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