xref: /linux/kernel/kexec_core.c (revision 981368e1440b76f68b1ac8f5fb14e739f80ecc4e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * kexec.c - kexec system call core code.
4  * Copyright (C) 2002-2004 Eric Biederman  <ebiederm@xmission.com>
5  */
6 
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 
9 #include <linux/btf.h>
10 #include <linux/capability.h>
11 #include <linux/mm.h>
12 #include <linux/file.h>
13 #include <linux/slab.h>
14 #include <linux/fs.h>
15 #include <linux/kexec.h>
16 #include <linux/mutex.h>
17 #include <linux/list.h>
18 #include <linux/highmem.h>
19 #include <linux/syscalls.h>
20 #include <linux/reboot.h>
21 #include <linux/ioport.h>
22 #include <linux/hardirq.h>
23 #include <linux/elf.h>
24 #include <linux/elfcore.h>
25 #include <linux/utsname.h>
26 #include <linux/numa.h>
27 #include <linux/suspend.h>
28 #include <linux/device.h>
29 #include <linux/freezer.h>
30 #include <linux/panic_notifier.h>
31 #include <linux/pm.h>
32 #include <linux/cpu.h>
33 #include <linux/uaccess.h>
34 #include <linux/io.h>
35 #include <linux/console.h>
36 #include <linux/vmalloc.h>
37 #include <linux/swap.h>
38 #include <linux/syscore_ops.h>
39 #include <linux/compiler.h>
40 #include <linux/hugetlb.h>
41 #include <linux/objtool.h>
42 #include <linux/kmsg_dump.h>
43 
44 #include <asm/page.h>
45 #include <asm/sections.h>
46 
47 #include <crypto/hash.h>
48 #include "kexec_internal.h"
49 
50 atomic_t __kexec_lock = ATOMIC_INIT(0);
51 
52 /* Flag to indicate we are going to kexec a new kernel */
53 bool kexec_in_progress = false;
54 
55 
56 /* Location of the reserved area for the crash kernel */
57 struct resource crashk_res = {
58 	.name  = "Crash kernel",
59 	.start = 0,
60 	.end   = 0,
61 	.flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
62 	.desc  = IORES_DESC_CRASH_KERNEL
63 };
64 struct resource crashk_low_res = {
65 	.name  = "Crash kernel",
66 	.start = 0,
67 	.end   = 0,
68 	.flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
69 	.desc  = IORES_DESC_CRASH_KERNEL
70 };
71 
72 int kexec_should_crash(struct task_struct *p)
73 {
74 	/*
75 	 * If crash_kexec_post_notifiers is enabled, don't run
76 	 * crash_kexec() here yet, which must be run after panic
77 	 * notifiers in panic().
78 	 */
79 	if (crash_kexec_post_notifiers)
80 		return 0;
81 	/*
82 	 * There are 4 panic() calls in make_task_dead() path, each of which
83 	 * corresponds to each of these 4 conditions.
84 	 */
85 	if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
86 		return 1;
87 	return 0;
88 }
89 
90 int kexec_crash_loaded(void)
91 {
92 	return !!kexec_crash_image;
93 }
94 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
95 
96 /*
97  * When kexec transitions to the new kernel there is a one-to-one
98  * mapping between physical and virtual addresses.  On processors
99  * where you can disable the MMU this is trivial, and easy.  For
100  * others it is still a simple predictable page table to setup.
101  *
102  * In that environment kexec copies the new kernel to its final
103  * resting place.  This means I can only support memory whose
104  * physical address can fit in an unsigned long.  In particular
105  * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
106  * If the assembly stub has more restrictive requirements
107  * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
108  * defined more restrictively in <asm/kexec.h>.
109  *
110  * The code for the transition from the current kernel to the
111  * new kernel is placed in the control_code_buffer, whose size
112  * is given by KEXEC_CONTROL_PAGE_SIZE.  In the best case only a single
113  * page of memory is necessary, but some architectures require more.
114  * Because this memory must be identity mapped in the transition from
115  * virtual to physical addresses it must live in the range
116  * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
117  * modifiable.
118  *
119  * The assembly stub in the control code buffer is passed a linked list
120  * of descriptor pages detailing the source pages of the new kernel,
121  * and the destination addresses of those source pages.  As this data
122  * structure is not used in the context of the current OS, it must
123  * be self-contained.
124  *
125  * The code has been made to work with highmem pages and will use a
126  * destination page in its final resting place (if it happens
127  * to allocate it).  The end product of this is that most of the
128  * physical address space, and most of RAM can be used.
129  *
130  * Future directions include:
131  *  - allocating a page table with the control code buffer identity
132  *    mapped, to simplify machine_kexec and make kexec_on_panic more
133  *    reliable.
134  */
135 
136 /*
137  * KIMAGE_NO_DEST is an impossible destination address..., for
138  * allocating pages whose destination address we do not care about.
139  */
140 #define KIMAGE_NO_DEST (-1UL)
141 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
142 
143 static struct page *kimage_alloc_page(struct kimage *image,
144 				       gfp_t gfp_mask,
145 				       unsigned long dest);
146 
147 int sanity_check_segment_list(struct kimage *image)
148 {
149 	int i;
150 	unsigned long nr_segments = image->nr_segments;
151 	unsigned long total_pages = 0;
152 	unsigned long nr_pages = totalram_pages();
153 
154 	/*
155 	 * Verify we have good destination addresses.  The caller is
156 	 * responsible for making certain we don't attempt to load
157 	 * the new image into invalid or reserved areas of RAM.  This
158 	 * just verifies it is an address we can use.
159 	 *
160 	 * Since the kernel does everything in page size chunks ensure
161 	 * the destination addresses are page aligned.  Too many
162 	 * special cases crop of when we don't do this.  The most
163 	 * insidious is getting overlapping destination addresses
164 	 * simply because addresses are changed to page size
165 	 * granularity.
166 	 */
167 	for (i = 0; i < nr_segments; i++) {
168 		unsigned long mstart, mend;
169 
170 		mstart = image->segment[i].mem;
171 		mend   = mstart + image->segment[i].memsz;
172 		if (mstart > mend)
173 			return -EADDRNOTAVAIL;
174 		if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
175 			return -EADDRNOTAVAIL;
176 		if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
177 			return -EADDRNOTAVAIL;
178 	}
179 
180 	/* Verify our destination addresses do not overlap.
181 	 * If we alloed overlapping destination addresses
182 	 * through very weird things can happen with no
183 	 * easy explanation as one segment stops on another.
184 	 */
185 	for (i = 0; i < nr_segments; i++) {
186 		unsigned long mstart, mend;
187 		unsigned long j;
188 
189 		mstart = image->segment[i].mem;
190 		mend   = mstart + image->segment[i].memsz;
191 		for (j = 0; j < i; j++) {
192 			unsigned long pstart, pend;
193 
194 			pstart = image->segment[j].mem;
195 			pend   = pstart + image->segment[j].memsz;
196 			/* Do the segments overlap ? */
197 			if ((mend > pstart) && (mstart < pend))
198 				return -EINVAL;
199 		}
200 	}
201 
202 	/* Ensure our buffer sizes are strictly less than
203 	 * our memory sizes.  This should always be the case,
204 	 * and it is easier to check up front than to be surprised
205 	 * later on.
206 	 */
207 	for (i = 0; i < nr_segments; i++) {
208 		if (image->segment[i].bufsz > image->segment[i].memsz)
209 			return -EINVAL;
210 	}
211 
212 	/*
213 	 * Verify that no more than half of memory will be consumed. If the
214 	 * request from userspace is too large, a large amount of time will be
215 	 * wasted allocating pages, which can cause a soft lockup.
216 	 */
217 	for (i = 0; i < nr_segments; i++) {
218 		if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
219 			return -EINVAL;
220 
221 		total_pages += PAGE_COUNT(image->segment[i].memsz);
222 	}
223 
224 	if (total_pages > nr_pages / 2)
225 		return -EINVAL;
226 
227 	/*
228 	 * Verify we have good destination addresses.  Normally
229 	 * the caller is responsible for making certain we don't
230 	 * attempt to load the new image into invalid or reserved
231 	 * areas of RAM.  But crash kernels are preloaded into a
232 	 * reserved area of ram.  We must ensure the addresses
233 	 * are in the reserved area otherwise preloading the
234 	 * kernel could corrupt things.
235 	 */
236 
237 	if (image->type == KEXEC_TYPE_CRASH) {
238 		for (i = 0; i < nr_segments; i++) {
239 			unsigned long mstart, mend;
240 
241 			mstart = image->segment[i].mem;
242 			mend = mstart + image->segment[i].memsz - 1;
243 			/* Ensure we are within the crash kernel limits */
244 			if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
245 			    (mend > phys_to_boot_phys(crashk_res.end)))
246 				return -EADDRNOTAVAIL;
247 		}
248 	}
249 
250 	return 0;
251 }
252 
253 struct kimage *do_kimage_alloc_init(void)
254 {
255 	struct kimage *image;
256 
257 	/* Allocate a controlling structure */
258 	image = kzalloc(sizeof(*image), GFP_KERNEL);
259 	if (!image)
260 		return NULL;
261 
262 	image->head = 0;
263 	image->entry = &image->head;
264 	image->last_entry = &image->head;
265 	image->control_page = ~0; /* By default this does not apply */
266 	image->type = KEXEC_TYPE_DEFAULT;
267 
268 	/* Initialize the list of control pages */
269 	INIT_LIST_HEAD(&image->control_pages);
270 
271 	/* Initialize the list of destination pages */
272 	INIT_LIST_HEAD(&image->dest_pages);
273 
274 	/* Initialize the list of unusable pages */
275 	INIT_LIST_HEAD(&image->unusable_pages);
276 
277 #ifdef CONFIG_CRASH_HOTPLUG
278 	image->hp_action = KEXEC_CRASH_HP_NONE;
279 	image->elfcorehdr_index = -1;
280 	image->elfcorehdr_updated = false;
281 #endif
282 
283 	return image;
284 }
285 
286 int kimage_is_destination_range(struct kimage *image,
287 					unsigned long start,
288 					unsigned long end)
289 {
290 	unsigned long i;
291 
292 	for (i = 0; i < image->nr_segments; i++) {
293 		unsigned long mstart, mend;
294 
295 		mstart = image->segment[i].mem;
296 		mend = mstart + image->segment[i].memsz;
297 		if ((end > mstart) && (start < mend))
298 			return 1;
299 	}
300 
301 	return 0;
302 }
303 
304 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
305 {
306 	struct page *pages;
307 
308 	if (fatal_signal_pending(current))
309 		return NULL;
310 	pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
311 	if (pages) {
312 		unsigned int count, i;
313 
314 		pages->mapping = NULL;
315 		set_page_private(pages, order);
316 		count = 1 << order;
317 		for (i = 0; i < count; i++)
318 			SetPageReserved(pages + i);
319 
320 		arch_kexec_post_alloc_pages(page_address(pages), count,
321 					    gfp_mask);
322 
323 		if (gfp_mask & __GFP_ZERO)
324 			for (i = 0; i < count; i++)
325 				clear_highpage(pages + i);
326 	}
327 
328 	return pages;
329 }
330 
331 static void kimage_free_pages(struct page *page)
332 {
333 	unsigned int order, count, i;
334 
335 	order = page_private(page);
336 	count = 1 << order;
337 
338 	arch_kexec_pre_free_pages(page_address(page), count);
339 
340 	for (i = 0; i < count; i++)
341 		ClearPageReserved(page + i);
342 	__free_pages(page, order);
343 }
344 
345 void kimage_free_page_list(struct list_head *list)
346 {
347 	struct page *page, *next;
348 
349 	list_for_each_entry_safe(page, next, list, lru) {
350 		list_del(&page->lru);
351 		kimage_free_pages(page);
352 	}
353 }
354 
355 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
356 							unsigned int order)
357 {
358 	/* Control pages are special, they are the intermediaries
359 	 * that are needed while we copy the rest of the pages
360 	 * to their final resting place.  As such they must
361 	 * not conflict with either the destination addresses
362 	 * or memory the kernel is already using.
363 	 *
364 	 * The only case where we really need more than one of
365 	 * these are for architectures where we cannot disable
366 	 * the MMU and must instead generate an identity mapped
367 	 * page table for all of the memory.
368 	 *
369 	 * At worst this runs in O(N) of the image size.
370 	 */
371 	struct list_head extra_pages;
372 	struct page *pages;
373 	unsigned int count;
374 
375 	count = 1 << order;
376 	INIT_LIST_HEAD(&extra_pages);
377 
378 	/* Loop while I can allocate a page and the page allocated
379 	 * is a destination page.
380 	 */
381 	do {
382 		unsigned long pfn, epfn, addr, eaddr;
383 
384 		pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
385 		if (!pages)
386 			break;
387 		pfn   = page_to_boot_pfn(pages);
388 		epfn  = pfn + count;
389 		addr  = pfn << PAGE_SHIFT;
390 		eaddr = epfn << PAGE_SHIFT;
391 		if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
392 			      kimage_is_destination_range(image, addr, eaddr)) {
393 			list_add(&pages->lru, &extra_pages);
394 			pages = NULL;
395 		}
396 	} while (!pages);
397 
398 	if (pages) {
399 		/* Remember the allocated page... */
400 		list_add(&pages->lru, &image->control_pages);
401 
402 		/* Because the page is already in it's destination
403 		 * location we will never allocate another page at
404 		 * that address.  Therefore kimage_alloc_pages
405 		 * will not return it (again) and we don't need
406 		 * to give it an entry in image->segment[].
407 		 */
408 	}
409 	/* Deal with the destination pages I have inadvertently allocated.
410 	 *
411 	 * Ideally I would convert multi-page allocations into single
412 	 * page allocations, and add everything to image->dest_pages.
413 	 *
414 	 * For now it is simpler to just free the pages.
415 	 */
416 	kimage_free_page_list(&extra_pages);
417 
418 	return pages;
419 }
420 
421 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
422 						      unsigned int order)
423 {
424 	/* Control pages are special, they are the intermediaries
425 	 * that are needed while we copy the rest of the pages
426 	 * to their final resting place.  As such they must
427 	 * not conflict with either the destination addresses
428 	 * or memory the kernel is already using.
429 	 *
430 	 * Control pages are also the only pags we must allocate
431 	 * when loading a crash kernel.  All of the other pages
432 	 * are specified by the segments and we just memcpy
433 	 * into them directly.
434 	 *
435 	 * The only case where we really need more than one of
436 	 * these are for architectures where we cannot disable
437 	 * the MMU and must instead generate an identity mapped
438 	 * page table for all of the memory.
439 	 *
440 	 * Given the low demand this implements a very simple
441 	 * allocator that finds the first hole of the appropriate
442 	 * size in the reserved memory region, and allocates all
443 	 * of the memory up to and including the hole.
444 	 */
445 	unsigned long hole_start, hole_end, size;
446 	struct page *pages;
447 
448 	pages = NULL;
449 	size = (1 << order) << PAGE_SHIFT;
450 	hole_start = (image->control_page + (size - 1)) & ~(size - 1);
451 	hole_end   = hole_start + size - 1;
452 	while (hole_end <= crashk_res.end) {
453 		unsigned long i;
454 
455 		cond_resched();
456 
457 		if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
458 			break;
459 		/* See if I overlap any of the segments */
460 		for (i = 0; i < image->nr_segments; i++) {
461 			unsigned long mstart, mend;
462 
463 			mstart = image->segment[i].mem;
464 			mend   = mstart + image->segment[i].memsz - 1;
465 			if ((hole_end >= mstart) && (hole_start <= mend)) {
466 				/* Advance the hole to the end of the segment */
467 				hole_start = (mend + (size - 1)) & ~(size - 1);
468 				hole_end   = hole_start + size - 1;
469 				break;
470 			}
471 		}
472 		/* If I don't overlap any segments I have found my hole! */
473 		if (i == image->nr_segments) {
474 			pages = pfn_to_page(hole_start >> PAGE_SHIFT);
475 			image->control_page = hole_end;
476 			break;
477 		}
478 	}
479 
480 	/* Ensure that these pages are decrypted if SME is enabled. */
481 	if (pages)
482 		arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
483 
484 	return pages;
485 }
486 
487 
488 struct page *kimage_alloc_control_pages(struct kimage *image,
489 					 unsigned int order)
490 {
491 	struct page *pages = NULL;
492 
493 	switch (image->type) {
494 	case KEXEC_TYPE_DEFAULT:
495 		pages = kimage_alloc_normal_control_pages(image, order);
496 		break;
497 	case KEXEC_TYPE_CRASH:
498 		pages = kimage_alloc_crash_control_pages(image, order);
499 		break;
500 	}
501 
502 	return pages;
503 }
504 
505 int kimage_crash_copy_vmcoreinfo(struct kimage *image)
506 {
507 	struct page *vmcoreinfo_page;
508 	void *safecopy;
509 
510 	if (image->type != KEXEC_TYPE_CRASH)
511 		return 0;
512 
513 	/*
514 	 * For kdump, allocate one vmcoreinfo safe copy from the
515 	 * crash memory. as we have arch_kexec_protect_crashkres()
516 	 * after kexec syscall, we naturally protect it from write
517 	 * (even read) access under kernel direct mapping. But on
518 	 * the other hand, we still need to operate it when crash
519 	 * happens to generate vmcoreinfo note, hereby we rely on
520 	 * vmap for this purpose.
521 	 */
522 	vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
523 	if (!vmcoreinfo_page) {
524 		pr_warn("Could not allocate vmcoreinfo buffer\n");
525 		return -ENOMEM;
526 	}
527 	safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
528 	if (!safecopy) {
529 		pr_warn("Could not vmap vmcoreinfo buffer\n");
530 		return -ENOMEM;
531 	}
532 
533 	image->vmcoreinfo_data_copy = safecopy;
534 	crash_update_vmcoreinfo_safecopy(safecopy);
535 
536 	return 0;
537 }
538 
539 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
540 {
541 	if (*image->entry != 0)
542 		image->entry++;
543 
544 	if (image->entry == image->last_entry) {
545 		kimage_entry_t *ind_page;
546 		struct page *page;
547 
548 		page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
549 		if (!page)
550 			return -ENOMEM;
551 
552 		ind_page = page_address(page);
553 		*image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
554 		image->entry = ind_page;
555 		image->last_entry = ind_page +
556 				      ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
557 	}
558 	*image->entry = entry;
559 	image->entry++;
560 	*image->entry = 0;
561 
562 	return 0;
563 }
564 
565 static int kimage_set_destination(struct kimage *image,
566 				   unsigned long destination)
567 {
568 	destination &= PAGE_MASK;
569 
570 	return kimage_add_entry(image, destination | IND_DESTINATION);
571 }
572 
573 
574 static int kimage_add_page(struct kimage *image, unsigned long page)
575 {
576 	page &= PAGE_MASK;
577 
578 	return kimage_add_entry(image, page | IND_SOURCE);
579 }
580 
581 
582 static void kimage_free_extra_pages(struct kimage *image)
583 {
584 	/* Walk through and free any extra destination pages I may have */
585 	kimage_free_page_list(&image->dest_pages);
586 
587 	/* Walk through and free any unusable pages I have cached */
588 	kimage_free_page_list(&image->unusable_pages);
589 
590 }
591 
592 void kimage_terminate(struct kimage *image)
593 {
594 	if (*image->entry != 0)
595 		image->entry++;
596 
597 	*image->entry = IND_DONE;
598 }
599 
600 #define for_each_kimage_entry(image, ptr, entry) \
601 	for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
602 		ptr = (entry & IND_INDIRECTION) ? \
603 			boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
604 
605 static void kimage_free_entry(kimage_entry_t entry)
606 {
607 	struct page *page;
608 
609 	page = boot_pfn_to_page(entry >> PAGE_SHIFT);
610 	kimage_free_pages(page);
611 }
612 
613 void kimage_free(struct kimage *image)
614 {
615 	kimage_entry_t *ptr, entry;
616 	kimage_entry_t ind = 0;
617 
618 	if (!image)
619 		return;
620 
621 	if (image->vmcoreinfo_data_copy) {
622 		crash_update_vmcoreinfo_safecopy(NULL);
623 		vunmap(image->vmcoreinfo_data_copy);
624 	}
625 
626 	kimage_free_extra_pages(image);
627 	for_each_kimage_entry(image, ptr, entry) {
628 		if (entry & IND_INDIRECTION) {
629 			/* Free the previous indirection page */
630 			if (ind & IND_INDIRECTION)
631 				kimage_free_entry(ind);
632 			/* Save this indirection page until we are
633 			 * done with it.
634 			 */
635 			ind = entry;
636 		} else if (entry & IND_SOURCE)
637 			kimage_free_entry(entry);
638 	}
639 	/* Free the final indirection page */
640 	if (ind & IND_INDIRECTION)
641 		kimage_free_entry(ind);
642 
643 	/* Handle any machine specific cleanup */
644 	machine_kexec_cleanup(image);
645 
646 	/* Free the kexec control pages... */
647 	kimage_free_page_list(&image->control_pages);
648 
649 	/*
650 	 * Free up any temporary buffers allocated. This might hit if
651 	 * error occurred much later after buffer allocation.
652 	 */
653 	if (image->file_mode)
654 		kimage_file_post_load_cleanup(image);
655 
656 	kfree(image);
657 }
658 
659 static kimage_entry_t *kimage_dst_used(struct kimage *image,
660 					unsigned long page)
661 {
662 	kimage_entry_t *ptr, entry;
663 	unsigned long destination = 0;
664 
665 	for_each_kimage_entry(image, ptr, entry) {
666 		if (entry & IND_DESTINATION)
667 			destination = entry & PAGE_MASK;
668 		else if (entry & IND_SOURCE) {
669 			if (page == destination)
670 				return ptr;
671 			destination += PAGE_SIZE;
672 		}
673 	}
674 
675 	return NULL;
676 }
677 
678 static struct page *kimage_alloc_page(struct kimage *image,
679 					gfp_t gfp_mask,
680 					unsigned long destination)
681 {
682 	/*
683 	 * Here we implement safeguards to ensure that a source page
684 	 * is not copied to its destination page before the data on
685 	 * the destination page is no longer useful.
686 	 *
687 	 * To do this we maintain the invariant that a source page is
688 	 * either its own destination page, or it is not a
689 	 * destination page at all.
690 	 *
691 	 * That is slightly stronger than required, but the proof
692 	 * that no problems will not occur is trivial, and the
693 	 * implementation is simply to verify.
694 	 *
695 	 * When allocating all pages normally this algorithm will run
696 	 * in O(N) time, but in the worst case it will run in O(N^2)
697 	 * time.   If the runtime is a problem the data structures can
698 	 * be fixed.
699 	 */
700 	struct page *page;
701 	unsigned long addr;
702 
703 	/*
704 	 * Walk through the list of destination pages, and see if I
705 	 * have a match.
706 	 */
707 	list_for_each_entry(page, &image->dest_pages, lru) {
708 		addr = page_to_boot_pfn(page) << PAGE_SHIFT;
709 		if (addr == destination) {
710 			list_del(&page->lru);
711 			return page;
712 		}
713 	}
714 	page = NULL;
715 	while (1) {
716 		kimage_entry_t *old;
717 
718 		/* Allocate a page, if we run out of memory give up */
719 		page = kimage_alloc_pages(gfp_mask, 0);
720 		if (!page)
721 			return NULL;
722 		/* If the page cannot be used file it away */
723 		if (page_to_boot_pfn(page) >
724 				(KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
725 			list_add(&page->lru, &image->unusable_pages);
726 			continue;
727 		}
728 		addr = page_to_boot_pfn(page) << PAGE_SHIFT;
729 
730 		/* If it is the destination page we want use it */
731 		if (addr == destination)
732 			break;
733 
734 		/* If the page is not a destination page use it */
735 		if (!kimage_is_destination_range(image, addr,
736 						  addr + PAGE_SIZE))
737 			break;
738 
739 		/*
740 		 * I know that the page is someones destination page.
741 		 * See if there is already a source page for this
742 		 * destination page.  And if so swap the source pages.
743 		 */
744 		old = kimage_dst_used(image, addr);
745 		if (old) {
746 			/* If so move it */
747 			unsigned long old_addr;
748 			struct page *old_page;
749 
750 			old_addr = *old & PAGE_MASK;
751 			old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
752 			copy_highpage(page, old_page);
753 			*old = addr | (*old & ~PAGE_MASK);
754 
755 			/* The old page I have found cannot be a
756 			 * destination page, so return it if it's
757 			 * gfp_flags honor the ones passed in.
758 			 */
759 			if (!(gfp_mask & __GFP_HIGHMEM) &&
760 			    PageHighMem(old_page)) {
761 				kimage_free_pages(old_page);
762 				continue;
763 			}
764 			page = old_page;
765 			break;
766 		}
767 		/* Place the page on the destination list, to be used later */
768 		list_add(&page->lru, &image->dest_pages);
769 	}
770 
771 	return page;
772 }
773 
774 static int kimage_load_normal_segment(struct kimage *image,
775 					 struct kexec_segment *segment)
776 {
777 	unsigned long maddr;
778 	size_t ubytes, mbytes;
779 	int result;
780 	unsigned char __user *buf = NULL;
781 	unsigned char *kbuf = NULL;
782 
783 	if (image->file_mode)
784 		kbuf = segment->kbuf;
785 	else
786 		buf = segment->buf;
787 	ubytes = segment->bufsz;
788 	mbytes = segment->memsz;
789 	maddr = segment->mem;
790 
791 	result = kimage_set_destination(image, maddr);
792 	if (result < 0)
793 		goto out;
794 
795 	while (mbytes) {
796 		struct page *page;
797 		char *ptr;
798 		size_t uchunk, mchunk;
799 
800 		page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
801 		if (!page) {
802 			result  = -ENOMEM;
803 			goto out;
804 		}
805 		result = kimage_add_page(image, page_to_boot_pfn(page)
806 								<< PAGE_SHIFT);
807 		if (result < 0)
808 			goto out;
809 
810 		ptr = kmap_local_page(page);
811 		/* Start with a clear page */
812 		clear_page(ptr);
813 		ptr += maddr & ~PAGE_MASK;
814 		mchunk = min_t(size_t, mbytes,
815 				PAGE_SIZE - (maddr & ~PAGE_MASK));
816 		uchunk = min(ubytes, mchunk);
817 
818 		/* For file based kexec, source pages are in kernel memory */
819 		if (image->file_mode)
820 			memcpy(ptr, kbuf, uchunk);
821 		else
822 			result = copy_from_user(ptr, buf, uchunk);
823 		kunmap_local(ptr);
824 		if (result) {
825 			result = -EFAULT;
826 			goto out;
827 		}
828 		ubytes -= uchunk;
829 		maddr  += mchunk;
830 		if (image->file_mode)
831 			kbuf += mchunk;
832 		else
833 			buf += mchunk;
834 		mbytes -= mchunk;
835 
836 		cond_resched();
837 	}
838 out:
839 	return result;
840 }
841 
842 static int kimage_load_crash_segment(struct kimage *image,
843 					struct kexec_segment *segment)
844 {
845 	/* For crash dumps kernels we simply copy the data from
846 	 * user space to it's destination.
847 	 * We do things a page at a time for the sake of kmap.
848 	 */
849 	unsigned long maddr;
850 	size_t ubytes, mbytes;
851 	int result;
852 	unsigned char __user *buf = NULL;
853 	unsigned char *kbuf = NULL;
854 
855 	result = 0;
856 	if (image->file_mode)
857 		kbuf = segment->kbuf;
858 	else
859 		buf = segment->buf;
860 	ubytes = segment->bufsz;
861 	mbytes = segment->memsz;
862 	maddr = segment->mem;
863 	while (mbytes) {
864 		struct page *page;
865 		char *ptr;
866 		size_t uchunk, mchunk;
867 
868 		page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
869 		if (!page) {
870 			result  = -ENOMEM;
871 			goto out;
872 		}
873 		arch_kexec_post_alloc_pages(page_address(page), 1, 0);
874 		ptr = kmap_local_page(page);
875 		ptr += maddr & ~PAGE_MASK;
876 		mchunk = min_t(size_t, mbytes,
877 				PAGE_SIZE - (maddr & ~PAGE_MASK));
878 		uchunk = min(ubytes, mchunk);
879 		if (mchunk > uchunk) {
880 			/* Zero the trailing part of the page */
881 			memset(ptr + uchunk, 0, mchunk - uchunk);
882 		}
883 
884 		/* For file based kexec, source pages are in kernel memory */
885 		if (image->file_mode)
886 			memcpy(ptr, kbuf, uchunk);
887 		else
888 			result = copy_from_user(ptr, buf, uchunk);
889 		kexec_flush_icache_page(page);
890 		kunmap_local(ptr);
891 		arch_kexec_pre_free_pages(page_address(page), 1);
892 		if (result) {
893 			result = -EFAULT;
894 			goto out;
895 		}
896 		ubytes -= uchunk;
897 		maddr  += mchunk;
898 		if (image->file_mode)
899 			kbuf += mchunk;
900 		else
901 			buf += mchunk;
902 		mbytes -= mchunk;
903 
904 		cond_resched();
905 	}
906 out:
907 	return result;
908 }
909 
910 int kimage_load_segment(struct kimage *image,
911 				struct kexec_segment *segment)
912 {
913 	int result = -ENOMEM;
914 
915 	switch (image->type) {
916 	case KEXEC_TYPE_DEFAULT:
917 		result = kimage_load_normal_segment(image, segment);
918 		break;
919 	case KEXEC_TYPE_CRASH:
920 		result = kimage_load_crash_segment(image, segment);
921 		break;
922 	}
923 
924 	return result;
925 }
926 
927 struct kexec_load_limit {
928 	/* Mutex protects the limit count. */
929 	struct mutex mutex;
930 	int limit;
931 };
932 
933 static struct kexec_load_limit load_limit_reboot = {
934 	.mutex = __MUTEX_INITIALIZER(load_limit_reboot.mutex),
935 	.limit = -1,
936 };
937 
938 static struct kexec_load_limit load_limit_panic = {
939 	.mutex = __MUTEX_INITIALIZER(load_limit_panic.mutex),
940 	.limit = -1,
941 };
942 
943 struct kimage *kexec_image;
944 struct kimage *kexec_crash_image;
945 static int kexec_load_disabled;
946 
947 #ifdef CONFIG_SYSCTL
948 static int kexec_limit_handler(struct ctl_table *table, int write,
949 			       void *buffer, size_t *lenp, loff_t *ppos)
950 {
951 	struct kexec_load_limit *limit = table->data;
952 	int val;
953 	struct ctl_table tmp = {
954 		.data = &val,
955 		.maxlen = sizeof(val),
956 		.mode = table->mode,
957 	};
958 	int ret;
959 
960 	if (write) {
961 		ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
962 		if (ret)
963 			return ret;
964 
965 		if (val < 0)
966 			return -EINVAL;
967 
968 		mutex_lock(&limit->mutex);
969 		if (limit->limit != -1 && val >= limit->limit)
970 			ret = -EINVAL;
971 		else
972 			limit->limit = val;
973 		mutex_unlock(&limit->mutex);
974 
975 		return ret;
976 	}
977 
978 	mutex_lock(&limit->mutex);
979 	val = limit->limit;
980 	mutex_unlock(&limit->mutex);
981 
982 	return proc_dointvec(&tmp, write, buffer, lenp, ppos);
983 }
984 
985 static struct ctl_table kexec_core_sysctls[] = {
986 	{
987 		.procname	= "kexec_load_disabled",
988 		.data		= &kexec_load_disabled,
989 		.maxlen		= sizeof(int),
990 		.mode		= 0644,
991 		/* only handle a transition from default "0" to "1" */
992 		.proc_handler	= proc_dointvec_minmax,
993 		.extra1		= SYSCTL_ONE,
994 		.extra2		= SYSCTL_ONE,
995 	},
996 	{
997 		.procname	= "kexec_load_limit_panic",
998 		.data		= &load_limit_panic,
999 		.mode		= 0644,
1000 		.proc_handler	= kexec_limit_handler,
1001 	},
1002 	{
1003 		.procname	= "kexec_load_limit_reboot",
1004 		.data		= &load_limit_reboot,
1005 		.mode		= 0644,
1006 		.proc_handler	= kexec_limit_handler,
1007 	},
1008 	{ }
1009 };
1010 
1011 static int __init kexec_core_sysctl_init(void)
1012 {
1013 	register_sysctl_init("kernel", kexec_core_sysctls);
1014 	return 0;
1015 }
1016 late_initcall(kexec_core_sysctl_init);
1017 #endif
1018 
1019 bool kexec_load_permitted(int kexec_image_type)
1020 {
1021 	struct kexec_load_limit *limit;
1022 
1023 	/*
1024 	 * Only the superuser can use the kexec syscall and if it has not
1025 	 * been disabled.
1026 	 */
1027 	if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1028 		return false;
1029 
1030 	/* Check limit counter and decrease it.*/
1031 	limit = (kexec_image_type == KEXEC_TYPE_CRASH) ?
1032 		&load_limit_panic : &load_limit_reboot;
1033 	mutex_lock(&limit->mutex);
1034 	if (!limit->limit) {
1035 		mutex_unlock(&limit->mutex);
1036 		return false;
1037 	}
1038 	if (limit->limit != -1)
1039 		limit->limit--;
1040 	mutex_unlock(&limit->mutex);
1041 
1042 	return true;
1043 }
1044 
1045 /*
1046  * No panic_cpu check version of crash_kexec().  This function is called
1047  * only when panic_cpu holds the current CPU number; this is the only CPU
1048  * which processes crash_kexec routines.
1049  */
1050 void __noclone __crash_kexec(struct pt_regs *regs)
1051 {
1052 	/* Take the kexec_lock here to prevent sys_kexec_load
1053 	 * running on one cpu from replacing the crash kernel
1054 	 * we are using after a panic on a different cpu.
1055 	 *
1056 	 * If the crash kernel was not located in a fixed area
1057 	 * of memory the xchg(&kexec_crash_image) would be
1058 	 * sufficient.  But since I reuse the memory...
1059 	 */
1060 	if (kexec_trylock()) {
1061 		if (kexec_crash_image) {
1062 			struct pt_regs fixed_regs;
1063 
1064 			crash_setup_regs(&fixed_regs, regs);
1065 			crash_save_vmcoreinfo();
1066 			machine_crash_shutdown(&fixed_regs);
1067 			machine_kexec(kexec_crash_image);
1068 		}
1069 		kexec_unlock();
1070 	}
1071 }
1072 STACK_FRAME_NON_STANDARD(__crash_kexec);
1073 
1074 __bpf_kfunc void crash_kexec(struct pt_regs *regs)
1075 {
1076 	int old_cpu, this_cpu;
1077 
1078 	/*
1079 	 * Only one CPU is allowed to execute the crash_kexec() code as with
1080 	 * panic().  Otherwise parallel calls of panic() and crash_kexec()
1081 	 * may stop each other.  To exclude them, we use panic_cpu here too.
1082 	 */
1083 	this_cpu = raw_smp_processor_id();
1084 	old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
1085 	if (old_cpu == PANIC_CPU_INVALID) {
1086 		/* This is the 1st CPU which comes here, so go ahead. */
1087 		__crash_kexec(regs);
1088 
1089 		/*
1090 		 * Reset panic_cpu to allow another panic()/crash_kexec()
1091 		 * call.
1092 		 */
1093 		atomic_set(&panic_cpu, PANIC_CPU_INVALID);
1094 	}
1095 }
1096 
1097 static inline resource_size_t crash_resource_size(const struct resource *res)
1098 {
1099 	return !res->end ? 0 : resource_size(res);
1100 }
1101 
1102 ssize_t crash_get_memory_size(void)
1103 {
1104 	ssize_t size = 0;
1105 
1106 	if (!kexec_trylock())
1107 		return -EBUSY;
1108 
1109 	size += crash_resource_size(&crashk_res);
1110 	size += crash_resource_size(&crashk_low_res);
1111 
1112 	kexec_unlock();
1113 	return size;
1114 }
1115 
1116 static int __crash_shrink_memory(struct resource *old_res,
1117 				 unsigned long new_size)
1118 {
1119 	struct resource *ram_res;
1120 
1121 	ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1122 	if (!ram_res)
1123 		return -ENOMEM;
1124 
1125 	ram_res->start = old_res->start + new_size;
1126 	ram_res->end   = old_res->end;
1127 	ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1128 	ram_res->name  = "System RAM";
1129 
1130 	if (!new_size) {
1131 		release_resource(old_res);
1132 		old_res->start = 0;
1133 		old_res->end   = 0;
1134 	} else {
1135 		crashk_res.end = ram_res->start - 1;
1136 	}
1137 
1138 	crash_free_reserved_phys_range(ram_res->start, ram_res->end);
1139 	insert_resource(&iomem_resource, ram_res);
1140 
1141 	return 0;
1142 }
1143 
1144 int crash_shrink_memory(unsigned long new_size)
1145 {
1146 	int ret = 0;
1147 	unsigned long old_size, low_size;
1148 
1149 	if (!kexec_trylock())
1150 		return -EBUSY;
1151 
1152 	if (kexec_crash_image) {
1153 		ret = -ENOENT;
1154 		goto unlock;
1155 	}
1156 
1157 	low_size = crash_resource_size(&crashk_low_res);
1158 	old_size = crash_resource_size(&crashk_res) + low_size;
1159 	new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
1160 	if (new_size >= old_size) {
1161 		ret = (new_size == old_size) ? 0 : -EINVAL;
1162 		goto unlock;
1163 	}
1164 
1165 	/*
1166 	 * (low_size > new_size) implies that low_size is greater than zero.
1167 	 * This also means that if low_size is zero, the else branch is taken.
1168 	 *
1169 	 * If low_size is greater than 0, (low_size > new_size) indicates that
1170 	 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
1171 	 * needs to be shrunken.
1172 	 */
1173 	if (low_size > new_size) {
1174 		ret = __crash_shrink_memory(&crashk_res, 0);
1175 		if (ret)
1176 			goto unlock;
1177 
1178 		ret = __crash_shrink_memory(&crashk_low_res, new_size);
1179 	} else {
1180 		ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
1181 	}
1182 
1183 	/* Swap crashk_res and crashk_low_res if needed */
1184 	if (!crashk_res.end && crashk_low_res.end) {
1185 		crashk_res.start = crashk_low_res.start;
1186 		crashk_res.end   = crashk_low_res.end;
1187 		release_resource(&crashk_low_res);
1188 		crashk_low_res.start = 0;
1189 		crashk_low_res.end   = 0;
1190 		insert_resource(&iomem_resource, &crashk_res);
1191 	}
1192 
1193 unlock:
1194 	kexec_unlock();
1195 	return ret;
1196 }
1197 
1198 void crash_save_cpu(struct pt_regs *regs, int cpu)
1199 {
1200 	struct elf_prstatus prstatus;
1201 	u32 *buf;
1202 
1203 	if ((cpu < 0) || (cpu >= nr_cpu_ids))
1204 		return;
1205 
1206 	/* Using ELF notes here is opportunistic.
1207 	 * I need a well defined structure format
1208 	 * for the data I pass, and I need tags
1209 	 * on the data to indicate what information I have
1210 	 * squirrelled away.  ELF notes happen to provide
1211 	 * all of that, so there is no need to invent something new.
1212 	 */
1213 	buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1214 	if (!buf)
1215 		return;
1216 	memset(&prstatus, 0, sizeof(prstatus));
1217 	prstatus.common.pr_pid = current->pid;
1218 	elf_core_copy_regs(&prstatus.pr_reg, regs);
1219 	buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1220 			      &prstatus, sizeof(prstatus));
1221 	final_note(buf);
1222 }
1223 
1224 /*
1225  * Move into place and start executing a preloaded standalone
1226  * executable.  If nothing was preloaded return an error.
1227  */
1228 int kernel_kexec(void)
1229 {
1230 	int error = 0;
1231 
1232 	if (!kexec_trylock())
1233 		return -EBUSY;
1234 	if (!kexec_image) {
1235 		error = -EINVAL;
1236 		goto Unlock;
1237 	}
1238 
1239 #ifdef CONFIG_KEXEC_JUMP
1240 	if (kexec_image->preserve_context) {
1241 		pm_prepare_console();
1242 		error = freeze_processes();
1243 		if (error) {
1244 			error = -EBUSY;
1245 			goto Restore_console;
1246 		}
1247 		suspend_console();
1248 		error = dpm_suspend_start(PMSG_FREEZE);
1249 		if (error)
1250 			goto Resume_console;
1251 		/* At this point, dpm_suspend_start() has been called,
1252 		 * but *not* dpm_suspend_end(). We *must* call
1253 		 * dpm_suspend_end() now.  Otherwise, drivers for
1254 		 * some devices (e.g. interrupt controllers) become
1255 		 * desynchronized with the actual state of the
1256 		 * hardware at resume time, and evil weirdness ensues.
1257 		 */
1258 		error = dpm_suspend_end(PMSG_FREEZE);
1259 		if (error)
1260 			goto Resume_devices;
1261 		error = suspend_disable_secondary_cpus();
1262 		if (error)
1263 			goto Enable_cpus;
1264 		local_irq_disable();
1265 		error = syscore_suspend();
1266 		if (error)
1267 			goto Enable_irqs;
1268 	} else
1269 #endif
1270 	{
1271 		kexec_in_progress = true;
1272 		kernel_restart_prepare("kexec reboot");
1273 		migrate_to_reboot_cpu();
1274 
1275 		/*
1276 		 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1277 		 * no further code needs to use CPU hotplug (which is true in
1278 		 * the reboot case). However, the kexec path depends on using
1279 		 * CPU hotplug again; so re-enable it here.
1280 		 */
1281 		cpu_hotplug_enable();
1282 		pr_notice("Starting new kernel\n");
1283 		machine_shutdown();
1284 	}
1285 
1286 	kmsg_dump(KMSG_DUMP_SHUTDOWN);
1287 	machine_kexec(kexec_image);
1288 
1289 #ifdef CONFIG_KEXEC_JUMP
1290 	if (kexec_image->preserve_context) {
1291 		syscore_resume();
1292  Enable_irqs:
1293 		local_irq_enable();
1294  Enable_cpus:
1295 		suspend_enable_secondary_cpus();
1296 		dpm_resume_start(PMSG_RESTORE);
1297  Resume_devices:
1298 		dpm_resume_end(PMSG_RESTORE);
1299  Resume_console:
1300 		resume_console();
1301 		thaw_processes();
1302  Restore_console:
1303 		pm_restore_console();
1304 	}
1305 #endif
1306 
1307  Unlock:
1308 	kexec_unlock();
1309 	return error;
1310 }
1311