xref: /linux/kernel/kexec_file.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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 <crypto/hash.h>
22 #include <crypto/sha.h>
23 #include <linux/syscalls.h>
24 #include <linux/vmalloc.h>
25 #include "kexec_internal.h"
26 
27 /*
28  * Declare these symbols weak so that if architecture provides a purgatory,
29  * these will be overridden.
30  */
31 char __weak kexec_purgatory[0];
32 size_t __weak kexec_purgatory_size = 0;
33 
34 static int kexec_calculate_store_digests(struct kimage *image);
35 
36 static int copy_file_from_fd(int fd, void **buf, unsigned long *buf_len)
37 {
38 	struct fd f = fdget(fd);
39 	int ret;
40 	struct kstat stat;
41 	loff_t pos;
42 	ssize_t bytes = 0;
43 
44 	if (!f.file)
45 		return -EBADF;
46 
47 	ret = vfs_getattr(&f.file->f_path, &stat);
48 	if (ret)
49 		goto out;
50 
51 	if (stat.size > INT_MAX) {
52 		ret = -EFBIG;
53 		goto out;
54 	}
55 
56 	/* Don't hand 0 to vmalloc, it whines. */
57 	if (stat.size == 0) {
58 		ret = -EINVAL;
59 		goto out;
60 	}
61 
62 	*buf = vmalloc(stat.size);
63 	if (!*buf) {
64 		ret = -ENOMEM;
65 		goto out;
66 	}
67 
68 	pos = 0;
69 	while (pos < stat.size) {
70 		bytes = kernel_read(f.file, pos, (char *)(*buf) + pos,
71 				    stat.size - pos);
72 		if (bytes < 0) {
73 			vfree(*buf);
74 			ret = bytes;
75 			goto out;
76 		}
77 
78 		if (bytes == 0)
79 			break;
80 		pos += bytes;
81 	}
82 
83 	if (pos != stat.size) {
84 		ret = -EBADF;
85 		vfree(*buf);
86 		goto out;
87 	}
88 
89 	*buf_len = pos;
90 out:
91 	fdput(f);
92 	return ret;
93 }
94 
95 /* Architectures can provide this probe function */
96 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
97 					 unsigned long buf_len)
98 {
99 	return -ENOEXEC;
100 }
101 
102 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
103 {
104 	return ERR_PTR(-ENOEXEC);
105 }
106 
107 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
108 {
109 	return -EINVAL;
110 }
111 
112 #ifdef CONFIG_KEXEC_VERIFY_SIG
113 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
114 					unsigned long buf_len)
115 {
116 	return -EKEYREJECTED;
117 }
118 #endif
119 
120 /* Apply relocations of type RELA */
121 int __weak
122 arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
123 				 unsigned int relsec)
124 {
125 	pr_err("RELA relocation unsupported.\n");
126 	return -ENOEXEC;
127 }
128 
129 /* Apply relocations of type REL */
130 int __weak
131 arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
132 			     unsigned int relsec)
133 {
134 	pr_err("REL relocation unsupported.\n");
135 	return -ENOEXEC;
136 }
137 
138 /*
139  * Free up memory used by kernel, initrd, and command line. This is temporary
140  * memory allocation which is not needed any more after these buffers have
141  * been loaded into separate segments and have been copied elsewhere.
142  */
143 void kimage_file_post_load_cleanup(struct kimage *image)
144 {
145 	struct purgatory_info *pi = &image->purgatory_info;
146 
147 	vfree(image->kernel_buf);
148 	image->kernel_buf = NULL;
149 
150 	vfree(image->initrd_buf);
151 	image->initrd_buf = NULL;
152 
153 	kfree(image->cmdline_buf);
154 	image->cmdline_buf = NULL;
155 
156 	vfree(pi->purgatory_buf);
157 	pi->purgatory_buf = NULL;
158 
159 	vfree(pi->sechdrs);
160 	pi->sechdrs = NULL;
161 
162 	/* See if architecture has anything to cleanup post load */
163 	arch_kimage_file_post_load_cleanup(image);
164 
165 	/*
166 	 * Above call should have called into bootloader to free up
167 	 * any data stored in kimage->image_loader_data. It should
168 	 * be ok now to free it up.
169 	 */
170 	kfree(image->image_loader_data);
171 	image->image_loader_data = NULL;
172 }
173 
174 /*
175  * In file mode list of segments is prepared by kernel. Copy relevant
176  * data from user space, do error checking, prepare segment list
177  */
178 static int
179 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
180 			     const char __user *cmdline_ptr,
181 			     unsigned long cmdline_len, unsigned flags)
182 {
183 	int ret = 0;
184 	void *ldata;
185 
186 	ret = copy_file_from_fd(kernel_fd, &image->kernel_buf,
187 				&image->kernel_buf_len);
188 	if (ret)
189 		return ret;
190 
191 	/* Call arch image probe handlers */
192 	ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
193 					    image->kernel_buf_len);
194 
195 	if (ret)
196 		goto out;
197 
198 #ifdef CONFIG_KEXEC_VERIFY_SIG
199 	ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
200 					   image->kernel_buf_len);
201 	if (ret) {
202 		pr_debug("kernel signature verification failed.\n");
203 		goto out;
204 	}
205 	pr_debug("kernel signature verification successful.\n");
206 #endif
207 	/* It is possible that there no initramfs is being loaded */
208 	if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
209 		ret = copy_file_from_fd(initrd_fd, &image->initrd_buf,
210 					&image->initrd_buf_len);
211 		if (ret)
212 			goto out;
213 	}
214 
215 	if (cmdline_len) {
216 		image->cmdline_buf = kzalloc(cmdline_len, GFP_KERNEL);
217 		if (!image->cmdline_buf) {
218 			ret = -ENOMEM;
219 			goto out;
220 		}
221 
222 		ret = copy_from_user(image->cmdline_buf, cmdline_ptr,
223 				     cmdline_len);
224 		if (ret) {
225 			ret = -EFAULT;
226 			goto out;
227 		}
228 
229 		image->cmdline_buf_len = cmdline_len;
230 
231 		/* command line should be a string with last byte null */
232 		if (image->cmdline_buf[cmdline_len - 1] != '\0') {
233 			ret = -EINVAL;
234 			goto out;
235 		}
236 	}
237 
238 	/* Call arch image load handlers */
239 	ldata = arch_kexec_kernel_image_load(image);
240 
241 	if (IS_ERR(ldata)) {
242 		ret = PTR_ERR(ldata);
243 		goto out;
244 	}
245 
246 	image->image_loader_data = ldata;
247 out:
248 	/* In case of error, free up all allocated memory in this function */
249 	if (ret)
250 		kimage_file_post_load_cleanup(image);
251 	return ret;
252 }
253 
254 static int
255 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
256 		       int initrd_fd, const char __user *cmdline_ptr,
257 		       unsigned long cmdline_len, unsigned long flags)
258 {
259 	int ret;
260 	struct kimage *image;
261 	bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
262 
263 	image = do_kimage_alloc_init();
264 	if (!image)
265 		return -ENOMEM;
266 
267 	image->file_mode = 1;
268 
269 	if (kexec_on_panic) {
270 		/* Enable special crash kernel control page alloc policy. */
271 		image->control_page = crashk_res.start;
272 		image->type = KEXEC_TYPE_CRASH;
273 	}
274 
275 	ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
276 					   cmdline_ptr, cmdline_len, flags);
277 	if (ret)
278 		goto out_free_image;
279 
280 	ret = sanity_check_segment_list(image);
281 	if (ret)
282 		goto out_free_post_load_bufs;
283 
284 	ret = -ENOMEM;
285 	image->control_code_page = kimage_alloc_control_pages(image,
286 					   get_order(KEXEC_CONTROL_PAGE_SIZE));
287 	if (!image->control_code_page) {
288 		pr_err("Could not allocate control_code_buffer\n");
289 		goto out_free_post_load_bufs;
290 	}
291 
292 	if (!kexec_on_panic) {
293 		image->swap_page = kimage_alloc_control_pages(image, 0);
294 		if (!image->swap_page) {
295 			pr_err("Could not allocate swap buffer\n");
296 			goto out_free_control_pages;
297 		}
298 	}
299 
300 	*rimage = image;
301 	return 0;
302 out_free_control_pages:
303 	kimage_free_page_list(&image->control_pages);
304 out_free_post_load_bufs:
305 	kimage_file_post_load_cleanup(image);
306 out_free_image:
307 	kfree(image);
308 	return ret;
309 }
310 
311 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
312 		unsigned long, cmdline_len, const char __user *, cmdline_ptr,
313 		unsigned long, flags)
314 {
315 	int ret = 0, i;
316 	struct kimage **dest_image, *image;
317 
318 	/* We only trust the superuser with rebooting the system. */
319 	if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
320 		return -EPERM;
321 
322 	/* Make sure we have a legal set of flags */
323 	if (flags != (flags & KEXEC_FILE_FLAGS))
324 		return -EINVAL;
325 
326 	image = NULL;
327 
328 	if (!mutex_trylock(&kexec_mutex))
329 		return -EBUSY;
330 
331 	dest_image = &kexec_image;
332 	if (flags & KEXEC_FILE_ON_CRASH)
333 		dest_image = &kexec_crash_image;
334 
335 	if (flags & KEXEC_FILE_UNLOAD)
336 		goto exchange;
337 
338 	/*
339 	 * In case of crash, new kernel gets loaded in reserved region. It is
340 	 * same memory where old crash kernel might be loaded. Free any
341 	 * current crash dump kernel before we corrupt it.
342 	 */
343 	if (flags & KEXEC_FILE_ON_CRASH)
344 		kimage_free(xchg(&kexec_crash_image, NULL));
345 
346 	ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
347 				     cmdline_len, flags);
348 	if (ret)
349 		goto out;
350 
351 	ret = machine_kexec_prepare(image);
352 	if (ret)
353 		goto out;
354 
355 	ret = kexec_calculate_store_digests(image);
356 	if (ret)
357 		goto out;
358 
359 	for (i = 0; i < image->nr_segments; i++) {
360 		struct kexec_segment *ksegment;
361 
362 		ksegment = &image->segment[i];
363 		pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
364 			 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
365 			 ksegment->memsz);
366 
367 		ret = kimage_load_segment(image, &image->segment[i]);
368 		if (ret)
369 			goto out;
370 	}
371 
372 	kimage_terminate(image);
373 
374 	/*
375 	 * Free up any temporary buffers allocated which are not needed
376 	 * after image has been loaded
377 	 */
378 	kimage_file_post_load_cleanup(image);
379 exchange:
380 	image = xchg(dest_image, image);
381 out:
382 	mutex_unlock(&kexec_mutex);
383 	kimage_free(image);
384 	return ret;
385 }
386 
387 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
388 				    struct kexec_buf *kbuf)
389 {
390 	struct kimage *image = kbuf->image;
391 	unsigned long temp_start, temp_end;
392 
393 	temp_end = min(end, kbuf->buf_max);
394 	temp_start = temp_end - kbuf->memsz;
395 
396 	do {
397 		/* align down start */
398 		temp_start = temp_start & (~(kbuf->buf_align - 1));
399 
400 		if (temp_start < start || temp_start < kbuf->buf_min)
401 			return 0;
402 
403 		temp_end = temp_start + kbuf->memsz - 1;
404 
405 		/*
406 		 * Make sure this does not conflict with any of existing
407 		 * segments
408 		 */
409 		if (kimage_is_destination_range(image, temp_start, temp_end)) {
410 			temp_start = temp_start - PAGE_SIZE;
411 			continue;
412 		}
413 
414 		/* We found a suitable memory range */
415 		break;
416 	} while (1);
417 
418 	/* If we are here, we found a suitable memory range */
419 	kbuf->mem = temp_start;
420 
421 	/* Success, stop navigating through remaining System RAM ranges */
422 	return 1;
423 }
424 
425 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
426 				     struct kexec_buf *kbuf)
427 {
428 	struct kimage *image = kbuf->image;
429 	unsigned long temp_start, temp_end;
430 
431 	temp_start = max(start, kbuf->buf_min);
432 
433 	do {
434 		temp_start = ALIGN(temp_start, kbuf->buf_align);
435 		temp_end = temp_start + kbuf->memsz - 1;
436 
437 		if (temp_end > end || temp_end > kbuf->buf_max)
438 			return 0;
439 		/*
440 		 * Make sure this does not conflict with any of existing
441 		 * segments
442 		 */
443 		if (kimage_is_destination_range(image, temp_start, temp_end)) {
444 			temp_start = temp_start + PAGE_SIZE;
445 			continue;
446 		}
447 
448 		/* We found a suitable memory range */
449 		break;
450 	} while (1);
451 
452 	/* If we are here, we found a suitable memory range */
453 	kbuf->mem = temp_start;
454 
455 	/* Success, stop navigating through remaining System RAM ranges */
456 	return 1;
457 }
458 
459 static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
460 {
461 	struct kexec_buf *kbuf = (struct kexec_buf *)arg;
462 	unsigned long sz = end - start + 1;
463 
464 	/* Returning 0 will take to next memory range */
465 	if (sz < kbuf->memsz)
466 		return 0;
467 
468 	if (end < kbuf->buf_min || start > kbuf->buf_max)
469 		return 0;
470 
471 	/*
472 	 * Allocate memory top down with-in ram range. Otherwise bottom up
473 	 * allocation.
474 	 */
475 	if (kbuf->top_down)
476 		return locate_mem_hole_top_down(start, end, kbuf);
477 	return locate_mem_hole_bottom_up(start, end, kbuf);
478 }
479 
480 /*
481  * Helper function for placing a buffer in a kexec segment. This assumes
482  * that kexec_mutex is held.
483  */
484 int kexec_add_buffer(struct kimage *image, char *buffer, unsigned long bufsz,
485 		     unsigned long memsz, unsigned long buf_align,
486 		     unsigned long buf_min, unsigned long buf_max,
487 		     bool top_down, unsigned long *load_addr)
488 {
489 
490 	struct kexec_segment *ksegment;
491 	struct kexec_buf buf, *kbuf;
492 	int ret;
493 
494 	/* Currently adding segment this way is allowed only in file mode */
495 	if (!image->file_mode)
496 		return -EINVAL;
497 
498 	if (image->nr_segments >= KEXEC_SEGMENT_MAX)
499 		return -EINVAL;
500 
501 	/*
502 	 * Make sure we are not trying to add buffer after allocating
503 	 * control pages. All segments need to be placed first before
504 	 * any control pages are allocated. As control page allocation
505 	 * logic goes through list of segments to make sure there are
506 	 * no destination overlaps.
507 	 */
508 	if (!list_empty(&image->control_pages)) {
509 		WARN_ON(1);
510 		return -EINVAL;
511 	}
512 
513 	memset(&buf, 0, sizeof(struct kexec_buf));
514 	kbuf = &buf;
515 	kbuf->image = image;
516 	kbuf->buffer = buffer;
517 	kbuf->bufsz = bufsz;
518 
519 	kbuf->memsz = ALIGN(memsz, PAGE_SIZE);
520 	kbuf->buf_align = max(buf_align, PAGE_SIZE);
521 	kbuf->buf_min = buf_min;
522 	kbuf->buf_max = buf_max;
523 	kbuf->top_down = top_down;
524 
525 	/* Walk the RAM ranges and allocate a suitable range for the buffer */
526 	if (image->type == KEXEC_TYPE_CRASH)
527 		ret = walk_iomem_res("Crash kernel",
528 				     IORESOURCE_MEM | IORESOURCE_BUSY,
529 				     crashk_res.start, crashk_res.end, kbuf,
530 				     locate_mem_hole_callback);
531 	else
532 		ret = walk_system_ram_res(0, -1, kbuf,
533 					  locate_mem_hole_callback);
534 	if (ret != 1) {
535 		/* A suitable memory range could not be found for buffer */
536 		return -EADDRNOTAVAIL;
537 	}
538 
539 	/* Found a suitable memory range */
540 	ksegment = &image->segment[image->nr_segments];
541 	ksegment->kbuf = kbuf->buffer;
542 	ksegment->bufsz = kbuf->bufsz;
543 	ksegment->mem = kbuf->mem;
544 	ksegment->memsz = kbuf->memsz;
545 	image->nr_segments++;
546 	*load_addr = ksegment->mem;
547 	return 0;
548 }
549 
550 /* Calculate and store the digest of segments */
551 static int kexec_calculate_store_digests(struct kimage *image)
552 {
553 	struct crypto_shash *tfm;
554 	struct shash_desc *desc;
555 	int ret = 0, i, j, zero_buf_sz, sha_region_sz;
556 	size_t desc_size, nullsz;
557 	char *digest;
558 	void *zero_buf;
559 	struct kexec_sha_region *sha_regions;
560 	struct purgatory_info *pi = &image->purgatory_info;
561 
562 	zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
563 	zero_buf_sz = PAGE_SIZE;
564 
565 	tfm = crypto_alloc_shash("sha256", 0, 0);
566 	if (IS_ERR(tfm)) {
567 		ret = PTR_ERR(tfm);
568 		goto out;
569 	}
570 
571 	desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
572 	desc = kzalloc(desc_size, GFP_KERNEL);
573 	if (!desc) {
574 		ret = -ENOMEM;
575 		goto out_free_tfm;
576 	}
577 
578 	sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
579 	sha_regions = vzalloc(sha_region_sz);
580 	if (!sha_regions)
581 		goto out_free_desc;
582 
583 	desc->tfm   = tfm;
584 	desc->flags = 0;
585 
586 	ret = crypto_shash_init(desc);
587 	if (ret < 0)
588 		goto out_free_sha_regions;
589 
590 	digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
591 	if (!digest) {
592 		ret = -ENOMEM;
593 		goto out_free_sha_regions;
594 	}
595 
596 	for (j = i = 0; i < image->nr_segments; i++) {
597 		struct kexec_segment *ksegment;
598 
599 		ksegment = &image->segment[i];
600 		/*
601 		 * Skip purgatory as it will be modified once we put digest
602 		 * info in purgatory.
603 		 */
604 		if (ksegment->kbuf == pi->purgatory_buf)
605 			continue;
606 
607 		ret = crypto_shash_update(desc, ksegment->kbuf,
608 					  ksegment->bufsz);
609 		if (ret)
610 			break;
611 
612 		/*
613 		 * Assume rest of the buffer is filled with zero and
614 		 * update digest accordingly.
615 		 */
616 		nullsz = ksegment->memsz - ksegment->bufsz;
617 		while (nullsz) {
618 			unsigned long bytes = nullsz;
619 
620 			if (bytes > zero_buf_sz)
621 				bytes = zero_buf_sz;
622 			ret = crypto_shash_update(desc, zero_buf, bytes);
623 			if (ret)
624 				break;
625 			nullsz -= bytes;
626 		}
627 
628 		if (ret)
629 			break;
630 
631 		sha_regions[j].start = ksegment->mem;
632 		sha_regions[j].len = ksegment->memsz;
633 		j++;
634 	}
635 
636 	if (!ret) {
637 		ret = crypto_shash_final(desc, digest);
638 		if (ret)
639 			goto out_free_digest;
640 		ret = kexec_purgatory_get_set_symbol(image, "sha_regions",
641 						sha_regions, sha_region_sz, 0);
642 		if (ret)
643 			goto out_free_digest;
644 
645 		ret = kexec_purgatory_get_set_symbol(image, "sha256_digest",
646 						digest, SHA256_DIGEST_SIZE, 0);
647 		if (ret)
648 			goto out_free_digest;
649 	}
650 
651 out_free_digest:
652 	kfree(digest);
653 out_free_sha_regions:
654 	vfree(sha_regions);
655 out_free_desc:
656 	kfree(desc);
657 out_free_tfm:
658 	kfree(tfm);
659 out:
660 	return ret;
661 }
662 
663 /* Actually load purgatory. Lot of code taken from kexec-tools */
664 static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
665 				  unsigned long max, int top_down)
666 {
667 	struct purgatory_info *pi = &image->purgatory_info;
668 	unsigned long align, buf_align, bss_align, buf_sz, bss_sz, bss_pad;
669 	unsigned long memsz, entry, load_addr, curr_load_addr, bss_addr, offset;
670 	unsigned char *buf_addr, *src;
671 	int i, ret = 0, entry_sidx = -1;
672 	const Elf_Shdr *sechdrs_c;
673 	Elf_Shdr *sechdrs = NULL;
674 	void *purgatory_buf = NULL;
675 
676 	/*
677 	 * sechdrs_c points to section headers in purgatory and are read
678 	 * only. No modifications allowed.
679 	 */
680 	sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
681 
682 	/*
683 	 * We can not modify sechdrs_c[] and its fields. It is read only.
684 	 * Copy it over to a local copy where one can store some temporary
685 	 * data and free it at the end. We need to modify ->sh_addr and
686 	 * ->sh_offset fields to keep track of permanent and temporary
687 	 * locations of sections.
688 	 */
689 	sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
690 	if (!sechdrs)
691 		return -ENOMEM;
692 
693 	memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
694 
695 	/*
696 	 * We seem to have multiple copies of sections. First copy is which
697 	 * is embedded in kernel in read only section. Some of these sections
698 	 * will be copied to a temporary buffer and relocated. And these
699 	 * sections will finally be copied to their final destination at
700 	 * segment load time.
701 	 *
702 	 * Use ->sh_offset to reflect section address in memory. It will
703 	 * point to original read only copy if section is not allocatable.
704 	 * Otherwise it will point to temporary copy which will be relocated.
705 	 *
706 	 * Use ->sh_addr to contain final address of the section where it
707 	 * will go during execution time.
708 	 */
709 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
710 		if (sechdrs[i].sh_type == SHT_NOBITS)
711 			continue;
712 
713 		sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
714 						sechdrs[i].sh_offset;
715 	}
716 
717 	/*
718 	 * Identify entry point section and make entry relative to section
719 	 * start.
720 	 */
721 	entry = pi->ehdr->e_entry;
722 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
723 		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
724 			continue;
725 
726 		if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
727 			continue;
728 
729 		/* Make entry section relative */
730 		if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
731 		    ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
732 		     pi->ehdr->e_entry)) {
733 			entry_sidx = i;
734 			entry -= sechdrs[i].sh_addr;
735 			break;
736 		}
737 	}
738 
739 	/* Determine how much memory is needed to load relocatable object. */
740 	buf_align = 1;
741 	bss_align = 1;
742 	buf_sz = 0;
743 	bss_sz = 0;
744 
745 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
746 		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
747 			continue;
748 
749 		align = sechdrs[i].sh_addralign;
750 		if (sechdrs[i].sh_type != SHT_NOBITS) {
751 			if (buf_align < align)
752 				buf_align = align;
753 			buf_sz = ALIGN(buf_sz, align);
754 			buf_sz += sechdrs[i].sh_size;
755 		} else {
756 			/* bss section */
757 			if (bss_align < align)
758 				bss_align = align;
759 			bss_sz = ALIGN(bss_sz, align);
760 			bss_sz += sechdrs[i].sh_size;
761 		}
762 	}
763 
764 	/* Determine the bss padding required to align bss properly */
765 	bss_pad = 0;
766 	if (buf_sz & (bss_align - 1))
767 		bss_pad = bss_align - (buf_sz & (bss_align - 1));
768 
769 	memsz = buf_sz + bss_pad + bss_sz;
770 
771 	/* Allocate buffer for purgatory */
772 	purgatory_buf = vzalloc(buf_sz);
773 	if (!purgatory_buf) {
774 		ret = -ENOMEM;
775 		goto out;
776 	}
777 
778 	if (buf_align < bss_align)
779 		buf_align = bss_align;
780 
781 	/* Add buffer to segment list */
782 	ret = kexec_add_buffer(image, purgatory_buf, buf_sz, memsz,
783 				buf_align, min, max, top_down,
784 				&pi->purgatory_load_addr);
785 	if (ret)
786 		goto out;
787 
788 	/* Load SHF_ALLOC sections */
789 	buf_addr = purgatory_buf;
790 	load_addr = curr_load_addr = pi->purgatory_load_addr;
791 	bss_addr = load_addr + buf_sz + bss_pad;
792 
793 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
794 		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
795 			continue;
796 
797 		align = sechdrs[i].sh_addralign;
798 		if (sechdrs[i].sh_type != SHT_NOBITS) {
799 			curr_load_addr = ALIGN(curr_load_addr, align);
800 			offset = curr_load_addr - load_addr;
801 			/* We already modifed ->sh_offset to keep src addr */
802 			src = (char *) sechdrs[i].sh_offset;
803 			memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
804 
805 			/* Store load address and source address of section */
806 			sechdrs[i].sh_addr = curr_load_addr;
807 
808 			/*
809 			 * This section got copied to temporary buffer. Update
810 			 * ->sh_offset accordingly.
811 			 */
812 			sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
813 
814 			/* Advance to the next address */
815 			curr_load_addr += sechdrs[i].sh_size;
816 		} else {
817 			bss_addr = ALIGN(bss_addr, align);
818 			sechdrs[i].sh_addr = bss_addr;
819 			bss_addr += sechdrs[i].sh_size;
820 		}
821 	}
822 
823 	/* Update entry point based on load address of text section */
824 	if (entry_sidx >= 0)
825 		entry += sechdrs[entry_sidx].sh_addr;
826 
827 	/* Make kernel jump to purgatory after shutdown */
828 	image->start = entry;
829 
830 	/* Used later to get/set symbol values */
831 	pi->sechdrs = sechdrs;
832 
833 	/*
834 	 * Used later to identify which section is purgatory and skip it
835 	 * from checksumming.
836 	 */
837 	pi->purgatory_buf = purgatory_buf;
838 	return ret;
839 out:
840 	vfree(sechdrs);
841 	vfree(purgatory_buf);
842 	return ret;
843 }
844 
845 static int kexec_apply_relocations(struct kimage *image)
846 {
847 	int i, ret;
848 	struct purgatory_info *pi = &image->purgatory_info;
849 	Elf_Shdr *sechdrs = pi->sechdrs;
850 
851 	/* Apply relocations */
852 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
853 		Elf_Shdr *section, *symtab;
854 
855 		if (sechdrs[i].sh_type != SHT_RELA &&
856 		    sechdrs[i].sh_type != SHT_REL)
857 			continue;
858 
859 		/*
860 		 * For section of type SHT_RELA/SHT_REL,
861 		 * ->sh_link contains section header index of associated
862 		 * symbol table. And ->sh_info contains section header
863 		 * index of section to which relocations apply.
864 		 */
865 		if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
866 		    sechdrs[i].sh_link >= pi->ehdr->e_shnum)
867 			return -ENOEXEC;
868 
869 		section = &sechdrs[sechdrs[i].sh_info];
870 		symtab = &sechdrs[sechdrs[i].sh_link];
871 
872 		if (!(section->sh_flags & SHF_ALLOC))
873 			continue;
874 
875 		/*
876 		 * symtab->sh_link contain section header index of associated
877 		 * string table.
878 		 */
879 		if (symtab->sh_link >= pi->ehdr->e_shnum)
880 			/* Invalid section number? */
881 			continue;
882 
883 		/*
884 		 * Respective architecture needs to provide support for applying
885 		 * relocations of type SHT_RELA/SHT_REL.
886 		 */
887 		if (sechdrs[i].sh_type == SHT_RELA)
888 			ret = arch_kexec_apply_relocations_add(pi->ehdr,
889 							       sechdrs, i);
890 		else if (sechdrs[i].sh_type == SHT_REL)
891 			ret = arch_kexec_apply_relocations(pi->ehdr,
892 							   sechdrs, i);
893 		if (ret)
894 			return ret;
895 	}
896 
897 	return 0;
898 }
899 
900 /* Load relocatable purgatory object and relocate it appropriately */
901 int kexec_load_purgatory(struct kimage *image, unsigned long min,
902 			 unsigned long max, int top_down,
903 			 unsigned long *load_addr)
904 {
905 	struct purgatory_info *pi = &image->purgatory_info;
906 	int ret;
907 
908 	if (kexec_purgatory_size <= 0)
909 		return -EINVAL;
910 
911 	if (kexec_purgatory_size < sizeof(Elf_Ehdr))
912 		return -ENOEXEC;
913 
914 	pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
915 
916 	if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
917 	    || pi->ehdr->e_type != ET_REL
918 	    || !elf_check_arch(pi->ehdr)
919 	    || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
920 		return -ENOEXEC;
921 
922 	if (pi->ehdr->e_shoff >= kexec_purgatory_size
923 	    || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
924 	    kexec_purgatory_size - pi->ehdr->e_shoff))
925 		return -ENOEXEC;
926 
927 	ret = __kexec_load_purgatory(image, min, max, top_down);
928 	if (ret)
929 		return ret;
930 
931 	ret = kexec_apply_relocations(image);
932 	if (ret)
933 		goto out;
934 
935 	*load_addr = pi->purgatory_load_addr;
936 	return 0;
937 out:
938 	vfree(pi->sechdrs);
939 	vfree(pi->purgatory_buf);
940 	return ret;
941 }
942 
943 static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
944 					    const char *name)
945 {
946 	Elf_Sym *syms;
947 	Elf_Shdr *sechdrs;
948 	Elf_Ehdr *ehdr;
949 	int i, k;
950 	const char *strtab;
951 
952 	if (!pi->sechdrs || !pi->ehdr)
953 		return NULL;
954 
955 	sechdrs = pi->sechdrs;
956 	ehdr = pi->ehdr;
957 
958 	for (i = 0; i < ehdr->e_shnum; i++) {
959 		if (sechdrs[i].sh_type != SHT_SYMTAB)
960 			continue;
961 
962 		if (sechdrs[i].sh_link >= ehdr->e_shnum)
963 			/* Invalid strtab section number */
964 			continue;
965 		strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
966 		syms = (Elf_Sym *)sechdrs[i].sh_offset;
967 
968 		/* Go through symbols for a match */
969 		for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
970 			if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
971 				continue;
972 
973 			if (strcmp(strtab + syms[k].st_name, name) != 0)
974 				continue;
975 
976 			if (syms[k].st_shndx == SHN_UNDEF ||
977 			    syms[k].st_shndx >= ehdr->e_shnum) {
978 				pr_debug("Symbol: %s has bad section index %d.\n",
979 						name, syms[k].st_shndx);
980 				return NULL;
981 			}
982 
983 			/* Found the symbol we are looking for */
984 			return &syms[k];
985 		}
986 	}
987 
988 	return NULL;
989 }
990 
991 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
992 {
993 	struct purgatory_info *pi = &image->purgatory_info;
994 	Elf_Sym *sym;
995 	Elf_Shdr *sechdr;
996 
997 	sym = kexec_purgatory_find_symbol(pi, name);
998 	if (!sym)
999 		return ERR_PTR(-EINVAL);
1000 
1001 	sechdr = &pi->sechdrs[sym->st_shndx];
1002 
1003 	/*
1004 	 * Returns the address where symbol will finally be loaded after
1005 	 * kexec_load_segment()
1006 	 */
1007 	return (void *)(sechdr->sh_addr + sym->st_value);
1008 }
1009 
1010 /*
1011  * Get or set value of a symbol. If "get_value" is true, symbol value is
1012  * returned in buf otherwise symbol value is set based on value in buf.
1013  */
1014 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1015 				   void *buf, unsigned int size, bool get_value)
1016 {
1017 	Elf_Sym *sym;
1018 	Elf_Shdr *sechdrs;
1019 	struct purgatory_info *pi = &image->purgatory_info;
1020 	char *sym_buf;
1021 
1022 	sym = kexec_purgatory_find_symbol(pi, name);
1023 	if (!sym)
1024 		return -EINVAL;
1025 
1026 	if (sym->st_size != size) {
1027 		pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1028 		       name, (unsigned long)sym->st_size, size);
1029 		return -EINVAL;
1030 	}
1031 
1032 	sechdrs = pi->sechdrs;
1033 
1034 	if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
1035 		pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1036 		       get_value ? "get" : "set");
1037 		return -EINVAL;
1038 	}
1039 
1040 	sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
1041 					sym->st_value;
1042 
1043 	if (get_value)
1044 		memcpy((void *)buf, sym_buf, size);
1045 	else
1046 		memcpy((void *)sym_buf, buf, size);
1047 
1048 	return 0;
1049 }
1050