xref: /openbmc/qemu/dump/dump.c (revision 7200fb21)
1 /*
2  * QEMU dump
3  *
4  * Copyright Fujitsu, Corp. 2011, 2012
5  *
6  * Authors:
7  *     Wen Congyang <wency@cn.fujitsu.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2 or later.
10  * See the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qemu/cutils.h"
16 #include "elf.h"
17 #include "qemu/bswap.h"
18 #include "exec/target_page.h"
19 #include "monitor/monitor.h"
20 #include "sysemu/dump.h"
21 #include "sysemu/runstate.h"
22 #include "sysemu/cpus.h"
23 #include "qapi/error.h"
24 #include "qapi/qapi-commands-dump.h"
25 #include "qapi/qapi-events-dump.h"
26 #include "qapi/qmp/qerror.h"
27 #include "qemu/error-report.h"
28 #include "qemu/main-loop.h"
29 #include "hw/misc/vmcoreinfo.h"
30 #include "migration/blocker.h"
31 #include "hw/core/cpu.h"
32 #include "win_dump.h"
33 
34 #include <zlib.h>
35 #ifdef CONFIG_LZO
36 #include <lzo/lzo1x.h>
37 #endif
38 #ifdef CONFIG_SNAPPY
39 #include <snappy-c.h>
40 #endif
41 #ifndef ELF_MACHINE_UNAME
42 #define ELF_MACHINE_UNAME "Unknown"
43 #endif
44 
45 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
46 
47 static Error *dump_migration_blocker;
48 
49 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size)   \
50     ((DIV_ROUND_UP((hdr_size), 4) +                     \
51       DIV_ROUND_UP((name_size), 4) +                    \
52       DIV_ROUND_UP((desc_size), 4)) * 4)
53 
54 static inline bool dump_is_64bit(DumpState *s)
55 {
56     return s->dump_info.d_class == ELFCLASS64;
57 }
58 
59 static inline bool dump_has_filter(DumpState *s)
60 {
61     return s->filter_area_length > 0;
62 }
63 
64 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
65 {
66     if (s->dump_info.d_endian == ELFDATA2LSB) {
67         val = cpu_to_le16(val);
68     } else {
69         val = cpu_to_be16(val);
70     }
71 
72     return val;
73 }
74 
75 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
76 {
77     if (s->dump_info.d_endian == ELFDATA2LSB) {
78         val = cpu_to_le32(val);
79     } else {
80         val = cpu_to_be32(val);
81     }
82 
83     return val;
84 }
85 
86 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
87 {
88     if (s->dump_info.d_endian == ELFDATA2LSB) {
89         val = cpu_to_le64(val);
90     } else {
91         val = cpu_to_be64(val);
92     }
93 
94     return val;
95 }
96 
97 static int dump_cleanup(DumpState *s)
98 {
99     if (s->dump_info.arch_cleanup_fn) {
100         s->dump_info.arch_cleanup_fn(s);
101     }
102 
103     guest_phys_blocks_free(&s->guest_phys_blocks);
104     memory_mapping_list_free(&s->list);
105     close(s->fd);
106     g_free(s->guest_note);
107     g_clear_pointer(&s->string_table_buf, g_array_unref);
108     s->guest_note = NULL;
109     if (s->resume) {
110         if (s->detached) {
111             qemu_mutex_lock_iothread();
112         }
113         vm_start();
114         if (s->detached) {
115             qemu_mutex_unlock_iothread();
116         }
117     }
118     migrate_del_blocker(&dump_migration_blocker);
119 
120     return 0;
121 }
122 
123 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
124 {
125     DumpState *s = opaque;
126     size_t written_size;
127 
128     written_size = qemu_write_full(s->fd, buf, size);
129     if (written_size != size) {
130         return -errno;
131     }
132 
133     return 0;
134 }
135 
136 static void prepare_elf64_header(DumpState *s, Elf64_Ehdr *elf_header)
137 {
138     /*
139      * phnum in the elf header is 16 bit, if we have more segments we
140      * set phnum to PN_XNUM and write the real number of segments to a
141      * special section.
142      */
143     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
144 
145     memset(elf_header, 0, sizeof(Elf64_Ehdr));
146     memcpy(elf_header, ELFMAG, SELFMAG);
147     elf_header->e_ident[EI_CLASS] = ELFCLASS64;
148     elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
149     elf_header->e_ident[EI_VERSION] = EV_CURRENT;
150     elf_header->e_type = cpu_to_dump16(s, ET_CORE);
151     elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
152     elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
153     elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
154     elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset);
155     elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
156     elf_header->e_phnum = cpu_to_dump16(s, phnum);
157     elf_header->e_shoff = cpu_to_dump64(s, s->shdr_offset);
158     elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
159     elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
160     elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
161 }
162 
163 static void prepare_elf32_header(DumpState *s, Elf32_Ehdr *elf_header)
164 {
165     /*
166      * phnum in the elf header is 16 bit, if we have more segments we
167      * set phnum to PN_XNUM and write the real number of segments to a
168      * special section.
169      */
170     uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
171 
172     memset(elf_header, 0, sizeof(Elf32_Ehdr));
173     memcpy(elf_header, ELFMAG, SELFMAG);
174     elf_header->e_ident[EI_CLASS] = ELFCLASS32;
175     elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
176     elf_header->e_ident[EI_VERSION] = EV_CURRENT;
177     elf_header->e_type = cpu_to_dump16(s, ET_CORE);
178     elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
179     elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
180     elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
181     elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset);
182     elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
183     elf_header->e_phnum = cpu_to_dump16(s, phnum);
184     elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset);
185     elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
186     elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
187     elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
188 }
189 
190 static void write_elf_header(DumpState *s, Error **errp)
191 {
192     Elf32_Ehdr elf32_header;
193     Elf64_Ehdr elf64_header;
194     size_t header_size;
195     void *header_ptr;
196     int ret;
197 
198     /* The NULL header and the shstrtab are always defined */
199     assert(s->shdr_num >= 2);
200     if (dump_is_64bit(s)) {
201         prepare_elf64_header(s, &elf64_header);
202         header_size = sizeof(elf64_header);
203         header_ptr = &elf64_header;
204     } else {
205         prepare_elf32_header(s, &elf32_header);
206         header_size = sizeof(elf32_header);
207         header_ptr = &elf32_header;
208     }
209 
210     ret = fd_write_vmcore(header_ptr, header_size, s);
211     if (ret < 0) {
212         error_setg_errno(errp, -ret, "dump: failed to write elf header");
213     }
214 }
215 
216 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
217                              int phdr_index, hwaddr offset,
218                              hwaddr filesz, Error **errp)
219 {
220     Elf64_Phdr phdr;
221     int ret;
222 
223     memset(&phdr, 0, sizeof(Elf64_Phdr));
224     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
225     phdr.p_offset = cpu_to_dump64(s, offset);
226     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
227     phdr.p_filesz = cpu_to_dump64(s, filesz);
228     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
229     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
230 
231     assert(memory_mapping->length >= filesz);
232 
233     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
234     if (ret < 0) {
235         error_setg_errno(errp, -ret,
236                          "dump: failed to write program header table");
237     }
238 }
239 
240 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
241                              int phdr_index, hwaddr offset,
242                              hwaddr filesz, Error **errp)
243 {
244     Elf32_Phdr phdr;
245     int ret;
246 
247     memset(&phdr, 0, sizeof(Elf32_Phdr));
248     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
249     phdr.p_offset = cpu_to_dump32(s, offset);
250     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
251     phdr.p_filesz = cpu_to_dump32(s, filesz);
252     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
253     phdr.p_vaddr =
254         cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
255 
256     assert(memory_mapping->length >= filesz);
257 
258     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
259     if (ret < 0) {
260         error_setg_errno(errp, -ret,
261                          "dump: failed to write program header table");
262     }
263 }
264 
265 static void prepare_elf64_phdr_note(DumpState *s, Elf64_Phdr *phdr)
266 {
267     memset(phdr, 0, sizeof(*phdr));
268     phdr->p_type = cpu_to_dump32(s, PT_NOTE);
269     phdr->p_offset = cpu_to_dump64(s, s->note_offset);
270     phdr->p_paddr = 0;
271     phdr->p_filesz = cpu_to_dump64(s, s->note_size);
272     phdr->p_memsz = cpu_to_dump64(s, s->note_size);
273     phdr->p_vaddr = 0;
274 }
275 
276 static inline int cpu_index(CPUState *cpu)
277 {
278     return cpu->cpu_index + 1;
279 }
280 
281 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
282                              Error **errp)
283 {
284     int ret;
285 
286     if (s->guest_note) {
287         ret = f(s->guest_note, s->guest_note_size, s);
288         if (ret < 0) {
289             error_setg(errp, "dump: failed to write guest note");
290         }
291     }
292 }
293 
294 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
295                               Error **errp)
296 {
297     CPUState *cpu;
298     int ret;
299     int id;
300 
301     CPU_FOREACH(cpu) {
302         id = cpu_index(cpu);
303         ret = cpu_write_elf64_note(f, cpu, id, s);
304         if (ret < 0) {
305             error_setg(errp, "dump: failed to write elf notes");
306             return;
307         }
308     }
309 
310     CPU_FOREACH(cpu) {
311         ret = cpu_write_elf64_qemunote(f, cpu, s);
312         if (ret < 0) {
313             error_setg(errp, "dump: failed to write CPU status");
314             return;
315         }
316     }
317 
318     write_guest_note(f, s, errp);
319 }
320 
321 static void prepare_elf32_phdr_note(DumpState *s, Elf32_Phdr *phdr)
322 {
323     memset(phdr, 0, sizeof(*phdr));
324     phdr->p_type = cpu_to_dump32(s, PT_NOTE);
325     phdr->p_offset = cpu_to_dump32(s, s->note_offset);
326     phdr->p_paddr = 0;
327     phdr->p_filesz = cpu_to_dump32(s, s->note_size);
328     phdr->p_memsz = cpu_to_dump32(s, s->note_size);
329     phdr->p_vaddr = 0;
330 }
331 
332 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
333                               Error **errp)
334 {
335     CPUState *cpu;
336     int ret;
337     int id;
338 
339     CPU_FOREACH(cpu) {
340         id = cpu_index(cpu);
341         ret = cpu_write_elf32_note(f, cpu, id, s);
342         if (ret < 0) {
343             error_setg(errp, "dump: failed to write elf notes");
344             return;
345         }
346     }
347 
348     CPU_FOREACH(cpu) {
349         ret = cpu_write_elf32_qemunote(f, cpu, s);
350         if (ret < 0) {
351             error_setg(errp, "dump: failed to write CPU status");
352             return;
353         }
354     }
355 
356     write_guest_note(f, s, errp);
357 }
358 
359 static void write_elf_phdr_note(DumpState *s, Error **errp)
360 {
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 
777     if (s->elf_section_data_size) {
778         s->elf_section_data = g_malloc0(s->elf_section_data_size);
779     }
780 
781     /* Adds the architecture defined section data to s->elf_section_data  */
782     if (s->dump_info.arch_sections_write_fn &&
783         s->elf_section_data_size) {
784         rc = s->dump_info.arch_sections_write_fn(s, s->elf_section_data);
785         if (rc) {
786             error_setg_errno(errp, rc,
787                              "dump: failed to get arch section data");
788             g_free(s->elf_section_data);
789             return;
790         }
791     }
792 
793     /* write sections to vmcore */
794     write_elf_sections(s, errp);
795 }
796 
797 static void create_vmcore(DumpState *s, Error **errp)
798 {
799     ERRP_GUARD();
800 
801     dump_begin(s, errp);
802     if (*errp) {
803         return;
804     }
805 
806     /* Iterate over memory and dump it to file */
807     dump_iterate(s, errp);
808     if (*errp) {
809         return;
810     }
811 
812     /* Write the section data */
813     dump_end(s, errp);
814 }
815 
816 static int write_start_flat_header(DumpState *s)
817 {
818     MakedumpfileHeader *mh;
819     int ret = 0;
820 
821     if (s->kdump_raw) {
822         return 0;
823     }
824 
825     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
826     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
827 
828     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
829            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
830 
831     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
832     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
833 
834     size_t written_size;
835     written_size = qemu_write_full(s->fd, mh, MAX_SIZE_MDF_HEADER);
836     if (written_size != MAX_SIZE_MDF_HEADER) {
837         ret = -1;
838     }
839 
840     g_free(mh);
841     return ret;
842 }
843 
844 static int write_end_flat_header(DumpState *s)
845 {
846     MakedumpfileDataHeader mdh;
847 
848     if (s->kdump_raw) {
849         return 0;
850     }
851 
852     mdh.offset = END_FLAG_FLAT_HEADER;
853     mdh.buf_size = END_FLAG_FLAT_HEADER;
854 
855     size_t written_size;
856     written_size = qemu_write_full(s->fd, &mdh, sizeof(mdh));
857     if (written_size != sizeof(mdh)) {
858         return -1;
859     }
860 
861     return 0;
862 }
863 
864 static int write_buffer(DumpState *s, off_t offset, const void *buf, size_t size)
865 {
866     size_t written_size;
867     MakedumpfileDataHeader mdh;
868     off_t seek_loc;
869 
870     if (s->kdump_raw) {
871         seek_loc = lseek(s->fd, offset, SEEK_SET);
872         if (seek_loc == (off_t) -1) {
873             return -1;
874         }
875     } else {
876         mdh.offset = cpu_to_be64(offset);
877         mdh.buf_size = cpu_to_be64(size);
878 
879         written_size = qemu_write_full(s->fd, &mdh, sizeof(mdh));
880         if (written_size != sizeof(mdh)) {
881             return -1;
882         }
883     }
884 
885     written_size = qemu_write_full(s->fd, buf, size);
886     if (written_size != size) {
887         return -1;
888     }
889 
890     return 0;
891 }
892 
893 static int buf_write_note(const void *buf, size_t size, void *opaque)
894 {
895     DumpState *s = opaque;
896 
897     /* note_buf is not enough */
898     if (s->note_buf_offset + size > s->note_size) {
899         return -1;
900     }
901 
902     memcpy(s->note_buf + s->note_buf_offset, buf, size);
903 
904     s->note_buf_offset += size;
905 
906     return 0;
907 }
908 
909 /*
910  * This function retrieves various sizes from an elf header.
911  *
912  * @note has to be a valid ELF note. The return sizes are unmodified
913  * (not padded or rounded up to be multiple of 4).
914  */
915 static void get_note_sizes(DumpState *s, const void *note,
916                            uint64_t *note_head_size,
917                            uint64_t *name_size,
918                            uint64_t *desc_size)
919 {
920     uint64_t note_head_sz;
921     uint64_t name_sz;
922     uint64_t desc_sz;
923 
924     if (dump_is_64bit(s)) {
925         const Elf64_Nhdr *hdr = note;
926         note_head_sz = sizeof(Elf64_Nhdr);
927         name_sz = cpu_to_dump64(s, hdr->n_namesz);
928         desc_sz = cpu_to_dump64(s, hdr->n_descsz);
929     } else {
930         const Elf32_Nhdr *hdr = note;
931         note_head_sz = sizeof(Elf32_Nhdr);
932         name_sz = cpu_to_dump32(s, hdr->n_namesz);
933         desc_sz = cpu_to_dump32(s, hdr->n_descsz);
934     }
935 
936     if (note_head_size) {
937         *note_head_size = note_head_sz;
938     }
939     if (name_size) {
940         *name_size = name_sz;
941     }
942     if (desc_size) {
943         *desc_size = desc_sz;
944     }
945 }
946 
947 static bool note_name_equal(DumpState *s,
948                             const uint8_t *note, const char *name)
949 {
950     int len = strlen(name) + 1;
951     uint64_t head_size, name_size;
952 
953     get_note_sizes(s, note, &head_size, &name_size, NULL);
954     head_size = ROUND_UP(head_size, 4);
955 
956     return name_size == len && memcmp(note + head_size, name, len) == 0;
957 }
958 
959 /* write common header, sub header and elf note to vmcore */
960 static void create_header32(DumpState *s, Error **errp)
961 {
962     ERRP_GUARD();
963     DiskDumpHeader32 *dh = NULL;
964     KdumpSubHeader32 *kh = NULL;
965     size_t size;
966     uint32_t block_size;
967     uint32_t sub_hdr_size;
968     uint32_t bitmap_blocks;
969     uint32_t status = 0;
970     uint64_t offset_note;
971 
972     /* write common header, the version of kdump-compressed format is 6th */
973     size = sizeof(DiskDumpHeader32);
974     dh = g_malloc0(size);
975 
976     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
977     dh->header_version = cpu_to_dump32(s, 6);
978     block_size = s->dump_info.page_size;
979     dh->block_size = cpu_to_dump32(s, block_size);
980     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
981     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
982     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
983     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
984     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
985     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
986     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
987     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
988     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
989 
990     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
991         status |= DUMP_DH_COMPRESSED_ZLIB;
992     }
993 #ifdef CONFIG_LZO
994     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
995         status |= DUMP_DH_COMPRESSED_LZO;
996     }
997 #endif
998 #ifdef CONFIG_SNAPPY
999     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1000         status |= DUMP_DH_COMPRESSED_SNAPPY;
1001     }
1002 #endif
1003     dh->status = cpu_to_dump32(s, status);
1004 
1005     if (write_buffer(s, 0, dh, size) < 0) {
1006         error_setg(errp, "dump: failed to write disk dump header");
1007         goto out;
1008     }
1009 
1010     /* write sub header */
1011     size = sizeof(KdumpSubHeader32);
1012     kh = g_malloc0(size);
1013 
1014     /* 64bit max_mapnr_64 */
1015     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1016     kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
1017     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1018 
1019     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1020     if (s->guest_note &&
1021         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1022         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1023 
1024         get_note_sizes(s, s->guest_note,
1025                        &hsize, &name_size, &size_vmcoreinfo_desc);
1026         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1027             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1028         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1029         kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
1030     }
1031 
1032     kh->offset_note = cpu_to_dump64(s, offset_note);
1033     kh->note_size = cpu_to_dump32(s, s->note_size);
1034 
1035     if (write_buffer(s, DISKDUMP_HEADER_BLOCKS *
1036                      block_size, kh, size) < 0) {
1037         error_setg(errp, "dump: failed to write kdump sub header");
1038         goto out;
1039     }
1040 
1041     /* write note */
1042     s->note_buf = g_malloc0(s->note_size);
1043     s->note_buf_offset = 0;
1044 
1045     /* use s->note_buf to store notes temporarily */
1046     write_elf32_notes(buf_write_note, s, errp);
1047     if (*errp) {
1048         goto out;
1049     }
1050     if (write_buffer(s, offset_note, s->note_buf,
1051                      s->note_size) < 0) {
1052         error_setg(errp, "dump: failed to write notes");
1053         goto out;
1054     }
1055 
1056     /* get offset of dump_bitmap */
1057     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1058                              block_size;
1059 
1060     /* get offset of page */
1061     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1062                      block_size;
1063 
1064 out:
1065     g_free(dh);
1066     g_free(kh);
1067     g_free(s->note_buf);
1068 }
1069 
1070 /* write common header, sub header and elf note to vmcore */
1071 static void create_header64(DumpState *s, Error **errp)
1072 {
1073     ERRP_GUARD();
1074     DiskDumpHeader64 *dh = NULL;
1075     KdumpSubHeader64 *kh = NULL;
1076     size_t size;
1077     uint32_t block_size;
1078     uint32_t sub_hdr_size;
1079     uint32_t bitmap_blocks;
1080     uint32_t status = 0;
1081     uint64_t offset_note;
1082 
1083     /* write common header, the version of kdump-compressed format is 6th */
1084     size = sizeof(DiskDumpHeader64);
1085     dh = g_malloc0(size);
1086 
1087     memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
1088     dh->header_version = cpu_to_dump32(s, 6);
1089     block_size = s->dump_info.page_size;
1090     dh->block_size = cpu_to_dump32(s, block_size);
1091     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
1092     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
1093     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
1094     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
1095     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
1096     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
1097     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
1098     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
1099     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
1100 
1101     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
1102         status |= DUMP_DH_COMPRESSED_ZLIB;
1103     }
1104 #ifdef CONFIG_LZO
1105     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
1106         status |= DUMP_DH_COMPRESSED_LZO;
1107     }
1108 #endif
1109 #ifdef CONFIG_SNAPPY
1110     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1111         status |= DUMP_DH_COMPRESSED_SNAPPY;
1112     }
1113 #endif
1114     dh->status = cpu_to_dump32(s, status);
1115 
1116     if (write_buffer(s, 0, dh, size) < 0) {
1117         error_setg(errp, "dump: failed to write disk dump header");
1118         goto out;
1119     }
1120 
1121     /* write sub header */
1122     size = sizeof(KdumpSubHeader64);
1123     kh = g_malloc0(size);
1124 
1125     /* 64bit max_mapnr_64 */
1126     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1127     kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
1128     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1129 
1130     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1131     if (s->guest_note &&
1132         note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1133         uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1134 
1135         get_note_sizes(s, s->guest_note,
1136                        &hsize, &name_size, &size_vmcoreinfo_desc);
1137         offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1138             (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1139         kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1140         kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
1141     }
1142 
1143     kh->offset_note = cpu_to_dump64(s, offset_note);
1144     kh->note_size = cpu_to_dump64(s, s->note_size);
1145 
1146     if (write_buffer(s, DISKDUMP_HEADER_BLOCKS *
1147                      block_size, kh, size) < 0) {
1148         error_setg(errp, "dump: failed to write kdump sub header");
1149         goto out;
1150     }
1151 
1152     /* write note */
1153     s->note_buf = g_malloc0(s->note_size);
1154     s->note_buf_offset = 0;
1155 
1156     /* use s->note_buf to store notes temporarily */
1157     write_elf64_notes(buf_write_note, s, errp);
1158     if (*errp) {
1159         goto out;
1160     }
1161 
1162     if (write_buffer(s, offset_note, s->note_buf,
1163                      s->note_size) < 0) {
1164         error_setg(errp, "dump: failed to write notes");
1165         goto out;
1166     }
1167 
1168     /* get offset of dump_bitmap */
1169     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1170                              block_size;
1171 
1172     /* get offset of page */
1173     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1174                      block_size;
1175 
1176 out:
1177     g_free(dh);
1178     g_free(kh);
1179     g_free(s->note_buf);
1180 }
1181 
1182 static void write_dump_header(DumpState *s, Error **errp)
1183 {
1184     if (dump_is_64bit(s)) {
1185         create_header64(s, errp);
1186     } else {
1187         create_header32(s, errp);
1188     }
1189 }
1190 
1191 static size_t dump_bitmap_get_bufsize(DumpState *s)
1192 {
1193     return s->dump_info.page_size;
1194 }
1195 
1196 /*
1197  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1198  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1199  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1200  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1201  * vmcore, ie. synchronizing un-sync bit into vmcore.
1202  */
1203 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1204                            uint8_t *buf, DumpState *s)
1205 {
1206     off_t old_offset, new_offset;
1207     off_t offset_bitmap1, offset_bitmap2;
1208     uint32_t byte, bit;
1209     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1210     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1211 
1212     /* should not set the previous place */
1213     assert(last_pfn <= pfn);
1214 
1215     /*
1216      * if the bit needed to be set is not cached in buf, flush the data in buf
1217      * to vmcore firstly.
1218      * making new_offset be bigger than old_offset can also sync remained data
1219      * into vmcore.
1220      */
1221     old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1222     new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1223 
1224     while (old_offset < new_offset) {
1225         /* calculate the offset and write dump_bitmap */
1226         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1227         if (write_buffer(s, offset_bitmap1, buf,
1228                          bitmap_bufsize) < 0) {
1229             return -1;
1230         }
1231 
1232         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1233         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1234                          old_offset;
1235         if (write_buffer(s, offset_bitmap2, buf,
1236                          bitmap_bufsize) < 0) {
1237             return -1;
1238         }
1239 
1240         memset(buf, 0, bitmap_bufsize);
1241         old_offset += bitmap_bufsize;
1242     }
1243 
1244     /* get the exact place of the bit in the buf, and set it */
1245     byte = (pfn % bits_per_buf) / CHAR_BIT;
1246     bit = (pfn % bits_per_buf) % CHAR_BIT;
1247     if (value) {
1248         buf[byte] |= 1u << bit;
1249     } else {
1250         buf[byte] &= ~(1u << bit);
1251     }
1252 
1253     return 0;
1254 }
1255 
1256 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1257 {
1258     int target_page_shift = ctz32(s->dump_info.page_size);
1259 
1260     return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1261 }
1262 
1263 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1264 {
1265     int target_page_shift = ctz32(s->dump_info.page_size);
1266 
1267     return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1268 }
1269 
1270 /*
1271  * Return the page frame number and the page content in *bufptr. bufptr can be
1272  * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated
1273  * memory. This is not necessarily the memory returned.
1274  */
1275 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1276                           uint8_t **bufptr, DumpState *s)
1277 {
1278     GuestPhysBlock *block = *blockptr;
1279     uint32_t page_size = s->dump_info.page_size;
1280     uint8_t *buf = NULL, *hbuf;
1281     hwaddr addr;
1282 
1283     /* block == NULL means the start of the iteration */
1284     if (!block) {
1285         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1286         *blockptr = block;
1287         addr = block->target_start;
1288         *pfnptr = dump_paddr_to_pfn(s, addr);
1289     } else {
1290         *pfnptr += 1;
1291         addr = dump_pfn_to_paddr(s, *pfnptr);
1292     }
1293     assert(block != NULL);
1294 
1295     while (1) {
1296         if (addr >= block->target_start && addr < block->target_end) {
1297             size_t n = MIN(block->target_end - addr, page_size - addr % page_size);
1298             hbuf = block->host_addr + (addr - block->target_start);
1299             if (!buf) {
1300                 if (n == page_size) {
1301                     /* this is a whole target page, go for it */
1302                     assert(addr % page_size == 0);
1303                     buf = hbuf;
1304                     break;
1305                 } else if (bufptr) {
1306                     assert(*bufptr);
1307                     buf = *bufptr;
1308                     memset(buf, 0, page_size);
1309                 } else {
1310                     return true;
1311                 }
1312             }
1313 
1314             memcpy(buf + addr % page_size, hbuf, n);
1315             addr += n;
1316             if (addr % page_size == 0 || addr >= block->target_end) {
1317                 /* we filled up the page or the current block is finished */
1318                 break;
1319             }
1320         } else {
1321             /* the next page is in the next block */
1322             *blockptr = block = QTAILQ_NEXT(block, next);
1323             if (!block) {
1324                 break;
1325             }
1326 
1327             addr = block->target_start;
1328             /* are we still in the same page? */
1329             if (dump_paddr_to_pfn(s, addr) != *pfnptr) {
1330                 if (buf) {
1331                     /* no, but we already filled something earlier, return it */
1332                     break;
1333                 } else {
1334                     /* else continue from there */
1335                     *pfnptr = dump_paddr_to_pfn(s, addr);
1336                 }
1337             }
1338         }
1339     }
1340 
1341     if (bufptr) {
1342         *bufptr = buf;
1343     }
1344 
1345     return buf != NULL;
1346 }
1347 
1348 static void write_dump_bitmap(DumpState *s, Error **errp)
1349 {
1350     int ret = 0;
1351     uint64_t last_pfn, pfn;
1352     void *dump_bitmap_buf;
1353     size_t num_dumpable;
1354     GuestPhysBlock *block_iter = NULL;
1355     size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1356     size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1357 
1358     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1359     dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1360 
1361     num_dumpable = 0;
1362     last_pfn = 0;
1363 
1364     /*
1365      * exam memory page by page, and set the bit in dump_bitmap corresponded
1366      * to the existing page.
1367      */
1368     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1369         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1370         if (ret < 0) {
1371             error_setg(errp, "dump: failed to set dump_bitmap");
1372             goto out;
1373         }
1374 
1375         last_pfn = pfn;
1376         num_dumpable++;
1377     }
1378 
1379     /*
1380      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1381      * set the remaining bits from last_pfn to the end of the bitmap buffer to
1382      * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1383      */
1384     if (num_dumpable > 0) {
1385         ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1386                               dump_bitmap_buf, s);
1387         if (ret < 0) {
1388             error_setg(errp, "dump: failed to sync dump_bitmap");
1389             goto out;
1390         }
1391     }
1392 
1393     /* number of dumpable pages that will be dumped later */
1394     s->num_dumpable = num_dumpable;
1395 
1396 out:
1397     g_free(dump_bitmap_buf);
1398 }
1399 
1400 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1401                                off_t offset)
1402 {
1403     data_cache->state = s;
1404     data_cache->data_size = 0;
1405     data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1406     data_cache->buf = g_malloc0(data_cache->buf_size);
1407     data_cache->offset = offset;
1408 }
1409 
1410 static int write_cache(DataCache *dc, const void *buf, size_t size,
1411                        bool flag_sync)
1412 {
1413     /*
1414      * dc->buf_size should not be less than size, otherwise dc will never be
1415      * enough
1416      */
1417     assert(size <= dc->buf_size);
1418 
1419     /*
1420      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1421      * otherwise check if the space is enough for caching data in buf, if not,
1422      * write the data in dc->buf to dc->state->fd and reset dc->buf
1423      */
1424     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1425         (flag_sync && dc->data_size > 0)) {
1426         if (write_buffer(dc->state, dc->offset, dc->buf, dc->data_size) < 0) {
1427             return -1;
1428         }
1429 
1430         dc->offset += dc->data_size;
1431         dc->data_size = 0;
1432     }
1433 
1434     if (!flag_sync) {
1435         memcpy(dc->buf + dc->data_size, buf, size);
1436         dc->data_size += size;
1437     }
1438 
1439     return 0;
1440 }
1441 
1442 static void free_data_cache(DataCache *data_cache)
1443 {
1444     g_free(data_cache->buf);
1445 }
1446 
1447 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1448 {
1449     switch (flag_compress) {
1450     case DUMP_DH_COMPRESSED_ZLIB:
1451         return compressBound(page_size);
1452 
1453     case DUMP_DH_COMPRESSED_LZO:
1454         /*
1455          * LZO will expand incompressible data by a little amount. Please check
1456          * the following URL to see the expansion calculation:
1457          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1458          */
1459         return page_size + page_size / 16 + 64 + 3;
1460 
1461 #ifdef CONFIG_SNAPPY
1462     case DUMP_DH_COMPRESSED_SNAPPY:
1463         return snappy_max_compressed_length(page_size);
1464 #endif
1465     }
1466     return 0;
1467 }
1468 
1469 static void write_dump_pages(DumpState *s, Error **errp)
1470 {
1471     int ret = 0;
1472     DataCache page_desc, page_data;
1473     size_t len_buf_out, size_out;
1474 #ifdef CONFIG_LZO
1475     lzo_bytep wrkmem = NULL;
1476 #endif
1477     uint8_t *buf_out = NULL;
1478     off_t offset_desc, offset_data;
1479     PageDescriptor pd, pd_zero;
1480     uint8_t *buf;
1481     GuestPhysBlock *block_iter = NULL;
1482     uint64_t pfn_iter;
1483     g_autofree uint8_t *page = NULL;
1484 
1485     /* get offset of page_desc and page_data in dump file */
1486     offset_desc = s->offset_page;
1487     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1488 
1489     prepare_data_cache(&page_desc, s, offset_desc);
1490     prepare_data_cache(&page_data, s, offset_data);
1491 
1492     /* prepare buffer to store compressed data */
1493     len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1494     assert(len_buf_out != 0);
1495 
1496 #ifdef CONFIG_LZO
1497     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1498 #endif
1499 
1500     buf_out = g_malloc(len_buf_out);
1501 
1502     /*
1503      * init zero page's page_desc and page_data, because every zero page
1504      * uses the same page_data
1505      */
1506     pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1507     pd_zero.flags = cpu_to_dump32(s, 0);
1508     pd_zero.offset = cpu_to_dump64(s, offset_data);
1509     pd_zero.page_flags = cpu_to_dump64(s, 0);
1510     buf = g_malloc0(s->dump_info.page_size);
1511     ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1512     g_free(buf);
1513     if (ret < 0) {
1514         error_setg(errp, "dump: failed to write page data (zero page)");
1515         goto out;
1516     }
1517 
1518     offset_data += s->dump_info.page_size;
1519     page = g_malloc(s->dump_info.page_size);
1520 
1521     /*
1522      * dump memory to vmcore page by page. zero page will all be resided in the
1523      * first page of page section
1524      */
1525     for (buf = page; get_next_page(&block_iter, &pfn_iter, &buf, s); buf = page) {
1526         /* check zero page */
1527         if (buffer_is_zero(buf, s->dump_info.page_size)) {
1528             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1529                               false);
1530             if (ret < 0) {
1531                 error_setg(errp, "dump: failed to write page desc");
1532                 goto out;
1533             }
1534         } else {
1535             /*
1536              * not zero page, then:
1537              * 1. compress the page
1538              * 2. write the compressed page into the cache of page_data
1539              * 3. get page desc of the compressed page and write it into the
1540              *    cache of page_desc
1541              *
1542              * only one compression format will be used here, for
1543              * s->flag_compress is set. But when compression fails to work,
1544              * we fall back to save in plaintext.
1545              */
1546              size_out = len_buf_out;
1547              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1548                     (compress2(buf_out, (uLongf *)&size_out, buf,
1549                                s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1550                     (size_out < s->dump_info.page_size)) {
1551                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1552                 pd.size  = cpu_to_dump32(s, size_out);
1553 
1554                 ret = write_cache(&page_data, buf_out, size_out, false);
1555                 if (ret < 0) {
1556                     error_setg(errp, "dump: failed to write page data");
1557                     goto out;
1558                 }
1559 #ifdef CONFIG_LZO
1560             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1561                     (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1562                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1563                     (size_out < s->dump_info.page_size)) {
1564                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1565                 pd.size  = cpu_to_dump32(s, size_out);
1566 
1567                 ret = write_cache(&page_data, buf_out, size_out, false);
1568                 if (ret < 0) {
1569                     error_setg(errp, "dump: failed to write page data");
1570                     goto out;
1571                 }
1572 #endif
1573 #ifdef CONFIG_SNAPPY
1574             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1575                     (snappy_compress((char *)buf, s->dump_info.page_size,
1576                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1577                     (size_out < s->dump_info.page_size)) {
1578                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1579                 pd.size  = cpu_to_dump32(s, size_out);
1580 
1581                 ret = write_cache(&page_data, buf_out, size_out, false);
1582                 if (ret < 0) {
1583                     error_setg(errp, "dump: failed to write page data");
1584                     goto out;
1585                 }
1586 #endif
1587             } else {
1588                 /*
1589                  * fall back to save in plaintext, size_out should be
1590                  * assigned the target's page size
1591                  */
1592                 pd.flags = cpu_to_dump32(s, 0);
1593                 size_out = s->dump_info.page_size;
1594                 pd.size = cpu_to_dump32(s, size_out);
1595 
1596                 ret = write_cache(&page_data, buf,
1597                                   s->dump_info.page_size, false);
1598                 if (ret < 0) {
1599                     error_setg(errp, "dump: failed to write page data");
1600                     goto out;
1601                 }
1602             }
1603 
1604             /* get and write page desc here */
1605             pd.page_flags = cpu_to_dump64(s, 0);
1606             pd.offset = cpu_to_dump64(s, offset_data);
1607             offset_data += size_out;
1608 
1609             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1610             if (ret < 0) {
1611                 error_setg(errp, "dump: failed to write page desc");
1612                 goto out;
1613             }
1614         }
1615         s->written_size += s->dump_info.page_size;
1616     }
1617 
1618     ret = write_cache(&page_desc, NULL, 0, true);
1619     if (ret < 0) {
1620         error_setg(errp, "dump: failed to sync cache for page_desc");
1621         goto out;
1622     }
1623     ret = write_cache(&page_data, NULL, 0, true);
1624     if (ret < 0) {
1625         error_setg(errp, "dump: failed to sync cache for page_data");
1626         goto out;
1627     }
1628 
1629 out:
1630     free_data_cache(&page_desc);
1631     free_data_cache(&page_data);
1632 
1633 #ifdef CONFIG_LZO
1634     g_free(wrkmem);
1635 #endif
1636 
1637     g_free(buf_out);
1638 }
1639 
1640 static void create_kdump_vmcore(DumpState *s, Error **errp)
1641 {
1642     ERRP_GUARD();
1643     int ret;
1644 
1645     /*
1646      * the kdump-compressed format is:
1647      *                                               File offset
1648      *  +------------------------------------------+ 0x0
1649      *  |    main header (struct disk_dump_header) |
1650      *  |------------------------------------------+ block 1
1651      *  |    sub header (struct kdump_sub_header)  |
1652      *  |------------------------------------------+ block 2
1653      *  |            1st-dump_bitmap               |
1654      *  |------------------------------------------+ block 2 + X blocks
1655      *  |            2nd-dump_bitmap               | (aligned by block)
1656      *  |------------------------------------------+ block 2 + 2 * X blocks
1657      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1658      *  |  page desc for pfn 1 (struct page_desc)  |
1659      *  |                    :                     |
1660      *  |------------------------------------------| (not aligned by block)
1661      *  |         page data (pfn 0)                |
1662      *  |         page data (pfn 1)                |
1663      *  |                    :                     |
1664      *  +------------------------------------------+
1665      */
1666 
1667     ret = write_start_flat_header(s);
1668     if (ret < 0) {
1669         error_setg(errp, "dump: failed to write start flat header");
1670         return;
1671     }
1672 
1673     write_dump_header(s, errp);
1674     if (*errp) {
1675         return;
1676     }
1677 
1678     write_dump_bitmap(s, errp);
1679     if (*errp) {
1680         return;
1681     }
1682 
1683     write_dump_pages(s, errp);
1684     if (*errp) {
1685         return;
1686     }
1687 
1688     ret = write_end_flat_header(s);
1689     if (ret < 0) {
1690         error_setg(errp, "dump: failed to write end flat header");
1691         return;
1692     }
1693 }
1694 
1695 static void get_max_mapnr(DumpState *s)
1696 {
1697     GuestPhysBlock *last_block;
1698 
1699     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1700     s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1701 }
1702 
1703 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1704 
1705 static void dump_state_prepare(DumpState *s)
1706 {
1707     /* zero the struct, setting status to active */
1708     *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1709 }
1710 
1711 bool qemu_system_dump_in_progress(void)
1712 {
1713     DumpState *state = &dump_state_global;
1714     return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1715 }
1716 
1717 /*
1718  * calculate total size of memory to be dumped (taking filter into
1719  * account.)
1720  */
1721 static int64_t dump_calculate_size(DumpState *s)
1722 {
1723     GuestPhysBlock *block;
1724     int64_t total = 0;
1725 
1726     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1727         total += dump_filtered_memblock_size(block,
1728                                              s->filter_area_begin,
1729                                              s->filter_area_length);
1730     }
1731 
1732     return total;
1733 }
1734 
1735 static void vmcoreinfo_update_phys_base(DumpState *s)
1736 {
1737     uint64_t size, note_head_size, name_size, phys_base;
1738     char **lines;
1739     uint8_t *vmci;
1740     size_t i;
1741 
1742     if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1743         return;
1744     }
1745 
1746     get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1747     note_head_size = ROUND_UP(note_head_size, 4);
1748 
1749     vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1750     *(vmci + size) = '\0';
1751 
1752     lines = g_strsplit((char *)vmci, "\n", -1);
1753     for (i = 0; lines[i]; i++) {
1754         const char *prefix = NULL;
1755 
1756         if (s->dump_info.d_machine == EM_X86_64) {
1757             prefix = "NUMBER(phys_base)=";
1758         } else if (s->dump_info.d_machine == EM_AARCH64) {
1759             prefix = "NUMBER(PHYS_OFFSET)=";
1760         }
1761 
1762         if (prefix && g_str_has_prefix(lines[i], prefix)) {
1763             if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1764                               &phys_base) < 0) {
1765                 warn_report("Failed to read %s", prefix);
1766             } else {
1767                 s->dump_info.phys_base = phys_base;
1768             }
1769             break;
1770         }
1771     }
1772 
1773     g_strfreev(lines);
1774 }
1775 
1776 static void dump_init(DumpState *s, int fd, bool has_format,
1777                       DumpGuestMemoryFormat format, bool paging, bool has_filter,
1778                       int64_t begin, int64_t length, bool kdump_raw,
1779                       Error **errp)
1780 {
1781     ERRP_GUARD();
1782     VMCoreInfoState *vmci = vmcoreinfo_find();
1783     CPUState *cpu;
1784     int nr_cpus;
1785     int ret;
1786 
1787     s->has_format = has_format;
1788     s->format = format;
1789     s->written_size = 0;
1790     s->kdump_raw = kdump_raw;
1791 
1792     /* kdump-compressed is conflict with paging and filter */
1793     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1794         assert(!paging && !has_filter);
1795     }
1796 
1797     if (runstate_is_running()) {
1798         vm_stop(RUN_STATE_SAVE_VM);
1799         s->resume = true;
1800     } else {
1801         s->resume = false;
1802     }
1803 
1804     /* If we use KVM, we should synchronize the registers before we get dump
1805      * info or physmap info.
1806      */
1807     cpu_synchronize_all_states();
1808     nr_cpus = 0;
1809     CPU_FOREACH(cpu) {
1810         nr_cpus++;
1811     }
1812 
1813     s->fd = fd;
1814     if (has_filter && !length) {
1815         error_setg(errp, "parameter 'length' expects a non-zero size");
1816         goto cleanup;
1817     }
1818     s->filter_area_begin = begin;
1819     s->filter_area_length = length;
1820 
1821     /* First index is 0, it's the special null name */
1822     s->string_table_buf = g_array_new(FALSE, TRUE, 1);
1823     /*
1824      * Allocate the null name, due to the clearing option set to true
1825      * it will be 0.
1826      */
1827     g_array_set_size(s->string_table_buf, 1);
1828 
1829     memory_mapping_list_init(&s->list);
1830 
1831     guest_phys_blocks_init(&s->guest_phys_blocks);
1832     guest_phys_blocks_append(&s->guest_phys_blocks);
1833     s->total_size = dump_calculate_size(s);
1834 #ifdef DEBUG_DUMP_GUEST_MEMORY
1835     fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1836 #endif
1837 
1838     /* it does not make sense to dump non-existent memory */
1839     if (!s->total_size) {
1840         error_setg(errp, "dump: no guest memory to dump");
1841         goto cleanup;
1842     }
1843 
1844     /* get dump info: endian, class and architecture.
1845      * If the target architecture is not supported, cpu_get_dump_info() will
1846      * return -1.
1847      */
1848     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1849     if (ret < 0) {
1850         error_setg(errp,
1851                    "dumping guest memory is not supported on this target");
1852         goto cleanup;
1853     }
1854 
1855     if (!s->dump_info.page_size) {
1856         s->dump_info.page_size = qemu_target_page_size();
1857     }
1858 
1859     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1860                                      s->dump_info.d_machine, nr_cpus);
1861     assert(s->note_size >= 0);
1862 
1863     /*
1864      * The goal of this block is to (a) update the previously guessed
1865      * phys_base, (b) copy the guest note out of the guest.
1866      * Failure to do so is not fatal for dumping.
1867      */
1868     if (vmci) {
1869         uint64_t addr, note_head_size, name_size, desc_size;
1870         uint32_t size;
1871         uint16_t guest_format;
1872 
1873         note_head_size = dump_is_64bit(s) ?
1874             sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr);
1875 
1876         guest_format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1877         size = le32_to_cpu(vmci->vmcoreinfo.size);
1878         addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1879         if (!vmci->has_vmcoreinfo) {
1880             warn_report("guest note is not present");
1881         } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1882             warn_report("guest note size is invalid: %" PRIu32, size);
1883         } else if (guest_format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1884             warn_report("guest note format is unsupported: %" PRIu16, guest_format);
1885         } else {
1886             s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1887             cpu_physical_memory_read(addr, s->guest_note, size);
1888 
1889             get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1890             s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1891                                                desc_size);
1892             if (name_size > MAX_GUEST_NOTE_SIZE ||
1893                 desc_size > MAX_GUEST_NOTE_SIZE ||
1894                 s->guest_note_size > size) {
1895                 warn_report("Invalid guest note header");
1896                 g_free(s->guest_note);
1897                 s->guest_note = NULL;
1898             } else {
1899                 vmcoreinfo_update_phys_base(s);
1900                 s->note_size += s->guest_note_size;
1901             }
1902         }
1903     }
1904 
1905     /* get memory mapping */
1906     if (paging) {
1907         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1908         if (*errp) {
1909             goto cleanup;
1910         }
1911     } else {
1912         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1913     }
1914 
1915     s->nr_cpus = nr_cpus;
1916 
1917     get_max_mapnr(s);
1918 
1919     uint64_t tmp;
1920     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1921                        s->dump_info.page_size);
1922     s->len_dump_bitmap = tmp * s->dump_info.page_size;
1923 
1924     /* init for kdump-compressed format */
1925     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1926         switch (format) {
1927         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1928             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1929             break;
1930 
1931         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1932 #ifdef CONFIG_LZO
1933             if (lzo_init() != LZO_E_OK) {
1934                 error_setg(errp, "failed to initialize the LZO library");
1935                 goto cleanup;
1936             }
1937 #endif
1938             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1939             break;
1940 
1941         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1942             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1943             break;
1944 
1945         default:
1946             s->flag_compress = 0;
1947         }
1948 
1949         return;
1950     }
1951 
1952     if (dump_has_filter(s)) {
1953         memory_mapping_filter(&s->list, s->filter_area_begin, s->filter_area_length);
1954     }
1955 
1956     /*
1957      * The first section header is always a special one in which most
1958      * fields are 0. The section header string table is also always
1959      * set.
1960      */
1961     s->shdr_num = 2;
1962 
1963     /*
1964      * Adds the number of architecture sections to shdr_num and sets
1965      * elf_section_data_size so we know the offsets and sizes of all
1966      * parts.
1967      */
1968     if (s->dump_info.arch_sections_add_fn) {
1969         s->dump_info.arch_sections_add_fn(s);
1970     }
1971 
1972     /*
1973      * calculate shdr_num so we know the offsets and sizes of all
1974      * parts.
1975      * Calculate phdr_num
1976      *
1977      * The absolute maximum amount of phdrs is UINT32_MAX - 1 as
1978      * sh_info is 32 bit. There's special handling once we go over
1979      * UINT16_MAX - 1 but that is handled in the ehdr and section
1980      * code.
1981      */
1982     s->phdr_num = 1; /* Reserve PT_NOTE */
1983     if (s->list.num <= UINT32_MAX - 1) {
1984         s->phdr_num += s->list.num;
1985     } else {
1986         s->phdr_num = UINT32_MAX;
1987     }
1988 
1989     /*
1990      * Now that the number of section and program headers is known we
1991      * can calculate the offsets of the headers and data.
1992      */
1993     if (dump_is_64bit(s)) {
1994         s->shdr_offset = sizeof(Elf64_Ehdr);
1995         s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
1996         s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
1997     } else {
1998         s->shdr_offset = sizeof(Elf32_Ehdr);
1999         s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
2000         s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
2001     }
2002     s->memory_offset = s->note_offset + s->note_size;
2003     s->section_offset = s->memory_offset + s->total_size;
2004 
2005     return;
2006 
2007 cleanup:
2008     dump_cleanup(s);
2009 }
2010 
2011 /* this operation might be time consuming. */
2012 static void dump_process(DumpState *s, Error **errp)
2013 {
2014     ERRP_GUARD();
2015     DumpQueryResult *result = NULL;
2016 
2017     if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2018         create_win_dump(s, errp);
2019     } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
2020         create_kdump_vmcore(s, errp);
2021     } else {
2022         create_vmcore(s, errp);
2023     }
2024 
2025     /* make sure status is written after written_size updates */
2026     smp_wmb();
2027     qatomic_set(&s->status,
2028                (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
2029 
2030     /* send DUMP_COMPLETED message (unconditionally) */
2031     result = qmp_query_dump(NULL);
2032     /* should never fail */
2033     assert(result);
2034     qapi_event_send_dump_completed(result,
2035                                    *errp ? error_get_pretty(*errp) : NULL);
2036     qapi_free_DumpQueryResult(result);
2037 
2038     dump_cleanup(s);
2039 }
2040 
2041 static void *dump_thread(void *data)
2042 {
2043     DumpState *s = (DumpState *)data;
2044     dump_process(s, NULL);
2045     return NULL;
2046 }
2047 
2048 DumpQueryResult *qmp_query_dump(Error **errp)
2049 {
2050     DumpQueryResult *result = g_new(DumpQueryResult, 1);
2051     DumpState *state = &dump_state_global;
2052     result->status = qatomic_read(&state->status);
2053     /* make sure we are reading status and written_size in order */
2054     smp_rmb();
2055     result->completed = state->written_size;
2056     result->total = state->total_size;
2057     return result;
2058 }
2059 
2060 void qmp_dump_guest_memory(bool paging, const char *protocol,
2061                            bool has_detach, bool detach,
2062                            bool has_begin, int64_t begin,
2063                            bool has_length, int64_t length,
2064                            bool has_format, DumpGuestMemoryFormat format,
2065                            Error **errp)
2066 {
2067     ERRP_GUARD();
2068     const char *p;
2069     int fd;
2070     DumpState *s;
2071     bool detach_p = false;
2072     bool kdump_raw = false;
2073 
2074     if (runstate_check(RUN_STATE_INMIGRATE)) {
2075         error_setg(errp, "Dump not allowed during incoming migration.");
2076         return;
2077     }
2078 
2079     /* if there is a dump in background, we should wait until the dump
2080      * finished */
2081     if (qemu_system_dump_in_progress()) {
2082         error_setg(errp, "There is a dump in process, please wait.");
2083         return;
2084     }
2085 
2086     /*
2087      * externally, we represent kdump-raw-* as separate formats, but internally
2088      * they are handled the same, except for the "raw" flag
2089      */
2090     if (has_format) {
2091         switch (format) {
2092         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_ZLIB:
2093             format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
2094             kdump_raw = true;
2095             break;
2096         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_LZO:
2097             format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
2098             kdump_raw = true;
2099             break;
2100         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_SNAPPY:
2101             format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
2102             kdump_raw = true;
2103             break;
2104         default:
2105             break;
2106         }
2107     }
2108 
2109     /*
2110      * kdump-compressed format need the whole memory dumped, so paging or
2111      * filter is not supported here.
2112      */
2113     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
2114         (paging || has_begin || has_length)) {
2115         error_setg(errp, "kdump-compressed format doesn't support paging or "
2116                          "filter");
2117         return;
2118     }
2119     if (has_begin && !has_length) {
2120         error_setg(errp, QERR_MISSING_PARAMETER, "length");
2121         return;
2122     }
2123     if (!has_begin && has_length) {
2124         error_setg(errp, QERR_MISSING_PARAMETER, "begin");
2125         return;
2126     }
2127     if (has_detach) {
2128         detach_p = detach;
2129     }
2130 
2131     /* check whether lzo/snappy is supported */
2132 #ifndef CONFIG_LZO
2133     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
2134         error_setg(errp, "kdump-lzo is not available now");
2135         return;
2136     }
2137 #endif
2138 
2139 #ifndef CONFIG_SNAPPY
2140     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
2141         error_setg(errp, "kdump-snappy is not available now");
2142         return;
2143     }
2144 #endif
2145 
2146     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
2147         && !win_dump_available(errp)) {
2148         return;
2149     }
2150 
2151     if (strstart(protocol, "fd:", &p)) {
2152         fd = monitor_get_fd(monitor_cur(), p, errp);
2153         if (fd == -1) {
2154             return;
2155         }
2156     } else if  (strstart(protocol, "file:", &p)) {
2157         fd = qemu_create(p, O_WRONLY | O_TRUNC | O_BINARY, S_IRUSR, errp);
2158         if (fd < 0) {
2159             return;
2160         }
2161     } else {
2162         error_setg(errp,
2163                    "parameter 'protocol' must start with 'file:' or 'fd:'");
2164         return;
2165     }
2166     if (kdump_raw && lseek(fd, 0, SEEK_CUR) == (off_t) -1) {
2167         close(fd);
2168         error_setg(errp, "kdump-raw formats require a seekable file");
2169         return;
2170     }
2171 
2172     if (!dump_migration_blocker) {
2173         error_setg(&dump_migration_blocker,
2174                    "Live migration disabled: dump-guest-memory in progress");
2175     }
2176 
2177     /*
2178      * Allows even for -only-migratable, but forbid migration during the
2179      * process of dump guest memory.
2180      */
2181     if (migrate_add_blocker_internal(&dump_migration_blocker, errp)) {
2182         /* Remember to release the fd before passing it over to dump state */
2183         close(fd);
2184         return;
2185     }
2186 
2187     s = &dump_state_global;
2188     dump_state_prepare(s);
2189 
2190     dump_init(s, fd, has_format, format, paging, has_begin,
2191               begin, length, kdump_raw, errp);
2192     if (*errp) {
2193         qatomic_set(&s->status, DUMP_STATUS_FAILED);
2194         return;
2195     }
2196 
2197     if (detach_p) {
2198         /* detached dump */
2199         s->detached = true;
2200         qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2201                            s, QEMU_THREAD_DETACHED);
2202     } else {
2203         /* sync dump */
2204         dump_process(s, errp);
2205     }
2206 }
2207 
2208 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2209 {
2210     DumpGuestMemoryCapability *cap =
2211                                   g_new0(DumpGuestMemoryCapability, 1);
2212     DumpGuestMemoryFormatList **tail = &cap->formats;
2213 
2214     /* elf is always available */
2215     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2216 
2217     /* kdump-zlib is always available */
2218     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2219     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_ZLIB);
2220 
2221     /* add new item if kdump-lzo is available */
2222 #ifdef CONFIG_LZO
2223     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2224     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_LZO);
2225 #endif
2226 
2227     /* add new item if kdump-snappy is available */
2228 #ifdef CONFIG_SNAPPY
2229     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2230     QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_SNAPPY);
2231 #endif
2232 
2233     if (win_dump_available(NULL)) {
2234         QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2235     }
2236 
2237     return cap;
2238 }
2239