xref: /linux/kernel/kexec_core.c (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
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 bool kexec_file_dbg_print;
56 
57 /*
58  * When kexec transitions to the new kernel there is a one-to-one
59  * mapping between physical and virtual addresses.  On processors
60  * where you can disable the MMU this is trivial, and easy.  For
61  * others it is still a simple predictable page table to setup.
62  *
63  * In that environment kexec copies the new kernel to its final
64  * resting place.  This means I can only support memory whose
65  * physical address can fit in an unsigned long.  In particular
66  * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
67  * If the assembly stub has more restrictive requirements
68  * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
69  * defined more restrictively in <asm/kexec.h>.
70  *
71  * The code for the transition from the current kernel to the
72  * new kernel is placed in the control_code_buffer, whose size
73  * is given by KEXEC_CONTROL_PAGE_SIZE.  In the best case only a single
74  * page of memory is necessary, but some architectures require more.
75  * Because this memory must be identity mapped in the transition from
76  * virtual to physical addresses it must live in the range
77  * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
78  * modifiable.
79  *
80  * The assembly stub in the control code buffer is passed a linked list
81  * of descriptor pages detailing the source pages of the new kernel,
82  * and the destination addresses of those source pages.  As this data
83  * structure is not used in the context of the current OS, it must
84  * be self-contained.
85  *
86  * The code has been made to work with highmem pages and will use a
87  * destination page in its final resting place (if it happens
88  * to allocate it).  The end product of this is that most of the
89  * physical address space, and most of RAM can be used.
90  *
91  * Future directions include:
92  *  - allocating a page table with the control code buffer identity
93  *    mapped, to simplify machine_kexec and make kexec_on_panic more
94  *    reliable.
95  */
96 
97 /*
98  * KIMAGE_NO_DEST is an impossible destination address..., for
99  * allocating pages whose destination address we do not care about.
100  */
101 #define KIMAGE_NO_DEST (-1UL)
102 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
103 
104 static struct page *kimage_alloc_page(struct kimage *image,
105 				       gfp_t gfp_mask,
106 				       unsigned long dest);
107 
sanity_check_segment_list(struct kimage * image)108 int sanity_check_segment_list(struct kimage *image)
109 {
110 	int i;
111 	unsigned long nr_segments = image->nr_segments;
112 	unsigned long total_pages = 0;
113 	unsigned long nr_pages = totalram_pages();
114 
115 	/*
116 	 * Verify we have good destination addresses.  The caller is
117 	 * responsible for making certain we don't attempt to load
118 	 * the new image into invalid or reserved areas of RAM.  This
119 	 * just verifies it is an address we can use.
120 	 *
121 	 * Since the kernel does everything in page size chunks ensure
122 	 * the destination addresses are page aligned.  Too many
123 	 * special cases crop of when we don't do this.  The most
124 	 * insidious is getting overlapping destination addresses
125 	 * simply because addresses are changed to page size
126 	 * granularity.
127 	 */
128 	for (i = 0; i < nr_segments; i++) {
129 		unsigned long mstart, mend;
130 
131 		mstart = image->segment[i].mem;
132 		mend   = mstart + image->segment[i].memsz;
133 		if (mstart > mend)
134 			return -EADDRNOTAVAIL;
135 		if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
136 			return -EADDRNOTAVAIL;
137 		if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
138 			return -EADDRNOTAVAIL;
139 	}
140 
141 	/* Verify our destination addresses do not overlap.
142 	 * If we alloed overlapping destination addresses
143 	 * through very weird things can happen with no
144 	 * easy explanation as one segment stops on another.
145 	 */
146 	for (i = 0; i < nr_segments; i++) {
147 		unsigned long mstart, mend;
148 		unsigned long j;
149 
150 		mstart = image->segment[i].mem;
151 		mend   = mstart + image->segment[i].memsz;
152 		for (j = 0; j < i; j++) {
153 			unsigned long pstart, pend;
154 
155 			pstart = image->segment[j].mem;
156 			pend   = pstart + image->segment[j].memsz;
157 			/* Do the segments overlap ? */
158 			if ((mend > pstart) && (mstart < pend))
159 				return -EINVAL;
160 		}
161 	}
162 
163 	/* Ensure our buffer sizes are strictly less than
164 	 * our memory sizes.  This should always be the case,
165 	 * and it is easier to check up front than to be surprised
166 	 * later on.
167 	 */
168 	for (i = 0; i < nr_segments; i++) {
169 		if (image->segment[i].bufsz > image->segment[i].memsz)
170 			return -EINVAL;
171 	}
172 
173 	/*
174 	 * Verify that no more than half of memory will be consumed. If the
175 	 * request from userspace is too large, a large amount of time will be
176 	 * wasted allocating pages, which can cause a soft lockup.
177 	 */
178 	for (i = 0; i < nr_segments; i++) {
179 		if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
180 			return -EINVAL;
181 
182 		total_pages += PAGE_COUNT(image->segment[i].memsz);
183 	}
184 
185 	if (total_pages > nr_pages / 2)
186 		return -EINVAL;
187 
188 #ifdef CONFIG_CRASH_DUMP
189 	/*
190 	 * Verify we have good destination addresses.  Normally
191 	 * the caller is responsible for making certain we don't
192 	 * attempt to load the new image into invalid or reserved
193 	 * areas of RAM.  But crash kernels are preloaded into a
194 	 * reserved area of ram.  We must ensure the addresses
195 	 * are in the reserved area otherwise preloading the
196 	 * kernel could corrupt things.
197 	 */
198 
199 	if (image->type == KEXEC_TYPE_CRASH) {
200 		for (i = 0; i < nr_segments; i++) {
201 			unsigned long mstart, mend;
202 
203 			mstart = image->segment[i].mem;
204 			mend = mstart + image->segment[i].memsz - 1;
205 			/* Ensure we are within the crash kernel limits */
206 			if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
207 			    (mend > phys_to_boot_phys(crashk_res.end)))
208 				return -EADDRNOTAVAIL;
209 		}
210 	}
211 #endif
212 
213 	return 0;
214 }
215 
do_kimage_alloc_init(void)216 struct kimage *do_kimage_alloc_init(void)
217 {
218 	struct kimage *image;
219 
220 	/* Allocate a controlling structure */
221 	image = kzalloc(sizeof(*image), GFP_KERNEL);
222 	if (!image)
223 		return NULL;
224 
225 	image->head = 0;
226 	image->entry = &image->head;
227 	image->last_entry = &image->head;
228 	image->control_page = ~0; /* By default this does not apply */
229 	image->type = KEXEC_TYPE_DEFAULT;
230 
231 	/* Initialize the list of control pages */
232 	INIT_LIST_HEAD(&image->control_pages);
233 
234 	/* Initialize the list of destination pages */
235 	INIT_LIST_HEAD(&image->dest_pages);
236 
237 	/* Initialize the list of unusable pages */
238 	INIT_LIST_HEAD(&image->unusable_pages);
239 
240 #ifdef CONFIG_CRASH_HOTPLUG
241 	image->hp_action = KEXEC_CRASH_HP_NONE;
242 	image->elfcorehdr_index = -1;
243 	image->elfcorehdr_updated = false;
244 #endif
245 
246 	return image;
247 }
248 
kimage_is_destination_range(struct kimage * image,unsigned long start,unsigned long end)249 int kimage_is_destination_range(struct kimage *image,
250 					unsigned long start,
251 					unsigned long end)
252 {
253 	unsigned long i;
254 
255 	for (i = 0; i < image->nr_segments; i++) {
256 		unsigned long mstart, mend;
257 
258 		mstart = image->segment[i].mem;
259 		mend = mstart + image->segment[i].memsz - 1;
260 		if ((end >= mstart) && (start <= mend))
261 			return 1;
262 	}
263 
264 	return 0;
265 }
266 
kimage_alloc_pages(gfp_t gfp_mask,unsigned int order)267 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
268 {
269 	struct page *pages;
270 
271 	if (fatal_signal_pending(current))
272 		return NULL;
273 	pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
274 	if (pages) {
275 		unsigned int count, i;
276 
277 		pages->mapping = NULL;
278 		set_page_private(pages, order);
279 		count = 1 << order;
280 		for (i = 0; i < count; i++)
281 			SetPageReserved(pages + i);
282 
283 		arch_kexec_post_alloc_pages(page_address(pages), count,
284 					    gfp_mask);
285 
286 		if (gfp_mask & __GFP_ZERO)
287 			for (i = 0; i < count; i++)
288 				clear_highpage(pages + i);
289 	}
290 
291 	return pages;
292 }
293 
kimage_free_pages(struct page * page)294 static void kimage_free_pages(struct page *page)
295 {
296 	unsigned int order, count, i;
297 
298 	order = page_private(page);
299 	count = 1 << order;
300 
301 	arch_kexec_pre_free_pages(page_address(page), count);
302 
303 	for (i = 0; i < count; i++)
304 		ClearPageReserved(page + i);
305 	__free_pages(page, order);
306 }
307 
kimage_free_page_list(struct list_head * list)308 void kimage_free_page_list(struct list_head *list)
309 {
310 	struct page *page, *next;
311 
312 	list_for_each_entry_safe(page, next, list, lru) {
313 		list_del(&page->lru);
314 		kimage_free_pages(page);
315 	}
316 }
317 
kimage_alloc_normal_control_pages(struct kimage * image,unsigned int order)318 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
319 							unsigned int order)
320 {
321 	/* Control pages are special, they are the intermediaries
322 	 * that are needed while we copy the rest of the pages
323 	 * to their final resting place.  As such they must
324 	 * not conflict with either the destination addresses
325 	 * or memory the kernel is already using.
326 	 *
327 	 * The only case where we really need more than one of
328 	 * these are for architectures where we cannot disable
329 	 * the MMU and must instead generate an identity mapped
330 	 * page table for all of the memory.
331 	 *
332 	 * At worst this runs in O(N) of the image size.
333 	 */
334 	struct list_head extra_pages;
335 	struct page *pages;
336 	unsigned int count;
337 
338 	count = 1 << order;
339 	INIT_LIST_HEAD(&extra_pages);
340 
341 	/* Loop while I can allocate a page and the page allocated
342 	 * is a destination page.
343 	 */
344 	do {
345 		unsigned long pfn, epfn, addr, eaddr;
346 
347 		pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
348 		if (!pages)
349 			break;
350 		pfn   = page_to_boot_pfn(pages);
351 		epfn  = pfn + count;
352 		addr  = pfn << PAGE_SHIFT;
353 		eaddr = (epfn << PAGE_SHIFT) - 1;
354 		if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
355 			      kimage_is_destination_range(image, addr, eaddr)) {
356 			list_add(&pages->lru, &extra_pages);
357 			pages = NULL;
358 		}
359 	} while (!pages);
360 
361 	if (pages) {
362 		/* Remember the allocated page... */
363 		list_add(&pages->lru, &image->control_pages);
364 
365 		/* Because the page is already in it's destination
366 		 * location we will never allocate another page at
367 		 * that address.  Therefore kimage_alloc_pages
368 		 * will not return it (again) and we don't need
369 		 * to give it an entry in image->segment[].
370 		 */
371 	}
372 	/* Deal with the destination pages I have inadvertently allocated.
373 	 *
374 	 * Ideally I would convert multi-page allocations into single
375 	 * page allocations, and add everything to image->dest_pages.
376 	 *
377 	 * For now it is simpler to just free the pages.
378 	 */
379 	kimage_free_page_list(&extra_pages);
380 
381 	return pages;
382 }
383 
384 #ifdef CONFIG_CRASH_DUMP
kimage_alloc_crash_control_pages(struct kimage * image,unsigned int order)385 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
386 						      unsigned int order)
387 {
388 	/* Control pages are special, they are the intermediaries
389 	 * that are needed while we copy the rest of the pages
390 	 * to their final resting place.  As such they must
391 	 * not conflict with either the destination addresses
392 	 * or memory the kernel is already using.
393 	 *
394 	 * Control pages are also the only pags we must allocate
395 	 * when loading a crash kernel.  All of the other pages
396 	 * are specified by the segments and we just memcpy
397 	 * into them directly.
398 	 *
399 	 * The only case where we really need more than one of
400 	 * these are for architectures where we cannot disable
401 	 * the MMU and must instead generate an identity mapped
402 	 * page table for all of the memory.
403 	 *
404 	 * Given the low demand this implements a very simple
405 	 * allocator that finds the first hole of the appropriate
406 	 * size in the reserved memory region, and allocates all
407 	 * of the memory up to and including the hole.
408 	 */
409 	unsigned long hole_start, hole_end, size;
410 	struct page *pages;
411 
412 	pages = NULL;
413 	size = (1 << order) << PAGE_SHIFT;
414 	hole_start = ALIGN(image->control_page, size);
415 	hole_end   = hole_start + size - 1;
416 	while (hole_end <= crashk_res.end) {
417 		unsigned long i;
418 
419 		cond_resched();
420 
421 		if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
422 			break;
423 		/* See if I overlap any of the segments */
424 		for (i = 0; i < image->nr_segments; i++) {
425 			unsigned long mstart, mend;
426 
427 			mstart = image->segment[i].mem;
428 			mend   = mstart + image->segment[i].memsz - 1;
429 			if ((hole_end >= mstart) && (hole_start <= mend)) {
430 				/* Advance the hole to the end of the segment */
431 				hole_start = ALIGN(mend, size);
432 				hole_end   = hole_start + size - 1;
433 				break;
434 			}
435 		}
436 		/* If I don't overlap any segments I have found my hole! */
437 		if (i == image->nr_segments) {
438 			pages = pfn_to_page(hole_start >> PAGE_SHIFT);
439 			image->control_page = hole_end + 1;
440 			break;
441 		}
442 	}
443 
444 	/* Ensure that these pages are decrypted if SME is enabled. */
445 	if (pages)
446 		arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
447 
448 	return pages;
449 }
450 #endif
451 
452 
kimage_alloc_control_pages(struct kimage * image,unsigned int order)453 struct page *kimage_alloc_control_pages(struct kimage *image,
454 					 unsigned int order)
455 {
456 	struct page *pages = NULL;
457 
458 	switch (image->type) {
459 	case KEXEC_TYPE_DEFAULT:
460 		pages = kimage_alloc_normal_control_pages(image, order);
461 		break;
462 #ifdef CONFIG_CRASH_DUMP
463 	case KEXEC_TYPE_CRASH:
464 		pages = kimage_alloc_crash_control_pages(image, order);
465 		break;
466 #endif
467 	}
468 
469 	return pages;
470 }
471 
kimage_add_entry(struct kimage * image,kimage_entry_t entry)472 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
473 {
474 	if (*image->entry != 0)
475 		image->entry++;
476 
477 	if (image->entry == image->last_entry) {
478 		kimage_entry_t *ind_page;
479 		struct page *page;
480 
481 		page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
482 		if (!page)
483 			return -ENOMEM;
484 
485 		ind_page = page_address(page);
486 		*image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
487 		image->entry = ind_page;
488 		image->last_entry = ind_page +
489 				      ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
490 	}
491 	*image->entry = entry;
492 	image->entry++;
493 	*image->entry = 0;
494 
495 	return 0;
496 }
497 
kimage_set_destination(struct kimage * image,unsigned long destination)498 static int kimage_set_destination(struct kimage *image,
499 				   unsigned long destination)
500 {
501 	destination &= PAGE_MASK;
502 
503 	return kimage_add_entry(image, destination | IND_DESTINATION);
504 }
505 
506 
kimage_add_page(struct kimage * image,unsigned long page)507 static int kimage_add_page(struct kimage *image, unsigned long page)
508 {
509 	page &= PAGE_MASK;
510 
511 	return kimage_add_entry(image, page | IND_SOURCE);
512 }
513 
514 
kimage_free_extra_pages(struct kimage * image)515 static void kimage_free_extra_pages(struct kimage *image)
516 {
517 	/* Walk through and free any extra destination pages I may have */
518 	kimage_free_page_list(&image->dest_pages);
519 
520 	/* Walk through and free any unusable pages I have cached */
521 	kimage_free_page_list(&image->unusable_pages);
522 
523 }
524 
kimage_terminate(struct kimage * image)525 void kimage_terminate(struct kimage *image)
526 {
527 	if (*image->entry != 0)
528 		image->entry++;
529 
530 	*image->entry = IND_DONE;
531 }
532 
533 #define for_each_kimage_entry(image, ptr, entry) \
534 	for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
535 		ptr = (entry & IND_INDIRECTION) ? \
536 			boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
537 
kimage_free_entry(kimage_entry_t entry)538 static void kimage_free_entry(kimage_entry_t entry)
539 {
540 	struct page *page;
541 
542 	page = boot_pfn_to_page(entry >> PAGE_SHIFT);
543 	kimage_free_pages(page);
544 }
545 
kimage_free(struct kimage * image)546 void kimage_free(struct kimage *image)
547 {
548 	kimage_entry_t *ptr, entry;
549 	kimage_entry_t ind = 0;
550 
551 	if (!image)
552 		return;
553 
554 #ifdef CONFIG_CRASH_DUMP
555 	if (image->vmcoreinfo_data_copy) {
556 		crash_update_vmcoreinfo_safecopy(NULL);
557 		vunmap(image->vmcoreinfo_data_copy);
558 	}
559 #endif
560 
561 	kimage_free_extra_pages(image);
562 	for_each_kimage_entry(image, ptr, entry) {
563 		if (entry & IND_INDIRECTION) {
564 			/* Free the previous indirection page */
565 			if (ind & IND_INDIRECTION)
566 				kimage_free_entry(ind);
567 			/* Save this indirection page until we are
568 			 * done with it.
569 			 */
570 			ind = entry;
571 		} else if (entry & IND_SOURCE)
572 			kimage_free_entry(entry);
573 	}
574 	/* Free the final indirection page */
575 	if (ind & IND_INDIRECTION)
576 		kimage_free_entry(ind);
577 
578 	/* Handle any machine specific cleanup */
579 	machine_kexec_cleanup(image);
580 
581 	/* Free the kexec control pages... */
582 	kimage_free_page_list(&image->control_pages);
583 
584 	/*
585 	 * Free up any temporary buffers allocated. This might hit if
586 	 * error occurred much later after buffer allocation.
587 	 */
588 	if (image->file_mode)
589 		kimage_file_post_load_cleanup(image);
590 
591 	kfree(image);
592 }
593 
kimage_dst_used(struct kimage * image,unsigned long page)594 static kimage_entry_t *kimage_dst_used(struct kimage *image,
595 					unsigned long page)
596 {
597 	kimage_entry_t *ptr, entry;
598 	unsigned long destination = 0;
599 
600 	for_each_kimage_entry(image, ptr, entry) {
601 		if (entry & IND_DESTINATION)
602 			destination = entry & PAGE_MASK;
603 		else if (entry & IND_SOURCE) {
604 			if (page == destination)
605 				return ptr;
606 			destination += PAGE_SIZE;
607 		}
608 	}
609 
610 	return NULL;
611 }
612 
kimage_alloc_page(struct kimage * image,gfp_t gfp_mask,unsigned long destination)613 static struct page *kimage_alloc_page(struct kimage *image,
614 					gfp_t gfp_mask,
615 					unsigned long destination)
616 {
617 	/*
618 	 * Here we implement safeguards to ensure that a source page
619 	 * is not copied to its destination page before the data on
620 	 * the destination page is no longer useful.
621 	 *
622 	 * To do this we maintain the invariant that a source page is
623 	 * either its own destination page, or it is not a
624 	 * destination page at all.
625 	 *
626 	 * That is slightly stronger than required, but the proof
627 	 * that no problems will not occur is trivial, and the
628 	 * implementation is simply to verify.
629 	 *
630 	 * When allocating all pages normally this algorithm will run
631 	 * in O(N) time, but in the worst case it will run in O(N^2)
632 	 * time.   If the runtime is a problem the data structures can
633 	 * be fixed.
634 	 */
635 	struct page *page;
636 	unsigned long addr;
637 
638 	/*
639 	 * Walk through the list of destination pages, and see if I
640 	 * have a match.
641 	 */
642 	list_for_each_entry(page, &image->dest_pages, lru) {
643 		addr = page_to_boot_pfn(page) << PAGE_SHIFT;
644 		if (addr == destination) {
645 			list_del(&page->lru);
646 			return page;
647 		}
648 	}
649 	page = NULL;
650 	while (1) {
651 		kimage_entry_t *old;
652 
653 		/* Allocate a page, if we run out of memory give up */
654 		page = kimage_alloc_pages(gfp_mask, 0);
655 		if (!page)
656 			return NULL;
657 		/* If the page cannot be used file it away */
658 		if (page_to_boot_pfn(page) >
659 				(KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
660 			list_add(&page->lru, &image->unusable_pages);
661 			continue;
662 		}
663 		addr = page_to_boot_pfn(page) << PAGE_SHIFT;
664 
665 		/* If it is the destination page we want use it */
666 		if (addr == destination)
667 			break;
668 
669 		/* If the page is not a destination page use it */
670 		if (!kimage_is_destination_range(image, addr,
671 						  addr + PAGE_SIZE - 1))
672 			break;
673 
674 		/*
675 		 * I know that the page is someones destination page.
676 		 * See if there is already a source page for this
677 		 * destination page.  And if so swap the source pages.
678 		 */
679 		old = kimage_dst_used(image, addr);
680 		if (old) {
681 			/* If so move it */
682 			unsigned long old_addr;
683 			struct page *old_page;
684 
685 			old_addr = *old & PAGE_MASK;
686 			old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
687 			copy_highpage(page, old_page);
688 			*old = addr | (*old & ~PAGE_MASK);
689 
690 			/* The old page I have found cannot be a
691 			 * destination page, so return it if it's
692 			 * gfp_flags honor the ones passed in.
693 			 */
694 			if (!(gfp_mask & __GFP_HIGHMEM) &&
695 			    PageHighMem(old_page)) {
696 				kimage_free_pages(old_page);
697 				continue;
698 			}
699 			page = old_page;
700 			break;
701 		}
702 		/* Place the page on the destination list, to be used later */
703 		list_add(&page->lru, &image->dest_pages);
704 	}
705 
706 	return page;
707 }
708 
kimage_load_normal_segment(struct kimage * image,struct kexec_segment * segment)709 static int kimage_load_normal_segment(struct kimage *image,
710 					 struct kexec_segment *segment)
711 {
712 	unsigned long maddr;
713 	size_t ubytes, mbytes;
714 	int result;
715 	unsigned char __user *buf = NULL;
716 	unsigned char *kbuf = NULL;
717 
718 	if (image->file_mode)
719 		kbuf = segment->kbuf;
720 	else
721 		buf = segment->buf;
722 	ubytes = segment->bufsz;
723 	mbytes = segment->memsz;
724 	maddr = segment->mem;
725 
726 	result = kimage_set_destination(image, maddr);
727 	if (result < 0)
728 		goto out;
729 
730 	while (mbytes) {
731 		struct page *page;
732 		char *ptr;
733 		size_t uchunk, mchunk;
734 
735 		page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
736 		if (!page) {
737 			result  = -ENOMEM;
738 			goto out;
739 		}
740 		result = kimage_add_page(image, page_to_boot_pfn(page)
741 								<< PAGE_SHIFT);
742 		if (result < 0)
743 			goto out;
744 
745 		ptr = kmap_local_page(page);
746 		/* Start with a clear page */
747 		clear_page(ptr);
748 		ptr += maddr & ~PAGE_MASK;
749 		mchunk = min_t(size_t, mbytes,
750 				PAGE_SIZE - (maddr & ~PAGE_MASK));
751 		uchunk = min(ubytes, mchunk);
752 
753 		if (uchunk) {
754 			/* For file based kexec, source pages are in kernel memory */
755 			if (image->file_mode)
756 				memcpy(ptr, kbuf, uchunk);
757 			else
758 				result = copy_from_user(ptr, buf, uchunk);
759 			ubytes -= uchunk;
760 			if (image->file_mode)
761 				kbuf += uchunk;
762 			else
763 				buf += uchunk;
764 		}
765 		kunmap_local(ptr);
766 		if (result) {
767 			result = -EFAULT;
768 			goto out;
769 		}
770 		maddr  += mchunk;
771 		mbytes -= mchunk;
772 
773 		cond_resched();
774 	}
775 out:
776 	return result;
777 }
778 
779 #ifdef CONFIG_CRASH_DUMP
kimage_load_crash_segment(struct kimage * image,struct kexec_segment * segment)780 static int kimage_load_crash_segment(struct kimage *image,
781 					struct kexec_segment *segment)
782 {
783 	/* For crash dumps kernels we simply copy the data from
784 	 * user space to it's destination.
785 	 * We do things a page at a time for the sake of kmap.
786 	 */
787 	unsigned long maddr;
788 	size_t ubytes, mbytes;
789 	int result;
790 	unsigned char __user *buf = NULL;
791 	unsigned char *kbuf = NULL;
792 
793 	result = 0;
794 	if (image->file_mode)
795 		kbuf = segment->kbuf;
796 	else
797 		buf = segment->buf;
798 	ubytes = segment->bufsz;
799 	mbytes = segment->memsz;
800 	maddr = segment->mem;
801 	while (mbytes) {
802 		struct page *page;
803 		char *ptr;
804 		size_t uchunk, mchunk;
805 
806 		page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
807 		if (!page) {
808 			result  = -ENOMEM;
809 			goto out;
810 		}
811 		arch_kexec_post_alloc_pages(page_address(page), 1, 0);
812 		ptr = kmap_local_page(page);
813 		ptr += maddr & ~PAGE_MASK;
814 		mchunk = min_t(size_t, mbytes,
815 				PAGE_SIZE - (maddr & ~PAGE_MASK));
816 		uchunk = min(ubytes, mchunk);
817 		if (mchunk > uchunk) {
818 			/* Zero the trailing part of the page */
819 			memset(ptr + uchunk, 0, mchunk - uchunk);
820 		}
821 
822 		if (uchunk) {
823 			/* For file based kexec, source pages are in kernel memory */
824 			if (image->file_mode)
825 				memcpy(ptr, kbuf, uchunk);
826 			else
827 				result = copy_from_user(ptr, buf, uchunk);
828 			ubytes -= uchunk;
829 			if (image->file_mode)
830 				kbuf += uchunk;
831 			else
832 				buf += uchunk;
833 		}
834 		kexec_flush_icache_page(page);
835 		kunmap_local(ptr);
836 		arch_kexec_pre_free_pages(page_address(page), 1);
837 		if (result) {
838 			result = -EFAULT;
839 			goto out;
840 		}
841 		maddr  += mchunk;
842 		mbytes -= mchunk;
843 
844 		cond_resched();
845 	}
846 out:
847 	return result;
848 }
849 #endif
850 
kimage_load_segment(struct kimage * image,struct kexec_segment * segment)851 int kimage_load_segment(struct kimage *image,
852 				struct kexec_segment *segment)
853 {
854 	int result = -ENOMEM;
855 
856 	switch (image->type) {
857 	case KEXEC_TYPE_DEFAULT:
858 		result = kimage_load_normal_segment(image, segment);
859 		break;
860 #ifdef CONFIG_CRASH_DUMP
861 	case KEXEC_TYPE_CRASH:
862 		result = kimage_load_crash_segment(image, segment);
863 		break;
864 #endif
865 	}
866 
867 	return result;
868 }
869 
870 struct kexec_load_limit {
871 	/* Mutex protects the limit count. */
872 	struct mutex mutex;
873 	int limit;
874 };
875 
876 static struct kexec_load_limit load_limit_reboot = {
877 	.mutex = __MUTEX_INITIALIZER(load_limit_reboot.mutex),
878 	.limit = -1,
879 };
880 
881 static struct kexec_load_limit load_limit_panic = {
882 	.mutex = __MUTEX_INITIALIZER(load_limit_panic.mutex),
883 	.limit = -1,
884 };
885 
886 struct kimage *kexec_image;
887 struct kimage *kexec_crash_image;
888 static int kexec_load_disabled;
889 
890 #ifdef CONFIG_SYSCTL
kexec_limit_handler(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)891 static int kexec_limit_handler(const struct ctl_table *table, int write,
892 			       void *buffer, size_t *lenp, loff_t *ppos)
893 {
894 	struct kexec_load_limit *limit = table->data;
895 	int val;
896 	struct ctl_table tmp = {
897 		.data = &val,
898 		.maxlen = sizeof(val),
899 		.mode = table->mode,
900 	};
901 	int ret;
902 
903 	if (write) {
904 		ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
905 		if (ret)
906 			return ret;
907 
908 		if (val < 0)
909 			return -EINVAL;
910 
911 		mutex_lock(&limit->mutex);
912 		if (limit->limit != -1 && val >= limit->limit)
913 			ret = -EINVAL;
914 		else
915 			limit->limit = val;
916 		mutex_unlock(&limit->mutex);
917 
918 		return ret;
919 	}
920 
921 	mutex_lock(&limit->mutex);
922 	val = limit->limit;
923 	mutex_unlock(&limit->mutex);
924 
925 	return proc_dointvec(&tmp, write, buffer, lenp, ppos);
926 }
927 
928 static struct ctl_table kexec_core_sysctls[] = {
929 	{
930 		.procname	= "kexec_load_disabled",
931 		.data		= &kexec_load_disabled,
932 		.maxlen		= sizeof(int),
933 		.mode		= 0644,
934 		/* only handle a transition from default "0" to "1" */
935 		.proc_handler	= proc_dointvec_minmax,
936 		.extra1		= SYSCTL_ONE,
937 		.extra2		= SYSCTL_ONE,
938 	},
939 	{
940 		.procname	= "kexec_load_limit_panic",
941 		.data		= &load_limit_panic,
942 		.mode		= 0644,
943 		.proc_handler	= kexec_limit_handler,
944 	},
945 	{
946 		.procname	= "kexec_load_limit_reboot",
947 		.data		= &load_limit_reboot,
948 		.mode		= 0644,
949 		.proc_handler	= kexec_limit_handler,
950 	},
951 };
952 
kexec_core_sysctl_init(void)953 static int __init kexec_core_sysctl_init(void)
954 {
955 	register_sysctl_init("kernel", kexec_core_sysctls);
956 	return 0;
957 }
958 late_initcall(kexec_core_sysctl_init);
959 #endif
960 
kexec_load_permitted(int kexec_image_type)961 bool kexec_load_permitted(int kexec_image_type)
962 {
963 	struct kexec_load_limit *limit;
964 
965 	/*
966 	 * Only the superuser can use the kexec syscall and if it has not
967 	 * been disabled.
968 	 */
969 	if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
970 		return false;
971 
972 	/* Check limit counter and decrease it.*/
973 	limit = (kexec_image_type == KEXEC_TYPE_CRASH) ?
974 		&load_limit_panic : &load_limit_reboot;
975 	mutex_lock(&limit->mutex);
976 	if (!limit->limit) {
977 		mutex_unlock(&limit->mutex);
978 		return false;
979 	}
980 	if (limit->limit != -1)
981 		limit->limit--;
982 	mutex_unlock(&limit->mutex);
983 
984 	return true;
985 }
986 
987 /*
988  * Move into place and start executing a preloaded standalone
989  * executable.  If nothing was preloaded return an error.
990  */
kernel_kexec(void)991 int kernel_kexec(void)
992 {
993 	int error = 0;
994 
995 	if (!kexec_trylock())
996 		return -EBUSY;
997 	if (!kexec_image) {
998 		error = -EINVAL;
999 		goto Unlock;
1000 	}
1001 
1002 #ifdef CONFIG_KEXEC_JUMP
1003 	if (kexec_image->preserve_context) {
1004 		pm_prepare_console();
1005 		error = freeze_processes();
1006 		if (error) {
1007 			error = -EBUSY;
1008 			goto Restore_console;
1009 		}
1010 		suspend_console();
1011 		error = dpm_suspend_start(PMSG_FREEZE);
1012 		if (error)
1013 			goto Resume_console;
1014 		/* At this point, dpm_suspend_start() has been called,
1015 		 * but *not* dpm_suspend_end(). We *must* call
1016 		 * dpm_suspend_end() now.  Otherwise, drivers for
1017 		 * some devices (e.g. interrupt controllers) become
1018 		 * desynchronized with the actual state of the
1019 		 * hardware at resume time, and evil weirdness ensues.
1020 		 */
1021 		error = dpm_suspend_end(PMSG_FREEZE);
1022 		if (error)
1023 			goto Resume_devices;
1024 		error = suspend_disable_secondary_cpus();
1025 		if (error)
1026 			goto Enable_cpus;
1027 		local_irq_disable();
1028 		error = syscore_suspend();
1029 		if (error)
1030 			goto Enable_irqs;
1031 	} else
1032 #endif
1033 	{
1034 		kexec_in_progress = true;
1035 		kernel_restart_prepare("kexec reboot");
1036 		migrate_to_reboot_cpu();
1037 		syscore_shutdown();
1038 
1039 		/*
1040 		 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1041 		 * no further code needs to use CPU hotplug (which is true in
1042 		 * the reboot case). However, the kexec path depends on using
1043 		 * CPU hotplug again; so re-enable it here.
1044 		 */
1045 		cpu_hotplug_enable();
1046 		pr_notice("Starting new kernel\n");
1047 		machine_shutdown();
1048 	}
1049 
1050 	kmsg_dump(KMSG_DUMP_SHUTDOWN);
1051 	machine_kexec(kexec_image);
1052 
1053 #ifdef CONFIG_KEXEC_JUMP
1054 	if (kexec_image->preserve_context) {
1055 		syscore_resume();
1056  Enable_irqs:
1057 		local_irq_enable();
1058  Enable_cpus:
1059 		suspend_enable_secondary_cpus();
1060 		dpm_resume_start(PMSG_RESTORE);
1061  Resume_devices:
1062 		dpm_resume_end(PMSG_RESTORE);
1063  Resume_console:
1064 		resume_console();
1065 		thaw_processes();
1066  Restore_console:
1067 		pm_restore_console();
1068 	}
1069 #endif
1070 
1071  Unlock:
1072 	kexec_unlock();
1073 	return error;
1074 }
1075