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 uint16_t cpu_to_dump16(DumpState *s, uint16_t val)
  40 {
  41     if (s->dump_info.d_endian == ELFDATA2LSB) {
  42         val = cpu_to_le16(val);
  43     } else {
  44         val = cpu_to_be16(val);
  45     }
  46
  47     return val;
  48 }
  49
  50 uint32_t cpu_to_dump32(DumpState *s, uint32_t val)
  51 {
  52     if (s->dump_info.d_endian == ELFDATA2LSB) {
  53         val = cpu_to_le32(val);
  54     } else {
  55         val = cpu_to_be32(val);
  56     }
  57
  58     return val;
  59 }
  60
  61 uint64_t cpu_to_dump64(DumpState *s, uint64_t val)
  62 {
  63     if (s->dump_info.d_endian == ELFDATA2LSB) {
  64         val = cpu_to_le64(val);
  65     } else {
  66         val = cpu_to_be64(val);
  67     }
  68
  69     return val;
  70 }
  71
  72 static int dump_cleanup(DumpState *s)
  73 {
  74     guest_phys_blocks_free(&s->guest_phys_blocks);
  75     memory_mapping_list_free(&s->list);
  76     close(s->fd);
  77     if (s->resume) {
  78         vm_start();
  79     }
  80
  81     return 0;
  82 }
  83
  84 static void dump_error(DumpState *s, const char *reason)
  85 {
  86     dump_cleanup(s);
  87 }
  88
  89 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
  90 {
  91     DumpState *s = opaque;
  92     size_t written_size;
  93
  94     written_size = qemu_write_full(s->fd, buf, size);
  95     if (written_size != size) {
  96         return -1;
  97     }
  98
  99     return 0;
 100 }
 101
 102 static int write_elf64_header(DumpState *s)
 103 {
 104     Elf64_Ehdr elf_header;
 105     int ret;
 106
 107     memset(&elf_header, 0, sizeof(Elf64_Ehdr));
 108     memcpy(&elf_header, ELFMAG, SELFMAG);
 109     elf_header.e_ident[EI_CLASS] = ELFCLASS64;
 110     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
 111     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
 112     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
 113     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
 114     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
 115     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
 116     elf_header.e_phoff = cpu_to_dump64(s, sizeof(Elf64_Ehdr));
 117     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr));
 118     elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
 119     if (s->have_section) {
 120         uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
 121
 122         elf_header.e_shoff = cpu_to_dump64(s, shoff);
 123         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr));
 124         elf_header.e_shnum = cpu_to_dump16(s, 1);
 125     }
 126
 127     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
 128     if (ret < 0) {
 129         dump_error(s, "dump: failed to write elf header.\n");
 130         return -1;
 131     }
 132
 133     return 0;
 134 }
 135
 136 static int write_elf32_header(DumpState *s)
 137 {
 138     Elf32_Ehdr elf_header;
 139     int ret;
 140
 141     memset(&elf_header, 0, sizeof(Elf32_Ehdr));
 142     memcpy(&elf_header, ELFMAG, SELFMAG);
 143     elf_header.e_ident[EI_CLASS] = ELFCLASS32;
 144     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
 145     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
 146     elf_header.e_type = cpu_to_dump16(s, ET_CORE);
 147     elf_header.e_machine = cpu_to_dump16(s, s->dump_info.d_machine);
 148     elf_header.e_version = cpu_to_dump32(s, EV_CURRENT);
 149     elf_header.e_ehsize = cpu_to_dump16(s, sizeof(elf_header));
 150     elf_header.e_phoff = cpu_to_dump32(s, sizeof(Elf32_Ehdr));
 151     elf_header.e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr));
 152     elf_header.e_phnum = cpu_to_dump16(s, s->phdr_num);
 153     if (s->have_section) {
 154         uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
 155
 156         elf_header.e_shoff = cpu_to_dump32(s, shoff);
 157         elf_header.e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr));
 158         elf_header.e_shnum = cpu_to_dump16(s, 1);
 159     }
 160
 161     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
 162     if (ret < 0) {
 163         dump_error(s, "dump: failed to write elf header.\n");
 164         return -1;
 165     }
 166
 167     return 0;
 168 }
 169
 170 static int write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
 171                             int phdr_index, hwaddr offset,
 172                             hwaddr filesz)
 173 {
 174     Elf64_Phdr phdr;
 175     int ret;
 176
 177     memset(&phdr, 0, sizeof(Elf64_Phdr));
 178     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
 179     phdr.p_offset = cpu_to_dump64(s, offset);
 180     phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr);
 181     phdr.p_filesz = cpu_to_dump64(s, filesz);
 182     phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length);
 183     phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr);
 184
 185     assert(memory_mapping->length >= filesz);
 186
 187     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
 188     if (ret < 0) {
 189         dump_error(s, "dump: failed to write program header table.\n");
 190         return -1;
 191     }
 192
 193     return 0;
 194 }
 195
 196 static int write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
 197                             int phdr_index, hwaddr offset,
 198                             hwaddr filesz)
 199 {
 200     Elf32_Phdr phdr;
 201     int ret;
 202
 203     memset(&phdr, 0, sizeof(Elf32_Phdr));
 204     phdr.p_type = cpu_to_dump32(s, PT_LOAD);
 205     phdr.p_offset = cpu_to_dump32(s, offset);
 206     phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr);
 207     phdr.p_filesz = cpu_to_dump32(s, filesz);
 208     phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length);
 209     phdr.p_vaddr = cpu_to_dump32(s, memory_mapping->virt_addr);
 210
 211     assert(memory_mapping->length >= filesz);
 212
 213     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
 214     if (ret < 0) {
 215         dump_error(s, "dump: failed to write program header table.\n");
 216         return -1;
 217     }
 218
 219     return 0;
 220 }
 221
 222 static int write_elf64_note(DumpState *s)
 223 {
 224     Elf64_Phdr phdr;
 225     hwaddr begin = s->memory_offset - s->note_size;
 226     int ret;
 227
 228     memset(&phdr, 0, sizeof(Elf64_Phdr));
 229     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
 230     phdr.p_offset = cpu_to_dump64(s, begin);
 231     phdr.p_paddr = 0;
 232     phdr.p_filesz = cpu_to_dump64(s, s->note_size);
 233     phdr.p_memsz = cpu_to_dump64(s, s->note_size);
 234     phdr.p_vaddr = 0;
 235
 236     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
 237     if (ret < 0) {
 238         dump_error(s, "dump: failed to write program header table.\n");
 239         return -1;
 240     }
 241
 242     return 0;
 243 }
 244
 245 static inline int cpu_index(CPUState *cpu)
 246 {
 247     return cpu->cpu_index + 1;
 248 }
 249
 250 static int write_elf64_notes(WriteCoreDumpFunction f, DumpState *s)
 251 {
 252     CPUState *cpu;
 253     int ret;
 254     int id;
 255
 256     CPU_FOREACH(cpu) {
 257         id = cpu_index(cpu);
 258         ret = cpu_write_elf64_note(f, cpu, id, s);
 259         if (ret < 0) {
 260             dump_error(s, "dump: failed to write elf notes.\n");
 261             return -1;
 262         }
 263     }
 264
 265     CPU_FOREACH(cpu) {
 266         ret = cpu_write_elf64_qemunote(f, cpu, s);
 267         if (ret < 0) {
 268             dump_error(s, "dump: failed to write CPU status.\n");
 269             return -1;
 270         }
 271     }
 272
 273     return 0;
 274 }
 275
 276 static int write_elf32_note(DumpState *s)
 277 {
 278     hwaddr begin = s->memory_offset - s->note_size;
 279     Elf32_Phdr phdr;
 280     int ret;
 281
 282     memset(&phdr, 0, sizeof(Elf32_Phdr));
 283     phdr.p_type = cpu_to_dump32(s, PT_NOTE);
 284     phdr.p_offset = cpu_to_dump32(s, begin);
 285     phdr.p_paddr = 0;
 286     phdr.p_filesz = cpu_to_dump32(s, s->note_size);
 287     phdr.p_memsz = cpu_to_dump32(s, s->note_size);
 288     phdr.p_vaddr = 0;
 289
 290     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
 291     if (ret < 0) {
 292         dump_error(s, "dump: failed to write program header table.\n");
 293         return -1;
 294     }
 295
 296     return 0;
 297 }
 298
 299 static int write_elf32_notes(WriteCoreDumpFunction f, DumpState *s)
 300 {
 301     CPUState *cpu;
 302     int ret;
 303     int id;
 304
 305     CPU_FOREACH(cpu) {
 306         id = cpu_index(cpu);
 307         ret = cpu_write_elf32_note(f, cpu, id, s);
 308         if (ret < 0) {
 309             dump_error(s, "dump: failed to write elf notes.\n");
 310             return -1;
 311         }
 312     }
 313
 314     CPU_FOREACH(cpu) {
 315         ret = cpu_write_elf32_qemunote(f, cpu, s);
 316         if (ret < 0) {
 317             dump_error(s, "dump: failed to write CPU status.\n");
 318             return -1;
 319         }
 320     }
 321
 322     return 0;
 323 }
 324
 325 static int write_elf_section(DumpState *s, int type)
 326 {
 327     Elf32_Shdr shdr32;
 328     Elf64_Shdr shdr64;
 329     int shdr_size;
 330     void *shdr;
 331     int ret;
 332
 333     if (type == 0) {
 334         shdr_size = sizeof(Elf32_Shdr);
 335         memset(&shdr32, 0, shdr_size);
 336         shdr32.sh_info = cpu_to_dump32(s, s->sh_info);
 337         shdr = &shdr32;
 338     } else {
 339         shdr_size = sizeof(Elf64_Shdr);
 340         memset(&shdr64, 0, shdr_size);
 341         shdr64.sh_info = cpu_to_dump32(s, s->sh_info);
 342         shdr = &shdr64;
 343     }
 344
 345     ret = fd_write_vmcore(&shdr, shdr_size, s);
 346     if (ret < 0) {
 347         dump_error(s, "dump: failed to write section header table.\n");
 348         return -1;
 349     }
 350
 351     return 0;
 352 }
 353
 354 static int write_data(DumpState *s, void *buf, int length)
 355 {
 356     int ret;
 357
 358     ret = fd_write_vmcore(buf, length, s);
 359     if (ret < 0) {
 360         dump_error(s, "dump: failed to save memory.\n");
 361         return -1;
 362     }
 363
 364     return 0;
 365 }
 366
 367 /* write the memroy to vmcore. 1 page per I/O. */
 368 static int write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
 369                         int64_t size)
 370 {
 371     int64_t i;
 372     int ret;
 373
 374     for (i = 0; i < size / TARGET_PAGE_SIZE; i++) {
 375         ret = write_data(s, block->host_addr + start + i * TARGET_PAGE_SIZE,
 376                          TARGET_PAGE_SIZE);
 377         if (ret < 0) {
 378             return ret;
 379         }
 380     }
 381
 382     if ((size % TARGET_PAGE_SIZE) != 0) {
 383         ret = write_data(s, block->host_addr + start + i * TARGET_PAGE_SIZE,
 384                          size % TARGET_PAGE_SIZE);
 385         if (ret < 0) {
 386             return ret;
 387         }
 388     }
 389
 390     return 0;
 391 }
 392
 393 /* get the memory's offset and size in the vmcore */
 394 static void get_offset_range(hwaddr phys_addr,
 395                              ram_addr_t mapping_length,
 396                              DumpState *s,
 397                              hwaddr *p_offset,
 398                              hwaddr *p_filesz)
 399 {
 400     GuestPhysBlock *block;
 401     hwaddr offset = s->memory_offset;
 402     int64_t size_in_block, start;
 403
 404     /* When the memory is not stored into vmcore, offset will be -1 */
 405     *p_offset = -1;
 406     *p_filesz = 0;
 407
 408     if (s->has_filter) {
 409         if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
 410             return;
 411         }
 412     }
 413
 414     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
 415         if (s->has_filter) {
 416             if (block->target_start >= s->begin + s->length ||
 417                 block->target_end <= s->begin) {
 418                 /* This block is out of the range */
 419                 continue;
 420             }
 421
 422             if (s->begin <= block->target_start) {
 423                 start = block->target_start;
 424             } else {
 425                 start = s->begin;
 426             }
 427
 428             size_in_block = block->target_end - start;
 429             if (s->begin + s->length < block->target_end) {
 430                 size_in_block -= block->target_end - (s->begin + s->length);
 431             }
 432         } else {
 433             start = block->target_start;
 434             size_in_block = block->target_end - block->target_start;
 435         }
 436
 437         if (phys_addr >= start && phys_addr < start + size_in_block) {
 438             *p_offset = phys_addr - start + offset;
 439
 440             /* The offset range mapped from the vmcore file must not spill over
 441              * the GuestPhysBlock, clamp it. The rest of the mapping will be
 442              * zero-filled in memory at load time; see
 443              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
 444              */
 445             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
 446                         mapping_length :
 447                         size_in_block - (phys_addr - start);
 448             return;
 449         }
 450
 451         offset += size_in_block;
 452     }
 453 }
 454
 455 static int write_elf_loads(DumpState *s)
 456 {
 457     hwaddr offset, filesz;
 458     MemoryMapping *memory_mapping;
 459     uint32_t phdr_index = 1;
 460     int ret;
 461     uint32_t max_index;
 462
 463     if (s->have_section) {
 464         max_index = s->sh_info;
 465     } else {
 466         max_index = s->phdr_num;
 467     }
 468
 469     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
 470         get_offset_range(memory_mapping->phys_addr,
 471                          memory_mapping->length,
 472                          s, &offset, &filesz);
 473         if (s->dump_info.d_class == ELFCLASS64) {
 474             ret = write_elf64_load(s, memory_mapping, phdr_index++, offset,
 475                                    filesz);
 476         } else {
 477             ret = write_elf32_load(s, memory_mapping, phdr_index++, offset,
 478                                    filesz);
 479         }
 480
 481         if (ret < 0) {
 482             return -1;
 483         }
 484
 485         if (phdr_index >= max_index) {
 486             break;
 487         }
 488     }
 489
 490     return 0;
 491 }
 492
 493 /* write elf header, PT_NOTE and elf note to vmcore. */
 494 static int dump_begin(DumpState *s)
 495 {
 496     int ret;
 497
 498     /*
 499      * the vmcore's format is:
 500      *   --------------
 501      *   |  elf header |
 502      *   --------------
 503      *   |  PT_NOTE    |
 504      *   --------------
 505      *   |  PT_LOAD    |
 506      *   --------------
 507      *   |  ......     |
 508      *   --------------
 509      *   |  PT_LOAD    |
 510      *   --------------
 511      *   |  sec_hdr    |
 512      *   --------------
 513      *   |  elf note   |
 514      *   --------------
 515      *   |  memory     |
 516      *   --------------
 517      *
 518      * we only know where the memory is saved after we write elf note into
 519      * vmcore.
 520      */
 521
 522     /* write elf header to vmcore */
 523     if (s->dump_info.d_class == ELFCLASS64) {
 524         ret = write_elf64_header(s);
 525     } else {
 526         ret = write_elf32_header(s);
 527     }
 528     if (ret < 0) {
 529         return -1;
 530     }
 531
 532     if (s->dump_info.d_class == ELFCLASS64) {
 533         /* write PT_NOTE to vmcore */
 534         if (write_elf64_note(s) < 0) {
 535             return -1;
 536         }
 537
 538         /* write all PT_LOAD to vmcore */
 539         if (write_elf_loads(s) < 0) {
 540             return -1;
 541         }
 542
 543         /* write section to vmcore */
 544         if (s->have_section) {
 545             if (write_elf_section(s, 1) < 0) {
 546                 return -1;
 547             }
 548         }
 549
 550         /* write notes to vmcore */
 551         if (write_elf64_notes(fd_write_vmcore, s) < 0) {
 552             return -1;
 553         }
 554
 555     } else {
 556         /* write PT_NOTE to vmcore */
 557         if (write_elf32_note(s) < 0) {
 558             return -1;
 559         }
 560
 561         /* write all PT_LOAD to vmcore */
 562         if (write_elf_loads(s) < 0) {
 563             return -1;
 564         }
 565
 566         /* write section to vmcore */
 567         if (s->have_section) {
 568             if (write_elf_section(s, 0) < 0) {
 569                 return -1;
 570             }
 571         }
 572
 573         /* write notes to vmcore */
 574         if (write_elf32_notes(fd_write_vmcore, s) < 0) {
 575             return -1;
 576         }
 577     }
 578
 579     return 0;
 580 }
 581
 582 /* write PT_LOAD to vmcore */
 583 static int dump_completed(DumpState *s)
 584 {
 585     dump_cleanup(s);
 586     return 0;
 587 }
 588
 589 static int get_next_block(DumpState *s, GuestPhysBlock *block)
 590 {
 591     while (1) {
 592         block = QTAILQ_NEXT(block, next);
 593         if (!block) {
 594             /* no more block */
 595             return 1;
 596         }
 597
 598         s->start = 0;
 599         s->next_block = block;
 600         if (s->has_filter) {
 601             if (block->target_start >= s->begin + s->length ||
 602                 block->target_end <= s->begin) {
 603                 /* This block is out of the range */
 604                 continue;
 605             }
 606
 607             if (s->begin > block->target_start) {
 608                 s->start = s->begin - block->target_start;
 609             }
 610         }
 611
 612         return 0;
 613     }
 614 }
 615
 616 /* write all memory to vmcore */
 617 static int dump_iterate(DumpState *s)
 618 {
 619     GuestPhysBlock *block;
 620     int64_t size;
 621     int ret;
 622
 623     while (1) {
 624         block = s->next_block;
 625
 626         size = block->target_end - block->target_start;
 627         if (s->has_filter) {
 628             size -= s->start;
 629             if (s->begin + s->length < block->target_end) {
 630                 size -= block->target_end - (s->begin + s->length);
 631             }
 632         }
 633         ret = write_memory(s, block, s->start, size);
 634         if (ret == -1) {
 635             return ret;
 636         }
 637
 638         ret = get_next_block(s, block);
 639         if (ret == 1) {
 640             dump_completed(s);
 641             return 0;
 642         }
 643     }
 644 }
 645
 646 static int create_vmcore(DumpState *s)
 647 {
 648     int ret;
 649
 650     ret = dump_begin(s);
 651     if (ret < 0) {
 652         return -1;
 653     }
 654
 655     ret = dump_iterate(s);
 656     if (ret < 0) {
 657         return -1;
 658     }
 659
 660     return 0;
 661 }
 662
 663 static int write_start_flat_header(int fd)
 664 {
 665     MakedumpfileHeader *mh;
 666     int ret = 0;
 667
 668     QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER);
 669     mh = g_malloc0(MAX_SIZE_MDF_HEADER);
 670
 671     memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE,
 672            MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE));
 673
 674     mh->type = cpu_to_be64(TYPE_FLAT_HEADER);
 675     mh->version = cpu_to_be64(VERSION_FLAT_HEADER);
 676
 677     size_t written_size;
 678     written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER);
 679     if (written_size != MAX_SIZE_MDF_HEADER) {
 680         ret = -1;
 681     }
 682
 683     g_free(mh);
 684     return ret;
 685 }
 686
 687 static int write_end_flat_header(int fd)
 688 {
 689     MakedumpfileDataHeader mdh;
 690
 691     mdh.offset = END_FLAG_FLAT_HEADER;
 692     mdh.buf_size = END_FLAG_FLAT_HEADER;
 693
 694     size_t written_size;
 695     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 696     if (written_size != sizeof(mdh)) {
 697         return -1;
 698     }
 699
 700     return 0;
 701 }
 702
 703 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
 704 {
 705     size_t written_size;
 706     MakedumpfileDataHeader mdh;
 707
 708     mdh.offset = cpu_to_be64(offset);
 709     mdh.buf_size = cpu_to_be64(size);
 710
 711     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 712     if (written_size != sizeof(mdh)) {
 713         return -1;
 714     }
 715
 716     written_size = qemu_write_full(fd, buf, size);
 717     if (written_size != size) {
 718         return -1;
 719     }
 720
 721     return 0;
 722 }
 723
 724 static int buf_write_note(const void *buf, size_t size, void *opaque)
 725 {
 726     DumpState *s = opaque;
 727
 728     /* note_buf is not enough */
 729     if (s->note_buf_offset + size > s->note_size) {
 730         return -1;
 731     }
 732
 733     memcpy(s->note_buf + s->note_buf_offset, buf, size);
 734
 735     s->note_buf_offset += size;
 736
 737     return 0;
 738 }
 739
 740 /* write common header, sub header and elf note to vmcore */
 741 static int create_header32(DumpState *s)
 742 {
 743     int ret = 0;
 744     DiskDumpHeader32 *dh = NULL;
 745     KdumpSubHeader32 *kh = NULL;
 746     size_t size;
 747     uint32_t block_size;
 748     uint32_t sub_hdr_size;
 749     uint32_t bitmap_blocks;
 750     uint32_t status = 0;
 751     uint64_t offset_note;
 752
 753     /* write common header, the version of kdump-compressed format is 6th */
 754     size = sizeof(DiskDumpHeader32);
 755     dh = g_malloc0(size);
 756
 757     strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
 758     dh->header_version = cpu_to_dump32(s, 6);
 759     block_size = TARGET_PAGE_SIZE;
 760     dh->block_size = cpu_to_dump32(s, block_size);
 761     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
 762     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 763     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
 764     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 765     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
 766     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
 767     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 768     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
 769     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
 770
 771     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 772         status |= DUMP_DH_COMPRESSED_ZLIB;
 773     }
 774 #ifdef CONFIG_LZO
 775     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 776         status |= DUMP_DH_COMPRESSED_LZO;
 777     }
 778 #endif
 779 #ifdef CONFIG_SNAPPY
 780     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 781         status |= DUMP_DH_COMPRESSED_SNAPPY;
 782     }
 783 #endif
 784     dh->status = cpu_to_dump32(s, status);
 785
 786     if (write_buffer(s->fd, 0, dh, size) < 0) {
 787         dump_error(s, "dump: failed to write disk dump header.\n");
 788         ret = -1;
 789         goto out;
 790     }
 791
 792     /* write sub header */
 793     size = sizeof(KdumpSubHeader32);
 794     kh = g_malloc0(size);
 795
 796     /* 64bit max_mapnr_64 */
 797     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
 798     kh->phys_base = cpu_to_dump32(s, PHYS_BASE);
 799     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
 800
 801     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 802     kh->offset_note = cpu_to_dump64(s, offset_note);
 803     kh->note_size = cpu_to_dump32(s, s->note_size);
 804
 805     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 806                      block_size, kh, size) < 0) {
 807         dump_error(s, "dump: failed to write kdump sub header.\n");
 808         ret = -1;
 809         goto out;
 810     }
 811
 812     /* write note */
 813     s->note_buf = g_malloc0(s->note_size);
 814     s->note_buf_offset = 0;
 815
 816     /* use s->note_buf to store notes temporarily */
 817     if (write_elf32_notes(buf_write_note, s) < 0) {
 818         ret = -1;
 819         goto out;
 820     }
 821
 822     if (write_buffer(s->fd, offset_note, s->note_buf,
 823                      s->note_size) < 0) {
 824         dump_error(s, "dump: failed to write notes");
 825         ret = -1;
 826         goto out;
 827     }
 828
 829     /* get offset of dump_bitmap */
 830     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 831                              block_size;
 832
 833     /* get offset of page */
 834     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 835                      block_size;
 836
 837 out:
 838     g_free(dh);
 839     g_free(kh);
 840     g_free(s->note_buf);
 841
 842     return ret;
 843 }
 844
 845 /* write common header, sub header and elf note to vmcore */
 846 static int create_header64(DumpState *s)
 847 {
 848     int ret = 0;
 849     DiskDumpHeader64 *dh = NULL;
 850     KdumpSubHeader64 *kh = NULL;
 851     size_t size;
 852     uint32_t block_size;
 853     uint32_t sub_hdr_size;
 854     uint32_t bitmap_blocks;
 855     uint32_t status = 0;
 856     uint64_t offset_note;
 857
 858     /* write common header, the version of kdump-compressed format is 6th */
 859     size = sizeof(DiskDumpHeader64);
 860     dh = g_malloc0(size);
 861
 862     strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
 863     dh->header_version = cpu_to_dump32(s, 6);
 864     block_size = TARGET_PAGE_SIZE;
 865     dh->block_size = cpu_to_dump32(s, block_size);
 866     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
 867     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 868     dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size);
 869     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 870     dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX));
 871     dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus);
 872     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 873     dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks);
 874     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
 875
 876     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 877         status |= DUMP_DH_COMPRESSED_ZLIB;
 878     }
 879 #ifdef CONFIG_LZO
 880     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 881         status |= DUMP_DH_COMPRESSED_LZO;
 882     }
 883 #endif
 884 #ifdef CONFIG_SNAPPY
 885     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 886         status |= DUMP_DH_COMPRESSED_SNAPPY;
 887     }
 888 #endif
 889     dh->status = cpu_to_dump32(s, status);
 890
 891     if (write_buffer(s->fd, 0, dh, size) < 0) {
 892         dump_error(s, "dump: failed to write disk dump header.\n");
 893         ret = -1;
 894         goto out;
 895     }
 896
 897     /* write sub header */
 898     size = sizeof(KdumpSubHeader64);
 899     kh = g_malloc0(size);
 900
 901     /* 64bit max_mapnr_64 */
 902     kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr);
 903     kh->phys_base = cpu_to_dump64(s, PHYS_BASE);
 904     kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL);
 905
 906     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 907     kh->offset_note = cpu_to_dump64(s, offset_note);
 908     kh->note_size = cpu_to_dump64(s, s->note_size);
 909
 910     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 911                      block_size, kh, size) < 0) {
 912         dump_error(s, "dump: failed to write kdump sub header.\n");
 913         ret = -1;
 914         goto out;
 915     }
 916
 917     /* write note */
 918     s->note_buf = g_malloc0(s->note_size);
 919     s->note_buf_offset = 0;
 920
 921     /* use s->note_buf to store notes temporarily */
 922     if (write_elf64_notes(buf_write_note, s) < 0) {
 923         ret = -1;
 924         goto out;
 925     }
 926
 927     if (write_buffer(s->fd, offset_note, s->note_buf,
 928                      s->note_size) < 0) {
 929         dump_error(s, "dump: failed to write notes");
 930         ret = -1;
 931         goto out;
 932     }
 933
 934     /* get offset of dump_bitmap */
 935     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 936                              block_size;
 937
 938     /* get offset of page */
 939     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 940                      block_size;
 941
 942 out:
 943     g_free(dh);
 944     g_free(kh);
 945     g_free(s->note_buf);
 946
 947     return ret;
 948 }
 949
 950 static int write_dump_header(DumpState *s)
 951 {
 952     if (s->dump_info.d_class == ELFCLASS32) {
 953         return create_header32(s);
 954     } else {
 955         return create_header64(s);
 956     }
 957 }
 958
 959 /*
 960  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
 961  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
 962  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
 963  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
 964  * vmcore, ie. synchronizing un-sync bit into vmcore.
 965  */
 966 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
 967                            uint8_t *buf, DumpState *s)
 968 {
 969     off_t old_offset, new_offset;
 970     off_t offset_bitmap1, offset_bitmap2;
 971     uint32_t byte, bit;
 972
 973     /* should not set the previous place */
 974     assert(last_pfn <= pfn);
 975
 976     /*
 977      * if the bit needed to be set is not cached in buf, flush the data in buf
 978      * to vmcore firstly.
 979      * making new_offset be bigger than old_offset can also sync remained data
 980      * into vmcore.
 981      */
 982     old_offset = BUFSIZE_BITMAP * (last_pfn / PFN_BUFBITMAP);
 983     new_offset = BUFSIZE_BITMAP * (pfn / PFN_BUFBITMAP);
 984
 985     while (old_offset < new_offset) {
 986         /* calculate the offset and write dump_bitmap */
 987         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
 988         if (write_buffer(s->fd, offset_bitmap1, buf,
 989                          BUFSIZE_BITMAP) < 0) {
 990             return -1;
 991         }
 992
 993         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
 994         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
 995                          old_offset;
 996         if (write_buffer(s->fd, offset_bitmap2, buf,
 997                          BUFSIZE_BITMAP) < 0) {
 998             return -1;
 999         }
1000
1001         memset(buf, 0, BUFSIZE_BITMAP);
1002         old_offset += BUFSIZE_BITMAP;
1003     }
1004
1005     /* get the exact place of the bit in the buf, and set it */
1006     byte = (pfn % PFN_BUFBITMAP) / CHAR_BIT;
1007     bit = (pfn % PFN_BUFBITMAP) % CHAR_BIT;
1008     if (value) {
1009         buf[byte] |= 1u << bit;
1010     } else {
1011         buf[byte] &= ~(1u << bit);
1012     }
1013
1014     return 0;
1015 }
1016
1017 /*
1018  * exam every page and return the page frame number and the address of the page.
1019  * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1020  * blocks, so block->target_start and block->target_end should be interal
1021  * multiples of the target page size.
1022  */
1023 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1024                           uint8_t **bufptr, DumpState *s)
1025 {
1026     GuestPhysBlock *block = *blockptr;
1027     hwaddr addr;
1028     uint8_t *buf;
1029
1030     /* block == NULL means the start of the iteration */
1031     if (!block) {
1032         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1033         *blockptr = block;
1034         assert((block->target_start & ~TARGET_PAGE_MASK) == 0);
1035         assert((block->target_end & ~TARGET_PAGE_MASK) == 0);
1036         *pfnptr = paddr_to_pfn(block->target_start);
1037         if (bufptr) {
1038             *bufptr = block->host_addr;
1039         }
1040         return true;
1041     }
1042
1043     *pfnptr = *pfnptr + 1;
1044     addr = pfn_to_paddr(*pfnptr);
1045
1046     if ((addr >= block->target_start) &&
1047         (addr + TARGET_PAGE_SIZE <= block->target_end)) {
1048         buf = block->host_addr + (addr - block->target_start);
1049     } else {
1050         /* the next page is in the next block */
1051         block = QTAILQ_NEXT(block, next);
1052         *blockptr = block;
1053         if (!block) {
1054             return false;
1055         }
1056         assert((block->target_start & ~TARGET_PAGE_MASK) == 0);
1057         assert((block->target_end & ~TARGET_PAGE_MASK) == 0);
1058         *pfnptr = paddr_to_pfn(block->target_start);
1059         buf = block->host_addr;
1060     }
1061
1062     if (bufptr) {
1063         *bufptr = buf;
1064     }
1065
1066     return true;
1067 }
1068
1069 static int write_dump_bitmap(DumpState *s)
1070 {
1071     int ret = 0;
1072     uint64_t last_pfn, pfn;
1073     void *dump_bitmap_buf;
1074     size_t num_dumpable;
1075     GuestPhysBlock *block_iter = NULL;
1076
1077     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1078     dump_bitmap_buf = g_malloc0(BUFSIZE_BITMAP);
1079
1080     num_dumpable = 0;
1081     last_pfn = 0;
1082
1083     /*
1084      * exam memory page by page, and set the bit in dump_bitmap corresponded
1085      * to the existing page.
1086      */
1087     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1088         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1089         if (ret < 0) {
1090             dump_error(s, "dump: failed to set dump_bitmap.\n");
1091             ret = -1;
1092             goto out;
1093         }
1094
1095         last_pfn = pfn;
1096         num_dumpable++;
1097     }
1098
1099     /*
1100      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1101      * set last_pfn + PFN_BUFBITMAP to 0 and those set but un-sync bit will be
1102      * synchronized into vmcore.
1103      */
1104     if (num_dumpable > 0) {
1105         ret = set_dump_bitmap(last_pfn, last_pfn + PFN_BUFBITMAP, false,
1106                               dump_bitmap_buf, s);
1107         if (ret < 0) {
1108             dump_error(s, "dump: failed to sync dump_bitmap.\n");
1109             ret = -1;
1110             goto out;
1111         }
1112     }
1113
1114     /* number of dumpable pages that will be dumped later */
1115     s->num_dumpable = num_dumpable;
1116
1117 out:
1118     g_free(dump_bitmap_buf);
1119
1120     return ret;
1121 }
1122
1123 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1124                                off_t offset)
1125 {
1126     data_cache->fd = s->fd;
1127     data_cache->data_size = 0;
1128     data_cache->buf_size = BUFSIZE_DATA_CACHE;
1129     data_cache->buf = g_malloc0(BUFSIZE_DATA_CACHE);
1130     data_cache->offset = offset;
1131 }
1132
1133 static int write_cache(DataCache *dc, const void *buf, size_t size,
1134                        bool flag_sync)
1135 {
1136     /*
1137      * dc->buf_size should not be less than size, otherwise dc will never be
1138      * enough
1139      */
1140     assert(size <= dc->buf_size);
1141
1142     /*
1143      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1144      * otherwise check if the space is enough for caching data in buf, if not,
1145      * write the data in dc->buf to dc->fd and reset dc->buf
1146      */
1147     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1148         (flag_sync && dc->data_size > 0)) {
1149         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1150             return -1;
1151         }
1152
1153         dc->offset += dc->data_size;
1154         dc->data_size = 0;
1155     }
1156
1157     if (!flag_sync) {
1158         memcpy(dc->buf + dc->data_size, buf, size);
1159         dc->data_size += size;
1160     }
1161
1162     return 0;
1163 }
1164
1165 static void free_data_cache(DataCache *data_cache)
1166 {
1167     g_free(data_cache->buf);
1168 }
1169
1170 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1171 {
1172     switch (flag_compress) {
1173     case DUMP_DH_COMPRESSED_ZLIB:
1174         return compressBound(page_size);
1175
1176     case DUMP_DH_COMPRESSED_LZO:
1177         /*
1178          * LZO will expand incompressible data by a little amount. Please check
1179          * the following URL to see the expansion calculation:
1180          * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1181          */
1182         return page_size + page_size / 16 + 64 + 3;
1183
1184 #ifdef CONFIG_SNAPPY
1185     case DUMP_DH_COMPRESSED_SNAPPY:
1186         return snappy_max_compressed_length(page_size);
1187 #endif
1188     }
1189     return 0;
1190 }
1191
1192 /*
1193  * check if the page is all 0
1194  */
1195 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1196 {
1197     return buffer_is_zero(buf, page_size);
1198 }
1199
1200 static int write_dump_pages(DumpState *s)
1201 {
1202     int ret = 0;
1203     DataCache page_desc, page_data;
1204     size_t len_buf_out, size_out;
1205 #ifdef CONFIG_LZO
1206     lzo_bytep wrkmem = NULL;
1207 #endif
1208     uint8_t *buf_out = NULL;
1209     off_t offset_desc, offset_data;
1210     PageDescriptor pd, pd_zero;
1211     uint8_t *buf;
1212     GuestPhysBlock *block_iter = NULL;
1213     uint64_t pfn_iter;
1214
1215     /* get offset of page_desc and page_data in dump file */
1216     offset_desc = s->offset_page;
1217     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1218
1219     prepare_data_cache(&page_desc, s, offset_desc);
1220     prepare_data_cache(&page_data, s, offset_data);
1221
1222     /* prepare buffer to store compressed data */
1223     len_buf_out = get_len_buf_out(TARGET_PAGE_SIZE, s->flag_compress);
1224     assert(len_buf_out != 0);
1225
1226 #ifdef CONFIG_LZO
1227     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1228 #endif
1229
1230     buf_out = g_malloc(len_buf_out);
1231
1232     /*
1233      * init zero page's page_desc and page_data, because every zero page
1234      * uses the same page_data
1235      */
1236     pd_zero.size = cpu_to_dump32(s, TARGET_PAGE_SIZE);
1237     pd_zero.flags = cpu_to_dump32(s, 0);
1238     pd_zero.offset = cpu_to_dump64(s, offset_data);
1239     pd_zero.page_flags = cpu_to_dump64(s, 0);
1240     buf = g_malloc0(TARGET_PAGE_SIZE);
1241     ret = write_cache(&page_data, buf, TARGET_PAGE_SIZE, false);
1242     g_free(buf);
1243     if (ret < 0) {
1244         dump_error(s, "dump: failed to write page data(zero page).\n");
1245         goto out;
1246     }
1247
1248     offset_data += TARGET_PAGE_SIZE;
1249
1250     /*
1251      * dump memory to vmcore page by page. zero page will all be resided in the
1252      * first page of page section
1253      */
1254     while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1255         /* check zero page */
1256         if (is_zero_page(buf, TARGET_PAGE_SIZE)) {
1257             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1258                               false);
1259             if (ret < 0) {
1260                 dump_error(s, "dump: failed to write page desc.\n");
1261                 goto out;
1262             }
1263         } else {
1264             /*
1265              * not zero page, then:
1266              * 1. compress the page
1267              * 2. write the compressed page into the cache of page_data
1268              * 3. get page desc of the compressed page and write it into the
1269              *    cache of page_desc
1270              *
1271              * only one compression format will be used here, for
1272              * s->flag_compress is set. But when compression fails to work,
1273              * we fall back to save in plaintext.
1274              */
1275              size_out = len_buf_out;
1276              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1277                     (compress2(buf_out, (uLongf *)&size_out, buf,
1278                                TARGET_PAGE_SIZE, Z_BEST_SPEED) == Z_OK) &&
1279                     (size_out < TARGET_PAGE_SIZE)) {
1280                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB);
1281                 pd.size  = cpu_to_dump32(s, size_out);
1282
1283                 ret = write_cache(&page_data, buf_out, size_out, false);
1284                 if (ret < 0) {
1285                     dump_error(s, "dump: failed to write page data.\n");
1286                     goto out;
1287                 }
1288 #ifdef CONFIG_LZO
1289             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1290                     (lzo1x_1_compress(buf, TARGET_PAGE_SIZE, buf_out,
1291                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1292                     (size_out < TARGET_PAGE_SIZE)) {
1293                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO);
1294                 pd.size  = cpu_to_dump32(s, size_out);
1295
1296                 ret = write_cache(&page_data, buf_out, size_out, false);
1297                 if (ret < 0) {
1298                     dump_error(s, "dump: failed to write page data.\n");
1299                     goto out;
1300                 }
1301 #endif
1302 #ifdef CONFIG_SNAPPY
1303             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1304                     (snappy_compress((char *)buf, TARGET_PAGE_SIZE,
1305                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1306                     (size_out < TARGET_PAGE_SIZE)) {
1307                 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY);
1308                 pd.size  = cpu_to_dump32(s, size_out);
1309
1310                 ret = write_cache(&page_data, buf_out, size_out, false);
1311                 if (ret < 0) {
1312                     dump_error(s, "dump: failed to write page data.\n");
1313                     goto out;
1314                 }
1315 #endif
1316             } else {
1317                 /*
1318                  * fall back to save in plaintext, size_out should be
1319                  * assigned TARGET_PAGE_SIZE
1320                  */
1321                 pd.flags = cpu_to_dump32(s, 0);
1322                 size_out = TARGET_PAGE_SIZE;
1323                 pd.size = cpu_to_dump32(s, size_out);
1324
1325                 ret = write_cache(&page_data, buf, TARGET_PAGE_SIZE, false);
1326                 if (ret < 0) {
1327                     dump_error(s, "dump: failed to write page data.\n");
1328                     goto out;
1329                 }
1330             }
1331
1332             /* get and write page desc here */
1333             pd.page_flags = cpu_to_dump64(s, 0);
1334             pd.offset = cpu_to_dump64(s, offset_data);
1335             offset_data += size_out;
1336
1337             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1338             if (ret < 0) {
1339                 dump_error(s, "dump: failed to write page desc.\n");
1340                 goto out;
1341             }
1342         }
1343     }
1344
1345     ret = write_cache(&page_desc, NULL, 0, true);
1346     if (ret < 0) {
1347         dump_error(s, "dump: failed to sync cache for page_desc.\n");
1348         goto out;
1349     }
1350     ret = write_cache(&page_data, NULL, 0, true);
1351     if (ret < 0) {
1352         dump_error(s, "dump: failed to sync cache for page_data.\n");
1353         goto out;
1354     }
1355
1356 out:
1357     free_data_cache(&page_desc);
1358     free_data_cache(&page_data);
1359
1360 #ifdef CONFIG_LZO
1361     g_free(wrkmem);
1362 #endif
1363
1364     g_free(buf_out);
1365
1366     return ret;
1367 }
1368
1369 static int create_kdump_vmcore(DumpState *s)
1370 {
1371     int ret;
1372
1373     /*
1374      * the kdump-compressed format is:
1375      *                                               File offset
1376      *  +------------------------------------------+ 0x0
1377      *  |    main header (struct disk_dump_header) |
1378      *  |------------------------------------------+ block 1
1379      *  |    sub header (struct kdump_sub_header)  |
1380      *  |------------------------------------------+ block 2
1381      *  |            1st-dump_bitmap               |
1382      *  |------------------------------------------+ block 2 + X blocks
1383      *  |            2nd-dump_bitmap               | (aligned by block)
1384      *  |------------------------------------------+ block 2 + 2 * X blocks
1385      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1386      *  |  page desc for pfn 1 (struct page_desc)  |
1387      *  |                    :                     |
1388      *  |------------------------------------------| (not aligned by block)
1389      *  |         page data (pfn 0)                |
1390      *  |         page data (pfn 1)                |
1391      *  |                    :                     |
1392      *  +------------------------------------------+
1393      */
1394
1395     ret = write_start_flat_header(s->fd);
1396     if (ret < 0) {
1397         dump_error(s, "dump: failed to write start flat header.\n");
1398         return -1;
1399     }
1400
1401     ret = write_dump_header(s);
1402     if (ret < 0) {
1403         return -1;
1404     }
1405
1406     ret = write_dump_bitmap(s);
1407     if (ret < 0) {
1408         return -1;
1409     }
1410
1411     ret = write_dump_pages(s);
1412     if (ret < 0) {
1413         return -1;
1414     }
1415
1416     ret = write_end_flat_header(s->fd);
1417     if (ret < 0) {
1418         dump_error(s, "dump: failed to write end flat header.\n");
1419         return -1;
1420     }
1421
1422     dump_completed(s);
1423
1424     return 0;
1425 }
1426
1427 static ram_addr_t get_start_block(DumpState *s)
1428 {
1429     GuestPhysBlock *block;
1430
1431     if (!s->has_filter) {
1432         s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1433         return 0;
1434     }
1435
1436     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1437         if (block->target_start >= s->begin + s->length ||
1438             block->target_end <= s->begin) {
1439             /* This block is out of the range */
1440             continue;
1441         }
1442
1443         s->next_block = block;
1444         if (s->begin > block->target_start) {
1445             s->start = s->begin - block->target_start;
1446         } else {
1447             s->start = 0;
1448         }
1449         return s->start;
1450     }
1451
1452     return -1;
1453 }
1454
1455 static void get_max_mapnr(DumpState *s)
1456 {
1457     GuestPhysBlock *last_block;
1458
1459     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head, GuestPhysBlockHead);
1460     s->max_mapnr = paddr_to_pfn(last_block->target_end);
1461 }
1462
1463 static int dump_init(DumpState *s, int fd, bool has_format,
1464                      DumpGuestMemoryFormat format, bool paging, bool has_filter,
1465                      int64_t begin, int64_t length, Error **errp)
1466 {
1467     CPUState *cpu;
1468     int nr_cpus;
1469     Error *err = NULL;
1470     int ret;
1471
1472     /* kdump-compressed is conflict with paging and filter */
1473     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1474         assert(!paging && !has_filter);
1475     }
1476
1477     if (runstate_is_running()) {
1478         vm_stop(RUN_STATE_SAVE_VM);
1479         s->resume = true;
1480     } else {
1481         s->resume = false;
1482     }
1483
1484     /* If we use KVM, we should synchronize the registers before we get dump
1485      * info or physmap info.
1486      */
1487     cpu_synchronize_all_states();
1488     nr_cpus = 0;
1489     CPU_FOREACH(cpu) {
1490         nr_cpus++;
1491     }
1492
1493     s->fd = fd;
1494     s->has_filter = has_filter;
1495     s->begin = begin;
1496     s->length = length;
1497
1498     memory_mapping_list_init(&s->list);
1499
1500     guest_phys_blocks_init(&s->guest_phys_blocks);
1501     guest_phys_blocks_append(&s->guest_phys_blocks);
1502
1503     s->start = get_start_block(s);
1504     if (s->start == -1) {
1505         error_set(errp, QERR_INVALID_PARAMETER, "begin");
1506         goto cleanup;
1507     }
1508
1509     /* get dump info: endian, class and architecture.
1510      * If the target architecture is not supported, cpu_get_dump_info() will
1511      * return -1.
1512      */
1513     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1514     if (ret < 0) {
1515         error_set(errp, QERR_UNSUPPORTED);
1516         goto cleanup;
1517     }
1518
1519     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1520                                      s->dump_info.d_machine, nr_cpus);
1521     if (s->note_size < 0) {
1522         error_set(errp, QERR_UNSUPPORTED);
1523         goto cleanup;
1524     }
1525
1526     /* get memory mapping */
1527     if (paging) {
1528         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1529         if (err != NULL) {
1530             error_propagate(errp, err);
1531             goto cleanup;
1532         }
1533     } else {
1534         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1535     }
1536
1537     s->nr_cpus = nr_cpus;
1538
1539     get_max_mapnr(s);
1540
1541     uint64_t tmp;
1542     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT), TARGET_PAGE_SIZE);
1543     s->len_dump_bitmap = tmp * TARGET_PAGE_SIZE;
1544
1545     /* init for kdump-compressed format */
1546     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1547         switch (format) {
1548         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1549             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1550             break;
1551
1552         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1553 #ifdef CONFIG_LZO
1554             if (lzo_init() != LZO_E_OK) {
1555                 error_setg(errp, "failed to initialize the LZO library");
1556                 goto cleanup;
1557             }
1558 #endif
1559             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1560             break;
1561
1562         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1563             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1564             break;
1565
1566         default:
1567             s->flag_compress = 0;
1568         }
1569
1570         return 0;
1571     }
1572
1573     if (s->has_filter) {
1574         memory_mapping_filter(&s->list, s->begin, s->length);
1575     }
1576
1577     /*
1578      * calculate phdr_num
1579      *
1580      * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1581      */
1582     s->phdr_num = 1; /* PT_NOTE */
1583     if (s->list.num < UINT16_MAX - 2) {
1584         s->phdr_num += s->list.num;
1585         s->have_section = false;
1586     } else {
1587         s->have_section = true;
1588         s->phdr_num = PN_XNUM;
1589         s->sh_info = 1; /* PT_NOTE */
1590
1591         /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1592         if (s->list.num <= UINT32_MAX - 1) {
1593             s->sh_info += s->list.num;
1594         } else {
1595             s->sh_info = UINT32_MAX;
1596         }
1597     }
1598
1599     if (s->dump_info.d_class == ELFCLASS64) {
1600         if (s->have_section) {
1601             s->memory_offset = sizeof(Elf64_Ehdr) +
1602                                sizeof(Elf64_Phdr) * s->sh_info +
1603                                sizeof(Elf64_Shdr) + s->note_size;
1604         } else {
1605             s->memory_offset = sizeof(Elf64_Ehdr) +
1606                                sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1607         }
1608     } else {
1609         if (s->have_section) {
1610             s->memory_offset = sizeof(Elf32_Ehdr) +
1611                                sizeof(Elf32_Phdr) * s->sh_info +
1612                                sizeof(Elf32_Shdr) + s->note_size;
1613         } else {
1614             s->memory_offset = sizeof(Elf32_Ehdr) +
1615                                sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1616         }
1617     }
1618
1619     return 0;
1620
1621 cleanup:
1622     dump_cleanup(s);
1623     return -1;
1624 }
1625
1626 void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
1627                            int64_t begin, bool has_length,
1628                            int64_t length, bool has_format,
1629                            DumpGuestMemoryFormat format, Error **errp)
1630 {
1631     const char *p;
1632     int fd = -1;
1633     DumpState *s;
1634     int ret;
1635
1636     /*
1637      * kdump-compressed format need the whole memory dumped, so paging or
1638      * filter is not supported here.
1639      */
1640     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1641         (paging || has_begin || has_length)) {
1642         error_setg(errp, "kdump-compressed format doesn't support paging or "
1643                          "filter");
1644         return;
1645     }
1646     if (has_begin && !has_length) {
1647         error_set(errp, QERR_MISSING_PARAMETER, "length");
1648         return;
1649     }
1650     if (!has_begin && has_length) {
1651         error_set(errp, QERR_MISSING_PARAMETER, "begin");
1652         return;
1653     }
1654
1655     /* check whether lzo/snappy is supported */
1656 #ifndef CONFIG_LZO
1657     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1658         error_setg(errp, "kdump-lzo is not available now");
1659         return;
1660     }
1661 #endif
1662
1663 #ifndef CONFIG_SNAPPY
1664     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1665         error_setg(errp, "kdump-snappy is not available now");
1666         return;
1667     }
1668 #endif
1669
1670 #if !defined(WIN32)
1671     if (strstart(file, "fd:", &p)) {
1672         fd = monitor_get_fd(cur_mon, p, errp);
1673         if (fd == -1) {
1674             return;
1675         }
1676     }
1677 #endif
1678
1679     if  (strstart(file, "file:", &p)) {
1680         fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1681         if (fd < 0) {
1682             error_setg_file_open(errp, errno, p);
1683             return;
1684         }
1685     }
1686
1687     if (fd == -1) {
1688         error_set(errp, QERR_INVALID_PARAMETER, "protocol");
1689         return;
1690     }
1691
1692     s = g_malloc0(sizeof(DumpState));
1693
1694     ret = dump_init(s, fd, has_format, format, paging, has_begin,
1695                     begin, length, errp);
1696     if (ret < 0) {
1697         g_free(s);
1698         return;
1699     }
1700
1701     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1702         if (create_kdump_vmcore(s) < 0) {
1703             error_set(errp, QERR_IO_ERROR);
1704         }
1705     } else {
1706         if (create_vmcore(s) < 0) {
1707             error_set(errp, QERR_IO_ERROR);
1708         }
1709     }
1710
1711     g_free(s);
1712 }
1713
1714 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
1715 {
1716     DumpGuestMemoryFormatList *item;
1717     DumpGuestMemoryCapability *cap =
1718                                   g_malloc0(sizeof(DumpGuestMemoryCapability));
1719
1720     /* elf is always available */
1721     item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1722     cap->formats = item;
1723     item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
1724
1725     /* kdump-zlib is always available */
1726     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1727     item = item->next;
1728     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
1729
1730     /* add new item if kdump-lzo is available */
1731 #ifdef CONFIG_LZO
1732     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1733     item = item->next;
1734     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
1735 #endif
1736
1737     /* add new item if kdump-snappy is available */
1738 #ifdef CONFIG_SNAPPY
1739     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1740     item = item->next;
1741     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
1742 #endif
1743
1744     return cap;
1745 }