION] = EV_CURRENT;
173     elf_header->e_type = cpu_to_dump16(s, ET_CORE);
174     elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
175     elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
176     elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
177     elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset);
178     elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
179     elf_header->e_phnum = cpu_to_dump16(s, phnum);
180     elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset);
181     elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
182     elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
183     elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
184 }
185 
186 static void write_elf_header(DumpState *s, Error **errp)
187 {
188     Elf32_Ehdr elf32_header;
189     Elf64_Ehdr elf64_header;
190     size_t header_size;
191     void *header_ptr;
192     int ret;
193 
194     /* The NULL header and the shstrtab are always defined */
195     assert(s->shdr_num >= 2);
196     if (dump_is_64bit(s)) {
197         prepare_elf64_header(s, &elf64_header);
198         header_size = sizeof(elf64_header);
199         header_ptr = &elf64_header;
200     } else {
201         prepare_elf32_header(s, &elf32_header);
202         header_size = sizeof(elf32_header);
203         header_ptr = &elf32_header;
204     }
205 
206     ret = fd_write_vmcore(header_ptr, header_size, s);
207     if (ret < 0) {
208         error_setg_errno(errp, -ret, "dump: failed to write elf header");
209     }
210 }
211 
212 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
213                              int phdr_index, hwaddr offset,
214                              hwaddr filesz, Error **errp)
215 {
216     Elf64_Phdr phdr;
217     int ret;
218 
219     memset(&phdr, 0, sizeof(Elf64_Phdr));
220     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
221     phdr.p_offset = cpu_to_dump64(s, offset);
222     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
223     phdr.p_filesz = cpu_to_dump64(s, filesz);
224     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
225     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
226 
227     assert(memory_mapping->length >= filesz);
228 
229     ret = fd_write_vmcore(&phdr, sizeof(Elf64_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_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
237                              int phdr_index, hwaddr offset,
238                              hwaddr filesz, Error **errp)
239 {
240     Elf32_Phdr phdr;
241     int ret;
242 
243     memset(&phdr, 0, sizeof(Elf32_Phdr));
244     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
245     phdr.p_offset = cpu_to_dump32(s, offset);
246     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
247     phdr.p_filesz = cpu_to_dump32(s, filesz);
248     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
249     phdr.p_vaddr =
250         cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
251 
252     assert(memory_mapping->length >= filesz);
253 
254     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
255     if (ret < 0) {
256         error_setg_errno(errp, -ret,
257                          "dump: failed to write program header table");
258     }
259 }
260 
261 static void prepare_elf64_phdr_note(DumpState *s, Elf64_Phdr *phdr)
262 {
263     memset(phdr, 0, sizeof(*phdr));
264     phdr->p_type = cpu_to_dump32(s, PT_NOTE);
265     phdr->p_offset = cpu_to_dump64(s, s->note_offset);
266     phdr->p_paddr = 0;
267     phdr->p_filesz = cpu_to_dump64(s, s->note_size);
268     phdr->p_memsz = cpu_to_dump64(s, s->note_size);
269     phdr->p_vaddr = 0;
270 }
271 
272 static inline int cpu_index(CPUState *cpu)
273 {
274     return cpu->cpu_index + 1;
275 }
276 
277 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
278                              Error **errp)
279 {
280     int ret;
281 
282     if (s->guest_note) {
283         ret = f(s->guest_note, s->guest_note_size, s);
284         if (ret < 0) {
285             error_setg(errp, "dump: failed to write guest note");
286         }
287     }
288 }
289 
290 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
291                               Error **errp)
292 {
293     CPUState *cpu;
294     int ret;
295     int id;
296 
297     CPU_FOREACH(cpu) {
298         id = cpu_index(cpu);
299         ret = cpu_write_elf64_note(f, cpu, id, s);
300         if (ret < 0) {
301             error_setg(errp, "dump: failed to write elf notes");
302             return;
303         }
304     }
305 
306     CPU_FOREACH(cpu) {
307         ret = cpu_write_elf64_qemunote(f, cpu, s);
308         if (ret < 0) {
309             error_setg(errp, "dump: failed to write CPU status");
310             return;
311         }
312     }
313 
314     write_guest_note(f, s, errp);
315 }
316 
317 static void prepare_elf32_phdr_note(DumpState *s, Elf32_Phdr *phdr)
318 {
319     memset(phdr, 0, sizeof(*phdr));
320     phdr->p_type = cpu_to_dump32(s, PT_NOTE);
321     phdr->p_offset = cpu_to_dump32(s, s->note_offset);
322     phdr->p_paddr = 0;
323     phdr->p_filesz = cpu_to_dump32(s, s->note_size);
324     phdr->p_memsz = cpu_to_dump32(s, s->note_size);
325     phdr->p_vaddr = 0;
326 }
327 
328 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
329                               Error **errp)
330 {
331     CPUState *cpu;
332     int ret;
333     int id;
334 
335     CPU_FOREACH(cpu) {
336         id = cpu_index(cpu);
337         ret = cpu_write_elf32_note(f, cpu, id, s);
338         if (ret < 0) {
339             error_setg(errp, "dump: failed to write elf notes");
340             return;
341         }
342     }
343 
344     CPU_FOREACH(cpu) {
345         ret = cpu_write_elf32_qemunote(f, cpu, s);
346         if (ret < 0) {
347             error_setg(errp, "dump: failed to write CPU status");
348             return;
349         }
350     }
351 
352     write_guest_note(f, s, errp);
353 }
354 
355 static void write_elf_phdr_note(DumpState *s, Error **errp)
356 {
357     Elf32_Phdr phdr32;
358     Elf64_Phdr phdr64;
359     void *phdr;
360     size_t size;
361     int ret;
362 
363     if (dump_is_64bit(s)) {
364         prepare_elf64_phdr_note(s, &phdr64);
365         size = sizeof(phdr64);
366         phdr = &phdr64;
367     } else {
368         prepare_elf32_phdr_note(s, &phdr32);
369         size = sizeof(phdr32);
370         phdr = &phdr32;
371     }
372 
373     ret = fd_write_vmcore(phdr, size, s);
374     if (ret < 0) {
375         error_setg_errno(errp, -ret,
376                          "dump: failed to write program header table");
377     }
378 }
379 
380 static void prepare_elf_section_hdr_zero(DumpState *s)
381 {
382     if (dump_is_64bit(s)) {
383         Elf64_Shdr *shdr64 = s->elf_section_hdrs;
384 
385         shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
386     } else {
387         Elf32_Shdr *shdr32 = s->elf_section_hdrs;
388 
389         shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
390     }
391 }
392 
393 static void prepare_elf_section_hdr_string(DumpState *s, void *buff)
394 {
395     uint64_t index = s->string_table_buf->len;
396     const char strtab[] = ".shstrtab";
397     Elf32_Shdr shdr32 = {};
398     Elf64_Shdr shdr64 = {};
399     int shdr_size;
400     void *shdr;
401 
402     g_array_append_vals(s->string_table_buf, strtab, sizeof(strtab));
403     if (dump_is_64bit(s)) {
404         shdr_size = sizeof(Elf64_Shdr);
405         shdr64.sh_type = SHT_STRTAB;
406         shdr64.sh_offset = s->section_offset + s->elf_section_data_size;
407         shdr64.sh_name = index;
408         shdr64.sh_size = s->string_table_buf->len;
409         shdr = &shdr64;
410     } else {
411         shdr_size = sizeof(Elf32_Shdr);
412         shdr32.sh_type = SHT_STRTAB;
413         shdr32.sh_offset = s->section_offset + s->elf_section_data_size;
414         shdr32.sh_name = index;
415         shdr32.sh_size = s->string_table_buf->len;
416         shdr = &shdr32;
417     }
418     memcpy(buff, shdr, shdr_size);
419 }
420 
421 static bool prepare_elf_section_hdrs(DumpState *s, Error **errp)
422 {
423     size_t len, sizeof_shdr;
424     void *buff_hdr;
425 
426     /*
427      * Section ordering:
428      * - HDR zero
429      * - Arch section hdrs
430      * - String table hdr
431      */
432     sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
433     len = sizeof_shdr * s->shdr_num;
434     s->elf_section_hdrs = g_malloc0(len);
435     buff_hdr = s->elf_section_hdrs;
436 
437     /*
438      * The first section header is ALWAYS a special initial section
439      * header.
440      *
441      * The header should be 0 with one exception being that if
442      * phdr_num is PN_XNUM then the sh_info field contains the real
443      * number of segment entries.
444      *
445      * As we zero allocate the buffer we will only need to modify
446      * sh_info for the PN_XNUM case.
447      */
448     if (s->phdr_num >= PN_XNUM) {
449         prepare_elf_section_hdr_zero(s);
450     }
451     buff_hdr += sizeof_shdr;
452 
453     /* Add architecture defined section headers */
454     if (s->dump_info.arch_sections_write_hdr_fn
455         && s->shdr_num > 2) {
456         buff_hdr += s->dump_info.arch_sections_write_hdr_fn(s, buff_hdr);
457 
458         if (s->shdr_num >= SHN_LORESERVE) {
459             error_setg_errno(errp, EINVAL,
460                              "dump: too many architecture defined sections");
461             return false;
462         }
463     }
464 
465     /*
466      * String table is the last section since strings are added via
467      * arch_sections_write_hdr().
468      */
469     prepare_elf_section_hdr_string(s, buff_hdr);
470     return true;
471 }
472 
473 static void write_elf_section_headers(DumpState *s, Error **errp)
474 {
475     size_t sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
476     int ret;
477 
478     if (!prepare_elf_section_hdrs(s, errp)) {
479         return;
480     }
481 
482     ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
483     if (ret < 0) {
484         error_setg_errno(errp, -ret, "dump: failed to write section headers");
485     }
486 
487     g_free(s->elf_section_hdrs);
488 }
489 
490 static void write_elf_sections(DumpState *s, Error **errp)
491 {
492     int ret;
493 
494     if (s->elf_section_data_size) {
495         /* Write architecture section data */
496         ret = fd_write_vmcore(s->elf_section_data,
497                               s->elf_section_data_size, s);
498         if (ret < 0) {
499             error_setg_errno(errp, -ret,
500                              "dump: failed to write architecture section data");
501             return;
502         }
503     }
504 
505     /* Write string table */
506     ret = fd_write_vmcore(s->string_table_buf->data,
507                           s->string_table_buf->len, s);
508     if (ret < 0) {
509         error_setg_errno(errp, -ret, "dump: failed to write string table data");
510     }
511 }
512 
513 static void write_data(DumpState *s, void *buf, int length, Error **errp)
514 {
515     int ret;
516 
517     ret = fd_write_vmcore(buf, length, s);
518     if (ret < 0) {
519         error_setg_errno(errp, -ret, "dump: failed to save memory");
520     } else {
521         s->written_size += length;
522     }
523 }
524 
525 /* write the memory to vmcore. 1 page per I/O. */
526 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
527                          int64_t size, Error **errp)
528 {
529     ERRP_GUARD();
530     int64_t i;
531 
532     for (i = 0; i < size / s->dump_info.page_size; i++) {
533         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
534                    s->dump_info.page_size, errp);
535         if (*errp) {
536             return;
537         }
538     }
539 
540     if ((size % s->dump_info.page_size) != 0) {
541         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
542                    size % s->dump_info.page_size, errp);
543         if (*errp) {
544             return;
545         }
546     }
547 }
548 
549 /* get the memory's offset and size in the vmcore */
550 static void get_offset_range(hwaddr phys_addr,
551                              ram_addr_t mapping_length,
552                              DumpState *s,
553                              hwaddr *p_offset,
554                              hwaddr *p_filesz)
555 {
556     GuestPhysBlock *block;
557     hwaddr offset = s->memory_offset;
558     int64_t size_in_block, start;
559 
560     /* When the memory is not stored into vmcore, offset will be -1 */
561     *p_offset = -1;
562     *p_filesz = 0;
563 
564     if (dump_has_filter(s)) {
565         if (phys_addr < s->filter_area_begin ||
566             phys_addr >= s->filter_area_begin + s->filter_area_length) {
567             return;
568         }
569     }
570 
571     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
572         if (dump_has_filter(s)) {
573             if (block->target_start >= s->filter_area_begin + s->filter_area_length ||
574                 block->target_end <= s->filter_area_begin) {
575                 /* This block is out of the range */
576                 continue;
577             }
578 
579             if (s->filter_area_begin <= block->target_start) {
580                 start = block->target_start;
581             } else {
582                 start = s->filter_area_begin;
583             }
584 
585             size_in_block = block->target_end - start;
586             if (s->filter_area_begin + s->filter_area_length < block->target_end) {
587                 size_in_block -= block->target_end - (s->filter_area_begin + s->filter_area_length);
588             }
589         } else {
590             start = block->target_start;
591             size_in_block = block->target_end - block->target_start;
592         }
593 
594         if (phys_addr >= start && phys_addr < start + size_in_block) {
595             *p_offset = phys_addr - start + offset;
596 
597             /* The offset range mapped from the vmcore file must not spill over
598              * the GuestPhysBlock, clamp it. The rest of the mapping will be
599              * zero-filled in memory at load time; see
600              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
601              */
602             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
603                         mapping_length :
604                         size_in_block - (phys_addr - start);
605             return;
606         }
607 
608         offset += size_in_block;
609     }
610 }
611 
612 static void write_elf_phdr_loads(DumpState *s, Error **errp)
613 {
614     ERRP_GUARD();
615     hwaddr offset, filesz;
616     MemoryMapping *memory_mapping;
617     uint32_t phdr_index = 1;
618 
619     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
620         get_offset_range(memory_mapping->phys_addr,
621                          memory_mapping->length,
622                          s, &offset, &filesz);
623         if (dump_is_64bit(s)) {
624             write_elf64_load(s, memory_mapping, phdr_index++, offset,
625                              filesz, errp);
626         } else {
627             write_elf32_load(s, memory_mapping, phdr_index++, offset,
628                              filesz, errp);
629         }
630 
631         if (*errp) {
632             return;
633         }
634 
635         if (phdr_index >= s->phdr_num) {
636             break;
637         }
638     }
639 }
640 
641 static void write_elf_notes(DumpState *s, Error **errp)
642 {
643     if (dump_is_64bit(s)) {
644         write_elf64_notes(fd_write_vmcore, s, errp);
645     } else {
646         write_elf32_notes(fd_write_vmcore, s, errp);
647     }
648 }
649 
650 /* write elf header, PT_NOTE and elf note to vmcore. */
651 static void dump_begin(DumpState *s, Error **errp)
652 {
653     ERRP_GUARD();
654 
655     /*
656      * the vmcore's format is:
657      *   --------------
658      *   |  elf header |
659      *   --------------
660      *   |  sctn_hdr   |
661      *   --------------
662      *   |  PT_NOTE    |
663      *   --------------
664      *   |  PT_LOAD    |
665      *   --------------
666      *   |  ......     |
667      *   --------------
668      *   |  PT_LOAD    |
669      *   --------------
670      *   |  elf note   |
671      *   --------------
672      *   |  memory     |
673      *   --------------
674      *
675      * we only know where the memory is saved after we write elf note into
676      * vmcore.
677      */
678 
679     /* write elf header to vmcore */
680     write_elf_header(s, errp);
681     if (*errp) {
682         return;
683     }
684 
685     /* write section headers to vmcore */
686     write_elf_section_headers(s, errp);
687     if (*errp) {
688         return;
689     }
690 
691     /* write PT_NOTE to vmcore */
692     write_elf_phdr_note(s, errp);
693     if (*errp) {
694         return;
695     }
696 
697     /* write all PT_LOADs to vmcore */
698     write_elf_phdr_loads(s, errp);
699     if (*errp) {
700         return;
701     }
702 
703     /* write notes to vmcore */
704     write_elf_notes(s, errp);
705 }
706 
707 int64_t dump_filtered_memblock_size(GuestPhysBlock *block,
708                                     int64_t filter_area_start,
709                                     int64_t filter_area_length)
710 {
711     int64_t size, left, right;
712 
713     /* No filter, return full size */
714     if (!filter_area_length) {
715         return block->target_end - block->target_start;
716     }
717 
718     /* calculate the overlapped region. */
719     left = MAX(filter_area_start, block->target_start);
720     right = MIN(filter_area_start + filter_area_length, block->target_end);
721     size = right - left;
722     size = size > 0 ? size : 0;
723 
724     return size;
725 }
726 
727 int64_t dump_filtered_memblock_start(GuestPhysBlock *block,
728                                      int64_t filter_area_start,
729                                      int64_t filter_area_length)
730 {
731     if (filter_area_length) {
732         /* return -1 if the block is not within filter area */
733         if (block->target_start >= filter_area_start + filter_area_length ||
734             block->target_end <= filter_area_start) {
735             return -1;
736         }
737 
738         if (filter_area_start > block->target_start) {
739             return filter_area_start - block->target_start;
740         }
741     }
742 
743     return 0;
744 }
745 
746 /* write all memory to vmcore */
747 static void dump_iterate(DumpState *s, Error **errp)
748 {
749     ERRP_GUARD();
750     GuestPhysBlock *block;
751     int64_t memblock_size, memblock_start;
752 
753     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
754         memblock_start = dump_filtered_memblock_start(block, s->filter_area_begin, s->filter_area_length);
755         if (memblock_start == -1) {
756             continue;
757         }
758 
759         memblock_size = dump_filtered_memblock_size(block, s->filter_area_begin, s->filter_area_length);
760 
761         /* Write the memory to file */
762         write_memory(s, block, memblock_start, memblock_size, errp);
763         if (*errp) {
764             return;
765         }
766     }
767 }
768 
769 static void dump_end(DumpState *s, Error **errp)
770 {
771     int rc;
772 
773     if (s->elf_section_data_size) {
774         s->elf_section_data = g_malloc0(s->elf_section_data_size);
775     }
776 
777     /* Adds the architecture defined section data to s->elf_section_data  */
778     if (s->dump_info.arch_sections_write_fn &&
779         s->elf_section_data_size) {
780         rc = s->dump_info.arch_sections_write_fn(s, s->elf_section_data);
781         if (rc) {
782             error_setg_errno(errp, rc,
783                              "dump: failed to get arch section data");
784             g_free(s->elf_section_data);
785             return;
786         }
787     }
788 
789     /* write sections to vmcore */
790     write_elf_sections(s, errp);
791 }
792 
793 static void create_vmcore(DumpState *s, Error **errp)
794 {
795     ERRP_GUARD();
796 
797     dump_begin(s, errp);
798     if (*errp) {
799         return;
800     }
801 
802     /* Iterate over memory and dump it to file */
803     dump_iterate(s, errp);
804     if (*errp) {
805         return;
806     }
807 
808     /* Write the section data */
809     dump_end(s, errp);
810 }
811 
812 static int write_start_flat_header(int fd)
813 {
814     MakedumpfileHeader *mh;
815     int ret = 0;
816 
817     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
818     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
819 
820     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
821            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
822 
823     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
824     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
825 
826     size_t written_size;
827     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
828     if (written_size != MAX_SIZE_MDF_HEADER) {
829         ret = -1;
830     }
831 
832     g_free(mh);
833     return ret;
834 }
835 
836 static int write_end_flat_header(int fd)
837 {
838     MakedumpfileDataHeader mdh;
839 
840     mdh.offset = END_FLAG_FLAT_HEADER;
841     mdh.buf_size = END_FLAG_FLAT_HEADER;
842 
843     size_t written_size;
844     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
845     if (written_size != sizeof(mdh)) {
846         return -1;
847     }
848 
849     return 0;
850 }
851 
852 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
853 {
854     size_t written_size;
855     MakedumpfileDataHeader mdh;
856 
857     mdh.offset = cpu_to_be64(offset);
858     mdh.buf_size = cpu_to_be64(size);
859 
860     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
861     if (written_size != sizeof(mdh)) {
862         return -1;
863     }
864 
865     written_size = qemu_write_full(fd, buf, size);
866     if (written_size != size) {
867         return -1;
868     }
869 
870     return 0;
871 }
872 
873 static int buf_write_note(const void *buf, size_t size, void *opaque)
874 {
875     DumpState *s = opaque;
876 
877     /* note_buf is not enough */
878     if (s->note_buf_offset + size > s->note_size) {
879         return -1;
880     }
881 
882     memcpy(s->note_buf + s->note_buf_offset, buf, size);
883 
884     s->note_buf_offset += size;
885 
886     return 0;
887 }
888 
889 /*
890  * This function retrieves various sizes from an elf header.
891  *
892  * @note has to be a valid ELF note. The return sizes are unmodified
893  * (not padded or rounded up to be multiple of 4).
894  */
895 static void get_note_sizes(DumpState *s, const void *note,
896                            uint64_t *note_head_size,
897                            uint64_t *name_size,
898                            uint64_t *desc_size)
899 {
900     uint64_t note_head_sz;
901     uint64_t name_sz;
902     uint64_t desc_sz;
903 
904     if (dump_is_64bit(s)) {
905         const Elf64_Nhdr *hdr = note;
906         note_head_sz = sizeof(Elf64_Nhdr);
907         name_sz = cpu_to_dump64(s, hdr->n_namesz);
908         desc_sz = cpu_to_dump64(s, hdr->n_descsz);
909     } else {
910         const Elf32_Nhdr *hdr = note;
911         note_head_sz = sizeof(Elf32_Nhdr);
912         name_sz = cpu_to_dump32(s, hdr->n_namesz);
913         desc_sz = cpu_to_dump32(s, hdr->n_descsz);
914     }
915 
916     if (note_head_size) {
917         *note_head_size = note_head_sz;
918     }
919     if (name_size) {
920         *name_size = name_sz;
921     }
922     if (desc_size) {
923         *desc_size = desc_sz;
924     }
925 }
926 
927 static bool note_name_equal(DumpState *s,
928                             const uint8_t *note, const char *name)
929 {
930     int len = strlen(name) + 1;
931     uint64_t head_size, name_size;
932 
933     get_note_sizes(s, note, &head_size, &name_size, NULL);
934     head_size = ROUND_UP(head_size, 4);
935 
936     return name_size == len && memcmp(note + head_size, name, len) == 0;
937 }
938 
939 /* write common header, sub header and elf note to vmcore */
940 static void create_header32(DumpState *s, Error **errp)
941 {
942     ERRP_GUARD();
943     DiskDumpHeader32 *dh = NULL;
944     KdumpSubHeader32 *kh = NULL;
945     size_t size;
946     uint32_t block_size;
947     uint32_t sub_hdr_size;
948     uint32_t bitmap_blocks;
949     uint32_t status = 0;
950     uint64_t offset_note;
951 
952     /* write common header, the version of kdump-compressed format is 6th */
953     size = sizeof(DiskDumpHeader32);
954     dh = g_malloc0(size);
955 
956     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
957     dh->header_version = cpu_to_dump32(s, 6);
958     block_size = s->dump_info.page_size;
959     dh->block_size = cpu_to_dump32(s, block_size);
960     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
961     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
962     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
963     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
964     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
965     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
966     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
967     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
968     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
969 
970     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
971         status |= DUMP_DH_COMPRESSED_ZLIB;
972     }
973 #ifdef CONFIG_LZO
974     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
975         status |= DUMP_DH_COMPRESSED_LZO;
976     }
977 #endif
978 #ifdef CONFIG_SNAPPY
979     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
980         status |= DUMP_DH_COMPRESSED_SNAPPY;
981     }
982 #endif
983     dh->status = cpu_to_dump32(s, status);
984 
985     if (write_buffer(s->fd, 0, dh, size) < 0) {
986         error_setg(errp, "dump: failed to write disk dump header");
987         goto out;
988     }
989 
990     /* write sub header */
991     size = sizeof(KdumpSubHeader32);
992     kh = g_malloc0(size);
993 
994     /* 64bit max_mapnr_64 */
995     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
996     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
997     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
998 
999     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1000     if (s->guest_note &&
1001         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1002         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1003 
1004         get_note_sizes(s, s->guest_note,
1005                        &hsize, &name_size, &size_vmcoreinfo_desc);
1006         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1007             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1008         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1009         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
1010     }
1011 
1012     kh->offset_note = cpu_to_dump64(s, offset_note);
1013     kh->note_size = cpu_to_dump32(s, s->note_size);
1014 
1015     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
1016                      block_size, kh, size) < 0) {
1017         error_setg(errp, "dump: failed to write kdump sub header");
1018         goto out;
1019     }
1020 
1021     /* write note */
1022     s->note_buf = g_malloc0(s->note_size);
1023     s->note_buf_offset = 0;
1024 
1025     /* use s->note_buf to store notes temporarily */
1026     write_elf32_notes(buf_write_note, s, errp);
1027     if (*errp) {
1028         goto out;
1029     }
1030     if (write_buffer(s->fd, offset_note, s->note_buf,
1031                      s->note_size) < 0) {
1032         error_setg(errp, "dump: failed to write notes");
1033         goto out;
1034     }
1035 
1036     /* get offset of dump_bitmap */
1037     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1038                              block_size;
1039 
1040     /* get offset of page */
1041     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1042                      block_size;
1043 
1044 out:
1045     g_free(dh);
1046     g_free(kh);
1047     g_free(s->note_buf);
1048 }
1049 
1050 /* write common header, sub header and elf note to vmcore */
1051 static void create_header64(DumpState *s, Error **errp)
1052 {
1053     ERRP_GUARD();
1054     DiskDumpHeader64 *dh = NULL;
1055     KdumpSubHeader64 *kh = NULL;
1056     size_t size;
1057     uint32_t block_size;
1058     uint32_t sub_hdr_size;
1059     uint32_t bitmap_blocks;
1060     uint32_t status = 0;
1061     uint64_t offset_note;
1062 
1063     /* write common header, the version of kdump-compressed format is 6th */
1064     size = sizeof(DiskDumpHeader64);
1065     dh = g_malloc0(size);
1066 
1067     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
1068     dh->header_version = cpu_to_dump32(s, 6);
1069     block_size = s->dump_info.page_size;
1070     dh->block_size = cpu_to_dump32(s, block_size);
1071     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
1072     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
1073     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
1074     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
1075     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
1076     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
1077     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
1078     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
1079     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
1080 
1081     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
1082         status |= DUMP_DH_COMPRESSED_ZLIB;
1083     }
1084 #ifdef CONFIG_LZO
1085     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
1086         status |= DUMP_DH_COMPRESSED_LZO;
1087     }
1088 #endif
1089 #ifdef CONFIG_SNAPPY
1090     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1091         status |= DUMP_DH_COMPRESSED_SNAPPY;
1092     }
1093 #endif
1094     dh->status = cpu_to_dump32(s, status);
1095 
1096     if (write_buffer(s->fd, 0, dh, size) < 0) {
1097         error_setg(errp, "dump: failed to write disk dump header");
1098         goto out;
1099     }
1100 
1101     /* write sub header */
1102     size = sizeof(KdumpSubHeader64);
1103     kh = g_malloc0(size);
1104 
1105     /* 64bit max_mapnr_64 */
1106     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1107     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
1108     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1109 
1110     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1111     if (s->guest_note &&
1112         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1113         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1114 
1115         get_note_sizes(s, s->guest_note,
1116                        &hsize, &name_size, &size_vmcoreinfo_desc);
1117         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1118             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1119         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1120         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
1121     }
1122 
1123     kh->offset_note = cpu_to_dump64(s, offset_note);
1124     kh->note_size = cpu_to_dump64(s, s->note_size);
1125 
1126     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
1127                      block_size, kh, size) < 0) {
1128         error_setg(errp, "dump: failed to write kdump sub header");
1129         goto out;
1130     }
1131 
1132     /* write note */
1133     s->note_buf = g_malloc0(s->note_size);
1134     s->note_buf_offset = 0;
1135 
1136     /* use s->note_buf to store notes temporarily */
1137     write_elf64_notes(buf_write_note, s, errp);
1138     if (*errp) {
1139         goto out;
1140     }
1141 
1142     if (write_buffer(s->fd, offset_note, s->note_buf,
1143                      s->note_size) < 0) {
1144         error_setg(errp, "dump: failed to write notes");
1145         goto out;
1146     }
1147 
1148     /* get offset of dump_bitmap */
1149     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1150                              block_size;
1151 
1152     /* get offset of page */
1153     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1154                      block_size;
1155 
1156 out:
1157     g_free(dh);
1158     g_free(kh);
1159     g_free(s->note_buf);
1160 }
1161 
1162 static void write_dump_header(DumpState *s, Error **errp)
1163 {
1164     if (dump_is_64bit(s)) {
1165         create_header64(s, errp);
1166     } else {
1167         create_header32(s, errp);
1168     }
1169 }
1170 
1171 static size_t dump_bitmap_get_bufsize(DumpState *s)
1172 {
1173     return s->dump_info.page_size;
1174 }
1175 
1176 /*
1177  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1178  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1179  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1180  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1181  * vmcore, ie. synchronizing un-sync bit into vmcore.
1182  */
1183 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1184                            uint8_t *buf, DumpState *s)
1185 {
1186     off_t old_offset, new_offset;
1187     off_t offset_bitmap1, offset_bitmap2;
1188     uint32_t byte, bit;
1189     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1190     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1191 
1192     /* should not set the previous place */
1193     assert(last_pfn <= pfn);
1194 
1195     /*
1196      * if the bit needed to be set is not cached in buf, flush the data in buf
1197      * to vmcore firstly.
1198      * making new_offset be bigger than old_offset can also sync remained data
1199      * into vmcore.
1200      */
1201     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1202     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1203 
1204     while (old_offset < new_offset) {
1205         /* calculate the offset and write dump_bitmap */
1206         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1207         if (write_buffer(s->fd, offset_bitmap1, buf,
1208                          bitmap_bufsize) < 0) {
1209             return -1;
1210         }
1211 
1212         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1213         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1214                          old_offset;
1215         if (write_buffer(s->fd, offset_bitmap2, buf,
1216                          bitmap_bufsize) < 0) {
1217             return -1;
1218         }
1219 
1220         memset(buf, 0, bitmap_bufsize);
1221         old_offset += bitmap_bufsize;
1222     }
1223 
1224     /* get the exact place of the bit in the buf, and set it */
1225     byte = (pfn % bits_per_buf) / CHAR_BIT;
1226     bit = (pfn % bits_per_buf) % CHAR_BIT;
1227     if (value) {
1228         buf[byte] |= 1u << bit;
1229     } else {
1230         buf[byte] &= ~(1u << bit);
1231     }
1232 
1233     return 0;
1234 }
1235 
1236 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1237 {
1238     int target_page_shift = ctz32(s->dump_info.page_size);
1239 
1240     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1241 }
1242 
1243 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1244 {
1245     int target_page_shift = ctz32(s->dump_info.page_size);
1246 
1247     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1248 }
1249 
1250 /*
1251  * Return the page frame number and the page content in *bufptr. bufptr can be
1252  * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated
1253  * memory. This is not necessarily the memory returned.
1254  */
1255 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1256                           uint8_t **bufptr, DumpState *s)
1257 {
1258     GuestPhysBlock *block = *blockptr;
1259     uint32_t page_size = s->dump_info.page_size;
1260     uint8_t *buf = NULL, *hbuf;
1261     hwaddr addr;
1262 
1263     /* block == NULL means the start of the iteration */
1264     if (!block) {
1265         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1266         *blockptr = block;
1267         addr = block->target_start;
1268         *pfnptr = dump_paddr_to_pfn(s, addr);
1269     } else {
1270         *pfnptr += 1;
1271         addr = dump_pfn_to_paddr(s, *pfnptr);
1272     }
1273     assert(block != NULL);
1274 
1275     while (1) {
1276         if (addr >= block->target_start && addr < block->target_end) {
1277             size_t n = MIN(block->target_end - addr, page_size - addr % page_size);
1278             hbuf = block->host_addr + (addr - block->target_start);
1279             if (!buf) {
1280                 if (n == page_size) {
1281                     /* this is a whole target page, go for it */
1282                     assert(addr % page_size == 0);
1283                     buf = hbuf;
1284                     break;
1285                 } else if (bufptr) {
1286                     assert(*bufptr);
1287                     buf = *bufptr;
1288                     memset(buf, 0, page_size);
1289                 } else {
1290                     return true;
1291                 }
1292             }
1293 
1294             memcpy(buf + addr % page_size, hbuf, n);
1295             addr += n;
1296             if (addr % page_size == 0 || addr >= block->target_end) {
1297                 /* we filled up the page or the current block is finished */
1298                 break;
1299             }
1300         } else {
1301             /* the next page is in the next block */
1302             *blockptr = block = QTAILQ_NEXT(block, next);
1303             if (!block) {
1304                 break;
1305             }
1306 
1307             addr = block->target_start;
1308             /* are we still in the same page? */
1309             if (dump_paddr_to_pfn(s, addr) != *pfnptr) {
1310                 if (buf) {
1311                     /* no, but we already filled something earlier, return it */
1312                     break;
1313                 } else {
1314                     /* else continue from there */
1315                     *pfnptr = dump_paddr_to_pfn(s, addr);
1316                 }
1317             }
1318         }
1319     }
1320 
1321     if (bufptr) {
1322         *bufptr = buf;
1323     }
1324 
1325     return buf != NULL;
1326 }
1327 
1328 static void write_dump_bitmap(DumpState *s, Error **errp)
1329 {
1330     int ret = 0;
1331     uint64_t last_pfn, pfn;
1332     void *dump_bitmap_buf;
1333     size_t num_dumpable;
1334     GuestPhysBlock *block_iter = NULL;
1335     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1336     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1337 
1338     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1339     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1340 
1341     num_dumpable = 0;
1342     last_pfn = 0;
1343 
1344     /*
1345      * exam memory page by page, and set the bit in dump_bitmap corresponded
1346      * to the existing page.
1347      */
1348     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1349         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1350         if (ret < 0) {
1351             error_setg(errp, "dump: failed to set dump_bitmap");
1352             goto out;
1353         }
1354 
1355         last_pfn = pfn;
1356         num_dumpable++;
1357     }
1358 
1359     /*
1360      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1361      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1362      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1363      */
1364     if (num_dumpable > 0) {
1365         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1366                               dump_bitmap_buf, s);
1367         if (ret < 0) {
1368             error_setg(errp, "dump: failed to sync dump_bitmap");
1369             goto out;
1370         }
1371     }
1372 
1373     /* number of dumpable pages that will be dumped later */
1374     s->num_dumpable = num_dumpable;
1375 
1376 out:
1377     g_free(dump_bitmap_buf);
1378 }
1379 
1380 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1381                                off_t offset)
1382 {
1383     data_cache->fd = s->fd;
1384     data_cache->data_size = 0;
1385     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1386     data_cache->buf = g_malloc0(data_cache->buf_size);
1387     data_cache->offset = offset;
1388 }
1389 
1390 static int write_cache(DataCache *dc, const void *buf, size_t size,
1391                        bool flag_sync)
1392 {
1393     /*
1394      * dc->buf_size should not be less than size, otherwise dc will never be
1395      * enough
1396      */
1397     assert(size <= dc->buf_size);
1398 
1399     /*
1400      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1401      * otherwise check if the space is enough for caching data in buf, if not,
1402      * write the data in dc->buf to dc->fd and reset dc->buf
1403      */
1404     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1405         (flag_sync && dc->data_size > 0)) {
1406         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1407             return -1;
1408         }
1409 
1410         dc->offset += dc->data_size;
1411         dc->data_size = 0;
1412     }
1413 
1414     if (!flag_sync) {
1415         memcpy(dc->buf + dc->data_size, buf, size);
1416         dc->data_size += size;
1417     }
1418 
1419     return 0;
1420 }
1421 
1422 static void free_data_cache(DataCache *data_cache)
1423 {
1424     g_free(data_cache->buf);
1425 }
1426 
1427 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1428 {
1429     switch (flag_compress) {
1430     case DUMP_DH_COMPRESSED_ZLIB:
1431         return compressBound(page_size);
1432 
1433     case DUMP_DH_COMPRESSED_LZO:
1434         /*
1435          * LZO will expand incompressible data by a little amount. Please check
1436          * the following URL to see the expansion calculation:
1437          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1438          */
1439         return page_size + page_size / 16 + 64 + 3;
1440 
1441 #ifdef CONFIG_SNAPPY
1442     case DUMP_DH_COMPRESSED_SNAPPY:
1443         return snappy_max_compressed_length(page_size);
1444 #endif
1445     }
1446     return 0;
1447 }
1448 
1449 static void write_dump_pages(DumpState *s, Error **errp)
1450 {
1451     int ret = 0;
1452     DataCache page_desc, page_data;
1453     size_t len_buf_out, size_out;
1454 #ifdef CONFIG_LZO
1455     lzo_bytep wrkmem = NULL;
1456 #endif
1457     uint8_t *buf_out = NULL;
1458     off_t offset_desc, offset_data;
1459     PageDescriptor pd, pd_zero;
1460     uint8_t *buf;
1461     GuestPhysBlock *block_iter = NULL;
1462     uint64_t pfn_iter;
1463     g_autofree uint8_t *page = NULL;
1464 
1465     /* get offset of page_desc and page_data in dump file */
1466     offset_desc = s->offset_page;
1467     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1468 
1469     prepare_data_cache(&page_desc, s, offset_desc);
1470     prepare_data_cache(&page_data, s, offset_data);
1471 
1472     /* prepare buffer to store compressed data */
1473     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1474     assert(len_buf_out != 0);
1475 
1476 #ifdef CONFIG_LZO
1477     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1478 #endif
1479 
1480     buf_out = g_malloc(len_buf_out);
1481 
1482     /*
1483      * init zero page's page_desc and page_data, because every zero page
1484      * uses the same page_data
1485      */
1486     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1487     pd_zero.flags = cpu_to_dump32(s, 0);
1488     pd_zero.offset = cpu_to_dump64(s, offset_data);
1489     pd_zero.page_flags = cpu_to_dump64(s, 0);
1490     buf = g_malloc0(s->dump_info.page_size);
1491     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1492     g_free(buf);
1493     if (ret < 0) {
1494         error_setg(errp, "dump: failed to write page data (zero page)");
1495         goto out;
1496     }
1497 
1498     offset_data += s->dump_info.page_size;
1499     page = g_malloc(s->dump_info.page_size);
1500 
1501     /*
1502      * dump memory to vmcore page by page. zero page will all be resided in the
1503      * first page of page section
1504      */
1505     for (buf = page; get_next_page(&block_iter, &pfn_iter, &buf, s); buf = page) {
1506         /* check zero page */
1507         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1508             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1509                               false);
1510             if (ret < 0) {
1511                 error_setg(errp, "dump: failed to write page desc");
1512                 goto out;
1513             }
1514         } else {
1515             /*
1516              * not zero page, then:
1517              * 1. compress the page
1518              * 2. write the compressed page into the cache of page_data
1519              * 3. get page desc of the compressed page and write it into the
1520              *    cache of page_desc
1521              *
1522              * only one compression format will be used here, for
1523              * s->flag_compress is set. But when compression fails to work,
1524              * we fall back to save in plaintext.
1525              */
1526              size_out = len_buf_out;
1527              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1528                     (compress2(buf_out, (uLongf *)&size_out, buf,
1529                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1530                     (size_out < s->dump_info.page_size)) {
1531                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1532                 pd.size  = cpu_to_dump32(s, size_out);
1533 
1534                 ret = write_cache(&page_data, buf_out, size_out, false);
1535                 if (ret < 0) {
1536                     error_setg(errp, "dump: failed to write page data");
1537                     goto out;
1538                 }
1539 #ifdef CONFIG_LZO
1540             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1541                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1542                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1543                     (size_out < s->dump_info.page_size)) {
1544                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1545                 pd.size  = cpu_to_dump32(s, size_out);
1546 
1547                 ret = write_cache(&page_data, buf_out, size_out, false);
1548                 if (ret < 0) {
1549                     error_setg(errp, "dump: failed to write page data");
1550                     goto out;
1551                 }
1552 #endif
1553 #ifdef CONFIG_SNAPPY
1554             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1555                     (snappy_compress((char *)buf, s->dump_info.page_size,
1556                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1557                     (size_out < s->dump_info.page_size)) {
1558                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1559                 pd.size  = cpu_to_dump32(s, size_out);
1560 
1561                 ret = write_cache(&page_data, buf_out, size_out, false);
1562                 if (ret < 0) {
1563                     error_setg(errp, "dump: failed to write page data");
1564                     goto out;
1565                 }
1566 #endif
1567             } else {
1568                 /*
1569                  * fall back to save in plaintext, size_out should be
1570                  * assigned the target's page size
1571                  */
1572                 pd.flags = cpu_to_dump32(s, 0);
1573                 size_out = s->dump_info.page_size;
1574                 pd.size = cpu_to_dump32(s, size_out);
1575 
1576                 ret = write_cache(&page_data, buf,
1577                                   s->dump_info.page_size, false);
1578                 if (ret < 0) {
1579                     error_setg(errp, "dump: failed to write page data");
1580                     goto out;
1581                 }
1582             }
1583 
1584             /* get and write page desc here */
1585             pd.page_flags = cpu_to_dump64(s, 0);
1586             pd.offset = cpu_to_dump64(s, offset_data);
1587             offset_data += size_out;
1588 
1589             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1590             if (ret < 0) {
1591                 error_setg(errp, "dump: failed to write page desc");
1592                 goto out;
1593             }
1594         }
1595         s->written_size += s->dump_info.page_size;
1596     }
1597 
1598     ret = write_cache(&page_desc, NULL, 0, true);
1599     if (ret < 0) {
1600         error_setg(errp, "dump: failed to sync cache for page_desc");
1601         goto out;
1602     }
1603     ret = write_cache(&page_data, NULL, 0, true);
1604     if (ret < 0) {
1605         error_setg(errp, "dump: failed to sync cache for page_data");
1606         goto out;
1607     }
1608 
1609 out:
1610     free_data_cache(&page_desc);
1611     free_data_cache(&page_data);
1612 
1613 #ifdef CONFIG_LZO
1614     g_free(wrkmem);
1615 #endif
1616 
1617     g_free(buf_out);
1618 }
1619 
1620 static void create_kdump_vmcore(DumpState *s, Error **errp)
1621 {
1622     ERRP_GUARD();
1623     int ret;
1624 
1625     /*
1626      * the kdump-compressed format is:
1627      *                                               File offset
1628      *  +------------------------------------------+ 0x0
1629      *  |    main header (struct disk_dump_header) |
1630      *  |------------------------------------------+ block 1
1631      *  |    sub header (struct kdump_sub_header)  |
1632      *  |------------------------------------------+ block 2
1633      *  |            1st-dump_bitmap               |
1634      *  |------------------------------------------+ block 2 + X blocks
1635      *  |            2nd-dump_bitmap               | (aligned by block)
1636      *  |------------------------------------------+ block 2 + 2 * X blocks
1637      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1638      *  |  page desc for pfn 1 (struct page_desc)  |
1639      *  |                    :                     |
1640      *  |------------------------------------------| (not aligned by block)
1641      *  |         page data (pfn 0)                |
1642      *  |         page data (pfn 1)                |
1643      *  |                    :                     |
1644      *  +------------------------------------------+
1645      */
1646 
1647     ret = write_start_flat_header(s->fd);
1648     if (ret < 0) {
1649         error_setg(errp, "dump: failed to write start flat header");
1650         return;
1651     }
1652 
1653     write_dump_header(s, errp);
1654     if (*errp) {
1655         return;
1656     }
1657 
1658     write_dump_bitmap(s, errp);
1659     if (*errp) {
1660         return;
1661     }
1662 
1663     write_dump_pages(s, errp);
1664     if (*errp) {
1665         return;
1666     }
1667 
1668     ret = write_end_flat_header(s->fd);
1669     if (ret < 0) {
1670         error_setg(errp, "dump: failed to write end flat header");
1671         return;
1672     }
1673 }
1674 
1675 static int validate_start_block(DumpState *s)
1676 {
1677     GuestPhysBlock *block;
1678 
1679     if (!dump_has_filter(s)) {
1680         return 0;
1681     }
1682 
1683     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1684         /* This block is out of the range */
1685         if (block->target_start >= s->filter_area_begin + s->filter_area_length ||
1686             block->target_end <= s->filter_area_begin) {
1687             continue;
1688         }
1689         return 0;
1690    }
1691 
1692     return -1;
1693 }
1694 
1695 static void get_max_mapnr(DumpState *s)
1696 {
1697     GuestPhysBlock *last_block;
1698 
1699     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1700     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1701 }
1702 
1703 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1704 
1705 static void dump_state_prepare(DumpState *s)
1706 {
1707     /* zero the struct, setting status to active */
1708     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1709 }
1710 
1711 bool qemu_system_dump_in_progress(void)
1712 {
1713     DumpState *state = &dump_state_global;
1714     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1715 }
1716 
1717 /*
1718  * calculate total size of memory to be dumped (taking filter into
1719  * account.)
1720  */
1721 static int64_t dump_calculate_size(DumpState *s)
1722 {
1723     GuestPhysBlock *block;
1724     int64_t total = 0;
1725 
1726     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1727         total += dump_filtered_memblock_size(block,
1728                                              s->filter_area_begin,
1729                                              s->filter_area_length);
1730     }
1731 
1732     return total;
1733 }
1734 
1735 static void vmcoreinfo_update_phys_base(DumpState *s)
1736 {
1737     uint64_t size, note_head_size, name_size, phys_base;
1738     char **lines;
1739     uint8_t *vmci;
1740     size_t i;
1741 
1742     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1743         return;
1744     }
1745 
1746     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1747     note_head_size = ROUND_UP(note_head_size, 4);
1748 
1749     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1750     *(vmci + size) = '\0';
1751 
1752     lines = g_strsplit((char *)vmci, "\n", -1);
1753     for (i = 0; lines[i]; i++) {
1754         const char *prefix = NULL;
1755 
1756         if (s->dump_info.d_machine == EM_X86_64) {
1757             prefix = "NUMBER(phys_base)=";
1758         } else if (s->dump_info.d_machine == EM_AARCH64) {
1759             prefix = "NUMBER(PHYS_OFFSET)=";
1760         }
1761 
1762         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1763             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1764                               &phys_base) < 0) {
1765                 warn_report("Failed to read %s", prefix);
1766             } else {
1767                 s->dump_info.phys_base = phys_base;
1768             }
1769             break;
1770         }
1771     }
1772 
1773     g_strfreev(lines);
1774 }
1775 
1776 static void dump_init(DumpState *s, int fd, bool has_format,
1777                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1778                       int64_t begin, int64_t length, Error **errp)
1779 {
1780     ERRP_GUARD();
1781     VMCoreInfoState *vmci = vmcoreinfo_find();
1782     CPUState *cpu;
1783     int nr_cpus;
1784     int ret;
1785 
1786     s->has_format = has_format;
1787     s->format = format;
1788     s->written_size = 0;
1789 
1790     /* kdump-compressed is conflict with paging and filter */
1791     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1792         assert(!paging && !has_filter);
1793     }
1794 
1795     if (runstate_is_running()) {
1796         vm_stop(RUN_STATE_SAVE_VM);
1797         s->resume = true;
1798     } else {
1799         s->resume = false;
1800     }
1801 
1802     /* If we use KVM, we should synchronize the registers before we get dump
1803      * info or physmap info.
1804      */
1805     cpu_synchronize_all_states();
1806     nr_cpus = 0;
1807     CPU_FOREACH(cpu) {
1808         nr_cpus++;
1809     }
1810 
1811     s->fd = fd;
1812     if (has_filter && !length) {
1813         error_setg(errp, QERR_INVALID_PARAMETER, "length");
1814         goto cleanup;
1815     }
1816     s->filter_area_begin = begin;
1817     s->filter_area_length = length;
1818 
1819     /* First index is 0, it's the special null name */
1820     s->string_table_buf = g_array_new(FALSE, TRUE, 1);
1821     /*
1822      * Allocate the null name, due to the clearing option set to true
1823      * it will be 0.
1824      */
1825     g_array_set_size(s->string_table_buf, 1);
1826 
1827     memory_mapping_list_init(&s->list);
1828 
1829     guest_phys_blocks_init(&s->guest_phys_blocks);
1830     guest_phys_blocks_append(&s->guest_phys_blocks);
1831     s->total_size = dump_calculate_size(s);
1832 #ifdef DEBUG_DUMP_GUEST_MEMORY
1833     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1834 #endif
1835 
1836     /* it does not make sense to dump non-existent memory */
1837     if (!s->total_size) {
1838         error_setg(errp, "dump: no guest memory to dump");
1839         goto cleanup;
1840     }
1841 
1842     /* Is the filter filtering everything? */
1843     if (validate_start_block(s) == -1) {
1844         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1845         goto cleanup;
1846     }
1847 
1848     /* get dump info: endian, class and architecture.
1849      * If the target architecture is not supported, cpu_get_dump_info() will
1850      * return -1.
1851      */
1852     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1853     if (ret < 0) {
1854         error_setg(errp,
1855                    "dumping guest memory is not supported on this target");
1856         goto cleanup;
1857     }
1858 
1859     if (!s->dump_info.page_size) {
1860         s->dump_info.page_size = qemu_target_page_size();
1861     }
1862 
1863     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1864                                      s->dump_info.d_machine, nr_cpus);
1865     assert(s->note_size >= 0);
1866 
1867     /*
1868      * The goal of this block is to (a) update the previously guessed
1869      * phys_base, (b) copy the guest note out of the guest.
1870      * Failure to do so is not fatal for dumping.
1871      */
1872     if (vmci) {
1873         uint64_t addr, note_head_size, name_size, desc_size;
1874         uint32_t size;
1875         uint16_t guest_format;
1876 
1877         note_head_size = dump_is_64bit(s) ?
1878             sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr);
1879 
1880         guest_format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1881         size = le32_to_cpu(vmci->vmcoreinfo.size);
1882         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1883         if (!vmci->has_vmcoreinfo) {
1884             warn_report("guest note is not present");
1885         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1886             warn_report("guest note size is invalid: %" PRIu32, size);
1887         } else if (guest_format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1888             warn_report("guest note format is unsupported: %" PRIu16, guest_format);
1889         } else {
1890             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1891             cpu_physical_memory_read(addr, s->guest_note, size);
1892 
1893             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1894             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1895                                                desc_size);
1896             if (name_size > MAX_GUEST_NOTE_SIZE ||
1897                 desc_size > MAX_GUEST_NOTE_SIZE ||
1898                 s->guest_note_size > size) {
1899                 warn_report("Invalid guest note header");
1900                 g_free(s->guest_note);
1901                 s->guest_note = NULL;
1902             } else {
1903                 vmcoreinfo_update_phys_base(s);
1904                 s->note_size += s->guest_note_size;
1905             }
1906         }
1907     }
1908 
1909     /* get memory mapping */
1910     if (paging) {
1911         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1912         if (*errp) {
1913             goto cleanup;
1914         }
1915     } else {
1916         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1917     }
1918 
1919     s->nr_cpus = nr_cpus;
1920 
1921     get_max_mapnr(s);
1922 
1923     uint64_t tmp;
1924     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1925                        s->dump_info.page_size);
1926     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1927 
1928     /* init for kdump-compressed format */
1929     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1930         switch (format) {
1931         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1932             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1933             break;
1934 
1935         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1936 #ifdef CONFIG_LZO
1937             if (lzo_init() != LZO_E_OK) {
1938                 error_setg(errp, "failed to initialize the LZO library");
1939                 goto cleanup;
1940             }
1941 #endif
1942             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1943             break;
1944 
1945         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1946             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1947             break;
1948 
1949         default:
1950             s->flag_compress = 0;
1951         }
1952 
1953         return;
1954     }
1955 
1956     if (dump_has_filter(s)) {
1957         memory_mapping_filter(&s->list, s->filter_area_begin, s->filter_area_length);
1958     }
1959 
1960     /*
1961      * The first section header is always a special one in which most
1962      * fields are 0. The section header string table is also always
1963      * set.
1964      */
1965     s->shdr_num = 2;
1966 
1967     /*
1968      * Adds the number of architecture sections to shdr_num and sets
1969      * elf_section_data_size so we know the offsets and sizes of all
1970      * parts.
1971      */
1972     if (s->dump_info.arch_sections_add_fn) {
1973         s->dump_info.arch_sections_add_fn(s);
1974     }
1975 
1976     /*
1977      * calculate shdr_num so we know the offsets and sizes of all
1978      * parts.
1979      * Calculate phdr_num
1980      *
1981      * The absolute maximum amount of phdrs is UINT32_MAX - 1 as
1982      * sh_info is 32 bit. There's special handling once we go over
1983      * UINT16_MAX - 1 but that is handled in the ehdr and section
1984      * code.
1985      */
1986     s->phdr_num = 1; /* Reserve PT_NOTE */
1987     if (s->list.num <= UINT32_MAX - 1) {
1988         s->phdr_num += s->list.num;
1989     } else {
1990         s->phdr_num = UINT32_MAX;
1991     }
1992 
1993     /*
1994      * Now that the number of section and program headers is known we
1995      * can calculate the offsets of the headers and data.
1996      */
1997     if (dump_is_64bit(s)) {
1998         s->shdr_offset = sizeof(Elf64_Ehdr);
1999         s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
2000         s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
2001     } else {
2002         s->shdr_offset = sizeof(Elf32_Ehdr);
2003         s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
2004         s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
2005     }
2006     s->memory_offset = s->note_offset + s->note_size;
2007     s->section_offset = s->memory_offset + s->total_size;
2008 
2009     return;
2010 
2011 cleanup:
2012     dump_cleanup(s);
2013 }
2014 
2015 /* this operation might be time consuming. */
2016 static void dump_process(DumpState *s, Error **errp)
2017 {
2018     ERRP_GUARD();
2019     DumpQueryResult *result = NULL;
2020 
2021     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2022         create_win_dump(s, errp);
2023     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
2024         create_kdump_vmcore(s, errp);
2025     } else {
2026         create_vmcore(s, errp);
2027     }
2028 
2029     /* make sure status is written after written_size updates */
2030     smp_wmb();
2031     qatomic_set(&s->status,
2032                (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
2033 
2034     /* send DUMP_COMPLETED message (unconditionally) */
2035     result = qmp_query_dump(NULL);
2036     /* should never fail */
2037     assert(result);
2038     qapi_event_send_dump_completed(result,
2039                                    *errp ? error_get_pretty(*errp) : NULL);
2040     qapi_free_DumpQueryResult(result);
2041 
2042     dump_cleanup(s);
2043 }
2044 
2045 static void *dump_thread(void *data)
2046 {
2047     DumpState *s = (DumpState *)data;
2048     dump_process(s, NULL);
2049     return NULL;
2050 }
2051 
2052 DumpQueryResult *qmp_query_dump(Error **errp)
2053 {
2054     DumpQueryResult *result = g_new(DumpQueryResult, 1);
2055     DumpState *state = &dump_state_global;
2056     result->status = qatomic_read(&state->status);
2057     /* make sure we are reading status and written_size in order */
2058     smp_rmb();
2059     result->completed = state->written_size;
2060     result->total = state->total_size;
2061     return result;
2062 }
2063 
2064 void qmp_dump_guest_memory(bool paging, const char *file,
2065                            bool has_detach, bool detach,
2066                            bool has_begin, int64_t begin, bool has_length,
2067                            int64_t length, bool has_format,
2068                            DumpGuestMemoryFormat format, Error **errp)
2069 {
2070     ERRP_GUARD();
2071     const char *p;
2072     int fd = -1;
2073     DumpState *s;
2074     bool detach_p = false;
2075 
2076     if (runstate_check(RUN_STATE_INMIGRATE)) {
2077         error_setg(errp, "Dump not allowed during incoming migration.");
2078         return;
2079     }
2080 
2081     /* if there is a dump in background, we should wait until the dump
2082      * finished */
2083     if (qemu_system_dump_in_progress()) {
2084         error_setg(errp, "There is a dump in process, please wait.");
2085         return;
2086     }
2087 
2088     /*
2089      * kdump-compressed format need the whole memory dumped, so paging or
2090      * filter is not supported here.
2091      */
2092     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
2093         (paging || has_begin || has_length)) {
2094         error_setg(errp, "kdump-compressed format doesn't support paging or "
2095                          "filter");
2096         return;
2097     }
2098     if (has_begin && !has_length) {
2099         error_setg(errp, QERR_MISSING_PARAMETER, "length");
2100         return;
2101     }
2102     if (!has_begin && has_length) {
2103         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
2104         return;
2105     }
2106     if (has_detach) {
2107         detach_p = detach;
2108     }
2109 
2110     /* check whether lzo/snappy is supported */
2111 #ifndef CONFIG_LZO
2112     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
2113         error_setg(errp, "kdump-lzo is not available now");
2114         return;
2115     }
2116 #endif
2117 
2118 #ifndef CONFIG_SNAPPY
2119     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
2120         error_setg(errp, "kdump-snappy is not available now");
2121         return;
2122     }
2123 #endif
2124 
2125     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
2126         && !win_dump_available(errp)) {
2127         return;
2128     }
2129 
2130 #if !defined(WIN32)
2131     if (strstart(file, "fd:", &p)) {
2132         fd = monitor_get_fd(monitor_cur(), p, errp);
2133         if (fd == -1) {
2134             return;
2135         }
2136     }
2137 #endif
2138 
2139     if  (strstart(file, "file:", &p)) {
2140         fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
2141         if (fd < 0) {
2142             error_setg_file_open(errp, errno, p);
2143             return;
2144         }
2145     }
2146 
2147     if (fd == -1) {
2148         error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
2149         return;
2150     }
2151 
2152     if (!dump_migration_blocker) {
2153         error_setg(&dump_migration_blocker,
2154                    "Live migration disabled: dump-guest-memory in progress");
2155     }
2156 
2157     /*
2158      * Allows even for -only-migratable, but forbid migration during the
2159      * process of dump guest memory.
2160      */
2161     if (migrate_add_blocker_internal(dump_migration_blocker, errp)) {
2162         /* Remember to release the fd before passing it over to dump state */
2163         close(fd);
2164         return;
2165     }
2166 
2167     s = &dump_state_global;
2168     dump_state_prepare(s);
2169 
2170     dump_init(s, fd, has_format, format, paging, has_begin,
2171               begin, length, errp);
2172     if (*errp) {
2173         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2174         return;
2175     }
2176 
2177     if (detach_p) {
2178         /* detached dump */
2179         s->detached = true;
2180         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2181                            s, QEMU_THREAD_DETACHED);
2182     } else {
2183         /* sync dump */
2184         dump_process(s, errp);
2185     }
2186 }
2187 
2188 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2189 {
2190     DumpGuestMemoryCapability *cap =
2191                                   g_new0(DumpGuestMemoryCapability, 1);
2192     DumpGuestMemoryFormatList **tail = &cap->formats;
2193 
2194     /* elf is always available */
2195     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2196 
2197     /* kdump-zlib is always available */
2198     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2199 
2200     /* add new item if kdump-lzo is available */
2201 #ifdef CONFIG_LZO
2202     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2203 #endif
2204 
2205     /* add new item if kdump-snappy is available */
2206 #ifdef CONFIG_SNAPPY
2207     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2208 #endif
2209 
2210     if (win_dump_available(NULL)) {
2211         QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2212     }
2213 
2214     return cap;
2215 }
2216