xref: /openbmc/qemu/dump/dump.c (revision 080741ab)
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     ERRP_GUARD();
361     Elf32_Phdr phdr32;
362     Elf64_Phdr phdr64;
363     void *phdr;
364     size_t size;
365     int ret;
366 
367     if (dump_is_64bit(s)) {
368         prepare_elf64_phdr_note(s, &phdr64);
369         size = sizeof(phdr64);
370         phdr = &phdr64;
371     } else {
372         prepare_elf32_phdr_note(s, &phdr32);
373         size = sizeof(phdr32);
374         phdr = &phdr32;
375     }
376 
377     ret = fd_write_vmcore(phdr, size, s);
378     if (ret < 0) {
379         error_setg_errno(errp, -ret,
380                          "dump: failed to write program header table");
381     }
382 }
383 
384 static void prepare_elf_section_hdr_zero(DumpState *s)
385 {
386     if (dump_is_64bit(s)) {
387         Elf64_Shdr *shdr64 = s->elf_section_hdrs;
388 
389         shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
390     } else {
391         Elf32_Shdr *shdr32 = s->elf_section_hdrs;
392 
393         shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
394     }
395 }
396 
397 static void prepare_elf_section_hdr_string(DumpState *s, void *buff)
398 {
399     uint64_t index = s->string_table_buf->len;
400     const char strtab[] = ".shstrtab";
401     Elf32_Shdr shdr32 = {};
402     Elf64_Shdr shdr64 = {};
403     int shdr_size;
404     void *shdr;
405 
406     g_array_append_vals(s->string_table_buf, strtab, sizeof(strtab));
407     if (dump_is_64bit(s)) {
408         shdr_size = sizeof(Elf64_Shdr);
409         shdr64.sh_type = SHT_STRTAB;
410         shdr64.sh_offset = s->section_offset + s->elf_section_data_size;
411         shdr64.sh_name = index;
412         shdr64.sh_size = s->string_table_buf->len;
413         shdr = &shdr64;
414     } else {
415         shdr_size = sizeof(Elf32_Shdr);
416         shdr32.sh_type = SHT_STRTAB;
417         shdr32.sh_offset = s->section_offset + s->elf_section_data_size;
418         shdr32.sh_name = index;
419         shdr32.sh_size = s->string_table_buf->len;
420         shdr = &shdr32;
421     }
422     memcpy(buff, shdr, shdr_size);
423 }
424 
425 static bool prepare_elf_section_hdrs(DumpState *s, Error **errp)
426 {
427     size_t len, sizeof_shdr;
428     void *buff_hdr;
429 
430     /*
431      * Section ordering:
432      * - HDR zero
433      * - Arch section hdrs
434      * - String table hdr
435      */
436     sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
437     len = sizeof_shdr * s->shdr_num;
438     s->elf_section_hdrs = g_malloc0(len);
439     buff_hdr = s->elf_section_hdrs;
440 
441     /*
442      * The first section header is ALWAYS a special initial section
443      * header.
444      *
445      * The header should be 0 with one exception being that if
446      * phdr_num is PN_XNUM then the sh_info field contains the real
447      * number of segment entries.
448      *
449      * As we zero allocate the buffer we will only need to modify
450      * sh_info for the PN_XNUM case.
451      */
452     if (s->phdr_num >= PN_XNUM) {
453         prepare_elf_section_hdr_zero(s);
454     }
455     buff_hdr += sizeof_shdr;
456 
457     /* Add architecture defined section headers */
458     if (s->dump_info.arch_sections_write_hdr_fn
459         && s->shdr_num > 2) {
460         buff_hdr += s->dump_info.arch_sections_write_hdr_fn(s, buff_hdr);
461 
462         if (s->shdr_num >= SHN_LORESERVE) {
463             error_setg_errno(errp, EINVAL,
464                              "dump: too many architecture defined sections");
465             return false;
466         }
467     }
468 
469     /*
470      * String table is the last section since strings are added via
471      * arch_sections_write_hdr().
472      */
473     prepare_elf_section_hdr_string(s, buff_hdr);
474     return true;
475 }
476 
477 static void write_elf_section_headers(DumpState *s, Error **errp)
478 {
479     size_t sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
480     int ret;
481 
482     if (!prepare_elf_section_hdrs(s, errp)) {
483         return;
484     }
485 
486     ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
487     if (ret < 0) {
488         error_setg_errno(errp, -ret, "dump: failed to write section headers");
489     }
490 
491     g_free(s->elf_section_hdrs);
492 }
493 
494 static void write_elf_sections(DumpState *s, Error **errp)
495 {
496     int ret;
497 
498     if (s->elf_section_data_size) {
499         /* Write architecture section data */
500         ret = fd_write_vmcore(s->elf_section_data,
501                               s->elf_section_data_size, s);
502         if (ret < 0) {
503             error_setg_errno(errp, -ret,
504                              "dump: failed to write architecture section data");
505             return;
506         }
507     }
508 
509     /* Write string table */
510     ret = fd_write_vmcore(s->string_table_buf->data,
511                           s->string_table_buf->len, s);
512     if (ret < 0) {
513         error_setg_errno(errp, -ret, "dump: failed to write string table data");
514     }
515 }
516 
517 static void write_data(DumpState *s, void *buf, int length, Error **errp)
518 {
519     int ret;
520 
521     ret = fd_write_vmcore(buf, length, s);
522     if (ret < 0) {
523         error_setg_errno(errp, -ret, "dump: failed to save memory");
524     } else {
525         s->written_size += length;
526     }
527 }
528 
529 /* write the memory to vmcore. 1 page per I/O. */
530 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
531                          int64_t size, Error **errp)
532 {
533     ERRP_GUARD();
534     int64_t i;
535 
536     for (i = 0; i < size / s->dump_info.page_size; i++) {
537         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
538                    s->dump_info.page_size, errp);
539         if (*errp) {
540             return;
541         }
542     }
543 
544     if ((size % s->dump_info.page_size) != 0) {
545         write_data(s, block->host_addr + start + i * s->dump_info.page_size,
546                    size % s->dump_info.page_size, errp);
547         if (*errp) {
548             return;
549         }
550     }
551 }
552 
553 /* get the memory's offset and size in the vmcore */
554 static void get_offset_range(hwaddr phys_addr,
555                              ram_addr_t mapping_length,
556                              DumpState *s,
557                              hwaddr *p_offset,
558                              hwaddr *p_filesz)
559 {
560     GuestPhysBlock *block;
561     hwaddr offset = s->memory_offset;
562     int64_t size_in_block, start;
563 
564     /* When the memory is not stored into vmcore, offset will be -1 */
565     *p_offset = -1;
566     *p_filesz = 0;
567 
568     if (dump_has_filter(s)) {
569         if (phys_addr < s->filter_area_begin ||
570             phys_addr >= s->filter_area_begin + s->filter_area_length) {
571             return;
572         }
573     }
574 
575     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
576         if (dump_has_filter(s)) {
577             if (block->target_start >= s->filter_area_begin + s->filter_area_length ||
578                 block->target_end <= s->filter_area_begin) {
579                 /* This block is out of the range */
580                 continue;
581             }
582 
583             if (s->filter_area_begin <= block->target_start) {
584                 start = block->target_start;
585             } else {
586                 start = s->filter_area_begin;
587             }
588 
589             size_in_block = block->target_end - start;
590             if (s->filter_area_begin + s->filter_area_length < block->target_end) {
591                 size_in_block -= block->target_end - (s->filter_area_begin + s->filter_area_length);
592             }
593         } else {
594             start = block->target_start;
595             size_in_block = block->target_end - block->target_start;
596         }
597 
598         if (phys_addr >= start && phys_addr < start + size_in_block) {
599             *p_offset = phys_addr - start + offset;
600 
601             /* The offset range mapped from the vmcore file must not spill over
602              * the GuestPhysBlock, clamp it. The rest of the mapping will be
603              * zero-filled in memory at load time; see
604              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
605              */
606             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
607                         mapping_length :
608                         size_in_block - (phys_addr - start);
609             return;
610         }
611 
612         offset += size_in_block;
613     }
614 }
615 
616 static void write_elf_phdr_loads(DumpState *s, Error **errp)
617 {
618     ERRP_GUARD();
619     hwaddr offset, filesz;
620     MemoryMapping *memory_mapping;
621     uint32_t phdr_index = 1;
622 
623     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
624         get_offset_range(memory_mapping->phys_addr,
625                          memory_mapping->length,
626                          s, &offset, &filesz);
627         if (dump_is_64bit(s)) {
628             write_elf64_load(s, memory_mapping, phdr_index++, offset,
629                              filesz, errp);
630         } else {
631             write_elf32_load(s, memory_mapping, phdr_index++, offset,
632                              filesz, errp);
633         }
634 
635         if (*errp) {
636             return;
637         }
638 
639         if (phdr_index >= s->phdr_num) {
640             break;
641         }
642     }
643 }
644 
645 static void write_elf_notes(DumpState *s, Error **errp)
646 {
647     if (dump_is_64bit(s)) {
648         write_elf64_notes(fd_write_vmcore, s, errp);
649     } else {
650         write_elf32_notes(fd_write_vmcore, s, errp);
651     }
652 }
653 
654 /* write elf header, PT_NOTE and elf note to vmcore. */
655 static void dump_begin(DumpState *s, Error **errp)
656 {
657     ERRP_GUARD();
658 
659     /*
660      * the vmcore's format is:
661      *   --------------
662      *   |  elf header |
663      *   --------------
664      *   |  sctn_hdr   |
665      *   --------------
666      *   |  PT_NOTE    |
667      *   --------------
668      *   |  PT_LOAD    |
669      *   --------------
670      *   |  ......     |
671      *   --------------
672      *   |  PT_LOAD    |
673      *   --------------
674      *   |  elf note   |
675      *   --------------
676      *   |  memory     |
677      *   --------------
678      *
679      * we only know where the memory is saved after we write elf note into
680      * vmcore.
681      */
682 
683     /* write elf header to vmcore */
684     write_elf_header(s, errp);
685     if (*errp) {
686         return;
687     }
688 
689     /* write section headers to vmcore */
690     write_elf_section_headers(s, errp);
691     if (*errp) {
692         return;
693     }
694 
695     /* write PT_NOTE to vmcore */
696     write_elf_phdr_note(s, errp);
697     if (*errp) {
698         return;
699     }
700 
701     /* write all PT_LOADs to vmcore */
702     write_elf_phdr_loads(s, errp);
703     if (*errp) {
704         return;
705     }
706 
707     /* write notes to vmcore */
708     write_elf_notes(s, errp);
709 }
710 
711 int64_t dump_filtered_memblock_size(GuestPhysBlock *block,
712                                     int64_t filter_area_start,
713                                     int64_t filter_area_length)
714 {
715     int64_t size, left, right;
716 
717     /* No filter, return full size */
718     if (!filter_area_length) {
719         return block->target_end - block->target_start;
720     }
721 
722     /* calculate the overlapped region. */
723     left = MAX(filter_area_start, block->target_start);
724     right = MIN(filter_area_start + filter_area_length, block->target_end);
725     size = right - left;
726     size = size > 0 ? size : 0;
727 
728     return size;
729 }
730 
731 int64_t dump_filtered_memblock_start(GuestPhysBlock *block,
732                                      int64_t filter_area_start,
733                                      int64_t filter_area_length)
734 {
735     if (filter_area_length) {
736         /* return -1 if the block is not within filter area */
737         if (block->target_start >= filter_area_start + filter_area_length ||
738             block->target_end <= filter_area_start) {
739             return -1;
740         }
741 
742         if (filter_area_start > block->target_start) {
743             return filter_area_start - block->target_start;
744         }
745     }
746 
747     return 0;
748 }
749 
750 /* write all memory to vmcore */
751 static void dump_iterate(DumpState *s, Error **errp)
752 {
753     ERRP_GUARD();
754     GuestPhysBlock *block;
755     int64_t memblock_size, memblock_start;
756 
757     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
758         memblock_start = dump_filtered_memblock_start(block, s->filter_area_begin, s->filter_area_length);
759         if (memblock_start == -1) {
760             continue;
761         }
762 
763         memblock_size = dump_filtered_memblock_size(block, s->filter_area_begin, s->filter_area_length);
764 
765         /* Write the memory to file */
766         write_memory(s, block, memblock_start, memblock_size, errp);
767         if (*errp) {
768             return;
769         }
770     }
771 }
772 
773 static void dump_end(DumpState *s, Error **errp)
774 {
775     int rc;
776     ERRP_GUARD();
777 
778     if (s->elf_section_data_size) {
779         s->elf_section_data = g_malloc0(s->elf_section_data_size);
780     }
781 
782     /* Adds the architecture defined section data to s->elf_section_data  */
783     if (s->dump_info.arch_sections_write_fn &&
784         s->elf_section_data_size) {
785         rc = s->dump_info.arch_sections_write_fn(s, s->elf_section_data);
786         if (rc) {
787             error_setg_errno(errp, rc,
788                              "dump: failed to get arch section data");
789             g_free(s->elf_section_data);
790             return;
791         }
792     }
793 
794     /* write sections to vmcore */
795     write_elf_sections(s, errp);
796 }
797 
798 static void create_vmcore(DumpState *s, Error **errp)
799 {
800     ERRP_GUARD();
801 
802     dump_begin(s, errp);
803     if (*errp) {
804         return;
805     }
806 
807     /* Iterate over memory and dump it to file */
808     dump_iterate(s, errp);
809     if (*errp) {
810         return;
811     }
812 
813     /* Write the section data */
814     dump_end(s, errp);
815 }
816 
817 static int write_start_flat_header(int fd)
818 {
819     MakedumpfileHeader *mh;
820     int ret = 0;
821 
822     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
823     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
824 
825     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
826            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
827 
828     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
829     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
830 
831     size_t written_size;
832     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
833     if (written_size != MAX_SIZE_MDF_HEADER) {
834         ret = -1;
835     }
836 
837     g_free(mh);
838     return ret;
839 }
840 
841 static int write_end_flat_header(int fd)
842 {
843     MakedumpfileDataHeader mdh;
844 
845     mdh.offset = END_FLAG_FLAT_HEADER;
846     mdh.buf_size = END_FLAG_FLAT_HEADER;
847 
848     size_t written_size;
849     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
850     if (written_size != sizeof(mdh)) {
851         return -1;
852     }
853 
854     return 0;
855 }
856 
857 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
858 {
859     size_t written_size;
860     MakedumpfileDataHeader mdh;
861 
862     mdh.offset = cpu_to_be64(offset);
863     mdh.buf_size = cpu_to_be64(size);
864 
865     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
866     if (written_size != sizeof(mdh)) {
867         return -1;
868     }
869 
870     written_size = qemu_write_full(fd, buf, size);
871     if (written_size != size) {
872         return -1;
873     }
874 
875     return 0;
876 }
877 
878 static int buf_write_note(const void *buf, size_t size, void *opaque)
879 {
880     DumpState *s = opaque;
881 
882     /* note_buf is not enough */
883     if (s->note_buf_offset + size > s->note_size) {
884         return -1;
885     }
886 
887     memcpy(s->note_buf + s->note_buf_offset, buf, size);
888 
889     s->note_buf_offset += size;
890 
891     return 0;
892 }
893 
894 /*
895  * This function retrieves various sizes from an elf header.
896  *
897  * @note has to be a valid ELF note. The return sizes are unmodified
898  * (not padded or rounded up to be multiple of 4).
899  */
900 static void get_note_sizes(DumpState *s, const void *note,
901                            uint64_t *note_head_size,
902                            uint64_t *name_size,
903                            uint64_t *desc_size)
904 {
905     uint64_t note_head_sz;
906     uint64_t name_sz;
907     uint64_t desc_sz;
908 
909     if (dump_is_64bit(s)) {
910         const Elf64_Nhdr *hdr = note;
911         note_head_sz = sizeof(Elf64_Nhdr);
912         name_sz = tswap64(hdr->n_namesz);
913         desc_sz = tswap64(hdr->n_descsz);
914     } else {
915         const Elf32_Nhdr *hdr = note;
916         note_head_sz = sizeof(Elf32_Nhdr);
917         name_sz = tswap32(hdr->n_namesz);
918         desc_sz = tswap32(hdr->n_descsz);
919     }
920 
921     if (note_head_size) {
922         *note_head_size = note_head_sz;
923     }
924     if (name_size) {
925         *name_size = name_sz;
926     }
927     if (desc_size) {
928         *desc_size = desc_sz;
929     }
930 }
931 
932 static bool note_name_equal(DumpState *s,
933                             const uint8_t *note, const char *name)
934 {
935     int len = strlen(name) + 1;
936     uint64_t head_size, name_size;
937 
938     get_note_sizes(s, note, &head_size, &name_size, NULL);
939     head_size = ROUND_UP(head_size, 4);
940 
941     return name_size == len && memcmp(note + head_size, name, len) == 0;
942 }
943 
944 /* write common header, sub header and elf note to vmcore */
945 static void create_header32(DumpState *s, Error **errp)
946 {
947     ERRP_GUARD();
948     DiskDumpHeader32 *dh = NULL;
949     KdumpSubHeader32 *kh = NULL;
950     size_t size;
951     uint32_t block_size;
952     uint32_t sub_hdr_size;
953     uint32_t bitmap_blocks;
954     uint32_t status = 0;
955     uint64_t offset_note;
956 
957     /* write common header, the version of kdump-compressed format is 6th */
958     size = sizeof(DiskDumpHeader32);
959     dh = g_malloc0(size);
960 
961     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
962     dh->header_version = cpu_to_dump32(s, 6);
963     block_size = s->dump_info.page_size;
964     dh->block_size = cpu_to_dump32(s, block_size);
965     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
966     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
967     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
968     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
969     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
970     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
971     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
972     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
973     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
974 
975     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
976         status |= DUMP_DH_COMPRESSED_ZLIB;
977     }
978 #ifdef CONFIG_LZO
979     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
980         status |= DUMP_DH_COMPRESSED_LZO;
981     }
982 #endif
983 #ifdef CONFIG_SNAPPY
984     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
985         status |= DUMP_DH_COMPRESSED_SNAPPY;
986     }
987 #endif
988     dh->status = cpu_to_dump32(s, status);
989 
990     if (write_buffer(s->fd, 0, dh, size) < 0) {
991         error_setg(errp, "dump: failed to write disk dump header");
992         goto out;
993     }
994 
995     /* write sub header */
996     size = sizeof(KdumpSubHeader32);
997     kh = g_malloc0(size);
998 
999     /* 64bit max_mapnr_64 */
1000     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1001     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
1002     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1003 
1004     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1005     if (s->guest_note &&
1006         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1007         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1008 
1009         get_note_sizes(s, s->guest_note,
1010                        &hsize, &name_size, &size_vmcoreinfo_desc);
1011         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1012             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1013         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1014         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
1015     }
1016 
1017     kh->offset_note = cpu_to_dump64(s, offset_note);
1018     kh->note_size = cpu_to_dump32(s, s->note_size);
1019 
1020     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
1021                      block_size, kh, size) < 0) {
1022         error_setg(errp, "dump: failed to write kdump sub header");
1023         goto out;
1024     }
1025 
1026     /* write note */
1027     s->note_buf = g_malloc0(s->note_size);
1028     s->note_buf_offset = 0;
1029 
1030     /* use s->note_buf to store notes temporarily */
1031     write_elf32_notes(buf_write_note, s, errp);
1032     if (*errp) {
1033         goto out;
1034     }
1035     if (write_buffer(s->fd, offset_note, s->note_buf,
1036                      s->note_size) < 0) {
1037         error_setg(errp, "dump: failed to write notes");
1038         goto out;
1039     }
1040 
1041     /* get offset of dump_bitmap */
1042     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1043                              block_size;
1044 
1045     /* get offset of page */
1046     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1047                      block_size;
1048 
1049 out:
1050     g_free(dh);
1051     g_free(kh);
1052     g_free(s->note_buf);
1053 }
1054 
1055 /* write common header, sub header and elf note to vmcore */
1056 static void create_header64(DumpState *s, Error **errp)
1057 {
1058     ERRP_GUARD();
1059     DiskDumpHeader64 *dh = NULL;
1060     KdumpSubHeader64 *kh = NULL;
1061     size_t size;
1062     uint32_t block_size;
1063     uint32_t sub_hdr_size;
1064     uint32_t bitmap_blocks;
1065     uint32_t status = 0;
1066     uint64_t offset_note;
1067 
1068     /* write common header, the version of kdump-compressed format is 6th */
1069     size = sizeof(DiskDumpHeader64);
1070     dh = g_malloc0(size);
1071 
1072     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
1073     dh->header_version = cpu_to_dump32(s, 6);
1074     block_size = s->dump_info.page_size;
1075     dh->block_size = cpu_to_dump32(s, block_size);
1076     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
1077     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
1078     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
1079     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
1080     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
1081     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
1082     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
1083     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
1084     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
1085 
1086     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
1087         status |= DUMP_DH_COMPRESSED_ZLIB;
1088     }
1089 #ifdef CONFIG_LZO
1090     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
1091         status |= DUMP_DH_COMPRESSED_LZO;
1092     }
1093 #endif
1094 #ifdef CONFIG_SNAPPY
1095     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1096         status |= DUMP_DH_COMPRESSED_SNAPPY;
1097     }
1098 #endif
1099     dh->status = cpu_to_dump32(s, status);
1100 
1101     if (write_buffer(s->fd, 0, dh, size) < 0) {
1102         error_setg(errp, "dump: failed to write disk dump header");
1103         goto out;
1104     }
1105 
1106     /* write sub header */
1107     size = sizeof(KdumpSubHeader64);
1108     kh = g_malloc0(size);
1109 
1110     /* 64bit max_mapnr_64 */
1111     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1112     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
1113     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1114 
1115     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1116     if (s->guest_note &&
1117         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1118         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1119 
1120         get_note_sizes(s, s->guest_note,
1121                        &hsize, &name_size, &size_vmcoreinfo_desc);
1122         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1123             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1124         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1125         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
1126     }
1127 
1128     kh->offset_note = cpu_to_dump64(s, offset_note);
1129     kh->note_size = cpu_to_dump64(s, s->note_size);
1130 
1131     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
1132                      block_size, kh, size) < 0) {
1133         error_setg(errp, "dump: failed to write kdump sub header");
1134         goto out;
1135     }
1136 
1137     /* write note */
1138     s->note_buf = g_malloc0(s->note_size);
1139     s->note_buf_offset = 0;
1140 
1141     /* use s->note_buf to store notes temporarily */
1142     write_elf64_notes(buf_write_note, s, errp);
1143     if (*errp) {
1144         goto out;
1145     }
1146 
1147     if (write_buffer(s->fd, offset_note, s->note_buf,
1148                      s->note_size) < 0) {
1149         error_setg(errp, "dump: failed to write notes");
1150         goto out;
1151     }
1152 
1153     /* get offset of dump_bitmap */
1154     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1155                              block_size;
1156 
1157     /* get offset of page */
1158     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1159                      block_size;
1160 
1161 out:
1162     g_free(dh);
1163     g_free(kh);
1164     g_free(s->note_buf);
1165 }
1166 
1167 static void write_dump_header(DumpState *s, Error **errp)
1168 {
1169     if (dump_is_64bit(s)) {
1170         create_header64(s, errp);
1171     } else {
1172         create_header32(s, errp);
1173     }
1174 }
1175 
1176 static size_t dump_bitmap_get_bufsize(DumpState *s)
1177 {
1178     return s->dump_info.page_size;
1179 }
1180 
1181 /*
1182  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1183  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1184  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1185  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1186  * vmcore, ie. synchronizing un-sync bit into vmcore.
1187  */
1188 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1189                            uint8_t *buf, DumpState *s)
1190 {
1191     off_t old_offset, new_offset;
1192     off_t offset_bitmap1, offset_bitmap2;
1193     uint32_t byte, bit;
1194     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1195     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1196 
1197     /* should not set the previous place */
1198     assert(last_pfn <= pfn);
1199 
1200     /*
1201      * if the bit needed to be set is not cached in buf, flush the data in buf
1202      * to vmcore firstly.
1203      * making new_offset be bigger than old_offset can also sync remained data
1204      * into vmcore.
1205      */
1206     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1207     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1208 
1209     while (old_offset < new_offset) {
1210         /* calculate the offset and write dump_bitmap */
1211         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1212         if (write_buffer(s->fd, offset_bitmap1, buf,
1213                          bitmap_bufsize) < 0) {
1214             return -1;
1215         }
1216 
1217         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1218         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1219                          old_offset;
1220         if (write_buffer(s->fd, offset_bitmap2, buf,
1221                          bitmap_bufsize) < 0) {
1222             return -1;
1223         }
1224 
1225         memset(buf, 0, bitmap_bufsize);
1226         old_offset += bitmap_bufsize;
1227     }
1228 
1229     /* get the exact place of the bit in the buf, and set it */
1230     byte = (pfn % bits_per_buf) / CHAR_BIT;
1231     bit = (pfn % bits_per_buf) % CHAR_BIT;
1232     if (value) {
1233         buf[byte] |= 1u << bit;
1234     } else {
1235         buf[byte] &= ~(1u << bit);
1236     }
1237 
1238     return 0;
1239 }
1240 
1241 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1242 {
1243     int target_page_shift = ctz32(s->dump_info.page_size);
1244 
1245     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1246 }
1247 
1248 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1249 {
1250     int target_page_shift = ctz32(s->dump_info.page_size);
1251 
1252     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1253 }
1254 
1255 /*
1256  * Return the page frame number and the page content in *bufptr. bufptr can be
1257  * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated
1258  * memory. This is not necessarily the memory returned.
1259  */
1260 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1261                           uint8_t **bufptr, DumpState *s)
1262 {
1263     GuestPhysBlock *block = *blockptr;
1264     uint32_t page_size = s->dump_info.page_size;
1265     uint8_t *buf = NULL, *hbuf;
1266     hwaddr addr;
1267 
1268     /* block == NULL means the start of the iteration */
1269     if (!block) {
1270         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1271         *blockptr = block;
1272         addr = block->target_start;
1273         *pfnptr = dump_paddr_to_pfn(s, addr);
1274     } else {
1275         *pfnptr += 1;
1276         addr = dump_pfn_to_paddr(s, *pfnptr);
1277     }
1278     assert(block != NULL);
1279 
1280     while (1) {
1281         if (addr >= block->target_start && addr < block->target_end) {
1282             size_t n = MIN(block->target_end - addr, page_size - addr % page_size);
1283             hbuf = block->host_addr + (addr - block->target_start);
1284             if (!buf) {
1285                 if (n == page_size) {
1286                     /* this is a whole target page, go for it */
1287                     assert(addr % page_size == 0);
1288                     buf = hbuf;
1289                     break;
1290                 } else if (bufptr) {
1291                     assert(*bufptr);
1292                     buf = *bufptr;
1293                     memset(buf, 0, page_size);
1294                 } else {
1295                     return true;
1296                 }
1297             }
1298 
1299             memcpy(buf + addr % page_size, hbuf, n);
1300             addr += n;
1301             if (addr % page_size == 0) {
1302                 /* we filled up the page */
1303                 break;
1304             }
1305         } else {
1306             /* the next page is in the next block */
1307             *blockptr = block = QTAILQ_NEXT(block, next);
1308             if (!block) {
1309                 break;
1310             }
1311 
1312             addr = block->target_start;
1313             /* are we still in the same page? */
1314             if (dump_paddr_to_pfn(s, addr) != *pfnptr) {
1315                 if (buf) {
1316                     /* no, but we already filled something earlier, return it */
1317                     break;
1318                 } else {
1319                     /* else continue from there */
1320                     *pfnptr = dump_paddr_to_pfn(s, addr);
1321                 }
1322             }
1323         }
1324     }
1325 
1326     if (bufptr) {
1327         *bufptr = buf;
1328     }
1329 
1330     return buf != NULL;
1331 }
1332 
1333 static void write_dump_bitmap(DumpState *s, Error **errp)
1334 {
1335     int ret = 0;
1336     uint64_t last_pfn, pfn;
1337     void *dump_bitmap_buf;
1338     size_t num_dumpable;
1339     GuestPhysBlock *block_iter = NULL;
1340     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1341     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1342 
1343     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1344     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1345 
1346     num_dumpable = 0;
1347     last_pfn = 0;
1348 
1349     /*
1350      * exam memory page by page, and set the bit in dump_bitmap corresponded
1351      * to the existing page.
1352      */
1353     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1354         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1355         if (ret < 0) {
1356             error_setg(errp, "dump: failed to set dump_bitmap");
1357             goto out;
1358         }
1359 
1360         last_pfn = pfn;
1361         num_dumpable++;
1362     }
1363 
1364     /*
1365      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1366      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1367      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1368      */
1369     if (num_dumpable > 0) {
1370         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1371                               dump_bitmap_buf, s);
1372         if (ret < 0) {
1373             error_setg(errp, "dump: failed to sync dump_bitmap");
1374             goto out;
1375         }
1376     }
1377 
1378     /* number of dumpable pages that will be dumped later */
1379     s->num_dumpable = num_dumpable;
1380 
1381 out:
1382     g_free(dump_bitmap_buf);
1383 }
1384 
1385 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1386                                off_t offset)
1387 {
1388     data_cache->fd = s->fd;
1389     data_cache->data_size = 0;
1390     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1391     data_cache->buf = g_malloc0(data_cache->buf_size);
1392     data_cache->offset = offset;
1393 }
1394 
1395 static int write_cache(DataCache *dc, const void *buf, size_t size,
1396                        bool flag_sync)
1397 {
1398     /*
1399      * dc->buf_size should not be less than size, otherwise dc will never be
1400      * enough
1401      */
1402     assert(size <= dc->buf_size);
1403 
1404     /*
1405      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1406      * otherwise check if the space is enough for caching data in buf, if not,
1407      * write the data in dc->buf to dc->fd and reset dc->buf
1408      */
1409     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1410         (flag_sync && dc->data_size > 0)) {
1411         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1412             return -1;
1413         }
1414 
1415         dc->offset += dc->data_size;
1416         dc->data_size = 0;
1417     }
1418 
1419     if (!flag_sync) {
1420         memcpy(dc->buf + dc->data_size, buf, size);
1421         dc->data_size += size;
1422     }
1423 
1424     return 0;
1425 }
1426 
1427 static void free_data_cache(DataCache *data_cache)
1428 {
1429     g_free(data_cache->buf);
1430 }
1431 
1432 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1433 {
1434     switch (flag_compress) {
1435     case DUMP_DH_COMPRESSED_ZLIB:
1436         return compressBound(page_size);
1437 
1438     case DUMP_DH_COMPRESSED_LZO:
1439         /*
1440          * LZO will expand incompressible data by a little amount. Please check
1441          * the following URL to see the expansion calculation:
1442          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1443          */
1444         return page_size + page_size / 16 + 64 + 3;
1445 
1446 #ifdef CONFIG_SNAPPY
1447     case DUMP_DH_COMPRESSED_SNAPPY:
1448         return snappy_max_compressed_length(page_size);
1449 #endif
1450     }
1451     return 0;
1452 }
1453 
1454 static void write_dump_pages(DumpState *s, Error **errp)
1455 {
1456     int ret = 0;
1457     DataCache page_desc, page_data;
1458     size_t len_buf_out, size_out;
1459 #ifdef CONFIG_LZO
1460     lzo_bytep wrkmem = NULL;
1461 #endif
1462     uint8_t *buf_out = NULL;
1463     off_t offset_desc, offset_data;
1464     PageDescriptor pd, pd_zero;
1465     uint8_t *buf;
1466     GuestPhysBlock *block_iter = NULL;
1467     uint64_t pfn_iter;
1468     g_autofree uint8_t *page = NULL;
1469 
1470     /* get offset of page_desc and page_data in dump file */
1471     offset_desc = s->offset_page;
1472     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1473 
1474     prepare_data_cache(&page_desc, s, offset_desc);
1475     prepare_data_cache(&page_data, s, offset_data);
1476 
1477     /* prepare buffer to store compressed data */
1478     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1479     assert(len_buf_out != 0);
1480 
1481 #ifdef CONFIG_LZO
1482     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1483 #endif
1484 
1485     buf_out = g_malloc(len_buf_out);
1486 
1487     /*
1488      * init zero page's page_desc and page_data, because every zero page
1489      * uses the same page_data
1490      */
1491     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1492     pd_zero.flags = cpu_to_dump32(s, 0);
1493     pd_zero.offset = cpu_to_dump64(s, offset_data);
1494     pd_zero.page_flags = cpu_to_dump64(s, 0);
1495     buf = g_malloc0(s->dump_info.page_size);
1496     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1497     g_free(buf);
1498     if (ret < 0) {
1499         error_setg(errp, "dump: failed to write page data (zero page)");
1500         goto out;
1501     }
1502 
1503     offset_data += s->dump_info.page_size;
1504     page = g_malloc(s->dump_info.page_size);
1505 
1506     /*
1507      * dump memory to vmcore page by page. zero page will all be resided in the
1508      * first page of page section
1509      */
1510     for (buf = page; get_next_page(&block_iter, &pfn_iter, &buf, s); buf = page) {
1511         /* check zero page */
1512         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1513             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1514                               false);
1515             if (ret < 0) {
1516                 error_setg(errp, "dump: failed to write page desc");
1517                 goto out;
1518             }
1519         } else {
1520             /*
1521              * not zero page, then:
1522              * 1. compress the page
1523              * 2. write the compressed page into the cache of page_data
1524              * 3. get page desc of the compressed page and write it into the
1525              *    cache of page_desc
1526              *
1527              * only one compression format will be used here, for
1528              * s->flag_compress is set. But when compression fails to work,
1529              * we fall back to save in plaintext.
1530              */
1531              size_out = len_buf_out;
1532              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1533                     (compress2(buf_out, (uLongf *)&size_out, buf,
1534                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1535                     (size_out < s->dump_info.page_size)) {
1536                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1537                 pd.size  = cpu_to_dump32(s, size_out);
1538 
1539                 ret = write_cache(&page_data, buf_out, size_out, false);
1540                 if (ret < 0) {
1541                     error_setg(errp, "dump: failed to write page data");
1542                     goto out;
1543                 }
1544 #ifdef CONFIG_LZO
1545             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1546                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1547                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1548                     (size_out < s->dump_info.page_size)) {
1549                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1550                 pd.size  = cpu_to_dump32(s, size_out);
1551 
1552                 ret = write_cache(&page_data, buf_out, size_out, false);
1553                 if (ret < 0) {
1554                     error_setg(errp, "dump: failed to write page data");
1555                     goto out;
1556                 }
1557 #endif
1558 #ifdef CONFIG_SNAPPY
1559             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1560                     (snappy_compress((char *)buf, s->dump_info.page_size,
1561                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1562                     (size_out < s->dump_info.page_size)) {
1563                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1564                 pd.size  = cpu_to_dump32(s, size_out);
1565 
1566                 ret = write_cache(&page_data, buf_out, size_out, false);
1567                 if (ret < 0) {
1568                     error_setg(errp, "dump: failed to write page data");
1569                     goto out;
1570                 }
1571 #endif
1572             } else {
1573                 /*
1574                  * fall back to save in plaintext, size_out should be
1575                  * assigned the target's page size
1576                  */
1577                 pd.flags = cpu_to_dump32(s, 0);
1578                 size_out = s->dump_info.page_size;
1579                 pd.size = cpu_to_dump32(s, size_out);
1580 
1581                 ret = write_cache(&page_data, buf,
1582                                   s->dump_info.page_size, false);
1583                 if (ret < 0) {
1584                     error_setg(errp, "dump: failed to write page data");
1585                     goto out;
1586                 }
1587             }
1588 
1589             /* get and write page desc here */
1590             pd.page_flags = cpu_to_dump64(s, 0);
1591             pd.offset = cpu_to_dump64(s, offset_data);
1592             offset_data += size_out;
1593 
1594             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1595             if (ret < 0) {
1596                 error_setg(errp, "dump: failed to write page desc");
1597                 goto out;
1598             }
1599         }
1600         s->written_size += s->dump_info.page_size;
1601     }
1602 
1603     ret = write_cache(&page_desc, NULL, 0, true);
1604     if (ret < 0) {
1605         error_setg(errp, "dump: failed to sync cache for page_desc");
1606         goto out;
1607     }
1608     ret = write_cache(&page_data, NULL, 0, true);
1609     if (ret < 0) {
1610         error_setg(errp, "dump: failed to sync cache for page_data");
1611         goto out;
1612     }
1613 
1614 out:
1615     free_data_cache(&page_desc);
1616     free_data_cache(&page_data);
1617 
1618 #ifdef CONFIG_LZO
1619     g_free(wrkmem);
1620 #endif
1621 
1622     g_free(buf_out);
1623 }
1624 
1625 static void create_kdump_vmcore(DumpState *s, Error **errp)
1626 {
1627     ERRP_GUARD();
1628     int ret;
1629 
1630     /*
1631      * the kdump-compressed format is:
1632      *                                               File offset
1633      *  +------------------------------------------+ 0x0
1634      *  |    main header (struct disk_dump_header) |
1635      *  |------------------------------------------+ block 1
1636      *  |    sub header (struct kdump_sub_header)  |
1637      *  |------------------------------------------+ block 2
1638      *  |            1st-dump_bitmap               |
1639      *  |------------------------------------------+ block 2 + X blocks
1640      *  |            2nd-dump_bitmap               | (aligned by block)
1641      *  |------------------------------------------+ block 2 + 2 * X blocks
1642      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1643      *  |  page desc for pfn 1 (struct page_desc)  |
1644      *  |                    :                     |
1645      *  |------------------------------------------| (not aligned by block)
1646      *  |         page data (pfn 0)                |
1647      *  |         page data (pfn 1)                |
1648      *  |                    :                     |
1649      *  +------------------------------------------+
1650      */
1651 
1652     ret = write_start_flat_header(s->fd);
1653     if (ret < 0) {
1654         error_setg(errp, "dump: failed to write start flat header");
1655         return;
1656     }
1657 
1658     write_dump_header(s, errp);
1659     if (*errp) {
1660         return;
1661     }
1662 
1663     write_dump_bitmap(s, errp);
1664     if (*errp) {
1665         return;
1666     }
1667 
1668     write_dump_pages(s, errp);
1669     if (*errp) {
1670         return;
1671     }
1672 
1673     ret = write_end_flat_header(s->fd);
1674     if (ret < 0) {
1675         error_setg(errp, "dump: failed to write end flat header");
1676         return;
1677     }
1678 }
1679 
1680 static int validate_start_block(DumpState *s)
1681 {
1682     GuestPhysBlock *block;
1683 
1684     if (!dump_has_filter(s)) {
1685         return 0;
1686     }
1687 
1688     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1689         /* This block is out of the range */
1690         if (block->target_start >= s->filter_area_begin + s->filter_area_length ||
1691             block->target_end <= s->filter_area_begin) {
1692             continue;
1693         }
1694         return 0;
1695    }
1696 
1697     return -1;
1698 }
1699 
1700 static void get_max_mapnr(DumpState *s)
1701 {
1702     GuestPhysBlock *last_block;
1703 
1704     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1705     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1706 }
1707 
1708 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1709 
1710 static void dump_state_prepare(DumpState *s)
1711 {
1712     /* zero the struct, setting status to active */
1713     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1714 }
1715 
1716 bool qemu_system_dump_in_progress(void)
1717 {
1718     DumpState *state = &dump_state_global;
1719     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1720 }
1721 
1722 /*
1723  * calculate total size of memory to be dumped (taking filter into
1724  * account.)
1725  */
1726 static int64_t dump_calculate_size(DumpState *s)
1727 {
1728     GuestPhysBlock *block;
1729     int64_t total = 0;
1730 
1731     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1732         total += dump_filtered_memblock_size(block,
1733                                              s->filter_area_begin,
1734                                              s->filter_area_length);
1735     }
1736 
1737     return total;
1738 }
1739 
1740 static void vmcoreinfo_update_phys_base(DumpState *s)
1741 {
1742     uint64_t size, note_head_size, name_size, phys_base;
1743     char **lines;
1744     uint8_t *vmci;
1745     size_t i;
1746 
1747     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1748         return;
1749     }
1750 
1751     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1752     note_head_size = ROUND_UP(note_head_size, 4);
1753 
1754     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1755     *(vmci + size) = '\0';
1756 
1757     lines = g_strsplit((char *)vmci, "\n", -1);
1758     for (i = 0; lines[i]; i++) {
1759         const char *prefix = NULL;
1760 
1761         if (s->dump_info.d_machine == EM_X86_64) {
1762             prefix = "NUMBER(phys_base)=";
1763         } else if (s->dump_info.d_machine == EM_AARCH64) {
1764             prefix = "NUMBER(PHYS_OFFSET)=";
1765         }
1766 
1767         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1768             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1769                               &phys_base) < 0) {
1770                 warn_report("Failed to read %s", prefix);
1771             } else {
1772                 s->dump_info.phys_base = phys_base;
1773             }
1774             break;
1775         }
1776     }
1777 
1778     g_strfreev(lines);
1779 }
1780 
1781 static void dump_init(DumpState *s, int fd, bool has_format,
1782                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1783                       int64_t begin, int64_t length, Error **errp)
1784 {
1785     ERRP_GUARD();
1786     VMCoreInfoState *vmci = vmcoreinfo_find();
1787     CPUState *cpu;
1788     int nr_cpus;
1789     int ret;
1790 
1791     s->has_format = has_format;
1792     s->format = format;
1793     s->written_size = 0;
1794 
1795     /* kdump-compressed is conflict with paging and filter */
1796     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1797         assert(!paging && !has_filter);
1798     }
1799 
1800     if (runstate_is_running()) {
1801         vm_stop(RUN_STATE_SAVE_VM);
1802         s->resume = true;
1803     } else {
1804         s->resume = false;
1805     }
1806 
1807     /* If we use KVM, we should synchronize the registers before we get dump
1808      * info or physmap info.
1809      */
1810     cpu_synchronize_all_states();
1811     nr_cpus = 0;
1812     CPU_FOREACH(cpu) {
1813         nr_cpus++;
1814     }
1815 
1816     s->fd = fd;
1817     if (has_filter && !length) {
1818         error_setg(errp, QERR_INVALID_PARAMETER, "length");
1819         goto cleanup;
1820     }
1821     s->filter_area_begin = begin;
1822     s->filter_area_length = length;
1823 
1824     /* First index is 0, it's the special null name */
1825     s->string_table_buf = g_array_new(FALSE, TRUE, 1);
1826     /*
1827      * Allocate the null name, due to the clearing option set to true
1828      * it will be 0.
1829      */
1830     g_array_set_size(s->string_table_buf, 1);
1831 
1832     memory_mapping_list_init(&s->list);
1833 
1834     guest_phys_blocks_init(&s->guest_phys_blocks);
1835     guest_phys_blocks_append(&s->guest_phys_blocks);
1836     s->total_size = dump_calculate_size(s);
1837 #ifdef DEBUG_DUMP_GUEST_MEMORY
1838     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1839 #endif
1840 
1841     /* it does not make sense to dump non-existent memory */
1842     if (!s->total_size) {
1843         error_setg(errp, "dump: no guest memory to dump");
1844         goto cleanup;
1845     }
1846 
1847     /* Is the filter filtering everything? */
1848     if (validate_start_block(s) == -1) {
1849         error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1850         goto cleanup;
1851     }
1852 
1853     /* get dump info: endian, class and architecture.
1854      * If the target architecture is not supported, cpu_get_dump_info() will
1855      * return -1.
1856      */
1857     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1858     if (ret < 0) {
1859         error_setg(errp, QERR_UNSUPPORTED);
1860         goto cleanup;
1861     }
1862 
1863     if (!s->dump_info.page_size) {
1864         s->dump_info.page_size = TARGET_PAGE_SIZE;
1865     }
1866 
1867     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1868                                      s->dump_info.d_machine, nr_cpus);
1869     if (s->note_size < 0) {
1870         error_setg(errp, QERR_UNSUPPORTED);
1871         goto cleanup;
1872     }
1873 
1874     /*
1875      * The goal of this block is to (a) update the previously guessed
1876      * phys_base, (b) copy the guest note out of the guest.
1877      * Failure to do so is not fatal for dumping.
1878      */
1879     if (vmci) {
1880         uint64_t addr, note_head_size, name_size, desc_size;
1881         uint32_t size;
1882         uint16_t format;
1883 
1884         note_head_size = dump_is_64bit(s) ?
1885             sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr);
1886 
1887         format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1888         size = le32_to_cpu(vmci->vmcoreinfo.size);
1889         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1890         if (!vmci->has_vmcoreinfo) {
1891             warn_report("guest note is not present");
1892         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1893             warn_report("guest note size is invalid: %" PRIu32, size);
1894         } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1895             warn_report("guest note format is unsupported: %" PRIu16, format);
1896         } else {
1897             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1898             cpu_physical_memory_read(addr, s->guest_note, size);
1899 
1900             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1901             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1902                                                desc_size);
1903             if (name_size > MAX_GUEST_NOTE_SIZE ||
1904                 desc_size > MAX_GUEST_NOTE_SIZE ||
1905                 s->guest_note_size > size) {
1906                 warn_report("Invalid guest note header");
1907                 g_free(s->guest_note);
1908                 s->guest_note = NULL;
1909             } else {
1910                 vmcoreinfo_update_phys_base(s);
1911                 s->note_size += s->guest_note_size;
1912             }
1913         }
1914     }
1915 
1916     /* get memory mapping */
1917     if (paging) {
1918         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1919         if (*errp) {
1920             goto cleanup;
1921         }
1922     } else {
1923         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1924     }
1925 
1926     s->nr_cpus = nr_cpus;
1927 
1928     get_max_mapnr(s);
1929 
1930     uint64_t tmp;
1931     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1932                        s->dump_info.page_size);
1933     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1934 
1935     /* init for kdump-compressed format */
1936     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1937         switch (format) {
1938         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1939             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1940             break;
1941 
1942         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1943 #ifdef CONFIG_LZO
1944             if (lzo_init() != LZO_E_OK) {
1945                 error_setg(errp, "failed to initialize the LZO library");
1946                 goto cleanup;
1947             }
1948 #endif
1949             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1950             break;
1951 
1952         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1953             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1954             break;
1955 
1956         default:
1957             s->flag_compress = 0;
1958         }
1959 
1960         return;
1961     }
1962 
1963     if (dump_has_filter(s)) {
1964         memory_mapping_filter(&s->list, s->filter_area_begin, s->filter_area_length);
1965     }
1966 
1967     /*
1968      * The first section header is always a special one in which most
1969      * fields are 0. The section header string table is also always
1970      * set.
1971      */
1972     s->shdr_num = 2;
1973 
1974     /*
1975      * Adds the number of architecture sections to shdr_num and sets
1976      * elf_section_data_size so we know the offsets and sizes of all
1977      * parts.
1978      */
1979     if (s->dump_info.arch_sections_add_fn) {
1980         s->dump_info.arch_sections_add_fn(s);
1981     }
1982 
1983     /*
1984      * calculate shdr_num so we know the offsets and sizes of all
1985      * parts.
1986      * Calculate phdr_num
1987      *
1988      * The absolute maximum amount of phdrs is UINT32_MAX - 1 as
1989      * sh_info is 32 bit. There's special handling once we go over
1990      * UINT16_MAX - 1 but that is handled in the ehdr and section
1991      * code.
1992      */
1993     s->phdr_num = 1; /* Reserve PT_NOTE */
1994     if (s->list.num <= UINT32_MAX - 1) {
1995         s->phdr_num += s->list.num;
1996     } else {
1997         s->phdr_num = UINT32_MAX;
1998     }
1999 
2000     /*
2001      * Now that the number of section and program headers is known we
2002      * can calculate the offsets of the headers and data.
2003      */
2004     if (dump_is_64bit(s)) {
2005         s->shdr_offset = sizeof(Elf64_Ehdr);
2006         s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
2007         s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
2008     } else {
2009         s->shdr_offset = sizeof(Elf32_Ehdr);
2010         s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
2011         s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
2012     }
2013     s->memory_offset = s->note_offset + s->note_size;
2014     s->section_offset = s->memory_offset + s->total_size;
2015 
2016     return;
2017 
2018 cleanup:
2019     dump_cleanup(s);
2020 }
2021 
2022 /* this operation might be time consuming. */
2023 static void dump_process(DumpState *s, Error **errp)
2024 {
2025     ERRP_GUARD();
2026     DumpQueryResult *result = NULL;
2027 
2028     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2029 #ifdef TARGET_X86_64
2030         create_win_dump(s, errp);
2031 #endif
2032     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
2033         create_kdump_vmcore(s, errp);
2034     } else {
2035         create_vmcore(s, errp);
2036     }
2037 
2038     /* make sure status is written after written_size updates */
2039     smp_wmb();
2040     qatomic_set(&s->status,
2041                (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
2042 
2043     /* send DUMP_COMPLETED message (unconditionally) */
2044     result = qmp_query_dump(NULL);
2045     /* should never fail */
2046     assert(result);
2047     qapi_event_send_dump_completed(result, !!*errp, (*errp ?
2048                                                      error_get_pretty(*errp) : NULL));
2049     qapi_free_DumpQueryResult(result);
2050 
2051     dump_cleanup(s);
2052 }
2053 
2054 static void *dump_thread(void *data)
2055 {
2056     DumpState *s = (DumpState *)data;
2057     dump_process(s, NULL);
2058     return NULL;
2059 }
2060 
2061 DumpQueryResult *qmp_query_dump(Error **errp)
2062 {
2063     DumpQueryResult *result = g_new(DumpQueryResult, 1);
2064     DumpState *state = &dump_state_global;
2065     result->status = qatomic_read(&state->status);
2066     /* make sure we are reading status and written_size in order */
2067     smp_rmb();
2068     result->completed = state->written_size;
2069     result->total = state->total_size;
2070     return result;
2071 }
2072 
2073 void qmp_dump_guest_memory(bool paging, const char *file,
2074                            bool has_detach, bool detach,
2075                            bool has_begin, int64_t begin, bool has_length,
2076                            int64_t length, bool has_format,
2077                            DumpGuestMemoryFormat format, Error **errp)
2078 {
2079     ERRP_GUARD();
2080     const char *p;
2081     int fd = -1;
2082     DumpState *s;
2083     bool detach_p = false;
2084 
2085     if (runstate_check(RUN_STATE_INMIGRATE)) {
2086         error_setg(errp, "Dump not allowed during incoming migration.");
2087         return;
2088     }
2089 
2090     /* if there is a dump in background, we should wait until the dump
2091      * finished */
2092     if (qemu_system_dump_in_progress()) {
2093         error_setg(errp, "There is a dump in process, please wait.");
2094         return;
2095     }
2096 
2097     /*
2098      * kdump-compressed format need the whole memory dumped, so paging or
2099      * filter is not supported here.
2100      */
2101     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
2102         (paging || has_begin || has_length)) {
2103         error_setg(errp, "kdump-compressed format doesn't support paging or "
2104                          "filter");
2105         return;
2106     }
2107     if (has_begin && !has_length) {
2108         error_setg(errp, QERR_MISSING_PARAMETER, "length");
2109         return;
2110     }
2111     if (!has_begin && has_length) {
2112         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
2113         return;
2114     }
2115     if (has_detach) {
2116         detach_p = detach;
2117     }
2118 
2119     /* check whether lzo/snappy is supported */
2120 #ifndef CONFIG_LZO
2121     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
2122         error_setg(errp, "kdump-lzo is not available now");
2123         return;
2124     }
2125 #endif
2126 
2127 #ifndef CONFIG_SNAPPY
2128     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
2129         error_setg(errp, "kdump-snappy is not available now");
2130         return;
2131     }
2132 #endif
2133 
2134 #ifndef TARGET_X86_64
2135     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2136         error_setg(errp, "Windows dump is only available for x86-64");
2137         return;
2138     }
2139 #endif
2140 
2141 #if !defined(WIN32)
2142     if (strstart(file, "fd:", &p)) {
2143         fd = monitor_get_fd(monitor_cur(), p, errp);
2144         if (fd == -1) {
2145             return;
2146         }
2147     }
2148 #endif
2149 
2150     if  (strstart(file, "file:", &p)) {
2151         fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
2152         if (fd < 0) {
2153             error_setg_file_open(errp, errno, p);
2154             return;
2155         }
2156     }
2157 
2158     if (fd == -1) {
2159         error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
2160         return;
2161     }
2162 
2163     if (!dump_migration_blocker) {
2164         error_setg(&dump_migration_blocker,
2165                    "Live migration disabled: dump-guest-memory in progress");
2166     }
2167 
2168     /*
2169      * Allows even for -only-migratable, but forbid migration during the
2170      * process of dump guest memory.
2171      */
2172     if (migrate_add_blocker_internal(dump_migration_blocker, errp)) {
2173         /* Remember to release the fd before passing it over to dump state */
2174         close(fd);
2175         return;
2176     }
2177 
2178     s = &dump_state_global;
2179     dump_state_prepare(s);
2180 
2181     dump_init(s, fd, has_format, format, paging, has_begin,
2182               begin, length, errp);
2183     if (*errp) {
2184         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2185         return;
2186     }
2187 
2188     if (detach_p) {
2189         /* detached dump */
2190         s->detached = true;
2191         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2192                            s, QEMU_THREAD_DETACHED);
2193     } else {
2194         /* sync dump */
2195         dump_process(s, errp);
2196     }
2197 }
2198 
2199 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2200 {
2201     DumpGuestMemoryCapability *cap =
2202                                   g_new0(DumpGuestMemoryCapability, 1);
2203     DumpGuestMemoryFormatList **tail = &cap->formats;
2204 
2205     /* elf is always available */
2206     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2207 
2208     /* kdump-zlib is always available */
2209     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2210 
2211     /* add new item if kdump-lzo is available */
2212 #ifdef CONFIG_LZO
2213     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2214 #endif
2215 
2216     /* add new item if kdump-snappy is available */
2217 #ifdef CONFIG_SNAPPY
2218     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2219 #endif
2220 
2221     /* Windows dump is available only if target is x86_64 */
2222 #ifdef TARGET_X86_64
2223     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2224 #endif
2225 
2226     return cap;
2227 }
2228