1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * kexec: kexec_file_load system call
4  *
5  * Copyright (C) 2014 Red Hat Inc.
6  * Authors:
7  *      Vivek Goyal <vgoyal@redhat.com>
8  */
9 
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 
12 #include <linux/capability.h>
13 #include <linux/mm.h>
14 #include <linux/file.h>
15 #include <linux/slab.h>
16 #include <linux/kexec.h>
17 #include <linux/memblock.h>
18 #include <linux/mutex.h>
19 #include <linux/list.h>
20 #include <linux/fs.h>
21 #include <linux/ima.h>
22 #include <crypto/hash.h>
23 #include <crypto/sha2.h>
24 #include <linux/elf.h>
25 #include <linux/elfcore.h>
26 #include <linux/kernel.h>
27 #include <linux/kernel_read_file.h>
28 #include <linux/syscalls.h>
29 #include <linux/vmalloc.h>
30 #include "kexec_internal.h"
31 
32 static int kexec_calculate_store_digests(struct kimage *image);
33 
34 /*
35  * Currently this is the only default function that is exported as some
36  * architectures need it to do additional handlings.
37  * In the future, other default functions may be exported too if required.
38  */
kexec_image_probe_default(struct kimage * image,void * buf,unsigned long buf_len)39 int kexec_image_probe_default(struct kimage *image, void *buf,
40 			      unsigned long buf_len)
41 {
42 	const struct kexec_file_ops * const *fops;
43 	int ret = -ENOEXEC;
44 
45 	for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
46 		ret = (*fops)->probe(buf, buf_len);
47 		if (!ret) {
48 			image->fops = *fops;
49 			return ret;
50 		}
51 	}
52 
53 	return ret;
54 }
55 
56 /* Architectures can provide this probe function */
arch_kexec_kernel_image_probe(struct kimage * image,void * buf,unsigned long buf_len)57 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
58 					 unsigned long buf_len)
59 {
60 	return kexec_image_probe_default(image, buf, buf_len);
61 }
62 
kexec_image_load_default(struct kimage * image)63 static void *kexec_image_load_default(struct kimage *image)
64 {
65 	if (!image->fops || !image->fops->load)
66 		return ERR_PTR(-ENOEXEC);
67 
68 	return image->fops->load(image, image->kernel_buf,
69 				 image->kernel_buf_len, image->initrd_buf,
70 				 image->initrd_buf_len, image->cmdline_buf,
71 				 image->cmdline_buf_len);
72 }
73 
arch_kexec_kernel_image_load(struct kimage * image)74 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
75 {
76 	return kexec_image_load_default(image);
77 }
78 
kexec_image_post_load_cleanup_default(struct kimage * image)79 int kexec_image_post_load_cleanup_default(struct kimage *image)
80 {
81 	if (!image->fops || !image->fops->cleanup)
82 		return 0;
83 
84 	return image->fops->cleanup(image->image_loader_data);
85 }
86 
arch_kimage_file_post_load_cleanup(struct kimage * image)87 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
88 {
89 	return kexec_image_post_load_cleanup_default(image);
90 }
91 
92 #ifdef CONFIG_KEXEC_SIG
kexec_image_verify_sig_default(struct kimage * image,void * buf,unsigned long buf_len)93 static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
94 					  unsigned long buf_len)
95 {
96 	if (!image->fops || !image->fops->verify_sig) {
97 		pr_debug("kernel loader does not support signature verification.\n");
98 		return -EKEYREJECTED;
99 	}
100 
101 	return image->fops->verify_sig(buf, buf_len);
102 }
103 
arch_kexec_kernel_verify_sig(struct kimage * image,void * buf,unsigned long buf_len)104 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
105 					unsigned long buf_len)
106 {
107 	return kexec_image_verify_sig_default(image, buf, buf_len);
108 }
109 #endif
110 
111 /*
112  * arch_kexec_apply_relocations_add - apply relocations of type RELA
113  * @pi:		Purgatory to be relocated.
114  * @section:	Section relocations applying to.
115  * @relsec:	Section containing RELAs.
116  * @symtab:	Corresponding symtab.
117  *
118  * Return: 0 on success, negative errno on error.
119  */
120 int __weak
arch_kexec_apply_relocations_add(struct purgatory_info * pi,Elf_Shdr * section,const Elf_Shdr * relsec,const Elf_Shdr * symtab)121 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section,
122 				 const Elf_Shdr *relsec, const Elf_Shdr *symtab)
123 {
124 	pr_err("RELA relocation unsupported.\n");
125 	return -ENOEXEC;
126 }
127 
128 /*
129  * arch_kexec_apply_relocations - apply relocations of type REL
130  * @pi:		Purgatory to be relocated.
131  * @section:	Section relocations applying to.
132  * @relsec:	Section containing RELs.
133  * @symtab:	Corresponding symtab.
134  *
135  * Return: 0 on success, negative errno on error.
136  */
137 int __weak
arch_kexec_apply_relocations(struct purgatory_info * pi,Elf_Shdr * section,const Elf_Shdr * relsec,const Elf_Shdr * symtab)138 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section,
139 			     const Elf_Shdr *relsec, const Elf_Shdr *symtab)
140 {
141 	pr_err("REL relocation unsupported.\n");
142 	return -ENOEXEC;
143 }
144 
145 /*
146  * Free up memory used by kernel, initrd, and command line. This is temporary
147  * memory allocation which is not needed any more after these buffers have
148  * been loaded into separate segments and have been copied elsewhere.
149  */
kimage_file_post_load_cleanup(struct kimage * image)150 void kimage_file_post_load_cleanup(struct kimage *image)
151 {
152 	struct purgatory_info *pi = &image->purgatory_info;
153 
154 	vfree(image->kernel_buf);
155 	image->kernel_buf = NULL;
156 
157 	vfree(image->initrd_buf);
158 	image->initrd_buf = NULL;
159 
160 	kfree(image->cmdline_buf);
161 	image->cmdline_buf = NULL;
162 
163 	vfree(pi->purgatory_buf);
164 	pi->purgatory_buf = NULL;
165 
166 	vfree(pi->sechdrs);
167 	pi->sechdrs = NULL;
168 
169 #ifdef CONFIG_IMA_KEXEC
170 	vfree(image->ima_buffer);
171 	image->ima_buffer = NULL;
172 #endif /* CONFIG_IMA_KEXEC */
173 
174 	/* See if architecture has anything to cleanup post load */
175 	arch_kimage_file_post_load_cleanup(image);
176 
177 	/*
178 	 * Above call should have called into bootloader to free up
179 	 * any data stored in kimage->image_loader_data. It should
180 	 * be ok now to free it up.
181 	 */
182 	kfree(image->image_loader_data);
183 	image->image_loader_data = NULL;
184 }
185 
186 #ifdef CONFIG_KEXEC_SIG
187 static int
kimage_validate_signature(struct kimage * image)188 kimage_validate_signature(struct kimage *image)
189 {
190 	int ret;
191 
192 	ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
193 					   image->kernel_buf_len);
194 	if (ret) {
195 
196 		if (IS_ENABLED(CONFIG_KEXEC_SIG_FORCE)) {
197 			pr_notice("Enforced kernel signature verification failed (%d).\n", ret);
198 			return ret;
199 		}
200 
201 		/*
202 		 * If IMA is guaranteed to appraise a signature on the kexec
203 		 * image, permit it even if the kernel is otherwise locked
204 		 * down.
205 		 */
206 		if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
207 		    security_locked_down(LOCKDOWN_KEXEC))
208 			return -EPERM;
209 
210 		pr_debug("kernel signature verification failed (%d).\n", ret);
211 	}
212 
213 	return 0;
214 }
215 #endif
216 
217 /*
218  * In file mode list of segments is prepared by kernel. Copy relevant
219  * data from user space, do error checking, prepare segment list
220  */
221 static int
kimage_file_prepare_segments(struct kimage * image,int kernel_fd,int initrd_fd,const char __user * cmdline_ptr,unsigned long cmdline_len,unsigned flags)222 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
223 			     const char __user *cmdline_ptr,
224 			     unsigned long cmdline_len, unsigned flags)
225 {
226 	int ret;
227 	void *ldata;
228 
229 	ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf,
230 				       INT_MAX, NULL, READING_KEXEC_IMAGE);
231 	if (ret < 0)
232 		return ret;
233 	image->kernel_buf_len = ret;
234 
235 	/* Call arch image probe handlers */
236 	ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
237 					    image->kernel_buf_len);
238 	if (ret)
239 		goto out;
240 
241 #ifdef CONFIG_KEXEC_SIG
242 	ret = kimage_validate_signature(image);
243 
244 	if (ret)
245 		goto out;
246 #endif
247 	/* It is possible that there no initramfs is being loaded */
248 	if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
249 		ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf,
250 					       INT_MAX, NULL,
251 					       READING_KEXEC_INITRAMFS);
252 		if (ret < 0)
253 			goto out;
254 		image->initrd_buf_len = ret;
255 		ret = 0;
256 	}
257 
258 	if (cmdline_len) {
259 		image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
260 		if (IS_ERR(image->cmdline_buf)) {
261 			ret = PTR_ERR(image->cmdline_buf);
262 			image->cmdline_buf = NULL;
263 			goto out;
264 		}
265 
266 		image->cmdline_buf_len = cmdline_len;
267 
268 		/* command line should be a string with last byte null */
269 		if (image->cmdline_buf[cmdline_len - 1] != '\0') {
270 			ret = -EINVAL;
271 			goto out;
272 		}
273 
274 		ima_kexec_cmdline(kernel_fd, image->cmdline_buf,
275 				  image->cmdline_buf_len - 1);
276 	}
277 
278 	/* IMA needs to pass the measurement list to the next kernel. */
279 	ima_add_kexec_buffer(image);
280 
281 	/* Call arch image load handlers */
282 	ldata = arch_kexec_kernel_image_load(image);
283 
284 	if (IS_ERR(ldata)) {
285 		ret = PTR_ERR(ldata);
286 		goto out;
287 	}
288 
289 	image->image_loader_data = ldata;
290 out:
291 	/* In case of error, free up all allocated memory in this function */
292 	if (ret)
293 		kimage_file_post_load_cleanup(image);
294 	return ret;
295 }
296 
297 static int
kimage_file_alloc_init(struct kimage ** rimage,int kernel_fd,int initrd_fd,const char __user * cmdline_ptr,unsigned long cmdline_len,unsigned long flags)298 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
299 		       int initrd_fd, const char __user *cmdline_ptr,
300 		       unsigned long cmdline_len, unsigned long flags)
301 {
302 	int ret;
303 	struct kimage *image;
304 	bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
305 
306 	image = do_kimage_alloc_init();
307 	if (!image)
308 		return -ENOMEM;
309 
310 	image->file_mode = 1;
311 
312 	if (kexec_on_panic) {
313 		/* Enable special crash kernel control page alloc policy. */
314 		image->control_page = crashk_res.start;
315 		image->type = KEXEC_TYPE_CRASH;
316 	}
317 
318 	ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
319 					   cmdline_ptr, cmdline_len, flags);
320 	if (ret)
321 		goto out_free_image;
322 
323 	ret = sanity_check_segment_list(image);
324 	if (ret)
325 		goto out_free_post_load_bufs;
326 
327 	ret = -ENOMEM;
328 	image->control_code_page = kimage_alloc_control_pages(image,
329 					   get_order(KEXEC_CONTROL_PAGE_SIZE));
330 	if (!image->control_code_page) {
331 		pr_err("Could not allocate control_code_buffer\n");
332 		goto out_free_post_load_bufs;
333 	}
334 
335 	if (!kexec_on_panic) {
336 		image->swap_page = kimage_alloc_control_pages(image, 0);
337 		if (!image->swap_page) {
338 			pr_err("Could not allocate swap buffer\n");
339 			goto out_free_control_pages;
340 		}
341 	}
342 
343 	*rimage = image;
344 	return 0;
345 out_free_control_pages:
346 	kimage_free_page_list(&image->control_pages);
347 out_free_post_load_bufs:
348 	kimage_file_post_load_cleanup(image);
349 out_free_image:
350 	kfree(image);
351 	return ret;
352 }
353 
SYSCALL_DEFINE5(kexec_file_load,int,kernel_fd,int,initrd_fd,unsigned long,cmdline_len,const char __user *,cmdline_ptr,unsigned long,flags)354 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
355 		unsigned long, cmdline_len, const char __user *, cmdline_ptr,
356 		unsigned long, flags)
357 {
358 	int ret = 0, i;
359 	struct kimage **dest_image, *image;
360 
361 	/* We only trust the superuser with rebooting the system. */
362 	if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
363 		return -EPERM;
364 
365 	/* Make sure we have a legal set of flags */
366 	if (flags != (flags & KEXEC_FILE_FLAGS))
367 		return -EINVAL;
368 
369 	image = NULL;
370 
371 	if (!mutex_trylock(&kexec_mutex))
372 		return -EBUSY;
373 
374 	dest_image = &kexec_image;
375 	if (flags & KEXEC_FILE_ON_CRASH) {
376 		dest_image = &kexec_crash_image;
377 		if (kexec_crash_image)
378 			arch_kexec_unprotect_crashkres();
379 	}
380 
381 	if (flags & KEXEC_FILE_UNLOAD)
382 		goto exchange;
383 
384 	/*
385 	 * In case of crash, new kernel gets loaded in reserved region. It is
386 	 * same memory where old crash kernel might be loaded. Free any
387 	 * current crash dump kernel before we corrupt it.
388 	 */
389 	if (flags & KEXEC_FILE_ON_CRASH)
390 		kimage_free(xchg(&kexec_crash_image, NULL));
391 
392 	ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
393 				     cmdline_len, flags);
394 	if (ret)
395 		goto out;
396 
397 	ret = machine_kexec_prepare(image);
398 	if (ret)
399 		goto out;
400 
401 	/*
402 	 * Some architecture(like S390) may touch the crash memory before
403 	 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
404 	 */
405 	ret = kimage_crash_copy_vmcoreinfo(image);
406 	if (ret)
407 		goto out;
408 
409 	ret = kexec_calculate_store_digests(image);
410 	if (ret)
411 		goto out;
412 
413 	for (i = 0; i < image->nr_segments; i++) {
414 		struct kexec_segment *ksegment;
415 
416 		ksegment = &image->segment[i];
417 		pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
418 			 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
419 			 ksegment->memsz);
420 
421 		ret = kimage_load_segment(image, &image->segment[i]);
422 		if (ret)
423 			goto out;
424 	}
425 
426 	kimage_terminate(image);
427 
428 	ret = machine_kexec_post_load(image);
429 	if (ret)
430 		goto out;
431 
432 	/*
433 	 * Free up any temporary buffers allocated which are not needed
434 	 * after image has been loaded
435 	 */
436 	kimage_file_post_load_cleanup(image);
437 exchange:
438 	image = xchg(dest_image, image);
439 out:
440 	if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
441 		arch_kexec_protect_crashkres();
442 
443 	mutex_unlock(&kexec_mutex);
444 	kimage_free(image);
445 	return ret;
446 }
447 
locate_mem_hole_top_down(unsigned long start,unsigned long end,struct kexec_buf * kbuf)448 static int locate_mem_hole_top_down(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_end = min(end, kbuf->buf_max);
455 	temp_start = temp_end - kbuf->memsz;
456 
457 	do {
458 		/* align down start */
459 		temp_start = temp_start & (~(kbuf->buf_align - 1));
460 
461 		if (temp_start < start || temp_start < kbuf->buf_min)
462 			return 0;
463 
464 		temp_end = temp_start + kbuf->memsz - 1;
465 
466 		/*
467 		 * Make sure this does not conflict with any of existing
468 		 * segments
469 		 */
470 		if (kimage_is_destination_range(image, temp_start, temp_end)) {
471 			temp_start = temp_start - PAGE_SIZE;
472 			continue;
473 		}
474 
475 		/* We found a suitable memory range */
476 		break;
477 	} while (1);
478 
479 	/* If we are here, we found a suitable memory range */
480 	kbuf->mem = temp_start;
481 
482 	/* Success, stop navigating through remaining System RAM ranges */
483 	return 1;
484 }
485 
locate_mem_hole_bottom_up(unsigned long start,unsigned long end,struct kexec_buf * kbuf)486 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
487 				     struct kexec_buf *kbuf)
488 {
489 	struct kimage *image = kbuf->image;
490 	unsigned long temp_start, temp_end;
491 
492 	temp_start = max(start, kbuf->buf_min);
493 
494 	do {
495 		temp_start = ALIGN(temp_start, kbuf->buf_align);
496 		temp_end = temp_start + kbuf->memsz - 1;
497 
498 		if (temp_end > end || temp_end > kbuf->buf_max)
499 			return 0;
500 		/*
501 		 * Make sure this does not conflict with any of existing
502 		 * segments
503 		 */
504 		if (kimage_is_destination_range(image, temp_start, temp_end)) {
505 			temp_start = temp_start + PAGE_SIZE;
506 			continue;
507 		}
508 
509 		/* We found a suitable memory range */
510 		break;
511 	} while (1);
512 
513 	/* If we are here, we found a suitable memory range */
514 	kbuf->mem = temp_start;
515 
516 	/* Success, stop navigating through remaining System RAM ranges */
517 	return 1;
518 }
519 
locate_mem_hole_callback(struct resource * res,void * arg)520 static int locate_mem_hole_callback(struct resource *res, void *arg)
521 {
522 	struct kexec_buf *kbuf = (struct kexec_buf *)arg;
523 	u64 start = res->start, end = res->end;
524 	unsigned long sz = end - start + 1;
525 
526 	/* Returning 0 will take to next memory range */
527 
528 	/* Don't use memory that will be detected and handled by a driver. */
529 	if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
530 		return 0;
531 
532 	if (sz < kbuf->memsz)
533 		return 0;
534 
535 	if (end < kbuf->buf_min || start > kbuf->buf_max)
536 		return 0;
537 
538 	/*
539 	 * Allocate memory top down with-in ram range. Otherwise bottom up
540 	 * allocation.
541 	 */
542 	if (kbuf->top_down)
543 		return locate_mem_hole_top_down(start, end, kbuf);
544 	return locate_mem_hole_bottom_up(start, end, kbuf);
545 }
546 
547 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
kexec_walk_memblock(struct kexec_buf * kbuf,int (* func)(struct resource *,void *))548 static int kexec_walk_memblock(struct kexec_buf *kbuf,
549 			       int (*func)(struct resource *, void *))
550 {
551 	int ret = 0;
552 	u64 i;
553 	phys_addr_t mstart, mend;
554 	struct resource res = { };
555 
556 	if (kbuf->image->type == KEXEC_TYPE_CRASH)
557 		return func(&crashk_res, kbuf);
558 
559 	if (kbuf->top_down) {
560 		for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
561 						&mstart, &mend, NULL) {
562 			/*
563 			 * In memblock, end points to the first byte after the
564 			 * range while in kexec, end points to the last byte
565 			 * in the range.
566 			 */
567 			res.start = mstart;
568 			res.end = mend - 1;
569 			ret = func(&res, kbuf);
570 			if (ret)
571 				break;
572 		}
573 	} else {
574 		for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
575 					&mstart, &mend, NULL) {
576 			/*
577 			 * In memblock, end points to the first byte after the
578 			 * range while in kexec, end points to the last byte
579 			 * in the range.
580 			 */
581 			res.start = mstart;
582 			res.end = mend - 1;
583 			ret = func(&res, kbuf);
584 			if (ret)
585 				break;
586 		}
587 	}
588 
589 	return ret;
590 }
591 #else
kexec_walk_memblock(struct kexec_buf * kbuf,int (* func)(struct resource *,void *))592 static int kexec_walk_memblock(struct kexec_buf *kbuf,
593 			       int (*func)(struct resource *, void *))
594 {
595 	return 0;
596 }
597 #endif
598 
599 /**
600  * kexec_walk_resources - call func(data) on free memory regions
601  * @kbuf:	Context info for the search. Also passed to @func.
602  * @func:	Function to call for each memory region.
603  *
604  * Return: The memory walk will stop when func returns a non-zero value
605  * and that value will be returned. If all free regions are visited without
606  * func returning non-zero, then zero will be returned.
607  */
kexec_walk_resources(struct kexec_buf * kbuf,int (* func)(struct resource *,void *))608 static int kexec_walk_resources(struct kexec_buf *kbuf,
609 				int (*func)(struct resource *, void *))
610 {
611 	if (kbuf->image->type == KEXEC_TYPE_CRASH)
612 		return walk_iomem_res_desc(crashk_res.desc,
613 					   IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
614 					   crashk_res.start, crashk_res.end,
615 					   kbuf, func);
616 	else
617 		return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
618 }
619 
620 /**
621  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
622  * @kbuf:	Parameters for the memory search.
623  *
624  * On success, kbuf->mem will have the start address of the memory region found.
625  *
626  * Return: 0 on success, negative errno on error.
627  */
kexec_locate_mem_hole(struct kexec_buf * kbuf)628 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
629 {
630 	int ret;
631 
632 	/* Arch knows where to place */
633 	if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
634 		return 0;
635 
636 	if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
637 		ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
638 	else
639 		ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
640 
641 	return ret == 1 ? 0 : -EADDRNOTAVAIL;
642 }
643 
644 /**
645  * arch_kexec_locate_mem_hole - Find free memory to place the segments.
646  * @kbuf:                       Parameters for the memory search.
647  *
648  * On success, kbuf->mem will have the start address of the memory region found.
649  *
650  * Return: 0 on success, negative errno on error.
651  */
arch_kexec_locate_mem_hole(struct kexec_buf * kbuf)652 int __weak arch_kexec_locate_mem_hole(struct kexec_buf *kbuf)
653 {
654 	return kexec_locate_mem_hole(kbuf);
655 }
656 
657 /**
658  * kexec_add_buffer - place a buffer in a kexec segment
659  * @kbuf:	Buffer contents and memory parameters.
660  *
661  * This function assumes that kexec_mutex is held.
662  * On successful return, @kbuf->mem will have the physical address of
663  * the buffer in memory.
664  *
665  * Return: 0 on success, negative errno on error.
666  */
kexec_add_buffer(struct kexec_buf * kbuf)667 int kexec_add_buffer(struct kexec_buf *kbuf)
668 {
669 	struct kexec_segment *ksegment;
670 	int ret;
671 
672 	/* Currently adding segment this way is allowed only in file mode */
673 	if (!kbuf->image->file_mode)
674 		return -EINVAL;
675 
676 	if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
677 		return -EINVAL;
678 
679 	/*
680 	 * Make sure we are not trying to add buffer after allocating
681 	 * control pages. All segments need to be placed first before
682 	 * any control pages are allocated. As control page allocation
683 	 * logic goes through list of segments to make sure there are
684 	 * no destination overlaps.
685 	 */
686 	if (!list_empty(&kbuf->image->control_pages)) {
687 		WARN_ON(1);
688 		return -EINVAL;
689 	}
690 
691 	/* Ensure minimum alignment needed for segments. */
692 	kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
693 	kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
694 
695 	/* Walk the RAM ranges and allocate a suitable range for the buffer */
696 	ret = arch_kexec_locate_mem_hole(kbuf);
697 	if (ret)
698 		return ret;
699 
700 	/* Found a suitable memory range */
701 	ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
702 	ksegment->kbuf = kbuf->buffer;
703 	ksegment->bufsz = kbuf->bufsz;
704 	ksegment->mem = kbuf->mem;
705 	ksegment->memsz = kbuf->memsz;
706 	kbuf->image->nr_segments++;
707 	return 0;
708 }
709 
710 /* Calculate and store the digest of segments */
kexec_calculate_store_digests(struct kimage * image)711 static int kexec_calculate_store_digests(struct kimage *image)
712 {
713 	struct crypto_shash *tfm;
714 	struct shash_desc *desc;
715 	int ret = 0, i, j, zero_buf_sz, sha_region_sz;
716 	size_t desc_size, nullsz;
717 	char *digest;
718 	void *zero_buf;
719 	struct kexec_sha_region *sha_regions;
720 	struct purgatory_info *pi = &image->purgatory_info;
721 
722 	if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
723 		return 0;
724 
725 	zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
726 	zero_buf_sz = PAGE_SIZE;
727 
728 	tfm = crypto_alloc_shash("sha256", 0, 0);
729 	if (IS_ERR(tfm)) {
730 		ret = PTR_ERR(tfm);
731 		goto out;
732 	}
733 
734 	desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
735 	desc = kzalloc(desc_size, GFP_KERNEL);
736 	if (!desc) {
737 		ret = -ENOMEM;
738 		goto out_free_tfm;
739 	}
740 
741 	sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
742 	sha_regions = vzalloc(sha_region_sz);
743 	if (!sha_regions) {
744 		ret = -ENOMEM;
745 		goto out_free_desc;
746 	}
747 
748 	desc->tfm   = tfm;
749 
750 	ret = crypto_shash_init(desc);
751 	if (ret < 0)
752 		goto out_free_sha_regions;
753 
754 	digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
755 	if (!digest) {
756 		ret = -ENOMEM;
757 		goto out_free_sha_regions;
758 	}
759 
760 	for (j = i = 0; i < image->nr_segments; i++) {
761 		struct kexec_segment *ksegment;
762 
763 		ksegment = &image->segment[i];
764 		/*
765 		 * Skip purgatory as it will be modified once we put digest
766 		 * info in purgatory.
767 		 */
768 		if (ksegment->kbuf == pi->purgatory_buf)
769 			continue;
770 
771 		ret = crypto_shash_update(desc, ksegment->kbuf,
772 					  ksegment->bufsz);
773 		if (ret)
774 			break;
775 
776 		/*
777 		 * Assume rest of the buffer is filled with zero and
778 		 * update digest accordingly.
779 		 */
780 		nullsz = ksegment->memsz - ksegment->bufsz;
781 		while (nullsz) {
782 			unsigned long bytes = nullsz;
783 
784 			if (bytes > zero_buf_sz)
785 				bytes = zero_buf_sz;
786 			ret = crypto_shash_update(desc, zero_buf, bytes);
787 			if (ret)
788 				break;
789 			nullsz -= bytes;
790 		}
791 
792 		if (ret)
793 			break;
794 
795 		sha_regions[j].start = ksegment->mem;
796 		sha_regions[j].len = ksegment->memsz;
797 		j++;
798 	}
799 
800 	if (!ret) {
801 		ret = crypto_shash_final(desc, digest);
802 		if (ret)
803 			goto out_free_digest;
804 		ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
805 						     sha_regions, sha_region_sz, 0);
806 		if (ret)
807 			goto out_free_digest;
808 
809 		ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
810 						     digest, SHA256_DIGEST_SIZE, 0);
811 		if (ret)
812 			goto out_free_digest;
813 	}
814 
815 out_free_digest:
816 	kfree(digest);
817 out_free_sha_regions:
818 	vfree(sha_regions);
819 out_free_desc:
820 	kfree(desc);
821 out_free_tfm:
822 	kfree(tfm);
823 out:
824 	return ret;
825 }
826 
827 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
828 /*
829  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
830  * @pi:		Purgatory to be loaded.
831  * @kbuf:	Buffer to setup.
832  *
833  * Allocates the memory needed for the buffer. Caller is responsible to free
834  * the memory after use.
835  *
836  * Return: 0 on success, negative errno on error.
837  */
kexec_purgatory_setup_kbuf(struct purgatory_info * pi,struct kexec_buf * kbuf)838 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
839 				      struct kexec_buf *kbuf)
840 {
841 	const Elf_Shdr *sechdrs;
842 	unsigned long bss_align;
843 	unsigned long bss_sz;
844 	unsigned long align;
845 	int i, ret;
846 
847 	sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
848 	kbuf->buf_align = bss_align = 1;
849 	kbuf->bufsz = bss_sz = 0;
850 
851 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
852 		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
853 			continue;
854 
855 		align = sechdrs[i].sh_addralign;
856 		if (sechdrs[i].sh_type != SHT_NOBITS) {
857 			if (kbuf->buf_align < align)
858 				kbuf->buf_align = align;
859 			kbuf->bufsz = ALIGN(kbuf->bufsz, align);
860 			kbuf->bufsz += sechdrs[i].sh_size;
861 		} else {
862 			if (bss_align < align)
863 				bss_align = align;
864 			bss_sz = ALIGN(bss_sz, align);
865 			bss_sz += sechdrs[i].sh_size;
866 		}
867 	}
868 	kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
869 	kbuf->memsz = kbuf->bufsz + bss_sz;
870 	if (kbuf->buf_align < bss_align)
871 		kbuf->buf_align = bss_align;
872 
873 	kbuf->buffer = vzalloc(kbuf->bufsz);
874 	if (!kbuf->buffer)
875 		return -ENOMEM;
876 	pi->purgatory_buf = kbuf->buffer;
877 
878 	ret = kexec_add_buffer(kbuf);
879 	if (ret)
880 		goto out;
881 
882 	return 0;
883 out:
884 	vfree(pi->purgatory_buf);
885 	pi->purgatory_buf = NULL;
886 	return ret;
887 }
888 
889 /*
890  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
891  * @pi:		Purgatory to be loaded.
892  * @kbuf:	Buffer prepared to store purgatory.
893  *
894  * Allocates the memory needed for the buffer. Caller is responsible to free
895  * the memory after use.
896  *
897  * Return: 0 on success, negative errno on error.
898  */
kexec_purgatory_setup_sechdrs(struct purgatory_info * pi,struct kexec_buf * kbuf)899 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
900 					 struct kexec_buf *kbuf)
901 {
902 	unsigned long bss_addr;
903 	unsigned long offset;
904 	Elf_Shdr *sechdrs;
905 	int i;
906 
907 	/*
908 	 * The section headers in kexec_purgatory are read-only. In order to
909 	 * have them modifiable make a temporary copy.
910 	 */
911 	sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
912 	if (!sechdrs)
913 		return -ENOMEM;
914 	memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
915 	       pi->ehdr->e_shnum * sizeof(Elf_Shdr));
916 	pi->sechdrs = sechdrs;
917 
918 	offset = 0;
919 	bss_addr = kbuf->mem + kbuf->bufsz;
920 	kbuf->image->start = pi->ehdr->e_entry;
921 
922 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
923 		unsigned long align;
924 		void *src, *dst;
925 
926 		if (!(sechdrs[i].sh_flags & SHF_ALLOC))
927 			continue;
928 
929 		align = sechdrs[i].sh_addralign;
930 		if (sechdrs[i].sh_type == SHT_NOBITS) {
931 			bss_addr = ALIGN(bss_addr, align);
932 			sechdrs[i].sh_addr = bss_addr;
933 			bss_addr += sechdrs[i].sh_size;
934 			continue;
935 		}
936 
937 		offset = ALIGN(offset, align);
938 		if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
939 		    pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
940 		    pi->ehdr->e_entry < (sechdrs[i].sh_addr
941 					 + sechdrs[i].sh_size)) {
942 			kbuf->image->start -= sechdrs[i].sh_addr;
943 			kbuf->image->start += kbuf->mem + offset;
944 		}
945 
946 		src = (void *)pi->ehdr + sechdrs[i].sh_offset;
947 		dst = pi->purgatory_buf + offset;
948 		memcpy(dst, src, sechdrs[i].sh_size);
949 
950 		sechdrs[i].sh_addr = kbuf->mem + offset;
951 		sechdrs[i].sh_offset = offset;
952 		offset += sechdrs[i].sh_size;
953 	}
954 
955 	return 0;
956 }
957 
kexec_apply_relocations(struct kimage * image)958 static int kexec_apply_relocations(struct kimage *image)
959 {
960 	int i, ret;
961 	struct purgatory_info *pi = &image->purgatory_info;
962 	const Elf_Shdr *sechdrs;
963 
964 	sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
965 
966 	for (i = 0; i < pi->ehdr->e_shnum; i++) {
967 		const Elf_Shdr *relsec;
968 		const Elf_Shdr *symtab;
969 		Elf_Shdr *section;
970 
971 		relsec = sechdrs + i;
972 
973 		if (relsec->sh_type != SHT_RELA &&
974 		    relsec->sh_type != SHT_REL)
975 			continue;
976 
977 		/*
978 		 * For section of type SHT_RELA/SHT_REL,
979 		 * ->sh_link contains section header index of associated
980 		 * symbol table. And ->sh_info contains section header
981 		 * index of section to which relocations apply.
982 		 */
983 		if (relsec->sh_info >= pi->ehdr->e_shnum ||
984 		    relsec->sh_link >= pi->ehdr->e_shnum)
985 			return -ENOEXEC;
986 
987 		section = pi->sechdrs + relsec->sh_info;
988 		symtab = sechdrs + relsec->sh_link;
989 
990 		if (!(section->sh_flags & SHF_ALLOC))
991 			continue;
992 
993 		/*
994 		 * symtab->sh_link contain section header index of associated
995 		 * string table.
996 		 */
997 		if (symtab->sh_link >= pi->ehdr->e_shnum)
998 			/* Invalid section number? */
999 			continue;
1000 
1001 		/*
1002 		 * Respective architecture needs to provide support for applying
1003 		 * relocations of type SHT_RELA/SHT_REL.
1004 		 */
1005 		if (relsec->sh_type == SHT_RELA)
1006 			ret = arch_kexec_apply_relocations_add(pi, section,
1007 							       relsec, symtab);
1008 		else if (relsec->sh_type == SHT_REL)
1009 			ret = arch_kexec_apply_relocations(pi, section,
1010 							   relsec, symtab);
1011 		if (ret)
1012 			return ret;
1013 	}
1014 
1015 	return 0;
1016 }
1017 
1018 /*
1019  * kexec_load_purgatory - Load and relocate the purgatory object.
1020  * @image:	Image to add the purgatory to.
1021  * @kbuf:	Memory parameters to use.
1022  *
1023  * Allocates the memory needed for image->purgatory_info.sechdrs and
1024  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1025  * to free the memory after use.
1026  *
1027  * Return: 0 on success, negative errno on error.
1028  */
kexec_load_purgatory(struct kimage * image,struct kexec_buf * kbuf)1029 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1030 {
1031 	struct purgatory_info *pi = &image->purgatory_info;
1032 	int ret;
1033 
1034 	if (kexec_purgatory_size <= 0)
1035 		return -EINVAL;
1036 
1037 	pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1038 
1039 	ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1040 	if (ret)
1041 		return ret;
1042 
1043 	ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1044 	if (ret)
1045 		goto out_free_kbuf;
1046 
1047 	ret = kexec_apply_relocations(image);
1048 	if (ret)
1049 		goto out;
1050 
1051 	return 0;
1052 out:
1053 	vfree(pi->sechdrs);
1054 	pi->sechdrs = NULL;
1055 out_free_kbuf:
1056 	vfree(pi->purgatory_buf);
1057 	pi->purgatory_buf = NULL;
1058 	return ret;
1059 }
1060 
1061 /*
1062  * kexec_purgatory_find_symbol - find a symbol in the purgatory
1063  * @pi:		Purgatory to search in.
1064  * @name:	Name of the symbol.
1065  *
1066  * Return: pointer to symbol in read-only symtab on success, NULL on error.
1067  */
kexec_purgatory_find_symbol(struct purgatory_info * pi,const char * name)1068 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1069 						  const char *name)
1070 {
1071 	const Elf_Shdr *sechdrs;
1072 	const Elf_Ehdr *ehdr;
1073 	const Elf_Sym *syms;
1074 	const char *strtab;
1075 	int i, k;
1076 
1077 	if (!pi->ehdr)
1078 		return NULL;
1079 
1080 	ehdr = pi->ehdr;
1081 	sechdrs = (void *)ehdr + ehdr->e_shoff;
1082 
1083 	for (i = 0; i < ehdr->e_shnum; i++) {
1084 		if (sechdrs[i].sh_type != SHT_SYMTAB)
1085 			continue;
1086 
1087 		if (sechdrs[i].sh_link >= ehdr->e_shnum)
1088 			/* Invalid strtab section number */
1089 			continue;
1090 		strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1091 		syms = (void *)ehdr + sechdrs[i].sh_offset;
1092 
1093 		/* Go through symbols for a match */
1094 		for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1095 			if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1096 				continue;
1097 
1098 			if (strcmp(strtab + syms[k].st_name, name) != 0)
1099 				continue;
1100 
1101 			if (syms[k].st_shndx == SHN_UNDEF ||
1102 			    syms[k].st_shndx >= ehdr->e_shnum) {
1103 				pr_debug("Symbol: %s has bad section index %d.\n",
1104 						name, syms[k].st_shndx);
1105 				return NULL;
1106 			}
1107 
1108 			/* Found the symbol we are looking for */
1109 			return &syms[k];
1110 		}
1111 	}
1112 
1113 	return NULL;
1114 }
1115 
kexec_purgatory_get_symbol_addr(struct kimage * image,const char * name)1116 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1117 {
1118 	struct purgatory_info *pi = &image->purgatory_info;
1119 	const Elf_Sym *sym;
1120 	Elf_Shdr *sechdr;
1121 
1122 	sym = kexec_purgatory_find_symbol(pi, name);
1123 	if (!sym)
1124 		return ERR_PTR(-EINVAL);
1125 
1126 	sechdr = &pi->sechdrs[sym->st_shndx];
1127 
1128 	/*
1129 	 * Returns the address where symbol will finally be loaded after
1130 	 * kexec_load_segment()
1131 	 */
1132 	return (void *)(sechdr->sh_addr + sym->st_value);
1133 }
1134 
1135 /*
1136  * Get or set value of a symbol. If "get_value" is true, symbol value is
1137  * returned in buf otherwise symbol value is set based on value in buf.
1138  */
kexec_purgatory_get_set_symbol(struct kimage * image,const char * name,void * buf,unsigned int size,bool get_value)1139 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1140 				   void *buf, unsigned int size, bool get_value)
1141 {
1142 	struct purgatory_info *pi = &image->purgatory_info;
1143 	const Elf_Sym *sym;
1144 	Elf_Shdr *sec;
1145 	char *sym_buf;
1146 
1147 	sym = kexec_purgatory_find_symbol(pi, name);
1148 	if (!sym)
1149 		return -EINVAL;
1150 
1151 	if (sym->st_size != size) {
1152 		pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1153 		       name, (unsigned long)sym->st_size, size);
1154 		return -EINVAL;
1155 	}
1156 
1157 	sec = pi->sechdrs + sym->st_shndx;
1158 
1159 	if (sec->sh_type == SHT_NOBITS) {
1160 		pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1161 		       get_value ? "get" : "set");
1162 		return -EINVAL;
1163 	}
1164 
1165 	sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1166 
1167 	if (get_value)
1168 		memcpy((void *)buf, sym_buf, size);
1169 	else
1170 		memcpy((void *)sym_buf, buf, size);
1171 
1172 	return 0;
1173 }
1174 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1175 
crash_exclude_mem_range(struct crash_mem * mem,unsigned long long mstart,unsigned long long mend)1176 int crash_exclude_mem_range(struct crash_mem *mem,
1177 			    unsigned long long mstart, unsigned long long mend)
1178 {
1179 	int i, j;
1180 	unsigned long long start, end, p_start, p_end;
1181 	struct crash_mem_range temp_range = {0, 0};
1182 
1183 	for (i = 0; i < mem->nr_ranges; i++) {
1184 		start = mem->ranges[i].start;
1185 		end = mem->ranges[i].end;
1186 		p_start = mstart;
1187 		p_end = mend;
1188 
1189 		if (mstart > end || mend < start)
1190 			continue;
1191 
1192 		/* Truncate any area outside of range */
1193 		if (mstart < start)
1194 			p_start = start;
1195 		if (mend > end)
1196 			p_end = end;
1197 
1198 		/* Found completely overlapping range */
1199 		if (p_start == start && p_end == end) {
1200 			mem->ranges[i].start = 0;
1201 			mem->ranges[i].end = 0;
1202 			if (i < mem->nr_ranges - 1) {
1203 				/* Shift rest of the ranges to left */
1204 				for (j = i; j < mem->nr_ranges - 1; j++) {
1205 					mem->ranges[j].start =
1206 						mem->ranges[j+1].start;
1207 					mem->ranges[j].end =
1208 							mem->ranges[j+1].end;
1209 				}
1210 
1211 				/*
1212 				 * Continue to check if there are another overlapping ranges
1213 				 * from the current position because of shifting the above
1214 				 * mem ranges.
1215 				 */
1216 				i--;
1217 				mem->nr_ranges--;
1218 				continue;
1219 			}
1220 			mem->nr_ranges--;
1221 			return 0;
1222 		}
1223 
1224 		if (p_start > start && p_end < end) {
1225 			/* Split original range */
1226 			mem->ranges[i].end = p_start - 1;
1227 			temp_range.start = p_end + 1;
1228 			temp_range.end = end;
1229 		} else if (p_start != start)
1230 			mem->ranges[i].end = p_start - 1;
1231 		else
1232 			mem->ranges[i].start = p_end + 1;
1233 		break;
1234 	}
1235 
1236 	/* If a split happened, add the split to array */
1237 	if (!temp_range.end)
1238 		return 0;
1239 
1240 	/* Split happened */
1241 	if (i == mem->max_nr_ranges - 1)
1242 		return -ENOMEM;
1243 
1244 	/* Location where new range should go */
1245 	j = i + 1;
1246 	if (j < mem->nr_ranges) {
1247 		/* Move over all ranges one slot towards the end */
1248 		for (i = mem->nr_ranges - 1; i >= j; i--)
1249 			mem->ranges[i + 1] = mem->ranges[i];
1250 	}
1251 
1252 	mem->ranges[j].start = temp_range.start;
1253 	mem->ranges[j].end = temp_range.end;
1254 	mem->nr_ranges++;
1255 	return 0;
1256 }
1257 
crash_prepare_elf64_headers(struct crash_mem * mem,int kernel_map,void ** addr,unsigned long * sz)1258 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1259 			  void **addr, unsigned long *sz)
1260 {
1261 	Elf64_Ehdr *ehdr;
1262 	Elf64_Phdr *phdr;
1263 	unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1264 	unsigned char *buf;
1265 	unsigned int cpu, i;
1266 	unsigned long long notes_addr;
1267 	unsigned long mstart, mend;
1268 
1269 	/* extra phdr for vmcoreinfo ELF note */
1270 	nr_phdr = nr_cpus + 1;
1271 	nr_phdr += mem->nr_ranges;
1272 
1273 	/*
1274 	 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1275 	 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1276 	 * I think this is required by tools like gdb. So same physical
1277 	 * memory will be mapped in two ELF headers. One will contain kernel
1278 	 * text virtual addresses and other will have __va(physical) addresses.
1279 	 */
1280 
1281 	nr_phdr++;
1282 	elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1283 	elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1284 
1285 	buf = vzalloc(elf_sz);
1286 	if (!buf)
1287 		return -ENOMEM;
1288 
1289 	ehdr = (Elf64_Ehdr *)buf;
1290 	phdr = (Elf64_Phdr *)(ehdr + 1);
1291 	memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1292 	ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1293 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1294 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1295 	ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1296 	memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1297 	ehdr->e_type = ET_CORE;
1298 	ehdr->e_machine = ELF_ARCH;
1299 	ehdr->e_version = EV_CURRENT;
1300 	ehdr->e_phoff = sizeof(Elf64_Ehdr);
1301 	ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1302 	ehdr->e_phentsize = sizeof(Elf64_Phdr);
1303 
1304 	/* Prepare one phdr of type PT_NOTE for each present CPU */
1305 	for_each_present_cpu(cpu) {
1306 		phdr->p_type = PT_NOTE;
1307 		notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1308 		phdr->p_offset = phdr->p_paddr = notes_addr;
1309 		phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1310 		(ehdr->e_phnum)++;
1311 		phdr++;
1312 	}
1313 
1314 	/* Prepare one PT_NOTE header for vmcoreinfo */
1315 	phdr->p_type = PT_NOTE;
1316 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1317 	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1318 	(ehdr->e_phnum)++;
1319 	phdr++;
1320 
1321 	/* Prepare PT_LOAD type program header for kernel text region */
1322 	if (kernel_map) {
1323 		phdr->p_type = PT_LOAD;
1324 		phdr->p_flags = PF_R|PF_W|PF_X;
1325 		phdr->p_vaddr = (unsigned long) _text;
1326 		phdr->p_filesz = phdr->p_memsz = _end - _text;
1327 		phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1328 		ehdr->e_phnum++;
1329 		phdr++;
1330 	}
1331 
1332 	/* Go through all the ranges in mem->ranges[] and prepare phdr */
1333 	for (i = 0; i < mem->nr_ranges; i++) {
1334 		mstart = mem->ranges[i].start;
1335 		mend = mem->ranges[i].end;
1336 
1337 		phdr->p_type = PT_LOAD;
1338 		phdr->p_flags = PF_R|PF_W|PF_X;
1339 		phdr->p_offset  = mstart;
1340 
1341 		phdr->p_paddr = mstart;
1342 		phdr->p_vaddr = (unsigned long) __va(mstart);
1343 		phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1344 		phdr->p_align = 0;
1345 		ehdr->e_phnum++;
1346 		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",
1347 			phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1348 			ehdr->e_phnum, phdr->p_offset);
1349 		phdr++;
1350 	}
1351 
1352 	*addr = buf;
1353 	*sz = elf_sz;
1354 	return 0;
1355 }
1356