1 /* 2 * Kexec bzImage loader 3 * 4 * Copyright (C) 2014 Red Hat Inc. 5 * Authors: 6 * Vivek Goyal <vgoyal@redhat.com> 7 * 8 * This source code is licensed under the GNU General Public License, 9 * Version 2. See the file COPYING for more details. 10 */ 11 12 #define pr_fmt(fmt) "kexec-bzImage64: " fmt 13 14 #include <linux/string.h> 15 #include <linux/printk.h> 16 #include <linux/errno.h> 17 #include <linux/slab.h> 18 #include <linux/kexec.h> 19 #include <linux/kernel.h> 20 #include <linux/mm.h> 21 #include <linux/efi.h> 22 #include <linux/verification.h> 23 24 #include <asm/bootparam.h> 25 #include <asm/setup.h> 26 #include <asm/crash.h> 27 #include <asm/efi.h> 28 #include <asm/e820/api.h> 29 #include <asm/kexec-bzimage64.h> 30 31 #define MAX_ELFCOREHDR_STR_LEN 30 /* elfcorehdr=0x<64bit-value> */ 32 33 /* 34 * Defines lowest physical address for various segments. Not sure where 35 * exactly these limits came from. Current bzimage64 loader in kexec-tools 36 * uses these so I am retaining it. It can be changed over time as we gain 37 * more insight. 38 */ 39 #define MIN_PURGATORY_ADDR 0x3000 40 #define MIN_BOOTPARAM_ADDR 0x3000 41 #define MIN_KERNEL_LOAD_ADDR 0x100000 42 #define MIN_INITRD_LOAD_ADDR 0x1000000 43 44 /* 45 * This is a place holder for all boot loader specific data structure which 46 * gets allocated in one call but gets freed much later during cleanup 47 * time. Right now there is only one field but it can grow as need be. 48 */ 49 struct bzimage64_data { 50 /* 51 * Temporary buffer to hold bootparams buffer. This should be 52 * freed once the bootparam segment has been loaded. 53 */ 54 void *bootparams_buf; 55 }; 56 57 static int setup_initrd(struct boot_params *params, 58 unsigned long initrd_load_addr, unsigned long initrd_len) 59 { 60 params->hdr.ramdisk_image = initrd_load_addr & 0xffffffffUL; 61 params->hdr.ramdisk_size = initrd_len & 0xffffffffUL; 62 63 params->ext_ramdisk_image = initrd_load_addr >> 32; 64 params->ext_ramdisk_size = initrd_len >> 32; 65 66 return 0; 67 } 68 69 static int setup_cmdline(struct kimage *image, struct boot_params *params, 70 unsigned long bootparams_load_addr, 71 unsigned long cmdline_offset, char *cmdline, 72 unsigned long cmdline_len) 73 { 74 char *cmdline_ptr = ((char *)params) + cmdline_offset; 75 unsigned long cmdline_ptr_phys, len = 0; 76 uint32_t cmdline_low_32, cmdline_ext_32; 77 78 if (image->type == KEXEC_TYPE_CRASH) { 79 len = sprintf(cmdline_ptr, 80 "elfcorehdr=0x%lx ", image->arch.elf_load_addr); 81 } 82 memcpy(cmdline_ptr + len, cmdline, cmdline_len); 83 cmdline_len += len; 84 85 cmdline_ptr[cmdline_len - 1] = '\0'; 86 87 pr_debug("Final command line is: %s\n", cmdline_ptr); 88 cmdline_ptr_phys = bootparams_load_addr + cmdline_offset; 89 cmdline_low_32 = cmdline_ptr_phys & 0xffffffffUL; 90 cmdline_ext_32 = cmdline_ptr_phys >> 32; 91 92 params->hdr.cmd_line_ptr = cmdline_low_32; 93 if (cmdline_ext_32) 94 params->ext_cmd_line_ptr = cmdline_ext_32; 95 96 return 0; 97 } 98 99 static int setup_e820_entries(struct boot_params *params) 100 { 101 unsigned int nr_e820_entries; 102 103 nr_e820_entries = e820_table_kexec->nr_entries; 104 105 /* TODO: Pass entries more than E820_MAX_ENTRIES_ZEROPAGE in bootparams setup data */ 106 if (nr_e820_entries > E820_MAX_ENTRIES_ZEROPAGE) 107 nr_e820_entries = E820_MAX_ENTRIES_ZEROPAGE; 108 109 params->e820_entries = nr_e820_entries; 110 memcpy(¶ms->e820_table, &e820_table_kexec->entries, nr_e820_entries*sizeof(struct e820_entry)); 111 112 return 0; 113 } 114 115 #ifdef CONFIG_EFI 116 static int setup_efi_info_memmap(struct boot_params *params, 117 unsigned long params_load_addr, 118 unsigned int efi_map_offset, 119 unsigned int efi_map_sz) 120 { 121 void *efi_map = (void *)params + efi_map_offset; 122 unsigned long efi_map_phys_addr = params_load_addr + efi_map_offset; 123 struct efi_info *ei = ¶ms->efi_info; 124 125 if (!efi_map_sz) 126 return 0; 127 128 efi_runtime_map_copy(efi_map, efi_map_sz); 129 130 ei->efi_memmap = efi_map_phys_addr & 0xffffffff; 131 ei->efi_memmap_hi = efi_map_phys_addr >> 32; 132 ei->efi_memmap_size = efi_map_sz; 133 134 return 0; 135 } 136 137 static int 138 prepare_add_efi_setup_data(struct boot_params *params, 139 unsigned long params_load_addr, 140 unsigned int efi_setup_data_offset) 141 { 142 unsigned long setup_data_phys; 143 struct setup_data *sd = (void *)params + efi_setup_data_offset; 144 struct efi_setup_data *esd = (void *)sd + sizeof(struct setup_data); 145 146 esd->fw_vendor = efi.fw_vendor; 147 esd->runtime = efi.runtime; 148 esd->tables = efi.config_table; 149 esd->smbios = efi.smbios; 150 151 sd->type = SETUP_EFI; 152 sd->len = sizeof(struct efi_setup_data); 153 154 /* Add setup data */ 155 setup_data_phys = params_load_addr + efi_setup_data_offset; 156 sd->next = params->hdr.setup_data; 157 params->hdr.setup_data = setup_data_phys; 158 159 return 0; 160 } 161 162 static int 163 setup_efi_state(struct boot_params *params, unsigned long params_load_addr, 164 unsigned int efi_map_offset, unsigned int efi_map_sz, 165 unsigned int efi_setup_data_offset) 166 { 167 struct efi_info *current_ei = &boot_params.efi_info; 168 struct efi_info *ei = ¶ms->efi_info; 169 170 if (!efi_enabled(EFI_RUNTIME_SERVICES)) 171 return 0; 172 173 if (!current_ei->efi_memmap_size) 174 return 0; 175 176 /* 177 * If 1:1 mapping is not enabled, second kernel can not setup EFI 178 * and use EFI run time services. User space will have to pass 179 * acpi_rsdp=<addr> on kernel command line to make second kernel boot 180 * without efi. 181 */ 182 if (efi_enabled(EFI_OLD_MEMMAP)) 183 return 0; 184 185 ei->efi_loader_signature = current_ei->efi_loader_signature; 186 ei->efi_systab = current_ei->efi_systab; 187 ei->efi_systab_hi = current_ei->efi_systab_hi; 188 189 ei->efi_memdesc_version = current_ei->efi_memdesc_version; 190 ei->efi_memdesc_size = efi_get_runtime_map_desc_size(); 191 192 setup_efi_info_memmap(params, params_load_addr, efi_map_offset, 193 efi_map_sz); 194 prepare_add_efi_setup_data(params, params_load_addr, 195 efi_setup_data_offset); 196 return 0; 197 } 198 #endif /* CONFIG_EFI */ 199 200 static int 201 setup_boot_parameters(struct kimage *image, struct boot_params *params, 202 unsigned long params_load_addr, 203 unsigned int efi_map_offset, unsigned int efi_map_sz, 204 unsigned int efi_setup_data_offset) 205 { 206 unsigned int nr_e820_entries; 207 unsigned long long mem_k, start, end; 208 int i, ret = 0; 209 210 /* Get subarch from existing bootparams */ 211 params->hdr.hardware_subarch = boot_params.hdr.hardware_subarch; 212 213 /* Copying screen_info will do? */ 214 memcpy(¶ms->screen_info, &boot_params.screen_info, 215 sizeof(struct screen_info)); 216 217 /* Fill in memsize later */ 218 params->screen_info.ext_mem_k = 0; 219 params->alt_mem_k = 0; 220 221 /* Default APM info */ 222 memset(¶ms->apm_bios_info, 0, sizeof(params->apm_bios_info)); 223 224 /* Default drive info */ 225 memset(¶ms->hd0_info, 0, sizeof(params->hd0_info)); 226 memset(¶ms->hd1_info, 0, sizeof(params->hd1_info)); 227 228 if (image->type == KEXEC_TYPE_CRASH) { 229 ret = crash_setup_memmap_entries(image, params); 230 if (ret) 231 return ret; 232 } else 233 setup_e820_entries(params); 234 235 nr_e820_entries = params->e820_entries; 236 237 for (i = 0; i < nr_e820_entries; i++) { 238 if (params->e820_table[i].type != E820_TYPE_RAM) 239 continue; 240 start = params->e820_table[i].addr; 241 end = params->e820_table[i].addr + params->e820_table[i].size - 1; 242 243 if ((start <= 0x100000) && end > 0x100000) { 244 mem_k = (end >> 10) - (0x100000 >> 10); 245 params->screen_info.ext_mem_k = mem_k; 246 params->alt_mem_k = mem_k; 247 if (mem_k > 0xfc00) 248 params->screen_info.ext_mem_k = 0xfc00; /* 64M*/ 249 if (mem_k > 0xffffffff) 250 params->alt_mem_k = 0xffffffff; 251 } 252 } 253 254 #ifdef CONFIG_EFI 255 /* Setup EFI state */ 256 setup_efi_state(params, params_load_addr, efi_map_offset, efi_map_sz, 257 efi_setup_data_offset); 258 #endif 259 260 /* Setup EDD info */ 261 memcpy(params->eddbuf, boot_params.eddbuf, 262 EDDMAXNR * sizeof(struct edd_info)); 263 params->eddbuf_entries = boot_params.eddbuf_entries; 264 265 memcpy(params->edd_mbr_sig_buffer, boot_params.edd_mbr_sig_buffer, 266 EDD_MBR_SIG_MAX * sizeof(unsigned int)); 267 268 return ret; 269 } 270 271 static int bzImage64_probe(const char *buf, unsigned long len) 272 { 273 int ret = -ENOEXEC; 274 struct setup_header *header; 275 276 /* kernel should be at least two sectors long */ 277 if (len < 2 * 512) { 278 pr_err("File is too short to be a bzImage\n"); 279 return ret; 280 } 281 282 header = (struct setup_header *)(buf + offsetof(struct boot_params, hdr)); 283 if (memcmp((char *)&header->header, "HdrS", 4) != 0) { 284 pr_err("Not a bzImage\n"); 285 return ret; 286 } 287 288 if (header->boot_flag != 0xAA55) { 289 pr_err("No x86 boot sector present\n"); 290 return ret; 291 } 292 293 if (header->version < 0x020C) { 294 pr_err("Must be at least protocol version 2.12\n"); 295 return ret; 296 } 297 298 if (!(header->loadflags & LOADED_HIGH)) { 299 pr_err("zImage not a bzImage\n"); 300 return ret; 301 } 302 303 if (!(header->xloadflags & XLF_KERNEL_64)) { 304 pr_err("Not a bzImage64. XLF_KERNEL_64 is not set.\n"); 305 return ret; 306 } 307 308 if (!(header->xloadflags & XLF_CAN_BE_LOADED_ABOVE_4G)) { 309 pr_err("XLF_CAN_BE_LOADED_ABOVE_4G is not set.\n"); 310 return ret; 311 } 312 313 /* 314 * Can't handle 32bit EFI as it does not allow loading kernel 315 * above 4G. This should be handled by 32bit bzImage loader 316 */ 317 if (efi_enabled(EFI_RUNTIME_SERVICES) && !efi_enabled(EFI_64BIT)) { 318 pr_debug("EFI is 32 bit. Can't load kernel above 4G.\n"); 319 return ret; 320 } 321 322 /* I've got a bzImage */ 323 pr_debug("It's a relocatable bzImage64\n"); 324 ret = 0; 325 326 return ret; 327 } 328 329 static void *bzImage64_load(struct kimage *image, char *kernel, 330 unsigned long kernel_len, char *initrd, 331 unsigned long initrd_len, char *cmdline, 332 unsigned long cmdline_len) 333 { 334 335 struct setup_header *header; 336 int setup_sects, kern16_size, ret = 0; 337 unsigned long setup_header_size, params_cmdline_sz; 338 struct boot_params *params; 339 unsigned long bootparam_load_addr, kernel_load_addr, initrd_load_addr; 340 struct bzimage64_data *ldata; 341 struct kexec_entry64_regs regs64; 342 void *stack; 343 unsigned int setup_hdr_offset = offsetof(struct boot_params, hdr); 344 unsigned int efi_map_offset, efi_map_sz, efi_setup_data_offset; 345 struct kexec_buf kbuf = { .image = image, .buf_max = ULONG_MAX, 346 .top_down = true }; 347 struct kexec_buf pbuf = { .image = image, .buf_min = MIN_PURGATORY_ADDR, 348 .buf_max = ULONG_MAX, .top_down = true }; 349 350 header = (struct setup_header *)(kernel + setup_hdr_offset); 351 setup_sects = header->setup_sects; 352 if (setup_sects == 0) 353 setup_sects = 4; 354 355 kern16_size = (setup_sects + 1) * 512; 356 if (kernel_len < kern16_size) { 357 pr_err("bzImage truncated\n"); 358 return ERR_PTR(-ENOEXEC); 359 } 360 361 if (cmdline_len > header->cmdline_size) { 362 pr_err("Kernel command line too long\n"); 363 return ERR_PTR(-EINVAL); 364 } 365 366 /* 367 * In case of crash dump, we will append elfcorehdr=<addr> to 368 * command line. Make sure it does not overflow 369 */ 370 if (cmdline_len + MAX_ELFCOREHDR_STR_LEN > header->cmdline_size) { 371 pr_debug("Appending elfcorehdr=<addr> to command line exceeds maximum allowed length\n"); 372 return ERR_PTR(-EINVAL); 373 } 374 375 /* Allocate and load backup region */ 376 if (image->type == KEXEC_TYPE_CRASH) { 377 ret = crash_load_segments(image); 378 if (ret) 379 return ERR_PTR(ret); 380 } 381 382 /* 383 * Load purgatory. For 64bit entry point, purgatory code can be 384 * anywhere. 385 */ 386 ret = kexec_load_purgatory(image, &pbuf); 387 if (ret) { 388 pr_err("Loading purgatory failed\n"); 389 return ERR_PTR(ret); 390 } 391 392 pr_debug("Loaded purgatory at 0x%lx\n", pbuf.mem); 393 394 395 /* 396 * Load Bootparams and cmdline and space for efi stuff. 397 * 398 * Allocate memory together for multiple data structures so 399 * that they all can go in single area/segment and we don't 400 * have to create separate segment for each. Keeps things 401 * little bit simple 402 */ 403 efi_map_sz = efi_get_runtime_map_size(); 404 params_cmdline_sz = sizeof(struct boot_params) + cmdline_len + 405 MAX_ELFCOREHDR_STR_LEN; 406 params_cmdline_sz = ALIGN(params_cmdline_sz, 16); 407 kbuf.bufsz = params_cmdline_sz + ALIGN(efi_map_sz, 16) + 408 sizeof(struct setup_data) + 409 sizeof(struct efi_setup_data); 410 411 params = kzalloc(kbuf.bufsz, GFP_KERNEL); 412 if (!params) 413 return ERR_PTR(-ENOMEM); 414 efi_map_offset = params_cmdline_sz; 415 efi_setup_data_offset = efi_map_offset + ALIGN(efi_map_sz, 16); 416 417 /* Copy setup header onto bootparams. Documentation/x86/boot.txt */ 418 setup_header_size = 0x0202 + kernel[0x0201] - setup_hdr_offset; 419 420 /* Is there a limit on setup header size? */ 421 memcpy(¶ms->hdr, (kernel + setup_hdr_offset), setup_header_size); 422 423 kbuf.buffer = params; 424 kbuf.memsz = kbuf.bufsz; 425 kbuf.buf_align = 16; 426 kbuf.buf_min = MIN_BOOTPARAM_ADDR; 427 ret = kexec_add_buffer(&kbuf); 428 if (ret) 429 goto out_free_params; 430 bootparam_load_addr = kbuf.mem; 431 pr_debug("Loaded boot_param, command line and misc at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 432 bootparam_load_addr, kbuf.bufsz, kbuf.bufsz); 433 434 /* Load kernel */ 435 kbuf.buffer = kernel + kern16_size; 436 kbuf.bufsz = kernel_len - kern16_size; 437 kbuf.memsz = PAGE_ALIGN(header->init_size); 438 kbuf.buf_align = header->kernel_alignment; 439 kbuf.buf_min = MIN_KERNEL_LOAD_ADDR; 440 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN; 441 ret = kexec_add_buffer(&kbuf); 442 if (ret) 443 goto out_free_params; 444 kernel_load_addr = kbuf.mem; 445 446 pr_debug("Loaded 64bit kernel at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 447 kernel_load_addr, kbuf.bufsz, kbuf.memsz); 448 449 /* Load initrd high */ 450 if (initrd) { 451 kbuf.buffer = initrd; 452 kbuf.bufsz = kbuf.memsz = initrd_len; 453 kbuf.buf_align = PAGE_SIZE; 454 kbuf.buf_min = MIN_INITRD_LOAD_ADDR; 455 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN; 456 ret = kexec_add_buffer(&kbuf); 457 if (ret) 458 goto out_free_params; 459 initrd_load_addr = kbuf.mem; 460 461 pr_debug("Loaded initrd at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 462 initrd_load_addr, initrd_len, initrd_len); 463 464 setup_initrd(params, initrd_load_addr, initrd_len); 465 } 466 467 setup_cmdline(image, params, bootparam_load_addr, 468 sizeof(struct boot_params), cmdline, cmdline_len); 469 470 /* bootloader info. Do we need a separate ID for kexec kernel loader? */ 471 params->hdr.type_of_loader = 0x0D << 4; 472 params->hdr.loadflags = 0; 473 474 /* Setup purgatory regs for entry */ 475 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", ®s64, 476 sizeof(regs64), 1); 477 if (ret) 478 goto out_free_params; 479 480 regs64.rbx = 0; /* Bootstrap Processor */ 481 regs64.rsi = bootparam_load_addr; 482 regs64.rip = kernel_load_addr + 0x200; 483 stack = kexec_purgatory_get_symbol_addr(image, "stack_end"); 484 if (IS_ERR(stack)) { 485 pr_err("Could not find address of symbol stack_end\n"); 486 ret = -EINVAL; 487 goto out_free_params; 488 } 489 490 regs64.rsp = (unsigned long)stack; 491 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", ®s64, 492 sizeof(regs64), 0); 493 if (ret) 494 goto out_free_params; 495 496 ret = setup_boot_parameters(image, params, bootparam_load_addr, 497 efi_map_offset, efi_map_sz, 498 efi_setup_data_offset); 499 if (ret) 500 goto out_free_params; 501 502 /* Allocate loader specific data */ 503 ldata = kzalloc(sizeof(struct bzimage64_data), GFP_KERNEL); 504 if (!ldata) { 505 ret = -ENOMEM; 506 goto out_free_params; 507 } 508 509 /* 510 * Store pointer to params so that it could be freed after loading 511 * params segment has been loaded and contents have been copied 512 * somewhere else. 513 */ 514 ldata->bootparams_buf = params; 515 return ldata; 516 517 out_free_params: 518 kfree(params); 519 return ERR_PTR(ret); 520 } 521 522 /* This cleanup function is called after various segments have been loaded */ 523 static int bzImage64_cleanup(void *loader_data) 524 { 525 struct bzimage64_data *ldata = loader_data; 526 527 if (!ldata) 528 return 0; 529 530 kfree(ldata->bootparams_buf); 531 ldata->bootparams_buf = NULL; 532 533 return 0; 534 } 535 536 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG 537 static int bzImage64_verify_sig(const char *kernel, unsigned long kernel_len) 538 { 539 return verify_pefile_signature(kernel, kernel_len, 540 VERIFY_USE_SECONDARY_KEYRING, 541 VERIFYING_KEXEC_PE_SIGNATURE); 542 } 543 #endif 544 545 const struct kexec_file_ops kexec_bzImage64_ops = { 546 .probe = bzImage64_probe, 547 .load = bzImage64_load, 548 .cleanup = bzImage64_cleanup, 549 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG 550 .verify_sig = bzImage64_verify_sig, 551 #endif 552 }; 553