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 #ifndef ELF_MACHINE_UNAME
  36 #define ELF_MACHINE_UNAME "Unknown"
  37 #endif
  38
  39 static uint16_t cpu_convert_to_target16(uint16_t val, int endian)
  40 {
  41     if (endian == ELFDATA2LSB) {
  42         val = cpu_to_le16(val);
  43     } else {
  44         val = cpu_to_be16(val);
  45     }
  46
  47     return val;
  48 }
  49
  50 static uint32_t cpu_convert_to_target32(uint32_t val, int endian)
  51 {
  52     if (endian == ELFDATA2LSB) {
  53         val = cpu_to_le32(val);
  54     } else {
  55         val = cpu_to_be32(val);
  56     }
  57
  58     return val;
  59 }
  60
  61 static uint64_t cpu_convert_to_target64(uint64_t val, int endian)
  62 {
  63     if (endian == ELFDATA2LSB) {
  64         val = cpu_to_le64(val);
  65     } else {
  66         val = cpu_to_be64(val);
  67     }
  68
  69     return val;
  70 }
  71
  72 typedef struct DumpState {
  73     GuestPhysBlockList guest_phys_blocks;
  74     ArchDumpInfo dump_info;
  75     MemoryMappingList list;
  76     uint16_t phdr_num;
  77     uint32_t sh_info;
  78     bool have_section;
  79     bool resume;
  80     ssize_t note_size;
  81     hwaddr memory_offset;
  82     int fd;
  83
  84     GuestPhysBlock *next_block;
  85     ram_addr_t start;
  86     bool has_filter;
  87     int64_t begin;
  88     int64_t length;
  89
  90     uint8_t *note_buf;          /* buffer for notes */
  91     size_t note_buf_offset;     /* the writing place in note_buf */
  92     uint32_t nr_cpus;           /* number of guest's cpu */
  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     MakedumpfileHeader *mh;
 713     int ret = 0;
 714
 715     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
 716     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
 717
 718     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
 719            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
 720
 721     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
 722     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
 723
 724     size_t written_size;
 725     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
 726     if (written_size != MAX_SIZE_MDF_HEADER) {
 727         ret = -1;
 728     }
 729
 730     g_free(mh);
 731     return ret;
 732 }
 733
 734 static int write_end_flat_header(int fd)
 735 {
 736     MakedumpfileDataHeader mdh;
 737
 738     mdh.offset = END_FLAG_FLAT_HEADER;
 739     mdh.buf_size = END_FLAG_FLAT_HEADER;
 740
 741     size_t written_size;
 742     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 743     if (written_size != sizeof(mdh)) {
 744         return -1;
 745     }
 746
 747     return 0;
 748 }
 749
 750 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
 751 {
 752     size_t written_size;
 753     MakedumpfileDataHeader mdh;
 754
 755     mdh.offset = cpu_to_be64(offset);
 756     mdh.buf_size = cpu_to_be64(size);
 757
 758     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 759     if (written_size != sizeof(mdh)) {
 760         return -1;
 761     }
 762
 763     written_size = qemu_write_full(fd, buf, size);
 764     if (written_size != size) {
 765         return -1;
 766     }
 767
 768     return 0;
 769 }
 770
 771 static int buf_write_note(const void *buf, size_t size, void *opaque)
 772 {
 773     DumpState *s = opaque;
 774
 775     /* note_buf is not enough */
 776     if (s->note_buf_offset + size > s->note_size) {
 777         return -1;
 778     }
 779
 780     memcpy(s->note_buf + s->note_buf_offset, buf, size);
 781
 782     s->note_buf_offset += size;
 783
 784     return 0;
 785 }
 786
 787 /* write common header, sub header and elf note to vmcore */
 788 static int create_header32(DumpState *s)
 789 {
 790     int ret = 0;
 791     DiskDumpHeader32 *dh = NULL;
 792     KdumpSubHeader32 *kh = NULL;
 793     size_t size;
 794     int endian = s->dump_info.d_endian;
 795     uint32_t block_size;
 796     uint32_t sub_hdr_size;
 797     uint32_t bitmap_blocks;
 798     uint32_t status = 0;
 799     uint64_t offset_note;
 800
 801     /* write common header, the version of kdump-compressed format is 6th */
 802     size = sizeof(DiskDumpHeader32);
 803     dh = g_malloc0(size);
 804
 805     strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
 806     dh->header_version = cpu_convert_to_target32(6, endian);
 807     block_size = TARGET_PAGE_SIZE;
 808     dh->block_size = cpu_convert_to_target32(block_size, endian);
 809     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
 810     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 811     dh->sub_hdr_size = cpu_convert_to_target32(sub_hdr_size, endian);
 812     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 813     dh->max_mapnr = cpu_convert_to_target32(MIN(s->max_mapnr, UINT_MAX),
 814                                             endian);
 815     dh->nr_cpus = cpu_convert_to_target32(s->nr_cpus, endian);
 816     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 817     dh->bitmap_blocks = cpu_convert_to_target32(bitmap_blocks, endian);
 818     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
 819
 820     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 821         status |= DUMP_DH_COMPRESSED_ZLIB;
 822     }
 823 #ifdef CONFIG_LZO
 824     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 825         status |= DUMP_DH_COMPRESSED_LZO;
 826     }
 827 #endif
 828 #ifdef CONFIG_SNAPPY
 829     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 830         status |= DUMP_DH_COMPRESSED_SNAPPY;
 831     }
 832 #endif
 833     dh->status = cpu_convert_to_target32(status, endian);
 834
 835     if (write_buffer(s->fd, 0, dh, size) < 0) {
 836         dump_error(s, "dump: failed to write disk dump header.\n");
 837         ret = -1;
 838         goto out;
 839     }
 840
 841     /* write sub header */
 842     size = sizeof(KdumpSubHeader32);
 843     kh = g_malloc0(size);
 844
 845     /* 64bit max_mapnr_64 */
 846     kh->max_mapnr_64 = cpu_convert_to_target64(s->max_mapnr, endian);
 847     kh->phys_base = cpu_convert_to_target32(PHYS_BASE, endian);
 848     kh->dump_level = cpu_convert_to_target32(DUMP_LEVEL, endian);
 849
 850     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 851     kh->offset_note = cpu_convert_to_target64(offset_note, endian);
 852     kh->note_size = cpu_convert_to_target32(s->note_size, endian);
 853
 854     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 855                      block_size, kh, size) < 0) {
 856         dump_error(s, "dump: failed to write kdump sub header.\n");
 857         ret = -1;
 858         goto out;
 859     }
 860
 861     /* write note */
 862     s->note_buf = g_malloc0(s->note_size);
 863     s->note_buf_offset = 0;
 864
 865     /* use s->note_buf to store notes temporarily */
 866     if (write_elf32_notes(buf_write_note, s) < 0) {
 867         ret = -1;
 868         goto out;
 869     }
 870
 871     if (write_buffer(s->fd, offset_note, s->note_buf,
 872                      s->note_size) < 0) {
 873         dump_error(s, "dump: failed to write notes");
 874         ret = -1;
 875         goto out;
 876     }
 877
 878     /* get offset of dump_bitmap */
 879     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 880                              block_size;
 881
 882     /* get offset of page */
 883     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 884                      block_size;
 885
 886 out:
 887     g_free(dh);
 888     g_free(kh);
 889     g_free(s->note_buf);
 890
 891     return ret;
 892 }
 893
 894 /* write common header, sub header and elf note to vmcore */
 895 static int create_header64(DumpState *s)
 896 {
 897     int ret = 0;
 898     DiskDumpHeader64 *dh = NULL;
 899     KdumpSubHeader64 *kh = NULL;
 900     size_t size;
 901     int endian = s->dump_info.d_endian;
 902     uint32_t block_size;
 903     uint32_t sub_hdr_size;
 904     uint32_t bitmap_blocks;
 905     uint32_t status = 0;
 906     uint64_t offset_note;
 907
 908     /* write common header, the version of kdump-compressed format is 6th */
 909     size = sizeof(DiskDumpHeader64);
 910     dh = g_malloc0(size);
 911
 912     strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
 913     dh->header_version = cpu_convert_to_target32(6, endian);
 914     block_size = TARGET_PAGE_SIZE;
 915     dh->block_size = cpu_convert_to_target32(block_size, endian);
 916     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
 917     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 918     dh->sub_hdr_size = cpu_convert_to_target32(sub_hdr_size, endian);
 919     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 920     dh->max_mapnr = cpu_convert_to_target32(MIN(s->max_mapnr, UINT_MAX),
 921                                             endian);
 922     dh->nr_cpus = cpu_convert_to_target32(s->nr_cpus, endian);
 923     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 924     dh->bitmap_blocks = cpu_convert_to_target32(bitmap_blocks, endian);
 925     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
 926
 927     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 928         status |= DUMP_DH_COMPRESSED_ZLIB;
 929     }
 930 #ifdef CONFIG_LZO
 931     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 932         status |= DUMP_DH_COMPRESSED_LZO;
 933     }
 934 #endif
 935 #ifdef CONFIG_SNAPPY
 936     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 937         status |= DUMP_DH_COMPRESSED_SNAPPY;
 938     }
 939 #endif
 940     dh->status = cpu_convert_to_target32(status, endian);
 941
 942     if (write_buffer(s->fd, 0, dh, size) < 0) {
 943         dump_error(s, "dump: failed to write disk dump header.\n");
 944         ret = -1;
 945         goto out;
 946     }
 947
 948     /* write sub header */
 949     size = sizeof(KdumpSubHeader64);
 950     kh = g_malloc0(size);
 951
 952     /* 64bit max_mapnr_64 */
 953     kh->max_mapnr_64 = cpu_convert_to_target64(s->max_mapnr, endian);
 954     kh->phys_base = cpu_convert_to_target64(PHYS_BASE, endian);
 955     kh->dump_level = cpu_convert_to_target32(DUMP_LEVEL, endian);
 956
 957     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 958     kh->offset_note = cpu_convert_to_target64(offset_note, endian);
 959     kh->note_size = cpu_convert_to_target64(s->note_size, endian);
 960
 961     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 962                      block_size, kh, size) < 0) {
 963         dump_error(s, "dump: failed to write kdump sub header.\n");
 964         ret = -1;
 965         goto out;
 966     }
 967
 968     /* write note */
 969     s->note_buf = g_malloc0(s->note_size);
 970     s->note_buf_offset = 0;
 971
 972     /* use s->note_buf to store notes temporarily */
 973     if (write_elf64_notes(buf_write_note, s) < 0) {
 974         ret = -1;
 975         goto out;
 976     }
 977
 978     if (write_buffer(s->fd, offset_note, s->note_buf,
 979                      s->note_size) < 0) {
 980         dump_error(s, "dump: failed to write notes");
 981         ret = -1;
 982         goto out;
 983     }
 984
 985     /* get offset of dump_bitmap */
 986     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 987                              block_size;
 988
 989     /* get offset of page */
 990     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 991                      block_size;
 992
 993 out:
 994     g_free(dh);
 995     g_free(kh);
 996     g_free(s->note_buf);
 997
 998     return ret;
 999 }
1000
1001 static int write_dump_header(DumpState *s)
1002 {
1003     if (s->dump_info.d_class == ELFCLASS32) {
1004         return create_header32(s);
1005     } else {
1006         return create_header64(s);
1007     }
1008 }
1009
1010 /*
1011  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1012  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1013  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1014  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1015  * vmcore, ie. synchronizing un-sync bit into vmcore.
1016  */
1017 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1018                            uint8_t *buf, DumpState *s)
1019 {
1020     off_t old_offset, new_offset;
1021     off_t offset_bitmap1, offset_bitmap2;
1022     uint32_t byte, bit;
1023
1024     /* should not set the previous place */
1025     assert(last_pfn <= pfn);
1026
1027     /*
1028      * if the bit needed to be set is not cached in buf, flush the data in buf
1029      * to vmcore firstly.
1030      * making new_offset be bigger than old_offset can also sync remained data
1031      * into vmcore.
1032      */
1033     old_offset = BUFSIZE_BITMAP * (last_pfn / PFN_BUFBITMAP);
1034     new_offset = BUFSIZE_BITMAP * (pfn / PFN_BUFBITMAP);
1035
1036     while (old_offset < new_offset) {
1037         /* calculate the offset and write dump_bitmap */
1038         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1039         if (write_buffer(s->fd, offset_bitmap1, buf,
1040                          BUFSIZE_BITMAP) < 0) {
1041             return -1;
1042         }
1043
1044         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1045         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1046                          old_offset;
1047         if (write_buffer(s->fd, offset_bitmap2, buf,
1048                          BUFSIZE_BITMAP) < 0) {
1049             return -1;
1050         }
1051
1052         memset(buf, 0, BUFSIZE_BITMAP);
1053         old_offset += BUFSIZE_BITMAP;
1054     }
1055
1056     /* get the exact place of the bit in the buf, and set it */
1057     byte = (pfn % PFN_BUFBITMAP) / CHAR_BIT;
1058     bit = (pfn % PFN_BUFBITMAP) % CHAR_BIT;
1059     if (value) {
1060         buf[byte] |= 1u << bit;
1061     } else {
1062         buf[byte] &= ~(1u << bit);
1063     }
1064
1065     return 0;
1066 }
1067
1068 /*
1069  * exam every page and return the page frame number and the address of the page.
1070  * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1071  * blocks, so block->target_start and block->target_end should be interal
1072  * multiples of the target page size.
1073  */
1074 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1075                           uint8_t **bufptr, DumpState *s)
1076 {
1077     GuestPhysBlock *block = *blockptr;
1078     hwaddr addr;
1079     uint8_t *buf;
1080
1081     /* block == NULL means the start of the iteration */
1082     if (!block) {
1083         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1084         *blockptr = block;
1085         assert((block->target_start & ~TARGET_PAGE_MASK) == 0);
1086         assert((block->target_end & ~TARGET_PAGE_MASK) == 0);
1087         *pfnptr = paddr_to_pfn(block->target_start);
1088         if (bufptr) {
1089             *bufptr = block->host_addr;
1090         }
1091         return true;
1092     }
1093
1094     *pfnptr = *pfnptr + 1;
1095     addr = pfn_to_paddr(*pfnptr);
1096
1097     if ((addr >= block->target_start) &&
1098         (addr + TARGET_PAGE_SIZE <= block->target_end)) {
1099         buf = block->host_addr + (addr - block->target_start);
1100     } else {
1101         /* the next page is in the next block */
1102         block = QTAILQ_NEXT(block, next);
1103         *blockptr = block;
1104         if (!block) {
1105             return false;
1106         }
1107         assert((block->target_start & ~TARGET_PAGE_MASK) == 0);
1108         assert((block->target_end & ~TARGET_PAGE_MASK) == 0);
1109         *pfnptr = paddr_to_pfn(block->target_start);
1110         buf = block->host_addr;
1111     }
1112
1113     if (bufptr) {
1114         *bufptr = buf;
1115     }
1116
1117     return true;
1118 }
1119
1120 static int write_dump_bitmap(DumpState *s)
1121 {
1122     int ret = 0;
1123     uint64_t last_pfn, pfn;
1124     void *dump_bitmap_buf;
1125     size_t num_dumpable;
1126     GuestPhysBlock *block_iter = NULL;
1127
1128     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1129     dump_bitmap_buf = g_malloc0(BUFSIZE_BITMAP);
1130
1131     num_dumpable = 0;
1132     last_pfn = 0;
1133
1134     /*
1135      * exam memory page by page, and set the bit in dump_bitmap corresponded
1136      * to the existing page.
1137      */
1138     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1139         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1140         if (ret < 0) {
1141             dump_error(s, "dump: failed to set dump_bitmap.\n");
1142             ret = -1;
1143             goto out;
1144         }
1145
1146         last_pfn = pfn;
1147         num_dumpable++;
1148     }
1149
1150     /*
1151      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1152      * set last_pfn + PFN_BUFBITMAP to 0 and those set but un-sync bit will be
1153      * synchronized into vmcore.
1154      */
1155     if (num_dumpable > 0) {
1156         ret = set_dump_bitmap(last_pfn, last_pfn + PFN_BUFBITMAP, false,
1157                               dump_bitmap_buf, s);
1158         if (ret < 0) {
1159             dump_error(s, "dump: failed to sync dump_bitmap.\n");
1160             ret = -1;
1161             goto out;
1162         }
1163     }
1164
1165     /* number of dumpable pages that will be dumped later */
1166     s->num_dumpable = num_dumpable;
1167
1168 out:
1169     g_free(dump_bitmap_buf);
1170
1171     return ret;
1172 }
1173
1174 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1175                                off_t offset)
1176 {
1177     data_cache->fd = s->fd;
1178     data_cache->data_size = 0;
1179     data_cache->buf_size = BUFSIZE_DATA_CACHE;
1180     data_cache->buf = g_malloc0(BUFSIZE_DATA_CACHE);
1181     data_cache->offset = offset;
1182 }
1183
1184 static int write_cache(DataCache *dc, const void *buf, size_t size,
1185                        bool flag_sync)
1186 {
1187     /*
1188      * dc->buf_size should not be less than size, otherwise dc will never be
1189      * enough
1190      */
1191     assert(size <= dc->buf_size);
1192
1193     /*
1194      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1195      * otherwise check if the space is enough for caching data in buf, if not,
1196      * write the data in dc->buf to dc->fd and reset dc->buf
1197      */
1198     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1199         (flag_sync && dc->data_size > 0)) {
1200         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1201             return -1;
1202         }
1203
1204         dc->offset += dc->data_size;
1205         dc->data_size = 0;
1206     }
1207
1208     if (!flag_sync) {
1209         memcpy(dc->buf + dc->data_size, buf, size);
1210         dc->data_size += size;
1211     }
1212
1213     return 0;
1214 }
1215
1216 static void free_data_cache(DataCache *data_cache)
1217 {
1218     g_free(data_cache->buf);
1219 }
1220
1221 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1222 {
1223     size_t len_buf_out_zlib, len_buf_out_lzo, len_buf_out_snappy;
1224     size_t len_buf_out;
1225
1226     /* init buf_out */
1227     len_buf_out_zlib = len_buf_out_lzo = len_buf_out_snappy = 0;
1228
1229     /* buf size for zlib */
1230     len_buf_out_zlib = compressBound(page_size);
1231
1232     /* buf size for lzo */
1233 #ifdef CONFIG_LZO
1234     if (flag_compress & DUMP_DH_COMPRESSED_LZO) {
1235         if (lzo_init() != LZO_E_OK) {
1236             /* return 0 to indicate lzo is unavailable */
1237             return 0;
1238         }
1239     }
1240
1241     /*
1242      * LZO will expand incompressible data by a little amount. please check the
1243      * following URL to see the expansion calculation:
1244      * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1245      */
1246     len_buf_out_lzo = page_size + page_size / 16 + 64 + 3;
1247 #endif
1248
1249 #ifdef CONFIG_SNAPPY
1250     /* buf size for snappy */
1251     len_buf_out_snappy = snappy_max_compressed_length(page_size);
1252 #endif
1253
1254     /* get the biggest that can store all kinds of compressed page */
1255     len_buf_out = MAX(len_buf_out_zlib,
1256                       MAX(len_buf_out_lzo, len_buf_out_snappy));
1257
1258     return len_buf_out;
1259 }
1260
1261 /*
1262  * check if the page is all 0
1263  */
1264 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1265 {
1266     return buffer_is_zero(buf, page_size);
1267 }
1268
1269 static int write_dump_pages(DumpState *s)
1270 {
1271     int ret = 0;
1272     DataCache page_desc, page_data;
1273     size_t len_buf_out, size_out;
1274 #ifdef CONFIG_LZO
1275     lzo_bytep wrkmem = NULL;
1276 #endif
1277     uint8_t *buf_out = NULL;
1278     off_t offset_desc, offset_data;
1279     PageDescriptor pd, pd_zero;
1280     uint8_t *buf;
1281     int endian = s->dump_info.d_endian;
1282     GuestPhysBlock *block_iter = NULL;
1283     uint64_t pfn_iter;
1284
1285     /* get offset of page_desc and page_data in dump file */
1286     offset_desc = s->offset_page;
1287     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1288
1289     prepare_data_cache(&page_desc, s, offset_desc);
1290     prepare_data_cache(&page_data, s, offset_data);
1291
1292     /* prepare buffer to store compressed data */
1293     len_buf_out = get_len_buf_out(TARGET_PAGE_SIZE, s->flag_compress);
1294     if (len_buf_out == 0) {
1295         dump_error(s, "dump: failed to get length of output buffer.\n");
1296         goto out;
1297     }
1298
1299 #ifdef CONFIG_LZO
1300     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1301 #endif
1302
1303     buf_out = g_malloc(len_buf_out);
1304
1305     /*
1306      * init zero page's page_desc and page_data, because every zero page
1307      * uses the same page_data
1308      */
1309     pd_zero.size = cpu_convert_to_target32(TARGET_PAGE_SIZE, endian);
1310     pd_zero.flags = cpu_convert_to_target32(0, endian);
1311     pd_zero.offset = cpu_convert_to_target64(offset_data, endian);
1312     pd_zero.page_flags = cpu_convert_to_target64(0, endian);
1313     buf = g_malloc0(TARGET_PAGE_SIZE);
1314     ret = write_cache(&page_data, buf, TARGET_PAGE_SIZE, false);
1315     g_free(buf);
1316     if (ret < 0) {
1317         dump_error(s, "dump: failed to write page data(zero page).\n");
1318         goto out;
1319     }
1320
1321     offset_data += TARGET_PAGE_SIZE;
1322
1323     /*
1324      * dump memory to vmcore page by page. zero page will all be resided in the
1325      * first page of page section
1326      */
1327     while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1328         /* check zero page */
1329         if (is_zero_page(buf, TARGET_PAGE_SIZE)) {
1330             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1331                               false);
1332             if (ret < 0) {
1333                 dump_error(s, "dump: failed to write page desc.\n");
1334                 goto out;
1335             }
1336         } else {
1337             /*
1338              * not zero page, then:
1339              * 1. compress the page
1340              * 2. write the compressed page into the cache of page_data
1341              * 3. get page desc of the compressed page and write it into the
1342              *    cache of page_desc
1343              *
1344              * only one compression format will be used here, for
1345              * s->flag_compress is set. But when compression fails to work,
1346              * we fall back to save in plaintext.
1347              */
1348              size_out = len_buf_out;
1349              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1350                  (compress2(buf_out, (uLongf *)&size_out, buf,
1351                             TARGET_PAGE_SIZE, Z_BEST_SPEED) == Z_OK) &&
1352                  (size_out < TARGET_PAGE_SIZE)) {
1353                 pd.flags = cpu_convert_to_target32(DUMP_DH_COMPRESSED_ZLIB,
1354                                                    endian);
1355                 pd.size  = cpu_convert_to_target32(size_out, endian);
1356
1357                 ret = write_cache(&page_data, buf_out, size_out, false);
1358                 if (ret < 0) {
1359                     dump_error(s, "dump: failed to write page data.\n");
1360                     goto out;
1361                 }
1362 #ifdef CONFIG_LZO
1363             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1364                     (lzo1x_1_compress(buf, TARGET_PAGE_SIZE, buf_out,
1365                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1366                     (size_out < TARGET_PAGE_SIZE)) {
1367                 pd.flags = cpu_convert_to_target32(DUMP_DH_COMPRESSED_LZO,
1368                                                    endian);
1369                 pd.size  = cpu_convert_to_target32(size_out, endian);
1370
1371                 ret = write_cache(&page_data, buf_out, size_out, false);
1372                 if (ret < 0) {
1373                     dump_error(s, "dump: failed to write page data.\n");
1374                     goto out;
1375                 }
1376 #endif
1377 #ifdef CONFIG_SNAPPY
1378             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1379                     (snappy_compress((char *)buf, TARGET_PAGE_SIZE,
1380                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1381                     (size_out < TARGET_PAGE_SIZE)) {
1382                 pd.flags = cpu_convert_to_target32(
1383                                         DUMP_DH_COMPRESSED_SNAPPY, endian);
1384                 pd.size  = cpu_convert_to_target32(size_out, endian);
1385
1386                 ret = write_cache(&page_data, buf_out, size_out, false);
1387                 if (ret < 0) {
1388                     dump_error(s, "dump: failed to write page data.\n");
1389                     goto out;
1390                 }
1391 #endif
1392             } else {
1393                 /*
1394                  * fall back to save in plaintext, size_out should be
1395                  * assigned TARGET_PAGE_SIZE
1396                  */
1397                 pd.flags = cpu_convert_to_target32(0, endian);
1398                 size_out = TARGET_PAGE_SIZE;
1399                 pd.size = cpu_convert_to_target32(size_out, endian);
1400
1401                 ret = write_cache(&page_data, buf, TARGET_PAGE_SIZE, false);
1402                 if (ret < 0) {
1403                     dump_error(s, "dump: failed to write page data.\n");
1404                     goto out;
1405                 }
1406             }
1407
1408             /* get and write page desc here */
1409             pd.page_flags = cpu_convert_to_target64(0, endian);
1410             pd.offset = cpu_convert_to_target64(offset_data, endian);
1411             offset_data += size_out;
1412
1413             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1414             if (ret < 0) {
1415                 dump_error(s, "dump: failed to write page desc.\n");
1416                 goto out;
1417             }
1418         }
1419     }
1420
1421     ret = write_cache(&page_desc, NULL, 0, true);
1422     if (ret < 0) {
1423         dump_error(s, "dump: failed to sync cache for page_desc.\n");
1424         goto out;
1425     }
1426     ret = write_cache(&page_data, NULL, 0, true);
1427     if (ret < 0) {
1428         dump_error(s, "dump: failed to sync cache for page_data.\n");
1429         goto out;
1430     }
1431
1432 out:
1433     free_data_cache(&page_desc);
1434     free_data_cache(&page_data);
1435
1436 #ifdef CONFIG_LZO
1437     g_free(wrkmem);
1438 #endif
1439
1440     g_free(buf_out);
1441
1442     return ret;
1443 }
1444
1445 static int create_kdump_vmcore(DumpState *s)
1446 {
1447     int ret;
1448
1449     /*
1450      * the kdump-compressed format is:
1451      *                                               File offset
1452      *  +------------------------------------------+ 0x0
1453      *  |    main header (struct disk_dump_header) |
1454      *  |------------------------------------------+ block 1
1455      *  |    sub header (struct kdump_sub_header)  |
1456      *  |------------------------------------------+ block 2
1457      *  |            1st-dump_bitmap               |
1458      *  |------------------------------------------+ block 2 + X blocks
1459      *  |            2nd-dump_bitmap               | (aligned by block)
1460      *  |------------------------------------------+ block 2 + 2 * X blocks
1461      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1462      *  |  page desc for pfn 1 (struct page_desc)  |
1463      *  |                    :                     |
1464      *  |------------------------------------------| (not aligned by block)
1465      *  |         page data (pfn 0)                |
1466      *  |         page data (pfn 1)                |
1467      *  |                    :                     |
1468      *  +------------------------------------------+
1469      */
1470
1471     ret = write_start_flat_header(s->fd);
1472     if (ret < 0) {
1473         dump_error(s, "dump: failed to write start flat header.\n");
1474         return -1;
1475     }
1476
1477     ret = write_dump_header(s);
1478     if (ret < 0) {
1479         return -1;
1480     }
1481
1482     ret = write_dump_bitmap(s);
1483     if (ret < 0) {
1484         return -1;
1485     }
1486
1487     ret = write_dump_pages(s);
1488     if (ret < 0) {
1489         return -1;
1490     }
1491
1492     ret = write_end_flat_header(s->fd);
1493     if (ret < 0) {
1494         dump_error(s, "dump: failed to write end flat header.\n");
1495         return -1;
1496     }
1497
1498     dump_completed(s);
1499
1500     return 0;
1501 }
1502
1503 static ram_addr_t get_start_block(DumpState *s)
1504 {
1505     GuestPhysBlock *block;
1506
1507     if (!s->has_filter) {
1508         s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1509         return 0;
1510     }
1511
1512     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1513         if (block->target_start >= s->begin + s->length ||
1514             block->target_end <= s->begin) {
1515             /* This block is out of the range */
1516             continue;
1517         }
1518
1519         s->next_block = block;
1520         if (s->begin > block->target_start) {
1521             s->start = s->begin - block->target_start;
1522         } else {
1523             s->start = 0;
1524         }
1525         return s->start;
1526     }
1527
1528     return -1;
1529 }
1530
1531 static void get_max_mapnr(DumpState *s)
1532 {
1533     GuestPhysBlock *last_block;
1534
1535     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head, GuestPhysBlockHead);
1536     s->max_mapnr = paddr_to_pfn(last_block->target_end);
1537 }
1538
1539 static int dump_init(DumpState *s, int fd, bool has_format,
1540                      DumpGuestMemoryFormat format, bool paging, bool has_filter,
1541                      int64_t begin, int64_t length, Error **errp)
1542 {
1543     CPUState *cpu;
1544     int nr_cpus;
1545     Error *err = NULL;
1546     int ret;
1547
1548     /* kdump-compressed is conflict with paging and filter */
1549     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1550         assert(!paging && !has_filter);
1551     }
1552
1553     if (runstate_is_running()) {
1554         vm_stop(RUN_STATE_SAVE_VM);
1555         s->resume = true;
1556     } else {
1557         s->resume = false;
1558     }
1559
1560     /* If we use KVM, we should synchronize the registers before we get dump
1561      * info or physmap info.
1562      */
1563     cpu_synchronize_all_states();
1564     nr_cpus = 0;
1565     CPU_FOREACH(cpu) {
1566         nr_cpus++;
1567     }
1568
1569     s->fd = fd;
1570     s->has_filter = has_filter;
1571     s->begin = begin;
1572     s->length = length;
1573
1574     guest_phys_blocks_init(&s->guest_phys_blocks);
1575     guest_phys_blocks_append(&s->guest_phys_blocks);
1576
1577     s->start = get_start_block(s);
1578     if (s->start == -1) {
1579         error_set(errp, QERR_INVALID_PARAMETER, "begin");
1580         goto cleanup;
1581     }
1582
1583     /* get dump info: endian, class and architecture.
1584      * If the target architecture is not supported, cpu_get_dump_info() will
1585      * return -1.
1586      */
1587     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1588     if (ret < 0) {
1589         error_set(errp, QERR_UNSUPPORTED);
1590         goto cleanup;
1591     }
1592
1593     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1594                                      s->dump_info.d_machine, nr_cpus);
1595     if (s->note_size < 0) {
1596         error_set(errp, QERR_UNSUPPORTED);
1597         goto cleanup;
1598     }
1599
1600     /* get memory mapping */
1601     memory_mapping_list_init(&s->list);
1602     if (paging) {
1603         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1604         if (err != NULL) {
1605             error_propagate(errp, err);
1606             goto cleanup;
1607         }
1608     } else {
1609         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1610     }
1611
1612     s->nr_cpus = nr_cpus;
1613
1614     get_max_mapnr(s);
1615
1616     uint64_t tmp;
1617     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT), TARGET_PAGE_SIZE);
1618     s->len_dump_bitmap = tmp * TARGET_PAGE_SIZE;
1619
1620     /* init for kdump-compressed format */
1621     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1622         switch (format) {
1623         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1624             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1625             break;
1626
1627         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1628             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1629             break;
1630
1631         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1632             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1633             break;
1634
1635         default:
1636             s->flag_compress = 0;
1637         }
1638
1639         return 0;
1640     }
1641
1642     if (s->has_filter) {
1643         memory_mapping_filter(&s->list, s->begin, s->length);
1644     }
1645
1646     /*
1647      * calculate phdr_num
1648      *
1649      * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1650      */
1651     s->phdr_num = 1; /* PT_NOTE */
1652     if (s->list.num < UINT16_MAX - 2) {
1653         s->phdr_num += s->list.num;
1654         s->have_section = false;
1655     } else {
1656         s->have_section = true;
1657         s->phdr_num = PN_XNUM;
1658         s->sh_info = 1; /* PT_NOTE */
1659
1660         /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1661         if (s->list.num <= UINT32_MAX - 1) {
1662             s->sh_info += s->list.num;
1663         } else {
1664             s->sh_info = UINT32_MAX;
1665         }
1666     }
1667
1668     if (s->dump_info.d_class == ELFCLASS64) {
1669         if (s->have_section) {
1670             s->memory_offset = sizeof(Elf64_Ehdr) +
1671                                sizeof(Elf64_Phdr) * s->sh_info +
1672                                sizeof(Elf64_Shdr) + s->note_size;
1673         } else {
1674             s->memory_offset = sizeof(Elf64_Ehdr) +
1675                                sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1676         }
1677     } else {
1678         if (s->have_section) {
1679             s->memory_offset = sizeof(Elf32_Ehdr) +
1680                                sizeof(Elf32_Phdr) * s->sh_info +
1681                                sizeof(Elf32_Shdr) + s->note_size;
1682         } else {
1683             s->memory_offset = sizeof(Elf32_Ehdr) +
1684                                sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1685         }
1686     }
1687
1688     return 0;
1689
1690 cleanup:
1691     guest_phys_blocks_free(&s->guest_phys_blocks);
1692
1693     if (s->resume) {
1694         vm_start();
1695     }
1696
1697     return -1;
1698 }
1699
1700 void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
1701                            int64_t begin, bool has_length,
1702                            int64_t length, bool has_format,
1703                            DumpGuestMemoryFormat format, Error **errp)
1704 {
1705     const char *p;
1706     int fd = -1;
1707     DumpState *s;
1708     int ret;
1709
1710     /*
1711      * kdump-compressed format need the whole memory dumped, so paging or
1712      * filter is not supported here.
1713      */
1714     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1715         (paging || has_begin || has_length)) {
1716         error_setg(errp, "kdump-compressed format doesn't support paging or "
1717                          "filter");
1718         return;
1719     }
1720     if (has_begin && !has_length) {
1721         error_set(errp, QERR_MISSING_PARAMETER, "length");
1722         return;
1723     }
1724     if (!has_begin && has_length) {
1725         error_set(errp, QERR_MISSING_PARAMETER, "begin");
1726         return;
1727     }
1728
1729     /* check whether lzo/snappy is supported */
1730 #ifndef CONFIG_LZO
1731     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1732         error_setg(errp, "kdump-lzo is not available now");
1733         return;
1734     }
1735 #endif
1736
1737 #ifndef CONFIG_SNAPPY
1738     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1739         error_setg(errp, "kdump-snappy is not available now");
1740         return;
1741     }
1742 #endif
1743
1744 #if !defined(WIN32)
1745     if (strstart(file, "fd:", &p)) {
1746         fd = monitor_get_fd(cur_mon, p, errp);
1747         if (fd == -1) {
1748             return;
1749         }
1750     }
1751 #endif
1752
1753     if  (strstart(file, "file:", &p)) {
1754         fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1755         if (fd < 0) {
1756             error_setg_file_open(errp, errno, p);
1757             return;
1758         }
1759     }
1760
1761     if (fd == -1) {
1762         error_set(errp, QERR_INVALID_PARAMETER, "protocol");
1763         return;
1764     }
1765
1766     s = g_malloc0(sizeof(DumpState));
1767
1768     ret = dump_init(s, fd, has_format, format, paging, has_begin,
1769                     begin, length, errp);
1770     if (ret < 0) {
1771         g_free(s);
1772         return;
1773     }
1774
1775     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1776         if (create_kdump_vmcore(s) < 0) {
1777             error_set(errp, QERR_IO_ERROR);
1778         }
1779     } else {
1780         if (create_vmcore(s) < 0) {
1781             error_set(errp, QERR_IO_ERROR);
1782         }
1783     }
1784
1785     g_free(s);
1786 }
1787
1788 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
1789 {
1790     DumpGuestMemoryFormatList *item;
1791     DumpGuestMemoryCapability *cap =
1792                                   g_malloc0(sizeof(DumpGuestMemoryCapability));
1793
1794     /* elf is always available */
1795     item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1796     cap->formats = item;
1797     item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
1798
1799     /* kdump-zlib is always available */
1800     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1801     item = item->next;
1802     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
1803
1804     /* add new item if kdump-lzo is available */
1805 #ifdef CONFIG_LZO
1806     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1807     item = item->next;
1808     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
1809 #endif
1810
1811     /* add new item if kdump-snappy is available */
1812 #ifdef CONFIG_SNAPPY
1813     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1814     item = item->next;
1815     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
1816 #endif
1817
1818     return cap;
1819 }