qemu/dump/dump.c
<<
>>
Prefs
   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-common.h"
  16#include "qemu/cutils.h"
  17#include "elf.h"
  18#include "exec/hwaddr.h"
  19#include "monitor/monitor.h"
  20#include "sysemu/kvm.h"
  21#include "sysemu/dump.h"
  22#include "sysemu/memory_mapping.h"
  23#include "sysemu/runstate.h"
  24#include "sysemu/cpus.h"
  25#include "qapi/error.h"
  26#include "qapi/qapi-commands-dump.h"
  27#include "qapi/qapi-events-dump.h"
  28#include "qapi/qmp/qerror.h"
  29#include "qemu/error-report.h"
  30#include "qemu/main-loop.h"
  31#include "hw/misc/vmcoreinfo.h"
  32
  33#ifdef TARGET_X86_64
  34#include "win_dump.h"
  35#endif
  36
  37#include <zlib.h>
  38#ifdef CONFIG_LZO
  39#include <lzo/lzo1x.h>
  40#endif
  41#ifdef CONFIG_SNAPPY
  42#include <snappy-c.h>
  43#endif
  44#ifndef ELF_MACHINE_UNAME
  45#define ELF_MACHINE_UNAME "Unknown"
  46#endif
  47
  48#define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
  49
  50#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
  55uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
  56{
  57    if (s->dump_info.d_endian == ELFDATA2LSB) {
  58        val = cpu_to_le16(val);
  59    } else {
  60        val = cpu_to_be16(val);
  61    }
  62
  63    return val;
  64}
  65
  66uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
  67{
  68    if (s->dump_info.d_endian == ELFDATA2LSB) {
  69        val = cpu_to_le32(val);
  70    } else {
  71        val = cpu_to_be32(val);
  72    }
  73
  74    return val;
  75}
  76
  77uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
  78{
  79    if (s->dump_info.d_endian == ELFDATA2LSB) {
  80        val = cpu_to_le64(val);
  81    } else {
  82        val = cpu_to_be64(val);
  83    }
  84
  85    return val;
  86}
  87
  88static int dump_cleanup(DumpState *s)
  89{
  90    guest_phys_blocks_free(&s->guest_phys_blocks);
  91    memory_mapping_list_free(&s->list);
  92    close(s->fd);
  93    g_free(s->guest_note);
  94    s->guest_note = NULL;
  95    if (s->resume) {
  96        if (s->detached) {
  97            qemu_mutex_lock_iothread();
  98        }
  99        vm_start();
 100        if (s->detached) {
 101            qemu_mutex_unlock_iothread();
 102        }
 103    }
 104
 105    return 0;
 106}
 107
 108static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
 109{
 110    DumpState *s = opaque;
 111    size_t written_size;
 112
 113    written_size = qemu_write_full(s->fd, buf, size);
 114    if (written_size != size) {
 115        return -errno;
 116    }
 117
 118    return 0;
 119}
 120
 121static void write_elf64_header(DumpState *s, Error **errp)
 122{
 123    Elf64_Ehdr elf_header;
 124    int ret;
 125
 126    memset(&elf_header, 0, sizeof(Elf64_Ehdr));
 127    memcpy(&elf_header, ELFMAG, SELFMAG);
 128    elf_header.e_ident[EI_CLASS] = ELFCLASS64;
 129    elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
 130    elf_header.e_ident[EI_VERSION] = EV_CURRENT;
 131    elf_header.e_type = cpu_to_dump16(s, ET_CORE);
 132    elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
 133    elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
 134    elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
 135    elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
 136    elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
 137    elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
 138    if (s->have_section) {
 139        uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
 140
 141        elf_header.e_shoff = cpu_to_dump64(s, shoff);
 142        elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
 143        elf_header.e_shnum = cpu_to_dump16(s, 1);
 144    }
 145
 146    ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
 147    if (ret < 0) {
 148        error_setg_errno(errp, -ret, "dump: failed to write elf header");
 149    }
 150}
 151
 152static void write_elf32_header(DumpState *s, Error **errp)
 153{
 154    Elf32_Ehdr elf_header;
 155    int ret;
 156
 157    memset(&elf_header, 0, sizeof(Elf32_Ehdr));
 158    memcpy(&elf_header, ELFMAG, SELFMAG);
 159    elf_header.e_ident[EI_CLASS] = ELFCLASS32;
 160    elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
 161    elf_header.e_ident[EI_VERSION] = EV_CURRENT;
 162    elf_header.e_type = cpu_to_dump16(s, ET_CORE);
 163    elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
 164    elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
 165    elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
 166    elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
 167    elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
 168    elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
 169    if (s->have_section) {
 170        uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
 171
 172        elf_header.e_shoff = cpu_to_dump32(s, shoff);
 173        elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
 174        elf_header.e_shnum = cpu_to_dump16(s, 1);
 175    }
 176
 177    ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
 178    if (ret < 0) {
 179        error_setg_errno(errp, -ret, "dump: failed to write elf header");
 180    }
 181}
 182
 183static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
 184                             int phdr_index, hwaddr offset,
 185                             hwaddr filesz, Error **errp)
 186{
 187    Elf64_Phdr phdr;
 188    int ret;
 189
 190    memset(&phdr, 0, sizeof(Elf64_Phdr));
 191    phdr.p_type = cpu_to_dump32(s, PT_LOAD);
 192    phdr.p_offset = cpu_to_dump64(s, offset);
 193    phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
 194    phdr.p_filesz = cpu_to_dump64(s, filesz);
 195    phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
 196    phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
 197
 198    assert(memory_mapping->length >= filesz);
 199
 200    ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
 201    if (ret < 0) {
 202        error_setg_errno(errp, -ret,
 203                         "dump: failed to write program header table");
 204    }
 205}
 206
 207static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
 208                             int phdr_index, hwaddr offset,
 209                             hwaddr filesz, Error **errp)
 210{
 211    Elf32_Phdr phdr;
 212    int ret;
 213
 214    memset(&phdr, 0, sizeof(Elf32_Phdr));
 215    phdr.p_type = cpu_to_dump32(s, PT_LOAD);
 216    phdr.p_offset = cpu_to_dump32(s, offset);
 217    phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
 218    phdr.p_filesz = cpu_to_dump32(s, filesz);
 219    phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
 220    phdr.p_vaddr =
 221        cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr;
 222
 223    assert(memory_mapping->length >= filesz);
 224
 225    ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
 226    if (ret < 0) {
 227        error_setg_errno(errp, -ret,
 228                         "dump: failed to write program header table");
 229    }
 230}
 231
 232static void write_elf64_note(DumpState *s, Error **errp)
 233{
 234    Elf64_Phdr phdr;
 235    hwaddr begin = s->memory_offset - s->note_size;
 236    int ret;
 237
 238    memset(&phdr, 0, sizeof(Elf64_Phdr));
 239    phdr.p_type = cpu_to_dump32(s, PT_NOTE);
 240    phdr.p_offset = cpu_to_dump64(s, begin);
 241    phdr.p_paddr = 0;
 242    phdr.p_filesz = cpu_to_dump64(s, s->note_size);
 243    phdr.p_memsz = cpu_to_dump64(s, s->note_size);
 244    phdr.p_vaddr = 0;
 245
 246    ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
 247    if (ret < 0) {
 248        error_setg_errno(errp, -ret,
 249                         "dump: failed to write program header table");
 250    }
 251}
 252
 253static inline int cpu_index(CPUState *cpu)
 254{
 255    return cpu->cpu_index + 1;
 256}
 257
 258static void write_guest_note(WriteCoreDumpFunction f, DumpState *s,
 259                             Error **errp)
 260{
 261    int ret;
 262
 263    if (s->guest_note) {
 264        ret = f(s->guest_note, s->guest_note_size, s);
 265        if (ret < 0) {
 266            error_setg(errp, "dump: failed to write guest note");
 267        }
 268    }
 269}
 270
 271static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s,
 272                              Error **errp)
 273{
 274    CPUState *cpu;
 275    int ret;
 276    int id;
 277
 278    CPU_FOREACH(cpu) {
 279        id = cpu_index(cpu);
 280        ret = cpu_write_elf64_note(f, cpu, id, s);
 281        if (ret < 0) {
 282            error_setg(errp, "dump: failed to write elf notes");
 283            return;
 284        }
 285    }
 286
 287    CPU_FOREACH(cpu) {
 288        ret = cpu_write_elf64_qemunote(f, cpu, s);
 289        if (ret < 0) {
 290            error_setg(errp, "dump: failed to write CPU status");
 291            return;
 292        }
 293    }
 294
 295    write_guest_note(f, s, errp);
 296}
 297
 298static void write_elf32_note(DumpState *s, Error **errp)
 299{
 300    hwaddr begin = s->memory_offset - s->note_size;
 301    Elf32_Phdr phdr;
 302    int ret;
 303
 304    memset(&phdr, 0, sizeof(Elf32_Phdr));
 305    phdr.p_type = cpu_to_dump32(s, PT_NOTE);
 306    phdr.p_offset = cpu_to_dump32(s, begin);
 307    phdr.p_paddr = 0;
 308    phdr.p_filesz = cpu_to_dump32(s, s->note_size);
 309    phdr.p_memsz = cpu_to_dump32(s, s->note_size);
 310    phdr.p_vaddr = 0;
 311
 312    ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
 313    if (ret < 0) {
 314        error_setg_errno(errp, -ret,
 315                         "dump: failed to write program header table");
 316    }
 317}
 318
 319static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s,
 320                              Error **errp)
 321{
 322    CPUState *cpu;
 323    int ret;
 324    int id;
 325
 326    CPU_FOREACH(cpu) {
 327        id = cpu_index(cpu);
 328        ret = cpu_write_elf32_note(f, cpu, id, s);
 329        if (ret < 0) {
 330            error_setg(errp, "dump: failed to write elf notes");
 331            return;
 332        }
 333    }
 334
 335    CPU_FOREACH(cpu) {
 336        ret = cpu_write_elf32_qemunote(f, cpu, s);
 337        if (ret < 0) {
 338            error_setg(errp, "dump: failed to write CPU status");
 339            return;
 340        }
 341    }
 342
 343    write_guest_note(f, s, errp);
 344}
 345
 346static void write_elf_section(DumpState *s, int type, Error **errp)
 347{
 348    Elf32_Shdr shdr32;
 349    Elf64_Shdr shdr64;
 350    int shdr_size;
 351    void *shdr;
 352    int ret;
 353
 354    if (type == 0) {
 355        shdr_size = sizeof(Elf32_Shdr);
 356        memset(&shdr32, 0, shdr_size);
 357        shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
 358        shdr = &shdr32;
 359    } else {
 360        shdr_size = sizeof(Elf64_Shdr);
 361        memset(&shdr64, 0, shdr_size);
 362        shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
 363        shdr = &shdr64;
 364    }
 365
 366    ret = fd_write_vmcore(shdr, shdr_size, s);
 367    if (ret < 0) {
 368        error_setg_errno(errp, -ret,
 369                         "dump: failed to write section header table");
 370    }
 371}
 372
 373static void write_data(DumpState *s, void *buf, int length, Error **errp)
 374{
 375    int ret;
 376
 377    ret = fd_write_vmcore(buf, length, s);
 378    if (ret < 0) {
 379        error_setg_errno(errp, -ret, "dump: failed to save memory");
 380    } else {
 381        s->written_size += length;
 382    }
 383}
 384
 385/* write the memory to vmcore. 1 page per I/O. */
 386static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
 387                         int64_t size, Error **errp)
 388{
 389    int64_t i;
 390    Error *local_err = NULL;
 391
 392    for (i = 0; i < size / s->dump_info.page_size; i++) {
 393        write_data(s, block->host_addr + start + i * s->dump_info.page_size,
 394                   s->dump_info.page_size, &local_err);
 395        if (local_err) {
 396            error_propagate(errp, local_err);
 397            return;
 398        }
 399    }
 400
 401    if ((size % s->dump_info.page_size) != 0) {
 402        write_data(s, block->host_addr + start + i * s->dump_info.page_size,
 403                   size % s->dump_info.page_size, &local_err);
 404        if (local_err) {
 405            error_propagate(errp, local_err);
 406            return;
 407        }
 408    }
 409}
 410
 411/* get the memory's offset and size in the vmcore */
 412static void get_offset_range(hwaddr phys_addr,
 413                             ram_addr_t mapping_length,
 414                             DumpState *s,
 415                             hwaddr *p_offset,
 416                             hwaddr *p_filesz)
 417{
 418    GuestPhysBlock *block;
 419    hwaddr offset = s->memory_offset;
 420    int64_t size_in_block, start;
 421
 422    /* When the memory is not stored into vmcore, offset will be -1 */
 423    *p_offset = -1;
 424    *p_filesz = 0;
 425
 426    if (s->has_filter) {
 427        if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
 428            return;
 429        }
 430    }
 431
 432    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
 433        if (s->has_filter) {
 434            if (block->target_start >= s->begin + s->length ||
 435                block->target_end <= s->begin) {
 436                /* This block is out of the range */
 437                continue;
 438            }
 439
 440            if (s->begin <= block->target_start) {
 441                start = block->target_start;
 442            } else {
 443                start = s->begin;
 444            }
 445
 446            size_in_block = block->target_end - start;
 447            if (s->begin + s->length < block->target_end) {
 448                size_in_block -= block->target_end - (s->begin + s->length);
 449            }
 450        } else {
 451            start = block->target_start;
 452            size_in_block = block->target_end - block->target_start;
 453        }
 454
 455        if (phys_addr >= start && phys_addr < start + size_in_block) {
 456            *p_offset = phys_addr - start + offset;
 457
 458            /* The offset range mapped from the vmcore file must not spill over
 459             * the GuestPhysBlock, clamp it. The rest of the mapping will be
 460             * zero-filled in memory at load time; see
 461             * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
 462             */
 463            *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
 464                        mapping_length :
 465                        size_in_block - (phys_addr - start);
 466            return;
 467        }
 468
 469        offset += size_in_block;
 470    }
 471}
 472
 473static void write_elf_loads(DumpState *s, Error **errp)
 474{
 475    hwaddr offset, filesz;
 476    MemoryMapping *memory_mapping;
 477    uint32_t phdr_index = 1;
 478    uint32_t max_index;
 479    Error *local_err = NULL;
 480
 481    if (s->have_section) {
 482        max_index = s->sh_info;
 483    } else {
 484        max_index = s->phdr_num;
 485    }
 486
 487    QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
 488        get_offset_range(memory_mapping->phys_addr,
 489                         memory_mapping->length,
 490                         s, &offset, &filesz);
 491        if (s->dump_info.d_class == ELFCLASS64) {
 492            write_elf64_load(s, memory_mapping, phdr_index++, offset,
 493                             filesz, &local_err);
 494        } else {
 495            write_elf32_load(s, memory_mapping, phdr_index++, offset,
 496                             filesz, &local_err);
 497        }
 498
 499        if (local_err) {
 500            error_propagate(errp, local_err);
 501            return;
 502        }
 503
 504        if (phdr_index >= max_index) {
 505            break;
 506        }
 507    }
 508}
 509
 510/* write elf header, PT_NOTE and elf note to vmcore. */
 511static void dump_begin(DumpState *s, Error **errp)
 512{
 513    Error *local_err = NULL;
 514
 515    /*
 516     * the vmcore's format is:
 517     *   --------------
 518     *   |  elf header |
 519     *   --------------
 520     *   |  PT_NOTE    |
 521     *   --------------
 522     *   |  PT_LOAD    |
 523     *   --------------
 524     *   |  ......     |
 525     *   --------------
 526     *   |  PT_LOAD    |
 527     *   --------------
 528     *   |  sec_hdr    |
 529     *   --------------
 530     *   |  elf note   |
 531     *   --------------
 532     *   |  memory     |
 533     *   --------------
 534     *
 535     * we only know where the memory is saved after we write elf note into
 536     * vmcore.
 537     */
 538
 539    /* write elf header to vmcore */
 540    if (s->dump_info.d_class == ELFCLASS64) {
 541        write_elf64_header(s, &local_err);
 542    } else {
 543        write_elf32_header(s, &local_err);
 544    }
 545    if (local_err) {
 546        error_propagate(errp, local_err);
 547        return;
 548    }
 549
 550    if (s->dump_info.d_class == ELFCLASS64) {
 551        /* write PT_NOTE to vmcore */
 552        write_elf64_note(s, &local_err);
 553        if (local_err) {
 554            error_propagate(errp, local_err);
 555            return;
 556        }
 557
 558        /* write all PT_LOAD to vmcore */
 559        write_elf_loads(s, &local_err);
 560        if (local_err) {
 561            error_propagate(errp, local_err);
 562            return;
 563        }
 564
 565        /* write section to vmcore */
 566        if (s->have_section) {
 567            write_elf_section(s, 1, &local_err);
 568            if (local_err) {
 569                error_propagate(errp, local_err);
 570                return;
 571            }
 572        }
 573
 574        /* write notes to vmcore */
 575        write_elf64_notes(fd_write_vmcore, s, &local_err);
 576        if (local_err) {
 577            error_propagate(errp, local_err);
 578            return;
 579        }
 580    } else {
 581        /* write PT_NOTE to vmcore */
 582        write_elf32_note(s, &local_err);
 583        if (local_err) {
 584            error_propagate(errp, local_err);
 585            return;
 586        }
 587
 588        /* write all PT_LOAD to vmcore */
 589        write_elf_loads(s, &local_err);
 590        if (local_err) {
 591            error_propagate(errp, local_err);
 592            return;
 593        }
 594
 595        /* write section to vmcore */
 596        if (s->have_section) {
 597            write_elf_section(s, 0, &local_err);
 598            if (local_err) {
 599                error_propagate(errp, local_err);
 600                return;
 601            }
 602        }
 603
 604        /* write notes to vmcore */
 605        write_elf32_notes(fd_write_vmcore, s, &local_err);
 606        if (local_err) {
 607            error_propagate(errp, local_err);
 608            return;
 609        }
 610    }
 611}
 612
 613static int get_next_block(DumpState *s, GuestPhysBlock *block)
 614{
 615    while (1) {
 616        block = QTAILQ_NEXT(block, next);
 617        if (!block) {
 618            /* no more block */
 619            return 1;
 620        }
 621
 622        s->start = 0;
 623        s->next_block = block;
 624        if (s->has_filter) {
 625            if (block->target_start >= s->begin + s->length ||
 626                block->target_end <= s->begin) {
 627                /* This block is out of the range */
 628                continue;
 629            }
 630
 631            if (s->begin > block->target_start) {
 632                s->start = s->begin - block->target_start;
 633            }
 634        }
 635
 636        return 0;
 637    }
 638}
 639
 640/* write all memory to vmcore */
 641static void dump_iterate(DumpState *s, Error **errp)
 642{
 643    GuestPhysBlock *block;
 644    int64_t size;
 645    Error *local_err = NULL;
 646
 647    do {
 648        block = s->next_block;
 649
 650        size = block->target_end - block->target_start;
 651        if (s->has_filter) {
 652            size -= s->start;
 653            if (s->begin + s->length < block->target_end) {
 654                size -= block->target_end - (s->begin + s->length);
 655            }
 656        }
 657        write_memory(s, block, s->start, size, &local_err);
 658        if (local_err) {
 659            error_propagate(errp, local_err);
 660            return;
 661        }
 662
 663    } while (!get_next_block(s, block));
 664}
 665
 666static void create_vmcore(DumpState *s, Error **errp)
 667{
 668    Error *local_err = NULL;
 669
 670    dump_begin(s, &local_err);
 671    if (local_err) {
 672        error_propagate(errp, local_err);
 673        return;
 674    }
 675
 676    dump_iterate(s, errp);
 677}
 678
 679static int write_start_flat_header(int fd)
 680{
 681    MakedumpfileHeader *mh;
 682    int ret = 0;
 683
 684    QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
 685    mh = g_malloc0(MAX_SIZE_MDF_HEADER);
 686
 687    memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
 688           MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
 689
 690    mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
 691    mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
 692
 693    size_t written_size;
 694    written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
 695    if (written_size != MAX_SIZE_MDF_HEADER) {
 696        ret = -1;
 697    }
 698
 699    g_free(mh);
 700    return ret;
 701}
 702
 703static int write_end_flat_header(int fd)
 704{
 705    MakedumpfileDataHeader mdh;
 706
 707    mdh.offset = END_FLAG_FLAT_HEADER;
 708    mdh.buf_size = END_FLAG_FLAT_HEADER;
 709
 710    size_t written_size;
 711    written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 712    if (written_size != sizeof(mdh)) {
 713        return -1;
 714    }
 715
 716    return 0;
 717}
 718
 719static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
 720{
 721    size_t written_size;
 722    MakedumpfileDataHeader mdh;
 723
 724    mdh.offset = cpu_to_be64(offset);
 725    mdh.buf_size = cpu_to_be64(size);
 726
 727    written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 728    if (written_size != sizeof(mdh)) {
 729        return -1;
 730    }
 731
 732    written_size = qemu_write_full(fd, buf, size);
 733    if (written_size != size) {
 734        return -1;
 735    }
 736
 737    return 0;
 738}
 739
 740static int buf_write_note(const void *buf, size_t size, void *opaque)
 741{
 742    DumpState *s = opaque;
 743
 744    /* note_buf is not enough */
 745    if (s->note_buf_offset + size > s->note_size) {
 746        return -1;
 747    }
 748
 749    memcpy(s->note_buf + s->note_buf_offset, buf, size);
 750
 751    s->note_buf_offset += size;
 752
 753    return 0;
 754}
 755
 756/*
 757 * This function retrieves various sizes from an elf header.
 758 *
 759 * @note has to be a valid ELF note. The return sizes are unmodified
 760 * (not padded or rounded up to be multiple of 4).
 761 */
 762static void get_note_sizes(DumpState *s, const void *note,
 763                           uint64_t *note_head_size,
 764                           uint64_t *name_size,
 765                           uint64_t *desc_size)
 766{
 767    uint64_t note_head_sz;
 768    uint64_t name_sz;
 769    uint64_t desc_sz;
 770
 771    if (s->dump_info.d_class == ELFCLASS64) {
 772        const Elf64_Nhdr *hdr = note;
 773        note_head_sz = sizeof(Elf64_Nhdr);
 774        name_sz = tswap64(hdr->n_namesz);
 775        desc_sz = tswap64(hdr->n_descsz);
 776    } else {
 777        const Elf32_Nhdr *hdr = note;
 778        note_head_sz = sizeof(Elf32_Nhdr);
 779        name_sz = tswap32(hdr->n_namesz);
 780        desc_sz = tswap32(hdr->n_descsz);
 781    }
 782
 783    if (note_head_size) {
 784        *note_head_size = note_head_sz;
 785    }
 786    if (name_size) {
 787        *name_size = name_sz;
 788    }
 789    if (desc_size) {
 790        *desc_size = desc_sz;
 791    }
 792}
 793
 794static bool note_name_equal(DumpState *s,
 795                            const uint8_t *note, const char *name)
 796{
 797    int len = strlen(name) + 1;
 798    uint64_t head_size, name_size;
 799
 800    get_note_sizes(s, note, &head_size, &name_size, NULL);
 801    head_size = ROUND_UP(head_size, 4);
 802
 803    return name_size == len && memcmp(note + head_size, name, len) == 0;
 804}
 805
 806/* write common header, sub header and elf note to vmcore */
 807static void create_header32(DumpState *s, Error **errp)
 808{
 809    DiskDumpHeader32 *dh = NULL;
 810    KdumpSubHeader32 *kh = NULL;
 811    size_t size;
 812    uint32_t block_size;
 813    uint32_t sub_hdr_size;
 814    uint32_t bitmap_blocks;
 815    uint32_t status = 0;
 816    uint64_t offset_note;
 817    Error *local_err = NULL;
 818
 819    /* write common header, the version of kdump-compressed format is 6th */
 820    size = sizeof(DiskDumpHeader32);
 821    dh = g_malloc0(size);
 822
 823    memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
 824    dh->header_version = cpu_to_dump32(s, 6);
 825    block_size = s->dump_info.page_size;
 826    dh->block_size = cpu_to_dump32(s, block_size);
 827    sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
 828    sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 829    dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
 830    /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 831    dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
 832    dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
 833    bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 834    dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
 835    strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
 836
 837    if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 838        status |= DUMP_DH_COMPRESSED_ZLIB;
 839    }
 840#ifdef CONFIG_LZO
 841    if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 842        status |= DUMP_DH_COMPRESSED_LZO;
 843    }
 844#endif
 845#ifdef CONFIG_SNAPPY
 846    if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 847        status |= DUMP_DH_COMPRESSED_SNAPPY;
 848    }
 849#endif
 850    dh->status = cpu_to_dump32(s, status);
 851
 852    if (write_buffer(s->fd, 0, dh, size) < 0) {
 853        error_setg(errp, "dump: failed to write disk dump header");
 854        goto out;
 855    }
 856
 857    /* write sub header */
 858    size = sizeof(KdumpSubHeader32);
 859    kh = g_malloc0(size);
 860
 861    /* 64bit max_mapnr_64 */
 862    kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
 863    kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base);
 864    kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
 865
 866    offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 867    if (s->guest_note &&
 868        note_name_equal(s, s->guest_note, "VMCOREINFO")) {
 869        uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
 870
 871        get_note_sizes(s, s->guest_note,
 872                       &hsize, &name_size, &size_vmcoreinfo_desc);
 873        offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
 874            (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
 875        kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
 876        kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc);
 877    }
 878
 879    kh->offset_note = cpu_to_dump64(s, offset_note);
 880    kh->note_size = cpu_to_dump32(s, s->note_size);
 881
 882    if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 883                     block_size, kh, size) < 0) {
 884        error_setg(errp, "dump: failed to write kdump sub header");
 885        goto out;
 886    }
 887
 888    /* write note */
 889    s->note_buf = g_malloc0(s->note_size);
 890    s->note_buf_offset = 0;
 891
 892    /* use s->note_buf to store notes temporarily */
 893    write_elf32_notes(buf_write_note, s, &local_err);
 894    if (local_err) {
 895        error_propagate(errp, local_err);
 896        goto out;
 897    }
 898    if (write_buffer(s->fd, offset_note, s->note_buf,
 899                     s->note_size) < 0) {
 900        error_setg(errp, "dump: failed to write notes");
 901        goto out;
 902    }
 903
 904    /* get offset of dump_bitmap */
 905    s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 906                             block_size;
 907
 908    /* get offset of page */
 909    s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 910                     block_size;
 911
 912out:
 913    g_free(dh);
 914    g_free(kh);
 915    g_free(s->note_buf);
 916}
 917
 918/* write common header, sub header and elf note to vmcore */
 919static void create_header64(DumpState *s, Error **errp)
 920{
 921    DiskDumpHeader64 *dh = NULL;
 922    KdumpSubHeader64 *kh = NULL;
 923    size_t size;
 924    uint32_t block_size;
 925    uint32_t sub_hdr_size;
 926    uint32_t bitmap_blocks;
 927    uint32_t status = 0;
 928    uint64_t offset_note;
 929    Error *local_err = NULL;
 930
 931    /* write common header, the version of kdump-compressed format is 6th */
 932    size = sizeof(DiskDumpHeader64);
 933    dh = g_malloc0(size);
 934
 935    memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN);
 936    dh->header_version = cpu_to_dump32(s, 6);
 937    block_size = s->dump_info.page_size;
 938    dh->block_size = cpu_to_dump32(s, block_size);
 939    sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
 940    sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 941    dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
 942    /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 943    dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
 944    dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
 945    bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 946    dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
 947    strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
 948
 949    if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 950        status |= DUMP_DH_COMPRESSED_ZLIB;
 951    }
 952#ifdef CONFIG_LZO
 953    if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 954        status |= DUMP_DH_COMPRESSED_LZO;
 955    }
 956#endif
 957#ifdef CONFIG_SNAPPY
 958    if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 959        status |= DUMP_DH_COMPRESSED_SNAPPY;
 960    }
 961#endif
 962    dh->status = cpu_to_dump32(s, status);
 963
 964    if (write_buffer(s->fd, 0, dh, size) < 0) {
 965        error_setg(errp, "dump: failed to write disk dump header");
 966        goto out;
 967    }
 968
 969    /* write sub header */
 970    size = sizeof(KdumpSubHeader64);
 971    kh = g_malloc0(size);
 972
 973    /* 64bit max_mapnr_64 */
 974    kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
 975    kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base);
 976    kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
 977
 978    offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 979    if (s->guest_note &&
 980        note_name_equal(s, s->guest_note, "VMCOREINFO")) {
 981        uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo;
 982
 983        get_note_sizes(s, s->guest_note,
 984                       &hsize, &name_size, &size_vmcoreinfo_desc);
 985        offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size +
 986            (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4;
 987        kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo);
 988        kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc);
 989    }
 990
 991    kh->offset_note = cpu_to_dump64(s, offset_note);
 992    kh->note_size = cpu_to_dump64(s, s->note_size);
 993
 994    if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 995                     block_size, kh, size) < 0) {
 996        error_setg(errp, "dump: failed to write kdump sub header");
 997        goto out;
 998    }
 999
1000    /* write note */
1001    s->note_buf = g_malloc0(s->note_size);
1002    s->note_buf_offset = 0;
1003
1004    /* use s->note_buf to store notes temporarily */
1005    write_elf64_notes(buf_write_note, s, &local_err);
1006    if (local_err) {
1007        error_propagate(errp, local_err);
1008        goto out;
1009    }
1010
1011    if (write_buffer(s->fd, offset_note, s->note_buf,
1012                     s->note_size) < 0) {
1013        error_setg(errp, "dump: failed to write notes");
1014        goto out;
1015    }
1016
1017    /* get offset of dump_bitmap */
1018    s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
1019                             block_size;
1020
1021    /* get offset of page */
1022    s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
1023                     block_size;
1024
1025out:
1026    g_free(dh);
1027    g_free(kh);
1028    g_free(s->note_buf);
1029}
1030
1031static void write_dump_header(DumpState *s, Error **errp)
1032{
1033    if (s->dump_info.d_class == ELFCLASS32) {
1034        create_header32(s, errp);
1035    } else {
1036        create_header64(s, errp);
1037    }
1038}
1039
1040static size_t dump_bitmap_get_bufsize(DumpState *s)
1041{
1042    return s->dump_info.page_size;
1043}
1044
1045/*
1046 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1047 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1048 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1049 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1050 * vmcore, ie. synchronizing un-sync bit into vmcore.
1051 */
1052static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1053                           uint8_t *buf, DumpState *s)
1054{
1055    off_t old_offset, new_offset;
1056    off_t offset_bitmap1, offset_bitmap2;
1057    uint32_t byte, bit;
1058    size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1059    size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1060
1061    /* should not set the previous place */
1062    assert(last_pfn <= pfn);
1063
1064    /*
1065     * if the bit needed to be set is not cached in buf, flush the data in buf
1066     * to vmcore firstly.
1067     * making new_offset be bigger than old_offset can also sync remained data
1068     * into vmcore.
1069     */
1070    old_offset = bitmap_bufsize * (last_pfn / bits_per_buf);
1071    new_offset = bitmap_bufsize * (pfn / bits_per_buf);
1072
1073    while (old_offset < new_offset) {
1074        /* calculate the offset and write dump_bitmap */
1075        offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1076        if (write_buffer(s->fd, offset_bitmap1, buf,
1077                         bitmap_bufsize) < 0) {
1078            return -1;
1079        }
1080
1081        /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1082        offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1083                         old_offset;
1084        if (write_buffer(s->fd, offset_bitmap2, buf,
1085                         bitmap_bufsize) < 0) {
1086            return -1;
1087        }
1088
1089        memset(buf, 0, bitmap_bufsize);
1090        old_offset += bitmap_bufsize;
1091    }
1092
1093    /* get the exact place of the bit in the buf, and set it */
1094    byte = (pfn % bits_per_buf) / CHAR_BIT;
1095    bit = (pfn % bits_per_buf) % CHAR_BIT;
1096    if (value) {
1097        buf[byte] |= 1u << bit;
1098    } else {
1099        buf[byte] &= ~(1u << bit);
1100    }
1101
1102    return 0;
1103}
1104
1105static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr)
1106{
1107    int target_page_shift = ctz32(s->dump_info.page_size);
1108
1109    return (addr >> target_page_shift) - ARCH_PFN_OFFSET;
1110}
1111
1112static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn)
1113{
1114    int target_page_shift = ctz32(s->dump_info.page_size);
1115
1116    return (pfn + ARCH_PFN_OFFSET) << target_page_shift;
1117}
1118
1119/*
1120 * exam every page and return the page frame number and the address of the page.
1121 * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1122 * blocks, so block->target_start and block->target_end should be interal
1123 * multiples of the target page size.
1124 */
1125static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1126                          uint8_t **bufptr, DumpState *s)
1127{
1128    GuestPhysBlock *block = *blockptr;
1129    hwaddr addr, target_page_mask = ~((hwaddr)s->dump_info.page_size - 1);
1130    uint8_t *buf;
1131
1132    /* block == NULL means the start of the iteration */
1133    if (!block) {
1134        block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1135        *blockptr = block;
1136        assert((block->target_start & ~target_page_mask) == 0);
1137        assert((block->target_end & ~target_page_mask) == 0);
1138        *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1139        if (bufptr) {
1140            *bufptr = block->host_addr;
1141        }
1142        return true;
1143    }
1144
1145    *pfnptr = *pfnptr + 1;
1146    addr = dump_pfn_to_paddr(s, *pfnptr);
1147
1148    if ((addr >= block->target_start) &&
1149        (addr + s->dump_info.page_size <= block->target_end)) {
1150        buf = block->host_addr + (addr - block->target_start);
1151    } else {
1152        /* the next page is in the next block */
1153        block = QTAILQ_NEXT(block, next);
1154        *blockptr = block;
1155        if (!block) {
1156            return false;
1157        }
1158        assert((block->target_start & ~target_page_mask) == 0);
1159        assert((block->target_end & ~target_page_mask) == 0);
1160        *pfnptr = dump_paddr_to_pfn(s, block->target_start);
1161        buf = block->host_addr;
1162    }
1163
1164    if (bufptr) {
1165        *bufptr = buf;
1166    }
1167
1168    return true;
1169}
1170
1171static void write_dump_bitmap(DumpState *s, Error **errp)
1172{
1173    int ret = 0;
1174    uint64_t last_pfn, pfn;
1175    void *dump_bitmap_buf;
1176    size_t num_dumpable;
1177    GuestPhysBlock *block_iter = NULL;
1178    size_t bitmap_bufsize = dump_bitmap_get_bufsize(s);
1179    size_t bits_per_buf = bitmap_bufsize * CHAR_BIT;
1180
1181    /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1182    dump_bitmap_buf = g_malloc0(bitmap_bufsize);
1183
1184    num_dumpable = 0;
1185    last_pfn = 0;
1186
1187    /*
1188     * exam memory page by page, and set the bit in dump_bitmap corresponded
1189     * to the existing page.
1190     */
1191    while (get_next_page(&block_iter, &pfn, NULL, s)) {
1192        ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1193        if (ret < 0) {
1194            error_setg(errp, "dump: failed to set dump_bitmap");
1195            goto out;
1196        }
1197
1198        last_pfn = pfn;
1199        num_dumpable++;
1200    }
1201
1202    /*
1203     * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1204     * set the remaining bits from last_pfn to the end of the bitmap buffer to
1205     * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1206     */
1207    if (num_dumpable > 0) {
1208        ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false,
1209                              dump_bitmap_buf, s);
1210        if (ret < 0) {
1211            error_setg(errp, "dump: failed to sync dump_bitmap");
1212            goto out;
1213        }
1214    }
1215
1216    /* number of dumpable pages that will be dumped later */
1217    s->num_dumpable = num_dumpable;
1218
1219out:
1220    g_free(dump_bitmap_buf);
1221}
1222
1223static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1224                               off_t offset)
1225{
1226    data_cache->fd = s->fd;
1227    data_cache->data_size = 0;
1228    data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s);
1229    data_cache->buf = g_malloc0(data_cache->buf_size);
1230    data_cache->offset = offset;
1231}
1232
1233static int write_cache(DataCache *dc, const void *buf, size_t size,
1234                       bool flag_sync)
1235{
1236    /*
1237     * dc->buf_size should not be less than size, otherwise dc will never be
1238     * enough
1239     */
1240    assert(size <= dc->buf_size);
1241
1242    /*
1243     * if flag_sync is set, synchronize data in dc->buf into vmcore.
1244     * otherwise check if the space is enough for caching data in buf, if not,
1245     * write the data in dc->buf to dc->fd and reset dc->buf
1246     */
1247    if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1248        (flag_sync && dc->data_size > 0)) {
1249        if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1250            return -1;
1251        }
1252
1253        dc->offset += dc->data_size;
1254        dc->data_size = 0;
1255    }
1256
1257    if (!flag_sync) {
1258        memcpy(dc->buf + dc->data_size, buf, size);
1259        dc->data_size += size;
1260    }
1261
1262    return 0;
1263}
1264
1265static void free_data_cache(DataCache *data_cache)
1266{
1267    g_free(data_cache->buf);
1268}
1269
1270static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1271{
1272    switch (flag_compress) {
1273    case DUMP_DH_COMPRESSED_ZLIB:
1274        return compressBound(page_size);
1275
1276    case DUMP_DH_COMPRESSED_LZO:
1277        /*
1278         * LZO will expand incompressible data by a little amount. Please check
1279         * the following URL to see the expansion calculation:
1280         * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1281         */
1282        return page_size + page_size / 16 + 64 + 3;
1283
1284#ifdef CONFIG_SNAPPY
1285    case DUMP_DH_COMPRESSED_SNAPPY:
1286        return snappy_max_compressed_length(page_size);
1287#endif
1288    }
1289    return 0;
1290}
1291
1292/*
1293 * check if the page is all 0
1294 */
1295static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1296{
1297    return buffer_is_zero(buf, page_size);
1298}
1299
1300static void write_dump_pages(DumpState *s, Error **errp)
1301{
1302    int ret = 0;
1303    DataCache page_desc, page_data;
1304    size_t len_buf_out, size_out;
1305#ifdef CONFIG_LZO
1306    lzo_bytep wrkmem = NULL;
1307#endif
1308    uint8_t *buf_out = NULL;
1309    off_t offset_desc, offset_data;
1310    PageDescriptor pd, pd_zero;
1311    uint8_t *buf;
1312    GuestPhysBlock *block_iter = NULL;
1313    uint64_t pfn_iter;
1314
1315    /* get offset of page_desc and page_data in dump file */
1316    offset_desc = s->offset_page;
1317    offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1318
1319    prepare_data_cache(&page_desc, s, offset_desc);
1320    prepare_data_cache(&page_data, s, offset_data);
1321
1322    /* prepare buffer to store compressed data */
1323    len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress);
1324    assert(len_buf_out != 0);
1325
1326#ifdef CONFIG_LZO
1327    wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1328#endif
1329
1330    buf_out = g_malloc(len_buf_out);
1331
1332    /*
1333     * init zero page's page_desc and page_data, because every zero page
1334     * uses the same page_data
1335     */
1336    pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size);
1337    pd_zero.flags = cpu_to_dump32(s, 0);
1338    pd_zero.offset = cpu_to_dump64(s, offset_data);
1339    pd_zero.page_flags = cpu_to_dump64(s, 0);
1340    buf = g_malloc0(s->dump_info.page_size);
1341    ret = write_cache(&page_data, buf, s->dump_info.page_size, false);
1342    g_free(buf);
1343    if (ret < 0) {
1344        error_setg(errp, "dump: failed to write page data (zero page)");
1345        goto out;
1346    }
1347
1348    offset_data += s->dump_info.page_size;
1349
1350    /*
1351     * dump memory to vmcore page by page. zero page will all be resided in the
1352     * first page of page section
1353     */
1354    while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1355        /* check zero page */
1356        if (is_zero_page(buf, s->dump_info.page_size)) {
1357            ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1358                              false);
1359            if (ret < 0) {
1360                error_setg(errp, "dump: failed to write page desc");
1361                goto out;
1362            }
1363        } else {
1364            /*
1365             * not zero page, then:
1366             * 1. compress the page
1367             * 2. write the compressed page into the cache of page_data
1368             * 3. get page desc of the compressed page and write it into the
1369             *    cache of page_desc
1370             *
1371             * only one compression format will be used here, for
1372             * s->flag_compress is set. But when compression fails to work,
1373             * we fall back to save in plaintext.
1374             */
1375             size_out = len_buf_out;
1376             if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1377                    (compress2(buf_out, (uLongf *)&size_out, buf,
1378                               s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) &&
1379                    (size_out < s->dump_info.page_size)) {
1380                pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1381                pd.size  = cpu_to_dump32(s, size_out);
1382
1383                ret = write_cache(&page_data, buf_out, size_out, false);
1384                if (ret < 0) {
1385                    error_setg(errp, "dump: failed to write page data");
1386                    goto out;
1387                }
1388#ifdef CONFIG_LZO
1389            } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1390                    (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out,
1391                    (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1392                    (size_out < s->dump_info.page_size)) {
1393                pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1394                pd.size  = cpu_to_dump32(s, size_out);
1395
1396                ret = write_cache(&page_data, buf_out, size_out, false);
1397                if (ret < 0) {
1398                    error_setg(errp, "dump: failed to write page data");
1399                    goto out;
1400                }
1401#endif
1402#ifdef CONFIG_SNAPPY
1403            } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1404                    (snappy_compress((char *)buf, s->dump_info.page_size,
1405                    (char *)buf_out, &size_out) == SNAPPY_OK) &&
1406                    (size_out < s->dump_info.page_size)) {
1407                pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1408                pd.size  = cpu_to_dump32(s, size_out);
1409
1410                ret = write_cache(&page_data, buf_out, size_out, false);
1411                if (ret < 0) {
1412                    error_setg(errp, "dump: failed to write page data");
1413                    goto out;
1414                }
1415#endif
1416            } else {
1417                /*
1418                 * fall back to save in plaintext, size_out should be
1419                 * assigned the target's page size
1420                 */
1421                pd.flags = cpu_to_dump32(s, 0);
1422                size_out = s->dump_info.page_size;
1423                pd.size = cpu_to_dump32(s, size_out);
1424
1425                ret = write_cache(&page_data, buf,
1426                                  s->dump_info.page_size, false);
1427                if (ret < 0) {
1428                    error_setg(errp, "dump: failed to write page data");
1429                    goto out;
1430                }
1431            }
1432
1433            /* get and write page desc here */
1434            pd.page_flags = cpu_to_dump64(s, 0);
1435            pd.offset = cpu_to_dump64(s, offset_data);
1436            offset_data += size_out;
1437
1438            ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1439            if (ret < 0) {
1440                error_setg(errp, "dump: failed to write page desc");
1441                goto out;
1442            }
1443        }
1444        s->written_size += s->dump_info.page_size;
1445    }
1446
1447    ret = write_cache(&page_desc, NULL, 0, true);
1448    if (ret < 0) {
1449        error_setg(errp, "dump: failed to sync cache for page_desc");
1450        goto out;
1451    }
1452    ret = write_cache(&page_data, NULL, 0, true);
1453    if (ret < 0) {
1454        error_setg(errp, "dump: failed to sync cache for page_data");
1455        goto out;
1456    }
1457
1458out:
1459    free_data_cache(&page_desc);
1460    free_data_cache(&page_data);
1461
1462#ifdef CONFIG_LZO
1463    g_free(wrkmem);
1464#endif
1465
1466    g_free(buf_out);
1467}
1468
1469static void create_kdump_vmcore(DumpState *s, Error **errp)
1470{
1471    int ret;
1472    Error *local_err = NULL;
1473
1474    /*
1475     * the kdump-compressed format is:
1476     *                                               File offset
1477     *  +------------------------------------------+ 0x0
1478     *  |    main header (struct disk_dump_header) |
1479     *  |------------------------------------------+ block 1
1480     *  |    sub header (struct kdump_sub_header)  |
1481     *  |------------------------------------------+ block 2
1482     *  |            1st-dump_bitmap               |
1483     *  |------------------------------------------+ block 2 + X blocks
1484     *  |            2nd-dump_bitmap               | (aligned by block)
1485     *  |------------------------------------------+ block 2 + 2 * X blocks
1486     *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1487     *  |  page desc for pfn 1 (struct page_desc)  |
1488     *  |                    :                     |
1489     *  |------------------------------------------| (not aligned by block)
1490     *  |         page data (pfn 0)                |
1491     *  |         page data (pfn 1)                |
1492     *  |                    :                     |
1493     *  +------------------------------------------+
1494     */
1495
1496    ret = write_start_flat_header(s->fd);
1497    if (ret < 0) {
1498        error_setg(errp, "dump: failed to write start flat header");
1499        return;
1500    }
1501
1502    write_dump_header(s, &local_err);
1503    if (local_err) {
1504        error_propagate(errp, local_err);
1505        return;
1506    }
1507
1508    write_dump_bitmap(s, &local_err);
1509    if (local_err) {
1510        error_propagate(errp, local_err);
1511        return;
1512    }
1513
1514    write_dump_pages(s, &local_err);
1515    if (local_err) {
1516        error_propagate(errp, local_err);
1517        return;
1518    }
1519
1520    ret = write_end_flat_header(s->fd);
1521    if (ret < 0) {
1522        error_setg(errp, "dump: failed to write end flat header");
1523        return;
1524    }
1525}
1526
1527static ram_addr_t get_start_block(DumpState *s)
1528{
1529    GuestPhysBlock *block;
1530
1531    if (!s->has_filter) {
1532        s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1533        return 0;
1534    }
1535
1536    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1537        if (block->target_start >= s->begin + s->length ||
1538            block->target_end <= s->begin) {
1539            /* This block is out of the range */
1540            continue;
1541        }
1542
1543        s->next_block = block;
1544        if (s->begin > block->target_start) {
1545            s->start = s->begin - block->target_start;
1546        } else {
1547            s->start = 0;
1548        }
1549        return s->start;
1550    }
1551
1552    return -1;
1553}
1554
1555static void get_max_mapnr(DumpState *s)
1556{
1557    GuestPhysBlock *last_block;
1558
1559    last_block = QTAILQ_LAST(&s->guest_phys_blocks.head);
1560    s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end);
1561}
1562
1563static DumpState dump_state_global = { .status = DUMP_STATUS_NONE };
1564
1565static void dump_state_prepare(DumpState *s)
1566{
1567    /* zero the struct, setting status to active */
1568    *s = (DumpState) { .status = DUMP_STATUS_ACTIVE };
1569}
1570
1571bool dump_in_progress(void)
1572{
1573    DumpState *state = &dump_state_global;
1574    return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE);
1575}
1576
1577/* calculate total size of memory to be dumped (taking filter into
1578 * acoount.) */
1579static int64_t dump_calculate_size(DumpState *s)
1580{
1581    GuestPhysBlock *block;
1582    int64_t size = 0, total = 0, left = 0, right = 0;
1583
1584    QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1585        if (s->has_filter) {
1586            /* calculate the overlapped region. */
1587            left = MAX(s->begin, block->target_start);
1588            right = MIN(s->begin + s->length, block->target_end);
1589            size = right - left;
1590            size = size > 0 ? size : 0;
1591        } else {
1592            /* count the whole region in */
1593            size = (block->target_end - block->target_start);
1594        }
1595        total += size;
1596    }
1597
1598    return total;
1599}
1600
1601static void vmcoreinfo_update_phys_base(DumpState *s)
1602{
1603    uint64_t size, note_head_size, name_size, phys_base;
1604    char **lines;
1605    uint8_t *vmci;
1606    size_t i;
1607
1608    if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) {
1609        return;
1610    }
1611
1612    get_note_sizes(s, s->guest_note, &note_head_size, &name_size, &size);
1613    note_head_size = ROUND_UP(note_head_size, 4);
1614
1615    vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4);
1616    *(vmci + size) = '\0';
1617
1618    lines = g_strsplit((char *)vmci, "\n", -1);
1619    for (i = 0; lines[i]; i++) {
1620        const char *prefix = NULL;
1621
1622        if (s->dump_info.d_machine == EM_X86_64) {
1623            prefix = "NUMBER(phys_base)=";
1624        } else if (s->dump_info.d_machine == EM_AARCH64) {
1625            prefix = "NUMBER(PHYS_OFFSET)=";
1626        }
1627
1628        if (prefix && g_str_has_prefix(lines[i], prefix)) {
1629            if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16,
1630                              &phys_base) < 0) {
1631                warn_report("Failed to read %s", prefix);
1632            } else {
1633                s->dump_info.phys_base = phys_base;
1634            }
1635            break;
1636        }
1637    }
1638
1639    g_strfreev(lines);
1640}
1641
1642static void dump_init(DumpState *s, int fd, bool has_format,
1643                      DumpGuestMemoryFormat format, bool paging, bool has_filter,
1644                      int64_t begin, int64_t length, Error **errp)
1645{
1646    VMCoreInfoState *vmci = vmcoreinfo_find();
1647    CPUState *cpu;
1648    int nr_cpus;
1649    Error *err = NULL;
1650    int ret;
1651
1652    s->has_format = has_format;
1653    s->format = format;
1654    s->written_size = 0;
1655
1656    /* kdump-compressed is conflict with paging and filter */
1657    if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1658        assert(!paging && !has_filter);
1659    }
1660
1661    if (runstate_is_running()) {
1662        vm_stop(RUN_STATE_SAVE_VM);
1663        s->resume = true;
1664    } else {
1665        s->resume = false;
1666    }
1667
1668    /* If we use KVM, we should synchronize the registers before we get dump
1669     * info or physmap info.
1670     */
1671    cpu_synchronize_all_states();
1672    nr_cpus = 0;
1673    CPU_FOREACH(cpu) {
1674        nr_cpus++;
1675    }
1676
1677    s->fd = fd;
1678    s->has_filter = has_filter;
1679    s->begin = begin;
1680    s->length = length;
1681
1682    memory_mapping_list_init(&s->list);
1683
1684    guest_phys_blocks_init(&s->guest_phys_blocks);
1685    guest_phys_blocks_append(&s->guest_phys_blocks);
1686    s->total_size = dump_calculate_size(s);
1687#ifdef DEBUG_DUMP_GUEST_MEMORY
1688    fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size);
1689#endif
1690
1691    /* it does not make sense to dump non-existent memory */
1692    if (!s->total_size) {
1693        error_setg(errp, "dump: no guest memory to dump");
1694        goto cleanup;
1695    }
1696
1697    s->start = get_start_block(s);
1698    if (s->start == -1) {
1699        error_setg(errp, QERR_INVALID_PARAMETER, "begin");
1700        goto cleanup;
1701    }
1702
1703    /* get dump info: endian, class and architecture.
1704     * If the target architecture is not supported, cpu_get_dump_info() will
1705     * return -1.
1706     */
1707    ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1708    if (ret < 0) {
1709        error_setg(errp, QERR_UNSUPPORTED);
1710        goto cleanup;
1711    }
1712
1713    if (!s->dump_info.page_size) {
1714        s->dump_info.page_size = TARGET_PAGE_SIZE;
1715    }
1716
1717    s->note_size = cpu_get_note_size(s->dump_info.d_class,
1718                                     s->dump_info.d_machine, nr_cpus);
1719    if (s->note_size < 0) {
1720        error_setg(errp, QERR_UNSUPPORTED);
1721        goto cleanup;
1722    }
1723
1724    /*
1725     * The goal of this block is to (a) update the previously guessed
1726     * phys_base, (b) copy the guest note out of the guest.
1727     * Failure to do so is not fatal for dumping.
1728     */
1729    if (vmci) {
1730        uint64_t addr, note_head_size, name_size, desc_size;
1731        uint32_t size;
1732        uint16_t format;
1733
1734        note_head_size = s->dump_info.d_class == ELFCLASS32 ?
1735            sizeof(Elf32_Nhdr) : sizeof(Elf64_Nhdr);
1736
1737        format = le16_to_cpu(vmci->vmcoreinfo.guest_format);
1738        size = le32_to_cpu(vmci->vmcoreinfo.size);
1739        addr = le64_to_cpu(vmci->vmcoreinfo.paddr);
1740        if (!vmci->has_vmcoreinfo) {
1741            warn_report("guest note is not present");
1742        } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) {
1743            warn_report("guest note size is invalid: %" PRIu32, size);
1744        } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) {
1745            warn_report("guest note format is unsupported: %" PRIu16, format);
1746        } else {
1747            s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */
1748            cpu_physical_memory_read(addr, s->guest_note, size);
1749
1750            get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size);
1751            s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size,
1752                                               desc_size);
1753            if (name_size > MAX_GUEST_NOTE_SIZE ||
1754                desc_size > MAX_GUEST_NOTE_SIZE ||
1755                s->guest_note_size > size) {
1756                warn_report("Invalid guest note header");
1757                g_free(s->guest_note);
1758                s->guest_note = NULL;
1759            } else {
1760                vmcoreinfo_update_phys_base(s);
1761                s->note_size += s->guest_note_size;
1762            }
1763        }
1764    }
1765
1766    /* get memory mapping */
1767    if (paging) {
1768        qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1769        if (err != NULL) {
1770            error_propagate(errp, err);
1771            goto cleanup;
1772        }
1773    } else {
1774        qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1775    }
1776
1777    s->nr_cpus = nr_cpus;
1778
1779    get_max_mapnr(s);
1780
1781    uint64_t tmp;
1782    tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT),
1783                       s->dump_info.page_size);
1784    s->len_dump_bitmap = tmp * s->dump_info.page_size;
1785
1786    /* init for kdump-compressed format */
1787    if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1788        switch (format) {
1789        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1790            s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1791            break;
1792
1793        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1794#ifdef CONFIG_LZO
1795            if (lzo_init() != LZO_E_OK) {
1796                error_setg(errp, "failed to initialize the LZO library");
1797                goto cleanup;
1798            }
1799#endif
1800            s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1801            break;
1802
1803        case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1804            s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1805            break;
1806
1807        default:
1808            s->flag_compress = 0;
1809        }
1810
1811        return;
1812    }
1813
1814    if (s->has_filter) {
1815        memory_mapping_filter(&s->list, s->begin, s->length);
1816    }
1817
1818    /*
1819     * calculate phdr_num
1820     *
1821     * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1822     */
1823    s->phdr_num = 1; /* PT_NOTE */
1824    if (s->list.num < UINT16_MAX - 2) {
1825        s->phdr_num += s->list.num;
1826        s->have_section = false;
1827    } else {
1828        s->have_section = true;
1829        s->phdr_num = PN_XNUM;
1830        s->sh_info = 1; /* PT_NOTE */
1831
1832        /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1833        if (s->list.num <= UINT32_MAX - 1) {
1834            s->sh_info += s->list.num;
1835        } else {
1836            s->sh_info = UINT32_MAX;
1837        }
1838    }
1839
1840    if (s->dump_info.d_class == ELFCLASS64) {
1841        if (s->have_section) {
1842            s->memory_offset = sizeof(Elf64_Ehdr) +
1843                               sizeof(Elf64_Phdr) * s->sh_info +
1844                               sizeof(Elf64_Shdr) + s->note_size;
1845        } else {
1846            s->memory_offset = sizeof(Elf64_Ehdr) +
1847                               sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1848        }
1849    } else {
1850        if (s->have_section) {
1851            s->memory_offset = sizeof(Elf32_Ehdr) +
1852                               sizeof(Elf32_Phdr) * s->sh_info +
1853                               sizeof(Elf32_Shdr) + s->note_size;
1854        } else {
1855            s->memory_offset = sizeof(Elf32_Ehdr) +
1856                               sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1857        }
1858    }
1859
1860    return;
1861
1862cleanup:
1863    dump_cleanup(s);
1864}
1865
1866/* this operation might be time consuming. */
1867static void dump_process(DumpState *s, Error **errp)
1868{
1869    Error *local_err = NULL;
1870    DumpQueryResult *result = NULL;
1871
1872    if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1873#ifdef TARGET_X86_64
1874        create_win_dump(s, &local_err);
1875#endif
1876    } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1877        create_kdump_vmcore(s, &local_err);
1878    } else {
1879        create_vmcore(s, &local_err);
1880    }
1881
1882    /* make sure status is written after written_size updates */
1883    smp_wmb();
1884    qatomic_set(&s->status,
1885               (local_err ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED));
1886
1887    /* send DUMP_COMPLETED message (unconditionally) */
1888    result = qmp_query_dump(NULL);
1889    /* should never fail */
1890    assert(result);
1891    qapi_event_send_dump_completed(result, !!local_err, (local_err ?
1892                                   error_get_pretty(local_err) : NULL));
1893    qapi_free_DumpQueryResult(result);
1894
1895    error_propagate(errp, local_err);
1896    dump_cleanup(s);
1897}
1898
1899static void *dump_thread(void *data)
1900{
1901    DumpState *s = (DumpState *)data;
1902    dump_process(s, NULL);
1903    return NULL;
1904}
1905
1906DumpQueryResult *qmp_query_dump(Error **errp)
1907{
1908    DumpQueryResult *result = g_new(DumpQueryResult, 1);
1909    DumpState *state = &dump_state_global;
1910    result->status = qatomic_read(&state->status);
1911    /* make sure we are reading status and written_size in order */
1912    smp_rmb();
1913    result->completed = state->written_size;
1914    result->total = state->total_size;
1915    return result;
1916}
1917
1918void qmp_dump_guest_memory(bool paging, const char *file,
1919                           bool has_detach, bool detach,
1920                           bool has_begin, int64_t begin, bool has_length,
1921                           int64_t length, bool has_format,
1922                           DumpGuestMemoryFormat format, Error **errp)
1923{
1924    const char *p;
1925    int fd = -1;
1926    DumpState *s;
1927    Error *local_err = NULL;
1928    bool detach_p = false;
1929
1930    if (runstate_check(RUN_STATE_INMIGRATE)) {
1931        error_setg(errp, "Dump not allowed during incoming migration.");
1932        return;
1933    }
1934
1935    /* if there is a dump in background, we should wait until the dump
1936     * finished */
1937    if (dump_in_progress()) {
1938        error_setg(errp, "There is a dump in process, please wait.");
1939        return;
1940    }
1941
1942    /*
1943     * kdump-compressed format need the whole memory dumped, so paging or
1944     * filter is not supported here.
1945     */
1946    if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1947        (paging || has_begin || has_length)) {
1948        error_setg(errp, "kdump-compressed format doesn't support paging or "
1949                         "filter");
1950        return;
1951    }
1952    if (has_begin && !has_length) {
1953        error_setg(errp, QERR_MISSING_PARAMETER, "length");
1954        return;
1955    }
1956    if (!has_begin && has_length) {
1957        error_setg(errp, QERR_MISSING_PARAMETER, "begin");
1958        return;
1959    }
1960    if (has_detach) {
1961        detach_p = detach;
1962    }
1963
1964    /* check whether lzo/snappy is supported */
1965#ifndef CONFIG_LZO
1966    if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1967        error_setg(errp, "kdump-lzo is not available now");
1968        return;
1969    }
1970#endif
1971
1972#ifndef CONFIG_SNAPPY
1973    if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1974        error_setg(errp, "kdump-snappy is not available now");
1975        return;
1976    }
1977#endif
1978
1979#ifndef TARGET_X86_64
1980    if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) {
1981        error_setg(errp, "Windows dump is only available for x86-64");
1982        return;
1983    }
1984#endif
1985
1986#if !defined(WIN32)
1987    if (strstart(file, "fd:", &p)) {
1988        fd = monitor_get_fd(monitor_cur(), p, errp);
1989        if (fd == -1) {
1990            return;
1991        }
1992    }
1993#endif
1994
1995    if  (strstart(file, "file:", &p)) {
1996        fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1997        if (fd < 0) {
1998            error_setg_file_open(errp, errno, p);
1999            return;
2000        }
2001    }
2002
2003    if (fd == -1) {
2004        error_setg(errp, QERR_INVALID_PARAMETER, "protocol");
2005        return;
2006    }
2007
2008    s = &dump_state_global;
2009    dump_state_prepare(s);
2010
2011    dump_init(s, fd, has_format, format, paging, has_begin,
2012              begin, length, &local_err);
2013    if (local_err) {
2014        error_propagate(errp, local_err);
2015        qatomic_set(&s->status, DUMP_STATUS_FAILED);
2016        return;
2017    }
2018
2019    if (detach_p) {
2020        /* detached dump */
2021        s->detached = true;
2022        qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread,
2023                           s, QEMU_THREAD_DETACHED);
2024    } else {
2025        /* sync dump */
2026        dump_process(s, errp);
2027    }
2028}
2029
2030DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
2031{
2032    DumpGuestMemoryCapability *cap =
2033                                  g_malloc0(sizeof(DumpGuestMemoryCapability));
2034    DumpGuestMemoryFormatList **tail = &cap->formats;
2035
2036    /* elf is always available */
2037    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF);
2038
2039    /* kdump-zlib is always available */
2040    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB);
2041
2042    /* add new item if kdump-lzo is available */
2043#ifdef CONFIG_LZO
2044    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO);
2045#endif
2046
2047    /* add new item if kdump-snappy is available */
2048#ifdef CONFIG_SNAPPY
2049    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY);
2050#endif
2051
2052    /* Windows dump is available only if target is x86_64 */
2053#ifdef TARGET_X86_64
2054    QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP);
2055#endif
2056
2057    return cap;
2058}
2059