xref: /openbmc/qemu/dump/dump.c (revision 6d2d454a)
1 /*
2  * QEMU dump
3  *
4  * Copyright Fujitsu, Corp. 2011, 2012
5  *
6  * Authors:
7  *     Wen Congyang <wency@cn.fujitsu.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
10  * See the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qemu/cutils.h"
16 #include "elf.h"
17 #include "exec/hwaddr.h"
18 #include "monitor/monitor.h"
19 #include "sysemu/kvm.h"
20 #include "sysemu/dump.h"
21 #include "sysemu/memory_mapping.h"
22 #include "sysemu/runstate.h"
23 #include "sysemu/cpus.h"
24 #include "qapi/error.h"
25 #include "qapi/qapi-commands-dump.h"
26 #include "qapi/qapi-events-dump.h"
27 #include "qapi/qmp/qerror.h"
28 #include "qemu/error-report.h"
29 #include "qemu/main-loop.h"
30 #include "hw/misc/vmcoreinfo.h"
31 #include "migration/blocker.h"
32 
33 #ifdef TARGET_X86_64
34 #include "win_dump.h"
35 #endif
36 
37 #include <zlib.h>
38 #ifdef CONFIG_LZO
39 #include <lzo/lzo1x.h>
40 #endif
41 #ifdef CONFIG_SNAPPY
42 #include <snappy-c.h>
43 #endif
44 #ifndef ELF_MACHINE_UNAME
45 #define ELF_MACHINE_UNAME "Unknown"
46 #endif
47 
48 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
49 
50 static Error *dump_migration_blocker;
51 
52 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size)   \
53     ((DIV_ROUND_UP((hdr_size), 4) +                     \
54       DIV_ROUND_UP((name_size), 4) +                    \
55       DIV_ROUND_UP((desc_size), 4)) * 4)
56 
57 static inline bool dump_is_64bit(DumpState *s)
58 {
59     return s->dump_info.d_class == ELFCLASS64;
60 }
61 
62 static inline bool dump_has_filter(DumpState *s)
63 {
64     return s->filter_area_length > 0;
65 }
66 
67 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
68 {
69     if (s->dump_info.d_endian == ELFDATA2LSB) {
70         val = cpu_to_le16(val);
71     } else {
72         val = cpu_to_be16(val);
73     }
74 
75     return val;
76 }
77 
78 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
79 {
80     if (s->dump_info.d_endian == ELFDATA2LSB) {
81         val = cpu_to_le32(val);
82     } else {
83         val = cpu_to_be32(val);
84     }
85 
86     return val;
87 }
88 
89 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
90 {
91     if (s->dump_info.d_endian == ELFDATA2LSB) {
92         val = cpu_to_le64(val);
93     } else {
94         val = cpu_to_be64(val);
95     }
96 
97     return val;
98 }
99 
100 static int dump_cleanup(DumpState *s)
101 {
102     guest_phys_blocks_free(&s->guest_phys_blocks);
103     memory_mapping_list_free(&s->list);
104     close(s->fd);
105     g_free(s->guest_note);
106     g_array_unref(s->string_table_buf);
107     s->guest_note = NULL;
108     if (s->resume) {
109         if (s->detached) {
110             qemu_mutex_lock_iothread();
111         }
112         vm_start();
113         if (s->detached) {
114             qemu_mutex_unlock_iothread();
115         }
116     }
117     migrate_del_blocker(dump_migration_blocker);
118 
119     return 0;
120 }
121 
122 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
123 {
124     DumpState *s = opaque;
125     size_t written_size;
126 
127     written_size = qemu_write_full(s->fd, buf, size);
128     if (written_size != size) {
129         return -errno;
130     }
131 
132     return 0;
133 }
134 
135 static void prepare_elf64_header(DumpState *s, Elf64_Ehdr *elf_header)
136 {
137     /*
138      * phnum in the elf header is 16 bit, if we have more segments we
139      * set phnum to PN_XNUM and write the real number of segments to a
140      * special section.
141      */
142     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
143 
144     memset(elf_header, 0, sizeof(Elf64_Ehdr));
145     memcpy(elf_header, ELFMAG, SELFMAG);
146     elf_header->e_ident[EI_CLASS] = ELFCLASS64;
147     elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
148     elf_header->e_ident[EI_VERSION] = EV_CURRENT;
149     elf_header->e_type = cpu_to_dump16(s, ET_CORE);
150     elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
151     elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
152     elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
153     elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset);
154     elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
155     elf_header->e_phnum = cpu_to_dump16(s, phnum);
156     elf_header->e_shoff = cpu_to_dump64(s, s->shdr_offset);
157     elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
158     elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
159     elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
160 }
161 
162 static void prepare_elf32_header(DumpState *s, Elf32_Ehdr *elf_header)
163 {
164     /*
165      * phnum in the elf header is 16 bit, if we have more segments we
166      * set phnum to PN_XNUM and write the real number of segments to a
167      * special section.
168      */
169     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
170 
171     memset(elf_header, 0, sizeof(Elf32_Ehdr));
172     memcpy(elf_header, ELFMAG, SELFMAG);
173     elf_header->e_ident[EI_CLASS] = ELFCLASS32;
174     elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
175     elf_header->e_ident[EI_VERSION] = EV_CURRENT;
176     elf_header->e_type = cpu_to_dump16(s, ET_CORE);
177     elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
178     elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
179     elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
180     elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset);
181     elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
182     elf_header->e_phnum = cpu_to_dump16(s, phnum);
183     elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset);
184     elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
185     elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
186     elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
187 }
188 
189 static void write_elf_header(DumpState *s, Error **errp)
190 {
191     Elf32_Ehdr elf32_header;
192     Elf64_Ehdr elf64_header;
193     size_t header_size;
194     void *header_ptr;
195     int ret;
196 
197     /* The NULL header and the shstrtab are always defined */
198     assert(s->shdr_num >= 2);
199     if (dump_is_64bit(s)) {
200         prepare_elf64_header(s, &elf64_header);
201         header_size = sizeof(elf64_header);
202         header_ptr = &elf64_header;
203     } else {
204         prepare_elf32_header(s, &elf32_header);
205         header_size = sizeof(elf32_header);
206         header_ptr = &elf32_header;
207     }
208 
209     ret = fd_write_vmcore(header_ptr, header_size, s);
210     if (ret < 0) {
211         error_setg_errno(errp, -ret, "dump: failed to write elf header");
212     }
213 }
214 
215 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
216                              int phdr_index, hwaddr offset,
217                              hwaddr filesz, Error **errp)
218 {
219     Elf64_Phdr phdr;
220     int ret;
221 
222     memset(&phdr, 0, sizeof(Elf64_Phdr));
223     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
224     phdr.p_offset = cpu_to_dump64(s, offset);
225     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
226     phdr.p_filesz = cpu_to_dump64(s, filesz);
227     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
228     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
229 
230     assert(memory_mapping->length >= filesz);
231 
232     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
233     if (ret < 0) {
234         error_setg_errno(errp, -ret,
235                          "dump: failed to write program header table");
236     }
237 }
238 
239 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
240                              int phdr_index, hwaddr offset,
241                              hwaddr filesz, Error **errp)
242 {
243     Elf32_Phdr phdr;
244     int ret;
245 
246     memset(&phdr, 0, sizeof(Elf32_Phdr));
247     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
248     phdr.p_offset = cpu_to_dump32(s, offset);
249     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
250     phdr.p_filesz = cpu_to_dump32(s, filesz);
251     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
252     phdr.p_vaddr =
253         cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
254 
255     assert(memory_mapping->length >= filesz);
256 
257     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
258     if (ret < 0) {
259         error_setg_errno(errp, -ret,
260                          "dump: failed to write program header table");
261     }
262 }
263 
264 static void prepare_elf64_phdr_note(DumpState *s, Elf64_Phdr *phdr)
265 {
266     memset(phdr, 0, sizeof(*phdr));
267     phdr->p_type = cpu_to_dump32(s, PT_NOTE);
268     phdr->p_offset = cpu_to_dump64(s, s->note_offset);
269     phdr->p_paddr = 0;
270     phdr->p_filesz = cpu_to_dump64(s, s->note_size);
271     phdr->p_memsz = cpu_to_dump64(s, s->note_size);
272     phdr->p_vaddr = 0;
273 }
274 
275 static inline int cpu_index(CPUState *cpu)
276 {
277     return cpu->cpu_index + 1;
278 }
279 
280 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
281                              Error **errp)
282 {
283     int ret;
284 
285     if (s->guest_note) {
286         ret = f(s->guest_note, s->guest_note_size, s);
287         if (ret < 0) {
288             error_setg(errp, "dump: failed to write guest note");
289         }
290     }
291 }
292 
293 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
294                               Error **errp)
295 {
296     CPUState *cpu;
297     int ret;
298     int id;
299 
300     CPU_FOREACH(cpu) {
301         id = cpu_index(cpu);
302         ret = cpu_write_elf64_note(f, cpu, id, s);
303         if (ret < 0) {
304             error_setg(errp, "dump: failed to write elf notes");
305             return;
306         }
307     }
308 
309     CPU_FOREACH(cpu) {
310         ret = cpu_write_elf64_qemunote(f, cpu, s);
311         if (ret < 0) {
312             error_setg(errp, "dump: failed to write CPU status");
313             return;
314         }
315     }
316 
317     write_guest_note(f, s, errp);
318 }
319 
320 static void prepare_elf32_phdr_note(DumpState *s, Elf32_Phdr *phdr)
321 {
322     memset(phdr, 0, sizeof(*phdr));
323     phdr->p_type = cpu_to_dump32(s, PT_NOTE);
324     phdr->p_offset = cpu_to_dump32(s, s->note_offset);
325     phdr->p_paddr = 0;
326     phdr->p_filesz = cpu_to_dump32(s, s->note_size);
327     phdr->p_memsz = cpu_to_dump32(s, s->note_size);
328     phdr->p_vaddr = 0;
329 }
330 
331 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
332                               Error **errp)
333 {
334     CPUState *cpu;
335     int ret;
336     int id;
337 
338     CPU_FOREACH(cpu) {
339         id = cpu_index(cpu);
340         ret = cpu_write_elf32_note(f, cpu, id, s);
341         if (ret < 0) {
342             error_setg(errp, "dump: failed to write elf notes");
343             return;
344         }
345     }
346 
347     CPU_FOREACH(cpu) {
348         ret = cpu_write_elf32_qemunote(f, cpu, s);
349         if (ret < 0) {
350             error_setg(errp, "dump: failed to write CPU status");
351             return;
352         }
353     }
354 
355     write_guest_note(f, s, errp);
356 }
357 
358 static void write_elf_phdr_note(DumpState *s, Error **errp)
359 {
360     Elf32_Phdr phdr32;
361     Elf64_Phdr phdr64;
362     void *phdr;
363     size_t size;
364     int ret;
365 
366     if (dump_is_64bit(s)) {
367         prepare_elf64_phdr_note(s, &phdr64);
368         size = sizeof(phdr64);
369         phdr = &phdr64;
370     } else {
371         prepare_elf32_phdr_note(s, &phdr32);
372         size = sizeof(phdr32);
373         phdr = &phdr32;
374     }
375 
376     ret = fd_write_vmcore(phdr, size, s);
377     if (ret < 0) {
378         error_setg_errno(errp, -ret,
379                          "dump: failed to write program header table");
380     }
381 }
382 
383 static void prepare_elf_section_hdr_zero(DumpState *s)
384 {
385     if (dump_is_64bit(s)) {
386         Elf64_Shdr *shdr64 = s->elf_section_hdrs;
387 
388         shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
389     } else {
390         Elf32_Shdr *shdr32 = s->elf_section_hdrs;
391 
392         shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
393     }
394 }
395 
396 static void prepare_elf_section_hdr_string(DumpState *s, void *buff)
397 {
398     uint64_t index = s->string_table_buf->len;
399     const char strtab[] = ".shstrtab";
400     Elf32_Shdr shdr32 = {};
401     Elf64_Shdr shdr64 = {};
402     int shdr_size;
403     void *shdr;
404 
405     g_array_append_vals(s->string_table_buf, strtab, sizeof(strtab));
406     if (dump_is_64bit(s)) {
407         shdr_size = sizeof(Elf64_Shdr);
408         shdr64.sh_type = SHT_STRTAB;
409         shdr64.sh_offset = s->section_offset + s->elf_section_data_size;
410         shdr64.sh_name = index;
411         shdr64.sh_size = s->string_table_buf->len;
412         shdr = &shdr64;
413     } else {
414         shdr_size = sizeof(Elf32_Shdr);
415         shdr32.sh_type = SHT_STRTAB;
416         shdr32.sh_offset = s->section_offset + s->elf_section_data_size;
417         shdr32.sh_name = index;
418         shdr32.sh_size = s->string_table_buf->len;
419         shdr = &shdr32;
420     }
421     memcpy(buff, shdr, shdr_size);
422 }
423 
424 static bool prepare_elf_section_hdrs(DumpState *s, Error **errp)
425 {
426     size_t len, sizeof_shdr;
427     void *buff_hdr;
428 
429     /*
430      * Section ordering:
431      * - HDR zero
432      * - Arch section hdrs
433      * - String table hdr
434      */
435     sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
436     len = sizeof_shdr * s->shdr_num;
437     s->elf_section_hdrs = g_malloc0(len);
438     buff_hdr = s->elf_section_hdrs;
439 
440     /*
441      * The first section header is ALWAYS a special initial section
442      * header.
443      *
444      * The header should be 0 with one exception being that if
445      * phdr_num is PN_XNUM then the sh_info field contains the real
446      * number of segment entries.
447      *
448      * As we zero allocate the buffer we will only need to modify
449      * sh_info for the PN_XNUM case.
450      */
451     if (s->phdr_num >= PN_XNUM) {
452         prepare_elf_section_hdr_zero(s);
453     }
454     buff_hdr += sizeof_shdr;
455 
456     /* Add architecture defined section headers */
457     if (s->dump_info.arch_sections_write_hdr_fn
458         && s->shdr_num > 2) {
459         buff_hdr += s->dump_info.arch_sections_write_hdr_fn(s, buff_hdr);
460 
461         if (s->shdr_num >= SHN_LORESERVE) {
462             error_setg_errno(errp, EINVAL,
463                              "dump: too many architecture defined sections");
464             return false;
465         }
466     }
467 
468     /*
469      * String table is the last section since strings are added via
470      * arch_sections_write_hdr().
471      */
472     prepare_elf_section_hdr_string(s, buff_hdr);
473     return true;
474 }
475 
476 static void write_elf_section_headers(DumpState *s, Error **errp)
477 {
478     size_t sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
479     int ret;
480 
481     if (!prepare_elf_section_hdrs(s, errp)) {
482         return;
483     }
484 
485     ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
486     if (ret < 0) {
487         error_setg_errno(errp, -ret, "dump: failed to write section headers");
488     }
489 
490     g_free(s->elf_section_hdrs);
491 }
492 
493 static void write_elf_sections(DumpState *s, Error **errp)
494 {
495     int ret;
496 
497     if (s->elf_section_data_size) {
498         /* Write architecture section data */
499         ret = fd_write_vmcore(s->elf_section_data,
500                               s->elf_section_data_size, s);
501         if (ret < 0) {
502             error_setg_errno(errp, -ret,
503                              "dump: failed to write architecture section data");
504             return;
505         }
506     }
507 
508     /* Write string table */
509     ret = fd_write_vmcore(s->string_table_buf->data,
510                           s->string_table_buf->len, s);
511     if (ret < 0) {
512         error_setg_errno(errp, -ret, "dump: failed to write string table data");
513     }
514 }
515 
516 static void write_data(DumpState *s, void *buf, int length, Error **errp)
517 {
518     int ret;
519 
520     ret = fd_write_vmcore(buf, length, s);
521     if (ret < 0) {
522         error_setg_errno(errp, -ret, "dump: failed to save memory");
523     } else {
524         s->written_size += length;
525     }
526 }
527 
528 /* write the memory to vmcore. 1 page per I/O. */
529 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
530                          int64_t size, Error **errp)
531 {
532     ERRP_GUARD();
533     int64_t i;
534 
535     for (i = 0; i < size / s->dump_info.page_size; i++) {
536         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
537                    s->dump_info.page_size, errp);
538         if (*errp) {
539             return;
540         }
541     }
542 
543     if ((size % s->dump_info.page_size) != 0) {
544         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
545                    size % s->dump_info.page_size, errp);
546         if (*errp) {
547             return;
548         }
549     }
550 }
551 
552 /* get the memory's offset and size in the vmcore */
553 static void get_offset_range(hwaddr phys_addr,
554                              ram_addr_t mapping_length,
555                              DumpState *s,
556                              hwaddr *p_offset,
557                              hwaddr *p_filesz)
558 {
559     GuestPhysBlock *block;
560     hwaddr offset = s->memory_offset;
561     int64_t size_in_block, start;
562 
563     /* When the memory is not stored into vmcore, offset will be -1 */
564     *p_offset = -1;
565     *p_filesz = 0;
566 
567     if (dump_has_filter(s)) {
568         if (phys_addr < s->filter_area_begin ||
569             phys_addr >= s->filter_area_begin + s->filter_area_length) {
570             return;
571         }
572     }
573 
574     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
575         if (dump_has_filter(s)) {
576             if (block->target_start >= s->filter_area_begin + s->filter_area_length ||
577                 block->target_end <= s->filter_area_begin) {
578                 /* This block is out of the range */
579                 continue;
580             }
581 
582             if (s->filter_area_begin <= block->target_start) {
583                 start = block->target_start;
584             } else {
585                 start = s->filter_area_begin;
586             }
587 
588             size_in_block = block->target_end - start;
589             if (s->filter_area_begin + s->filter_area_length < block->target_end) {
590                 size_in_block -= block->target_end - (s->filter_area_begin + s->filter_area_length);
591             }
592         } else {
593             start = block->target_start;
594             size_in_block = block->target_end - block->target_start;
595         }
596 
597         if (phys_addr >= start && phys_addr < start + size_in_block) {
598             *p_offset = phys_addr - start + offset;
599 
600             /* The offset range mapped from the vmcore file must not spill over
601              * the GuestPhysBlock, clamp it. The rest of the mapping will be
602              * zero-filled in memory at load time; see
603              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
604              */
605             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
606                         mapping_length :
607                         size_in_block - (phys_addr - start);
608             return;
609         }
610 
611         offset += size_in_block;
612     }
613 }
614 
615 static void write_elf_phdr_loads(DumpState *s, Error **errp)
616 {
617     ERRP_GUARD();
618     hwaddr offset, filesz;
619     MemoryMapping *memory_mapping;
620     uint32_t phdr_index = 1;
621 
622     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
623         get_offset_range(memory_mapping->phys_addr,
624                          memory_mapping->length,
625                          s, &offset, &filesz);
626         if (dump_is_64bit(s)) {
627             write_elf64_load(s, memory_mapping, phdr_index++, offset,
628                              filesz, errp);
629         } else {
630             write_elf32_load(s, memory_mapping, phdr_index++, offset,
631                              filesz, errp);
632         }
633 
634         if (*errp) {
635             return;
636         }
637 
638         if (phdr_index >= s->phdr_num) {
639             break;
640         }
641     }
642 }
643 
644 static void write_elf_notes(DumpState *s, Error **errp)
645 {
646     if (dump_is_64bit(s)) {
647         write_elf64_notes(fd_write_vmcore, s, errp);
648     } else {
649         write_elf32_notes(fd_write_vmcore, s, errp);
650     }
651 }
652 
653 /* write elf header, PT_NOTE and elf note to vmcore. */
654 static void dump_begin(DumpState *s, Error **errp)
655 {
656     ERRP_GUARD();
657 
658     /*
659      * the vmcore's format is:
660      *   --------------
661      *   |  elf header |
662      *   --------------
663      *   |  sctn_hdr   |
664      *   --------------
665      *   |  PT_NOTE    |
666      *   --------------
667      *   |  PT_LOAD    |
668      *   --------------
669      *   |  ......     |
670      *   --------------
671      *   |  PT_LOAD    |
672      *   --------------
673      *   |  elf note   |
674      *   --------------
675      *   |  memory     |
676      *   --------------
677      *
678      * we only know where the memory is saved after we write elf note into
679      * vmcore.
680      */
681 
682     /* write elf header to vmcore */
683     write_elf_header(s, errp);
684     if (*errp) {
685         return;
686     }
687 
688     /* write section headers to vmcore */
689     write_elf_section_headers(s, errp);
690     if (*errp) {
691         return;
692     }
693 
694     /* write PT_NOTE to vmcore */
695     write_elf_phdr_note(s, errp);
696     if (*errp) {
697         return;
698     }
699 
700     /* write all PT_LOADs to vmcore */
701     write_elf_phdr_loads(s, errp);
702     if (*errp) {
703         return;
704     }
705 
706     /* write notes to vmcore */
707     write_elf_notes(s, errp);
708 }
709 
710 int64_t dump_filtered_memblock_size(GuestPhysBlock *block,
711                                     int64_t filter_area_start,
712                                     int64_t filter_area_length)
713 {
714     int64_t size, left, right;
715 
716     /* No filter, return full size */
717     if (!filter_area_length) {
718         return block->target_end - block->target_start;
719     }
720 
721     /* calculate the overlapped region. */
722     left = MAX(filter_area_start, block->target_start);
723     right = MIN(filter_area_start + filter_area_length, block->target_end);
724     size = right - left;
725     size = size > 0 ? size : 0;
726 
727     return size;
728 }
729 
730 int64_t dump_filtered_memblock_start(GuestPhysBlock *block,
731                                      int64_t filter_area_start,
732                                      int64_t filter_area_length)
733 {
734     if (filter_area_length) {
735         /* return -1 if the block is not within filter area */
736         if (block->target_start >= filter_area_start + filter_area_length ||
737             block->target_end <= filter_area_start) {
738             return -1;
739         }
740 
741         if (filter_area_start > block->target_start) {
742             return filter_area_start - block->target_start;
743         }
744     }
745 
746     return 0;
747 }
748 
749 /* write all memory to vmcore */
750 static void dump_iterate(DumpState *s, Error **errp)
751 {
752     ERRP_GUARD();
753     GuestPhysBlock *block;
754     int64_t memblock_size, memblock_start;
755 
756     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
757         memblock_start = dump_filtered_memblock_start(block, s->filter_area_begin, s->filter_area_length);
758         if (memblock_start == -1) {
759             continue;
760         }
761 
762         memblock_size = dump_filtered_memblock_size(block, s->filter_area_begin, s->filter_area_length);
763 
764         /* Write the memory to file */
765         write_memory(s, block, memblock_start, memblock_size, errp);
766         if (*errp) {
767             return;
768         }
769     }
770 }
771 
772 static void dump_end(DumpState *s, Error **errp)
773 {
774     int rc;
775 
776     if (s->elf_section_data_size) {
777         s->elf_section_data = g_malloc0(s->elf_section_data_size);
778     }
779 
780     /* Adds the architecture defined section data to s->elf_section_data  */
781     if (s->dump_info.arch_sections_write_fn &&
782         s->elf_section_data_size) {
783         rc = s->dump_info.arch_sections_write_fn(s, s->elf_section_data);
784         if (rc) {
785             error_setg_errno(errp, rc,
786                              "dump: failed to get arch section data");
787             g_free(s->elf_section_data);
788             return;
789         }
790     }
791 
792     /* write sections to vmcore */
793     write_elf_sections(s, errp);
794 }
795 
796 static void create_vmcore(DumpState *s, Error **errp)
797 {
798     ERRP_GUARD();
799 
800     dump_begin(s, errp);
801     if (*errp) {
802         return;
803     }
804 
805     /* Iterate over memory and dump it to file */
806     dump_iterate(s, errp);
807     if (*errp) {
808         return;
809     }
810 
811     /* Write the section data */
812     dump_end(s, errp);
813 }
814 
815 static int write_start_flat_header(int fd)
816 {
817     MakedumpfileHeader *mh;
818     int ret = 0;
819 
820     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
821     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
822 
823     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
824            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
825 
826     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
827     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
828 
829     size_t written_size;
830     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
831     if (written_size != MAX_SIZE_MDF_HEADER) {
832         ret = -1;
833     }
834 
835     g_free(mh);
836     return ret;
837 }
838 
839 static int write_end_flat_header(int fd)
840 {
841     MakedumpfileDataHeader mdh;
842 
843     mdh.offset = END_FLAG_FLAT_HEADER;
844     mdh.buf_size = END_FLAG_FLAT_HEADER;
845 
846     size_t written_size;
847     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
848     if (written_size != sizeof(mdh)) {
849         return -1;
850     }
851 
852     return 0;
853 }
854 
855 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
856 {
857     size_t written_size;
858     MakedumpfileDataHeader mdh;
859 
860     mdh.offset = cpu_to_be64(offset);
861     mdh.buf_size = cpu_to_be64(size);
862 
863     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
864     if (written_size != sizeof(mdh)) {
865         return -1;
866     }
867 
868     written_size = qemu_write_full(fd, buf, size);
869     if (written_size != size) {
870         return -1;
871     }
872 
873     return 0;
874 }
875 
876 static int buf_write_note(const void *buf, size_t size, void *opaque)
877 {
878     DumpState *s = opaque;
879 
880     /* note_buf is not enough */
881     if (s->note_buf_offset + size > s->note_size) {
882         return -1;
883     }
884 
885     memcpy(s->note_buf + s->note_buf_offset, buf, size);
886 
887     s->note_buf_offset += size;
888 
889     return 0;
890 }
891 
892 /*
893  * This function retrieves various sizes from an elf header.
894  *
895  * @note has to be a valid ELF note. The return sizes are unmodified
896  * (not padded or rounded up to be multiple of 4).
897  */
898 static void get_note_sizes(DumpState *s, const void *note,
899                            uint64_t *note_head_size,
900                            uint64_t *name_size,
901                            uint64_t *desc_size)
902 {
903     uint64_t note_head_sz;
904     uint64_t name_sz;
905     uint64_t desc_sz;
906 
907     if (dump_is_64bit(s)) {
908         const Elf64_Nhdr *hdr = note;
909         note_head_sz = sizeof(Elf64_Nhdr);
910         name_sz = tswap64(hdr->n_namesz);
911         desc_sz = tswap64(hdr->n_descsz);
912     } else {
913         const Elf32_Nhdr *hdr = note;
914         note_head_sz = sizeof(Elf32_Nhdr);
915         name_sz = tswap32(hdr->n_namesz);
916         desc_sz = tswap32(hdr->n_descsz);
917     }
918 
919     if (note_head_size) {
920         *note_head_size = note_head_sz;
921     }
922     if (name_size) {
923         *name_size = name_sz;
924     }
925     if (desc_size) {
926         *desc_size = desc_sz;
927     }
928 }
929 
930 static bool note_name_equal(DumpState *s,
931                             const uint8_t *note, const char *name)
932 {
933     int len = strlen(name) + 1;
934     uint64_t head_size, name_size;
935 
936     get_note_sizes(s, note, &head_size, &name_size, NULL);
937     head_size = ROUND_UP(head_size, 4);
938 
939     return name_size == len && memcmp(note + head_size, name, len) == 0;
940 }
941 
942 /* write common header, sub header and elf note to vmcore */
943 static void create_header32(DumpState *s, Error **errp)
944 {
945     ERRP_GUARD();
946     DiskDumpHeader32 *dh = NULL;
947     KdumpSubHeader32 *kh = NULL;
948     size_t size;
949     uint32_t block_size;
950     uint32_t sub_hdr_size;
951     uint32_t bitmap_blocks;
952     uint32_t status = 0;
953     uint64_t offset_note;
954 
955     /* write common header, the version of kdump-compressed format is 6th */
956     size = sizeof(DiskDumpHeader32);
957     dh = g_malloc0(size);
958 
959     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
960     dh->header_version = cpu_to_dump32(s, 6);
961     block_size = s->dump_info.page_size;
962     dh->block_size = cpu_to_dump32(s, block_size);
963     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
964     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
965     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
966     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
967     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
968     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
969     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
970     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
971     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
972 
973     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
974         status |= DUMP_DH_COMPRESSED_ZLIB;
975     }
976 #ifdef CONFIG_LZO
977     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
978         status |= DUMP_DH_COMPRESSED_LZO;
979     }
980 #endif
981 #ifdef CONFIG_SNAPPY
982     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
983         status |= DUMP_DH_COMPRESSED_SNAPPY;
984     }
985 #endif
986     dh->status = cpu_to_dump32(s, status);
987 
988     if (write_buffer(s->fd, 0, dh, size) < 0) {
989         error_setg(errp, "dump: failed to write disk dump header");
990         goto out;
991     }
992 
993     /* write sub header */
994     size = sizeof(KdumpSubHeader32);
995     kh = g_malloc0(size);
996 
997     /* 64bit max_mapnr_64 */
998     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
999     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
1000     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1001 
1002     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1003     if (s->guest_note &&
1004         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1005         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1006 
1007         get_note_sizes(s, s->guest_note,
1008                        &hsize, &name_size, &size_vmcoreinfo_desc);
1009         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1010             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1011         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1012         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
1013     }
1014 
1015     kh->offset_note = cpu_to_dump64(s, offset_note);
1016     kh->note_size = cpu_to_dump32(s, s->note_size);
1017 
1018     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
1019                      block_size, kh, size) < 0) {
1020         error_setg(errp, "dump: failed to write kdump sub header");
1021         goto out;
1022     }
1023 
1024     /* write note */
1025     s->note_buf = g_malloc0(s->note_size);
1026     s->note_buf_offset = 0;
1027 
1028     /* use s->note_buf to store notes temporarily */
1029     write_elf32_notes(buf_write_note, s, errp);
1030     if (*errp) {
1031         goto out;
1032     }
1033     if (write_buffer(s->fd, offset_note, s->note_buf,
1034                      s->note_size) < 0) {
1035         error_setg(errp, "dump: failed to write notes");
1036         goto out;
1037     }
1038 
1039     /* get offset of dump_bitmap */
1040     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1041                              block_size;
1042 
1043     /* get offset of page */
1044     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1045                      block_size;
1046 
1047 out:
1048     g_free(dh);
1049     g_free(kh);
1050     g_free(s->note_buf);
1051 }
1052 
1053 /* write common header, sub header and elf note to vmcore */
1054 static void create_header64(DumpState *s, Error **errp)
1055 {
1056     ERRP_GUARD();
1057     DiskDumpHeader64 *dh = NULL;
1058     KdumpSubHeader64 *kh = NULL;
1059     size_t size;
1060     uint32_t block_size;
1061     uint32_t sub_hdr_size;
1062     uint32_t bitmap_blocks;
1063     uint32_t status = 0;
1064     uint64_t offset_note;
1065 
1066     /* write common header, the version of kdump-compressed format is 6th */
1067     size = sizeof(DiskDumpHeader64);
1068     dh = g_malloc0(size);
1069 
1070     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
1071     dh->header_version = cpu_to_dump32(s, 6);
1072     block_size = s->dump_info.page_size;
1073     dh->block_size = cpu_to_dump32(s, block_size);
1074     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
1075     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
1076     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
1077     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
1078     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
1079     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
1080     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
1081     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
1082     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
1083 
1084     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
1085         status |= DUMP_DH_COMPRESSED_ZLIB;
1086     }
1087 #ifdef CONFIG_LZO
1088     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
1089         status |= DUMP_DH_COMPRESSED_LZO;
1090     }
1091 #endif
1092 #ifdef CONFIG_SNAPPY
1093     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1094         status |= DUMP_DH_COMPRESSED_SNAPPY;
1095     }
1096 #endif
1097     dh->status = cpu_to_dump32(s, status);
1098 
1099     if (write_buffer(s->fd, 0, dh, size) < 0) {
1100         error_setg(errp, "dump: failed to write disk dump header");
1101         goto out;
1102     }
1103 
1104     /* write sub header */
1105     size = sizeof(KdumpSubHeader64);
1106     kh = g_malloc0(size);
1107 
1108     /* 64bit max_mapnr_64 */
1109     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1110     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
1111     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1112 
1113     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1114     if (s->guest_note &&
1115         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1116         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1117 
1118         get_note_sizes(s, s->guest_note,
1119                        &hsize, &name_size, &size_vmcoreinfo_desc);
1120         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1121             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1122         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1123         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
1124     }
1125 
1126     kh->offset_note = cpu_to_dump64(s, offset_note);
1127     kh->note_size = cpu_to_dump64(s, s->note_size);
1128 
1129     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
1130                      block_size, kh, size) < 0) {
1131         error_setg(errp, "dump: failed to write kdump sub header");
1132         goto out;
1133     }
1134 
1135     /* write note */
1136     s->note_buf = g_malloc0(s->note_size);
1137     s->note_buf_offset = 0;
1138 
1139     /* use s->note_buf to store notes temporarily */
1140     write_elf64_notes(buf_write_note, s, errp);
1141     if (*errp) {
1142         goto out;
1143     }
1144 
1145     if (write_buffer(s->fd, offset_note, s->note_buf,
1146                      s->note_size) < 0) {
1147         error_setg(errp, "dump: failed to write notes");
1148         goto out;
1149     }
1150 
1151     /* get offset of dump_bitmap */
1152     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1153                              block_size;
1154 
1155     /* get offset of page */
1156     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1157                      block_size;
1158 
1159 out:
1160     g_free(dh);
1161     g_free(kh);
1162     g_free(s->note_buf);
1163 }
1164 
1165 static void write_dump_header(DumpState *s, Error **errp)
1166 {
1167     if (dump_is_64bit(s)) {
1168         create_header64(s, errp);
1169     } else {
1170         create_header32(s, errp);
1171     }
1172 }
1173 
1174 static size_t dump_bitmap_get_bufsize(DumpState *s)
1175 {
1176     return s->dump_info.page_size;
1177 }
1178 
1179 /*
1180  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1181  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1182  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1183  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1184  * vmcore, ie. synchronizing un-sync bit into vmcore.
1185  */
1186 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1187                            uint8_t *buf, DumpState *s)
1188 {
1189     off_t old_offset, new_offset;
1190     off_t offset_bitmap1, offset_bitmap2;
1191     uint32_t byte, bit;
1192     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1193     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1194 
1195     /* should not set the previous place */
1196     assert(last_pfn <= pfn);
1197 
1198     /*
1199      * if the bit needed to be set is not cached in buf, flush the data in buf
1200      * to vmcore firstly.
1201      * making new_offset be bigger than old_offset can also sync remained data
1202      * into vmcore.
1203      */
1204     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1205     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1206 
1207     while (old_offset < new_offset) {
1208         /* calculate the offset and write dump_bitmap */
1209         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1210         if (write_buffer(s->fd, offset_bitmap1, buf,
1211                          bitmap_bufsize) < 0) {
1212             return -1;
1213         }
1214 
1215         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1216         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1217                          old_offset;
1218         if (write_buffer(s->fd, offset_bitmap2, buf,
1219                          bitmap_bufsize) < 0) {
1220             return -1;
1221         }
1222 
1223         memset(buf, 0, bitmap_bufsize);
1224         old_offset += bitmap_bufsize;
1225     }
1226 
1227     /* get the exact place of the bit in the buf, and set it */
1228     byte = (pfn % bits_per_buf) / CHAR_BIT;
1229     bit = (pfn % bits_per_buf) % CHAR_BIT;
1230     if (value) {
1231         buf[byte] |= 1u << bit;
1232     } else {
1233         buf[byte] &= ~(1u << bit);
1234     }
1235 
1236     return 0;
1237 }
1238 
1239 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1240 {
1241     int target_page_shift = ctz32(s->dump_info.page_size);
1242 
1243     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1244 }
1245 
1246 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1247 {
1248     int target_page_shift = ctz32(s->dump_info.page_size);
1249 
1250     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1251 }
1252 
1253 /*
1254  * Return the page frame number and the page content in *bufptr. bufptr can be
1255  * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated
1256  * memory. This is not necessarily the memory returned.
1257  */
1258 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1259                           uint8_t **bufptr, DumpState *s)
1260 {
1261     GuestPhysBlock *block = *blockptr;
1262     uint32_t page_size = s->dump_info.page_size;
1263     uint8_t *buf = NULL, *hbuf;
1264     hwaddr addr;
1265 
1266     /* block == NULL means the start of the iteration */
1267     if (!block) {
1268         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1269         *blockptr = block;
1270         addr = block->target_start;
1271         *pfnptr = dump_paddr_to_pfn(s, addr);
1272     } else {
1273         *pfnptr += 1;
1274         addr = dump_pfn_to_paddr(s, *pfnptr);
1275     }
1276     assert(block != NULL);
1277 
1278     while (1) {
1279         if (addr >= block->target_start && addr < block->target_end) {
1280             size_t n = MIN(block->target_end - addr, page_size - addr % page_size);
1281             hbuf = block->host_addr + (addr - block->target_start);
1282             if (!buf) {
1283                 if (n == page_size) {
1284                     /* this is a whole target page, go for it */
1285                     assert(addr % page_size == 0);
1286                     buf = hbuf;
1287                     break;
1288                 } else if (bufptr) {
1289                     assert(*bufptr);
1290                     buf = *bufptr;
1291                     memset(buf, 0, page_size);
1292                 } else {
1293                     return true;
1294                 }
1295             }
1296 
1297             memcpy(buf + addr % page_size, hbuf, n);
1298             addr += n;
1299             if (addr % page_size == 0) {
1300                 /* we filled up the page */
1301                 break;
1302             }
1303         } else {
1304             /* the next page is in the next block */
1305             *blockptr = block = QTAILQ_NEXT(block, next);
1306             if (!block) {
1307                 break;
1308             }
1309 
1310             addr = block->target_start;
1311             /* are we still in the same page? */
1312             if (dump_paddr_to_pfn(s, addr) != *pfnptr) {
1313                 if (buf) {
1314                     /* no, but we already filled something earlier, return it */
1315                     break;
1316                 } else {
1317                     /* else continue from there */
1318                     *pfnptr = dump_paddr_to_pfn(s, addr);
1319                 }
1320             }
1321         }
1322     }
1323 
1324     if (bufptr) {
1325         *bufptr = buf;
1326     }
1327 
1328     return buf != NULL;
1329 }
1330 
1331 static void write_dump_bitmap(DumpState *s, Error **errp)
1332 {
1333     int ret = 0;
1334     uint64_t last_pfn, pfn;
1335     void *dump_bitmap_buf;
1336     size_t num_dumpable;
1337     GuestPhysBlock *block_iter = NULL;
1338     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1339     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1340 
1341     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1342     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1343 
1344     num_dumpable = 0;
1345     last_pfn = 0;
1346 
1347     /*
1348      * exam memory page by page, and set the bit in dump_bitmap corresponded
1349      * to the existing page.
1350      */
1351     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1352         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1353         if (ret < 0) {
1354             error_setg(errp, "dump: failed to set dump_bitmap");
1355             goto out;
1356         }
1357 
1358         last_pfn = pfn;
1359         num_dumpable++;
1360     }
1361 
1362     /*
1363      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1364      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1365      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1366      */
1367     if (num_dumpable > 0) {
1368         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1369                               dump_bitmap_buf, s);
1370         if (ret < 0) {
1371             error_setg(errp, "dump: failed to sync dump_bitmap");
1372             goto out;
1373         }
1374     }
1375 
1376     /* number of dumpable pages that will be dumped later */
1377     s->num_dumpable = num_dumpable;
1378 
1379 out:
1380     g_free(dump_bitmap_buf);
1381 }
1382 
1383 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1384                                off_t offset)
1385 {
1386     data_cache->fd = s->fd;
1387     data_cache->data_size = 0;
1388     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1389     data_cache->buf = g_malloc0(data_cache->buf_size);
1390     data_cache->offset = offset;
1391 }
1392 
1393 static int write_cache(DataCache *dc, const void *buf, size_t size,
1394                        bool flag_sync)
1395 {
1396     /*
1397      * dc->buf_size should not be less than size, otherwise dc will never be
1398      * enough
1399      */
1400     assert(size <= dc->buf_size);
1401 
1402     /*
1403      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1404      * otherwise check if the space is enough for caching data in buf, if not,
1405      * write the data in dc->buf to dc->fd and reset dc->buf
1406      */
1407     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1408         (flag_sync && dc->data_size > 0)) {
1409         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1410             return -1;
1411         }
1412 
1413         dc->offset += dc->data_size;
1414         dc->data_size = 0;
1415     }
1416 
1417     if (!flag_sync) {
1418         memcpy(dc->buf + dc->data_size, buf, size);
1419         dc->data_size += size;
1420     }
1421 
1422     return 0;
1423 }
1424 
1425 static void free_data_cache(DataCache *data_cache)
1426 {
1427     g_free(data_cache->buf);
1428 }
1429 
1430 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1431 {
1432     switch (flag_compress) {
1433     case DUMP_DH_COMPRESSED_ZLIB:
1434         return compressBound(page_size);
1435 
1436     case DUMP_DH_COMPRESSED_LZO:
1437         /*
1438          * LZO will expand incompressible data by a little amount. Please check
1439          * the following URL to see the expansion calculation:
1440          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1441          */
1442         return page_size + page_size / 16 + 64 + 3;
1443 
1444 #ifdef CONFIG_SNAPPY
1445     case DUMP_DH_COMPRESSED_SNAPPY:
1446         return snappy_max_compressed_length(page_size);
1447 #endif
1448     }
1449     return 0;
1450 }
1451 
1452 static void write_dump_pages(DumpState *s, Error **errp)
1453 {
1454     int ret = 0;
1455     DataCache page_desc, page_data;
1456     size_t len_buf_out, size_out;
1457 #ifdef CONFIG_LZO
1458     lzo_bytep wrkmem = NULL;
1459 #endif
1460     uint8_t *buf_out = NULL;
1461     off_t offset_desc, offset_data;
1462     PageDescriptor pd, pd_zero;
1463     uint8_t *buf;
1464     GuestPhysBlock *block_iter = NULL;
1465     uint64_t pfn_iter;
1466     g_autofree uint8_t *page = NULL;
1467 
1468     /* get offset of page_desc and page_data in dump file */
1469     offset_desc = s->offset_page;
1470     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1471 
1472     prepare_data_cache(&page_desc, s, offset_desc);
1473     prepare_data_cache(&page_data, s, offset_data);
1474 
1475     /* prepare buffer to store compressed data */
1476     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1477     assert(len_buf_out != 0);
1478 
1479 #ifdef CONFIG_LZO
1480     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1481 #endif
1482 
1483     buf_out = g_malloc(len_buf_out);
1484 
1485     /*
1486      * init zero page's page_desc and page_data, because every zero page
1487      * uses the same page_data
1488      */
1489     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1490     pd_zero.flags = cpu_to_dump32(s, 0);
1491     pd_zero.offset = cpu_to_dump64(s, offset_data);
1492     pd_zero.page_flags = cpu_to_dump64(s, 0);
1493     buf = g_malloc0(s->dump_info.page_size);
1494     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1495     g_free(buf);
1496     if (ret < 0) {
1497         error_setg(errp, "dump: failed to write page data (zero page)");
1498         goto out;
1499     }
1500 
1501     offset_data += s->dump_info.page_size;
1502     page = g_malloc(s->dump_info.page_size);
1503 
1504     /*
1505      * dump memory to vmcore page by page. zero page will all be resided in the
1506      * first page of page section
1507      */
1508     for (buf = page; get_next_page(&block_iter, &pfn_iter, &buf, s); buf = page) {
1509         /* check zero page */
1510         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1511             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1512                               false);
1513             if (ret < 0) {
1514                 error_setg(errp, "dump: failed to write page desc");
1515                 goto out;
1516             }
1517         } else {
1518             /*
1519              * not zero page, then:
1520              * 1. compress the page
1521              * 2. write the compressed page into the cache of page_data
1522              * 3. get page desc of the compressed page and write it into the
1523              *    cache of page_desc
1524              *
1525              * only one compression format will be used here, for
1526              * s->flag_compress is set. But when compression fails to work,
1527              * we fall back to save in plaintext.
1528              */
1529              size_out = len_buf_out;
1530              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1531                     (compress2(buf_out, (uLongf *)&size_out, buf,
1532                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1533                     (size_out < s->dump_info.page_size)) {
1534                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1535                 pd.size  = cpu_to_dump32(s, size_out);
1536 
1537                 ret = write_cache(&page_data, buf_out, size_out, false);
1538                 if (ret < 0) {
1539                     error_setg(errp, "dump: failed to write page data");
1540                     goto out;
1541                 }
1542 #ifdef CONFIG_LZO
1543             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1544                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1545                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1546                     (size_out < s->dump_info.page_size)) {
1547                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1548                 pd.size  = cpu_to_dump32(s, size_out);
1549 
1550                 ret = write_cache(&page_data, buf_out, size_out, false);
1551                 if (ret < 0) {
1552                     error_setg(errp, "dump: failed to write page data");
1553                     goto out;
1554                 }
1555 #endif
1556 #ifdef CONFIG_SNAPPY
1557             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1558                     (snappy_compress((char *)buf, s->dump_info.page_size,
1559                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1560                     (size_out < s->dump_info.page_size)) {
1561                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1562                 pd.size  = cpu_to_dump32(s, size_out);
1563 
1564                 ret = write_cache(&page_data, buf_out, size_out, false);
1565                 if (ret < 0) {
1566                     error_setg(errp, "dump: failed to write page data");
1567                     goto out;
1568                 }
1569 #endif
1570             } else {
1571                 /*
1572                  * fall back to save in plaintext, size_out should be
1573                  * assigned the target's page size
1574                  */
1575                 pd.flags = cpu_to_dump32(s, 0);
1576                 size_out = s->dump_info.page_size;
1577                 pd.size = cpu_to_dump32(s, size_out);
1578 
1579                 ret = write_cache(&page_data, buf,
1580                                   s->dump_info.page_size, false);
1581                 if (ret < 0) {
1582                     error_setg(errp, "dump: failed to write page data");
1583                     goto out;
1584                 }
1585             }
1586 
1587             /* get and write page desc here */
1588             pd.page_flags = cpu_to_dump64(s, 0);
1589             pd.offset = cpu_to_dump64(s, offset_data);
1590             offset_data += size_out;
1591 
1592             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1593             if (ret < 0) {
1594                 error_setg(errp, "dump: failed to write page desc");
1595                 goto out;
1596             }
1597         }
1598         s->written_size += s->dump_info.page_size;
1599     }
1600 
1601     ret = write_cache(&page_desc, NULL, 0, true);
1602     if (ret < 0) {
1603         error_setg(errp, "dump: failed to sync cache for page_desc");
1604         goto out;
1605     }
1606     ret = write_cache(&page_data, NULL, 0, true);
1607     if (ret < 0) {
1608         error_setg(errp, "dump: failed to sync cache for page_data");
1609         goto out;
1610     }
1611 
1612 out:
1613     free_data_cache(&page_desc);
1614     free_data_cache(&page_data);
1615 
1616 #ifdef CONFIG_LZO
1617     g_free(wrkmem);
1618 #endif
1619 
1620     g_free(buf_out);
1621 }
1622 
1623 static void create_kdump_vmcore(DumpState *s, Error **errp)
1624 {
1625     ERRP_GUARD();
1626     int ret;
1627 
1628     /*
1629      * the kdump-compressed format is:
1630      *                                               File offset
1631      *  +------------------------------------------+ 0x0
1632      *  |    main header (struct disk_dump_header) |
1633      *  |------------------------------------------+ block 1
1634      *  |    sub header (struct kdump_sub_header)  |
1635      *  |------------------------------------------+ block 2
1636      *  |            1st-dump_bitmap               |
1637      *  |------------------------------------------+ block 2 + X blocks
1638      *  |            2nd-dump_bitmap               | (aligned by block)
1639      *  |------------------------------------------+ block 2 + 2 * X blocks
1640      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1641      *  |  page desc for pfn 1 (struct page_desc)  |
1642      *  |                    :                     |
1643      *  |------------------------------------------| (not aligned by block)
1644      *  |         page data (pfn 0)                |
1645      *  |         page data (pfn 1)                |
1646      *  |                    :                     |
1647      *  +------------------------------------------+
1648      */
1649 
1650     ret = write_start_flat_header(s->fd);
1651     if (ret < 0) {
1652         error_setg(errp, "dump: failed to write start flat header");
1653         return;
1654     }
1655 
1656     write_dump_header(s, errp);
1657     if (*errp) {
1658         return;
1659     }
1660 
1661     write_dump_bitmap(s, errp);
1662     if (*errp) {
1663         return;
1664     }
1665 
1666     write_dump_pages(s, errp);
1667     if (*errp) {
1668         return;
1669     }
1670 
1671     ret = write_end_flat_header(s->fd);
1672     if (ret < 0) {
1673         error_setg(errp, "dump: failed to write end flat header");
1674         return;
1675     }
1676 }
1677 
1678 static int validate_start_block(DumpState *s)
1679 {
1680     GuestPhysBlock *block;
1681 
1682     if (!dump_has_filter(s)) {
1683         return 0;
1684     }
1685 
1686     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1687         /* This block is out of the range */
1688         if (block->target_start >= s->filter_area_begin + s->filter_area_length ||
1689             block->target_end <= s->filter_area_begin) {
1690             continue;
1691         }
1692         return 0;
1693    }
1694 
1695     return -1;
1696 }
1697 
1698 static void get_max_mapnr(DumpState *s)
1699 {
1700     GuestPhysBlock *last_block;
1701 
1702     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1703     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1704 }
1705 
1706 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1707 
1708 static void dump_state_prepare(DumpState *s)
1709 {
1710     /* zero the struct, setting status to active */
1711     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1712 }
1713 
1714 bool qemu_system_dump_in_progress(void)
1715 {
1716     DumpState *state = &dump_state_global;
1717     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1718 }
1719 
1720 /*
1721  * calculate total size of memory to be dumped (taking filter into
1722  * account.)
1723  */
1724 static int64_t dump_calculate_size(DumpState *s)
1725 {
1726     GuestPhysBlock *block;
1727     int64_t total = 0;
1728 
1729     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1730         total += dump_filtered_memblock_size(block,
1731                                              s->filter_area_begin,
1732                                              s->filter_area_length);
1733     }
1734 
1735     return total;
1736 }
1737 
1738 static void vmcoreinfo_update_phys_base(DumpState *s)
1739 {
1740     uint64_t size, note_head_size, name_size, phys_base;
1741     char **lines;
1742     uint8_t *vmci;
1743     size_t i;
1744 
1745     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1746         return;
1747     }
1748 
1749     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1750     note_head_size = ROUND_UP(note_head_size, 4);
1751 
1752     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1753     *(vmci + size) = '\0';
1754 
1755     lines = g_strsplit((char *)vmci, "\n", -1);
1756     for (i = 0; lines[i]; i++) {
1757         const char *prefix = NULL;
1758 
1759         if (s->dump_info.d_machine == EM_X86_64) {
1760             prefix = "NUMBER(phys_base)=";
1761         } else if (s->dump_info.d_machine == EM_AARCH64) {
1762             prefix = "NUMBER(PHYS_OFFSET)=";
1763         }
1764 
1765         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1766             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1767                               &phys_base) < 0) {
1768                 warn_report("Failed to read %s", prefix);
1769             } else {
1770                 s->dump_info.phys_base = phys_base;
1771             }
1772             break;
1773         }
1774     }
1775 
1776     g_strfreev(lines);
1777 }
1778 
1779 static void dump_init(DumpState *s, int fd, bool has_format,
1780                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1781                       int64_t begin, int64_t length, Error **errp)
1782 {
1783     ERRP_GUARD();
1784     VMCoreInfoState *vmci = vmcoreinfo_find();
1785     CPUState *cpu;
1786     int nr_cpus;
1787     int ret;
1788 
1789     s->has_format = has_format;
1790     s->format = format;
1791     s->written_size = 0;
1792 
1793     /* kdump-compressed is conflict with paging and filter */
1794     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1795         assert(!paging && !has_filter);
1796     }
1797 
1798     if (runstate_is_running()) {
1799         vm_stop(RUN_STATE_SAVE_VM);
1800         s->resume = true;
1801     } else {
1802         s->resume = false;
1803     }
1804 
1805     /* If we use KVM, we should synchronize the registers before we get dump
1806      * info or physmap info.
1807      */
1808     cpu_synchronize_all_states();
1809     nr_cpus = 0;
1810     CPU_FOREACH(cpu) {
1811         nr_cpus++;
1812     }
1813 
1814     s->fd = fd;
1815     if (has_filter && !length) {
1816         error_setg(errp, QERR_INVALID_PARAMETER, "length");
1817         goto cleanup;
1818     }
1819     s->filter_area_begin = begin;
1820     s->filter_area_length = length;
1821 
1822     /* First index is 0, it's the special null name */
1823     s->string_table_buf = g_array_new(FALSE, TRUE, 1);
1824     /*
1825      * Allocate the null name, due to the clearing option set to true
1826      * it will be 0.
1827      */
1828     g_array_set_size(s->string_table_buf, 1);
1829 
1830     memory_mapping_list_init(&s->list);
1831 
1832     guest_phys_blocks_init(&s->guest_phys_blocks);
1833     guest_phys_blocks_append(&s->guest_phys_blocks);
1834     s->total_size = dump_calculate_size(s);
1835 #ifdef DEBUG_DUMP_GUEST_MEMORY
1836     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1837 #endif
1838 
1839     /* it does not make sense to dump non-existent memory */
1840     if (!s->total_size) {
1841         error_setg(errp, "dump: no guest memory to dump");
1842         goto cleanup;
1843     }
1844 
1845     /* Is the filter filtering everything? */
1846     if (validate_start_block(s) == -1) {
1847         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1848         goto cleanup;
1849     }
1850 
1851     /* get dump info: endian, class and architecture.
1852      * If the target architecture is not supported, cpu_get_dump_info() will
1853      * return -1.
1854      */
1855     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1856     if (ret < 0) {
1857         error_setg(errp,
1858                    "dumping guest memory is not supported on this target");
1859         goto cleanup;
1860     }
1861 
1862     if (!s->dump_info.page_size) {
1863         s->dump_info.page_size = TARGET_PAGE_SIZE;
1864     }
1865 
1866     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1867                                      s->dump_info.d_machine, nr_cpus);
1868     assert(s->note_size >= 0);
1869 
1870     /*
1871      * The goal of this block is to (a) update the previously guessed
1872      * phys_base, (b) copy the guest note out of the guest.
1873      * Failure to do so is not fatal for dumping.
1874      */
1875     if (vmci) {
1876         uint64_t addr, note_head_size, name_size, desc_size;
1877         uint32_t size;
1878         uint16_t format;
1879 
1880         note_head_size = dump_is_64bit(s) ?
1881             sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr);
1882 
1883         format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1884         size = le32_to_cpu(vmci->vmcoreinfo.size);
1885         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1886         if (!vmci->has_vmcoreinfo) {
1887             warn_report("guest note is not present");
1888         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1889             warn_report("guest note size is invalid: %" PRIu32, size);
1890         } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1891             warn_report("guest note format is unsupported: %" PRIu16, format);
1892         } else {
1893             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1894             cpu_physical_memory_read(addr, s->guest_note, size);
1895 
1896             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1897             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1898                                                desc_size);
1899             if (name_size > MAX_GUEST_NOTE_SIZE ||
1900                 desc_size > MAX_GUEST_NOTE_SIZE ||
1901                 s->guest_note_size > size) {
1902                 warn_report("Invalid guest note header");
1903                 g_free(s->guest_note);
1904                 s->guest_note = NULL;
1905             } else {
1906                 vmcoreinfo_update_phys_base(s);
1907                 s->note_size += s->guest_note_size;
1908             }
1909         }
1910     }
1911 
1912     /* get memory mapping */
1913     if (paging) {
1914         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1915         if (*errp) {
1916             goto cleanup;
1917         }
1918     } else {
1919         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1920     }
1921 
1922     s->nr_cpus = nr_cpus;
1923 
1924     get_max_mapnr(s);
1925 
1926     uint64_t tmp;
1927     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1928                        s->dump_info.page_size);
1929     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1930 
1931     /* init for kdump-compressed format */
1932     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1933         switch (format) {
1934         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1935             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1936             break;
1937 
1938         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1939 #ifdef CONFIG_LZO
1940             if (lzo_init() != LZO_E_OK) {
1941                 error_setg(errp, "failed to initialize the LZO library");
1942                 goto cleanup;
1943             }
1944 #endif
1945             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1946             break;
1947 
1948         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1949             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1950             break;
1951 
1952         default:
1953             s->flag_compress = 0;
1954         }
1955 
1956         return;
1957     }
1958 
1959     if (dump_has_filter(s)) {
1960         memory_mapping_filter(&s->list, s->filter_area_begin, s->filter_area_length);
1961     }
1962 
1963     /*
1964      * The first section header is always a special one in which most
1965      * fields are 0. The section header string table is also always
1966      * set.
1967      */
1968     s->shdr_num = 2;
1969 
1970     /*
1971      * Adds the number of architecture sections to shdr_num and sets
1972      * elf_section_data_size so we know the offsets and sizes of all
1973      * parts.
1974      */
1975     if (s->dump_info.arch_sections_add_fn) {
1976         s->dump_info.arch_sections_add_fn(s);
1977     }
1978 
1979     /*
1980      * calculate shdr_num so we know the offsets and sizes of all
1981      * parts.
1982      * Calculate phdr_num
1983      *
1984      * The absolute maximum amount of phdrs is UINT32_MAX - 1 as
1985      * sh_info is 32 bit. There's special handling once we go over
1986      * UINT16_MAX - 1 but that is handled in the ehdr and section
1987      * code.
1988      */
1989     s->phdr_num = 1; /* Reserve PT_NOTE */
1990     if (s->list.num <= UINT32_MAX - 1) {
1991         s->phdr_num += s->list.num;
1992     } else {
1993         s->phdr_num = UINT32_MAX;
1994     }
1995 
1996     /*
1997      * Now that the number of section and program headers is known we
1998      * can calculate the offsets of the headers and data.
1999      */
2000     if (dump_is_64bit(s)) {
2001         s->shdr_offset = sizeof(Elf64_Ehdr);
2002         s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
2003         s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
2004     } else {
2005         s->shdr_offset = sizeof(Elf32_Ehdr);
2006         s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
2007         s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
2008     }
2009     s->memory_offset = s->note_offset + s->note_size;
2010     s->section_offset = s->memory_offset + s->total_size;
2011 
2012     return;
2013 
2014 cleanup:
2015     dump_cleanup(s);
2016 }
2017 
2018 /* this operation might be time consuming. */
2019 static void dump_process(DumpState *s, Error **errp)
2020 {
2021     ERRP_GUARD();
2022     DumpQueryResult *result = NULL;
2023 
2024     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2025 #ifdef TARGET_X86_64
2026         create_win_dump(s, errp);
2027 #endif
2028     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
2029         create_kdump_vmcore(s, errp);
2030     } else {
2031         create_vmcore(s, errp);
2032     }
2033 
2034     /* make sure status is written after written_size updates */
2035     smp_wmb();
2036     qatomic_set(&s->status,
2037                (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
2038 
2039     /* send DUMP_COMPLETED message (unconditionally) */
2040     result = qmp_query_dump(NULL);
2041     /* should never fail */
2042     assert(result);
2043     qapi_event_send_dump_completed(result,
2044                                    *errp ? error_get_pretty(*errp) : NULL);
2045     qapi_free_DumpQueryResult(result);
2046 
2047     dump_cleanup(s);
2048 }
2049 
2050 static void *dump_thread(void *data)
2051 {
2052     DumpState *s = (DumpState *)data;
2053     dump_process(s, NULL);
2054     return NULL;
2055 }
2056 
2057 DumpQueryResult *qmp_query_dump(Error **errp)
2058 {
2059     DumpQueryResult *result = g_new(DumpQueryResult, 1);
2060     DumpState *state = &dump_state_global;
2061     result->status = qatomic_read(&state->status);
2062     /* make sure we are reading status and written_size in order */
2063     smp_rmb();
2064     result->completed = state->written_size;
2065     result->total = state->total_size;
2066     return result;
2067 }
2068 
2069 void qmp_dump_guest_memory(bool paging, const char *file,
2070                            bool has_detach, bool detach,
2071                            bool has_begin, int64_t begin, bool has_length,
2072                            int64_t length, bool has_format,
2073                            DumpGuestMemoryFormat format, Error **errp)
2074 {
2075     ERRP_GUARD();
2076     const char *p;
2077     int fd = -1;
2078     DumpState *s;
2079     bool detach_p = false;
2080 
2081     if (runstate_check(RUN_STATE_INMIGRATE)) {
2082         error_setg(errp, "Dump not allowed during incoming migration.");
2083         return;
2084     }
2085 
2086     /* if there is a dump in background, we should wait until the dump
2087      * finished */
2088     if (qemu_system_dump_in_progress()) {
2089         error_setg(errp, "There is a dump in process, please wait.");
2090         return;
2091     }
2092 
2093     /*
2094      * kdump-compressed format need the whole memory dumped, so paging or
2095      * filter is not supported here.
2096      */
2097     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
2098         (paging || has_begin || has_length)) {
2099         error_setg(errp, "kdump-compressed format doesn't support paging or "
2100                          "filter");
2101         return;
2102     }
2103     if (has_begin && !has_length) {
2104         error_setg(errp, QERR_MISSING_PARAMETER, "length");
2105         return;
2106     }
2107     if (!has_begin && has_length) {
2108         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
2109         return;
2110     }
2111     if (has_detach) {
2112         detach_p = detach;
2113     }
2114 
2115     /* check whether lzo/snappy is supported */
2116 #ifndef CONFIG_LZO
2117     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
2118         error_setg(errp, "kdump-lzo is not available now");
2119         return;
2120     }
2121 #endif
2122 
2123 #ifndef CONFIG_SNAPPY
2124     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
2125         error_setg(errp, "kdump-snappy is not available now");
2126         return;
2127     }
2128 #endif
2129 
2130 #ifndef TARGET_X86_64
2131     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2132         error_setg(errp, "Windows dump is only available for x86-64");
2133         return;
2134     }
2135 #endif
2136 
2137 #if !defined(WIN32)
2138     if (strstart(file, "fd:", &p)) {
2139         fd = monitor_get_fd(monitor_cur(), p, errp);
2140         if (fd == -1) {
2141             return;
2142         }
2143     }
2144 #endif
2145 
2146     if  (strstart(file, "file:", &p)) {
2147         fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
2148         if (fd < 0) {
2149             error_setg_file_open(errp, errno, p);
2150             return;
2151         }
2152     }
2153 
2154     if (fd == -1) {
2155         error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
2156         return;
2157     }
2158 
2159     if (!dump_migration_blocker) {
2160         error_setg(&dump_migration_blocker,
2161                    "Live migration disabled: dump-guest-memory in progress");
2162     }
2163 
2164     /*
2165      * Allows even for -only-migratable, but forbid migration during the
2166      * process of dump guest memory.
2167      */
2168     if (migrate_add_blocker_internal(dump_migration_blocker, errp)) {
2169         /* Remember to release the fd before passing it over to dump state */
2170         close(fd);
2171         return;
2172     }
2173 
2174     s = &dump_state_global;
2175     dump_state_prepare(s);
2176 
2177     dump_init(s, fd, has_format, format, paging, has_begin,
2178               begin, length, errp);
2179     if (*errp) {
2180         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2181         return;
2182     }
2183 
2184     if (detach_p) {
2185         /* detached dump */
2186         s->detached = true;
2187         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2188                            s, QEMU_THREAD_DETACHED);
2189     } else {
2190         /* sync dump */
2191         dump_process(s, errp);
2192     }
2193 }
2194 
2195 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2196 {
2197     DumpGuestMemoryCapability *cap =
2198                                   g_new0(DumpGuestMemoryCapability, 1);
2199     DumpGuestMemoryFormatList **tail = &cap->formats;
2200 
2201     /* elf is always available */
2202     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2203 
2204     /* kdump-zlib is always available */
2205     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2206 
2207     /* add new item if kdump-lzo is available */
2208 #ifdef CONFIG_LZO
2209     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2210 #endif
2211 
2212     /* add new item if kdump-snappy is available */
2213 #ifdef CONFIG_SNAPPY
2214     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2215 #endif
2216 
2217     /* Windows dump is available only if target is x86_64 */
2218 #ifdef TARGET_X86_64
2219     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2220 #endif
2221 
2222     return cap;
2223 }
2224