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/osdep.h" 15 #include "qemu/cutils.h" 16 #include "elf.h" 17 #include "qemu/bswap.h" 18 #include "exec/target_page.h" 19 #include "monitor/monitor.h" 20 #include "sysemu/dump.h" 21 #include "sysemu/runstate.h" 22 #include "sysemu/cpus.h" 23 #include "qapi/error.h" 24 #include "qapi/qapi-commands-dump.h" 25 #include "qapi/qapi-events-dump.h" 26 #include "qapi/qmp/qerror.h" 27 #include "qemu/main-loop.h" 28 #include "hw/misc/vmcoreinfo.h" 29 #include "migration/blocker.h" 30 #include "hw/core/cpu.h" 31 32 #ifdef TARGET_X86_64 33 #include "win_dump.h" 34 #endif 35 36 #include <zlib.h> 37 #ifdef CONFIG_LZO 38 #include <lzo/lzo1x.h> 39 #endif 40 #ifdef CONFIG_SNAPPY 41 #include <snappy-c.h> 42 #endif 43 #ifndef ELF_MACHINE_UNAME 44 #define ELF_MACHINE_UNAME "Unknown" 45 #endif 46 47 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */ 48 49 static Error *dump_migration_blocker; 50 51 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size) \ 52 ((DIV_ROUND_UP((hdr_size), 4) + \ 53 DIV_ROUND_UP((name_size), 4) + \ 54 DIV_ROUND_UP((desc_size), 4)) * 4) 55 56 static inline bool dump_is_64bit(DumpState *s) 57 { 58 return s->dump_info.d_class == ELFCLASS64; 59 } 60 61 static inline bool dump_has_filter(DumpState *s) 62 { 63 return s->filter_area_length > 0; 64 } 65 66 uint16_t cpu_to_dump16(DumpState *s, uint16_t val) 67 { 68 if (s->dump_info.d_endian == ELFDATA2LSB) { 69 val = cpu_to_le16(val); 70 } else { 71 val = cpu_to_be16(val); 72 } 73 74 return val; 75 } 76 77 uint32_t cpu_to_dump32(DumpState *s, uint32_t val) 78 { 79 if (s->dump_info.d_endian == ELFDATA2LSB) { 80 val = cpu_to_le32(val); 81 } else { 82 val = cpu_to_be32(val); 83 } 84 85 return val; 86 } 87 88 uint64_t cpu_to_dump64(DumpState *s, uint64_t val) 89 { 90 if (s->dump_info.d_endian == ELFDATA2LSB) { 91 val = cpu_to_le64(val); 92 } else { 93 val = cpu_to_be64(val); 94 } 95 96 return val; 97 } 98 99 static int dump_cleanup(DumpState *s) 100 { 101 guest_phys_blocks_free(&s->guest_phys_blocks); 102 memory_mapping_list_free(&s->list); 103 close(s->fd); 104 g_free(s->guest_note); 105 g_array_unref(s->string_table_buf); 106 s->guest_note = NULL; 107 if (s->resume) { 108 if (s->detached) { 109 qemu_mutex_lock_iothread(); 110 } 111 vm_start(); 112 if (s->detached) { 113 qemu_mutex_unlock_iothread(); 114 } 115 } 116 migrate_del_blocker(dump_migration_blocker); 117 118 return 0; 119 } 120 121 static int fd_write_vmcore(const void *buf, size_t size, void *opaque) 122 { 123 DumpState *s = opaque; 124 size_t written_size; 125 126 written_size = qemu_write_full(s->fd, buf, size); 127 if (written_size != size) { 128 return -errno; 129 } 130 131 return 0; 132 } 133 134 static void prepare_elf64_header(DumpState *s, Elf64_Ehdr *elf_header) 135 { 136 /* 137 * phnum in the elf header is 16 bit, if we have more segments we 138 * set phnum to PN_XNUM and write the real number of segments to a 139 * special section. 140 */ 141 uint16_t phnum = MIN(s->phdr_num, PN_XNUM); 142 143 memset(elf_header, 0, sizeof(Elf64_Ehdr)); 144 memcpy(elf_header, ELFMAG, SELFMAG); 145 elf_header->e_ident[EI_CLASS] = ELFCLASS64; 146 elf_header->e_ident[EI_DATA] = s->dump_info.d_endian; 147 elf_header->e_ident[EI_VERSION] = EV_CURRENT; 148 elf_header->e_type = cpu_to_dump16(s, ET_CORE); 149 elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine); 150 elf_header->e_version = cpu_to_dump32(s, EV_CURRENT); 151 elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header)); 152 elf_header->e_phoff = cpu_to_dump64(s, s->phdr_offset); 153 elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf64_Phdr)); 154 elf_header->e_phnum = cpu_to_dump16(s, phnum); 155 elf_header->e_shoff = cpu_to_dump64(s, s->shdr_offset); 156 elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf64_Shdr)); 157 elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num); 158 elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1); 159 } 160 161 static void prepare_elf32_header(DumpState *s, Elf32_Ehdr *elf_header) 162 { 163 /* 164 * phnum in the elf header is 16 bit, if we have more segments we 165 * set phnum to PN_XNUM and write the real number of segments to a 166 * special section. 167 */ 168 uint16_t phnum = MIN(s->phdr_num, PN_XNUM); 169 170 memset(elf_header, 0, sizeof(Elf32_Ehdr)); 171 memcpy(elf_header, ELFMAG, SELFMAG); 172 elf_header->e_ident[EI_CLASS] = ELFCLASS32; 173 elf_header->e_ident[EI_DATA] = s->dump_info.d_endian; 174 elf_header->e_ident[EI_VERSION] = EV_CURRENT; 175 elf_header->e_type = cpu_to_dump16(s, ET_CORE); 176 elf_header->e_machine = cpu_to_dump16(s, s->dump_info.d_machine); 177 elf_header->e_version = cpu_to_dump32(s, EV_CURRENT); 178 elf_header->e_ehsize = cpu_to_dump16(s, sizeof(elf_header)); 179 elf_header->e_phoff = cpu_to_dump32(s, s->phdr_offset); 180 elf_header->e_phentsize = cpu_to_dump16(s, sizeof(Elf32_Phdr)); 181 elf_header->e_phnum = cpu_to_dump16(s, phnum); 182 elf_header->e_shoff = cpu_to_dump32(s, s->shdr_offset); 183 elf_header->e_shentsize = cpu_to_dump16(s, sizeof(Elf32_Shdr)); 184 elf_header->e_shnum = cpu_to_dump16(s, s->shdr_num); 185 elf_header->e_shstrndx = cpu_to_dump16(s, s->shdr_num - 1); 186 } 187 188 static void write_elf_header(DumpState *s, Error **errp) 189 { 190 Elf32_Ehdr elf32_header; 191 Elf64_Ehdr elf64_header; 192 size_t header_size; 193 void *header_ptr; 194 int ret; 195 196 /* The NULL header and the shstrtab are always defined */ 197 assert(s->shdr_num >= 2); 198 if (dump_is_64bit(s)) { 199 prepare_elf64_header(s, &elf64_header); 200 header_size = sizeof(elf64_header); 201 header_ptr = &elf64_header; 202 } else { 203 prepare_elf32_header(s, &elf32_header); 204 header_size = sizeof(elf32_header); 205 header_ptr = &elf32_header; 206 } 207 208 ret = fd_write_vmcore(header_ptr, header_size, s); 209 if (ret < 0) { 210 error_setg_errno(errp, -ret, "dump: failed to write elf header"); 211 } 212 } 213 214 static void write_elf64_load(DumpState *s, MemoryMapping *memory_mapping, 215 int phdr_index, hwaddr offset, 216 hwaddr filesz, Error **errp) 217 { 218 Elf64_Phdr phdr; 219 int ret; 220 221 memset(&phdr, 0, sizeof(Elf64_Phdr)); 222 phdr.p_type = cpu_to_dump32(s, PT_LOAD); 223 phdr.p_offset = cpu_to_dump64(s, offset); 224 phdr.p_paddr = cpu_to_dump64(s, memory_mapping->phys_addr); 225 phdr.p_filesz = cpu_to_dump64(s, filesz); 226 phdr.p_memsz = cpu_to_dump64(s, memory_mapping->length); 227 phdr.p_vaddr = cpu_to_dump64(s, memory_mapping->virt_addr) ?: phdr.p_paddr; 228 229 assert(memory_mapping->length >= filesz); 230 231 ret = fd_write_vmcore(&phdr, sizeof(Elf64_Phdr), s); 232 if (ret < 0) { 233 error_setg_errno(errp, -ret, 234 "dump: failed to write program header table"); 235 } 236 } 237 238 static void write_elf32_load(DumpState *s, MemoryMapping *memory_mapping, 239 int phdr_index, hwaddr offset, 240 hwaddr filesz, Error **errp) 241 { 242 Elf32_Phdr phdr; 243 int ret; 244 245 memset(&phdr, 0, sizeof(Elf32_Phdr)); 246 phdr.p_type = cpu_to_dump32(s, PT_LOAD); 247 phdr.p_offset = cpu_to_dump32(s, offset); 248 phdr.p_paddr = cpu_to_dump32(s, memory_mapping->phys_addr); 249 phdr.p_filesz = cpu_to_dump32(s, filesz); 250 phdr.p_memsz = cpu_to_dump32(s, memory_mapping->length); 251 phdr.p_vaddr = 252 cpu_to_dump32(s, memory_mapping->virt_addr) ?: phdr.p_paddr; 253 254 assert(memory_mapping->length >= filesz); 255 256 ret = fd_write_vmcore(&phdr, sizeof(Elf32_Phdr), s); 257 if (ret < 0) { 258 error_setg_errno(errp, -ret, 259 "dump: failed to write program header table"); 260 } 261 } 262 263 static void prepare_elf64_phdr_note(DumpState *s, Elf64_Phdr *phdr) 264 { 265 memset(phdr, 0, sizeof(*phdr)); 266 phdr->p_type = cpu_to_dump32(s, PT_NOTE); 267 phdr->p_offset = cpu_to_dump64(s, s->note_offset); 268 phdr->p_paddr = 0; 269 phdr->p_filesz = cpu_to_dump64(s, s->note_size); 270 phdr->p_memsz = cpu_to_dump64(s, s->note_size); 271 phdr->p_vaddr = 0; 272 } 273 274 static inline int cpu_index(CPUState *cpu) 275 { 276 return cpu->cpu_index + 1; 277 } 278 279 static void write_guest_note(WriteCoreDumpFunction f, DumpState *s, 280 Error **errp) 281 { 282 int ret; 283 284 if (s->guest_note) { 285 ret = f(s->guest_note, s->guest_note_size, s); 286 if (ret < 0) { 287 error_setg(errp, "dump: failed to write guest note"); 288 } 289 } 290 } 291 292 static void write_elf64_notes(WriteCoreDumpFunction f, DumpState *s, 293 Error **errp) 294 { 295 CPUState *cpu; 296 int ret; 297 int id; 298 299 CPU_FOREACH(cpu) { 300 id = cpu_index(cpu); 301 ret = cpu_write_elf64_note(f, cpu, id, s); 302 if (ret < 0) { 303 error_setg(errp, "dump: failed to write elf notes"); 304 return; 305 } 306 } 307 308 CPU_FOREACH(cpu) { 309 ret = cpu_write_elf64_qemunote(f, cpu, s); 310 if (ret < 0) { 311 error_setg(errp, "dump: failed to write CPU status"); 312 return; 313 } 314 } 315 316 write_guest_note(f, s, errp); 317 } 318 319 static void prepare_elf32_phdr_note(DumpState *s, Elf32_Phdr *phdr) 320 { 321 memset(phdr, 0, sizeof(*phdr)); 322 phdr->p_type = cpu_to_dump32(s, PT_NOTE); 323 phdr->p_offset = cpu_to_dump32(s, s->note_offset); 324 phdr->p_paddr = 0; 325 phdr->p_filesz = cpu_to_dump32(s, s->note_size); 326 phdr->p_memsz = cpu_to_dump32(s, s->note_size); 327 phdr->p_vaddr = 0; 328 } 329 330 static void write_elf32_notes(WriteCoreDumpFunction f, DumpState *s, 331 Error **errp) 332 { 333 CPUState *cpu; 334 int ret; 335 int id; 336 337 CPU_FOREACH(cpu) { 338 id = cpu_index(cpu); 339 ret = cpu_write_elf32_note(f, cpu, id, s); 340 if (ret < 0) { 341 error_setg(errp, "dump: failed to write elf notes"); 342 return; 343 } 344 } 345 346 CPU_FOREACH(cpu) { 347 ret = cpu_write_elf32_qemunote(f, cpu, s); 348 if (ret < 0) { 349 error_setg(errp, "dump: failed to write CPU status"); 350 return; 351 } 352 } 353 354 write_guest_note(f, s, errp); 355 } 356 357 static void write_elf_phdr_note(DumpState *s, Error **errp) 358 { 359 Elf32_Phdr phdr32; 360 Elf64_Phdr phdr64; 361 void *phdr; 362 size_t size; 363 int ret; 364 365 if (dump_is_64bit(s)) { 366 prepare_elf64_phdr_note(s, &phdr64); 367 size = sizeof(phdr64); 368 phdr = &phdr64; 369 } else { 370 prepare_elf32_phdr_note(s, &phdr32); 371 size = sizeof(phdr32); 372 phdr = &phdr32; 373 } 374 375 ret = fd_write_vmcore(phdr, size, s); 376 if (ret < 0) { 377 error_setg_errno(errp, -ret, 378 "dump: failed to write program header table"); 379 } 380 } 381 382 static void prepare_elf_section_hdr_zero(DumpState *s) 383 { 384 if (dump_is_64bit(s)) { 385 Elf64_Shdr *shdr64 = s->elf_section_hdrs; 386 387 shdr64->sh_info = cpu_to_dump32(s, s->phdr_num); 388 } else { 389 Elf32_Shdr *shdr32 = s->elf_section_hdrs; 390 391 shdr32->sh_info = cpu_to_dump32(s, s->phdr_num); 392 } 393 } 394 395 static void prepare_elf_section_hdr_string(DumpState *s, void *buff) 396 { 397 uint64_t index = s->string_table_buf->len; 398 const char strtab[] = ".shstrtab"; 399 Elf32_Shdr shdr32 = {}; 400 Elf64_Shdr shdr64 = {}; 401 int shdr_size; 402 void *shdr; 403 404 g_array_append_vals(s->string_table_buf, strtab, sizeof(strtab)); 405 if (dump_is_64bit(s)) { 406 shdr_size = sizeof(Elf64_Shdr); 407 shdr64.sh_type = SHT_STRTAB; 408 shdr64.sh_offset = s->section_offset + s->elf_section_data_size; 409 shdr64.sh_name = index; 410 shdr64.sh_size = s->string_table_buf->len; 411 shdr = &shdr64; 412 } else { 413 shdr_size = sizeof(Elf32_Shdr); 414 shdr32.sh_type = SHT_STRTAB; 415 shdr32.sh_offset = s->section_offset + s->elf_section_data_size; 416 shdr32.sh_name = index; 417 shdr32.sh_size = s->string_table_buf->len; 418 shdr = &shdr32; 419 } 420 memcpy(buff, shdr, shdr_size); 421 } 422 423 static bool prepare_elf_section_hdrs(DumpState *s, Error **errp) 424 { 425 size_t len, sizeof_shdr; 426 void *buff_hdr; 427 428 /* 429 * Section ordering: 430 * - HDR zero 431 * - Arch section hdrs 432 * - String table hdr 433 */ 434 sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr); 435 len = sizeof_shdr * s->shdr_num; 436 s->elf_section_hdrs = g_malloc0(len); 437 buff_hdr = s->elf_section_hdrs; 438 439 /* 440 * The first section header is ALWAYS a special initial section 441 * header. 442 * 443 * The header should be 0 with one exception being that if 444 * phdr_num is PN_XNUM then the sh_info field contains the real 445 * number of segment entries. 446 * 447 * As we zero allocate the buffer we will only need to modify 448 * sh_info for the PN_XNUM case. 449 */ 450 if (s->phdr_num >= PN_XNUM) { 451 prepare_elf_section_hdr_zero(s); 452 } 453 buff_hdr += sizeof_shdr; 454 455 /* Add architecture defined section headers */ 456 if (s->dump_info.arch_sections_write_hdr_fn 457 && s->shdr_num > 2) { 458 buff_hdr += s->dump_info.arch_sections_write_hdr_fn(s, buff_hdr); 459 460 if (s->shdr_num >= SHN_LORESERVE) { 461 error_setg_errno(errp, EINVAL, 462 "dump: too many architecture defined sections"); 463 return false; 464 } 465 } 466 467 /* 468 * String table is the last section since strings are added via 469 * arch_sections_write_hdr(). 470 */ 471 prepare_elf_section_hdr_string(s, buff_hdr); 472 return true; 473 } 474 475 static void write_elf_section_headers(DumpState *s, Error **errp) 476 { 477 size_t sizeof_shdr = dump_is_64bit(s) ? sizeof(Elf64_Shdr) : sizeof(Elf32_Shdr); 478 int ret; 479 480 if (!prepare_elf_section_hdrs(s, errp)) { 481 return; 482 } 483 484 ret = fd_write_vmcore(s->elf_section_hdrs, s->shdr_num * sizeof_shdr, s); 485 if (ret < 0) { 486 error_setg_errno(errp, -ret, "dump: failed to write section headers"); 487 } 488 489 g_free(s->elf_section_hdrs); 490 } 491 492 static void write_elf_sections(DumpState *s, Error **errp) 493 { 494 int ret; 495 496 if (s->elf_section_data_size) { 497 /* Write architecture section data */ 498 ret = fd_write_vmcore(s->elf_section_data, 499 s->elf_section_data_size, s); 500 if (ret < 0) { 501 error_setg_errno(errp, -ret, 502 "dump: failed to write architecture section data"); 503 return; 504 } 505 } 506 507 /* Write string table */ 508 ret = fd_write_vmcore(s->string_table_buf->data, 509 s->string_table_buf->len, s); 510 if (ret < 0) { 511 error_setg_errno(errp, -ret, "dump: failed to write string table data"); 512 } 513 } 514 515 static void write_data(DumpState *s, void *buf, int length, Error **errp) 516 { 517 int ret; 518 519 ret = fd_write_vmcore(buf, length, s); 520 if (ret < 0) { 521 error_setg_errno(errp, -ret, "dump: failed to save memory"); 522 } else { 523 s->written_size += length; 524 } 525 } 526 527 /* write the memory to vmcore. 1 page per I/O. */ 528 static void write_memory(DumpState *s, GuestPhysBlock *block, ram_addr_t start, 529 int64_t size, Error **errp) 530 { 531 ERRP_GUARD(); 532 int64_t i; 533 534 for (i = 0; i < size / s->dump_info.page_size; i++) { 535 write_data(s, block->host_addr + start + i * s->dump_info.page_size, 536 s->dump_info.page_size, errp); 537 if (*errp) { 538 return; 539 } 540 } 541 542 if ((size % s->dump_info.page_size) != 0) { 543 write_data(s, block->host_addr + start + i * s->dump_info.page_size, 544 size % s->dump_info.page_size, errp); 545 if (*errp) { 546 return; 547 } 548 } 549 } 550 551 /* get the memory's offset and size in the vmcore */ 552 static void get_offset_range(hwaddr phys_addr, 553 ram_addr_t mapping_length, 554 DumpState *s, 555 hwaddr *p_offset, 556 hwaddr *p_filesz) 557 { 558 GuestPhysBlock *block; 559 hwaddr offset = s->memory_offset; 560 int64_t size_in_block, start; 561 562 /* When the memory is not stored into vmcore, offset will be -1 */ 563 *p_offset = -1; 564 *p_filesz = 0; 565 566 if (dump_has_filter(s)) { 567 if (phys_addr < s->filter_area_begin || 568 phys_addr >= s->filter_area_begin + s->filter_area_length) { 569 return; 570 } 571 } 572 573 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) { 574 if (dump_has_filter(s)) { 575 if (block->target_start >= s->filter_area_begin + s->filter_area_length || 576 block->target_end <= s->filter_area_begin) { 577 /* This block is out of the range */ 578 continue; 579 } 580 581 if (s->filter_area_begin <= block->target_start) { 582 start = block->target_start; 583 } else { 584 start = s->filter_area_begin; 585 } 586 587 size_in_block = block->target_end - start; 588 if (s->filter_area_begin + s->filter_area_length < block->target_end) { 589 size_in_block -= block->target_end - (s->filter_area_begin + s->filter_area_length); 590 } 591 } else { 592 start = block->target_start; 593 size_in_block = block->target_end - block->target_start; 594 } 595 596 if (phys_addr >= start && phys_addr < start + size_in_block) { 597 *p_offset = phys_addr - start + offset; 598 599 /* The offset range mapped from the vmcore file must not spill over 600 * the GuestPhysBlock, clamp it. The rest of the mapping will be 601 * zero-filled in memory at load time; see 602 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>. 603 */ 604 *p_filesz = phys_addr + mapping_length <= start + size_in_block ? 605 mapping_length : 606 size_in_block - (phys_addr - start); 607 return; 608 } 609 610 offset += size_in_block; 611 } 612 } 613 614 static void write_elf_phdr_loads(DumpState *s, Error **errp) 615 { 616 ERRP_GUARD(); 617 hwaddr offset, filesz; 618 MemoryMapping *memory_mapping; 619 uint32_t phdr_index = 1; 620 621 QTAILQ_FOREACH(memory_mapping, &s->list.head, next) { 622 get_offset_range(memory_mapping->phys_addr, 623 memory_mapping->length, 624 s, &offset, &filesz); 625 if (dump_is_64bit(s)) { 626 write_elf64_load(s, memory_mapping, phdr_index++, offset, 627 filesz, errp); 628 } else { 629 write_elf32_load(s, memory_mapping, phdr_index++, offset, 630 filesz, errp); 631 } 632 633 if (*errp) { 634 return; 635 } 636 637 if (phdr_index >= s->phdr_num) { 638 break; 639 } 640 } 641 } 642 643 static void write_elf_notes(DumpState *s, Error **errp) 644 { 645 if (dump_is_64bit(s)) { 646 write_elf64_notes(fd_write_vmcore, s, errp); 647 } else { 648 write_elf32_notes(fd_write_vmcore, s, errp); 649 } 650 } 651 652 /* write elf header, PT_NOTE and elf note to vmcore. */ 653 static void dump_begin(DumpState *s, Error **errp) 654 { 655 ERRP_GUARD(); 656 657 /* 658 * the vmcore's format is: 659 * -------------- 660 * | elf header | 661 * -------------- 662 * | sctn_hdr | 663 * -------------- 664 * | PT_NOTE | 665 * -------------- 666 * | PT_LOAD | 667 * -------------- 668 * | ...... | 669 * -------------- 670 * | PT_LOAD | 671 * -------------- 672 * | elf note | 673 * -------------- 674 * | memory | 675 * -------------- 676 * 677 * we only know where the memory is saved after we write elf note into 678 * vmcore. 679 */ 680 681 /* write elf header to vmcore */ 682 write_elf_header(s, errp); 683 if (*errp) { 684 return; 685 } 686 687 /* write section headers to vmcore */ 688 write_elf_section_headers(s, errp); 689 if (*errp) { 690 return; 691 } 692 693 /* write PT_NOTE to vmcore */ 694 write_elf_phdr_note(s, errp); 695 if (*errp) { 696 return; 697 } 698 699 /* write all PT_LOADs to vmcore */ 700 write_elf_phdr_loads(s, errp); 701 if (*errp) { 702 return; 703 } 704 705 /* write notes to vmcore */ 706 write_elf_notes(s, errp); 707 } 708 709 int64_t dump_filtered_memblock_size(GuestPhysBlock *block, 710 int64_t filter_area_start, 711 int64_t filter_area_length) 712 { 713 int64_t size, left, right; 714 715 /* No filter, return full size */ 716 if (!filter_area_length) { 717 return block->target_end - block->target_start; 718 } 719 720 /* calculate the overlapped region. */ 721 left = MAX(filter_area_start, block->target_start); 722 right = MIN(filter_area_start + filter_area_length, block->target_end); 723 size = right - left; 724 size = size > 0 ? size : 0; 725 726 return size; 727 } 728 729 int64_t dump_filtered_memblock_start(GuestPhysBlock *block, 730 int64_t filter_area_start, 731 int64_t filter_area_length) 732 { 733 if (filter_area_length) { 734 /* return -1 if the block is not within filter area */ 735 if (block->target_start >= filter_area_start + filter_area_length || 736 block->target_end <= filter_area_start) { 737 return -1; 738 } 739 740 if (filter_area_start > block->target_start) { 741 return filter_area_start - block->target_start; 742 } 743 } 744 745 return 0; 746 } 747 748 /* write all memory to vmcore */ 749 static void dump_iterate(DumpState *s, Error **errp) 750 { 751 ERRP_GUARD(); 752 GuestPhysBlock *block; 753 int64_t memblock_size, memblock_start; 754 755 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) { 756 memblock_start = dump_filtered_memblock_start(block, s->filter_area_begin, s->filter_area_length); 757 if (memblock_start == -1) { 758 continue; 759 } 760 761 memblock_size = dump_filtered_memblock_size(block, s->filter_area_begin, s->filter_area_length); 762 763 /* Write the memory to file */ 764 write_memory(s, block, memblock_start, memblock_size, errp); 765 if (*errp) { 766 return; 767 } 768 } 769 } 770 771 static void dump_end(DumpState *s, Error **errp) 772 { 773 int rc; 774 775 if (s->elf_section_data_size) { 776 s->elf_section_data = g_malloc0(s->elf_section_data_size); 777 } 778 779 /* Adds the architecture defined section data to s->elf_section_data */ 780 if (s->dump_info.arch_sections_write_fn && 781 s->elf_section_data_size) { 782 rc = s->dump_info.arch_sections_write_fn(s, s->elf_section_data); 783 if (rc) { 784 error_setg_errno(errp, rc, 785 "dump: failed to get arch section data"); 786 g_free(s->elf_section_data); 787 return; 788 } 789 } 790 791 /* write sections to vmcore */ 792 write_elf_sections(s, errp); 793 } 794 795 static void create_vmcore(DumpState *s, Error **errp) 796 { 797 ERRP_GUARD(); 798 799 dump_begin(s, errp); 800 if (*errp) { 801 return; 802 } 803 804 /* Iterate over memory and dump it to file */ 805 dump_iterate(s, errp); 806 if (*errp) { 807 return; 808 } 809 810 /* Write the section data */ 811 dump_end(s, errp); 812 } 813 814 static int write_start_flat_header(int fd) 815 { 816 MakedumpfileHeader *mh; 817 int ret = 0; 818 819 QEMU_BUILD_BUG_ON(sizeof *mh > MAX_SIZE_MDF_HEADER); 820 mh = g_malloc0(MAX_SIZE_MDF_HEADER); 821 822 memcpy(mh->signature, MAKEDUMPFILE_SIGNATURE, 823 MIN(sizeof mh->signature, sizeof MAKEDUMPFILE_SIGNATURE)); 824 825 mh->type = cpu_to_be64(TYPE_FLAT_HEADER); 826 mh->version = cpu_to_be64(VERSION_FLAT_HEADER); 827 828 size_t written_size; 829 written_size = qemu_write_full(fd, mh, MAX_SIZE_MDF_HEADER); 830 if (written_size != MAX_SIZE_MDF_HEADER) { 831 ret = -1; 832 } 833 834 g_free(mh); 835 return ret; 836 } 837 838 static int write_end_flat_header(int fd) 839 { 840 MakedumpfileDataHeader mdh; 841 842 mdh.offset = END_FLAG_FLAT_HEADER; 843 mdh.buf_size = END_FLAG_FLAT_HEADER; 844 845 size_t written_size; 846 written_size = qemu_write_full(fd, &mdh, sizeof(mdh)); 847 if (written_size != sizeof(mdh)) { 848 return -1; 849 } 850 851 return 0; 852 } 853 854 static int write_buffer(int fd, off_t offset, const void *buf, size_t size) 855 { 856 size_t written_size; 857 MakedumpfileDataHeader mdh; 858 859 mdh.offset = cpu_to_be64(offset); 860 mdh.buf_size = cpu_to_be64(size); 861 862 written_size = qemu_write_full(fd, &mdh, sizeof(mdh)); 863 if (written_size != sizeof(mdh)) { 864 return -1; 865 } 866 867 written_size = qemu_write_full(fd, buf, size); 868 if (written_size != size) { 869 return -1; 870 } 871 872 return 0; 873 } 874 875 static int buf_write_note(const void *buf, size_t size, void *opaque) 876 { 877 DumpState *s = opaque; 878 879 /* note_buf is not enough */ 880 if (s->note_buf_offset + size > s->note_size) { 881 return -1; 882 } 883 884 memcpy(s->note_buf + s->note_buf_offset, buf, size); 885 886 s->note_buf_offset += size; 887 888 return 0; 889 } 890 891 /* 892 * This function retrieves various sizes from an elf header. 893 * 894 * @note has to be a valid ELF note. The return sizes are unmodified 895 * (not padded or rounded up to be multiple of 4). 896 */ 897 static void get_note_sizes(DumpState *s, const void *note, 898 uint64_t *note_head_size, 899 uint64_t *name_size, 900 uint64_t *desc_size) 901 { 902 uint64_t note_head_sz; 903 uint64_t name_sz; 904 uint64_t desc_sz; 905 906 if (dump_is_64bit(s)) { 907 const Elf64_Nhdr *hdr = note; 908 note_head_sz = sizeof(Elf64_Nhdr); 909 name_sz = cpu_to_dump64(s, hdr->n_namesz); 910 desc_sz = cpu_to_dump64(s, hdr->n_descsz); 911 } else { 912 const Elf32_Nhdr *hdr = note; 913 note_head_sz = sizeof(Elf32_Nhdr); 914 name_sz = cpu_to_dump32(s, hdr->n_namesz); 915 desc_sz = cpu_to_dump32(s, hdr->n_descsz); 916 } 917 918 if (note_head_size) { 919 *note_head_size = note_head_sz; 920 } 921 if (name_size) { 922 *name_size = name_sz; 923 } 924 if (desc_size) { 925 *desc_size = desc_sz; 926 } 927 } 928 929 static bool note_name_equal(DumpState *s, 930 const uint8_t *note, const char *name) 931 { 932 int len = strlen(name) + 1; 933 uint64_t head_size, name_size; 934 935 get_note_sizes(s, note, &head_size, &name_size, NULL); 936 head_size = ROUND_UP(head_size, 4); 937 938 return name_size == len && memcmp(note + head_size, name, len) == 0; 939 } 940 941 /* write common header, sub header and elf note to vmcore */ 942 static void create_header32(DumpState *s, Error **errp) 943 { 944 ERRP_GUARD(); 945 DiskDumpHeader32 *dh = NULL; 946 KdumpSubHeader32 *kh = NULL; 947 size_t size; 948 uint32_t block_size; 949 uint32_t sub_hdr_size; 950 uint32_t bitmap_blocks; 951 uint32_t status = 0; 952 uint64_t offset_note; 953 954 /* write common header, the version of kdump-compressed format is 6th */ 955 size = sizeof(DiskDumpHeader32); 956 dh = g_malloc0(size); 957 958 memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN); 959 dh->header_version = cpu_to_dump32(s, 6); 960 block_size = s->dump_info.page_size; 961 dh->block_size = cpu_to_dump32(s, block_size); 962 sub_hdr_size = sizeof(struct KdumpSubHeader32) + s->note_size; 963 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size); 964 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size); 965 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */ 966 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX)); 967 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus); 968 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2; 969 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks); 970 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine)); 971 972 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) { 973 status |= DUMP_DH_COMPRESSED_ZLIB; 974 } 975 #ifdef CONFIG_LZO 976 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) { 977 status |= DUMP_DH_COMPRESSED_LZO; 978 } 979 #endif 980 #ifdef CONFIG_SNAPPY 981 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) { 982 status |= DUMP_DH_COMPRESSED_SNAPPY; 983 } 984 #endif 985 dh->status = cpu_to_dump32(s, status); 986 987 if (write_buffer(s->fd, 0, dh, size) < 0) { 988 error_setg(errp, "dump: failed to write disk dump header"); 989 goto out; 990 } 991 992 /* write sub header */ 993 size = sizeof(KdumpSubHeader32); 994 kh = g_malloc0(size); 995 996 /* 64bit max_mapnr_64 */ 997 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr); 998 kh->phys_base = cpu_to_dump32(s, s->dump_info.phys_base); 999 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL); 1000 1001 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size; 1002 if (s->guest_note && 1003 note_name_equal(s, s->guest_note, "VMCOREINFO")) { 1004 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo; 1005 1006 get_note_sizes(s, s->guest_note, 1007 &hsize, &name_size, &size_vmcoreinfo_desc); 1008 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size + 1009 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4; 1010 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo); 1011 kh->size_vmcoreinfo = cpu_to_dump32(s, size_vmcoreinfo_desc); 1012 } 1013 1014 kh->offset_note = cpu_to_dump64(s, offset_note); 1015 kh->note_size = cpu_to_dump32(s, s->note_size); 1016 1017 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS * 1018 block_size, kh, size) < 0) { 1019 error_setg(errp, "dump: failed to write kdump sub header"); 1020 goto out; 1021 } 1022 1023 /* write note */ 1024 s->note_buf = g_malloc0(s->note_size); 1025 s->note_buf_offset = 0; 1026 1027 /* use s->note_buf to store notes temporarily */ 1028 write_elf32_notes(buf_write_note, s, errp); 1029 if (*errp) { 1030 goto out; 1031 } 1032 if (write_buffer(s->fd, offset_note, s->note_buf, 1033 s->note_size) < 0) { 1034 error_setg(errp, "dump: failed to write notes"); 1035 goto out; 1036 } 1037 1038 /* get offset of dump_bitmap */ 1039 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) * 1040 block_size; 1041 1042 /* get offset of page */ 1043 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) * 1044 block_size; 1045 1046 out: 1047 g_free(dh); 1048 g_free(kh); 1049 g_free(s->note_buf); 1050 } 1051 1052 /* write common header, sub header and elf note to vmcore */ 1053 static void create_header64(DumpState *s, Error **errp) 1054 { 1055 ERRP_GUARD(); 1056 DiskDumpHeader64 *dh = NULL; 1057 KdumpSubHeader64 *kh = NULL; 1058 size_t size; 1059 uint32_t block_size; 1060 uint32_t sub_hdr_size; 1061 uint32_t bitmap_blocks; 1062 uint32_t status = 0; 1063 uint64_t offset_note; 1064 1065 /* write common header, the version of kdump-compressed format is 6th */ 1066 size = sizeof(DiskDumpHeader64); 1067 dh = g_malloc0(size); 1068 1069 memcpy(dh->signature, KDUMP_SIGNATURE, SIG_LEN); 1070 dh->header_version = cpu_to_dump32(s, 6); 1071 block_size = s->dump_info.page_size; 1072 dh->block_size = cpu_to_dump32(s, block_size); 1073 sub_hdr_size = sizeof(struct KdumpSubHeader64) + s->note_size; 1074 sub_hdr_size = DIV_ROUND_UP(sub_hdr_size, block_size); 1075 dh->sub_hdr_size = cpu_to_dump32(s, sub_hdr_size); 1076 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */ 1077 dh->max_mapnr = cpu_to_dump32(s, MIN(s->max_mapnr, UINT_MAX)); 1078 dh->nr_cpus = cpu_to_dump32(s, s->nr_cpus); 1079 bitmap_blocks = DIV_ROUND_UP(s->len_dump_bitmap, block_size) * 2; 1080 dh->bitmap_blocks = cpu_to_dump32(s, bitmap_blocks); 1081 strncpy(dh->utsname.machine, ELF_MACHINE_UNAME, sizeof(dh->utsname.machine)); 1082 1083 if (s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) { 1084 status |= DUMP_DH_COMPRESSED_ZLIB; 1085 } 1086 #ifdef CONFIG_LZO 1087 if (s->flag_compress & DUMP_DH_COMPRESSED_LZO) { 1088 status |= DUMP_DH_COMPRESSED_LZO; 1089 } 1090 #endif 1091 #ifdef CONFIG_SNAPPY 1092 if (s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) { 1093 status |= DUMP_DH_COMPRESSED_SNAPPY; 1094 } 1095 #endif 1096 dh->status = cpu_to_dump32(s, status); 1097 1098 if (write_buffer(s->fd, 0, dh, size) < 0) { 1099 error_setg(errp, "dump: failed to write disk dump header"); 1100 goto out; 1101 } 1102 1103 /* write sub header */ 1104 size = sizeof(KdumpSubHeader64); 1105 kh = g_malloc0(size); 1106 1107 /* 64bit max_mapnr_64 */ 1108 kh->max_mapnr_64 = cpu_to_dump64(s, s->max_mapnr); 1109 kh->phys_base = cpu_to_dump64(s, s->dump_info.phys_base); 1110 kh->dump_level = cpu_to_dump32(s, DUMP_LEVEL); 1111 1112 offset_note = DISKDUMP_HEADER_BLOCKS * block_size + size; 1113 if (s->guest_note && 1114 note_name_equal(s, s->guest_note, "VMCOREINFO")) { 1115 uint64_t hsize, name_size, size_vmcoreinfo_desc, offset_vmcoreinfo; 1116 1117 get_note_sizes(s, s->guest_note, 1118 &hsize, &name_size, &size_vmcoreinfo_desc); 1119 offset_vmcoreinfo = offset_note + s->note_size - s->guest_note_size + 1120 (DIV_ROUND_UP(hsize, 4) + DIV_ROUND_UP(name_size, 4)) * 4; 1121 kh->offset_vmcoreinfo = cpu_to_dump64(s, offset_vmcoreinfo); 1122 kh->size_vmcoreinfo = cpu_to_dump64(s, size_vmcoreinfo_desc); 1123 } 1124 1125 kh->offset_note = cpu_to_dump64(s, offset_note); 1126 kh->note_size = cpu_to_dump64(s, s->note_size); 1127 1128 if (write_buffer(s->fd, DISKDUMP_HEADER_BLOCKS * 1129 block_size, kh, size) < 0) { 1130 error_setg(errp, "dump: failed to write kdump sub header"); 1131 goto out; 1132 } 1133 1134 /* write note */ 1135 s->note_buf = g_malloc0(s->note_size); 1136 s->note_buf_offset = 0; 1137 1138 /* use s->note_buf to store notes temporarily */ 1139 write_elf64_notes(buf_write_note, s, errp); 1140 if (*errp) { 1141 goto out; 1142 } 1143 1144 if (write_buffer(s->fd, offset_note, s->note_buf, 1145 s->note_size) < 0) { 1146 error_setg(errp, "dump: failed to write notes"); 1147 goto out; 1148 } 1149 1150 /* get offset of dump_bitmap */ 1151 s->offset_dump_bitmap = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size) * 1152 block_size; 1153 1154 /* get offset of page */ 1155 s->offset_page = (DISKDUMP_HEADER_BLOCKS + sub_hdr_size + bitmap_blocks) * 1156 block_size; 1157 1158 out: 1159 g_free(dh); 1160 g_free(kh); 1161 g_free(s->note_buf); 1162 } 1163 1164 static void write_dump_header(DumpState *s, Error **errp) 1165 { 1166 if (dump_is_64bit(s)) { 1167 create_header64(s, errp); 1168 } else { 1169 create_header32(s, errp); 1170 } 1171 } 1172 1173 static size_t dump_bitmap_get_bufsize(DumpState *s) 1174 { 1175 return s->dump_info.page_size; 1176 } 1177 1178 /* 1179 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be 1180 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0. 1181 * set_dump_bitmap will always leave the recently set bit un-sync. And setting 1182 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into 1183 * vmcore, ie. synchronizing un-sync bit into vmcore. 1184 */ 1185 static int set_dump_bitmap(uint64_t last_pfn, uint64_t pfn, bool value, 1186 uint8_t *buf, DumpState *s) 1187 { 1188 off_t old_offset, new_offset; 1189 off_t offset_bitmap1, offset_bitmap2; 1190 uint32_t byte, bit; 1191 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s); 1192 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT; 1193 1194 /* should not set the previous place */ 1195 assert(last_pfn <= pfn); 1196 1197 /* 1198 * if the bit needed to be set is not cached in buf, flush the data in buf 1199 * to vmcore firstly. 1200 * making new_offset be bigger than old_offset can also sync remained data 1201 * into vmcore. 1202 */ 1203 old_offset = bitmap_bufsize * (last_pfn / bits_per_buf); 1204 new_offset = bitmap_bufsize * (pfn / bits_per_buf); 1205 1206 while (old_offset < new_offset) { 1207 /* calculate the offset and write dump_bitmap */ 1208 offset_bitmap1 = s->offset_dump_bitmap + old_offset; 1209 if (write_buffer(s->fd, offset_bitmap1, buf, 1210 bitmap_bufsize) < 0) { 1211 return -1; 1212 } 1213 1214 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */ 1215 offset_bitmap2 = s->offset_dump_bitmap + s->len_dump_bitmap + 1216 old_offset; 1217 if (write_buffer(s->fd, offset_bitmap2, buf, 1218 bitmap_bufsize) < 0) { 1219 return -1; 1220 } 1221 1222 memset(buf, 0, bitmap_bufsize); 1223 old_offset += bitmap_bufsize; 1224 } 1225 1226 /* get the exact place of the bit in the buf, and set it */ 1227 byte = (pfn % bits_per_buf) / CHAR_BIT; 1228 bit = (pfn % bits_per_buf) % CHAR_BIT; 1229 if (value) { 1230 buf[byte] |= 1u << bit; 1231 } else { 1232 buf[byte] &= ~(1u << bit); 1233 } 1234 1235 return 0; 1236 } 1237 1238 static uint64_t dump_paddr_to_pfn(DumpState *s, uint64_t addr) 1239 { 1240 int target_page_shift = ctz32(s->dump_info.page_size); 1241 1242 return (addr >> target_page_shift) - ARCH_PFN_OFFSET; 1243 } 1244 1245 static uint64_t dump_pfn_to_paddr(DumpState *s, uint64_t pfn) 1246 { 1247 int target_page_shift = ctz32(s->dump_info.page_size); 1248 1249 return (pfn + ARCH_PFN_OFFSET) << target_page_shift; 1250 } 1251 1252 /* 1253 * Return the page frame number and the page content in *bufptr. bufptr can be 1254 * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated 1255 * memory. This is not necessarily the memory returned. 1256 */ 1257 static bool get_next_page(GuestPhysBlock **blockptr, uint64_t *pfnptr, 1258 uint8_t **bufptr, DumpState *s) 1259 { 1260 GuestPhysBlock *block = *blockptr; 1261 uint32_t page_size = s->dump_info.page_size; 1262 uint8_t *buf = NULL, *hbuf; 1263 hwaddr addr; 1264 1265 /* block == NULL means the start of the iteration */ 1266 if (!block) { 1267 block = QTAILQ_FIRST(&s->guest_phys_blocks.head); 1268 *blockptr = block; 1269 addr = block->target_start; 1270 *pfnptr = dump_paddr_to_pfn(s, addr); 1271 } else { 1272 *pfnptr += 1; 1273 addr = dump_pfn_to_paddr(s, *pfnptr); 1274 } 1275 assert(block != NULL); 1276 1277 while (1) { 1278 if (addr >= block->target_start && addr < block->target_end) { 1279 size_t n = MIN(block->target_end - addr, page_size - addr % page_size); 1280 hbuf = block->host_addr + (addr - block->target_start); 1281 if (!buf) { 1282 if (n == page_size) { 1283 /* this is a whole target page, go for it */ 1284 assert(addr % page_size == 0); 1285 buf = hbuf; 1286 break; 1287 } else if (bufptr) { 1288 assert(*bufptr); 1289 buf = *bufptr; 1290 memset(buf, 0, page_size); 1291 } else { 1292 return true; 1293 } 1294 } 1295 1296 memcpy(buf + addr % page_size, hbuf, n); 1297 addr += n; 1298 if (addr % page_size == 0) { 1299 /* we filled up the page */ 1300 break; 1301 } 1302 } else { 1303 /* the next page is in the next block */ 1304 *blockptr = block = QTAILQ_NEXT(block, next); 1305 if (!block) { 1306 break; 1307 } 1308 1309 addr = block->target_start; 1310 /* are we still in the same page? */ 1311 if (dump_paddr_to_pfn(s, addr) != *pfnptr) { 1312 if (buf) { 1313 /* no, but we already filled something earlier, return it */ 1314 break; 1315 } else { 1316 /* else continue from there */ 1317 *pfnptr = dump_paddr_to_pfn(s, addr); 1318 } 1319 } 1320 } 1321 } 1322 1323 if (bufptr) { 1324 *bufptr = buf; 1325 } 1326 1327 return buf != NULL; 1328 } 1329 1330 static void write_dump_bitmap(DumpState *s, Error **errp) 1331 { 1332 int ret = 0; 1333 uint64_t last_pfn, pfn; 1334 void *dump_bitmap_buf; 1335 size_t num_dumpable; 1336 GuestPhysBlock *block_iter = NULL; 1337 size_t bitmap_bufsize = dump_bitmap_get_bufsize(s); 1338 size_t bits_per_buf = bitmap_bufsize * CHAR_BIT; 1339 1340 /* dump_bitmap_buf is used to store dump_bitmap temporarily */ 1341 dump_bitmap_buf = g_malloc0(bitmap_bufsize); 1342 1343 num_dumpable = 0; 1344 last_pfn = 0; 1345 1346 /* 1347 * exam memory page by page, and set the bit in dump_bitmap corresponded 1348 * to the existing page. 1349 */ 1350 while (get_next_page(&block_iter, &pfn, NULL, s)) { 1351 ret = set_dump_bitmap(last_pfn, pfn, true, dump_bitmap_buf, s); 1352 if (ret < 0) { 1353 error_setg(errp, "dump: failed to set dump_bitmap"); 1354 goto out; 1355 } 1356 1357 last_pfn = pfn; 1358 num_dumpable++; 1359 } 1360 1361 /* 1362 * set_dump_bitmap will always leave the recently set bit un-sync. Here we 1363 * set the remaining bits from last_pfn to the end of the bitmap buffer to 1364 * 0. With those set, the un-sync bit will be synchronized into the vmcore. 1365 */ 1366 if (num_dumpable > 0) { 1367 ret = set_dump_bitmap(last_pfn, last_pfn + bits_per_buf, false, 1368 dump_bitmap_buf, s); 1369 if (ret < 0) { 1370 error_setg(errp, "dump: failed to sync dump_bitmap"); 1371 goto out; 1372 } 1373 } 1374 1375 /* number of dumpable pages that will be dumped later */ 1376 s->num_dumpable = num_dumpable; 1377 1378 out: 1379 g_free(dump_bitmap_buf); 1380 } 1381 1382 static void prepare_data_cache(DataCache *data_cache, DumpState *s, 1383 off_t offset) 1384 { 1385 data_cache->fd = s->fd; 1386 data_cache->data_size = 0; 1387 data_cache->buf_size = 4 * dump_bitmap_get_bufsize(s); 1388 data_cache->buf = g_malloc0(data_cache->buf_size); 1389 data_cache->offset = offset; 1390 } 1391 1392 static int write_cache(DataCache *dc, const void *buf, size_t size, 1393 bool flag_sync) 1394 { 1395 /* 1396 * dc->buf_size should not be less than size, otherwise dc will never be 1397 * enough 1398 */ 1399 assert(size <= dc->buf_size); 1400 1401 /* 1402 * if flag_sync is set, synchronize data in dc->buf into vmcore. 1403 * otherwise check if the space is enough for caching data in buf, if not, 1404 * write the data in dc->buf to dc->fd and reset dc->buf 1405 */ 1406 if ((!flag_sync && dc->data_size + size > dc->buf_size) || 1407 (flag_sync && dc->data_size > 0)) { 1408 if (write_buffer(dc->fd, dc->offset, dc->buf, dc->data_size) < 0) { 1409 return -1; 1410 } 1411 1412 dc->offset += dc->data_size; 1413 dc->data_size = 0; 1414 } 1415 1416 if (!flag_sync) { 1417 memcpy(dc->buf + dc->data_size, buf, size); 1418 dc->data_size += size; 1419 } 1420 1421 return 0; 1422 } 1423 1424 static void free_data_cache(DataCache *data_cache) 1425 { 1426 g_free(data_cache->buf); 1427 } 1428 1429 static size_t get_len_buf_out(size_t page_size, uint32_t flag_compress) 1430 { 1431 switch (flag_compress) { 1432 case DUMP_DH_COMPRESSED_ZLIB: 1433 return compressBound(page_size); 1434 1435 case DUMP_DH_COMPRESSED_LZO: 1436 /* 1437 * LZO will expand incompressible data by a little amount. Please check 1438 * the following URL to see the expansion calculation: 1439 * http://www.oberhumer.com/opensource/lzo/lzofaq.php 1440 */ 1441 return page_size + page_size / 16 + 64 + 3; 1442 1443 #ifdef CONFIG_SNAPPY 1444 case DUMP_DH_COMPRESSED_SNAPPY: 1445 return snappy_max_compressed_length(page_size); 1446 #endif 1447 } 1448 return 0; 1449 } 1450 1451 static void write_dump_pages(DumpState *s, Error **errp) 1452 { 1453 int ret = 0; 1454 DataCache page_desc, page_data; 1455 size_t len_buf_out, size_out; 1456 #ifdef CONFIG_LZO 1457 lzo_bytep wrkmem = NULL; 1458 #endif 1459 uint8_t *buf_out = NULL; 1460 off_t offset_desc, offset_data; 1461 PageDescriptor pd, pd_zero; 1462 uint8_t *buf; 1463 GuestPhysBlock *block_iter = NULL; 1464 uint64_t pfn_iter; 1465 g_autofree uint8_t *page = NULL; 1466 1467 /* get offset of page_desc and page_data in dump file */ 1468 offset_desc = s->offset_page; 1469 offset_data = offset_desc + sizeof(PageDescriptor) * s->num_dumpable; 1470 1471 prepare_data_cache(&page_desc, s, offset_desc); 1472 prepare_data_cache(&page_data, s, offset_data); 1473 1474 /* prepare buffer to store compressed data */ 1475 len_buf_out = get_len_buf_out(s->dump_info.page_size, s->flag_compress); 1476 assert(len_buf_out != 0); 1477 1478 #ifdef CONFIG_LZO 1479 wrkmem = g_malloc(LZO1X_1_MEM_COMPRESS); 1480 #endif 1481 1482 buf_out = g_malloc(len_buf_out); 1483 1484 /* 1485 * init zero page's page_desc and page_data, because every zero page 1486 * uses the same page_data 1487 */ 1488 pd_zero.size = cpu_to_dump32(s, s->dump_info.page_size); 1489 pd_zero.flags = cpu_to_dump32(s, 0); 1490 pd_zero.offset = cpu_to_dump64(s, offset_data); 1491 pd_zero.page_flags = cpu_to_dump64(s, 0); 1492 buf = g_malloc0(s->dump_info.page_size); 1493 ret = write_cache(&page_data, buf, s->dump_info.page_size, false); 1494 g_free(buf); 1495 if (ret < 0) { 1496 error_setg(errp, "dump: failed to write page data (zero page)"); 1497 goto out; 1498 } 1499 1500 offset_data += s->dump_info.page_size; 1501 page = g_malloc(s->dump_info.page_size); 1502 1503 /* 1504 * dump memory to vmcore page by page. zero page will all be resided in the 1505 * first page of page section 1506 */ 1507 for (buf = page; get_next_page(&block_iter, &pfn_iter, &buf, s); buf = page) { 1508 /* check zero page */ 1509 if (buffer_is_zero(buf, s->dump_info.page_size)) { 1510 ret = write_cache(&page_desc, &pd_zero, sizeof(PageDescriptor), 1511 false); 1512 if (ret < 0) { 1513 error_setg(errp, "dump: failed to write page desc"); 1514 goto out; 1515 } 1516 } else { 1517 /* 1518 * not zero page, then: 1519 * 1. compress the page 1520 * 2. write the compressed page into the cache of page_data 1521 * 3. get page desc of the compressed page and write it into the 1522 * cache of page_desc 1523 * 1524 * only one compression format will be used here, for 1525 * s->flag_compress is set. But when compression fails to work, 1526 * we fall back to save in plaintext. 1527 */ 1528 size_out = len_buf_out; 1529 if ((s->flag_compress & DUMP_DH_COMPRESSED_ZLIB) && 1530 (compress2(buf_out, (uLongf *)&size_out, buf, 1531 s->dump_info.page_size, Z_BEST_SPEED) == Z_OK) && 1532 (size_out < s->dump_info.page_size)) { 1533 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_ZLIB); 1534 pd.size = cpu_to_dump32(s, size_out); 1535 1536 ret = write_cache(&page_data, buf_out, size_out, false); 1537 if (ret < 0) { 1538 error_setg(errp, "dump: failed to write page data"); 1539 goto out; 1540 } 1541 #ifdef CONFIG_LZO 1542 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_LZO) && 1543 (lzo1x_1_compress(buf, s->dump_info.page_size, buf_out, 1544 (lzo_uint *)&size_out, wrkmem) == LZO_E_OK) && 1545 (size_out < s->dump_info.page_size)) { 1546 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_LZO); 1547 pd.size = cpu_to_dump32(s, size_out); 1548 1549 ret = write_cache(&page_data, buf_out, size_out, false); 1550 if (ret < 0) { 1551 error_setg(errp, "dump: failed to write page data"); 1552 goto out; 1553 } 1554 #endif 1555 #ifdef CONFIG_SNAPPY 1556 } else if ((s->flag_compress & DUMP_DH_COMPRESSED_SNAPPY) && 1557 (snappy_compress((char *)buf, s->dump_info.page_size, 1558 (char *)buf_out, &size_out) == SNAPPY_OK) && 1559 (size_out < s->dump_info.page_size)) { 1560 pd.flags = cpu_to_dump32(s, DUMP_DH_COMPRESSED_SNAPPY); 1561 pd.size = cpu_to_dump32(s, size_out); 1562 1563 ret = write_cache(&page_data, buf_out, size_out, false); 1564 if (ret < 0) { 1565 error_setg(errp, "dump: failed to write page data"); 1566 goto out; 1567 } 1568 #endif 1569 } else { 1570 /* 1571 * fall back to save in plaintext, size_out should be 1572 * assigned the target's page size 1573 */ 1574 pd.flags = cpu_to_dump32(s, 0); 1575 size_out = s->dump_info.page_size; 1576 pd.size = cpu_to_dump32(s, size_out); 1577 1578 ret = write_cache(&page_data, buf, 1579 s->dump_info.page_size, false); 1580 if (ret < 0) { 1581 error_setg(errp, "dump: failed to write page data"); 1582 goto out; 1583 } 1584 } 1585 1586 /* get and write page desc here */ 1587 pd.page_flags = cpu_to_dump64(s, 0); 1588 pd.offset = cpu_to_dump64(s, offset_data); 1589 offset_data += size_out; 1590 1591 ret = write_cache(&page_desc, &pd, sizeof(PageDescriptor), false); 1592 if (ret < 0) { 1593 error_setg(errp, "dump: failed to write page desc"); 1594 goto out; 1595 } 1596 } 1597 s->written_size += s->dump_info.page_size; 1598 } 1599 1600 ret = write_cache(&page_desc, NULL, 0, true); 1601 if (ret < 0) { 1602 error_setg(errp, "dump: failed to sync cache for page_desc"); 1603 goto out; 1604 } 1605 ret = write_cache(&page_data, NULL, 0, true); 1606 if (ret < 0) { 1607 error_setg(errp, "dump: failed to sync cache for page_data"); 1608 goto out; 1609 } 1610 1611 out: 1612 free_data_cache(&page_desc); 1613 free_data_cache(&page_data); 1614 1615 #ifdef CONFIG_LZO 1616 g_free(wrkmem); 1617 #endif 1618 1619 g_free(buf_out); 1620 } 1621 1622 static void create_kdump_vmcore(DumpState *s, Error **errp) 1623 { 1624 ERRP_GUARD(); 1625 int ret; 1626 1627 /* 1628 * the kdump-compressed format is: 1629 * File offset 1630 * +------------------------------------------+ 0x0 1631 * | main header (struct disk_dump_header) | 1632 * |------------------------------------------+ block 1 1633 * | sub header (struct kdump_sub_header) | 1634 * |------------------------------------------+ block 2 1635 * | 1st-dump_bitmap | 1636 * |------------------------------------------+ block 2 + X blocks 1637 * | 2nd-dump_bitmap | (aligned by block) 1638 * |------------------------------------------+ block 2 + 2 * X blocks 1639 * | page desc for pfn 0 (struct page_desc) | (aligned by block) 1640 * | page desc for pfn 1 (struct page_desc) | 1641 * | : | 1642 * |------------------------------------------| (not aligned by block) 1643 * | page data (pfn 0) | 1644 * | page data (pfn 1) | 1645 * | : | 1646 * +------------------------------------------+ 1647 */ 1648 1649 ret = write_start_flat_header(s->fd); 1650 if (ret < 0) { 1651 error_setg(errp, "dump: failed to write start flat header"); 1652 return; 1653 } 1654 1655 write_dump_header(s, errp); 1656 if (*errp) { 1657 return; 1658 } 1659 1660 write_dump_bitmap(s, errp); 1661 if (*errp) { 1662 return; 1663 } 1664 1665 write_dump_pages(s, errp); 1666 if (*errp) { 1667 return; 1668 } 1669 1670 ret = write_end_flat_header(s->fd); 1671 if (ret < 0) { 1672 error_setg(errp, "dump: failed to write end flat header"); 1673 return; 1674 } 1675 } 1676 1677 static int validate_start_block(DumpState *s) 1678 { 1679 GuestPhysBlock *block; 1680 1681 if (!dump_has_filter(s)) { 1682 return 0; 1683 } 1684 1685 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) { 1686 /* This block is out of the range */ 1687 if (block->target_start >= s->filter_area_begin + s->filter_area_length || 1688 block->target_end <= s->filter_area_begin) { 1689 continue; 1690 } 1691 return 0; 1692 } 1693 1694 return -1; 1695 } 1696 1697 static void get_max_mapnr(DumpState *s) 1698 { 1699 GuestPhysBlock *last_block; 1700 1701 last_block = QTAILQ_LAST(&s->guest_phys_blocks.head); 1702 s->max_mapnr = dump_paddr_to_pfn(s, last_block->target_end); 1703 } 1704 1705 static DumpState dump_state_global = { .status = DUMP_STATUS_NONE }; 1706 1707 static void dump_state_prepare(DumpState *s) 1708 { 1709 /* zero the struct, setting status to active */ 1710 *s = (DumpState) { .status = DUMP_STATUS_ACTIVE }; 1711 } 1712 1713 bool qemu_system_dump_in_progress(void) 1714 { 1715 DumpState *state = &dump_state_global; 1716 return (qatomic_read(&state->status) == DUMP_STATUS_ACTIVE); 1717 } 1718 1719 /* 1720 * calculate total size of memory to be dumped (taking filter into 1721 * account.) 1722 */ 1723 static int64_t dump_calculate_size(DumpState *s) 1724 { 1725 GuestPhysBlock *block; 1726 int64_t total = 0; 1727 1728 QTAILQ_FOREACH(block, &s->guest_phys_blocks.head, next) { 1729 total += dump_filtered_memblock_size(block, 1730 s->filter_area_begin, 1731 s->filter_area_length); 1732 } 1733 1734 return total; 1735 } 1736 1737 static void vmcoreinfo_update_phys_base(DumpState *s) 1738 { 1739 uint64_t size, note_head_size, name_size, phys_base; 1740 char **lines; 1741 uint8_t *vmci; 1742 size_t i; 1743 1744 if (!note_name_equal(s, s->guest_note, "VMCOREINFO")) { 1745 return; 1746 } 1747 1748 get_note_sizes(s, s->guest_note, ¬e_head_size, &name_size, &size); 1749 note_head_size = ROUND_UP(note_head_size, 4); 1750 1751 vmci = s->guest_note + note_head_size + ROUND_UP(name_size, 4); 1752 *(vmci + size) = '\0'; 1753 1754 lines = g_strsplit((char *)vmci, "\n", -1); 1755 for (i = 0; lines[i]; i++) { 1756 const char *prefix = NULL; 1757 1758 if (s->dump_info.d_machine == EM_X86_64) { 1759 prefix = "NUMBER(phys_base)="; 1760 } else if (s->dump_info.d_machine == EM_AARCH64) { 1761 prefix = "NUMBER(PHYS_OFFSET)="; 1762 } 1763 1764 if (prefix && g_str_has_prefix(lines[i], prefix)) { 1765 if (qemu_strtou64(lines[i] + strlen(prefix), NULL, 16, 1766 &phys_base) < 0) { 1767 warn_report("Failed to read %s", prefix); 1768 } else { 1769 s->dump_info.phys_base = phys_base; 1770 } 1771 break; 1772 } 1773 } 1774 1775 g_strfreev(lines); 1776 } 1777 1778 static void dump_init(DumpState *s, int fd, bool has_format, 1779 DumpGuestMemoryFormat format, bool paging, bool has_filter, 1780 int64_t begin, int64_t length, Error **errp) 1781 { 1782 ERRP_GUARD(); 1783 VMCoreInfoState *vmci = vmcoreinfo_find(); 1784 CPUState *cpu; 1785 int nr_cpus; 1786 int ret; 1787 1788 s->has_format = has_format; 1789 s->format = format; 1790 s->written_size = 0; 1791 1792 /* kdump-compressed is conflict with paging and filter */ 1793 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) { 1794 assert(!paging && !has_filter); 1795 } 1796 1797 if (runstate_is_running()) { 1798 vm_stop(RUN_STATE_SAVE_VM); 1799 s->resume = true; 1800 } else { 1801 s->resume = false; 1802 } 1803 1804 /* If we use KVM, we should synchronize the registers before we get dump 1805 * info or physmap info. 1806 */ 1807 cpu_synchronize_all_states(); 1808 nr_cpus = 0; 1809 CPU_FOREACH(cpu) { 1810 nr_cpus++; 1811 } 1812 1813 s->fd = fd; 1814 if (has_filter && !length) { 1815 error_setg(errp, QERR_INVALID_PARAMETER, "length"); 1816 goto cleanup; 1817 } 1818 s->filter_area_begin = begin; 1819 s->filter_area_length = length; 1820 1821 /* First index is 0, it's the special null name */ 1822 s->string_table_buf = g_array_new(FALSE, TRUE, 1); 1823 /* 1824 * Allocate the null name, due to the clearing option set to true 1825 * it will be 0. 1826 */ 1827 g_array_set_size(s->string_table_buf, 1); 1828 1829 memory_mapping_list_init(&s->list); 1830 1831 guest_phys_blocks_init(&s->guest_phys_blocks); 1832 guest_phys_blocks_append(&s->guest_phys_blocks); 1833 s->total_size = dump_calculate_size(s); 1834 #ifdef DEBUG_DUMP_GUEST_MEMORY 1835 fprintf(stderr, "DUMP: total memory to dump: %lu\n", s->total_size); 1836 #endif 1837 1838 /* it does not make sense to dump non-existent memory */ 1839 if (!s->total_size) { 1840 error_setg(errp, "dump: no guest memory to dump"); 1841 goto cleanup; 1842 } 1843 1844 /* Is the filter filtering everything? */ 1845 if (validate_start_block(s) == -1) { 1846 error_setg(errp, QERR_INVALID_PARAMETER, "begin"); 1847 goto cleanup; 1848 } 1849 1850 /* get dump info: endian, class and architecture. 1851 * If the target architecture is not supported, cpu_get_dump_info() will 1852 * return -1. 1853 */ 1854 ret = cpu_get_dump_info(&s->dump_info, &s->guest_phys_blocks); 1855 if (ret < 0) { 1856 error_setg(errp, 1857 "dumping guest memory is not supported on this target"); 1858 goto cleanup; 1859 } 1860 1861 if (!s->dump_info.page_size) { 1862 s->dump_info.page_size = qemu_target_page_size(); 1863 } 1864 1865 s->note_size = cpu_get_note_size(s->dump_info.d_class, 1866 s->dump_info.d_machine, nr_cpus); 1867 assert(s->note_size >= 0); 1868 1869 /* 1870 * The goal of this block is to (a) update the previously guessed 1871 * phys_base, (b) copy the guest note out of the guest. 1872 * Failure to do so is not fatal for dumping. 1873 */ 1874 if (vmci) { 1875 uint64_t addr, note_head_size, name_size, desc_size; 1876 uint32_t size; 1877 uint16_t format; 1878 1879 note_head_size = dump_is_64bit(s) ? 1880 sizeof(Elf64_Nhdr) : sizeof(Elf32_Nhdr); 1881 1882 format = le16_to_cpu(vmci->vmcoreinfo.guest_format); 1883 size = le32_to_cpu(vmci->vmcoreinfo.size); 1884 addr = le64_to_cpu(vmci->vmcoreinfo.paddr); 1885 if (!vmci->has_vmcoreinfo) { 1886 warn_report("guest note is not present"); 1887 } else if (size < note_head_size || size > MAX_GUEST_NOTE_SIZE) { 1888 warn_report("guest note size is invalid: %" PRIu32, size); 1889 } else if (format != FW_CFG_VMCOREINFO_FORMAT_ELF) { 1890 warn_report("guest note format is unsupported: %" PRIu16, format); 1891 } else { 1892 s->guest_note = g_malloc(size + 1); /* +1 for adding \0 */ 1893 cpu_physical_memory_read(addr, s->guest_note, size); 1894 1895 get_note_sizes(s, s->guest_note, NULL, &name_size, &desc_size); 1896 s->guest_note_size = ELF_NOTE_SIZE(note_head_size, name_size, 1897 desc_size); 1898 if (name_size > MAX_GUEST_NOTE_SIZE || 1899 desc_size > MAX_GUEST_NOTE_SIZE || 1900 s->guest_note_size > size) { 1901 warn_report("Invalid guest note header"); 1902 g_free(s->guest_note); 1903 s->guest_note = NULL; 1904 } else { 1905 vmcoreinfo_update_phys_base(s); 1906 s->note_size += s->guest_note_size; 1907 } 1908 } 1909 } 1910 1911 /* get memory mapping */ 1912 if (paging) { 1913 qemu_get_guest_memory_mapping(&s->list, &s->guest_phys_blocks, errp); 1914 if (*errp) { 1915 goto cleanup; 1916 } 1917 } else { 1918 qemu_get_guest_simple_memory_mapping(&s->list, &s->guest_phys_blocks); 1919 } 1920 1921 s->nr_cpus = nr_cpus; 1922 1923 get_max_mapnr(s); 1924 1925 uint64_t tmp; 1926 tmp = DIV_ROUND_UP(DIV_ROUND_UP(s->max_mapnr, CHAR_BIT), 1927 s->dump_info.page_size); 1928 s->len_dump_bitmap = tmp * s->dump_info.page_size; 1929 1930 /* init for kdump-compressed format */ 1931 if (has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) { 1932 switch (format) { 1933 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB: 1934 s->flag_compress = DUMP_DH_COMPRESSED_ZLIB; 1935 break; 1936 1937 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO: 1938 #ifdef CONFIG_LZO 1939 if (lzo_init() != LZO_E_OK) { 1940 error_setg(errp, "failed to initialize the LZO library"); 1941 goto cleanup; 1942 } 1943 #endif 1944 s->flag_compress = DUMP_DH_COMPRESSED_LZO; 1945 break; 1946 1947 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY: 1948 s->flag_compress = DUMP_DH_COMPRESSED_SNAPPY; 1949 break; 1950 1951 default: 1952 s->flag_compress = 0; 1953 } 1954 1955 return; 1956 } 1957 1958 if (dump_has_filter(s)) { 1959 memory_mapping_filter(&s->list, s->filter_area_begin, s->filter_area_length); 1960 } 1961 1962 /* 1963 * The first section header is always a special one in which most 1964 * fields are 0. The section header string table is also always 1965 * set. 1966 */ 1967 s->shdr_num = 2; 1968 1969 /* 1970 * Adds the number of architecture sections to shdr_num and sets 1971 * elf_section_data_size so we know the offsets and sizes of all 1972 * parts. 1973 */ 1974 if (s->dump_info.arch_sections_add_fn) { 1975 s->dump_info.arch_sections_add_fn(s); 1976 } 1977 1978 /* 1979 * calculate shdr_num so we know the offsets and sizes of all 1980 * parts. 1981 * Calculate phdr_num 1982 * 1983 * The absolute maximum amount of phdrs is UINT32_MAX - 1 as 1984 * sh_info is 32 bit. There's special handling once we go over 1985 * UINT16_MAX - 1 but that is handled in the ehdr and section 1986 * code. 1987 */ 1988 s->phdr_num = 1; /* Reserve PT_NOTE */ 1989 if (s->list.num <= UINT32_MAX - 1) { 1990 s->phdr_num += s->list.num; 1991 } else { 1992 s->phdr_num = UINT32_MAX; 1993 } 1994 1995 /* 1996 * Now that the number of section and program headers is known we 1997 * can calculate the offsets of the headers and data. 1998 */ 1999 if (dump_is_64bit(s)) { 2000 s->shdr_offset = sizeof(Elf64_Ehdr); 2001 s->phdr_offset = s->shdr_offset + sizeof(Elf64_Shdr) * s->shdr_num; 2002 s->note_offset = s->phdr_offset + sizeof(Elf64_Phdr) * s->phdr_num; 2003 } else { 2004 s->shdr_offset = sizeof(Elf32_Ehdr); 2005 s->phdr_offset = s->shdr_offset + sizeof(Elf32_Shdr) * s->shdr_num; 2006 s->note_offset = s->phdr_offset + sizeof(Elf32_Phdr) * s->phdr_num; 2007 } 2008 s->memory_offset = s->note_offset + s->note_size; 2009 s->section_offset = s->memory_offset + s->total_size; 2010 2011 return; 2012 2013 cleanup: 2014 dump_cleanup(s); 2015 } 2016 2017 /* this operation might be time consuming. */ 2018 static void dump_process(DumpState *s, Error **errp) 2019 { 2020 ERRP_GUARD(); 2021 DumpQueryResult *result = NULL; 2022 2023 if (s->has_format && s->format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) { 2024 #ifdef TARGET_X86_64 2025 create_win_dump(s, errp); 2026 #endif 2027 } else if (s->has_format && s->format != DUMP_GUEST_MEMORY_FORMAT_ELF) { 2028 create_kdump_vmcore(s, errp); 2029 } else { 2030 create_vmcore(s, errp); 2031 } 2032 2033 /* make sure status is written after written_size updates */ 2034 smp_wmb(); 2035 qatomic_set(&s->status, 2036 (*errp ? DUMP_STATUS_FAILED : DUMP_STATUS_COMPLETED)); 2037 2038 /* send DUMP_COMPLETED message (unconditionally) */ 2039 result = qmp_query_dump(NULL); 2040 /* should never fail */ 2041 assert(result); 2042 qapi_event_send_dump_completed(result, 2043 *errp ? error_get_pretty(*errp) : NULL); 2044 qapi_free_DumpQueryResult(result); 2045 2046 dump_cleanup(s); 2047 } 2048 2049 static void *dump_thread(void *data) 2050 { 2051 DumpState *s = (DumpState *)data; 2052 dump_process(s, NULL); 2053 return NULL; 2054 } 2055 2056 DumpQueryResult *qmp_query_dump(Error **errp) 2057 { 2058 DumpQueryResult *result = g_new(DumpQueryResult, 1); 2059 DumpState *state = &dump_state_global; 2060 result->status = qatomic_read(&state->status); 2061 /* make sure we are reading status and written_size in order */ 2062 smp_rmb(); 2063 result->completed = state->written_size; 2064 result->total = state->total_size; 2065 return result; 2066 } 2067 2068 void qmp_dump_guest_memory(bool paging, const char *file, 2069 bool has_detach, bool detach, 2070 bool has_begin, int64_t begin, bool has_length, 2071 int64_t length, bool has_format, 2072 DumpGuestMemoryFormat format, Error **errp) 2073 { 2074 ERRP_GUARD(); 2075 const char *p; 2076 int fd = -1; 2077 DumpState *s; 2078 bool detach_p = false; 2079 2080 if (runstate_check(RUN_STATE_INMIGRATE)) { 2081 error_setg(errp, "Dump not allowed during incoming migration."); 2082 return; 2083 } 2084 2085 /* if there is a dump in background, we should wait until the dump 2086 * finished */ 2087 if (qemu_system_dump_in_progress()) { 2088 error_setg(errp, "There is a dump in process, please wait."); 2089 return; 2090 } 2091 2092 /* 2093 * kdump-compressed format need the whole memory dumped, so paging or 2094 * filter is not supported here. 2095 */ 2096 if ((has_format && format != DUMP_GUEST_MEMORY_FORMAT_ELF) && 2097 (paging || has_begin || has_length)) { 2098 error_setg(errp, "kdump-compressed format doesn't support paging or " 2099 "filter"); 2100 return; 2101 } 2102 if (has_begin && !has_length) { 2103 error_setg(errp, QERR_MISSING_PARAMETER, "length"); 2104 return; 2105 } 2106 if (!has_begin && has_length) { 2107 error_setg(errp, QERR_MISSING_PARAMETER, "begin"); 2108 return; 2109 } 2110 if (has_detach) { 2111 detach_p = detach; 2112 } 2113 2114 /* check whether lzo/snappy is supported */ 2115 #ifndef CONFIG_LZO 2116 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO) { 2117 error_setg(errp, "kdump-lzo is not available now"); 2118 return; 2119 } 2120 #endif 2121 2122 #ifndef CONFIG_SNAPPY 2123 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY) { 2124 error_setg(errp, "kdump-snappy is not available now"); 2125 return; 2126 } 2127 #endif 2128 2129 #ifndef TARGET_X86_64 2130 if (has_format && format == DUMP_GUEST_MEMORY_FORMAT_WIN_DMP) { 2131 error_setg(errp, "Windows dump is only available for x86-64"); 2132 return; 2133 } 2134 #endif 2135 2136 #if !defined(WIN32) 2137 if (strstart(file, "fd:", &p)) { 2138 fd = monitor_get_fd(monitor_cur(), p, errp); 2139 if (fd == -1) { 2140 return; 2141 } 2142 } 2143 #endif 2144 2145 if (strstart(file, "file:", &p)) { 2146 fd = qemu_open_old(p, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR); 2147 if (fd < 0) { 2148 error_setg_file_open(errp, errno, p); 2149 return; 2150 } 2151 } 2152 2153 if (fd == -1) { 2154 error_setg(errp, QERR_INVALID_PARAMETER, "protocol"); 2155 return; 2156 } 2157 2158 if (!dump_migration_blocker) { 2159 error_setg(&dump_migration_blocker, 2160 "Live migration disabled: dump-guest-memory in progress"); 2161 } 2162 2163 /* 2164 * Allows even for -only-migratable, but forbid migration during the 2165 * process of dump guest memory. 2166 */ 2167 if (migrate_add_blocker_internal(dump_migration_blocker, errp)) { 2168 /* Remember to release the fd before passing it over to dump state */ 2169 close(fd); 2170 return; 2171 } 2172 2173 s = &dump_state_global; 2174 dump_state_prepare(s); 2175 2176 dump_init(s, fd, has_format, format, paging, has_begin, 2177 begin, length, errp); 2178 if (*errp) { 2179 qatomic_set(&s->status, DUMP_STATUS_FAILED); 2180 return; 2181 } 2182 2183 if (detach_p) { 2184 /* detached dump */ 2185 s->detached = true; 2186 qemu_thread_create(&s->dump_thread, "dump_thread", dump_thread, 2187 s, QEMU_THREAD_DETACHED); 2188 } else { 2189 /* sync dump */ 2190 dump_process(s, errp); 2191 } 2192 } 2193 2194 DumpGuestMemoryCapability *qmp_query_dump_guest_memory_capability(Error **errp) 2195 { 2196 DumpGuestMemoryCapability *cap = 2197 g_new0(DumpGuestMemoryCapability, 1); 2198 DumpGuestMemoryFormatList **tail = &cap->formats; 2199 2200 /* elf is always available */ 2201 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_ELF); 2202 2203 /* kdump-zlib is always available */ 2204 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB); 2205 2206 /* add new item if kdump-lzo is available */ 2207 #ifdef CONFIG_LZO 2208 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO); 2209 #endif 2210 2211 /* add new item if kdump-snappy is available */ 2212 #ifdef CONFIG_SNAPPY 2213 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY); 2214 #endif 2215 2216 /* Windows dump is available only if target is x86_64 */ 2217 #ifdef TARGET_X86_64 2218 QAPI_LIST_APPEND(tail, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP); 2219 #endif 2220 2221 return cap; 2222 } 2223