1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kexec bzImage loader
4  *
5  * Copyright (C) 2014 Red Hat Inc.
6  * Authors:
7  *      Vivek Goyal <vgoyal@redhat.com>
8  */
9 
10 #define pr_fmt(fmt)	"kexec-bzImage64: " fmt
11 
12 #include <linux/string.h>
13 #include <linux/printk.h>
14 #include <linux/errno.h>
15 #include <linux/slab.h>
16 #include <linux/kexec.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/efi.h>
20 #include <linux/verification.h>
21 
22 #include <asm/bootparam.h>
23 #include <asm/setup.h>
24 #include <asm/crash.h>
25 #include <asm/efi.h>
26 #include <asm/e820/api.h>
27 #include <asm/kexec-bzimage64.h>
28 
29 #define MAX_ELFCOREHDR_STR_LEN	30	/* elfcorehdr=0x<64bit-value> */
30 
31 /*
32  * Defines lowest physical address for various segments. Not sure where
33  * exactly these limits came from. Current bzimage64 loader in kexec-tools
34  * uses these so I am retaining it. It can be changed over time as we gain
35  * more insight.
36  */
37 #define MIN_PURGATORY_ADDR	0x3000
38 #define MIN_BOOTPARAM_ADDR	0x3000
39 #define MIN_KERNEL_LOAD_ADDR	0x100000
40 #define MIN_INITRD_LOAD_ADDR	0x1000000
41 
42 /*
43  * This is a place holder for all boot loader specific data structure which
44  * gets allocated in one call but gets freed much later during cleanup
45  * time. Right now there is only one field but it can grow as need be.
46  */
47 struct bzimage64_data {
48 	/*
49 	 * Temporary buffer to hold bootparams buffer. This should be
50 	 * freed once the bootparam segment has been loaded.
51 	 */
52 	void *bootparams_buf;
53 };
54 
55 static int setup_initrd(struct boot_params *params,
56 		unsigned long initrd_load_addr, unsigned long initrd_len)
57 {
58 	params->hdr.ramdisk_image = initrd_load_addr & 0xffffffffUL;
59 	params->hdr.ramdisk_size = initrd_len & 0xffffffffUL;
60 
61 	params->ext_ramdisk_image = initrd_load_addr >> 32;
62 	params->ext_ramdisk_size = initrd_len >> 32;
63 
64 	return 0;
65 }
66 
67 static int setup_cmdline(struct kimage *image, struct boot_params *params,
68 			 unsigned long bootparams_load_addr,
69 			 unsigned long cmdline_offset, char *cmdline,
70 			 unsigned long cmdline_len)
71 {
72 	char *cmdline_ptr = ((char *)params) + cmdline_offset;
73 	unsigned long cmdline_ptr_phys, len = 0;
74 	uint32_t cmdline_low_32, cmdline_ext_32;
75 
76 	if (image->type == KEXEC_TYPE_CRASH) {
77 		len = sprintf(cmdline_ptr,
78 			"elfcorehdr=0x%lx ", image->arch.elf_load_addr);
79 	}
80 	memcpy(cmdline_ptr + len, cmdline, cmdline_len);
81 	cmdline_len += len;
82 
83 	cmdline_ptr[cmdline_len - 1] = '\0';
84 
85 	pr_debug("Final command line is: %s\n", cmdline_ptr);
86 	cmdline_ptr_phys = bootparams_load_addr + cmdline_offset;
87 	cmdline_low_32 = cmdline_ptr_phys & 0xffffffffUL;
88 	cmdline_ext_32 = cmdline_ptr_phys >> 32;
89 
90 	params->hdr.cmd_line_ptr = cmdline_low_32;
91 	if (cmdline_ext_32)
92 		params->ext_cmd_line_ptr = cmdline_ext_32;
93 
94 	return 0;
95 }
96 
97 static int setup_e820_entries(struct boot_params *params)
98 {
99 	unsigned int nr_e820_entries;
100 
101 	nr_e820_entries = e820_table_kexec->nr_entries;
102 
103 	/* TODO: Pass entries more than E820_MAX_ENTRIES_ZEROPAGE in bootparams setup data */
104 	if (nr_e820_entries > E820_MAX_ENTRIES_ZEROPAGE)
105 		nr_e820_entries = E820_MAX_ENTRIES_ZEROPAGE;
106 
107 	params->e820_entries = nr_e820_entries;
108 	memcpy(&params->e820_table, &e820_table_kexec->entries, nr_e820_entries*sizeof(struct e820_entry));
109 
110 	return 0;
111 }
112 
113 #ifdef CONFIG_EFI
114 static int setup_efi_info_memmap(struct boot_params *params,
115 				  unsigned long params_load_addr,
116 				  unsigned int efi_map_offset,
117 				  unsigned int efi_map_sz)
118 {
119 	void *efi_map = (void *)params + efi_map_offset;
120 	unsigned long efi_map_phys_addr = params_load_addr + efi_map_offset;
121 	struct efi_info *ei = &params->efi_info;
122 
123 	if (!efi_map_sz)
124 		return 0;
125 
126 	efi_runtime_map_copy(efi_map, efi_map_sz);
127 
128 	ei->efi_memmap = efi_map_phys_addr & 0xffffffff;
129 	ei->efi_memmap_hi = efi_map_phys_addr >> 32;
130 	ei->efi_memmap_size = efi_map_sz;
131 
132 	return 0;
133 }
134 
135 static int
136 prepare_add_efi_setup_data(struct boot_params *params,
137 		       unsigned long params_load_addr,
138 		       unsigned int efi_setup_data_offset)
139 {
140 	unsigned long setup_data_phys;
141 	struct setup_data *sd = (void *)params + efi_setup_data_offset;
142 	struct efi_setup_data *esd = (void *)sd + sizeof(struct setup_data);
143 
144 	esd->fw_vendor = efi.fw_vendor;
145 	esd->runtime = efi.runtime;
146 	esd->tables = efi.config_table;
147 	esd->smbios = efi.smbios;
148 
149 	sd->type = SETUP_EFI;
150 	sd->len = sizeof(struct efi_setup_data);
151 
152 	/* Add setup data */
153 	setup_data_phys = params_load_addr + efi_setup_data_offset;
154 	sd->next = params->hdr.setup_data;
155 	params->hdr.setup_data = setup_data_phys;
156 
157 	return 0;
158 }
159 
160 static int
161 setup_efi_state(struct boot_params *params, unsigned long params_load_addr,
162 		unsigned int efi_map_offset, unsigned int efi_map_sz,
163 		unsigned int efi_setup_data_offset)
164 {
165 	struct efi_info *current_ei = &boot_params.efi_info;
166 	struct efi_info *ei = &params->efi_info;
167 
168 	if (!efi_enabled(EFI_RUNTIME_SERVICES))
169 		return 0;
170 
171 	if (!current_ei->efi_memmap_size)
172 		return 0;
173 
174 	/*
175 	 * If 1:1 mapping is not enabled, second kernel can not setup EFI
176 	 * and use EFI run time services. User space will have to pass
177 	 * acpi_rsdp=<addr> on kernel command line to make second kernel boot
178 	 * without efi.
179 	 */
180 	if (efi_enabled(EFI_OLD_MEMMAP))
181 		return 0;
182 
183 	ei->efi_loader_signature = current_ei->efi_loader_signature;
184 	ei->efi_systab = current_ei->efi_systab;
185 	ei->efi_systab_hi = current_ei->efi_systab_hi;
186 
187 	ei->efi_memdesc_version = current_ei->efi_memdesc_version;
188 	ei->efi_memdesc_size = efi_get_runtime_map_desc_size();
189 
190 	setup_efi_info_memmap(params, params_load_addr, efi_map_offset,
191 			      efi_map_sz);
192 	prepare_add_efi_setup_data(params, params_load_addr,
193 				   efi_setup_data_offset);
194 	return 0;
195 }
196 #endif /* CONFIG_EFI */
197 
198 static int
199 setup_boot_parameters(struct kimage *image, struct boot_params *params,
200 		      unsigned long params_load_addr,
201 		      unsigned int efi_map_offset, unsigned int efi_map_sz,
202 		      unsigned int efi_setup_data_offset)
203 {
204 	unsigned int nr_e820_entries;
205 	unsigned long long mem_k, start, end;
206 	int i, ret = 0;
207 
208 	/* Get subarch from existing bootparams */
209 	params->hdr.hardware_subarch = boot_params.hdr.hardware_subarch;
210 
211 	/* Copying screen_info will do? */
212 	memcpy(&params->screen_info, &boot_params.screen_info,
213 				sizeof(struct screen_info));
214 
215 	/* Fill in memsize later */
216 	params->screen_info.ext_mem_k = 0;
217 	params->alt_mem_k = 0;
218 
219 	/* Always fill in RSDP: it is either 0 or a valid value */
220 	params->acpi_rsdp_addr = boot_params.acpi_rsdp_addr;
221 
222 	/* Default APM info */
223 	memset(&params->apm_bios_info, 0, sizeof(params->apm_bios_info));
224 
225 	/* Default drive info */
226 	memset(&params->hd0_info, 0, sizeof(params->hd0_info));
227 	memset(&params->hd1_info, 0, sizeof(params->hd1_info));
228 
229 	if (image->type == KEXEC_TYPE_CRASH) {
230 		ret = crash_setup_memmap_entries(image, params);
231 		if (ret)
232 			return ret;
233 	} else
234 		setup_e820_entries(params);
235 
236 	nr_e820_entries = params->e820_entries;
237 
238 	for (i = 0; i < nr_e820_entries; i++) {
239 		if (params->e820_table[i].type != E820_TYPE_RAM)
240 			continue;
241 		start = params->e820_table[i].addr;
242 		end = params->e820_table[i].addr + params->e820_table[i].size - 1;
243 
244 		if ((start <= 0x100000) && end > 0x100000) {
245 			mem_k = (end >> 10) - (0x100000 >> 10);
246 			params->screen_info.ext_mem_k = mem_k;
247 			params->alt_mem_k = mem_k;
248 			if (mem_k > 0xfc00)
249 				params->screen_info.ext_mem_k = 0xfc00; /* 64M*/
250 			if (mem_k > 0xffffffff)
251 				params->alt_mem_k = 0xffffffff;
252 		}
253 	}
254 
255 #ifdef CONFIG_EFI
256 	/* Setup EFI state */
257 	setup_efi_state(params, params_load_addr, efi_map_offset, efi_map_sz,
258 			efi_setup_data_offset);
259 #endif
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 	if (!(header->xloadflags & XLF_5LEVEL) && pgtable_l5_enabled()) {
323 		pr_err("bzImage cannot handle 5-level paging mode.\n");
324 		return ret;
325 	}
326 
327 	/* I've got a bzImage */
328 	pr_debug("It's a relocatable bzImage64\n");
329 	ret = 0;
330 
331 	return ret;
332 }
333 
334 static void *bzImage64_load(struct kimage *image, char *kernel,
335 			    unsigned long kernel_len, char *initrd,
336 			    unsigned long initrd_len, char *cmdline,
337 			    unsigned long cmdline_len)
338 {
339 
340 	struct setup_header *header;
341 	int setup_sects, kern16_size, ret = 0;
342 	unsigned long setup_header_size, params_cmdline_sz;
343 	struct boot_params *params;
344 	unsigned long bootparam_load_addr, kernel_load_addr, initrd_load_addr;
345 	struct bzimage64_data *ldata;
346 	struct kexec_entry64_regs regs64;
347 	void *stack;
348 	unsigned int setup_hdr_offset = offsetof(struct boot_params, hdr);
349 	unsigned int efi_map_offset, efi_map_sz, efi_setup_data_offset;
350 	struct kexec_buf kbuf = { .image = image, .buf_max = ULONG_MAX,
351 				  .top_down = true };
352 	struct kexec_buf pbuf = { .image = image, .buf_min = MIN_PURGATORY_ADDR,
353 				  .buf_max = ULONG_MAX, .top_down = true };
354 
355 	header = (struct setup_header *)(kernel + setup_hdr_offset);
356 	setup_sects = header->setup_sects;
357 	if (setup_sects == 0)
358 		setup_sects = 4;
359 
360 	kern16_size = (setup_sects + 1) * 512;
361 	if (kernel_len < kern16_size) {
362 		pr_err("bzImage truncated\n");
363 		return ERR_PTR(-ENOEXEC);
364 	}
365 
366 	if (cmdline_len > header->cmdline_size) {
367 		pr_err("Kernel command line too long\n");
368 		return ERR_PTR(-EINVAL);
369 	}
370 
371 	/*
372 	 * In case of crash dump, we will append elfcorehdr=<addr> to
373 	 * command line. Make sure it does not overflow
374 	 */
375 	if (cmdline_len + MAX_ELFCOREHDR_STR_LEN > header->cmdline_size) {
376 		pr_debug("Appending elfcorehdr=<addr> to command line exceeds maximum allowed length\n");
377 		return ERR_PTR(-EINVAL);
378 	}
379 
380 	/* Allocate and load backup region */
381 	if (image->type == KEXEC_TYPE_CRASH) {
382 		ret = crash_load_segments(image);
383 		if (ret)
384 			return ERR_PTR(ret);
385 	}
386 
387 	/*
388 	 * Load purgatory. For 64bit entry point, purgatory  code can be
389 	 * anywhere.
390 	 */
391 	ret = kexec_load_purgatory(image, &pbuf);
392 	if (ret) {
393 		pr_err("Loading purgatory failed\n");
394 		return ERR_PTR(ret);
395 	}
396 
397 	pr_debug("Loaded purgatory at 0x%lx\n", pbuf.mem);
398 
399 
400 	/*
401 	 * Load Bootparams and cmdline and space for efi stuff.
402 	 *
403 	 * Allocate memory together for multiple data structures so
404 	 * that they all can go in single area/segment and we don't
405 	 * have to create separate segment for each. Keeps things
406 	 * little bit simple
407 	 */
408 	efi_map_sz = efi_get_runtime_map_size();
409 	params_cmdline_sz = sizeof(struct boot_params) + cmdline_len +
410 				MAX_ELFCOREHDR_STR_LEN;
411 	params_cmdline_sz = ALIGN(params_cmdline_sz, 16);
412 	kbuf.bufsz = params_cmdline_sz + ALIGN(efi_map_sz, 16) +
413 				sizeof(struct setup_data) +
414 				sizeof(struct efi_setup_data);
415 
416 	params = kzalloc(kbuf.bufsz, GFP_KERNEL);
417 	if (!params)
418 		return ERR_PTR(-ENOMEM);
419 	efi_map_offset = params_cmdline_sz;
420 	efi_setup_data_offset = efi_map_offset + ALIGN(efi_map_sz, 16);
421 
422 	/* Copy setup header onto bootparams. Documentation/x86/boot.rst */
423 	setup_header_size = 0x0202 + kernel[0x0201] - setup_hdr_offset;
424 
425 	/* Is there a limit on setup header size? */
426 	memcpy(&params->hdr, (kernel + setup_hdr_offset), setup_header_size);
427 
428 	kbuf.buffer = params;
429 	kbuf.memsz = kbuf.bufsz;
430 	kbuf.buf_align = 16;
431 	kbuf.buf_min = MIN_BOOTPARAM_ADDR;
432 	ret = kexec_add_buffer(&kbuf);
433 	if (ret)
434 		goto out_free_params;
435 	bootparam_load_addr = kbuf.mem;
436 	pr_debug("Loaded boot_param, command line and misc at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
437 		 bootparam_load_addr, kbuf.bufsz, kbuf.bufsz);
438 
439 	/* Load kernel */
440 	kbuf.buffer = kernel + kern16_size;
441 	kbuf.bufsz =  kernel_len - kern16_size;
442 	kbuf.memsz = PAGE_ALIGN(header->init_size);
443 	kbuf.buf_align = header->kernel_alignment;
444 	kbuf.buf_min = MIN_KERNEL_LOAD_ADDR;
445 	kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
446 	ret = kexec_add_buffer(&kbuf);
447 	if (ret)
448 		goto out_free_params;
449 	kernel_load_addr = kbuf.mem;
450 
451 	pr_debug("Loaded 64bit kernel at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
452 		 kernel_load_addr, kbuf.bufsz, kbuf.memsz);
453 
454 	/* Load initrd high */
455 	if (initrd) {
456 		kbuf.buffer = initrd;
457 		kbuf.bufsz = kbuf.memsz = initrd_len;
458 		kbuf.buf_align = PAGE_SIZE;
459 		kbuf.buf_min = MIN_INITRD_LOAD_ADDR;
460 		kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
461 		ret = kexec_add_buffer(&kbuf);
462 		if (ret)
463 			goto out_free_params;
464 		initrd_load_addr = kbuf.mem;
465 
466 		pr_debug("Loaded initrd at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
467 				initrd_load_addr, initrd_len, initrd_len);
468 
469 		setup_initrd(params, initrd_load_addr, initrd_len);
470 	}
471 
472 	setup_cmdline(image, params, bootparam_load_addr,
473 		      sizeof(struct boot_params), cmdline, cmdline_len);
474 
475 	/* bootloader info. Do we need a separate ID for kexec kernel loader? */
476 	params->hdr.type_of_loader = 0x0D << 4;
477 	params->hdr.loadflags = 0;
478 
479 	/* Setup purgatory regs for entry */
480 	ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", &regs64,
481 					     sizeof(regs64), 1);
482 	if (ret)
483 		goto out_free_params;
484 
485 	regs64.rbx = 0; /* Bootstrap Processor */
486 	regs64.rsi = bootparam_load_addr;
487 	regs64.rip = kernel_load_addr + 0x200;
488 	stack = kexec_purgatory_get_symbol_addr(image, "stack_end");
489 	if (IS_ERR(stack)) {
490 		pr_err("Could not find address of symbol stack_end\n");
491 		ret = -EINVAL;
492 		goto out_free_params;
493 	}
494 
495 	regs64.rsp = (unsigned long)stack;
496 	ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", &regs64,
497 					     sizeof(regs64), 0);
498 	if (ret)
499 		goto out_free_params;
500 
501 	ret = setup_boot_parameters(image, params, bootparam_load_addr,
502 				    efi_map_offset, efi_map_sz,
503 				    efi_setup_data_offset);
504 	if (ret)
505 		goto out_free_params;
506 
507 	/* Allocate loader specific data */
508 	ldata = kzalloc(sizeof(struct bzimage64_data), GFP_KERNEL);
509 	if (!ldata) {
510 		ret = -ENOMEM;
511 		goto out_free_params;
512 	}
513 
514 	/*
515 	 * Store pointer to params so that it could be freed after loading
516 	 * params segment has been loaded and contents have been copied
517 	 * somewhere else.
518 	 */
519 	ldata->bootparams_buf = params;
520 	return ldata;
521 
522 out_free_params:
523 	kfree(params);
524 	return ERR_PTR(ret);
525 }
526 
527 /* This cleanup function is called after various segments have been loaded */
528 static int bzImage64_cleanup(void *loader_data)
529 {
530 	struct bzimage64_data *ldata = loader_data;
531 
532 	if (!ldata)
533 		return 0;
534 
535 	kfree(ldata->bootparams_buf);
536 	ldata->bootparams_buf = NULL;
537 
538 	return 0;
539 }
540 
541 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG
542 static int bzImage64_verify_sig(const char *kernel, unsigned long kernel_len)
543 {
544 	int ret;
545 
546 	ret = verify_pefile_signature(kernel, kernel_len,
547 				      VERIFY_USE_SECONDARY_KEYRING,
548 				      VERIFYING_KEXEC_PE_SIGNATURE);
549 	if (ret == -ENOKEY && IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING)) {
550 		ret = verify_pefile_signature(kernel, kernel_len,
551 					      VERIFY_USE_PLATFORM_KEYRING,
552 					      VERIFYING_KEXEC_PE_SIGNATURE);
553 	}
554 	return ret;
555 }
556 #endif
557 
558 const struct kexec_file_ops kexec_bzImage64_ops = {
559 	.probe = bzImage64_probe,
560 	.load = bzImage64_load,
561 	.cleanup = bzImage64_cleanup,
562 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG
563 	.verify_sig = bzImage64_verify_sig,
564 #endif
565 };
566