qemu

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
#include "qemu/range.h"
34

35
#include <zlib.h>
36
#ifdef CONFIG_LZO
37
#include <lzo/lzo1x.h>
38
#endif
39
#ifdef CONFIG_SNAPPY
40
#include <snappy-c.h>
41
#endif
42
#ifndef ELF_MACHINE_UNAME
43
#define ELF_MACHINE_UNAME "Unknown"
44
#endif
45

46
#define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
47

48
static Error *dump_migration_blocker;
49

50
#define ELF_NOTE_SIZE(hdr_size, name_size, desc_size)   \
51
    ((DIV_ROUND_UP((hdr_size), 4) +                     \
52
      DIV_ROUND_UP((name_size), 4) +                    \
53
      DIV_ROUND_UP((desc_size), 4)) * 4)
54

55
static inline bool dump_is_64bit(DumpState *s)
56
{
57
    return s->dump_info.d_class == ELFCLASS64;
58
}
59

60
static inline bool dump_has_filter(DumpState *s)
61
{
62
    return s->filter_area_length > 0;
63
}
64

65
uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
66
{
67
    if (s->dump_info.d_endian == ELFDATA2LSB) {
68
        val = cpu_to_le16(val);
69
    } else {
70
        val = cpu_to_be16(val);
71
    }
72

73
    return val;
74
}
75

76
uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
77
{
78
    if (s->dump_info.d_endian == ELFDATA2LSB) {
79
        val = cpu_to_le32(val);
80
    } else {
81
        val = cpu_to_be32(val);
82
    }
83

84
    return val;
85
}
86

87
uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
88
{
89
    if (s->dump_info.d_endian == ELFDATA2LSB) {
90
        val = cpu_to_le64(val);
91
    } else {
92
        val = cpu_to_be64(val);
93
    }
94

95
    return val;
96
}
97

98
static int dump_cleanup(DumpState *s)
99
{
100
    if (s->dump_info.arch_cleanup_fn) {
101
        s->dump_info.arch_cleanup_fn(s);
102
    }
103

104
    guest_phys_blocks_free(&s->guest_phys_blocks);
105
    memory_mapping_list_free(&s->list);
106
    close(s->fd);
107
    g_free(s->guest_note);
108
    g_clear_pointer(&s->string_table_buf, g_array_unref);
109
    s->guest_note = NULL;
110
    if (s->resume) {
111
        if (s->detached) {
112
            bql_lock();
113
        }
114
        vm_start();
115
        if (s->detached) {
116
            bql_unlock();
117
        }
118
    }
119
    migrate_del_blocker(&dump_migration_blocker);
120

121
    return 0;
122
}
123

124
static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
125
{
126
    DumpState *s = opaque;
127
    size_t written_size;
128

129
    written_size = qemu_write_full(s->fd, buf, size);
130
    if (written_size != size) {
131
        return -errno;
132
    }
133

134
    return 0;
135
}
136

137
static void prepare_elf64_header(DumpState *s, Elf64_Ehdr *elf_header)
138
{
139
    /*
140
     * phnum in the elf header is 16 bit, if we have more segments we
141
     * set phnum to PN_XNUM and write the real number of segments to a
142
     * special section.
143
     */
144
    uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
145

146
    memset(elf_header, 0, sizeof(Elf64_Ehdr));
147
    memcpy(elf_header, ELFMAG, SELFMAG);
148
    elf_header->e_ident[EI_CLASS] = ELFCLASS64;
149
    elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
150
    elf_header->e_ident[EI_VERSION] = EV_CURRENT;
151
    elf_header->e_type = cpu_to_dump16(s, ET_CORE);
152
    elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
153
    elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
154
    elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
155
    elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset);
156
    elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
157
    elf_header->e_phnum = cpu_to_dump16(s, phnum);
158
    elf_header->e_shoff = cpu_to_dump64(s, s->shdr_offset);
159
    elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
160
    elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
161
    elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
162
}
163

164
static void prepare_elf32_header(DumpState *s, Elf32_Ehdr *elf_header)
165
{
166
    /*
167
     * phnum in the elf header is 16 bit, if we have more segments we
168
     * set phnum to PN_XNUM and write the real number of segments to a
169
     * special section.
170
     */
171
    uint16_t phnum = MIN(s->phdr_num, PN_XNUM);
172

173
    memset(elf_header, 0, sizeof(Elf32_Ehdr));
174
    memcpy(elf_header, ELFMAG, SELFMAG);
175
    elf_header->e_ident[EI_CLASS] = ELFCLASS32;
176
    elf_header->e_ident[EI_DATA] = s->dump_info.d_endian;
177
    elf_header->e_ident[EI_VERSION] = EV_CURRENT;
178
    elf_header->e_type = cpu_to_dump16(s, ET_CORE);
179
    elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
180
    elf_header->e_version = cpu_to_dump32(s, EV_CURRENT);
181
    elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
182
    elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset);
183
    elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
184
    elf_header->e_phnum = cpu_to_dump16(s, phnum);
185
    elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset);
186
    elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
187
    elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num);
188
    elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1);
189
}
190

191
static void write_elf_header(DumpState *s, Error **errp)
192
{
193
    Elf32_Ehdr elf32_header;
194
    Elf64_Ehdr elf64_header;
195
    size_t header_size;
196
    void *header_ptr;
197
    int ret;
198

199
    /* The NULL header and the shstrtab are always defined */
200
    assert(s->shdr_num >= 2);
201
    if (dump_is_64bit(s)) {
202
        prepare_elf64_header(s, &elf64_header);
203
        header_size = sizeof(elf64_header);
204
        header_ptr = &elf64_header;
205
    } else {
206
        prepare_elf32_header(s, &elf32_header);
207
        header_size = sizeof(elf32_header);
208
        header_ptr = &elf32_header;
209
    }
210

211
    ret = fd_write_vmcore(header_ptr, header_size, s);
212
    if (ret < 0) {
213
        error_setg_errno(errp, -ret, "dump: failed to write elf header");
214
    }
215
}
216

217
static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
218
                             int phdr_index, hwaddr offset,
219
                             hwaddr filesz, Error **errp)
220
{
221
    Elf64_Phdr phdr;
222
    int ret;
223

224
    memset(&phdr, 0, sizeof(Elf64_Phdr));
225
    phdr.p_type = cpu_to_dump32(s, PT_LOAD);
226
    phdr.p_offset = cpu_to_dump64(s, offset);
227
    phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
228
    phdr.p_filesz = cpu_to_dump64(s, filesz);
229
    phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
230
    phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
231

232
    assert(memory_mapping->length >= filesz);
233

234
    ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
235
    if (ret < 0) {
236
        error_setg_errno(errp, -ret,
237
                         "dump: failed to write program header table");
238
    }
239
}
240

241
static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
242
                             int phdr_index, hwaddr offset,
243
                             hwaddr filesz, Error **errp)
244
{
245
    Elf32_Phdr phdr;
246
    int ret;
247

248
    memset(&phdr, 0, sizeof(Elf32_Phdr));
249
    phdr.p_type = cpu_to_dump32(s, PT_LOAD);
250
    phdr.p_offset = cpu_to_dump32(s, offset);
251
    phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
252
    phdr.p_filesz = cpu_to_dump32(s, filesz);
253
    phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
254
    phdr.p_vaddr =
255
        cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
256

257
    assert(memory_mapping->length >= filesz);
258

259
    ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
260
    if (ret < 0) {
261
        error_setg_errno(errp, -ret,
262
                         "dump: failed to write program header table");
263
    }
264
}
265

266
static void prepare_elf64_phdr_note(DumpState *s, Elf64_Phdr *phdr)
267
{
268
    memset(phdr, 0, sizeof(*phdr));
269
    phdr->p_type = cpu_to_dump32(s, PT_NOTE);
270
    phdr->p_offset = cpu_to_dump64(s, s->note_offset);
271
    phdr->p_paddr = 0;
272
    phdr->p_filesz = cpu_to_dump64(s, s->note_size);
273
    phdr->p_memsz = cpu_to_dump64(s, s->note_size);
274
    phdr->p_vaddr = 0;
275
}
276

277
static inline int cpu_index(CPUState *cpu)
278
{
279
    return cpu->cpu_index + 1;
280
}
281

282
static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
283
                             Error **errp)
284
{
285
    int ret;
286

287
    if (s->guest_note) {
288
        ret = f(s->guest_note, s->guest_note_size, s);
289
        if (ret < 0) {
290
            error_setg(errp, "dump: failed to write guest note");
291
        }
292
    }
293
}
294

295
static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
296
                              Error **errp)
297
{
298
    CPUState *cpu;
299
    int ret;
300
    int id;
301

302
    CPU_FOREACH(cpu) {
303
        id = cpu_index(cpu);
304
        ret = cpu_write_elf64_note(f, cpu, id, s);
305
        if (ret < 0) {
306
            error_setg(errp, "dump: failed to write elf notes");
307
            return;
308
        }
309
    }
310

311
    CPU_FOREACH(cpu) {
312
        ret = cpu_write_elf64_qemunote(f, cpu, s);
313
        if (ret < 0) {
314
            error_setg(errp, "dump: failed to write CPU status");
315
            return;
316
        }
317
    }
318

319
    write_guest_note(f, s, errp);
320
}
321

322
static void prepare_elf32_phdr_note(DumpState *s, Elf32_Phdr *phdr)
323
{
324
    memset(phdr, 0, sizeof(*phdr));
325
    phdr->p_type = cpu_to_dump32(s, PT_NOTE);
326
    phdr->p_offset = cpu_to_dump32(s, s->note_offset);
327
    phdr->p_paddr = 0;
328
    phdr->p_filesz = cpu_to_dump32(s, s->note_size);
329
    phdr->p_memsz = cpu_to_dump32(s, s->note_size);
330
    phdr->p_vaddr = 0;
331
}
332

333
static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
334
                              Error **errp)
335
{
336
    CPUState *cpu;
337
    int ret;
338
    int id;
339

340
    CPU_FOREACH(cpu) {
341
        id = cpu_index(cpu);
342
        ret = cpu_write_elf32_note(f, cpu, id, s);
343
        if (ret < 0) {
344
            error_setg(errp, "dump: failed to write elf notes");
345
            return;
346
        }
347
    }
348

349
    CPU_FOREACH(cpu) {
350
        ret = cpu_write_elf32_qemunote(f, cpu, s);
351
        if (ret < 0) {
352
            error_setg(errp, "dump: failed to write CPU status");
353
            return;
354
        }
355
    }
356

357
    write_guest_note(f, s, errp);
358
}
359

360
static void write_elf_phdr_note(DumpState *s, Error **errp)
361
{
362
    Elf32_Phdr phdr32;
363
    Elf64_Phdr phdr64;
364
    void *phdr;
365
    size_t size;
366
    int ret;
367

368
    if (dump_is_64bit(s)) {
369
        prepare_elf64_phdr_note(s, &phdr64);
370
        size = sizeof(phdr64);
371
        phdr = &phdr64;
372
    } else {
373
        prepare_elf32_phdr_note(s, &phdr32);
374
        size = sizeof(phdr32);
375
        phdr = &phdr32;
376
    }
377

378
    ret = fd_write_vmcore(phdr, size, s);
379
    if (ret < 0) {
380
        error_setg_errno(errp, -ret,
381
                         "dump: failed to write program header table");
382
    }
383
}
384

385
static void prepare_elf_section_hdr_zero(DumpState *s)
386
{
387
    if (dump_is_64bit(s)) {
388
        Elf64_Shdr *shdr64 = s->elf_section_hdrs;
389

390
        shdr64->sh_info = cpu_to_dump32(s, s->phdr_num);
391
    } else {
392
        Elf32_Shdr *shdr32 = s->elf_section_hdrs;
393

394
        shdr32->sh_info = cpu_to_dump32(s, s->phdr_num);
395
    }
396
}
397

398
static void prepare_elf_section_hdr_string(DumpState *s, void *buff)
399
{
400
    uint64_t index = s->string_table_buf->len;
401
    const char strtab[] = ".shstrtab";
402
    Elf32_Shdr shdr32 = {};
403
    Elf64_Shdr shdr64 = {};
404
    int shdr_size;
405
    void *shdr;
406

407
    g_array_append_vals(s->string_table_buf, strtab, sizeof(strtab));
408
    if (dump_is_64bit(s)) {
409
        shdr_size = sizeof(Elf64_Shdr);
410
        shdr64.sh_type = SHT_STRTAB;
411
        shdr64.sh_offset = s->section_offset + s->elf_section_data_size;
412
        shdr64.sh_name = index;
413
        shdr64.sh_size = s->string_table_buf->len;
414
        shdr = &shdr64;
415
    } else {
416
        shdr_size = sizeof(Elf32_Shdr);
417
        shdr32.sh_type = SHT_STRTAB;
418
        shdr32.sh_offset = s->section_offset + s->elf_section_data_size;
419
        shdr32.sh_name = index;
420
        shdr32.sh_size = s->string_table_buf->len;
421
        shdr = &shdr32;
422
    }
423
    memcpy(buff, shdr, shdr_size);
424
}
425

426
static bool prepare_elf_section_hdrs(DumpState *s, Error **errp)
427
{
428
    size_t len, sizeof_shdr;
429
    void *buff_hdr;
430

431
    /*
432
     * Section ordering:
433
     * - HDR zero
434
     * - Arch section hdrs
435
     * - String table hdr
436
     */
437
    sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
438
    len = sizeof_shdr * s->shdr_num;
439
    s->elf_section_hdrs = g_malloc0(len);
440
    buff_hdr = s->elf_section_hdrs;
441

442
    /*
443
     * The first section header is ALWAYS a special initial section
444
     * header.
445
     *
446
     * The header should be 0 with one exception being that if
447
     * phdr_num is PN_XNUM then the sh_info field contains the real
448
     * number of segment entries.
449
     *
450
     * As we zero allocate the buffer we will only need to modify
451
     * sh_info for the PN_XNUM case.
452
     */
453
    if (s->phdr_num >= PN_XNUM) {
454
        prepare_elf_section_hdr_zero(s);
455
    }
456
    buff_hdr += sizeof_shdr;
457

458
    /* Add architecture defined section headers */
459
    if (s->dump_info.arch_sections_write_hdr_fn
460
        && s->shdr_num > 2) {
461
        buff_hdr += s->dump_info.arch_sections_write_hdr_fn(s, buff_hdr);
462

463
        if (s->shdr_num >= SHN_LORESERVE) {
464
            error_setg_errno(errp, EINVAL,
465
                             "dump: too many architecture defined sections");
466
            return false;
467
        }
468
    }
469

470
    /*
471
     * String table is the last section since strings are added via
472
     * arch_sections_write_hdr().
473
     */
474
    prepare_elf_section_hdr_string(s, buff_hdr);
475
    return true;
476
}
477

478
static void write_elf_section_headers(DumpState *s, Error **errp)
479
{
480
    size_t sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr);
481
    int ret;
482

483
    if (!prepare_elf_section_hdrs(s, errp)) {
484
        return;
485
    }
486

487
    ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s);
488
    if (ret < 0) {
489
        error_setg_errno(errp, -ret, "dump: failed to write section headers");
490
    }
491

492
    g_free(s->elf_section_hdrs);
493
}
494

495
static void write_elf_sections(DumpState *s, Error **errp)
496
{
497
    int ret;
498

499
    if (s->elf_section_data_size) {
500
        /* Write architecture section data */
501
        ret = fd_write_vmcore(s->elf_section_data,
502
                              s->elf_section_data_size, s);
503
        if (ret < 0) {
504
            error_setg_errno(errp, -ret,
505
                             "dump: failed to write architecture section data");
506
            return;
507
        }
508
    }
509

510
    /* Write string table */
511
    ret = fd_write_vmcore(s->string_table_buf->data,
512
                          s->string_table_buf->len, s);
513
    if (ret < 0) {
514
        error_setg_errno(errp, -ret, "dump: failed to write string table data");
515
    }
516
}
517

518
static void write_data(DumpState *s, void *buf, int length, Error **errp)
519
{
520
    int ret;
521

522
    ret = fd_write_vmcore(buf, length, s);
523
    if (ret < 0) {
524
        error_setg_errno(errp, -ret, "dump: failed to save memory");
525
    } else {
526
        s->written_size += length;
527
    }
528
}
529

530
/* write the memory to vmcore. 1 page per I/O. */
531
static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
532
                         int64_t size, Error **errp)
533
{
534
    ERRP_GUARD();
535
    int64_t i;
536

537
    for (i = 0; i < size / s->dump_info.page_size; i++) {
538
        write_data(s, block->host_addr + start + i * s->dump_info.page_size,
539
                   s->dump_info.page_size, errp);
540
        if (*errp) {
541
            return;
542
        }
543
    }
544

545
    if ((size % s->dump_info.page_size) != 0) {
546
        write_data(s, block->host_addr + start + i * s->dump_info.page_size,
547
                   size % s->dump_info.page_size, errp);
548
        if (*errp) {
549
            return;
550
        }
551
    }
552
}
553

554
/* get the memory's offset and size in the vmcore */
555
static void get_offset_range(hwaddr phys_addr,
556
                             ram_addr_t mapping_length,
557
                             DumpState *s,
558
                             hwaddr *p_offset,
559
                             hwaddr *p_filesz)
560
{
561
    GuestPhysBlock *block;
562
    hwaddr offset = s->memory_offset;
563
    int64_t size_in_block, start;
564

565
    /* When the memory is not stored into vmcore, offset will be -1 */
566
    *p_offset = -1;
567
    *p_filesz = 0;
568

569
    if (dump_has_filter(s)) {
570
        if (phys_addr < s->filter_area_begin ||
571
            phys_addr >= s->filter_area_begin + s->filter_area_length) {
572
            return;
573
        }
574
    }
575

576
    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
577
        if (dump_has_filter(s)) {
578
            if (!ranges_overlap(block->target_start,
579
                                block->target_end - block->target_start,
580
                                s->filter_area_begin,
581
                                s->filter_area_length)) {
582
                /* This block is out of the range */
583
                continue;
584
            }
585

586
            if (s->filter_area_begin <= block->target_start) {
587
                start = block->target_start;
588
            } else {
589
                start = s->filter_area_begin;
590
            }
591

592
            size_in_block = block->target_end - start;
593
            if (s->filter_area_begin + s->filter_area_length < block->target_end) {
594
                size_in_block -= block->target_end - (s->filter_area_begin + s->filter_area_length);
595
            }
596
        } else {
597
            start = block->target_start;
598
            size_in_block = block->target_end - block->target_start;
599
        }
600

601
        if (phys_addr >= start && phys_addr < start + size_in_block) {
602
            *p_offset = phys_addr - start + offset;
603

604
            /* The offset range mapped from the vmcore file must not spill over
605
             * the GuestPhysBlock, clamp it. The rest of the mapping will be
606
             * zero-filled in memory at load time; see
607
             * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
608
             */
609
            *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
610
                        mapping_length :
611
                        size_in_block - (phys_addr - start);
612
            return;
613
        }
614

615
        offset += size_in_block;
616
    }
617
}
618

619
static void write_elf_phdr_loads(DumpState *s, Error **errp)
620
{
621
    ERRP_GUARD();
622
    hwaddr offset, filesz;
623
    MemoryMapping *memory_mapping;
624
    uint32_t phdr_index = 1;
625

626
    QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
627
        get_offset_range(memory_mapping->phys_addr,
628
                         memory_mapping->length,
629
                         s, &offset, &filesz);
630
        if (dump_is_64bit(s)) {
631
            write_elf64_load(s, memory_mapping, phdr_index++, offset,
632
                             filesz, errp);
633
        } else {
634
            write_elf32_load(s, memory_mapping, phdr_index++, offset,
635
                             filesz, errp);
636
        }
637

638
        if (*errp) {
639
            return;
640
        }
641

642
        if (phdr_index >= s->phdr_num) {
643
            break;
644
        }
645
    }
646
}
647

648
static void write_elf_notes(DumpState *s, Error **errp)
649
{
650
    if (dump_is_64bit(s)) {
651
        write_elf64_notes(fd_write_vmcore, s, errp);
652
    } else {
653
        write_elf32_notes(fd_write_vmcore, s, errp);
654
    }
655
}
656

657
/* write elf header, PT_NOTE and elf note to vmcore. */
658
static void dump_begin(DumpState *s, Error **errp)
659
{
660
    ERRP_GUARD();
661

662
    /*
663
     * the vmcore's format is:
664
     *   --------------
665
     *   |  elf header |
666
     *   --------------
667
     *   |  sctn_hdr   |
668
     *   --------------
669
     *   |  PT_NOTE    |
670
     *   --------------
671
     *   |  PT_LOAD    |
672
     *   --------------
673
     *   |  ......     |
674
     *   --------------
675
     *   |  PT_LOAD    |
676
     *   --------------
677
     *   |  elf note   |
678
     *   --------------
679
     *   |  memory     |
680
     *   --------------
681
     *
682
     * we only know where the memory is saved after we write elf note into
683
     * vmcore.
684
     */
685

686
    /* write elf header to vmcore */
687
    write_elf_header(s, errp);
688
    if (*errp) {
689
        return;
690
    }
691

692
    /* write section headers to vmcore */
693
    write_elf_section_headers(s, errp);
694
    if (*errp) {
695
        return;
696
    }
697

698
    /* write PT_NOTE to vmcore */
699
    write_elf_phdr_note(s, errp);
700
    if (*errp) {
701
        return;
702
    }
703

704
    /* write all PT_LOADs to vmcore */
705
    write_elf_phdr_loads(s, errp);
706
    if (*errp) {
707
        return;
708
    }
709

710
    /* write notes to vmcore */
711
    write_elf_notes(s, errp);
712
}
713

714
int64_t dump_filtered_memblock_size(GuestPhysBlock *block,
715
                                    int64_t filter_area_start,
716
                                    int64_t filter_area_length)
717
{
718
    int64_t size, left, right;
719

720
    /* No filter, return full size */
721
    if (!filter_area_length) {
722
        return block->target_end - block->target_start;
723
    }
724

725
    /* calculate the overlapped region. */
726
    left = MAX(filter_area_start, block->target_start);
727
    right = MIN(filter_area_start + filter_area_length, block->target_end);
728
    size = right - left;
729
    size = size > 0 ? size : 0;
730

731
    return size;
732
}
733

734
int64_t dump_filtered_memblock_start(GuestPhysBlock *block,
735
                                     int64_t filter_area_start,
736
                                     int64_t filter_area_length)
737
{
738
    if (filter_area_length) {
739
        /* return -1 if the block is not within filter area */
740
        if (!ranges_overlap(block->target_start,
741
                            block->target_end - block->target_start,
742
                            filter_area_start, filter_area_length)) {
743
            return -1;
744
        }
745

746
        if (filter_area_start > block->target_start) {
747
            return filter_area_start - block->target_start;
748
        }
749
    }
750

751
    return 0;
752
}
753

754
/* write all memory to vmcore */
755
static void dump_iterate(DumpState *s, Error **errp)
756
{
757
    ERRP_GUARD();
758
    GuestPhysBlock *block;
759
    int64_t memblock_size, memblock_start;
760

761
    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
762
        memblock_start = dump_filtered_memblock_start(block, s->filter_area_begin, s->filter_area_length);
763
        if (memblock_start == -1) {
764
            continue;
765
        }
766

767
        memblock_size = dump_filtered_memblock_size(block, s->filter_area_begin, s->filter_area_length);
768

769
        /* Write the memory to file */
770
        write_memory(s, block, memblock_start, memblock_size, errp);
771
        if (*errp) {
772
            return;
773
        }
774
    }
775
}
776

777
static void dump_end(DumpState *s, Error **errp)
778
{
779
    int rc;
780

781
    if (s->elf_section_data_size) {
782
        s->elf_section_data = g_malloc0(s->elf_section_data_size);
783
    }
784

785
    /* Adds the architecture defined section data to s->elf_section_data  */
786
    if (s->dump_info.arch_sections_write_fn &&
787
        s->elf_section_data_size) {
788
        rc = s->dump_info.arch_sections_write_fn(s, s->elf_section_data);
789
        if (rc) {
790
            error_setg_errno(errp, rc,
791
                             "dump: failed to get arch section data");
792
            g_free(s->elf_section_data);
793
            return;
794
        }
795
    }
796

797
    /* write sections to vmcore */
798
    write_elf_sections(s, errp);
799
}
800

801
static void create_vmcore(DumpState *s, Error **errp)
802
{
803
    ERRP_GUARD();
804

805
    dump_begin(s, errp);
806
    if (*errp) {
807
        return;
808
    }
809

810
    /* Iterate over memory and dump it to file */
811
    dump_iterate(s, errp);
812
    if (*errp) {
813
        return;
814
    }
815

816
    /* Write the section data */
817
    dump_end(s, errp);
818
}
819

820
static int write_start_flat_header(DumpState *s)
821
{
822
    MakedumpfileHeader *mh;
823
    int ret = 0;
824

825
    if (s->kdump_raw) {
826
        return 0;
827
    }
828

829
    QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
830
    mh = g_malloc0(MAX_SIZE_MDF_HEADER);
831

832
    memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
833
           MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
834

835
    mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
836
    mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
837

838
    size_t written_size;
839
    written_size = qemu_write_full(s->fd, mh, MAX_SIZE_MDF_HEADER);
840
    if (written_size != MAX_SIZE_MDF_HEADER) {
841
        ret = -1;
842
    }
843

844
    g_free(mh);
845
    return ret;
846
}
847

848
static int write_end_flat_header(DumpState *s)
849
{
850
    MakedumpfileDataHeader mdh;
851

852
    if (s->kdump_raw) {
853
        return 0;
854
    }
855

856
    mdh.offset = END_FLAG_FLAT_HEADER;
857
    mdh.buf_size = END_FLAG_FLAT_HEADER;
858

859
    size_t written_size;
860
    written_size = qemu_write_full(s->fd, &mdh, sizeof(mdh));
861
    if (written_size != sizeof(mdh)) {
862
        return -1;
863
    }
864

865
    return 0;
866
}
867

868
static int write_buffer(DumpState *s, off_t offset, const void *buf, size_t size)
869
{
870
    size_t written_size;
871
    MakedumpfileDataHeader mdh;
872
    off_t seek_loc;
873

874
    if (s->kdump_raw) {
875
        seek_loc = lseek(s->fd, offset, SEEK_SET);
876
        if (seek_loc == (off_t) -1) {
877
            return -1;
878
        }
879
    } else {
880
        mdh.offset = cpu_to_be64(offset);
881
        mdh.buf_size = cpu_to_be64(size);
882

883
        written_size = qemu_write_full(s->fd, &mdh, sizeof(mdh));
884
        if (written_size != sizeof(mdh)) {
885
            return -1;
886
        }
887
    }
888

889
    written_size = qemu_write_full(s->fd, buf, size);
890
    if (written_size != size) {
891
        return -1;
892
    }
893

894
    return 0;
895
}
896

897
static int buf_write_note(const void *buf, size_t size, void *opaque)
898
{
899
    DumpState *s = opaque;
900

901
    /* note_buf is not enough */
902
    if (s->note_buf_offset + size > s->note_size) {
903
        return -1;
904
    }
905

906
    memcpy(s->note_buf + s->note_buf_offset, buf, size);
907

908
    s->note_buf_offset += size;
909

910
    return 0;
911
}
912

913
/*
914
 * This function retrieves various sizes from an elf header.
915
 *
916
 * @note has to be a valid ELF note. The return sizes are unmodified
917
 * (not padded or rounded up to be multiple of 4).
918
 */
919
static void get_note_sizes(DumpState *s, const void *note,
920
                           uint64_t *note_head_size,
921
                           uint64_t *name_size,
922
                           uint64_t *desc_size)
923
{
924
    uint64_t note_head_sz;
925
    uint64_t name_sz;
926
    uint64_t desc_sz;
927

928
    if (dump_is_64bit(s)) {
929
        const Elf64_Nhdr *hdr = note;
930
        note_head_sz = sizeof(Elf64_Nhdr);
931
        name_sz = cpu_to_dump64(s, hdr->n_namesz);
932
        desc_sz = cpu_to_dump64(s, hdr->n_descsz);
933
    } else {
934
        const Elf32_Nhdr *hdr = note;
935
        note_head_sz = sizeof(Elf32_Nhdr);
936
        name_sz = cpu_to_dump32(s, hdr->n_namesz);
937
        desc_sz = cpu_to_dump32(s, hdr->n_descsz);
938
    }
939

940
    if (note_head_size) {
941
        *note_head_size = note_head_sz;
942
    }
943
    if (name_size) {
944
        *name_size = name_sz;
945
    }
946
    if (desc_size) {
947
        *desc_size = desc_sz;
948
    }
949
}
950

951
static bool note_name_equal(DumpState *s,
952
                            const uint8_t *note, const char *name)
953
{
954
    int len = strlen(name) + 1;
955
    uint64_t head_size, name_size;
956

957
    get_note_sizes(s, note, &head_size, &name_size, NULL);
958
    head_size = ROUND_UP(head_size, 4);
959

960
    return name_size == len && memcmp(note + head_size, name, len) == 0;
961
}
962

963
/* write common header, sub header and elf note to vmcore */
964
static void create_header32(DumpState *s, Error **errp)
965
{
966
    ERRP_GUARD();
967
    DiskDumpHeader32 *dh = NULL;
968
    KdumpSubHeader32 *kh = NULL;
969
    size_t size;
970
    uint32_t block_size;
971
    uint32_t sub_hdr_size;
972
    uint32_t bitmap_blocks;
973
    uint32_t status = 0;
974
    uint64_t offset_note;
975

976
    /* write common header, the version of kdump-compressed format is 6th */
977
    size = sizeof(DiskDumpHeader32);
978
    dh = g_malloc0(size);
979

980
    memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
981
    dh->header_version = cpu_to_dump32(s, 6);
982
    block_size = s->dump_info.page_size;
983
    dh->block_size = cpu_to_dump32(s, block_size);
984
    sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
985
    sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
986
    dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
987
    /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
988
    dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
989
    dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
990
    bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
991
    dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
992
    strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
993

994
    if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
995
        status |= DUMP_DH_COMPRESSED_ZLIB;
996
    }
997
#ifdef CONFIG_LZO
998
    if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
999
        status |= DUMP_DH_COMPRESSED_LZO;
1000
    }
1001
#endif
1002
#ifdef CONFIG_SNAPPY
1003
    if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1004
        status |= DUMP_DH_COMPRESSED_SNAPPY;
1005
    }
1006
#endif
1007
    dh->status = cpu_to_dump32(s, status);
1008

1009
    if (write_buffer(s, 0, dh, size) < 0) {
1010
        error_setg(errp, "dump: failed to write disk dump header");
1011
        goto out;
1012
    }
1013

1014
    /* write sub header */
1015
    size = sizeof(KdumpSubHeader32);
1016
    kh = g_malloc0(size);
1017

1018
    /* 64bit max_mapnr_64 */
1019
    kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1020
    kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
1021
    kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1022

1023
    offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1024
    if (s->guest_note &&
1025
        note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1026
        uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1027

1028
        get_note_sizes(s, s->guest_note,
1029
                       &hsize, &name_size, &size_vmcoreinfo_desc);
1030
        offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1031
            (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1032
        kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1033
        kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
1034
    }
1035

1036
    kh->offset_note = cpu_to_dump64(s, offset_note);
1037
    kh->note_size = cpu_to_dump32(s, s->note_size);
1038

1039
    if (write_buffer(s, DISKDUMP_HEADER_BLOCKS *
1040
                     block_size, kh, size) < 0) {
1041
        error_setg(errp, "dump: failed to write kdump sub header");
1042
        goto out;
1043
    }
1044

1045
    /* write note */
1046
    s->note_buf = g_malloc0(s->note_size);
1047
    s->note_buf_offset = 0;
1048

1049
    /* use s->note_buf to store notes temporarily */
1050
    write_elf32_notes(buf_write_note, s, errp);
1051
    if (*errp) {
1052
        goto out;
1053
    }
1054
    if (write_buffer(s, offset_note, s->note_buf,
1055
                     s->note_size) < 0) {
1056
        error_setg(errp, "dump: failed to write notes");
1057
        goto out;
1058
    }
1059

1060
    /* get offset of dump_bitmap */
1061
    s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1062
                             block_size;
1063

1064
    /* get offset of page */
1065
    s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1066
                     block_size;
1067

1068
out:
1069
    g_free(dh);
1070
    g_free(kh);
1071
    g_free(s->note_buf);
1072
}
1073

1074
/* write common header, sub header and elf note to vmcore */
1075
static void create_header64(DumpState *s, Error **errp)
1076
{
1077
    ERRP_GUARD();
1078
    DiskDumpHeader64 *dh = NULL;
1079
    KdumpSubHeader64 *kh = NULL;
1080
    size_t size;
1081
    uint32_t block_size;
1082
    uint32_t sub_hdr_size;
1083
    uint32_t bitmap_blocks;
1084
    uint32_t status = 0;
1085
    uint64_t offset_note;
1086

1087
    /* write common header, the version of kdump-compressed format is 6th */
1088
    size = sizeof(DiskDumpHeader64);
1089
    dh = g_malloc0(size);
1090

1091
    memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
1092
    dh->header_version = cpu_to_dump32(s, 6);
1093
    block_size = s->dump_info.page_size;
1094
    dh->block_size = cpu_to_dump32(s, block_size);
1095
    sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
1096
    sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
1097
    dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
1098
    /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
1099
    dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
1100
    dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
1101
    bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
1102
    dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
1103
    strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
1104

1105
    if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
1106
        status |= DUMP_DH_COMPRESSED_ZLIB;
1107
    }
1108
#ifdef CONFIG_LZO
1109
    if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
1110
        status |= DUMP_DH_COMPRESSED_LZO;
1111
    }
1112
#endif
1113
#ifdef CONFIG_SNAPPY
1114
    if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
1115
        status |= DUMP_DH_COMPRESSED_SNAPPY;
1116
    }
1117
#endif
1118
    dh->status = cpu_to_dump32(s, status);
1119

1120
    if (write_buffer(s, 0, dh, size) < 0) {
1121
        error_setg(errp, "dump: failed to write disk dump header");
1122
        goto out;
1123
    }
1124

1125
    /* write sub header */
1126
    size = sizeof(KdumpSubHeader64);
1127
    kh = g_malloc0(size);
1128

1129
    /* 64bit max_mapnr_64 */
1130
    kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
1131
    kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
1132
    kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
1133

1134
    offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
1135
    if (s->guest_note &&
1136
        note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1137
        uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
1138

1139
        get_note_sizes(s, s->guest_note,
1140
                       &hsize, &name_size, &size_vmcoreinfo_desc);
1141
        offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
1142
            (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
1143
        kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
1144
        kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
1145
    }
1146

1147
    kh->offset_note = cpu_to_dump64(s, offset_note);
1148
    kh->note_size = cpu_to_dump64(s, s->note_size);
1149

1150
    if (write_buffer(s, DISKDUMP_HEADER_BLOCKS *
1151
                     block_size, kh, size) < 0) {
1152
        error_setg(errp, "dump: failed to write kdump sub header");
1153
        goto out;
1154
    }
1155

1156
    /* write note */
1157
    s->note_buf = g_malloc0(s->note_size);
1158
    s->note_buf_offset = 0;
1159

1160
    /* use s->note_buf to store notes temporarily */
1161
    write_elf64_notes(buf_write_note, s, errp);
1162
    if (*errp) {
1163
        goto out;
1164
    }
1165

1166
    if (write_buffer(s, offset_note, s->note_buf,
1167
                     s->note_size) < 0) {
1168
        error_setg(errp, "dump: failed to write notes");
1169
        goto out;
1170
    }
1171

1172
    /* get offset of dump_bitmap */
1173
    s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1174
                             block_size;
1175

1176
    /* get offset of page */
1177
    s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1178
                     block_size;
1179

1180
out:
1181
    g_free(dh);
1182
    g_free(kh);
1183
    g_free(s->note_buf);
1184
}
1185

1186
static void write_dump_header(DumpState *s, Error **errp)
1187
{
1188
    if (dump_is_64bit(s)) {
1189
        create_header64(s, errp);
1190
    } else {
1191
        create_header32(s, errp);
1192
    }
1193
}
1194

1195
static size_t dump_bitmap_get_bufsize(DumpState *s)
1196
{
1197
    return s->dump_info.page_size;
1198
}
1199

1200
/*
1201
 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1202
 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1203
 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1204
 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1205
 * vmcore, ie. synchronizing un-sync bit into vmcore.
1206
 */
1207
static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1208
                           uint8_t *buf, DumpState *s)
1209
{
1210
    off_t old_offset, new_offset;
1211
    off_t offset_bitmap1, offset_bitmap2;
1212
    uint32_t byte, bit;
1213
    size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1214
    size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1215

1216
    /* should not set the previous place */
1217
    assert(last_pfn <= pfn);
1218

1219
    /*
1220
     * if the bit needed to be set is not cached in buf, flush the data in buf
1221
     * to vmcore firstly.
1222
     * making new_offset be bigger than old_offset can also sync remained data
1223
     * into vmcore.
1224
     */
1225
    old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1226
    new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1227

1228
    while (old_offset < new_offset) {
1229
        /* calculate the offset and write dump_bitmap */
1230
        offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1231
        if (write_buffer(s, offset_bitmap1, buf,
1232
                         bitmap_bufsize) < 0) {
1233
            return -1;
1234
        }
1235

1236
        /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1237
        offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1238
                         old_offset;
1239
        if (write_buffer(s, offset_bitmap2, buf,
1240
                         bitmap_bufsize) < 0) {
1241
            return -1;
1242
        }
1243

1244
        memset(buf, 0, bitmap_bufsize);
1245
        old_offset += bitmap_bufsize;
1246
    }
1247

1248
    /* get the exact place of the bit in the buf, and set it */
1249
    byte = (pfn % bits_per_buf) / CHAR_BIT;
1250
    bit = (pfn % bits_per_buf) % CHAR_BIT;
1251
    if (value) {
1252
        buf[byte] |= 1u << bit;
1253
    } else {
1254
        buf[byte] &= ~(1u << bit);
1255
    }
1256

1257
    return 0;
1258
}
1259

1260
static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1261
{
1262
    int target_page_shift = ctz32(s->dump_info.page_size);
1263

1264
    return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1265
}
1266

1267
static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1268
{
1269
    int target_page_shift = ctz32(s->dump_info.page_size);
1270

1271
    return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1272
}
1273

1274
/*
1275
 * Return the page frame number and the page content in *bufptr. bufptr can be
1276
 * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated
1277
 * memory. This is not necessarily the memory returned.
1278
 */
1279
static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1280
                          uint8_t **bufptr, DumpState *s)
1281
{
1282
    GuestPhysBlock *block = *blockptr;
1283
    uint32_t page_size = s->dump_info.page_size;
1284
    uint8_t *buf = NULL, *hbuf;
1285
    hwaddr addr;
1286

1287
    /* block == NULL means the start of the iteration */
1288
    if (!block) {
1289
        block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1290
        *blockptr = block;
1291
        addr = block->target_start;
1292
        *pfnptr = dump_paddr_to_pfn(s, addr);
1293
    } else {
1294
        *pfnptr += 1;
1295
        addr = dump_pfn_to_paddr(s, *pfnptr);
1296
    }
1297
    assert(block != NULL);
1298

1299
    while (1) {
1300
        if (addr >= block->target_start && addr < block->target_end) {
1301
            size_t n = MIN(block->target_end - addr, page_size - addr % page_size);
1302
            hbuf = block->host_addr + (addr - block->target_start);
1303
            if (!buf) {
1304
                if (n == page_size) {
1305
                    /* this is a whole target page, go for it */
1306
                    assert(addr % page_size == 0);
1307
                    buf = hbuf;
1308
                    break;
1309
                } else if (bufptr) {
1310
                    assert(*bufptr);
1311
                    buf = *bufptr;
1312
                    memset(buf, 0, page_size);
1313
                } else {
1314
                    return true;
1315
                }
1316
            }
1317

1318
            memcpy(buf + addr % page_size, hbuf, n);
1319
            addr += n;
1320
            if (addr % page_size == 0 || addr >= block->target_end) {
1321
                /* we filled up the page or the current block is finished */
1322
                break;
1323
            }
1324
        } else {
1325
            /* the next page is in the next block */
1326
            *blockptr = block = QTAILQ_NEXT(block, next);
1327
            if (!block) {
1328
                break;
1329
            }
1330

1331
            addr = block->target_start;
1332
            /* are we still in the same page? */
1333
            if (dump_paddr_to_pfn(s, addr) != *pfnptr) {
1334
                if (buf) {
1335
                    /* no, but we already filled something earlier, return it */
1336
                    break;
1337
                } else {
1338
                    /* else continue from there */
1339
                    *pfnptr = dump_paddr_to_pfn(s, addr);
1340
                }
1341
            }
1342
        }
1343
    }
1344

1345
    if (bufptr) {
1346
        *bufptr = buf;
1347
    }
1348

1349
    return buf != NULL;
1350
}
1351

1352
static void write_dump_bitmap(DumpState *s, Error **errp)
1353
{
1354
    int ret = 0;
1355
    uint64_t last_pfn, pfn;
1356
    void *dump_bitmap_buf;
1357
    size_t num_dumpable;
1358
    GuestPhysBlock *block_iter = NULL;
1359
    size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1360
    size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1361

1362
    /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1363
    dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1364

1365
    num_dumpable = 0;
1366
    last_pfn = 0;
1367

1368
    /*
1369
     * exam memory page by page, and set the bit in dump_bitmap corresponded
1370
     * to the existing page.
1371
     */
1372
    while (get_next_page(&block_iter, &pfn, NULL, s)) {
1373
        ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1374
        if (ret < 0) {
1375
            error_setg(errp, "dump: failed to set dump_bitmap");
1376
            goto out;
1377
        }
1378

1379
        last_pfn = pfn;
1380
        num_dumpable++;
1381
    }
1382

1383
    /*
1384
     * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1385
     * set the remaining bits from last_pfn to the end of the bitmap buffer to
1386
     * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1387
     */
1388
    if (num_dumpable > 0) {
1389
        ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1390
                              dump_bitmap_buf, s);
1391
        if (ret < 0) {
1392
            error_setg(errp, "dump: failed to sync dump_bitmap");
1393
            goto out;
1394
        }
1395
    }
1396

1397
    /* number of dumpable pages that will be dumped later */
1398
    s->num_dumpable = num_dumpable;
1399

1400
out:
1401
    g_free(dump_bitmap_buf);
1402
}
1403

1404
static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1405
                               off_t offset)
1406
{
1407
    data_cache->state = s;
1408
    data_cache->data_size = 0;
1409
    data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1410
    data_cache->buf = g_malloc0(data_cache->buf_size);
1411
    data_cache->offset = offset;
1412
}
1413

1414
static int write_cache(DataCache *dc, const void *buf, size_t size,
1415
                       bool flag_sync)
1416
{
1417
    /*
1418
     * dc->buf_size should not be less than size, otherwise dc will never be
1419
     * enough
1420
     */
1421
    assert(size <= dc->buf_size);
1422

1423
    /*
1424
     * if flag_sync is set, synchronize data in dc->buf into vmcore.
1425
     * otherwise check if the space is enough for caching data in buf, if not,
1426
     * write the data in dc->buf to dc->state->fd and reset dc->buf
1427
     */
1428
    if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1429
        (flag_sync && dc->data_size > 0)) {
1430
        if (write_buffer(dc->state, dc->offset, dc->buf, dc->data_size) < 0) {
1431
            return -1;
1432
        }
1433

1434
        dc->offset += dc->data_size;
1435
        dc->data_size = 0;
1436
    }
1437

1438
    if (!flag_sync) {
1439
        memcpy(dc->buf + dc->data_size, buf, size);
1440
        dc->data_size += size;
1441
    }
1442

1443
    return 0;
1444
}
1445

1446
static void free_data_cache(DataCache *data_cache)
1447
{
1448
    g_free(data_cache->buf);
1449
}
1450

1451
static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1452
{
1453
    switch (flag_compress) {
1454
    case DUMP_DH_COMPRESSED_ZLIB:
1455
        return compressBound(page_size);
1456

1457
    case DUMP_DH_COMPRESSED_LZO:
1458
        /*
1459
         * LZO will expand incompressible data by a little amount. Please check
1460
         * the following URL to see the expansion calculation:
1461
         * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1462
         */
1463
        return page_size + page_size / 16 + 64 + 3;
1464

1465
#ifdef CONFIG_SNAPPY
1466
    case DUMP_DH_COMPRESSED_SNAPPY:
1467
        return snappy_max_compressed_length(page_size);
1468
#endif
1469
    }
1470
    return 0;
1471
}
1472

1473
static void write_dump_pages(DumpState *s, Error **errp)
1474
{
1475
    int ret = 0;
1476
    DataCache page_desc, page_data;
1477
    size_t len_buf_out, size_out;
1478
#ifdef CONFIG_LZO
1479
    lzo_bytep wrkmem = NULL;
1480
#endif
1481
    uint8_t *buf_out = NULL;
1482
    off_t offset_desc, offset_data;
1483
    PageDescriptor pd, pd_zero;
1484
    uint8_t *buf;
1485
    GuestPhysBlock *block_iter = NULL;
1486
    uint64_t pfn_iter;
1487
    g_autofree uint8_t *page = NULL;
1488

1489
    /* get offset of page_desc and page_data in dump file */
1490
    offset_desc = s->offset_page;
1491
    offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1492

1493
    prepare_data_cache(&page_desc, s, offset_desc);
1494
    prepare_data_cache(&page_data, s, offset_data);
1495

1496
    /* prepare buffer to store compressed data */
1497
    len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1498
    assert(len_buf_out != 0);
1499

1500
#ifdef CONFIG_LZO
1501
    wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1502
#endif
1503

1504
    buf_out = g_malloc(len_buf_out);
1505

1506
    /*
1507
     * init zero page's page_desc and page_data, because every zero page
1508
     * uses the same page_data
1509
     */
1510
    pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1511
    pd_zero.flags = cpu_to_dump32(s, 0);
1512
    pd_zero.offset = cpu_to_dump64(s, offset_data);
1513
    pd_zero.page_flags = cpu_to_dump64(s, 0);
1514
    buf = g_malloc0(s->dump_info.page_size);
1515
    ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1516
    g_free(buf);
1517
    if (ret < 0) {
1518
        error_setg(errp, "dump: failed to write page data (zero page)");
1519
        goto out;
1520
    }
1521

1522
    offset_data += s->dump_info.page_size;
1523
    page = g_malloc(s->dump_info.page_size);
1524

1525
    /*
1526
     * dump memory to vmcore page by page. zero page will all be resided in the
1527
     * first page of page section
1528
     */
1529
    for (buf = page; get_next_page(&block_iter, &pfn_iter, &buf, s); buf = page) {
1530
        /* check zero page */
1531
        if (buffer_is_zero(buf, s->dump_info.page_size)) {
1532
            ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1533
                              false);
1534
            if (ret < 0) {
1535
                error_setg(errp, "dump: failed to write page desc");
1536
                goto out;
1537
            }
1538
        } else {
1539
            /*
1540
             * not zero page, then:
1541
             * 1. compress the page
1542
             * 2. write the compressed page into the cache of page_data
1543
             * 3. get page desc of the compressed page and write it into the
1544
             *    cache of page_desc
1545
             *
1546
             * only one compression format will be used here, for
1547
             * s->flag_compress is set. But when compression fails to work,
1548
             * we fall back to save in plaintext.
1549
             */
1550
             size_out = len_buf_out;
1551
             if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1552
                    (compress2(buf_out, (uLongf *)&size_out, buf,
1553
                               s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1554
                    (size_out < s->dump_info.page_size)) {
1555
                pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1556
                pd.size  = cpu_to_dump32(s, size_out);
1557

1558
                ret = write_cache(&page_data, buf_out, size_out, false);
1559
                if (ret < 0) {
1560
                    error_setg(errp, "dump: failed to write page data");
1561
                    goto out;
1562
                }
1563
#ifdef CONFIG_LZO
1564
            } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1565
                    (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1566
                    (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1567
                    (size_out < s->dump_info.page_size)) {
1568
                pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1569
                pd.size  = cpu_to_dump32(s, size_out);
1570

1571
                ret = write_cache(&page_data, buf_out, size_out, false);
1572
                if (ret < 0) {
1573
                    error_setg(errp, "dump: failed to write page data");
1574
                    goto out;
1575
                }
1576
#endif
1577
#ifdef CONFIG_SNAPPY
1578
            } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1579
                    (snappy_compress((char *)buf, s->dump_info.page_size,
1580
                    (char *)buf_out, &size_out) == SNAPPY_OK) &&
1581
                    (size_out < s->dump_info.page_size)) {
1582
                pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1583
                pd.size  = cpu_to_dump32(s, size_out);
1584

1585
                ret = write_cache(&page_data, buf_out, size_out, false);
1586
                if (ret < 0) {
1587
                    error_setg(errp, "dump: failed to write page data");
1588
                    goto out;
1589
                }
1590
#endif
1591
            } else {
1592
                /*
1593
                 * fall back to save in plaintext, size_out should be
1594
                 * assigned the target's page size
1595
                 */
1596
                pd.flags = cpu_to_dump32(s, 0);
1597
                size_out = s->dump_info.page_size;
1598
                pd.size = cpu_to_dump32(s, size_out);
1599

1600
                ret = write_cache(&page_data, buf,
1601
                                  s->dump_info.page_size, false);
1602
                if (ret < 0) {
1603
                    error_setg(errp, "dump: failed to write page data");
1604
                    goto out;
1605
                }
1606
            }
1607

1608
            /* get and write page desc here */
1609
            pd.page_flags = cpu_to_dump64(s, 0);
1610
            pd.offset = cpu_to_dump64(s, offset_data);
1611
            offset_data += size_out;
1612

1613
            ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1614
            if (ret < 0) {
1615
                error_setg(errp, "dump: failed to write page desc");
1616
                goto out;
1617
            }
1618
        }
1619
        s->written_size += s->dump_info.page_size;
1620
    }
1621

1622
    ret = write_cache(&page_desc, NULL, 0, true);
1623
    if (ret < 0) {
1624
        error_setg(errp, "dump: failed to sync cache for page_desc");
1625
        goto out;
1626
    }
1627
    ret = write_cache(&page_data, NULL, 0, true);
1628
    if (ret < 0) {
1629
        error_setg(errp, "dump: failed to sync cache for page_data");
1630
        goto out;
1631
    }
1632

1633
out:
1634
    free_data_cache(&page_desc);
1635
    free_data_cache(&page_data);
1636

1637
#ifdef CONFIG_LZO
1638
    g_free(wrkmem);
1639
#endif
1640

1641
    g_free(buf_out);
1642
}
1643

1644
static void create_kdump_vmcore(DumpState *s, Error **errp)
1645
{
1646
    ERRP_GUARD();
1647
    int ret;
1648

1649
    /*
1650
     * the kdump-compressed format is:
1651
     *                                               File offset
1652
     *  +------------------------------------------+ 0x0
1653
     *  |    main header (struct disk_dump_header) |
1654
     *  |------------------------------------------+ block 1
1655
     *  |    sub header (struct kdump_sub_header)  |
1656
     *  |------------------------------------------+ block 2
1657
     *  |            1st-dump_bitmap               |
1658
     *  |------------------------------------------+ block 2 + X blocks
1659
     *  |            2nd-dump_bitmap               | (aligned by block)
1660
     *  |------------------------------------------+ block 2 + 2 * X blocks
1661
     *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1662
     *  |  page desc for pfn 1 (struct page_desc)  |
1663
     *  |                    :                     |
1664
     *  |------------------------------------------| (not aligned by block)
1665
     *  |         page data (pfn 0)                |
1666
     *  |         page data (pfn 1)                |
1667
     *  |                    :                     |
1668
     *  +------------------------------------------+
1669
     */
1670

1671
    ret = write_start_flat_header(s);
1672
    if (ret < 0) {
1673
        error_setg(errp, "dump: failed to write start flat header");
1674
        return;
1675
    }
1676

1677
    write_dump_header(s, errp);
1678
    if (*errp) {
1679
        return;
1680
    }
1681

1682
    write_dump_bitmap(s, errp);
1683
    if (*errp) {
1684
        return;
1685
    }
1686

1687
    write_dump_pages(s, errp);
1688
    if (*errp) {
1689
        return;
1690
    }
1691

1692
    ret = write_end_flat_header(s);
1693
    if (ret < 0) {
1694
        error_setg(errp, "dump: failed to write end flat header");
1695
        return;
1696
    }
1697
}
1698

1699
static void get_max_mapnr(DumpState *s)
1700
{
1701
    GuestPhysBlock *last_block;
1702

1703
    last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1704
    s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1705
}
1706

1707
static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1708

1709
static void dump_state_prepare(DumpState *s)
1710
{
1711
    /* zero the struct, setting status to active */
1712
    *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1713
}
1714

1715
bool qemu_system_dump_in_progress(void)
1716
{
1717
    DumpState *state = &dump_state_global;
1718
    return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1719
}
1720

1721
/*
1722
 * calculate total size of memory to be dumped (taking filter into
1723
 * account.)
1724
 */
1725
static int64_t dump_calculate_size(DumpState *s)
1726
{
1727
    GuestPhysBlock *block;
1728
    int64_t total = 0;
1729

1730
    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1731
        total += dump_filtered_memblock_size(block,
1732
                                             s->filter_area_begin,
1733
                                             s->filter_area_length);
1734
    }
1735

1736
    return total;
1737
}
1738

1739
static void vmcoreinfo_update_phys_base(DumpState *s)
1740
{
1741
    uint64_t size, note_head_size, name_size, phys_base;
1742
    char **lines;
1743
    uint8_t *vmci;
1744
    size_t i;
1745

1746
    if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1747
        return;
1748
    }
1749

1750
    get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1751
    note_head_size = ROUND_UP(note_head_size, 4);
1752

1753
    vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1754
    *(vmci + size) = '\0';
1755

1756
    lines = g_strsplit((char *)vmci, "\n", -1);
1757
    for (i = 0; lines[i]; i++) {
1758
        const char *prefix = NULL;
1759

1760
        if (s->dump_info.d_machine == EM_X86_64) {
1761
            prefix = "NUMBER(phys_base)=";
1762
        } else if (s->dump_info.d_machine == EM_AARCH64) {
1763
            prefix = "NUMBER(PHYS_OFFSET)=";
1764
        }
1765

1766
        if (prefix && g_str_has_prefix(lines[i], prefix)) {
1767
            if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1768
                              &phys_base) < 0) {
1769
                warn_report("Failed to read %s", prefix);
1770
            } else {
1771
                s->dump_info.phys_base = phys_base;
1772
            }
1773
            break;
1774
        }
1775
    }
1776

1777
    g_strfreev(lines);
1778
}
1779

1780
static void dump_init(DumpState *s, int fd, bool has_format,
1781
                      DumpGuestMemoryFormat format, bool paging, bool has_filter,
1782
                      int64_t begin, int64_t length, bool kdump_raw,
1783
                      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
    s->kdump_raw = kdump_raw;
1795

1796
    /* kdump-compressed is conflict with paging and filter */
1797
    if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1798
        assert(!paging && !has_filter);
1799
    }
1800

1801
    if (runstate_is_running()) {
1802
        vm_stop(RUN_STATE_SAVE_VM);
1803
        s->resume = true;
1804
    } else {
1805
        s->resume = false;
1806
    }
1807

1808
    /* If we use KVM, we should synchronize the registers before we get dump
1809
     * info or physmap info.
1810
     */
1811
    cpu_synchronize_all_states();
1812
    nr_cpus = 0;
1813
    CPU_FOREACH(cpu) {
1814
        nr_cpus++;
1815
    }
1816

1817
    s->fd = fd;
1818
    if (has_filter && !length) {
1819
        error_setg(errp, "parameter 'length' expects a non-zero size");
1820
        goto cleanup;
1821
    }
1822
    s->filter_area_begin = begin;
1823
    s->filter_area_length = length;
1824

1825
    /* First index is 0, it's the special null name */
1826
    s->string_table_buf = g_array_new(FALSE, TRUE, 1);
1827
    /*
1828
     * Allocate the null name, due to the clearing option set to true
1829
     * it will be 0.
1830
     */
1831
    g_array_set_size(s->string_table_buf, 1);
1832

1833
    memory_mapping_list_init(&s->list);
1834

1835
    guest_phys_blocks_init(&s->guest_phys_blocks);
1836
    guest_phys_blocks_append(&s->guest_phys_blocks);
1837
    s->total_size = dump_calculate_size(s);
1838
#ifdef DEBUG_DUMP_GUEST_MEMORY
1839
    fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1840
#endif
1841

1842
    /* it does not make sense to dump non-existent memory */
1843
    if (!s->total_size) {
1844
        error_setg(errp, "dump: no guest memory to dump");
1845
        goto cleanup;
1846
    }
1847

1848
    /* get dump info: endian, class and architecture.
1849
     * If the target architecture is not supported, cpu_get_dump_info() will
1850
     * return -1.
1851
     */
1852
    ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1853
    if (ret < 0) {
1854
        error_setg(errp,
1855
                   "dumping guest memory is not supported on this target");
1856
        goto cleanup;
1857
    }
1858

1859
    if (!s->dump_info.page_size) {
1860
        s->dump_info.page_size = qemu_target_page_size();
1861
    }
1862

1863
    s->note_size = cpu_get_note_size(s->dump_info.d_class,
1864
                                     s->dump_info.d_machine, nr_cpus);
1865
    assert(s->note_size >= 0);
1866

1867
    /*
1868
     * The goal of this block is to (a) update the previously guessed
1869
     * phys_base, (b) copy the guest note out of the guest.
1870
     * Failure to do so is not fatal for dumping.
1871
     */
1872
    if (vmci) {
1873
        uint64_t addr, note_head_size, name_size, desc_size;
1874
        uint32_t size;
1875
        uint16_t guest_format;
1876

1877
        note_head_size = dump_is_64bit(s) ?
1878
            sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr);
1879

1880
        guest_format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1881
        size = le32_to_cpu(vmci->vmcoreinfo.size);
1882
        addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1883
        if (!vmci->has_vmcoreinfo) {
1884
            warn_report("guest note is not present");
1885
        } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1886
            warn_report("guest note size is invalid: %" PRIu32, size);
1887
        } else if (guest_format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1888
            warn_report("guest note format is unsupported: %" PRIu16, guest_format);
1889
        } else {
1890
            s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1891
            cpu_physical_memory_read(addr, s->guest_note, size);
1892

1893
            get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1894
            s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1895
                                               desc_size);
1896
            if (name_size > MAX_GUEST_NOTE_SIZE ||
1897
                desc_size > MAX_GUEST_NOTE_SIZE ||
1898
                s->guest_note_size > size) {
1899
                warn_report("Invalid guest note header");
1900
                g_free(s->guest_note);
1901
                s->guest_note = NULL;
1902
            } else {
1903
                vmcoreinfo_update_phys_base(s);
1904
                s->note_size += s->guest_note_size;
1905
            }
1906
        }
1907
    }
1908

1909
    /* get memory mapping */
1910
    if (paging) {
1911
        qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp);
1912
        if (*errp) {
1913
            goto cleanup;
1914
        }
1915
    } else {
1916
        qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1917
    }
1918

1919
    s->nr_cpus = nr_cpus;
1920

1921
    get_max_mapnr(s);
1922

1923
    uint64_t tmp;
1924
    tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1925
                       s->dump_info.page_size);
1926
    s->len_dump_bitmap = tmp * s->dump_info.page_size;
1927

1928
    /* init for kdump-compressed format */
1929
    if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1930
        switch (format) {
1931
        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1932
            s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1933
            break;
1934

1935
        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1936
#ifdef CONFIG_LZO
1937
            if (lzo_init() != LZO_E_OK) {
1938
                error_setg(errp, "failed to initialize the LZO library");
1939
                goto cleanup;
1940
            }
1941
#endif
1942
            s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1943
            break;
1944

1945
        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1946
            s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1947
            break;
1948

1949
        default:
1950
            s->flag_compress = 0;
1951
        }
1952

1953
        return;
1954
    }
1955

1956
    if (dump_has_filter(s)) {
1957
        memory_mapping_filter(&s->list, s->filter_area_begin, s->filter_area_length);
1958
    }
1959

1960
    /*
1961
     * The first section header is always a special one in which most
1962
     * fields are 0. The section header string table is also always
1963
     * set.
1964
     */
1965
    s->shdr_num = 2;
1966

1967
    /*
1968
     * Adds the number of architecture sections to shdr_num and sets
1969
     * elf_section_data_size so we know the offsets and sizes of all
1970
     * parts.
1971
     */
1972
    if (s->dump_info.arch_sections_add_fn) {
1973
        s->dump_info.arch_sections_add_fn(s);
1974
    }
1975

1976
    /*
1977
     * calculate shdr_num so we know the offsets and sizes of all
1978
     * parts.
1979
     * Calculate phdr_num
1980
     *
1981
     * The absolute maximum amount of phdrs is UINT32_MAX - 1 as
1982
     * sh_info is 32 bit. There's special handling once we go over
1983
     * UINT16_MAX - 1 but that is handled in the ehdr and section
1984
     * code.
1985
     */
1986
    s->phdr_num = 1; /* Reserve PT_NOTE */
1987
    if (s->list.num <= UINT32_MAX - 1) {
1988
        s->phdr_num += s->list.num;
1989
    } else {
1990
        s->phdr_num = UINT32_MAX;
1991
    }
1992

1993
    /*
1994
     * Now that the number of section and program headers is known we
1995
     * can calculate the offsets of the headers and data.
1996
     */
1997
    if (dump_is_64bit(s)) {
1998
        s->shdr_offset = sizeof(Elf64_Ehdr);
1999
        s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num;
2000
        s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num;
2001
    } else {
2002
        s->shdr_offset = sizeof(Elf32_Ehdr);
2003
        s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num;
2004
        s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num;
2005
    }
2006
    s->memory_offset = s->note_offset + s->note_size;
2007
    s->section_offset = s->memory_offset + s->total_size;
2008

2009
    return;
2010

2011
cleanup:
2012
    dump_cleanup(s);
2013
}
2014

2015
/* this operation might be time consuming. */
2016
static void dump_process(DumpState *s, Error **errp)
2017
{
2018
    ERRP_GUARD();
2019
    DumpQueryResult *result = NULL;
2020

2021
    if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
2022
        create_win_dump(s, errp);
2023
    } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
2024
        create_kdump_vmcore(s, errp);
2025
    } else {
2026
        create_vmcore(s, errp);
2027
    }
2028

2029
    /* make sure status is written after written_size updates */
2030
    smp_wmb();
2031
    qatomic_set(&s->status,
2032
               (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
2033

2034
    /* send DUMP_COMPLETED message (unconditionally) */
2035
    result = qmp_query_dump(NULL);
2036
    /* should never fail */
2037
    assert(result);
2038
    qapi_event_send_dump_completed(result,
2039
                                   *errp ? error_get_pretty(*errp) : NULL);
2040
    qapi_free_DumpQueryResult(result);
2041

2042
    dump_cleanup(s);
2043
}
2044

2045
static void *dump_thread(void *data)
2046
{
2047
    DumpState *s = (DumpState *)data;
2048
    dump_process(s, NULL);
2049
    return NULL;
2050
}
2051

2052
DumpQueryResult *qmp_query_dump(Error **errp)
2053
{
2054
    DumpQueryResult *result = g_new(DumpQueryResult, 1);
2055
    DumpState *state = &dump_state_global;
2056
    result->status = qatomic_read(&state->status);
2057
    /* make sure we are reading status and written_size in order */
2058
    smp_rmb();
2059
    result->completed = state->written_size;
2060
    result->total = state->total_size;
2061
    return result;
2062
}
2063

2064
void qmp_dump_guest_memory(bool paging, const char *protocol,
2065
                           bool has_detach, bool detach,
2066
                           bool has_begin, int64_t begin,
2067
                           bool has_length, int64_t length,
2068
                           bool has_format, DumpGuestMemoryFormat format,
2069
                           Error **errp)
2070
{
2071
    ERRP_GUARD();
2072
    const char *p;
2073
    int fd;
2074
    DumpState *s;
2075
    bool detach_p = false;
2076
    bool kdump_raw = false;
2077

2078
    if (runstate_check(RUN_STATE_INMIGRATE)) {
2079
        error_setg(errp, "Dump not allowed during incoming migration.");
2080
        return;
2081
    }
2082

2083
    /* if there is a dump in background, we should wait until the dump
2084
     * finished */
2085
    if (qemu_system_dump_in_progress()) {
2086
        error_setg(errp, "There is a dump in process, please wait.");
2087
        return;
2088
    }
2089

2090
    /*
2091
     * externally, we represent kdump-raw-* as separate formats, but internally
2092
     * they are handled the same, except for the "raw" flag
2093
     */
2094
    if (has_format) {
2095
        switch (format) {
2096
        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_ZLIB:
2097
            format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
2098
            kdump_raw = true;
2099
            break;
2100
        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_LZO:
2101
            format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
2102
            kdump_raw = true;
2103
            break;
2104
        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_SNAPPY:
2105
            format = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
2106
            kdump_raw = true;
2107
            break;
2108
        default:
2109
            break;
2110
        }
2111
    }
2112

2113
    /*
2114
     * kdump-compressed format need the whole memory dumped, so paging or
2115
     * filter is not supported here.
2116
     */
2117
    if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
2118
        (paging || has_begin || has_length)) {
2119
        error_setg(errp, "kdump-compressed format doesn't support paging or "
2120
                         "filter");
2121
        return;
2122
    }
2123
    if (has_begin && !has_length) {
2124
        error_setg(errp, QERR_MISSING_PARAMETER, "length");
2125
        return;
2126
    }
2127
    if (!has_begin && has_length) {
2128
        error_setg(errp, QERR_MISSING_PARAMETER, "begin");
2129
        return;
2130
    }
2131
    if (has_detach) {
2132
        detach_p = detach;
2133
    }
2134

2135
    /* check whether lzo/snappy is supported */
2136
#ifndef CONFIG_LZO
2137
    if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
2138
        error_setg(errp, "kdump-lzo is not available now");
2139
        return;
2140
    }
2141
#endif
2142

2143
#ifndef CONFIG_SNAPPY
2144
    if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
2145
        error_setg(errp, "kdump-snappy is not available now");
2146
        return;
2147
    }
2148
#endif
2149

2150
    if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
2151
        && !win_dump_available(errp)) {
2152
        return;
2153
    }
2154

2155
    if (strstart(protocol, "fd:", &p)) {
2156
        fd = monitor_get_fd(monitor_cur(), p, errp);
2157
        if (fd == -1) {
2158
            return;
2159
        }
2160
    } else if  (strstart(protocol, "file:", &p)) {
2161
        fd = qemu_create(p, O_WRONLY | O_TRUNC | O_BINARY, S_IRUSR, errp);
2162
        if (fd < 0) {
2163
            return;
2164
        }
2165
    } else {
2166
        error_setg(errp,
2167
                   "parameter 'protocol' must start with 'file:' or 'fd:'");
2168
        return;
2169
    }
2170
    if (kdump_raw && lseek(fd, 0, SEEK_CUR) == (off_t) -1) {
2171
        close(fd);
2172
        error_setg(errp, "kdump-raw formats require a seekable file");
2173
        return;
2174
    }
2175

2176
    if (!dump_migration_blocker) {
2177
        error_setg(&dump_migration_blocker,
2178
                   "Live migration disabled: dump-guest-memory in progress");
2179
    }
2180

2181
    /*
2182
     * Allows even for -only-migratable, but forbid migration during the
2183
     * process of dump guest memory.
2184
     */
2185
    if (migrate_add_blocker_internal(&dump_migration_blocker, errp)) {
2186
        /* Remember to release the fd before passing it over to dump state */
2187
        close(fd);
2188
        return;
2189
    }
2190

2191
    s = &dump_state_global;
2192
    dump_state_prepare(s);
2193

2194
    dump_init(s, fd, has_format, format, paging, has_begin,
2195
              begin, length, kdump_raw, errp);
2196
    if (*errp) {
2197
        qatomic_set(&s->status, DUMP_STATUS_FAILED);
2198
        return;
2199
    }
2200

2201
    if (detach_p) {
2202
        /* detached dump */
2203
        s->detached = true;
2204
        qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2205
                           s, QEMU_THREAD_DETACHED);
2206
    } else {
2207
        /* sync dump */
2208
        dump_process(s, errp);
2209
    }
2210
}
2211

2212
DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2213
{
2214
    DumpGuestMemoryCapability *cap =
2215
                                  g_new0(DumpGuestMemoryCapability, 1);
2216
    DumpGuestMemoryFormatList **tail = &cap->formats;
2217

2218
    /* elf is always available */
2219
    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2220

2221
    /* kdump-zlib is always available */
2222
    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2223
    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_ZLIB);
2224

2225
    /* add new item if kdump-lzo is available */
2226
#ifdef CONFIG_LZO
2227
    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2228
    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_LZO);
2229
#endif
2230

2231
    /* add new item if kdump-snappy is available */
2232
#ifdef CONFIG_SNAPPY
2233
    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2234
    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_RAW_SNAPPY);
2235
#endif
2236

2237
    if (win_dump_available(NULL)) {
2238
        QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2239
    }
2240

2241
    return cap;
2242
}
2243