xref: /linux/kernel/kexec_file.c (revision a8b70ccf10e38775785d9cb12ead916474549f99)
1 /*
2  * kexec: kexec_file_load system call
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) KBUILD_MODNAME ": " fmt
13 
14 #include <linux/capability.h>
15 #include <linux/mm.h>
16 #include <linux/file.h>
17 #include <linux/slab.h>
18 #include <linux/kexec.h>
19 #include <linux/mutex.h>
20 #include <linux/list.h>
21 #include <linux/fs.h>
22 #include <linux/ima.h>
23 #include <crypto/hash.h>
24 #include <crypto/sha.h>
25 #include <linux/elf.h>
26 #include <linux/elfcore.h>
27 #include <linux/kernel.h>
28 #include <linux/kexec.h>
29 #include <linux/slab.h>
30 #include <linux/syscalls.h>
31 #include <linux/vmalloc.h>
32 #include "kexec_internal.h"
33 
34 static int kexec_calculate_store_digests(struct kimage *image);
35 
36 /*
37  * Currently this is the only default function that is exported as some
38  * architectures need it to do additional handlings.
39  * In the future, other default functions may be exported too if required.
40  */
41 int kexec_image_probe_default(struct kimage *image, void *buf,
42 			      unsigned long buf_len)
43 {
44 	const struct kexec_file_ops * const *fops;
45 	int ret = -ENOEXEC;
46 
47 	for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
48 		ret = (*fops)->probe(buf, buf_len);
49 		if (!ret) {
50 			image->fops = *fops;
51 			return ret;
52 		}
53 	}
54 
55 	return ret;
56 }
57 
58 /* Architectures can provide this probe function */
59 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
60 					 unsigned long buf_len)
61 {
62 	return kexec_image_probe_default(image, buf, buf_len);
63 }
64 
65 static void *kexec_image_load_default(struct kimage *image)
66 {
67 	if (!image->fops || !image->fops->load)
68 		return ERR_PTR(-ENOEXEC);
69 
70 	return image->fops->load(image, image->kernel_buf,
71 				 image->kernel_buf_len, image->initrd_buf,
72 				 image->initrd_buf_len, image->cmdline_buf,
73 				 image->cmdline_buf_len);
74 }
75 
76 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
77 {
78 	return kexec_image_load_default(image);
79 }
80 
81 static int kexec_image_post_load_cleanup_default(struct kimage *image)
82 {
83 	if (!image->fops || !image->fops->cleanup)
84 		return 0;
85 
86 	return image->fops->cleanup(image->image_loader_data);
87 }
88 
89 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
90 {
91 	return kexec_image_post_load_cleanup_default(image);
92 }
93 
94 #ifdef CONFIG_KEXEC_VERIFY_SIG
95 static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
96 					  unsigned long buf_len)
97 {
98 	if (!image->fops || !image->fops->verify_sig) {
99 		pr_debug("kernel loader does not support signature verification.\n");
100 		return -EKEYREJECTED;
101 	}
102 
103 	return image->fops->verify_sig(buf, buf_len);
104 }
105 
106 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
107 					unsigned long buf_len)
108 {
109 	return kexec_image_verify_sig_default(image, buf, buf_len);
110 }
111 #endif
112 
113 /*
114  * arch_kexec_apply_relocations_add - apply relocations of type RELA
115  * @pi:		Purgatory to be relocated.
116  * @section:	Section relocations applying to.
117  * @relsec:	Section containing RELAs.
118  * @symtab:	Corresponding symtab.
119  *
120  * Return: 0 on success, negative errno on error.
121  */
122 int __weak
123 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section,
124 				 const Elf_Shdr *relsec, const Elf_Shdr *symtab)
125 {
126 	pr_err("RELA relocation unsupported.\n");
127 	return -ENOEXEC;
128 }
129 
130 /*
131  * arch_kexec_apply_relocations - apply relocations of type REL
132  * @pi:		Purgatory to be relocated.
133  * @section:	Section relocations applying to.
134  * @relsec:	Section containing RELs.
135  * @symtab:	Corresponding symtab.
136  *
137  * Return: 0 on success, negative errno on error.
138  */
139 int __weak
140 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section,
141 			     const Elf_Shdr *relsec, const Elf_Shdr *symtab)
142 {
143 	pr_err("REL relocation unsupported.\n");
144 	return -ENOEXEC;
145 }
146 
147 /*
148  * Free up memory used by kernel, initrd, and command line. This is temporary
149  * memory allocation which is not needed any more after these buffers have
150  * been loaded into separate segments and have been copied elsewhere.
151  */
152 void kimage_file_post_load_cleanup(struct kimage *image)
153 {
154 	struct purgatory_info *pi = &image->purgatory_info;
155 
156 	vfree(image->kernel_buf);
157 	image->kernel_buf = NULL;
158 
159 	vfree(image->initrd_buf);
160 	image->initrd_buf = NULL;
161 
162 	kfree(image->cmdline_buf);
163 	image->cmdline_buf = NULL;
164 
165 	vfree(pi->purgatory_buf);
166 	pi->purgatory_buf = NULL;
167 
168 	vfree(pi->sechdrs);
169 	pi->sechdrs = NULL;
170 
171 	/* See if architecture has anything to cleanup post load */
172 	arch_kimage_file_post_load_cleanup(image);
173 
174 	/*
175 	 * Above call should have called into bootloader to free up
176 	 * any data stored in kimage->image_loader_data. It should
177 	 * be ok now to free it up.
178 	 */
179 	kfree(image->image_loader_data);
180 	image->image_loader_data = NULL;
181 }
182 
183 /*
184  * In file mode list of segments is prepared by kernel. Copy relevant
185  * data from user space, do error checking, prepare segment list
186  */
187 static int
188 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
189 			     const char __user *cmdline_ptr,
190 			     unsigned long cmdline_len, unsigned flags)
191 {
192 	int ret = 0;
193 	void *ldata;
194 	loff_t size;
195 
196 	ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
197 				       &size, INT_MAX, READING_KEXEC_IMAGE);
198 	if (ret)
199 		return ret;
200 	image->kernel_buf_len = size;
201 
202 	/* IMA needs to pass the measurement list to the next kernel. */
203 	ima_add_kexec_buffer(image);
204 
205 	/* Call arch image probe handlers */
206 	ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
207 					    image->kernel_buf_len);
208 	if (ret)
209 		goto out;
210 
211 #ifdef CONFIG_KEXEC_VERIFY_SIG
212 	ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
213 					   image->kernel_buf_len);
214 	if (ret) {
215 		pr_debug("kernel signature verification failed.\n");
216 		goto out;
217 	}
218 	pr_debug("kernel signature verification successful.\n");
219 #endif
220 	/* It is possible that there no initramfs is being loaded */
221 	if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
222 		ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
223 					       &size, INT_MAX,
224 					       READING_KEXEC_INITRAMFS);
225 		if (ret)
226 			goto out;
227 		image->initrd_buf_len = size;
228 	}
229 
230 	if (cmdline_len) {
231 		image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
232 		if (IS_ERR(image->cmdline_buf)) {
233 			ret = PTR_ERR(image->cmdline_buf);
234 			image->cmdline_buf = NULL;
235 			goto out;
236 		}
237 
238 		image->cmdline_buf_len = cmdline_len;
239 
240 		/* command line should be a string with last byte null */
241 		if (image->cmdline_buf[cmdline_len - 1] != '\0') {
242 			ret = -EINVAL;
243 			goto out;
244 		}
245 	}
246 
247 	/* Call arch image load handlers */
248 	ldata = arch_kexec_kernel_image_load(image);
249 
250 	if (IS_ERR(ldata)) {
251 		ret = PTR_ERR(ldata);
252 		goto out;
253 	}
254 
255 	image->image_loader_data = ldata;
256 out:
257 	/* In case of error, free up all allocated memory in this function */
258 	if (ret)
259 		kimage_file_post_load_cleanup(image);
260 	return ret;
261 }
262 
263 static int
264 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
265 		       int initrd_fd, const char __user *cmdline_ptr,
266 		       unsigned long cmdline_len, unsigned long flags)
267 {
268 	int ret;
269 	struct kimage *image;
270 	bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
271 
272 	image = do_kimage_alloc_init();
273 	if (!image)
274 		return -ENOMEM;
275 
276 	image->file_mode = 1;
277 
278 	if (kexec_on_panic) {
279 		/* Enable special crash kernel control page alloc policy. */
280 		image->control_page = crashk_res.start;
281 		image->type = KEXEC_TYPE_CRASH;
282 	}
283 
284 	ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
285 					   cmdline_ptr, cmdline_len, flags);
286 	if (ret)
287 		goto out_free_image;
288 
289 	ret = sanity_check_segment_list(image);
290 	if (ret)
291 		goto out_free_post_load_bufs;
292 
293 	ret = -ENOMEM;
294 	image->control_code_page = kimage_alloc_control_pages(image,
295 					   get_order(KEXEC_CONTROL_PAGE_SIZE));
296 	if (!image->control_code_page) {
297 		pr_err("Could not allocate control_code_buffer\n");
298 		goto out_free_post_load_bufs;
299 	}
300 
301 	if (!kexec_on_panic) {
302 		image->swap_page = kimage_alloc_control_pages(image, 0);
303 		if (!image->swap_page) {
304 			pr_err("Could not allocate swap buffer\n");
305 			goto out_free_control_pages;
306 		}
307 	}
308 
309 	*rimage = image;
310 	return 0;
311 out_free_control_pages:
312 	kimage_free_page_list(&image->control_pages);
313 out_free_post_load_bufs:
314 	kimage_file_post_load_cleanup(image);
315 out_free_image:
316 	kfree(image);
317 	return ret;
318 }
319 
320 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
321 		unsigned long, cmdline_len, const char __user *, cmdline_ptr,
322 		unsigned long, flags)
323 {
324 	int ret = 0, i;
325 	struct kimage **dest_image, *image;
326 
327 	/* We only trust the superuser with rebooting the system. */
328 	if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
329 		return -EPERM;
330 
331 	/* Make sure we have a legal set of flags */
332 	if (flags != (flags & KEXEC_FILE_FLAGS))
333 		return -EINVAL;
334 
335 	image = NULL;
336 
337 	if (!mutex_trylock(&kexec_mutex))
338 		return -EBUSY;
339 
340 	dest_image = &kexec_image;
341 	if (flags & KEXEC_FILE_ON_CRASH) {
342 		dest_image = &kexec_crash_image;
343 		if (kexec_crash_image)
344 			arch_kexec_unprotect_crashkres();
345 	}
346 
347 	if (flags & KEXEC_FILE_UNLOAD)
348 		goto exchange;
349 
350 	/*
351 	 * In case of crash, new kernel gets loaded in reserved region. It is
352 	 * same memory where old crash kernel might be loaded. Free any
353 	 * current crash dump kernel before we corrupt it.
354 	 */
355 	if (flags & KEXEC_FILE_ON_CRASH)
356 		kimage_free(xchg(&kexec_crash_image, NULL));
357 
358 	ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
359 				     cmdline_len, flags);
360 	if (ret)
361 		goto out;
362 
363 	ret = machine_kexec_prepare(image);
364 	if (ret)
365 		goto out;
366 
367 	/*
368 	 * Some architecture(like S390) may touch the crash memory before
369 	 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
370 	 */
371 	ret = kimage_crash_copy_vmcoreinfo(image);
372 	if (ret)
373 		goto out;
374 
375 	ret = kexec_calculate_store_digests(image);
376 	if (ret)
377 		goto out;
378 
379 	for (i = 0; i < image->nr_segments; i++) {
380 		struct kexec_segment *ksegment;
381 
382 		ksegment = &image->segment[i];
383 		pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
384 			 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
385 			 ksegment->memsz);
386 
387 		ret = kimage_load_segment(image, &image->segment[i]);
388 		if (ret)
389 			goto out;
390 	}
391 
392 	kimage_terminate(image);
393 
394 	/*
395 	 * Free up any temporary buffers allocated which are not needed
396 	 * after image has been loaded
397 	 */
398 	kimage_file_post_load_cleanup(image);
399 exchange:
400 	image = xchg(dest_image, image);
401 out:
402 	if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
403 		arch_kexec_protect_crashkres();
404 
405 	mutex_unlock(&kexec_mutex);
406 	kimage_free(image);
407 	return ret;
408 }
409 
410 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
411 				    struct kexec_buf *kbuf)
412 {
413 	struct kimage *image = kbuf->image;
414 	unsigned long temp_start, temp_end;
415 
416 	temp_end = min(end, kbuf->buf_max);
417 	temp_start = temp_end - kbuf->memsz;
418 
419 	do {
420 		/* align down start */
421 		temp_start = temp_start & (~(kbuf->buf_align - 1));
422 
423 		if (temp_start < start || temp_start < kbuf->buf_min)
424 			return 0;
425 
426 		temp_end = temp_start + kbuf->memsz - 1;
427 
428 		/*
429 		 * Make sure this does not conflict with any of existing
430 		 * segments
431 		 */
432 		if (kimage_is_destination_range(image, temp_start, temp_end)) {
433 			temp_start = temp_start - PAGE_SIZE;
434 			continue;
435 		}
436 
437 		/* We found a suitable memory range */
438 		break;
439 	} while (1);
440 
441 	/* If we are here, we found a suitable memory range */
442 	kbuf->mem = temp_start;
443 
444 	/* Success, stop navigating through remaining System RAM ranges */
445 	return 1;
446 }
447 
448 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
449 				     struct kexec_buf *kbuf)
450 {
451 	struct kimage *image = kbuf->image;
452 	unsigned long temp_start, temp_end;
453 
454 	temp_start = max(start, kbuf->buf_min);
455 
456 	do {
457 		temp_start = ALIGN(temp_start, kbuf->buf_align);
458 		temp_end = temp_start + kbuf->memsz - 1;
459 
460 		if (temp_end > end || temp_end > kbuf->buf_max)
461 			return 0;
462 		/*
463 		 * Make sure this does not conflict with any of existing
464 		 * segments
465 		 */
466 		if (kimage_is_destination_range(image, temp_start, temp_end)) {
467 			temp_start = temp_start + PAGE_SIZE;
468 			continue;
469 		}
470 
471 		/* We found a suitable memory range */
472 		break;
473 	} while (1);
474 
475 	/* If we are here, we found a suitable memory range */
476 	kbuf->mem = temp_start;
477 
478 	/* Success, stop navigating through remaining System RAM ranges */
479 	return 1;
480 }
481 
482 static int locate_mem_hole_callback(struct resource *res, void *arg)
483 {
484 	struct kexec_buf *kbuf = (struct kexec_buf *)arg;
485 	u64 start = res->start, end = res->end;
486 	unsigned long sz = end - start + 1;
487 
488 	/* Returning 0 will take to next memory range */
489 	if (sz < kbuf->memsz)
490 		return 0;
491 
492 	if (end < kbuf->buf_min || start > kbuf->buf_max)
493 		return 0;
494 
495 	/*
496 	 * Allocate memory top down with-in ram range. Otherwise bottom up
497 	 * allocation.
498 	 */
499 	if (kbuf->top_down)
500 		return locate_mem_hole_top_down(start, end, kbuf);
501 	return locate_mem_hole_bottom_up(start, end, kbuf);
502 }
503 
504 /**
505  * arch_kexec_walk_mem - call func(data) on free memory regions
506  * @kbuf:	Context info for the search. Also passed to @func.
507  * @func:	Function to call for each memory region.
508  *
509  * Return: The memory walk will stop when func returns a non-zero value
510  * and that value will be returned. If all free regions are visited without
511  * func returning non-zero, then zero will be returned.
512  */
513 int __weak arch_kexec_walk_mem(struct kexec_buf *kbuf,
514 			       int (*func)(struct resource *, void *))
515 {
516 	if (kbuf->image->type == KEXEC_TYPE_CRASH)
517 		return walk_iomem_res_desc(crashk_res.desc,
518 					   IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
519 					   crashk_res.start, crashk_res.end,
520 					   kbuf, func);
521 	else
522 		return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
523 }
524 
525 /**
526  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
527  * @kbuf:	Parameters for the memory search.
528  *
529  * On success, kbuf->mem will have the start address of the memory region found.
530  *
531  * Return: 0 on success, negative errno on error.
532  */
533 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
534 {
535 	int ret;
536 
537 	ret = arch_kexec_walk_mem(kbuf, locate_mem_hole_callback);
538 
539 	return ret == 1 ? 0 : -EADDRNOTAVAIL;
540 }
541 
542 /**
543  * kexec_add_buffer - place a buffer in a kexec segment
544  * @kbuf:	Buffer contents and memory parameters.
545  *
546  * This function assumes that kexec_mutex is held.
547  * On successful return, @kbuf->mem will have the physical address of
548  * the buffer in memory.
549  *
550  * Return: 0 on success, negative errno on error.
551  */
552 int kexec_add_buffer(struct kexec_buf *kbuf)
553 {
554 
555 	struct kexec_segment *ksegment;
556 	int ret;
557 
558 	/* Currently adding segment this way is allowed only in file mode */
559 	if (!kbuf->image->file_mode)
560 		return -EINVAL;
561 
562 	if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
563 		return -EINVAL;
564 
565 	/*
566 	 * Make sure we are not trying to add buffer after allocating
567 	 * control pages. All segments need to be placed first before
568 	 * any control pages are allocated. As control page allocation
569 	 * logic goes through list of segments to make sure there are
570 	 * no destination overlaps.
571 	 */
572 	if (!list_empty(&kbuf->image->control_pages)) {
573 		WARN_ON(1);
574 		return -EINVAL;
575 	}
576 
577 	/* Ensure minimum alignment needed for segments. */
578 	kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
579 	kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
580 
581 	/* Walk the RAM ranges and allocate a suitable range for the buffer */
582 	ret = kexec_locate_mem_hole(kbuf);
583 	if (ret)
584 		return ret;
585 
586 	/* Found a suitable memory range */
587 	ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
588 	ksegment->kbuf = kbuf->buffer;
589 	ksegment->bufsz = kbuf->bufsz;
590 	ksegment->mem = kbuf->mem;
591 	ksegment->memsz = kbuf->memsz;
592 	kbuf->image->nr_segments++;
593 	return 0;
594 }
595 
596 /* Calculate and store the digest of segments */
597 static int kexec_calculate_store_digests(struct kimage *image)
598 {
599 	struct crypto_shash *tfm;
600 	struct shash_desc *desc;
601 	int ret = 0, i, j, zero_buf_sz, sha_region_sz;
602 	size_t desc_size, nullsz;
603 	char *digest;
604 	void *zero_buf;
605 	struct kexec_sha_region *sha_regions;
606 	struct purgatory_info *pi = &image->purgatory_info;
607 
608 	if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
609 		return 0;
610 
611 	zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
612 	zero_buf_sz = PAGE_SIZE;
613 
614 	tfm = crypto_alloc_shash("sha256", 0, 0);
615 	if (IS_ERR(tfm)) {
616 		ret = PTR_ERR(tfm);
617 		goto out;
618 	}
619 
620 	desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
621 	desc = kzalloc(desc_size, GFP_KERNEL);
622 	if (!desc) {
623 		ret = -ENOMEM;
624 		goto out_free_tfm;
625 	}
626 
627 	sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
628 	sha_regions = vzalloc(sha_region_sz);
629 	if (!sha_regions)
630 		goto out_free_desc;
631 
632 	desc->tfm   = tfm;
633 	desc->flags = 0;
634 
635 	ret = crypto_shash_init(desc);
636 	if (ret < 0)
637 		goto out_free_sha_regions;
638 
639 	digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
640 	if (!digest) {
641 		ret = -ENOMEM;
642 		goto out_free_sha_regions;
643 	}
644 
645 	for (j = i = 0; i < image->nr_segments; i++) {
646 		struct kexec_segment *ksegment;
647 
648 		ksegment = &image->segment[i];
649 		/*
650 		 * Skip purgatory as it will be modified once we put digest
651 		 * info in purgatory.
652 		 */
653 		if (ksegment->kbuf == pi->purgatory_buf)
654 			continue;
655 
656 		ret = crypto_shash_update(desc, ksegment->kbuf,
657 					  ksegment->bufsz);
658 		if (ret)
659 			break;
660 
661 		/*
662 		 * Assume rest of the buffer is filled with zero and
663 		 * update digest accordingly.
664 		 */
665 		nullsz = ksegment->memsz - ksegment->bufsz;
666 		while (nullsz) {
667 			unsigned long bytes = nullsz;
668 
669 			if (bytes > zero_buf_sz)
670 				bytes = zero_buf_sz;
671 			ret = crypto_shash_update(desc, zero_buf, bytes);
672 			if (ret)
673 				break;
674 			nullsz -= bytes;
675 		}
676 
677 		if (ret)
678 			break;
679 
680 		sha_regions[j].start = ksegment->mem;
681 		sha_regions[j].len = ksegment->memsz;
682 		j++;
683 	}
684 
685 	if (!ret) {
686 		ret = crypto_shash_final(desc, digest);
687 		if (ret)
688 			goto out_free_digest;
689 		ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
690 						     sha_regions, sha_region_sz, 0);
691 		if (ret)
692 			goto out_free_digest;
693 
694 		ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
695 						     digest, SHA256_DIGEST_SIZE, 0);
696 		if (ret)
697 			goto out_free_digest;
698 	}
699 
700 out_free_digest:
701 	kfree(digest);
702 out_free_sha_regions:
703 	vfree(sha_regions);
704 out_free_desc:
705 	kfree(desc);
706 out_free_tfm:
707 	kfree(tfm);
708 out:
709 	return ret;
710 }
711 
712 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
713 /*
714  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
715  * @pi:		Purgatory to be loaded.
716  * @kbuf:	Buffer to setup.
717  *
718  * Allocates the memory needed for the buffer. Caller is responsible to free
719  * the memory after use.
720  *
721  * Return: 0 on success, negative errno on error.
722  */
723 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
724 				      struct kexec_buf *kbuf)
725 {
726 	const Elf_Shdr *sechdrs;
727 	unsigned long bss_align;
728 	unsigned long bss_sz;
729 	unsigned long align;
730 	int i, ret;
731 
732 	sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
733 	kbuf->buf_align = bss_align = 1;
734 	kbuf->bufsz = bss_sz = 0;
735 
736 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
737 		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
738 			continue;
739 
740 		align = sechdrs[i].sh_addralign;
741 		if (sechdrs[i].sh_type != SHT_NOBITS) {
742 			if (kbuf->buf_align < align)
743 				kbuf->buf_align = align;
744 			kbuf->bufsz = ALIGN(kbuf->bufsz, align);
745 			kbuf->bufsz += sechdrs[i].sh_size;
746 		} else {
747 			if (bss_align < align)
748 				bss_align = align;
749 			bss_sz = ALIGN(bss_sz, align);
750 			bss_sz += sechdrs[i].sh_size;
751 		}
752 	}
753 	kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
754 	kbuf->memsz = kbuf->bufsz + bss_sz;
755 	if (kbuf->buf_align < bss_align)
756 		kbuf->buf_align = bss_align;
757 
758 	kbuf->buffer = vzalloc(kbuf->bufsz);
759 	if (!kbuf->buffer)
760 		return -ENOMEM;
761 	pi->purgatory_buf = kbuf->buffer;
762 
763 	ret = kexec_add_buffer(kbuf);
764 	if (ret)
765 		goto out;
766 
767 	return 0;
768 out:
769 	vfree(pi->purgatory_buf);
770 	pi->purgatory_buf = NULL;
771 	return ret;
772 }
773 
774 /*
775  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
776  * @pi:		Purgatory to be loaded.
777  * @kbuf:	Buffer prepared to store purgatory.
778  *
779  * Allocates the memory needed for the buffer. Caller is responsible to free
780  * the memory after use.
781  *
782  * Return: 0 on success, negative errno on error.
783  */
784 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
785 					 struct kexec_buf *kbuf)
786 {
787 	unsigned long bss_addr;
788 	unsigned long offset;
789 	Elf_Shdr *sechdrs;
790 	int i;
791 
792 	/*
793 	 * The section headers in kexec_purgatory are read-only. In order to
794 	 * have them modifiable make a temporary copy.
795 	 */
796 	sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
797 	if (!sechdrs)
798 		return -ENOMEM;
799 	memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
800 	       pi->ehdr->e_shnum * sizeof(Elf_Shdr));
801 	pi->sechdrs = sechdrs;
802 
803 	offset = 0;
804 	bss_addr = kbuf->mem + kbuf->bufsz;
805 	kbuf->image->start = pi->ehdr->e_entry;
806 
807 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
808 		unsigned long align;
809 		void *src, *dst;
810 
811 		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
812 			continue;
813 
814 		align = sechdrs[i].sh_addralign;
815 		if (sechdrs[i].sh_type == SHT_NOBITS) {
816 			bss_addr = ALIGN(bss_addr, align);
817 			sechdrs[i].sh_addr = bss_addr;
818 			bss_addr += sechdrs[i].sh_size;
819 			continue;
820 		}
821 
822 		offset = ALIGN(offset, align);
823 		if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
824 		    pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
825 		    pi->ehdr->e_entry < (sechdrs[i].sh_addr
826 					 + sechdrs[i].sh_size)) {
827 			kbuf->image->start -= sechdrs[i].sh_addr;
828 			kbuf->image->start += kbuf->mem + offset;
829 		}
830 
831 		src = (void *)pi->ehdr + sechdrs[i].sh_offset;
832 		dst = pi->purgatory_buf + offset;
833 		memcpy(dst, src, sechdrs[i].sh_size);
834 
835 		sechdrs[i].sh_addr = kbuf->mem + offset;
836 		sechdrs[i].sh_offset = offset;
837 		offset += sechdrs[i].sh_size;
838 	}
839 
840 	return 0;
841 }
842 
843 static int kexec_apply_relocations(struct kimage *image)
844 {
845 	int i, ret;
846 	struct purgatory_info *pi = &image->purgatory_info;
847 	const Elf_Shdr *sechdrs;
848 
849 	sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
850 
851 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
852 		const Elf_Shdr *relsec;
853 		const Elf_Shdr *symtab;
854 		Elf_Shdr *section;
855 
856 		relsec = sechdrs + i;
857 
858 		if (relsec->sh_type != SHT_RELA &&
859 		    relsec->sh_type != SHT_REL)
860 			continue;
861 
862 		/*
863 		 * For section of type SHT_RELA/SHT_REL,
864 		 * ->sh_link contains section header index of associated
865 		 * symbol table. And ->sh_info contains section header
866 		 * index of section to which relocations apply.
867 		 */
868 		if (relsec->sh_info >= pi->ehdr->e_shnum ||
869 		    relsec->sh_link >= pi->ehdr->e_shnum)
870 			return -ENOEXEC;
871 
872 		section = pi->sechdrs + relsec->sh_info;
873 		symtab = sechdrs + relsec->sh_link;
874 
875 		if (!(section->sh_flags & SHF_ALLOC))
876 			continue;
877 
878 		/*
879 		 * symtab->sh_link contain section header index of associated
880 		 * string table.
881 		 */
882 		if (symtab->sh_link >= pi->ehdr->e_shnum)
883 			/* Invalid section number? */
884 			continue;
885 
886 		/*
887 		 * Respective architecture needs to provide support for applying
888 		 * relocations of type SHT_RELA/SHT_REL.
889 		 */
890 		if (relsec->sh_type == SHT_RELA)
891 			ret = arch_kexec_apply_relocations_add(pi, section,
892 							       relsec, symtab);
893 		else if (relsec->sh_type == SHT_REL)
894 			ret = arch_kexec_apply_relocations(pi, section,
895 							   relsec, symtab);
896 		if (ret)
897 			return ret;
898 	}
899 
900 	return 0;
901 }
902 
903 /*
904  * kexec_load_purgatory - Load and relocate the purgatory object.
905  * @image:	Image to add the purgatory to.
906  * @kbuf:	Memory parameters to use.
907  *
908  * Allocates the memory needed for image->purgatory_info.sechdrs and
909  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
910  * to free the memory after use.
911  *
912  * Return: 0 on success, negative errno on error.
913  */
914 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
915 {
916 	struct purgatory_info *pi = &image->purgatory_info;
917 	int ret;
918 
919 	if (kexec_purgatory_size <= 0)
920 		return -EINVAL;
921 
922 	pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
923 
924 	ret = kexec_purgatory_setup_kbuf(pi, kbuf);
925 	if (ret)
926 		return ret;
927 
928 	ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
929 	if (ret)
930 		goto out_free_kbuf;
931 
932 	ret = kexec_apply_relocations(image);
933 	if (ret)
934 		goto out;
935 
936 	return 0;
937 out:
938 	vfree(pi->sechdrs);
939 	pi->sechdrs = NULL;
940 out_free_kbuf:
941 	vfree(pi->purgatory_buf);
942 	pi->purgatory_buf = NULL;
943 	return ret;
944 }
945 
946 /*
947  * kexec_purgatory_find_symbol - find a symbol in the purgatory
948  * @pi:		Purgatory to search in.
949  * @name:	Name of the symbol.
950  *
951  * Return: pointer to symbol in read-only symtab on success, NULL on error.
952  */
953 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
954 						  const char *name)
955 {
956 	const Elf_Shdr *sechdrs;
957 	const Elf_Ehdr *ehdr;
958 	const Elf_Sym *syms;
959 	const char *strtab;
960 	int i, k;
961 
962 	if (!pi->ehdr)
963 		return NULL;
964 
965 	ehdr = pi->ehdr;
966 	sechdrs = (void *)ehdr + ehdr->e_shoff;
967 
968 	for (i = 0; i < ehdr->e_shnum; i++) {
969 		if (sechdrs[i].sh_type != SHT_SYMTAB)
970 			continue;
971 
972 		if (sechdrs[i].sh_link >= ehdr->e_shnum)
973 			/* Invalid strtab section number */
974 			continue;
975 		strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
976 		syms = (void *)ehdr + sechdrs[i].sh_offset;
977 
978 		/* Go through symbols for a match */
979 		for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
980 			if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
981 				continue;
982 
983 			if (strcmp(strtab + syms[k].st_name, name) != 0)
984 				continue;
985 
986 			if (syms[k].st_shndx == SHN_UNDEF ||
987 			    syms[k].st_shndx >= ehdr->e_shnum) {
988 				pr_debug("Symbol: %s has bad section index %d.\n",
989 						name, syms[k].st_shndx);
990 				return NULL;
991 			}
992 
993 			/* Found the symbol we are looking for */
994 			return &syms[k];
995 		}
996 	}
997 
998 	return NULL;
999 }
1000 
1001 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1002 {
1003 	struct purgatory_info *pi = &image->purgatory_info;
1004 	const Elf_Sym *sym;
1005 	Elf_Shdr *sechdr;
1006 
1007 	sym = kexec_purgatory_find_symbol(pi, name);
1008 	if (!sym)
1009 		return ERR_PTR(-EINVAL);
1010 
1011 	sechdr = &pi->sechdrs[sym->st_shndx];
1012 
1013 	/*
1014 	 * Returns the address where symbol will finally be loaded after
1015 	 * kexec_load_segment()
1016 	 */
1017 	return (void *)(sechdr->sh_addr + sym->st_value);
1018 }
1019 
1020 /*
1021  * Get or set value of a symbol. If "get_value" is true, symbol value is
1022  * returned in buf otherwise symbol value is set based on value in buf.
1023  */
1024 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1025 				   void *buf, unsigned int size, bool get_value)
1026 {
1027 	struct purgatory_info *pi = &image->purgatory_info;
1028 	const Elf_Sym *sym;
1029 	Elf_Shdr *sec;
1030 	char *sym_buf;
1031 
1032 	sym = kexec_purgatory_find_symbol(pi, name);
1033 	if (!sym)
1034 		return -EINVAL;
1035 
1036 	if (sym->st_size != size) {
1037 		pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1038 		       name, (unsigned long)sym->st_size, size);
1039 		return -EINVAL;
1040 	}
1041 
1042 	sec = pi->sechdrs + sym->st_shndx;
1043 
1044 	if (sec->sh_type == SHT_NOBITS) {
1045 		pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1046 		       get_value ? "get" : "set");
1047 		return -EINVAL;
1048 	}
1049 
1050 	sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1051 
1052 	if (get_value)
1053 		memcpy((void *)buf, sym_buf, size);
1054 	else
1055 		memcpy((void *)sym_buf, buf, size);
1056 
1057 	return 0;
1058 }
1059 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1060 
1061 int crash_exclude_mem_range(struct crash_mem *mem,
1062 			    unsigned long long mstart, unsigned long long mend)
1063 {
1064 	int i, j;
1065 	unsigned long long start, end;
1066 	struct crash_mem_range temp_range = {0, 0};
1067 
1068 	for (i = 0; i < mem->nr_ranges; i++) {
1069 		start = mem->ranges[i].start;
1070 		end = mem->ranges[i].end;
1071 
1072 		if (mstart > end || mend < start)
1073 			continue;
1074 
1075 		/* Truncate any area outside of range */
1076 		if (mstart < start)
1077 			mstart = start;
1078 		if (mend > end)
1079 			mend = end;
1080 
1081 		/* Found completely overlapping range */
1082 		if (mstart == start && mend == end) {
1083 			mem->ranges[i].start = 0;
1084 			mem->ranges[i].end = 0;
1085 			if (i < mem->nr_ranges - 1) {
1086 				/* Shift rest of the ranges to left */
1087 				for (j = i; j < mem->nr_ranges - 1; j++) {
1088 					mem->ranges[j].start =
1089 						mem->ranges[j+1].start;
1090 					mem->ranges[j].end =
1091 							mem->ranges[j+1].end;
1092 				}
1093 			}
1094 			mem->nr_ranges--;
1095 			return 0;
1096 		}
1097 
1098 		if (mstart > start && mend < end) {
1099 			/* Split original range */
1100 			mem->ranges[i].end = mstart - 1;
1101 			temp_range.start = mend + 1;
1102 			temp_range.end = end;
1103 		} else if (mstart != start)
1104 			mem->ranges[i].end = mstart - 1;
1105 		else
1106 			mem->ranges[i].start = mend + 1;
1107 		break;
1108 	}
1109 
1110 	/* If a split happened, add the split to array */
1111 	if (!temp_range.end)
1112 		return 0;
1113 
1114 	/* Split happened */
1115 	if (i == mem->max_nr_ranges - 1)
1116 		return -ENOMEM;
1117 
1118 	/* Location where new range should go */
1119 	j = i + 1;
1120 	if (j < mem->nr_ranges) {
1121 		/* Move over all ranges one slot towards the end */
1122 		for (i = mem->nr_ranges - 1; i >= j; i--)
1123 			mem->ranges[i + 1] = mem->ranges[i];
1124 	}
1125 
1126 	mem->ranges[j].start = temp_range.start;
1127 	mem->ranges[j].end = temp_range.end;
1128 	mem->nr_ranges++;
1129 	return 0;
1130 }
1131 
1132 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1133 			  void **addr, unsigned long *sz)
1134 {
1135 	Elf64_Ehdr *ehdr;
1136 	Elf64_Phdr *phdr;
1137 	unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1138 	unsigned char *buf;
1139 	unsigned int cpu, i;
1140 	unsigned long long notes_addr;
1141 	unsigned long mstart, mend;
1142 
1143 	/* extra phdr for vmcoreinfo elf note */
1144 	nr_phdr = nr_cpus + 1;
1145 	nr_phdr += mem->nr_ranges;
1146 
1147 	/*
1148 	 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1149 	 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1150 	 * I think this is required by tools like gdb. So same physical
1151 	 * memory will be mapped in two elf headers. One will contain kernel
1152 	 * text virtual addresses and other will have __va(physical) addresses.
1153 	 */
1154 
1155 	nr_phdr++;
1156 	elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1157 	elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1158 
1159 	buf = vzalloc(elf_sz);
1160 	if (!buf)
1161 		return -ENOMEM;
1162 
1163 	ehdr = (Elf64_Ehdr *)buf;
1164 	phdr = (Elf64_Phdr *)(ehdr + 1);
1165 	memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1166 	ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1167 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1168 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1169 	ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1170 	memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1171 	ehdr->e_type = ET_CORE;
1172 	ehdr->e_machine = ELF_ARCH;
1173 	ehdr->e_version = EV_CURRENT;
1174 	ehdr->e_phoff = sizeof(Elf64_Ehdr);
1175 	ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1176 	ehdr->e_phentsize = sizeof(Elf64_Phdr);
1177 
1178 	/* Prepare one phdr of type PT_NOTE for each present cpu */
1179 	for_each_present_cpu(cpu) {
1180 		phdr->p_type = PT_NOTE;
1181 		notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1182 		phdr->p_offset = phdr->p_paddr = notes_addr;
1183 		phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1184 		(ehdr->e_phnum)++;
1185 		phdr++;
1186 	}
1187 
1188 	/* Prepare one PT_NOTE header for vmcoreinfo */
1189 	phdr->p_type = PT_NOTE;
1190 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1191 	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1192 	(ehdr->e_phnum)++;
1193 	phdr++;
1194 
1195 	/* Prepare PT_LOAD type program header for kernel text region */
1196 	if (kernel_map) {
1197 		phdr->p_type = PT_LOAD;
1198 		phdr->p_flags = PF_R|PF_W|PF_X;
1199 		phdr->p_vaddr = (Elf64_Addr)_text;
1200 		phdr->p_filesz = phdr->p_memsz = _end - _text;
1201 		phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1202 		ehdr->e_phnum++;
1203 		phdr++;
1204 	}
1205 
1206 	/* Go through all the ranges in mem->ranges[] and prepare phdr */
1207 	for (i = 0; i < mem->nr_ranges; i++) {
1208 		mstart = mem->ranges[i].start;
1209 		mend = mem->ranges[i].end;
1210 
1211 		phdr->p_type = PT_LOAD;
1212 		phdr->p_flags = PF_R|PF_W|PF_X;
1213 		phdr->p_offset  = mstart;
1214 
1215 		phdr->p_paddr = mstart;
1216 		phdr->p_vaddr = (unsigned long long) __va(mstart);
1217 		phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1218 		phdr->p_align = 0;
1219 		ehdr->e_phnum++;
1220 		phdr++;
1221 		pr_debug("Crash PT_LOAD elf header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n",
1222 			phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1223 			ehdr->e_phnum, phdr->p_offset);
1224 	}
1225 
1226 	*addr = buf;
1227 	*sz = elf_sz;
1228 	return 0;
1229 }
1230