dump.c

   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-common.h"
  15 #include "elf.h"
  16 #include "cpu.h"
  17 #include "exec/cpu-all.h"
  18 #include "exec/hwaddr.h"
  19 #include "monitor/monitor.h"
  20 #include "sysemu/kvm.h"
  21 #include "sysemu/dump.h"
  22 #include "sysemu/sysemu.h"
  23 #include "sysemu/memory_mapping.h"
  24 #include "sysemu/cpus.h"
  25 #include "qapi/error.h"
  26 #include "qmp-commands.h"
  27
  28 #include <zlib.h>
  29 #ifdef CONFIG_LZO
  30 #include <lzo/lzo1x.h>
  31 #endif
  32 #ifdef CONFIG_SNAPPY
  33 #include <snappy-c.h>
  34 #endif
  35
  36 static uint16_t cpu_convert_to_target16(uint16_t val, int endian)
  37 {
  38     if (endian == ELFDATA2LSB) {
  39         val = cpu_to_le16(val);
  40     } else {
  41         val = cpu_to_be16(val);
  42     }
  43
  44     return val;
  45 }
  46
  47 static uint32_t cpu_convert_to_target32(uint32_t val, int endian)
  48 {
  49     if (endian == ELFDATA2LSB) {
  50         val = cpu_to_le32(val);
  51     } else {
  52         val = cpu_to_be32(val);
  53     }
  54
  55     return val;
  56 }
  57
  58 static uint64_t cpu_convert_to_target64(uint64_t val, int endian)
  59 {
  60     if (endian == ELFDATA2LSB) {
  61         val = cpu_to_le64(val);
  62     } else {
  63         val = cpu_to_be64(val);
  64     }
  65
  66     return val;
  67 }
  68
  69 typedef struct DumpState {
  70     GuestPhysBlockList guest_phys_blocks;
  71     ArchDumpInfo dump_info;
  72     MemoryMappingList list;
  73     uint16_t phdr_num;
  74     uint32_t sh_info;
  75     bool have_section;
  76     bool resume;
  77     ssize_t note_size;
  78     hwaddr memory_offset;
  79     int fd;
  80
  81     GuestPhysBlock *next_block;
  82     ram_addr_t start;
  83     bool has_filter;
  84     int64_t begin;
  85     int64_t length;
  86     Error **errp;
  87
  88     uint8_t *note_buf;          /* buffer for notes */
  89     size_t note_buf_offset;     /* the writing place in note_buf */
  90     uint32_t nr_cpus;           /* number of guest's cpu */
  91     size_t page_size;           /* guest's page size */
  92     uint32_t page_shift;        /* guest's page shift */
  93     uint64_t max_mapnr;         /* the biggest guest's phys-mem's number */
  94     size_t len_dump_bitmap;     /* the size of the place used to store
  95                                    dump_bitmap in vmcore */
  96     off_t offset_dump_bitmap;   /* offset of dump_bitmap part in vmcore */
  97     off_t offset_page;          /* offset of page part in vmcore */
  98     size_t num_dumpable;        /* number of page that can be dumped */
  99     uint32_t flag_compress;     /* indicate the compression format */
 100 } DumpState;
 101
 102 static int dump_cleanup(DumpState *s)
 103 {
 104     int ret = 0;
 105
 106     guest_phys_blocks_free(&s->guest_phys_blocks);
 107     memory_mapping_list_free(&s->list);
 108     if (s->fd != -1) {
 109         close(s->fd);
 110     }
 111     if (s->resume) {
 112         vm_start();
 113     }
 114
 115     return ret;
 116 }
 117
 118 static void dump_error(DumpState *s, const char *reason)
 119 {
 120     dump_cleanup(s);
 121 }
 122
 123 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
 124 {
 125     DumpState *s = opaque;
 126     size_t written_size;
 127
 128     written_size = qemu_write_full(s->fd, buf, size);
 129     if (written_size != size) {
 130         return -1;
 131     }
 132
 133     return 0;
 134 }
 135
 136 static int write_elf64_header(DumpState *s)
 137 {
 138     Elf64_Ehdr elf_header;
 139     int ret;
 140     int endian = s->dump_info.d_endian;
 141
 142     memset(&elf_header, 0, sizeof(Elf64_Ehdr));
 143     memcpy(&elf_header, ELFMAG, SELFMAG);
 144     elf_header.e_ident[EI_CLASS] = ELFCLASS64;
 145     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
 146     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
 147     elf_header.e_type = cpu_convert_to_target16(ET_CORE, endian);
 148     elf_header.e_machine = cpu_convert_to_target16(s->dump_info.d_machine,
 149                                                    endian);
 150     elf_header.e_version = cpu_convert_to_target32(EV_CURRENT, endian);
 151     elf_header.e_ehsize = cpu_convert_to_target16(sizeof(elf_header), endian);
 152     elf_header.e_phoff = cpu_convert_to_target64(sizeof(Elf64_Ehdr), endian);
 153     elf_header.e_phentsize = cpu_convert_to_target16(sizeof(Elf64_Phdr),
 154                                                      endian);
 155     elf_header.e_phnum = cpu_convert_to_target16(s->phdr_num, endian);
 156     if (s->have_section) {
 157         uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
 158
 159         elf_header.e_shoff = cpu_convert_to_target64(shoff, endian);
 160         elf_header.e_shentsize = cpu_convert_to_target16(sizeof(Elf64_Shdr),
 161                                                          endian);
 162         elf_header.e_shnum = cpu_convert_to_target16(1, endian);
 163     }
 164
 165     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
 166     if (ret < 0) {
 167         dump_error(s, "dump: failed to write elf header.\n");
 168         return -1;
 169     }
 170
 171     return 0;
 172 }
 173
 174 static int write_elf32_header(DumpState *s)
 175 {
 176     Elf32_Ehdr elf_header;
 177     int ret;
 178     int endian = s->dump_info.d_endian;
 179
 180     memset(&elf_header, 0, sizeof(Elf32_Ehdr));
 181     memcpy(&elf_header, ELFMAG, SELFMAG);
 182     elf_header.e_ident[EI_CLASS] = ELFCLASS32;
 183     elf_header.e_ident[EI_DATA] = endian;
 184     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
 185     elf_header.e_type = cpu_convert_to_target16(ET_CORE, endian);
 186     elf_header.e_machine = cpu_convert_to_target16(s->dump_info.d_machine,
 187                                                    endian);
 188     elf_header.e_version = cpu_convert_to_target32(EV_CURRENT, endian);
 189     elf_header.e_ehsize = cpu_convert_to_target16(sizeof(elf_header), endian);
 190     elf_header.e_phoff = cpu_convert_to_target32(sizeof(Elf32_Ehdr), endian);
 191     elf_header.e_phentsize = cpu_convert_to_target16(sizeof(Elf32_Phdr),
 192                                                      endian);
 193     elf_header.e_phnum = cpu_convert_to_target16(s->phdr_num, endian);
 194     if (s->have_section) {
 195         uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
 196
 197         elf_header.e_shoff = cpu_convert_to_target32(shoff, endian);
 198         elf_header.e_shentsize = cpu_convert_to_target16(sizeof(Elf32_Shdr),
 199                                                          endian);
 200         elf_header.e_shnum = cpu_convert_to_target16(1, endian);
 201     }
 202
 203     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
 204     if (ret < 0) {
 205         dump_error(s, "dump: failed to write elf header.\n");
 206         return -1;
 207     }
 208
 209     return 0;
 210 }
 211
 212 static int write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
 213                             int phdr_index, hwaddr offset,
 214                             hwaddr filesz)
 215 {
 216     Elf64_Phdr phdr;
 217     int ret;
 218     int endian = s->dump_info.d_endian;
 219
 220     memset(&phdr, 0, sizeof(Elf64_Phdr));
 221     phdr.p_type = cpu_convert_to_target32(PT_LOAD, endian);
 222     phdr.p_offset = cpu_convert_to_target64(offset, endian);
 223     phdr.p_paddr = cpu_convert_to_target64(memory_mapping->phys_addr, endian);
 224     phdr.p_filesz = cpu_convert_to_target64(filesz, endian);
 225     phdr.p_memsz = cpu_convert_to_target64(memory_mapping->length, endian);
 226     phdr.p_vaddr = cpu_convert_to_target64(memory_mapping->virt_addr, endian);
 227
 228     assert(memory_mapping->length >= filesz);
 229
 230     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
 231     if (ret < 0) {
 232         dump_error(s, "dump: failed to write program header table.\n");
 233         return -1;
 234     }
 235
 236     return 0;
 237 }
 238
 239 static int write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
 240                             int phdr_index, hwaddr offset,
 241                             hwaddr filesz)
 242 {
 243     Elf32_Phdr phdr;
 244     int ret;
 245     int endian = s->dump_info.d_endian;
 246
 247     memset(&phdr, 0, sizeof(Elf32_Phdr));
 248     phdr.p_type = cpu_convert_to_target32(PT_LOAD, endian);
 249     phdr.p_offset = cpu_convert_to_target32(offset, endian);
 250     phdr.p_paddr = cpu_convert_to_target32(memory_mapping->phys_addr, endian);
 251     phdr.p_filesz = cpu_convert_to_target32(filesz, endian);
 252     phdr.p_memsz = cpu_convert_to_target32(memory_mapping->length, endian);
 253     phdr.p_vaddr = cpu_convert_to_target32(memory_mapping->virt_addr, endian);
 254
 255     assert(memory_mapping->length >= filesz);
 256
 257     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
 258     if (ret < 0) {
 259         dump_error(s, "dump: failed to write program header table.\n");
 260         return -1;
 261     }
 262
 263     return 0;
 264 }
 265
 266 static int write_elf64_note(DumpState *s)
 267 {
 268     Elf64_Phdr phdr;
 269     int endian = s->dump_info.d_endian;
 270     hwaddr begin = s->memory_offset - s->note_size;
 271     int ret;
 272
 273     memset(&phdr, 0, sizeof(Elf64_Phdr));
 274     phdr.p_type = cpu_convert_to_target32(PT_NOTE, endian);
 275     phdr.p_offset = cpu_convert_to_target64(begin, endian);
 276     phdr.p_paddr = 0;
 277     phdr.p_filesz = cpu_convert_to_target64(s->note_size, endian);
 278     phdr.p_memsz = cpu_convert_to_target64(s->note_size, endian);
 279     phdr.p_vaddr = 0;
 280
 281     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
 282     if (ret < 0) {
 283         dump_error(s, "dump: failed to write program header table.\n");
 284         return -1;
 285     }
 286
 287     return 0;
 288 }
 289
 290 static inline int cpu_index(CPUState *cpu)
 291 {
 292     return cpu->cpu_index + 1;
 293 }
 294
 295 static int write_elf64_notes(WriteCoreDumpFunction f, DumpState *s)
 296 {
 297     CPUState *cpu;
 298     int ret;
 299     int id;
 300
 301     CPU_FOREACH(cpu) {
 302         id = cpu_index(cpu);
 303         ret = cpu_write_elf64_note(f, cpu, id, s);
 304         if (ret < 0) {
 305             dump_error(s, "dump: failed to write elf notes.\n");
 306             return -1;
 307         }
 308     }
 309
 310     CPU_FOREACH(cpu) {
 311         ret = cpu_write_elf64_qemunote(f, cpu, s);
 312         if (ret < 0) {
 313             dump_error(s, "dump: failed to write CPU status.\n");
 314             return -1;
 315         }
 316     }
 317
 318     return 0;
 319 }
 320
 321 static int write_elf32_note(DumpState *s)
 322 {
 323     hwaddr begin = s->memory_offset - s->note_size;
 324     Elf32_Phdr phdr;
 325     int endian = s->dump_info.d_endian;
 326     int ret;
 327
 328     memset(&phdr, 0, sizeof(Elf32_Phdr));
 329     phdr.p_type = cpu_convert_to_target32(PT_NOTE, endian);
 330     phdr.p_offset = cpu_convert_to_target32(begin, endian);
 331     phdr.p_paddr = 0;
 332     phdr.p_filesz = cpu_convert_to_target32(s->note_size, endian);
 333     phdr.p_memsz = cpu_convert_to_target32(s->note_size, endian);
 334     phdr.p_vaddr = 0;
 335
 336     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
 337     if (ret < 0) {
 338         dump_error(s, "dump: failed to write program header table.\n");
 339         return -1;
 340     }
 341
 342     return 0;
 343 }
 344
 345 static int write_elf32_notes(WriteCoreDumpFunction f, DumpState *s)
 346 {
 347     CPUState *cpu;
 348     int ret;
 349     int id;
 350
 351     CPU_FOREACH(cpu) {
 352         id = cpu_index(cpu);
 353         ret = cpu_write_elf32_note(f, cpu, id, s);
 354         if (ret < 0) {
 355             dump_error(s, "dump: failed to write elf notes.\n");
 356             return -1;
 357         }
 358     }
 359
 360     CPU_FOREACH(cpu) {
 361         ret = cpu_write_elf32_qemunote(f, cpu, s);
 362         if (ret < 0) {
 363             dump_error(s, "dump: failed to write CPU status.\n");
 364             return -1;
 365         }
 366     }
 367
 368     return 0;
 369 }
 370
 371 static int write_elf_section(DumpState *s, int type)
 372 {
 373     Elf32_Shdr shdr32;
 374     Elf64_Shdr shdr64;
 375     int endian = s->dump_info.d_endian;
 376     int shdr_size;
 377     void *shdr;
 378     int ret;
 379
 380     if (type == 0) {
 381         shdr_size = sizeof(Elf32_Shdr);
 382         memset(&shdr32, 0, shdr_size);
 383         shdr32.sh_info = cpu_convert_to_target32(s->sh_info, endian);
 384         shdr = &shdr32;
 385     } else {
 386         shdr_size = sizeof(Elf64_Shdr);
 387         memset(&shdr64, 0, shdr_size);
 388         shdr64.sh_info = cpu_convert_to_target32(s->sh_info, endian);
 389         shdr = &shdr64;
 390     }
 391
 392     ret = fd_write_vmcore(&shdr, shdr_size, s);
 393     if (ret < 0) {
 394         dump_error(s, "dump: failed to write section header table.\n");
 395         return -1;
 396     }
 397
 398     return 0;
 399 }
 400
 401 static int write_data(DumpState *s, void *buf, int length)
 402 {
 403     int ret;
 404
 405     ret = fd_write_vmcore(buf, length, s);
 406     if (ret < 0) {
 407         dump_error(s, "dump: failed to save memory.\n");
 408         return -1;
 409     }
 410
 411     return 0;
 412 }
 413
 414 /* write the memroy to vmcore. 1 page per I/O. */
 415 static int write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
 416                         int64_t size)
 417 {
 418     int64_t i;
 419     int ret;
 420
 421     for (i = 0; i < size / TARGET_PAGE_SIZE; i++) {
 422         ret = write_data(s, block->host_addr + start + i * TARGET_PAGE_SIZE,
 423                          TARGET_PAGE_SIZE);
 424         if (ret < 0) {
 425             return ret;
 426         }
 427     }
 428
 429     if ((size % TARGET_PAGE_SIZE) != 0) {
 430         ret = write_data(s, block->host_addr + start + i * TARGET_PAGE_SIZE,
 431                          size % TARGET_PAGE_SIZE);
 432         if (ret < 0) {
 433             return ret;
 434         }
 435     }
 436
 437     return 0;
 438 }
 439
 440 /* get the memory's offset and size in the vmcore */
 441 static void get_offset_range(hwaddr phys_addr,
 442                              ram_addr_t mapping_length,
 443                              DumpState *s,
 444                              hwaddr *p_offset,
 445                              hwaddr *p_filesz)
 446 {
 447     GuestPhysBlock *block;
 448     hwaddr offset = s->memory_offset;
 449     int64_t size_in_block, start;
 450
 451     /* When the memory is not stored into vmcore, offset will be -1 */
 452     *p_offset = -1;
 453     *p_filesz = 0;
 454
 455     if (s->has_filter) {
 456         if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
 457             return;
 458         }
 459     }
 460
 461     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
 462         if (s->has_filter) {
 463             if (block->target_start >= s->begin + s->length ||
 464                 block->target_end <= s->begin) {
 465                 /* This block is out of the range */
 466                 continue;
 467             }
 468
 469             if (s->begin <= block->target_start) {
 470                 start = block->target_start;
 471             } else {
 472                 start = s->begin;
 473             }
 474
 475             size_in_block = block->target_end - start;
 476             if (s->begin + s->length < block->target_end) {
 477                 size_in_block -= block->target_end - (s->begin + s->length);
 478             }
 479         } else {
 480             start = block->target_start;
 481             size_in_block = block->target_end - block->target_start;
 482         }
 483
 484         if (phys_addr >= start && phys_addr < start + size_in_block) {
 485             *p_offset = phys_addr - start + offset;
 486
 487             /* The offset range mapped from the vmcore file must not spill over
 488              * the GuestPhysBlock, clamp it. The rest of the mapping will be
 489              * zero-filled in memory at load time; see
 490              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
 491              */
 492             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
 493                         mapping_length :
 494                         size_in_block - (phys_addr - start);
 495             return;
 496         }
 497
 498         offset += size_in_block;
 499     }
 500 }
 501
 502 static int write_elf_loads(DumpState *s)
 503 {
 504     hwaddr offset, filesz;
 505     MemoryMapping *memory_mapping;
 506     uint32_t phdr_index = 1;
 507     int ret;
 508     uint32_t max_index;
 509
 510     if (s->have_section) {
 511         max_index = s->sh_info;
 512     } else {
 513         max_index = s->phdr_num;
 514     }
 515
 516     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
 517         get_offset_range(memory_mapping->phys_addr,
 518                          memory_mapping->length,
 519                          s, &offset, &filesz);
 520         if (s->dump_info.d_class == ELFCLASS64) {
 521             ret = write_elf64_load(s, memory_mapping, phdr_index++, offset,
 522                                    filesz);
 523         } else {
 524             ret = write_elf32_load(s, memory_mapping, phdr_index++, offset,
 525                                    filesz);
 526         }
 527
 528         if (ret < 0) {
 529             return -1;
 530         }
 531
 532         if (phdr_index >= max_index) {
 533             break;
 534         }
 535     }
 536
 537     return 0;
 538 }
 539
 540 /* write elf header, PT_NOTE and elf note to vmcore. */
 541 static int dump_begin(DumpState *s)
 542 {
 543     int ret;
 544
 545     /*
 546      * the vmcore's format is:
 547      *   --------------
 548      *   |  elf header |
 549      *   --------------
 550      *   |  PT_NOTE    |
 551      *   --------------
 552      *   |  PT_LOAD    |
 553      *   --------------
 554      *   |  ......     |
 555      *   --------------
 556      *   |  PT_LOAD    |
 557      *   --------------
 558      *   |  sec_hdr    |
 559      *   --------------
 560      *   |  elf note   |
 561      *   --------------
 562      *   |  memory     |
 563      *   --------------
 564      *
 565      * we only know where the memory is saved after we write elf note into
 566      * vmcore.
 567      */
 568
 569     /* write elf header to vmcore */
 570     if (s->dump_info.d_class == ELFCLASS64) {
 571         ret = write_elf64_header(s);
 572     } else {
 573         ret = write_elf32_header(s);
 574     }
 575     if (ret < 0) {
 576         return -1;
 577     }
 578
 579     if (s->dump_info.d_class == ELFCLASS64) {
 580         /* write PT_NOTE to vmcore */
 581         if (write_elf64_note(s) < 0) {
 582             return -1;
 583         }
 584
 585         /* write all PT_LOAD to vmcore */
 586         if (write_elf_loads(s) < 0) {
 587             return -1;
 588         }
 589
 590         /* write section to vmcore */
 591         if (s->have_section) {
 592             if (write_elf_section(s, 1) < 0) {
 593                 return -1;
 594             }
 595         }
 596
 597         /* write notes to vmcore */
 598         if (write_elf64_notes(fd_write_vmcore, s) < 0) {
 599             return -1;
 600         }
 601
 602     } else {
 603         /* write PT_NOTE to vmcore */
 604         if (write_elf32_note(s) < 0) {
 605             return -1;
 606         }
 607
 608         /* write all PT_LOAD to vmcore */
 609         if (write_elf_loads(s) < 0) {
 610             return -1;
 611         }
 612
 613         /* write section to vmcore */
 614         if (s->have_section) {
 615             if (write_elf_section(s, 0) < 0) {
 616                 return -1;
 617             }
 618         }
 619
 620         /* write notes to vmcore */
 621         if (write_elf32_notes(fd_write_vmcore, s) < 0) {
 622             return -1;
 623         }
 624     }
 625
 626     return 0;
 627 }
 628
 629 /* write PT_LOAD to vmcore */
 630 static int dump_completed(DumpState *s)
 631 {
 632     dump_cleanup(s);
 633     return 0;
 634 }
 635
 636 static int get_next_block(DumpState *s, GuestPhysBlock *block)
 637 {
 638     while (1) {
 639         block = QTAILQ_NEXT(block, next);
 640         if (!block) {
 641             /* no more block */
 642             return 1;
 643         }
 644
 645         s->start = 0;
 646         s->next_block = block;
 647         if (s->has_filter) {
 648             if (block->target_start >= s->begin + s->length ||
 649                 block->target_end <= s->begin) {
 650                 /* This block is out of the range */
 651                 continue;
 652             }
 653
 654             if (s->begin > block->target_start) {
 655                 s->start = s->begin - block->target_start;
 656             }
 657         }
 658
 659         return 0;
 660     }
 661 }
 662
 663 /* write all memory to vmcore */
 664 static int dump_iterate(DumpState *s)
 665 {
 666     GuestPhysBlock *block;
 667     int64_t size;
 668     int ret;
 669
 670     while (1) {
 671         block = s->next_block;
 672
 673         size = block->target_end - block->target_start;
 674         if (s->has_filter) {
 675             size -= s->start;
 676             if (s->begin + s->length < block->target_end) {
 677                 size -= block->target_end - (s->begin + s->length);
 678             }
 679         }
 680         ret = write_memory(s, block, s->start, size);
 681         if (ret == -1) {
 682             return ret;
 683         }
 684
 685         ret = get_next_block(s, block);
 686         if (ret == 1) {
 687             dump_completed(s);
 688             return 0;
 689         }
 690     }
 691 }
 692
 693 static int create_vmcore(DumpState *s)
 694 {
 695     int ret;
 696
 697     ret = dump_begin(s);
 698     if (ret < 0) {
 699         return -1;
 700     }
 701
 702     ret = dump_iterate(s);
 703     if (ret < 0) {
 704         return -1;
 705     }
 706
 707     return 0;
 708 }
 709
 710 static int write_start_flat_header(int fd)
 711 {
 712     uint8_t *buf;
 713     MakedumpfileHeader mh;
 714     int ret = 0;
 715
 716     memset(&mh, 0, sizeof(mh));
 717     strncpy(mh.signature, MAKEDUMPFILE_SIGNATURE,
 718             strlen(MAKEDUMPFILE_SIGNATURE));
 719
 720     mh.type = cpu_to_be64(TYPE_FLAT_HEADER);
 721     mh.version = cpu_to_be64(VERSION_FLAT_HEADER);
 722
 723     buf = g_malloc0(MAX_SIZE_MDF_HEADER);
 724     memcpy(buf, &mh, sizeof(mh));
 725
 726     size_t written_size;
 727     written_size = qemu_write_full(fd, buf, MAX_SIZE_MDF_HEADER);
 728     if (written_size != MAX_SIZE_MDF_HEADER) {
 729         ret = -1;
 730     }
 731
 732     g_free(buf);
 733     return ret;
 734 }
 735
 736 static int write_end_flat_header(int fd)
 737 {
 738     MakedumpfileDataHeader mdh;
 739
 740     mdh.offset = END_FLAG_FLAT_HEADER;
 741     mdh.buf_size = END_FLAG_FLAT_HEADER;
 742
 743     size_t written_size;
 744     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 745     if (written_size != sizeof(mdh)) {
 746         return -1;
 747     }
 748
 749     return 0;
 750 }
 751
 752 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
 753 {
 754     size_t written_size;
 755     MakedumpfileDataHeader mdh;
 756
 757     mdh.offset = cpu_to_be64(offset);
 758     mdh.buf_size = cpu_to_be64(size);
 759
 760     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 761     if (written_size != sizeof(mdh)) {
 762         return -1;
 763     }
 764
 765     written_size = qemu_write_full(fd, buf, size);
 766     if (written_size != size) {
 767         return -1;
 768     }
 769
 770     return 0;
 771 }
 772
 773 static int buf_write_note(const void *buf, size_t size, void *opaque)
 774 {
 775     DumpState *s = opaque;
 776
 777     /* note_buf is not enough */
 778     if (s->note_buf_offset + size > s->note_size) {
 779         return -1;
 780     }
 781
 782     memcpy(s->note_buf + s->note_buf_offset, buf, size);
 783
 784     s->note_buf_offset += size;
 785
 786     return 0;
 787 }
 788
 789 /* write common header, sub header and elf note to vmcore */
 790 static int create_header32(DumpState *s)
 791 {
 792     int ret = 0;
 793     DiskDumpHeader32 *dh = NULL;
 794     KdumpSubHeader32 *kh = NULL;
 795     size_t size;
 796     int endian = s->dump_info.d_endian;
 797     uint32_t block_size;
 798     uint32_t sub_hdr_size;
 799     uint32_t bitmap_blocks;
 800     uint32_t status = 0;
 801     uint64_t offset_note;
 802
 803     /* write common header, the version of kdump-compressed format is 6th */
 804     size = sizeof(DiskDumpHeader32);
 805     dh = g_malloc0(size);
 806
 807     strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
 808     dh->header_version = cpu_convert_to_target32(6, endian);
 809     block_size = s->page_size;
 810     dh->block_size = cpu_convert_to_target32(block_size, endian);
 811     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
 812     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 813     dh->sub_hdr_size = cpu_convert_to_target32(sub_hdr_size, endian);
 814     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 815     dh->max_mapnr = cpu_convert_to_target32(MIN(s->max_mapnr, UINT_MAX),
 816                                             endian);
 817     dh->nr_cpus = cpu_convert_to_target32(s->nr_cpus, endian);
 818     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 819     dh->bitmap_blocks = cpu_convert_to_target32(bitmap_blocks, endian);
 820     memcpy(&(dh->utsname.machine), "i686", 4);
 821
 822     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 823         status |= DUMP_DH_COMPRESSED_ZLIB;
 824     }
 825 #ifdef CONFIG_LZO
 826     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 827         status |= DUMP_DH_COMPRESSED_LZO;
 828     }
 829 #endif
 830 #ifdef CONFIG_SNAPPY
 831     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 832         status |= DUMP_DH_COMPRESSED_SNAPPY;
 833     }
 834 #endif
 835     dh->status = cpu_convert_to_target32(status, endian);
 836
 837     if (write_buffer(s->fd, 0, dh, size) < 0) {
 838         dump_error(s, "dump: failed to write disk dump header.\n");
 839         ret = -1;
 840         goto out;
 841     }
 842
 843     /* write sub header */
 844     size = sizeof(KdumpSubHeader32);
 845     kh = g_malloc0(size);
 846
 847     /* 64bit max_mapnr_64 */
 848     kh->max_mapnr_64 = cpu_convert_to_target64(s->max_mapnr, endian);
 849     kh->phys_base = cpu_convert_to_target32(PHYS_BASE, endian);
 850     kh->dump_level = cpu_convert_to_target32(DUMP_LEVEL, endian);
 851
 852     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 853     kh->offset_note = cpu_convert_to_target64(offset_note, endian);
 854     kh->note_size = cpu_convert_to_target32(s->note_size, endian);
 855
 856     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 857                      block_size, kh, size) < 0) {
 858         dump_error(s, "dump: failed to write kdump sub header.\n");
 859         ret = -1;
 860         goto out;
 861     }
 862
 863     /* write note */
 864     s->note_buf = g_malloc0(s->note_size);
 865     s->note_buf_offset = 0;
 866
 867     /* use s->note_buf to store notes temporarily */
 868     if (write_elf32_notes(buf_write_note, s) < 0) {
 869         ret = -1;
 870         goto out;
 871     }
 872
 873     if (write_buffer(s->fd, offset_note, s->note_buf,
 874                      s->note_size) < 0) {
 875         dump_error(s, "dump: failed to write notes");
 876         ret = -1;
 877         goto out;
 878     }
 879
 880     /* get offset of dump_bitmap */
 881     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 882                              block_size;
 883
 884     /* get offset of page */
 885     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 886                      block_size;
 887
 888 out:
 889     g_free(dh);
 890     g_free(kh);
 891     g_free(s->note_buf);
 892
 893     return ret;
 894 }
 895
 896 /* write common header, sub header and elf note to vmcore */
 897 static int create_header64(DumpState *s)
 898 {
 899     int ret = 0;
 900     DiskDumpHeader64 *dh = NULL;
 901     KdumpSubHeader64 *kh = NULL;
 902     size_t size;
 903     int endian = s->dump_info.d_endian;
 904     uint32_t block_size;
 905     uint32_t sub_hdr_size;
 906     uint32_t bitmap_blocks;
 907     uint32_t status = 0;
 908     uint64_t offset_note;
 909
 910     /* write common header, the version of kdump-compressed format is 6th */
 911     size = sizeof(DiskDumpHeader64);
 912     dh = g_malloc0(size);
 913
 914     strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
 915     dh->header_version = cpu_convert_to_target32(6, endian);
 916     block_size = s->page_size;
 917     dh->block_size = cpu_convert_to_target32(block_size, endian);
 918     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
 919     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 920     dh->sub_hdr_size = cpu_convert_to_target32(sub_hdr_size, endian);
 921     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 922     dh->max_mapnr = cpu_convert_to_target32(MIN(s->max_mapnr, UINT_MAX),
 923                                             endian);
 924     dh->nr_cpus = cpu_convert_to_target32(s->nr_cpus, endian);
 925     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 926     dh->bitmap_blocks = cpu_convert_to_target32(bitmap_blocks, endian);
 927     memcpy(&(dh->utsname.machine), "x86_64", 6);
 928
 929     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 930         status |= DUMP_DH_COMPRESSED_ZLIB;
 931     }
 932 #ifdef CONFIG_LZO
 933     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 934         status |= DUMP_DH_COMPRESSED_LZO;
 935     }
 936 #endif
 937 #ifdef CONFIG_SNAPPY
 938     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 939         status |= DUMP_DH_COMPRESSED_SNAPPY;
 940     }
 941 #endif
 942     dh->status = cpu_convert_to_target32(status, endian);
 943
 944     if (write_buffer(s->fd, 0, dh, size) < 0) {
 945         dump_error(s, "dump: failed to write disk dump header.\n");
 946         ret = -1;
 947         goto out;
 948     }
 949
 950     /* write sub header */
 951     size = sizeof(KdumpSubHeader64);
 952     kh = g_malloc0(size);
 953
 954     /* 64bit max_mapnr_64 */
 955     kh->max_mapnr_64 = cpu_convert_to_target64(s->max_mapnr, endian);
 956     kh->phys_base = cpu_convert_to_target64(PHYS_BASE, endian);
 957     kh->dump_level = cpu_convert_to_target32(DUMP_LEVEL, endian);
 958
 959     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 960     kh->offset_note = cpu_convert_to_target64(offset_note, endian);
 961     kh->note_size = cpu_convert_to_target64(s->note_size, endian);
 962
 963     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 964                      block_size, kh, size) < 0) {
 965         dump_error(s, "dump: failed to write kdump sub header.\n");
 966         ret = -1;
 967         goto out;
 968     }
 969
 970     /* write note */
 971     s->note_buf = g_malloc0(s->note_size);
 972     s->note_buf_offset = 0;
 973
 974     /* use s->note_buf to store notes temporarily */
 975     if (write_elf64_notes(buf_write_note, s) < 0) {
 976         ret = -1;
 977         goto out;
 978     }
 979
 980     if (write_buffer(s->fd, offset_note, s->note_buf,
 981                      s->note_size) < 0) {
 982         dump_error(s, "dump: failed to write notes");
 983         ret = -1;
 984         goto out;
 985     }
 986
 987     /* get offset of dump_bitmap */
 988     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 989                              block_size;
 990
 991     /* get offset of page */
 992     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 993                      block_size;
 994
 995 out:
 996     g_free(dh);
 997     g_free(kh);
 998     g_free(s->note_buf);
 999
1000     return ret;
1001 }
1002
1003 static int write_dump_header(DumpState *s)
1004 {
1005     if (s->dump_info.d_machine == EM_386) {
1006         return create_header32(s);
1007     } else {
1008         return create_header64(s);
1009     }
1010 }
1011
1012 /*
1013  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1014  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1015  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1016  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1017  * vmcore, ie. synchronizing un-sync bit into vmcore.
1018  */
1019 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1020                            uint8_t *buf, DumpState *s)
1021 {
1022     off_t old_offset, new_offset;
1023     off_t offset_bitmap1, offset_bitmap2;
1024     uint32_t byte, bit;
1025
1026     /* should not set the previous place */
1027     assert(last_pfn <= pfn);
1028
1029     /*
1030      * if the bit needed to be set is not cached in buf, flush the data in buf
1031      * to vmcore firstly.
1032      * making new_offset be bigger than old_offset can also sync remained data
1033      * into vmcore.
1034      */
1035     old_offset = BUFSIZE_BITMAP * (last_pfn / PFN_BUFBITMAP);
1036     new_offset = BUFSIZE_BITMAP * (pfn / PFN_BUFBITMAP);
1037
1038     while (old_offset < new_offset) {
1039         /* calculate the offset and write dump_bitmap */
1040         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1041         if (write_buffer(s->fd, offset_bitmap1, buf,
1042                          BUFSIZE_BITMAP) < 0) {
1043             return -1;
1044         }
1045
1046         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1047         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1048                          old_offset;
1049         if (write_buffer(s->fd, offset_bitmap2, buf,
1050                          BUFSIZE_BITMAP) < 0) {
1051             return -1;
1052         }
1053
1054         memset(buf, 0, BUFSIZE_BITMAP);
1055         old_offset += BUFSIZE_BITMAP;
1056     }
1057
1058     /* get the exact place of the bit in the buf, and set it */
1059     byte = (pfn % PFN_BUFBITMAP) / CHAR_BIT;
1060     bit = (pfn % PFN_BUFBITMAP) % CHAR_BIT;
1061     if (value) {
1062         buf[byte] |= 1u << bit;
1063     } else {
1064         buf[byte] &= ~(1u << bit);
1065     }
1066
1067     return 0;
1068 }
1069
1070 /*
1071  * exam every page and return the page frame number and the address of the page.
1072  * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1073  * blocks, so block->target_start and block->target_end should be interal
1074  * multiples of the target page size.
1075  */
1076 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1077                           uint8_t **bufptr, DumpState *s)
1078 {
1079     GuestPhysBlock *block = *blockptr;
1080     hwaddr addr;
1081     uint8_t *buf;
1082
1083     /* block == NULL means the start of the iteration */
1084     if (!block) {
1085         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1086         *blockptr = block;
1087         assert(block->target_start % s->page_size == 0);
1088         assert(block->target_end % s->page_size == 0);
1089         *pfnptr = paddr_to_pfn(block->target_start, s->page_shift);
1090         if (bufptr) {
1091             *bufptr = block->host_addr;
1092         }
1093         return true;
1094     }
1095
1096     *pfnptr = *pfnptr + 1;
1097     addr = pfn_to_paddr(*pfnptr, s->page_shift);
1098
1099     if ((addr >= block->target_start) &&
1100         (addr + s->page_size <= block->target_end)) {
1101         buf = block->host_addr + (addr - block->target_start);
1102     } else {
1103         /* the next page is in the next block */
1104         block = QTAILQ_NEXT(block, next);
1105         *blockptr = block;
1106         if (!block) {
1107             return false;
1108         }
1109         assert(block->target_start % s->page_size == 0);
1110         assert(block->target_end % s->page_size == 0);
1111         *pfnptr = paddr_to_pfn(block->target_start, s->page_shift);
1112         buf = block->host_addr;
1113     }
1114
1115     if (bufptr) {
1116         *bufptr = buf;
1117     }
1118
1119     return true;
1120 }
1121
1122 static int write_dump_bitmap(DumpState *s)
1123 {
1124     int ret = 0;
1125     uint64_t last_pfn, pfn;
1126     void *dump_bitmap_buf;
1127     size_t num_dumpable;
1128     GuestPhysBlock *block_iter = NULL;
1129
1130     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1131     dump_bitmap_buf = g_malloc0(BUFSIZE_BITMAP);
1132
1133     num_dumpable = 0;
1134     last_pfn = 0;
1135
1136     /*
1137      * exam memory page by page, and set the bit in dump_bitmap corresponded
1138      * to the existing page.
1139      */
1140     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1141         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1142         if (ret < 0) {
1143             dump_error(s, "dump: failed to set dump_bitmap.\n");
1144             ret = -1;
1145             goto out;
1146         }
1147
1148         last_pfn = pfn;
1149         num_dumpable++;
1150     }
1151
1152     /*
1153      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1154      * set last_pfn + PFN_BUFBITMAP to 0 and those set but un-sync bit will be
1155      * synchronized into vmcore.
1156      */
1157     if (num_dumpable > 0) {
1158         ret = set_dump_bitmap(last_pfn, last_pfn + PFN_BUFBITMAP, false,
1159                               dump_bitmap_buf, s);
1160         if (ret < 0) {
1161             dump_error(s, "dump: failed to sync dump_bitmap.\n");
1162             ret = -1;
1163             goto out;
1164         }
1165     }
1166
1167     /* number of dumpable pages that will be dumped later */
1168     s->num_dumpable = num_dumpable;
1169
1170 out:
1171     g_free(dump_bitmap_buf);
1172
1173     return ret;
1174 }
1175
1176 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1177                                off_t offset)
1178 {
1179     data_cache->fd = s->fd;
1180     data_cache->data_size = 0;
1181     data_cache->buf_size = BUFSIZE_DATA_CACHE;
1182     data_cache->buf = g_malloc0(BUFSIZE_DATA_CACHE);
1183     data_cache->offset = offset;
1184 }
1185
1186 static int write_cache(DataCache *dc, const void *buf, size_t size,
1187                        bool flag_sync)
1188 {
1189     /*
1190      * dc->buf_size should not be less than size, otherwise dc will never be
1191      * enough
1192      */
1193     assert(size <= dc->buf_size);
1194
1195     /*
1196      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1197      * otherwise check if the space is enough for caching data in buf, if not,
1198      * write the data in dc->buf to dc->fd and reset dc->buf
1199      */
1200     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1201         (flag_sync && dc->data_size > 0)) {
1202         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1203             return -1;
1204         }
1205
1206         dc->offset += dc->data_size;
1207         dc->data_size = 0;
1208     }
1209
1210     if (!flag_sync) {
1211         memcpy(dc->buf + dc->data_size, buf, size);
1212         dc->data_size += size;
1213     }
1214
1215     return 0;
1216 }
1217
1218 static void free_data_cache(DataCache *data_cache)
1219 {
1220     g_free(data_cache->buf);
1221 }
1222
1223 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1224 {
1225     size_t len_buf_out_zlib, len_buf_out_lzo, len_buf_out_snappy;
1226     size_t len_buf_out;
1227
1228     /* init buf_out */
1229     len_buf_out_zlib = len_buf_out_lzo = len_buf_out_snappy = 0;
1230
1231     /* buf size for zlib */
1232     len_buf_out_zlib = compressBound(page_size);
1233
1234     /* buf size for lzo */
1235 #ifdef CONFIG_LZO
1236     if (flag_compress & DUMP_DH_COMPRESSED_LZO) {
1237         if (lzo_init() != LZO_E_OK) {
1238             /* return 0 to indicate lzo is unavailable */
1239             return 0;
1240         }
1241     }
1242
1243     /*
1244      * LZO will expand incompressible data by a little amount. please check the
1245      * following URL to see the expansion calculation:
1246      * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1247      */
1248     len_buf_out_lzo = page_size + page_size / 16 + 64 + 3;
1249 #endif
1250
1251 #ifdef CONFIG_SNAPPY
1252     /* buf size for snappy */
1253     len_buf_out_snappy = snappy_max_compressed_length(page_size);
1254 #endif
1255
1256     /* get the biggest that can store all kinds of compressed page */
1257     len_buf_out = MAX(len_buf_out_zlib,
1258                       MAX(len_buf_out_lzo, len_buf_out_snappy));
1259
1260     return len_buf_out;
1261 }
1262
1263 /*
1264  * check if the page is all 0
1265  */
1266 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1267 {
1268     return buffer_is_zero(buf, page_size);
1269 }
1270
1271 static int write_dump_pages(DumpState *s)
1272 {
1273     int ret = 0;
1274     DataCache page_desc, page_data;
1275     size_t len_buf_out, size_out;
1276 #ifdef CONFIG_LZO
1277     lzo_bytep wrkmem = NULL;
1278 #endif
1279     uint8_t *buf_out = NULL;
1280     off_t offset_desc, offset_data;
1281     PageDescriptor pd, pd_zero;
1282     uint8_t *buf;
1283     int endian = s->dump_info.d_endian;
1284     GuestPhysBlock *block_iter = NULL;
1285     uint64_t pfn_iter;
1286
1287     /* get offset of page_desc and page_data in dump file */
1288     offset_desc = s->offset_page;
1289     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1290
1291     prepare_data_cache(&page_desc, s, offset_desc);
1292     prepare_data_cache(&page_data, s, offset_data);
1293
1294     /* prepare buffer to store compressed data */
1295     len_buf_out = get_len_buf_out(s->page_size, s->flag_compress);
1296     if (len_buf_out == 0) {
1297         dump_error(s, "dump: failed to get length of output buffer.\n");
1298         goto out;
1299     }
1300
1301 #ifdef CONFIG_LZO
1302     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1303 #endif
1304
1305     buf_out = g_malloc(len_buf_out);
1306
1307     /*
1308      * init zero page's page_desc and page_data, because every zero page
1309      * uses the same page_data
1310      */
1311     pd_zero.size = cpu_convert_to_target32(s->page_size, endian);
1312     pd_zero.flags = cpu_convert_to_target32(0, endian);
1313     pd_zero.offset = cpu_convert_to_target64(offset_data, endian);
1314     pd_zero.page_flags = cpu_convert_to_target64(0, endian);
1315     buf = g_malloc0(s->page_size);
1316     ret = write_cache(&page_data, buf, s->page_size, false);
1317     g_free(buf);
1318     if (ret < 0) {
1319         dump_error(s, "dump: failed to write page data(zero page).\n");
1320         goto out;
1321     }
1322
1323     offset_data += s->page_size;
1324
1325     /*
1326      * dump memory to vmcore page by page. zero page will all be resided in the
1327      * first page of page section
1328      */
1329     while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1330         /* check zero page */
1331         if (is_zero_page(buf, s->page_size)) {
1332             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1333                               false);
1334             if (ret < 0) {
1335                 dump_error(s, "dump: failed to write page desc.\n");
1336                 goto out;
1337             }
1338         } else {
1339             /*
1340              * not zero page, then:
1341              * 1. compress the page
1342              * 2. write the compressed page into the cache of page_data
1343              * 3. get page desc of the compressed page and write it into the
1344              *    cache of page_desc
1345              *
1346              * only one compression format will be used here, for
1347              * s->flag_compress is set. But when compression fails to work,
1348              * we fall back to save in plaintext.
1349              */
1350              size_out = len_buf_out;
1351              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1352                     (compress2(buf_out, (uLongf *)&size_out, buf, s->page_size,
1353                     Z_BEST_SPEED) == Z_OK) && (size_out < s->page_size)) {
1354                 pd.flags = cpu_convert_to_target32(DUMP_DH_COMPRESSED_ZLIB,
1355                                                    endian);
1356                 pd.size  = cpu_convert_to_target32(size_out, endian);
1357
1358                 ret = write_cache(&page_data, buf_out, size_out, false);
1359                 if (ret < 0) {
1360                     dump_error(s, "dump: failed to write page data.\n");
1361                     goto out;
1362                 }
1363 #ifdef CONFIG_LZO
1364             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1365                     (lzo1x_1_compress(buf, s->page_size, buf_out,
1366                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1367                     (size_out < s->page_size)) {
1368                 pd.flags = cpu_convert_to_target32(DUMP_DH_COMPRESSED_LZO,
1369                                                    endian);
1370                 pd.size  = cpu_convert_to_target32(size_out, endian);
1371
1372                 ret = write_cache(&page_data, buf_out, size_out, false);
1373                 if (ret < 0) {
1374                     dump_error(s, "dump: failed to write page data.\n");
1375                     goto out;
1376                 }
1377 #endif
1378 #ifdef CONFIG_SNAPPY
1379             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1380                     (snappy_compress((char *)buf, s->page_size,
1381                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1382                     (size_out < s->page_size)) {
1383                 pd.flags = cpu_convert_to_target32(
1384                                         DUMP_DH_COMPRESSED_SNAPPY, endian);
1385                 pd.size  = cpu_convert_to_target32(size_out, endian);
1386
1387                 ret = write_cache(&page_data, buf_out, size_out, false);
1388                 if (ret < 0) {
1389                     dump_error(s, "dump: failed to write page data.\n");
1390                     goto out;
1391                 }
1392 #endif
1393             } else {
1394                 /*
1395                  * fall back to save in plaintext, size_out should be
1396                  * assigned to s->page_size
1397                  */
1398                 pd.flags = cpu_convert_to_target32(0, endian);
1399                 size_out = s->page_size;
1400                 pd.size = cpu_convert_to_target32(size_out, endian);
1401
1402                 ret = write_cache(&page_data, buf, s->page_size, false);
1403                 if (ret < 0) {
1404                     dump_error(s, "dump: failed to write page data.\n");
1405                     goto out;
1406                 }
1407             }
1408
1409             /* get and write page desc here */
1410             pd.page_flags = cpu_convert_to_target64(0, endian);
1411             pd.offset = cpu_convert_to_target64(offset_data, endian);
1412             offset_data += size_out;
1413
1414             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1415             if (ret < 0) {
1416                 dump_error(s, "dump: failed to write page desc.\n");
1417                 goto out;
1418             }
1419         }
1420     }
1421
1422     ret = write_cache(&page_desc, NULL, 0, true);
1423     if (ret < 0) {
1424         dump_error(s, "dump: failed to sync cache for page_desc.\n");
1425         goto out;
1426     }
1427     ret = write_cache(&page_data, NULL, 0, true);
1428     if (ret < 0) {
1429         dump_error(s, "dump: failed to sync cache for page_data.\n");
1430         goto out;
1431     }
1432
1433 out:
1434     free_data_cache(&page_desc);
1435     free_data_cache(&page_data);
1436
1437 #ifdef CONFIG_LZO
1438     g_free(wrkmem);
1439 #endif
1440
1441     g_free(buf_out);
1442
1443     return ret;
1444 }
1445
1446 static int create_kdump_vmcore(DumpState *s)
1447 {
1448     int ret;
1449
1450     /*
1451      * the kdump-compressed format is:
1452      *                                               File offset
1453      *  +------------------------------------------+ 0x0
1454      *  |    main header (struct disk_dump_header) |
1455      *  |------------------------------------------+ block 1
1456      *  |    sub header (struct kdump_sub_header)  |
1457      *  |------------------------------------------+ block 2
1458      *  |            1st-dump_bitmap               |
1459      *  |------------------------------------------+ block 2 + X blocks
1460      *  |            2nd-dump_bitmap               | (aligned by block)
1461      *  |------------------------------------------+ block 2 + 2 * X blocks
1462      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1463      *  |  page desc for pfn 1 (struct page_desc)  |
1464      *  |                    :                     |
1465      *  |------------------------------------------| (not aligned by block)
1466      *  |         page data (pfn 0)                |
1467      *  |         page data (pfn 1)                |
1468      *  |                    :                     |
1469      *  +------------------------------------------+
1470      */
1471
1472     ret = write_start_flat_header(s->fd);
1473     if (ret < 0) {
1474         dump_error(s, "dump: failed to write start flat header.\n");
1475         return -1;
1476     }
1477
1478     ret = write_dump_header(s);
1479     if (ret < 0) {
1480         return -1;
1481     }
1482
1483     ret = write_dump_bitmap(s);
1484     if (ret < 0) {
1485         return -1;
1486     }
1487
1488     ret = write_dump_pages(s);
1489     if (ret < 0) {
1490         return -1;
1491     }
1492
1493     ret = write_end_flat_header(s->fd);
1494     if (ret < 0) {
1495         dump_error(s, "dump: failed to write end flat header.\n");
1496         return -1;
1497     }
1498
1499     dump_completed(s);
1500
1501     return 0;
1502 }
1503
1504 static ram_addr_t get_start_block(DumpState *s)
1505 {
1506     GuestPhysBlock *block;
1507
1508     if (!s->has_filter) {
1509         s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1510         return 0;
1511     }
1512
1513     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1514         if (block->target_start >= s->begin + s->length ||
1515             block->target_end <= s->begin) {
1516             /* This block is out of the range */
1517             continue;
1518         }
1519
1520         s->next_block = block;
1521         if (s->begin > block->target_start) {
1522             s->start = s->begin - block->target_start;
1523         } else {
1524             s->start = 0;
1525         }
1526         return s->start;
1527     }
1528
1529     return -1;
1530 }
1531
1532 static void get_max_mapnr(DumpState *s)
1533 {
1534     GuestPhysBlock *last_block;
1535
1536     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head, GuestPhysBlockHead);
1537     s->max_mapnr = paddr_to_pfn(last_block->target_end, s->page_shift);
1538 }
1539
1540 static int dump_init(DumpState *s, int fd, bool has_format,
1541                      DumpGuestMemoryFormat format, bool paging, bool has_filter,
1542                      int64_t begin, int64_t length, Error **errp)
1543 {
1544     CPUState *cpu;
1545     int nr_cpus;
1546     Error *err = NULL;
1547     int ret;
1548
1549     /* kdump-compressed is conflict with paging and filter */
1550     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1551         assert(!paging && !has_filter);
1552     }
1553
1554     if (runstate_is_running()) {
1555         vm_stop(RUN_STATE_SAVE_VM);
1556         s->resume = true;
1557     } else {
1558         s->resume = false;
1559     }
1560
1561     /* If we use KVM, we should synchronize the registers before we get dump
1562      * info or physmap info.
1563      */
1564     cpu_synchronize_all_states();
1565     nr_cpus = 0;
1566     CPU_FOREACH(cpu) {
1567         nr_cpus++;
1568     }
1569
1570     s->errp = errp;
1571     s->fd = fd;
1572     s->has_filter = has_filter;
1573     s->begin = begin;
1574     s->length = length;
1575
1576     guest_phys_blocks_init(&s->guest_phys_blocks);
1577     guest_phys_blocks_append(&s->guest_phys_blocks);
1578
1579     s->start = get_start_block(s);
1580     if (s->start == -1) {
1581         error_set(errp, QERR_INVALID_PARAMETER, "begin");
1582         goto cleanup;
1583     }
1584
1585     /* get dump info: endian, class and architecture.
1586      * If the target architecture is not supported, cpu_get_dump_info() will
1587      * return -1.
1588      */
1589     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1590     if (ret < 0) {
1591         error_set(errp, QERR_UNSUPPORTED);
1592         goto cleanup;
1593     }
1594
1595     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1596                                      s->dump_info.d_machine, nr_cpus);
1597     if (s->note_size < 0) {
1598         error_set(errp, QERR_UNSUPPORTED);
1599         goto cleanup;
1600     }
1601
1602     /* get memory mapping */
1603     memory_mapping_list_init(&s->list);
1604     if (paging) {
1605         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1606         if (err != NULL) {
1607             error_propagate(errp, err);
1608             goto cleanup;
1609         }
1610     } else {
1611         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1612     }
1613
1614     s->nr_cpus = nr_cpus;
1615     s->page_size = TARGET_PAGE_SIZE;
1616     s->page_shift = ffs(s->page_size) - 1;
1617
1618     get_max_mapnr(s);
1619
1620     uint64_t tmp;
1621     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT), s->page_size);
1622     s->len_dump_bitmap = tmp * s->page_size;
1623
1624     /* init for kdump-compressed format */
1625     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1626         switch (format) {
1627         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1628             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1629             break;
1630
1631         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1632             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1633             break;
1634
1635         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1636             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1637             break;
1638
1639         default:
1640             s->flag_compress = 0;
1641         }
1642
1643         return 0;
1644     }
1645
1646     if (s->has_filter) {
1647         memory_mapping_filter(&s->list, s->begin, s->length);
1648     }
1649
1650     /*
1651      * calculate phdr_num
1652      *
1653      * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1654      */
1655     s->phdr_num = 1; /* PT_NOTE */
1656     if (s->list.num < UINT16_MAX - 2) {
1657         s->phdr_num += s->list.num;
1658         s->have_section = false;
1659     } else {
1660         s->have_section = true;
1661         s->phdr_num = PN_XNUM;
1662         s->sh_info = 1; /* PT_NOTE */
1663
1664         /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1665         if (s->list.num <= UINT32_MAX - 1) {
1666             s->sh_info += s->list.num;
1667         } else {
1668             s->sh_info = UINT32_MAX;
1669         }
1670     }
1671
1672     if (s->dump_info.d_class == ELFCLASS64) {
1673         if (s->have_section) {
1674             s->memory_offset = sizeof(Elf64_Ehdr) +
1675                                sizeof(Elf64_Phdr) * s->sh_info +
1676                                sizeof(Elf64_Shdr) + s->note_size;
1677         } else {
1678             s->memory_offset = sizeof(Elf64_Ehdr) +
1679                                sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1680         }
1681     } else {
1682         if (s->have_section) {
1683             s->memory_offset = sizeof(Elf32_Ehdr) +
1684                                sizeof(Elf32_Phdr) * s->sh_info +
1685                                sizeof(Elf32_Shdr) + s->note_size;
1686         } else {
1687             s->memory_offset = sizeof(Elf32_Ehdr) +
1688                                sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1689         }
1690     }
1691
1692     return 0;
1693
1694 cleanup:
1695     guest_phys_blocks_free(&s->guest_phys_blocks);
1696
1697     if (s->resume) {
1698         vm_start();
1699     }
1700
1701     return -1;
1702 }
1703
1704 void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
1705                            int64_t begin, bool has_length,
1706                            int64_t length, bool has_format,
1707                            DumpGuestMemoryFormat format, Error **errp)
1708 {
1709     const char *p;
1710     int fd = -1;
1711     DumpState *s;
1712     int ret;
1713
1714     /*
1715      * kdump-compressed format need the whole memory dumped, so paging or
1716      * filter is not supported here.
1717      */
1718     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1719         (paging || has_begin || has_length)) {
1720         error_setg(errp, "kdump-compressed format doesn't support paging or "
1721                          "filter");
1722         return;
1723     }
1724     if (has_begin && !has_length) {
1725         error_set(errp, QERR_MISSING_PARAMETER, "length");
1726         return;
1727     }
1728     if (!has_begin && has_length) {
1729         error_set(errp, QERR_MISSING_PARAMETER, "begin");
1730         return;
1731     }
1732
1733     /* check whether lzo/snappy is supported */
1734 #ifndef CONFIG_LZO
1735     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1736         error_setg(errp, "kdump-lzo is not available now");
1737         return;
1738     }
1739 #endif
1740
1741 #ifndef CONFIG_SNAPPY
1742     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1743         error_setg(errp, "kdump-snappy is not available now");
1744         return;
1745     }
1746 #endif
1747
1748 #if !defined(WIN32)
1749     if (strstart(file, "fd:", &p)) {
1750         fd = monitor_get_fd(cur_mon, p, errp);
1751         if (fd == -1) {
1752             return;
1753         }
1754     }
1755 #endif
1756
1757     if  (strstart(file, "file:", &p)) {
1758         fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1759         if (fd < 0) {
1760             error_setg_file_open(errp, errno, p);
1761             return;
1762         }
1763     }
1764
1765     if (fd == -1) {
1766         error_set(errp, QERR_INVALID_PARAMETER, "protocol");
1767         return;
1768     }
1769
1770     s = g_malloc0(sizeof(DumpState));
1771
1772     ret = dump_init(s, fd, has_format, format, paging, has_begin,
1773                     begin, length, errp);
1774     if (ret < 0) {
1775         g_free(s);
1776         return;
1777     }
1778
1779     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1780         if (create_kdump_vmcore(s) < 0 && !error_is_set(s->errp)) {
1781             error_set(errp, QERR_IO_ERROR);
1782         }
1783     } else {
1784         if (create_vmcore(s) < 0 && !error_is_set(s->errp)) {
1785             error_set(errp, QERR_IO_ERROR);
1786         }
1787     }
1788
1789     g_free(s);
1790 }