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/verify_pefile.h> 23#include <keys/system_keyring.h> 24 25#include <asm/bootparam.h> 26#include <asm/setup.h> 27#include <asm/crash.h> 28#include <asm/efi.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 */ 49struct 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 57static 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 69static 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; 76 uint32_t cmdline_low_32, cmdline_ext_32; 77 78 memcpy(cmdline_ptr, cmdline, cmdline_len); 79 if (image->type == KEXEC_TYPE_CRASH) { 80 len = sprintf(cmdline_ptr + cmdline_len - 1, 81 " elfcorehdr=0x%lx", image->arch.elf_load_addr); 82 cmdline_len += len; 83 } 84 cmdline_ptr[cmdline_len - 1] = '\0'; 85 86 pr_debug("Final command line is: %s\n", cmdline_ptr); 87 cmdline_ptr_phys = bootparams_load_addr + cmdline_offset; 88 cmdline_low_32 = cmdline_ptr_phys & 0xffffffffUL; 89 cmdline_ext_32 = cmdline_ptr_phys >> 32; 90 91 params->hdr.cmd_line_ptr = cmdline_low_32; 92 if (cmdline_ext_32) 93 params->ext_cmd_line_ptr = cmdline_ext_32; 94 95 return 0; 96} 97 98static int setup_e820_entries(struct boot_params *params) 99{ 100 unsigned int nr_e820_entries; 101 102 nr_e820_entries = e820_saved.nr_map; 103 104 /* TODO: Pass entries more than E820MAX in bootparams setup data */ 105 if (nr_e820_entries > E820MAX) 106 nr_e820_entries = E820MAX; 107 108 params->e820_entries = nr_e820_entries; 109 memcpy(¶ms->e820_map, &e820_saved.map, 110 nr_e820_entries * sizeof(struct e820entry)); 111 112 return 0; 113} 114 115#ifdef CONFIG_EFI 116static 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 137static int 138prepare_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 162static int 163setup_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 (!current_ei->efi_memmap_size) 171 return 0; 172 173 /* 174 * If 1:1 mapping is not enabled, second kernel can not setup EFI 175 * and use EFI run time services. User space will have to pass 176 * acpi_rsdp=<addr> on kernel command line to make second kernel boot 177 * without efi. 178 */ 179 if (efi_enabled(EFI_OLD_MEMMAP)) 180 return 0; 181 182 ei->efi_loader_signature = current_ei->efi_loader_signature; 183 ei->efi_systab = current_ei->efi_systab; 184 ei->efi_systab_hi = current_ei->efi_systab_hi; 185 186 ei->efi_memdesc_version = current_ei->efi_memdesc_version; 187 ei->efi_memdesc_size = efi_get_runtime_map_desc_size(); 188 189 setup_efi_info_memmap(params, params_load_addr, efi_map_offset, 190 efi_map_sz); 191 prepare_add_efi_setup_data(params, params_load_addr, 192 efi_setup_data_offset); 193 return 0; 194} 195#endif /* CONFIG_EFI */ 196 197static int 198setup_boot_parameters(struct kimage *image, struct boot_params *params, 199 unsigned long params_load_addr, 200 unsigned int efi_map_offset, unsigned int efi_map_sz, 201 unsigned int efi_setup_data_offset) 202{ 203 unsigned int nr_e820_entries; 204 unsigned long long mem_k, start, end; 205 int i, ret = 0; 206 207 /* Get subarch from existing bootparams */ 208 params->hdr.hardware_subarch = boot_params.hdr.hardware_subarch; 209 210 /* Copying screen_info will do? */ 211 memcpy(¶ms->screen_info, &boot_params.screen_info, 212 sizeof(struct screen_info)); 213 214 /* Fill in memsize later */ 215 params->screen_info.ext_mem_k = 0; 216 params->alt_mem_k = 0; 217 218 /* Default APM info */ 219 memset(¶ms->apm_bios_info, 0, sizeof(params->apm_bios_info)); 220 221 /* Default drive info */ 222 memset(¶ms->hd0_info, 0, sizeof(params->hd0_info)); 223 memset(¶ms->hd1_info, 0, sizeof(params->hd1_info)); 224 225 /* Default sysdesc table */ 226 params->sys_desc_table.length = 0; 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_map[i].type != E820_RAM) 239 continue; 240 start = params->e820_map[i].addr; 241 end = params->e820_map[i].addr + params->e820_map[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 271static int bzImage64_probe(const char *buf, unsigned long len) 272{ 273 int ret = -ENOEXEC; 274 struct setup_header *header; 275 276 /* kernel should be atleast 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 329static 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, params_misc_sz; 338 struct boot_params *params; 339 unsigned long bootparam_load_addr, kernel_load_addr, initrd_load_addr; 340 unsigned long purgatory_load_addr; 341 unsigned long kernel_bufsz, kernel_memsz, kernel_align; 342 char *kernel_buf; 343 struct bzimage64_data *ldata; 344 struct kexec_entry64_regs regs64; 345 void *stack; 346 unsigned int setup_hdr_offset = offsetof(struct boot_params, hdr); 347 unsigned int efi_map_offset, efi_map_sz, efi_setup_data_offset; 348 349 header = (struct setup_header *)(kernel + setup_hdr_offset); 350 setup_sects = header->setup_sects; 351 if (setup_sects == 0) 352 setup_sects = 4; 353 354 kern16_size = (setup_sects + 1) * 512; 355 if (kernel_len < kern16_size) { 356 pr_err("bzImage truncated\n"); 357 return ERR_PTR(-ENOEXEC); 358 } 359 360 if (cmdline_len > header->cmdline_size) { 361 pr_err("Kernel command line too long\n"); 362 return ERR_PTR(-EINVAL); 363 } 364 365 /* 366 * In case of crash dump, we will append elfcorehdr=<addr> to 367 * command line. Make sure it does not overflow 368 */ 369 if (cmdline_len + MAX_ELFCOREHDR_STR_LEN > header->cmdline_size) { 370 pr_debug("Appending elfcorehdr=<addr> to command line exceeds maximum allowed length\n"); 371 return ERR_PTR(-EINVAL); 372 } 373 374 /* Allocate and load backup region */ 375 if (image->type == KEXEC_TYPE_CRASH) { 376 ret = crash_load_segments(image); 377 if (ret) 378 return ERR_PTR(ret); 379 } 380 381 /* 382 * Load purgatory. For 64bit entry point, purgatory code can be 383 * anywhere. 384 */ 385 ret = kexec_load_purgatory(image, MIN_PURGATORY_ADDR, ULONG_MAX, 1, 386 &purgatory_load_addr); 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", purgatory_load_addr); 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 efi_map_sz = ALIGN(efi_map_sz, 16); 405 params_cmdline_sz = sizeof(struct boot_params) + cmdline_len + 406 MAX_ELFCOREHDR_STR_LEN; 407 params_cmdline_sz = ALIGN(params_cmdline_sz, 16); 408 params_misc_sz = params_cmdline_sz + efi_map_sz + 409 sizeof(struct setup_data) + 410 sizeof(struct efi_setup_data); 411 412 params = kzalloc(params_misc_sz, GFP_KERNEL); 413 if (!params) 414 return ERR_PTR(-ENOMEM); 415 efi_map_offset = params_cmdline_sz; 416 efi_setup_data_offset = efi_map_offset + efi_map_sz; 417 418 /* Copy setup header onto bootparams. Documentation/x86/boot.txt */ 419 setup_header_size = 0x0202 + kernel[0x0201] - setup_hdr_offset; 420 421 /* Is there a limit on setup header size? */ 422 memcpy(¶ms->hdr, (kernel + setup_hdr_offset), setup_header_size); 423 424 ret = kexec_add_buffer(image, (char *)params, params_misc_sz, 425 params_misc_sz, 16, MIN_BOOTPARAM_ADDR, 426 ULONG_MAX, 1, &bootparam_load_addr); 427 if (ret) 428 goto out_free_params; 429 pr_debug("Loaded boot_param, command line and misc at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 430 bootparam_load_addr, params_misc_sz, params_misc_sz); 431 432 /* Load kernel */ 433 kernel_buf = kernel + kern16_size; 434 kernel_bufsz = kernel_len - kern16_size; 435 kernel_memsz = PAGE_ALIGN(header->init_size); 436 kernel_align = header->kernel_alignment; 437 438 ret = kexec_add_buffer(image, kernel_buf, 439 kernel_bufsz, kernel_memsz, kernel_align, 440 MIN_KERNEL_LOAD_ADDR, ULONG_MAX, 1, 441 &kernel_load_addr); 442 if (ret) 443 goto out_free_params; 444 445 pr_debug("Loaded 64bit kernel at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 446 kernel_load_addr, kernel_memsz, kernel_memsz); 447 448 /* Load initrd high */ 449 if (initrd) { 450 ret = kexec_add_buffer(image, initrd, initrd_len, initrd_len, 451 PAGE_SIZE, MIN_INITRD_LOAD_ADDR, 452 ULONG_MAX, 1, &initrd_load_addr); 453 if (ret) 454 goto out_free_params; 455 456 pr_debug("Loaded initrd at 0x%lx bufsz=0x%lx memsz=0x%lx\n", 457 initrd_load_addr, initrd_len, initrd_len); 458 459 setup_initrd(params, initrd_load_addr, initrd_len); 460 } 461 462 setup_cmdline(image, params, bootparam_load_addr, 463 sizeof(struct boot_params), cmdline, cmdline_len); 464 465 /* bootloader info. Do we need a separate ID for kexec kernel loader? */ 466 params->hdr.type_of_loader = 0x0D << 4; 467 params->hdr.loadflags = 0; 468 469 /* Setup purgatory regs for entry */ 470 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", ®s64, 471 sizeof(regs64), 1); 472 if (ret) 473 goto out_free_params; 474 475 regs64.rbx = 0; /* Bootstrap Processor */ 476 regs64.rsi = bootparam_load_addr; 477 regs64.rip = kernel_load_addr + 0x200; 478 stack = kexec_purgatory_get_symbol_addr(image, "stack_end"); 479 if (IS_ERR(stack)) { 480 pr_err("Could not find address of symbol stack_end\n"); 481 ret = -EINVAL; 482 goto out_free_params; 483 } 484 485 regs64.rsp = (unsigned long)stack; 486 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", ®s64, 487 sizeof(regs64), 0); 488 if (ret) 489 goto out_free_params; 490 491 ret = setup_boot_parameters(image, params, bootparam_load_addr, 492 efi_map_offset, efi_map_sz, 493 efi_setup_data_offset); 494 if (ret) 495 goto out_free_params; 496 497 /* Allocate loader specific data */ 498 ldata = kzalloc(sizeof(struct bzimage64_data), GFP_KERNEL); 499 if (!ldata) { 500 ret = -ENOMEM; 501 goto out_free_params; 502 } 503 504 /* 505 * Store pointer to params so that it could be freed after loading 506 * params segment has been loaded and contents have been copied 507 * somewhere else. 508 */ 509 ldata->bootparams_buf = params; 510 return ldata; 511 512out_free_params: 513 kfree(params); 514 return ERR_PTR(ret); 515} 516 517/* This cleanup function is called after various segments have been loaded */ 518static int bzImage64_cleanup(void *loader_data) 519{ 520 struct bzimage64_data *ldata = loader_data; 521 522 if (!ldata) 523 return 0; 524 525 kfree(ldata->bootparams_buf); 526 ldata->bootparams_buf = NULL; 527 528 return 0; 529} 530 531#ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG 532static int bzImage64_verify_sig(const char *kernel, unsigned long kernel_len) 533{ 534 bool trusted; 535 int ret; 536 537 ret = verify_pefile_signature(kernel, kernel_len, 538 system_trusted_keyring, &trusted); 539 if (ret < 0) 540 return ret; 541 if (!trusted) 542 return -EKEYREJECTED; 543 return 0; 544} 545#endif 546 547struct kexec_file_ops kexec_bzImage64_ops = { 548 .probe = bzImage64_probe, 549 .load = bzImage64_load, 550 .cleanup = bzImage64_cleanup, 551#ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG 552 .verify_sig = bzImage64_verify_sig, 553#endif 554}; 555