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     size_t page_size;           /* guest's page size */
  94     uint32_t page_shift;        /* guest's page shift */
  95     uint64_t max_mapnr;         /* the biggest guest's phys-mem's number */
  96     size_t len_dump_bitmap;     /* the size of the place used to store
  97                                    dump_bitmap in vmcore */
  98     off_t offset_dump_bitmap;   /* offset of dump_bitmap part in vmcore */
  99     off_t offset_page;          /* offset of page part in vmcore */
 100     size_t num_dumpable;        /* number of page that can be dumped */
 101     uint32_t flag_compress;     /* indicate the compression format */
 102 } DumpState;
 103
 104 static int dump_cleanup(DumpState *s)
 105 {
 106     int ret = 0;
 107
 108     guest_phys_blocks_free(&s->guest_phys_blocks);
 109     memory_mapping_list_free(&s->list);
 110     if (s->fd != -1) {
 111         close(s->fd);
 112     }
 113     if (s->resume) {
 114         vm_start();
 115     }
 116
 117     return ret;
 118 }
 119
 120 static void dump_error(DumpState *s, const char *reason)
 121 {
 122     dump_cleanup(s);
 123 }
 124
 125 static int fd_write_vmcore(const void *buf, size_t size, void *opaque)
 126 {
 127     DumpState *s = opaque;
 128     size_t written_size;
 129
 130     written_size = qemu_write_full(s->fd, buf, size);
 131     if (written_size != size) {
 132         return -1;
 133     }
 134
 135     return 0;
 136 }
 137
 138 static int write_elf64_header(DumpState *s)
 139 {
 140     Elf64_Ehdr elf_header;
 141     int ret;
 142     int endian = s->dump_info.d_endian;
 143
 144     memset(&elf_header, 0, sizeof(Elf64_Ehdr));
 145     memcpy(&elf_header, ELFMAG, SELFMAG);
 146     elf_header.e_ident[EI_CLASS] = ELFCLASS64;
 147     elf_header.e_ident[EI_DATA] = s->dump_info.d_endian;
 148     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
 149     elf_header.e_type = cpu_convert_to_target16(ET_CORE, endian);
 150     elf_header.e_machine = cpu_convert_to_target16(s->dump_info.d_machine,
 151                                                    endian);
 152     elf_header.e_version = cpu_convert_to_target32(EV_CURRENT, endian);
 153     elf_header.e_ehsize = cpu_convert_to_target16(sizeof(elf_header), endian);
 154     elf_header.e_phoff = cpu_convert_to_target64(sizeof(Elf64_Ehdr), endian);
 155     elf_header.e_phentsize = cpu_convert_to_target16(sizeof(Elf64_Phdr),
 156                                                      endian);
 157     elf_header.e_phnum = cpu_convert_to_target16(s->phdr_num, endian);
 158     if (s->have_section) {
 159         uint64_t shoff = sizeof(Elf64_Ehdr) + sizeof(Elf64_Phdr) * s->sh_info;
 160
 161         elf_header.e_shoff = cpu_convert_to_target64(shoff, endian);
 162         elf_header.e_shentsize = cpu_convert_to_target16(sizeof(Elf64_Shdr),
 163                                                          endian);
 164         elf_header.e_shnum = cpu_convert_to_target16(1, endian);
 165     }
 166
 167     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
 168     if (ret < 0) {
 169         dump_error(s, "dump: failed to write elf header.\n");
 170         return -1;
 171     }
 172
 173     return 0;
 174 }
 175
 176 static int write_elf32_header(DumpState *s)
 177 {
 178     Elf32_Ehdr elf_header;
 179     int ret;
 180     int endian = s->dump_info.d_endian;
 181
 182     memset(&elf_header, 0, sizeof(Elf32_Ehdr));
 183     memcpy(&elf_header, ELFMAG, SELFMAG);
 184     elf_header.e_ident[EI_CLASS] = ELFCLASS32;
 185     elf_header.e_ident[EI_DATA] = endian;
 186     elf_header.e_ident[EI_VERSION] = EV_CURRENT;
 187     elf_header.e_type = cpu_convert_to_target16(ET_CORE, endian);
 188     elf_header.e_machine = cpu_convert_to_target16(s->dump_info.d_machine,
 189                                                    endian);
 190     elf_header.e_version = cpu_convert_to_target32(EV_CURRENT, endian);
 191     elf_header.e_ehsize = cpu_convert_to_target16(sizeof(elf_header), endian);
 192     elf_header.e_phoff = cpu_convert_to_target32(sizeof(Elf32_Ehdr), endian);
 193     elf_header.e_phentsize = cpu_convert_to_target16(sizeof(Elf32_Phdr),
 194                                                      endian);
 195     elf_header.e_phnum = cpu_convert_to_target16(s->phdr_num, endian);
 196     if (s->have_section) {
 197         uint32_t shoff = sizeof(Elf32_Ehdr) + sizeof(Elf32_Phdr) * s->sh_info;
 198
 199         elf_header.e_shoff = cpu_convert_to_target32(shoff, endian);
 200         elf_header.e_shentsize = cpu_convert_to_target16(sizeof(Elf32_Shdr),
 201                                                          endian);
 202         elf_header.e_shnum = cpu_convert_to_target16(1, endian);
 203     }
 204
 205     ret = fd_write_vmcore(&elf_header, sizeof(elf_header), s);
 206     if (ret < 0) {
 207         dump_error(s, "dump: failed to write elf header.\n");
 208         return -1;
 209     }
 210
 211     return 0;
 212 }
 213
 214 static int write_elf64_load(DumpState *s, MemoryMapping *memory_mapping,
 215                             int phdr_index, hwaddr offset,
 216                             hwaddr filesz)
 217 {
 218     Elf64_Phdr phdr;
 219     int ret;
 220     int endian = s->dump_info.d_endian;
 221
 222     memset(&phdr, 0, sizeof(Elf64_Phdr));
 223     phdr.p_type = cpu_convert_to_target32(PT_LOAD, endian);
 224     phdr.p_offset = cpu_convert_to_target64(offset, endian);
 225     phdr.p_paddr = cpu_convert_to_target64(memory_mapping->phys_addr, endian);
 226     phdr.p_filesz = cpu_convert_to_target64(filesz, endian);
 227     phdr.p_memsz = cpu_convert_to_target64(memory_mapping->length, endian);
 228     phdr.p_vaddr = cpu_convert_to_target64(memory_mapping->virt_addr, endian);
 229
 230     assert(memory_mapping->length >= filesz);
 231
 232     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
 233     if (ret < 0) {
 234         dump_error(s, "dump: failed to write program header table.\n");
 235         return -1;
 236     }
 237
 238     return 0;
 239 }
 240
 241 static int write_elf32_load(DumpState *s, MemoryMapping *memory_mapping,
 242                             int phdr_index, hwaddr offset,
 243                             hwaddr filesz)
 244 {
 245     Elf32_Phdr phdr;
 246     int ret;
 247     int endian = s->dump_info.d_endian;
 248
 249     memset(&phdr, 0, sizeof(Elf32_Phdr));
 250     phdr.p_type = cpu_convert_to_target32(PT_LOAD, endian);
 251     phdr.p_offset = cpu_convert_to_target32(offset, endian);
 252     phdr.p_paddr = cpu_convert_to_target32(memory_mapping->phys_addr, endian);
 253     phdr.p_filesz = cpu_convert_to_target32(filesz, endian);
 254     phdr.p_memsz = cpu_convert_to_target32(memory_mapping->length, endian);
 255     phdr.p_vaddr = cpu_convert_to_target32(memory_mapping->virt_addr, endian);
 256
 257     assert(memory_mapping->length >= filesz);
 258
 259     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
 260     if (ret < 0) {
 261         dump_error(s, "dump: failed to write program header table.\n");
 262         return -1;
 263     }
 264
 265     return 0;
 266 }
 267
 268 static int write_elf64_note(DumpState *s)
 269 {
 270     Elf64_Phdr phdr;
 271     int endian = s->dump_info.d_endian;
 272     hwaddr begin = s->memory_offset - s->note_size;
 273     int ret;
 274
 275     memset(&phdr, 0, sizeof(Elf64_Phdr));
 276     phdr.p_type = cpu_convert_to_target32(PT_NOTE, endian);
 277     phdr.p_offset = cpu_convert_to_target64(begin, endian);
 278     phdr.p_paddr = 0;
 279     phdr.p_filesz = cpu_convert_to_target64(s->note_size, endian);
 280     phdr.p_memsz = cpu_convert_to_target64(s->note_size, endian);
 281     phdr.p_vaddr = 0;
 282
 283     ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s);
 284     if (ret < 0) {
 285         dump_error(s, "dump: failed to write program header table.\n");
 286         return -1;
 287     }
 288
 289     return 0;
 290 }
 291
 292 static inline int cpu_index(CPUState *cpu)
 293 {
 294     return cpu->cpu_index + 1;
 295 }
 296
 297 static int write_elf64_notes(WriteCoreDumpFunction f, DumpState *s)
 298 {
 299     CPUState *cpu;
 300     int ret;
 301     int id;
 302
 303     CPU_FOREACH(cpu) {
 304         id = cpu_index(cpu);
 305         ret = cpu_write_elf64_note(f, cpu, id, s);
 306         if (ret < 0) {
 307             dump_error(s, "dump: failed to write elf notes.\n");
 308             return -1;
 309         }
 310     }
 311
 312     CPU_FOREACH(cpu) {
 313         ret = cpu_write_elf64_qemunote(f, cpu, s);
 314         if (ret < 0) {
 315             dump_error(s, "dump: failed to write CPU status.\n");
 316             return -1;
 317         }
 318     }
 319
 320     return 0;
 321 }
 322
 323 static int write_elf32_note(DumpState *s)
 324 {
 325     hwaddr begin = s->memory_offset - s->note_size;
 326     Elf32_Phdr phdr;
 327     int endian = s->dump_info.d_endian;
 328     int ret;
 329
 330     memset(&phdr, 0, sizeof(Elf32_Phdr));
 331     phdr.p_type = cpu_convert_to_target32(PT_NOTE, endian);
 332     phdr.p_offset = cpu_convert_to_target32(begin, endian);
 333     phdr.p_paddr = 0;
 334     phdr.p_filesz = cpu_convert_to_target32(s->note_size, endian);
 335     phdr.p_memsz = cpu_convert_to_target32(s->note_size, endian);
 336     phdr.p_vaddr = 0;
 337
 338     ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s);
 339     if (ret < 0) {
 340         dump_error(s, "dump: failed to write program header table.\n");
 341         return -1;
 342     }
 343
 344     return 0;
 345 }
 346
 347 static int write_elf32_notes(WriteCoreDumpFunction f, DumpState *s)
 348 {
 349     CPUState *cpu;
 350     int ret;
 351     int id;
 352
 353     CPU_FOREACH(cpu) {
 354         id = cpu_index(cpu);
 355         ret = cpu_write_elf32_note(f, cpu, id, s);
 356         if (ret < 0) {
 357             dump_error(s, "dump: failed to write elf notes.\n");
 358             return -1;
 359         }
 360     }
 361
 362     CPU_FOREACH(cpu) {
 363         ret = cpu_write_elf32_qemunote(f, cpu, s);
 364         if (ret < 0) {
 365             dump_error(s, "dump: failed to write CPU status.\n");
 366             return -1;
 367         }
 368     }
 369
 370     return 0;
 371 }
 372
 373 static int write_elf_section(DumpState *s, int type)
 374 {
 375     Elf32_Shdr shdr32;
 376     Elf64_Shdr shdr64;
 377     int endian = s->dump_info.d_endian;
 378     int shdr_size;
 379     void *shdr;
 380     int ret;
 381
 382     if (type == 0) {
 383         shdr_size = sizeof(Elf32_Shdr);
 384         memset(&shdr32, 0, shdr_size);
 385         shdr32.sh_info = cpu_convert_to_target32(s->sh_info, endian);
 386         shdr = &shdr32;
 387     } else {
 388         shdr_size = sizeof(Elf64_Shdr);
 389         memset(&shdr64, 0, shdr_size);
 390         shdr64.sh_info = cpu_convert_to_target32(s->sh_info, endian);
 391         shdr = &shdr64;
 392     }
 393
 394     ret = fd_write_vmcore(&shdr, shdr_size, s);
 395     if (ret < 0) {
 396         dump_error(s, "dump: failed to write section header table.\n");
 397         return -1;
 398     }
 399
 400     return 0;
 401 }
 402
 403 static int write_data(DumpState *s, void *buf, int length)
 404 {
 405     int ret;
 406
 407     ret = fd_write_vmcore(buf, length, s);
 408     if (ret < 0) {
 409         dump_error(s, "dump: failed to save memory.\n");
 410         return -1;
 411     }
 412
 413     return 0;
 414 }
 415
 416 /* write the memroy to vmcore. 1 page per I/O. */
 417 static int write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start,
 418                         int64_t size)
 419 {
 420     int64_t i;
 421     int ret;
 422
 423     for (i = 0; i < size / TARGET_PAGE_SIZE; i++) {
 424         ret = write_data(s, block->host_addr + start + i * TARGET_PAGE_SIZE,
 425                          TARGET_PAGE_SIZE);
 426         if (ret < 0) {
 427             return ret;
 428         }
 429     }
 430
 431     if ((size % TARGET_PAGE_SIZE) != 0) {
 432         ret = write_data(s, block->host_addr + start + i * TARGET_PAGE_SIZE,
 433                          size % TARGET_PAGE_SIZE);
 434         if (ret < 0) {
 435             return ret;
 436         }
 437     }
 438
 439     return 0;
 440 }
 441
 442 /* get the memory's offset and size in the vmcore */
 443 static void get_offset_range(hwaddr phys_addr,
 444                              ram_addr_t mapping_length,
 445                              DumpState *s,
 446                              hwaddr *p_offset,
 447                              hwaddr *p_filesz)
 448 {
 449     GuestPhysBlock *block;
 450     hwaddr offset = s->memory_offset;
 451     int64_t size_in_block, start;
 452
 453     /* When the memory is not stored into vmcore, offset will be -1 */
 454     *p_offset = -1;
 455     *p_filesz = 0;
 456
 457     if (s->has_filter) {
 458         if (phys_addr < s->begin || phys_addr >= s->begin + s->length) {
 459             return;
 460         }
 461     }
 462
 463     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
 464         if (s->has_filter) {
 465             if (block->target_start >= s->begin + s->length ||
 466                 block->target_end <= s->begin) {
 467                 /* This block is out of the range */
 468                 continue;
 469             }
 470
 471             if (s->begin <= block->target_start) {
 472                 start = block->target_start;
 473             } else {
 474                 start = s->begin;
 475             }
 476
 477             size_in_block = block->target_end - start;
 478             if (s->begin + s->length < block->target_end) {
 479                 size_in_block -= block->target_end - (s->begin + s->length);
 480             }
 481         } else {
 482             start = block->target_start;
 483             size_in_block = block->target_end - block->target_start;
 484         }
 485
 486         if (phys_addr >= start && phys_addr < start + size_in_block) {
 487             *p_offset = phys_addr - start + offset;
 488
 489             /* The offset range mapped from the vmcore file must not spill over
 490              * the GuestPhysBlock, clamp it. The rest of the mapping will be
 491              * zero-filled in memory at load time; see
 492              * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
 493              */
 494             *p_filesz = phys_addr + mapping_length <= start + size_in_block ?
 495                         mapping_length :
 496                         size_in_block - (phys_addr - start);
 497             return;
 498         }
 499
 500         offset += size_in_block;
 501     }
 502 }
 503
 504 static int write_elf_loads(DumpState *s)
 505 {
 506     hwaddr offset, filesz;
 507     MemoryMapping *memory_mapping;
 508     uint32_t phdr_index = 1;
 509     int ret;
 510     uint32_t max_index;
 511
 512     if (s->have_section) {
 513         max_index = s->sh_info;
 514     } else {
 515         max_index = s->phdr_num;
 516     }
 517
 518     QTAILQ_FOREACH(memory_mapping, &s->list.head, next) {
 519         get_offset_range(memory_mapping->phys_addr,
 520                          memory_mapping->length,
 521                          s, &offset, &filesz);
 522         if (s->dump_info.d_class == ELFCLASS64) {
 523             ret = write_elf64_load(s, memory_mapping, phdr_index++, offset,
 524                                    filesz);
 525         } else {
 526             ret = write_elf32_load(s, memory_mapping, phdr_index++, offset,
 527                                    filesz);
 528         }
 529
 530         if (ret < 0) {
 531             return -1;
 532         }
 533
 534         if (phdr_index >= max_index) {
 535             break;
 536         }
 537     }
 538
 539     return 0;
 540 }
 541
 542 /* write elf header, PT_NOTE and elf note to vmcore. */
 543 static int dump_begin(DumpState *s)
 544 {
 545     int ret;
 546
 547     /*
 548      * the vmcore's format is:
 549      *   --------------
 550      *   |  elf header |
 551      *   --------------
 552      *   |  PT_NOTE    |
 553      *   --------------
 554      *   |  PT_LOAD    |
 555      *   --------------
 556      *   |  ......     |
 557      *   --------------
 558      *   |  PT_LOAD    |
 559      *   --------------
 560      *   |  sec_hdr    |
 561      *   --------------
 562      *   |  elf note   |
 563      *   --------------
 564      *   |  memory     |
 565      *   --------------
 566      *
 567      * we only know where the memory is saved after we write elf note into
 568      * vmcore.
 569      */
 570
 571     /* write elf header to vmcore */
 572     if (s->dump_info.d_class == ELFCLASS64) {
 573         ret = write_elf64_header(s);
 574     } else {
 575         ret = write_elf32_header(s);
 576     }
 577     if (ret < 0) {
 578         return -1;
 579     }
 580
 581     if (s->dump_info.d_class == ELFCLASS64) {
 582         /* write PT_NOTE to vmcore */
 583         if (write_elf64_note(s) < 0) {
 584             return -1;
 585         }
 586
 587         /* write all PT_LOAD to vmcore */
 588         if (write_elf_loads(s) < 0) {
 589             return -1;
 590         }
 591
 592         /* write section to vmcore */
 593         if (s->have_section) {
 594             if (write_elf_section(s, 1) < 0) {
 595                 return -1;
 596             }
 597         }
 598
 599         /* write notes to vmcore */
 600         if (write_elf64_notes(fd_write_vmcore, s) < 0) {
 601             return -1;
 602         }
 603
 604     } else {
 605         /* write PT_NOTE to vmcore */
 606         if (write_elf32_note(s) < 0) {
 607             return -1;
 608         }
 609
 610         /* write all PT_LOAD to vmcore */
 611         if (write_elf_loads(s) < 0) {
 612             return -1;
 613         }
 614
 615         /* write section to vmcore */
 616         if (s->have_section) {
 617             if (write_elf_section(s, 0) < 0) {
 618                 return -1;
 619             }
 620         }
 621
 622         /* write notes to vmcore */
 623         if (write_elf32_notes(fd_write_vmcore, s) < 0) {
 624             return -1;
 625         }
 626     }
 627
 628     return 0;
 629 }
 630
 631 /* write PT_LOAD to vmcore */
 632 static int dump_completed(DumpState *s)
 633 {
 634     dump_cleanup(s);
 635     return 0;
 636 }
 637
 638 static int get_next_block(DumpState *s, GuestPhysBlock *block)
 639 {
 640     while (1) {
 641         block = QTAILQ_NEXT(block, next);
 642         if (!block) {
 643             /* no more block */
 644             return 1;
 645         }
 646
 647         s->start = 0;
 648         s->next_block = block;
 649         if (s->has_filter) {
 650             if (block->target_start >= s->begin + s->length ||
 651                 block->target_end <= s->begin) {
 652                 /* This block is out of the range */
 653                 continue;
 654             }
 655
 656             if (s->begin > block->target_start) {
 657                 s->start = s->begin - block->target_start;
 658             }
 659         }
 660
 661         return 0;
 662     }
 663 }
 664
 665 /* write all memory to vmcore */
 666 static int dump_iterate(DumpState *s)
 667 {
 668     GuestPhysBlock *block;
 669     int64_t size;
 670     int ret;
 671
 672     while (1) {
 673         block = s->next_block;
 674
 675         size = block->target_end - block->target_start;
 676         if (s->has_filter) {
 677             size -= s->start;
 678             if (s->begin + s->length < block->target_end) {
 679                 size -= block->target_end - (s->begin + s->length);
 680             }
 681         }
 682         ret = write_memory(s, block, s->start, size);
 683         if (ret == -1) {
 684             return ret;
 685         }
 686
 687         ret = get_next_block(s, block);
 688         if (ret == 1) {
 689             dump_completed(s);
 690             return 0;
 691         }
 692     }
 693 }
 694
 695 static int create_vmcore(DumpState *s)
 696 {
 697     int ret;
 698
 699     ret = dump_begin(s);
 700     if (ret < 0) {
 701         return -1;
 702     }
 703
 704     ret = dump_iterate(s);
 705     if (ret < 0) {
 706         return -1;
 707     }
 708
 709     return 0;
 710 }
 711
 712 static int write_start_flat_header(int fd)
 713 {
 714     uint8_t *buf;
 715     MakedumpfileHeader mh;
 716     int ret = 0;
 717
 718     memset(&mh, 0, sizeof(mh));
 719     strncpy(mh.signature, MAKEDUMPFILE_SIGNATURE,
 720             strlen(MAKEDUMPFILE_SIGNATURE));
 721
 722     mh.type = cpu_to_be64(TYPE_FLAT_HEADER);
 723     mh.version = cpu_to_be64(VERSION_FLAT_HEADER);
 724
 725     buf = g_malloc0(MAX_SIZE_MDF_HEADER);
 726     memcpy(buf, &mh, sizeof(mh));
 727
 728     size_t written_size;
 729     written_size = qemu_write_full(fd, buf, MAX_SIZE_MDF_HEADER);
 730     if (written_size != MAX_SIZE_MDF_HEADER) {
 731         ret = -1;
 732     }
 733
 734     g_free(buf);
 735     return ret;
 736 }
 737
 738 static int write_end_flat_header(int fd)
 739 {
 740     MakedumpfileDataHeader mdh;
 741
 742     mdh.offset = END_FLAG_FLAT_HEADER;
 743     mdh.buf_size = END_FLAG_FLAT_HEADER;
 744
 745     size_t written_size;
 746     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 747     if (written_size != sizeof(mdh)) {
 748         return -1;
 749     }
 750
 751     return 0;
 752 }
 753
 754 static int write_buffer(int fd, off_t offset, const void *buf, size_t size)
 755 {
 756     size_t written_size;
 757     MakedumpfileDataHeader mdh;
 758
 759     mdh.offset = cpu_to_be64(offset);
 760     mdh.buf_size = cpu_to_be64(size);
 761
 762     written_size = qemu_write_full(fd, &mdh, sizeof(mdh));
 763     if (written_size != sizeof(mdh)) {
 764         return -1;
 765     }
 766
 767     written_size = qemu_write_full(fd, buf, size);
 768     if (written_size != size) {
 769         return -1;
 770     }
 771
 772     return 0;
 773 }
 774
 775 static int buf_write_note(const void *buf, size_t size, void *opaque)
 776 {
 777     DumpState *s = opaque;
 778
 779     /* note_buf is not enough */
 780     if (s->note_buf_offset + size > s->note_size) {
 781         return -1;
 782     }
 783
 784     memcpy(s->note_buf + s->note_buf_offset, buf, size);
 785
 786     s->note_buf_offset += size;
 787
 788     return 0;
 789 }
 790
 791 /* write common header, sub header and elf note to vmcore */
 792 static int create_header32(DumpState *s)
 793 {
 794     int ret = 0;
 795     DiskDumpHeader32 *dh = NULL;
 796     KdumpSubHeader32 *kh = NULL;
 797     size_t size;
 798     int endian = s->dump_info.d_endian;
 799     uint32_t block_size;
 800     uint32_t sub_hdr_size;
 801     uint32_t bitmap_blocks;
 802     uint32_t status = 0;
 803     uint64_t offset_note;
 804
 805     /* write common header, the version of kdump-compressed format is 6th */
 806     size = sizeof(DiskDumpHeader32);
 807     dh = g_malloc0(size);
 808
 809     strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
 810     dh->header_version = cpu_convert_to_target32(6, endian);
 811     block_size = s->page_size;
 812     dh->block_size = cpu_convert_to_target32(block_size, endian);
 813     sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size;
 814     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 815     dh->sub_hdr_size = cpu_convert_to_target32(sub_hdr_size, endian);
 816     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 817     dh->max_mapnr = cpu_convert_to_target32(MIN(s->max_mapnr, UINT_MAX),
 818                                             endian);
 819     dh->nr_cpus = cpu_convert_to_target32(s->nr_cpus, endian);
 820     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 821     dh->bitmap_blocks = cpu_convert_to_target32(bitmap_blocks, endian);
 822     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
 823
 824     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 825         status |= DUMP_DH_COMPRESSED_ZLIB;
 826     }
 827 #ifdef CONFIG_LZO
 828     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 829         status |= DUMP_DH_COMPRESSED_LZO;
 830     }
 831 #endif
 832 #ifdef CONFIG_SNAPPY
 833     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 834         status |= DUMP_DH_COMPRESSED_SNAPPY;
 835     }
 836 #endif
 837     dh->status = cpu_convert_to_target32(status, endian);
 838
 839     if (write_buffer(s->fd, 0, dh, size) < 0) {
 840         dump_error(s, "dump: failed to write disk dump header.\n");
 841         ret = -1;
 842         goto out;
 843     }
 844
 845     /* write sub header */
 846     size = sizeof(KdumpSubHeader32);
 847     kh = g_malloc0(size);
 848
 849     /* 64bit max_mapnr_64 */
 850     kh->max_mapnr_64 = cpu_convert_to_target64(s->max_mapnr, endian);
 851     kh->phys_base = cpu_convert_to_target32(PHYS_BASE, endian);
 852     kh->dump_level = cpu_convert_to_target32(DUMP_LEVEL, endian);
 853
 854     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 855     kh->offset_note = cpu_convert_to_target64(offset_note, endian);
 856     kh->note_size = cpu_convert_to_target32(s->note_size, endian);
 857
 858     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 859                      block_size, kh, size) < 0) {
 860         dump_error(s, "dump: failed to write kdump sub header.\n");
 861         ret = -1;
 862         goto out;
 863     }
 864
 865     /* write note */
 866     s->note_buf = g_malloc0(s->note_size);
 867     s->note_buf_offset = 0;
 868
 869     /* use s->note_buf to store notes temporarily */
 870     if (write_elf32_notes(buf_write_note, s) < 0) {
 871         ret = -1;
 872         goto out;
 873     }
 874
 875     if (write_buffer(s->fd, offset_note, s->note_buf,
 876                      s->note_size) < 0) {
 877         dump_error(s, "dump: failed to write notes");
 878         ret = -1;
 879         goto out;
 880     }
 881
 882     /* get offset of dump_bitmap */
 883     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 884                              block_size;
 885
 886     /* get offset of page */
 887     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 888                      block_size;
 889
 890 out:
 891     g_free(dh);
 892     g_free(kh);
 893     g_free(s->note_buf);
 894
 895     return ret;
 896 }
 897
 898 /* write common header, sub header and elf note to vmcore */
 899 static int create_header64(DumpState *s)
 900 {
 901     int ret = 0;
 902     DiskDumpHeader64 *dh = NULL;
 903     KdumpSubHeader64 *kh = NULL;
 904     size_t size;
 905     int endian = s->dump_info.d_endian;
 906     uint32_t block_size;
 907     uint32_t sub_hdr_size;
 908     uint32_t bitmap_blocks;
 909     uint32_t status = 0;
 910     uint64_t offset_note;
 911
 912     /* write common header, the version of kdump-compressed format is 6th */
 913     size = sizeof(DiskDumpHeader64);
 914     dh = g_malloc0(size);
 915
 916     strncpy(dh->signature, KDUMP_SIGNATURE, strlen(KDUMP_SIGNATURE));
 917     dh->header_version = cpu_convert_to_target32(6, endian);
 918     block_size = s->page_size;
 919     dh->block_size = cpu_convert_to_target32(block_size, endian);
 920     sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size;
 921     sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size);
 922     dh->sub_hdr_size = cpu_convert_to_target32(sub_hdr_size, endian);
 923     /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
 924     dh->max_mapnr = cpu_convert_to_target32(MIN(s->max_mapnr, UINT_MAX),
 925                                             endian);
 926     dh->nr_cpus = cpu_convert_to_target32(s->nr_cpus, endian);
 927     bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2;
 928     dh->bitmap_blocks = cpu_convert_to_target32(bitmap_blocks, endian);
 929     strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine));
 930
 931     if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) {
 932         status |= DUMP_DH_COMPRESSED_ZLIB;
 933     }
 934 #ifdef CONFIG_LZO
 935     if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) {
 936         status |= DUMP_DH_COMPRESSED_LZO;
 937     }
 938 #endif
 939 #ifdef CONFIG_SNAPPY
 940     if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) {
 941         status |= DUMP_DH_COMPRESSED_SNAPPY;
 942     }
 943 #endif
 944     dh->status = cpu_convert_to_target32(status, endian);
 945
 946     if (write_buffer(s->fd, 0, dh, size) < 0) {
 947         dump_error(s, "dump: failed to write disk dump header.\n");
 948         ret = -1;
 949         goto out;
 950     }
 951
 952     /* write sub header */
 953     size = sizeof(KdumpSubHeader64);
 954     kh = g_malloc0(size);
 955
 956     /* 64bit max_mapnr_64 */
 957     kh->max_mapnr_64 = cpu_convert_to_target64(s->max_mapnr, endian);
 958     kh->phys_base = cpu_convert_to_target64(PHYS_BASE, endian);
 959     kh->dump_level = cpu_convert_to_target32(DUMP_LEVEL, endian);
 960
 961     offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size;
 962     kh->offset_note = cpu_convert_to_target64(offset_note, endian);
 963     kh->note_size = cpu_convert_to_target64(s->note_size, endian);
 964
 965     if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS *
 966                      block_size, kh, size) < 0) {
 967         dump_error(s, "dump: failed to write kdump sub header.\n");
 968         ret = -1;
 969         goto out;
 970     }
 971
 972     /* write note */
 973     s->note_buf = g_malloc0(s->note_size);
 974     s->note_buf_offset = 0;
 975
 976     /* use s->note_buf to store notes temporarily */
 977     if (write_elf64_notes(buf_write_note, s) < 0) {
 978         ret = -1;
 979         goto out;
 980     }
 981
 982     if (write_buffer(s->fd, offset_note, s->note_buf,
 983                      s->note_size) < 0) {
 984         dump_error(s, "dump: failed to write notes");
 985         ret = -1;
 986         goto out;
 987     }
 988
 989     /* get offset of dump_bitmap */
 990     s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) *
 991                              block_size;
 992
 993     /* get offset of page */
 994     s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) *
 995                      block_size;
 996
 997 out:
 998     g_free(dh);
 999     g_free(kh);
1000     g_free(s->note_buf);
1001
1002     return ret;
1003 }
1004
1005 static int write_dump_header(DumpState *s)
1006 {
1007     if (s->dump_info.d_machine == EM_386) {
1008         return create_header32(s);
1009     } else {
1010         return create_header64(s);
1011     }
1012 }
1013
1014 /*
1015  * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1016  * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1017  * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1018  * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1019  * vmcore, ie. synchronizing un-sync bit into vmcore.
1020  */
1021 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value,
1022                            uint8_t *buf, DumpState *s)
1023 {
1024     off_t old_offset, new_offset;
1025     off_t offset_bitmap1, offset_bitmap2;
1026     uint32_t byte, bit;
1027
1028     /* should not set the previous place */
1029     assert(last_pfn <= pfn);
1030
1031     /*
1032      * if the bit needed to be set is not cached in buf, flush the data in buf
1033      * to vmcore firstly.
1034      * making new_offset be bigger than old_offset can also sync remained data
1035      * into vmcore.
1036      */
1037     old_offset = BUFSIZE_BITMAP * (last_pfn / PFN_BUFBITMAP);
1038     new_offset = BUFSIZE_BITMAP * (pfn / PFN_BUFBITMAP);
1039
1040     while (old_offset < new_offset) {
1041         /* calculate the offset and write dump_bitmap */
1042         offset_bitmap1 = s->offset_dump_bitmap + old_offset;
1043         if (write_buffer(s->fd, offset_bitmap1, buf,
1044                          BUFSIZE_BITMAP) < 0) {
1045             return -1;
1046         }
1047
1048         /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1049         offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap +
1050                          old_offset;
1051         if (write_buffer(s->fd, offset_bitmap2, buf,
1052                          BUFSIZE_BITMAP) < 0) {
1053             return -1;
1054         }
1055
1056         memset(buf, 0, BUFSIZE_BITMAP);
1057         old_offset += BUFSIZE_BITMAP;
1058     }
1059
1060     /* get the exact place of the bit in the buf, and set it */
1061     byte = (pfn % PFN_BUFBITMAP) / CHAR_BIT;
1062     bit = (pfn % PFN_BUFBITMAP) % CHAR_BIT;
1063     if (value) {
1064         buf[byte] |= 1u << bit;
1065     } else {
1066         buf[byte] &= ~(1u << bit);
1067     }
1068
1069     return 0;
1070 }
1071
1072 /*
1073  * exam every page and return the page frame number and the address of the page.
1074  * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1075  * blocks, so block->target_start and block->target_end should be interal
1076  * multiples of the target page size.
1077  */
1078 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr,
1079                           uint8_t **bufptr, DumpState *s)
1080 {
1081     GuestPhysBlock *block = *blockptr;
1082     hwaddr addr;
1083     uint8_t *buf;
1084
1085     /* block == NULL means the start of the iteration */
1086     if (!block) {
1087         block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1088         *blockptr = block;
1089         assert(block->target_start % s->page_size == 0);
1090         assert(block->target_end % s->page_size == 0);
1091         *pfnptr = paddr_to_pfn(block->target_start, s->page_shift);
1092         if (bufptr) {
1093             *bufptr = block->host_addr;
1094         }
1095         return true;
1096     }
1097
1098     *pfnptr = *pfnptr + 1;
1099     addr = pfn_to_paddr(*pfnptr, s->page_shift);
1100
1101     if ((addr >= block->target_start) &&
1102         (addr + s->page_size <= block->target_end)) {
1103         buf = block->host_addr + (addr - block->target_start);
1104     } else {
1105         /* the next page is in the next block */
1106         block = QTAILQ_NEXT(block, next);
1107         *blockptr = block;
1108         if (!block) {
1109             return false;
1110         }
1111         assert(block->target_start % s->page_size == 0);
1112         assert(block->target_end % s->page_size == 0);
1113         *pfnptr = paddr_to_pfn(block->target_start, s->page_shift);
1114         buf = block->host_addr;
1115     }
1116
1117     if (bufptr) {
1118         *bufptr = buf;
1119     }
1120
1121     return true;
1122 }
1123
1124 static int write_dump_bitmap(DumpState *s)
1125 {
1126     int ret = 0;
1127     uint64_t last_pfn, pfn;
1128     void *dump_bitmap_buf;
1129     size_t num_dumpable;
1130     GuestPhysBlock *block_iter = NULL;
1131
1132     /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1133     dump_bitmap_buf = g_malloc0(BUFSIZE_BITMAP);
1134
1135     num_dumpable = 0;
1136     last_pfn = 0;
1137
1138     /*
1139      * exam memory page by page, and set the bit in dump_bitmap corresponded
1140      * to the existing page.
1141      */
1142     while (get_next_page(&block_iter, &pfn, NULL, s)) {
1143         ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s);
1144         if (ret < 0) {
1145             dump_error(s, "dump: failed to set dump_bitmap.\n");
1146             ret = -1;
1147             goto out;
1148         }
1149
1150         last_pfn = pfn;
1151         num_dumpable++;
1152     }
1153
1154     /*
1155      * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1156      * set last_pfn + PFN_BUFBITMAP to 0 and those set but un-sync bit will be
1157      * synchronized into vmcore.
1158      */
1159     if (num_dumpable > 0) {
1160         ret = set_dump_bitmap(last_pfn, last_pfn + PFN_BUFBITMAP, false,
1161                               dump_bitmap_buf, s);
1162         if (ret < 0) {
1163             dump_error(s, "dump: failed to sync dump_bitmap.\n");
1164             ret = -1;
1165             goto out;
1166         }
1167     }
1168
1169     /* number of dumpable pages that will be dumped later */
1170     s->num_dumpable = num_dumpable;
1171
1172 out:
1173     g_free(dump_bitmap_buf);
1174
1175     return ret;
1176 }
1177
1178 static void prepare_data_cache(DataCache *data_cache, DumpState *s,
1179                                off_t offset)
1180 {
1181     data_cache->fd = s->fd;
1182     data_cache->data_size = 0;
1183     data_cache->buf_size = BUFSIZE_DATA_CACHE;
1184     data_cache->buf = g_malloc0(BUFSIZE_DATA_CACHE);
1185     data_cache->offset = offset;
1186 }
1187
1188 static int write_cache(DataCache *dc, const void *buf, size_t size,
1189                        bool flag_sync)
1190 {
1191     /*
1192      * dc->buf_size should not be less than size, otherwise dc will never be
1193      * enough
1194      */
1195     assert(size <= dc->buf_size);
1196
1197     /*
1198      * if flag_sync is set, synchronize data in dc->buf into vmcore.
1199      * otherwise check if the space is enough for caching data in buf, if not,
1200      * write the data in dc->buf to dc->fd and reset dc->buf
1201      */
1202     if ((!flag_sync && dc->data_size + size > dc->buf_size) ||
1203         (flag_sync && dc->data_size > 0)) {
1204         if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) {
1205             return -1;
1206         }
1207
1208         dc->offset += dc->data_size;
1209         dc->data_size = 0;
1210     }
1211
1212     if (!flag_sync) {
1213         memcpy(dc->buf + dc->data_size, buf, size);
1214         dc->data_size += size;
1215     }
1216
1217     return 0;
1218 }
1219
1220 static void free_data_cache(DataCache *data_cache)
1221 {
1222     g_free(data_cache->buf);
1223 }
1224
1225 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress)
1226 {
1227     size_t len_buf_out_zlib, len_buf_out_lzo, len_buf_out_snappy;
1228     size_t len_buf_out;
1229
1230     /* init buf_out */
1231     len_buf_out_zlib = len_buf_out_lzo = len_buf_out_snappy = 0;
1232
1233     /* buf size for zlib */
1234     len_buf_out_zlib = compressBound(page_size);
1235
1236     /* buf size for lzo */
1237 #ifdef CONFIG_LZO
1238     if (flag_compress & DUMP_DH_COMPRESSED_LZO) {
1239         if (lzo_init() != LZO_E_OK) {
1240             /* return 0 to indicate lzo is unavailable */
1241             return 0;
1242         }
1243     }
1244
1245     /*
1246      * LZO will expand incompressible data by a little amount. please check the
1247      * following URL to see the expansion calculation:
1248      * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1249      */
1250     len_buf_out_lzo = page_size + page_size / 16 + 64 + 3;
1251 #endif
1252
1253 #ifdef CONFIG_SNAPPY
1254     /* buf size for snappy */
1255     len_buf_out_snappy = snappy_max_compressed_length(page_size);
1256 #endif
1257
1258     /* get the biggest that can store all kinds of compressed page */
1259     len_buf_out = MAX(len_buf_out_zlib,
1260                       MAX(len_buf_out_lzo, len_buf_out_snappy));
1261
1262     return len_buf_out;
1263 }
1264
1265 /*
1266  * check if the page is all 0
1267  */
1268 static inline bool is_zero_page(const uint8_t *buf, size_t page_size)
1269 {
1270     return buffer_is_zero(buf, page_size);
1271 }
1272
1273 static int write_dump_pages(DumpState *s)
1274 {
1275     int ret = 0;
1276     DataCache page_desc, page_data;
1277     size_t len_buf_out, size_out;
1278 #ifdef CONFIG_LZO
1279     lzo_bytep wrkmem = NULL;
1280 #endif
1281     uint8_t *buf_out = NULL;
1282     off_t offset_desc, offset_data;
1283     PageDescriptor pd, pd_zero;
1284     uint8_t *buf;
1285     int endian = s->dump_info.d_endian;
1286     GuestPhysBlock *block_iter = NULL;
1287     uint64_t pfn_iter;
1288
1289     /* get offset of page_desc and page_data in dump file */
1290     offset_desc = s->offset_page;
1291     offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable;
1292
1293     prepare_data_cache(&page_desc, s, offset_desc);
1294     prepare_data_cache(&page_data, s, offset_data);
1295
1296     /* prepare buffer to store compressed data */
1297     len_buf_out = get_len_buf_out(s->page_size, s->flag_compress);
1298     if (len_buf_out == 0) {
1299         dump_error(s, "dump: failed to get length of output buffer.\n");
1300         goto out;
1301     }
1302
1303 #ifdef CONFIG_LZO
1304     wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS);
1305 #endif
1306
1307     buf_out = g_malloc(len_buf_out);
1308
1309     /*
1310      * init zero page's page_desc and page_data, because every zero page
1311      * uses the same page_data
1312      */
1313     pd_zero.size = cpu_convert_to_target32(s->page_size, endian);
1314     pd_zero.flags = cpu_convert_to_target32(0, endian);
1315     pd_zero.offset = cpu_convert_to_target64(offset_data, endian);
1316     pd_zero.page_flags = cpu_convert_to_target64(0, endian);
1317     buf = g_malloc0(s->page_size);
1318     ret = write_cache(&page_data, buf, s->page_size, false);
1319     g_free(buf);
1320     if (ret < 0) {
1321         dump_error(s, "dump: failed to write page data(zero page).\n");
1322         goto out;
1323     }
1324
1325     offset_data += s->page_size;
1326
1327     /*
1328      * dump memory to vmcore page by page. zero page will all be resided in the
1329      * first page of page section
1330      */
1331     while (get_next_page(&block_iter, &pfn_iter, &buf, s)) {
1332         /* check zero page */
1333         if (is_zero_page(buf, s->page_size)) {
1334             ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor),
1335                               false);
1336             if (ret < 0) {
1337                 dump_error(s, "dump: failed to write page desc.\n");
1338                 goto out;
1339             }
1340         } else {
1341             /*
1342              * not zero page, then:
1343              * 1. compress the page
1344              * 2. write the compressed page into the cache of page_data
1345              * 3. get page desc of the compressed page and write it into the
1346              *    cache of page_desc
1347              *
1348              * only one compression format will be used here, for
1349              * s->flag_compress is set. But when compression fails to work,
1350              * we fall back to save in plaintext.
1351              */
1352              size_out = len_buf_out;
1353              if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) &&
1354                     (compress2(buf_out, (uLongf *)&size_out, buf, s->page_size,
1355                     Z_BEST_SPEED) == Z_OK) && (size_out < s->page_size)) {
1356                 pd.flags = cpu_convert_to_target32(DUMP_DH_COMPRESSED_ZLIB,
1357                                                    endian);
1358                 pd.size  = cpu_convert_to_target32(size_out, endian);
1359
1360                 ret = write_cache(&page_data, buf_out, size_out, false);
1361                 if (ret < 0) {
1362                     dump_error(s, "dump: failed to write page data.\n");
1363                     goto out;
1364                 }
1365 #ifdef CONFIG_LZO
1366             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) &&
1367                     (lzo1x_1_compress(buf, s->page_size, buf_out,
1368                     (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) &&
1369                     (size_out < s->page_size)) {
1370                 pd.flags = cpu_convert_to_target32(DUMP_DH_COMPRESSED_LZO,
1371                                                    endian);
1372                 pd.size  = cpu_convert_to_target32(size_out, endian);
1373
1374                 ret = write_cache(&page_data, buf_out, size_out, false);
1375                 if (ret < 0) {
1376                     dump_error(s, "dump: failed to write page data.\n");
1377                     goto out;
1378                 }
1379 #endif
1380 #ifdef CONFIG_SNAPPY
1381             } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) &&
1382                     (snappy_compress((char *)buf, s->page_size,
1383                     (char *)buf_out, &size_out) == SNAPPY_OK) &&
1384                     (size_out < s->page_size)) {
1385                 pd.flags = cpu_convert_to_target32(
1386                                         DUMP_DH_COMPRESSED_SNAPPY, endian);
1387                 pd.size  = cpu_convert_to_target32(size_out, endian);
1388
1389                 ret = write_cache(&page_data, buf_out, size_out, false);
1390                 if (ret < 0) {
1391                     dump_error(s, "dump: failed to write page data.\n");
1392                     goto out;
1393                 }
1394 #endif
1395             } else {
1396                 /*
1397                  * fall back to save in plaintext, size_out should be
1398                  * assigned to s->page_size
1399                  */
1400                 pd.flags = cpu_convert_to_target32(0, endian);
1401                 size_out = s->page_size;
1402                 pd.size = cpu_convert_to_target32(size_out, endian);
1403
1404                 ret = write_cache(&page_data, buf, s->page_size, false);
1405                 if (ret < 0) {
1406                     dump_error(s, "dump: failed to write page data.\n");
1407                     goto out;
1408                 }
1409             }
1410
1411             /* get and write page desc here */
1412             pd.page_flags = cpu_convert_to_target64(0, endian);
1413             pd.offset = cpu_convert_to_target64(offset_data, endian);
1414             offset_data += size_out;
1415
1416             ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false);
1417             if (ret < 0) {
1418                 dump_error(s, "dump: failed to write page desc.\n");
1419                 goto out;
1420             }
1421         }
1422     }
1423
1424     ret = write_cache(&page_desc, NULL, 0, true);
1425     if (ret < 0) {
1426         dump_error(s, "dump: failed to sync cache for page_desc.\n");
1427         goto out;
1428     }
1429     ret = write_cache(&page_data, NULL, 0, true);
1430     if (ret < 0) {
1431         dump_error(s, "dump: failed to sync cache for page_data.\n");
1432         goto out;
1433     }
1434
1435 out:
1436     free_data_cache(&page_desc);
1437     free_data_cache(&page_data);
1438
1439 #ifdef CONFIG_LZO
1440     g_free(wrkmem);
1441 #endif
1442
1443     g_free(buf_out);
1444
1445     return ret;
1446 }
1447
1448 static int create_kdump_vmcore(DumpState *s)
1449 {
1450     int ret;
1451
1452     /*
1453      * the kdump-compressed format is:
1454      *                                               File offset
1455      *  +------------------------------------------+ 0x0
1456      *  |    main header (struct disk_dump_header) |
1457      *  |------------------------------------------+ block 1
1458      *  |    sub header (struct kdump_sub_header)  |
1459      *  |------------------------------------------+ block 2
1460      *  |            1st-dump_bitmap               |
1461      *  |------------------------------------------+ block 2 + X blocks
1462      *  |            2nd-dump_bitmap               | (aligned by block)
1463      *  |------------------------------------------+ block 2 + 2 * X blocks
1464      *  |  page desc for pfn 0 (struct page_desc)  | (aligned by block)
1465      *  |  page desc for pfn 1 (struct page_desc)  |
1466      *  |                    :                     |
1467      *  |------------------------------------------| (not aligned by block)
1468      *  |         page data (pfn 0)                |
1469      *  |         page data (pfn 1)                |
1470      *  |                    :                     |
1471      *  +------------------------------------------+
1472      */
1473
1474     ret = write_start_flat_header(s->fd);
1475     if (ret < 0) {
1476         dump_error(s, "dump: failed to write start flat header.\n");
1477         return -1;
1478     }
1479
1480     ret = write_dump_header(s);
1481     if (ret < 0) {
1482         return -1;
1483     }
1484
1485     ret = write_dump_bitmap(s);
1486     if (ret < 0) {
1487         return -1;
1488     }
1489
1490     ret = write_dump_pages(s);
1491     if (ret < 0) {
1492         return -1;
1493     }
1494
1495     ret = write_end_flat_header(s->fd);
1496     if (ret < 0) {
1497         dump_error(s, "dump: failed to write end flat header.\n");
1498         return -1;
1499     }
1500
1501     dump_completed(s);
1502
1503     return 0;
1504 }
1505
1506 static ram_addr_t get_start_block(DumpState *s)
1507 {
1508     GuestPhysBlock *block;
1509
1510     if (!s->has_filter) {
1511         s->next_block = QTAILQ_FIRST(&s->guest_phys_blocks.head);
1512         return 0;
1513     }
1514
1515     QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) {
1516         if (block->target_start >= s->begin + s->length ||
1517             block->target_end <= s->begin) {
1518             /* This block is out of the range */
1519             continue;
1520         }
1521
1522         s->next_block = block;
1523         if (s->begin > block->target_start) {
1524             s->start = s->begin - block->target_start;
1525         } else {
1526             s->start = 0;
1527         }
1528         return s->start;
1529     }
1530
1531     return -1;
1532 }
1533
1534 static void get_max_mapnr(DumpState *s)
1535 {
1536     GuestPhysBlock *last_block;
1537
1538     last_block = QTAILQ_LAST(&s->guest_phys_blocks.head, GuestPhysBlockHead);
1539     s->max_mapnr = paddr_to_pfn(last_block->target_end, s->page_shift);
1540 }
1541
1542 static int dump_init(DumpState *s, int fd, bool has_format,
1543                      DumpGuestMemoryFormat format, bool paging, bool has_filter,
1544                      int64_t begin, int64_t length, Error **errp)
1545 {
1546     CPUState *cpu;
1547     int nr_cpus;
1548     Error *err = NULL;
1549     int ret;
1550
1551     /* kdump-compressed is conflict with paging and filter */
1552     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1553         assert(!paging && !has_filter);
1554     }
1555
1556     if (runstate_is_running()) {
1557         vm_stop(RUN_STATE_SAVE_VM);
1558         s->resume = true;
1559     } else {
1560         s->resume = false;
1561     }
1562
1563     /* If we use KVM, we should synchronize the registers before we get dump
1564      * info or physmap info.
1565      */
1566     cpu_synchronize_all_states();
1567     nr_cpus = 0;
1568     CPU_FOREACH(cpu) {
1569         nr_cpus++;
1570     }
1571
1572     s->fd = fd;
1573     s->has_filter = has_filter;
1574     s->begin = begin;
1575     s->length = length;
1576
1577     guest_phys_blocks_init(&s->guest_phys_blocks);
1578     guest_phys_blocks_append(&s->guest_phys_blocks);
1579
1580     s->start = get_start_block(s);
1581     if (s->start == -1) {
1582         error_set(errp, QERR_INVALID_PARAMETER, "begin");
1583         goto cleanup;
1584     }
1585
1586     /* get dump info: endian, class and architecture.
1587      * If the target architecture is not supported, cpu_get_dump_info() will
1588      * return -1.
1589      */
1590     ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks);
1591     if (ret < 0) {
1592         error_set(errp, QERR_UNSUPPORTED);
1593         goto cleanup;
1594     }
1595
1596     s->note_size = cpu_get_note_size(s->dump_info.d_class,
1597                                      s->dump_info.d_machine, nr_cpus);
1598     if (s->note_size < 0) {
1599         error_set(errp, QERR_UNSUPPORTED);
1600         goto cleanup;
1601     }
1602
1603     /* get memory mapping */
1604     memory_mapping_list_init(&s->list);
1605     if (paging) {
1606         qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, &err);
1607         if (err != NULL) {
1608             error_propagate(errp, err);
1609             goto cleanup;
1610         }
1611     } else {
1612         qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks);
1613     }
1614
1615     s->nr_cpus = nr_cpus;
1616     s->page_size = TARGET_PAGE_SIZE;
1617     s->page_shift = ffs(s->page_size) - 1;
1618
1619     get_max_mapnr(s);
1620
1621     uint64_t tmp;
1622     tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT), s->page_size);
1623     s->len_dump_bitmap = tmp * s->page_size;
1624
1625     /* init for kdump-compressed format */
1626     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1627         switch (format) {
1628         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB:
1629             s->flag_compress = DUMP_DH_COMPRESSED_ZLIB;
1630             break;
1631
1632         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO:
1633             s->flag_compress = DUMP_DH_COMPRESSED_LZO;
1634             break;
1635
1636         case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY:
1637             s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY;
1638             break;
1639
1640         default:
1641             s->flag_compress = 0;
1642         }
1643
1644         return 0;
1645     }
1646
1647     if (s->has_filter) {
1648         memory_mapping_filter(&s->list, s->begin, s->length);
1649     }
1650
1651     /*
1652      * calculate phdr_num
1653      *
1654      * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1655      */
1656     s->phdr_num = 1; /* PT_NOTE */
1657     if (s->list.num < UINT16_MAX - 2) {
1658         s->phdr_num += s->list.num;
1659         s->have_section = false;
1660     } else {
1661         s->have_section = true;
1662         s->phdr_num = PN_XNUM;
1663         s->sh_info = 1; /* PT_NOTE */
1664
1665         /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1666         if (s->list.num <= UINT32_MAX - 1) {
1667             s->sh_info += s->list.num;
1668         } else {
1669             s->sh_info = UINT32_MAX;
1670         }
1671     }
1672
1673     if (s->dump_info.d_class == ELFCLASS64) {
1674         if (s->have_section) {
1675             s->memory_offset = sizeof(Elf64_Ehdr) +
1676                                sizeof(Elf64_Phdr) * s->sh_info +
1677                                sizeof(Elf64_Shdr) + s->note_size;
1678         } else {
1679             s->memory_offset = sizeof(Elf64_Ehdr) +
1680                                sizeof(Elf64_Phdr) * s->phdr_num + s->note_size;
1681         }
1682     } else {
1683         if (s->have_section) {
1684             s->memory_offset = sizeof(Elf32_Ehdr) +
1685                                sizeof(Elf32_Phdr) * s->sh_info +
1686                                sizeof(Elf32_Shdr) + s->note_size;
1687         } else {
1688             s->memory_offset = sizeof(Elf32_Ehdr) +
1689                                sizeof(Elf32_Phdr) * s->phdr_num + s->note_size;
1690         }
1691     }
1692
1693     return 0;
1694
1695 cleanup:
1696     guest_phys_blocks_free(&s->guest_phys_blocks);
1697
1698     if (s->resume) {
1699         vm_start();
1700     }
1701
1702     return -1;
1703 }
1704
1705 void qmp_dump_guest_memory(bool paging, const char *file, bool has_begin,
1706                            int64_t begin, bool has_length,
1707                            int64_t length, bool has_format,
1708                            DumpGuestMemoryFormat format, Error **errp)
1709 {
1710     const char *p;
1711     int fd = -1;
1712     DumpState *s;
1713     int ret;
1714
1715     /*
1716      * kdump-compressed format need the whole memory dumped, so paging or
1717      * filter is not supported here.
1718      */
1719     if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) &&
1720         (paging || has_begin || has_length)) {
1721         error_setg(errp, "kdump-compressed format doesn't support paging or "
1722                          "filter");
1723         return;
1724     }
1725     if (has_begin && !has_length) {
1726         error_set(errp, QERR_MISSING_PARAMETER, "length");
1727         return;
1728     }
1729     if (!has_begin && has_length) {
1730         error_set(errp, QERR_MISSING_PARAMETER, "begin");
1731         return;
1732     }
1733
1734     /* check whether lzo/snappy is supported */
1735 #ifndef CONFIG_LZO
1736     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) {
1737         error_setg(errp, "kdump-lzo is not available now");
1738         return;
1739     }
1740 #endif
1741
1742 #ifndef CONFIG_SNAPPY
1743     if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) {
1744         error_setg(errp, "kdump-snappy is not available now");
1745         return;
1746     }
1747 #endif
1748
1749 #if !defined(WIN32)
1750     if (strstart(file, "fd:", &p)) {
1751         fd = monitor_get_fd(cur_mon, p, errp);
1752         if (fd == -1) {
1753             return;
1754         }
1755     }
1756 #endif
1757
1758     if  (strstart(file, "file:", &p)) {
1759         fd = qemu_open(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR);
1760         if (fd < 0) {
1761             error_setg_file_open(errp, errno, p);
1762             return;
1763         }
1764     }
1765
1766     if (fd == -1) {
1767         error_set(errp, QERR_INVALID_PARAMETER, "protocol");
1768         return;
1769     }
1770
1771     s = g_malloc0(sizeof(DumpState));
1772
1773     ret = dump_init(s, fd, has_format, format, paging, has_begin,
1774                     begin, length, errp);
1775     if (ret < 0) {
1776         g_free(s);
1777         return;
1778     }
1779
1780     if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) {
1781         if (create_kdump_vmcore(s) < 0) {
1782             error_set(errp, QERR_IO_ERROR);
1783         }
1784     } else {
1785         if (create_vmcore(s) < 0) {
1786             error_set(errp, QERR_IO_ERROR);
1787         }
1788     }
1789
1790     g_free(s);
1791 }
1792
1793 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp)
1794 {
1795     DumpGuestMemoryFormatList *item;
1796     DumpGuestMemoryCapability *cap =
1797                                   g_malloc0(sizeof(DumpGuestMemoryCapability));
1798
1799     /* elf is always available */
1800     item = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1801     cap->formats = item;
1802     item->value = DUMP_GUEST_MEMORY_FORMAT_ELF;
1803
1804     /* kdump-zlib is always available */
1805     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1806     item = item->next;
1807     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB;
1808
1809     /* add new item if kdump-lzo is available */
1810 #ifdef CONFIG_LZO
1811     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1812     item = item->next;
1813     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO;
1814 #endif
1815
1816     /* add new item if kdump-snappy is available */
1817 #ifdef CONFIG_SNAPPY
1818     item->next = g_malloc0(sizeof(DumpGuestMemoryFormatList));
1819     item = item->next;
1820     item->value = DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY;
1821 #endif
1822
1823     return cap;
1824 }