Loading...
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/capability.h>
10#include <linux/mm.h>
11#include <linux/file.h>
12#include <linux/slab.h>
13#include <linux/fs.h>
14#include <linux/kexec.h>
15#include <linux/mutex.h>
16#include <linux/list.h>
17#include <linux/highmem.h>
18#include <linux/syscalls.h>
19#include <linux/reboot.h>
20#include <linux/ioport.h>
21#include <linux/hardirq.h>
22#include <linux/elf.h>
23#include <linux/elfcore.h>
24#include <linux/utsname.h>
25#include <linux/numa.h>
26#include <linux/suspend.h>
27#include <linux/device.h>
28#include <linux/freezer.h>
29#include <linux/pm.h>
30#include <linux/cpu.h>
31#include <linux/uaccess.h>
32#include <linux/io.h>
33#include <linux/console.h>
34#include <linux/vmalloc.h>
35#include <linux/swap.h>
36#include <linux/syscore_ops.h>
37#include <linux/compiler.h>
38#include <linux/hugetlb.h>
39#include <linux/frame.h>
40
41#include <asm/page.h>
42#include <asm/sections.h>
43
44#include <crypto/hash.h>
45#include <crypto/sha.h>
46#include "kexec_internal.h"
47
48DEFINE_MUTEX(kexec_mutex);
49
50/* Per cpu memory for storing cpu states in case of system crash. */
51note_buf_t __percpu *crash_notes;
52
53/* Flag to indicate we are going to kexec a new kernel */
54bool kexec_in_progress = false;
55
56
57/* Location of the reserved area for the crash kernel */
58struct resource crashk_res = {
59 .name = "Crash kernel",
60 .start = 0,
61 .end = 0,
62 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
63 .desc = IORES_DESC_CRASH_KERNEL
64};
65struct resource crashk_low_res = {
66 .name = "Crash kernel",
67 .start = 0,
68 .end = 0,
69 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
70 .desc = IORES_DESC_CRASH_KERNEL
71};
72
73int kexec_should_crash(struct task_struct *p)
74{
75 /*
76 * If crash_kexec_post_notifiers is enabled, don't run
77 * crash_kexec() here yet, which must be run after panic
78 * notifiers in panic().
79 */
80 if (crash_kexec_post_notifiers)
81 return 0;
82 /*
83 * There are 4 panic() calls in do_exit() path, each of which
84 * corresponds to each of these 4 conditions.
85 */
86 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
87 return 1;
88 return 0;
89}
90
91int kexec_crash_loaded(void)
92{
93 return !!kexec_crash_image;
94}
95EXPORT_SYMBOL_GPL(kexec_crash_loaded);
96
97/*
98 * When kexec transitions to the new kernel there is a one-to-one
99 * mapping between physical and virtual addresses. On processors
100 * where you can disable the MMU this is trivial, and easy. For
101 * others it is still a simple predictable page table to setup.
102 *
103 * In that environment kexec copies the new kernel to its final
104 * resting place. This means I can only support memory whose
105 * physical address can fit in an unsigned long. In particular
106 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
107 * If the assembly stub has more restrictive requirements
108 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
109 * defined more restrictively in <asm/kexec.h>.
110 *
111 * The code for the transition from the current kernel to the
112 * the new kernel is placed in the control_code_buffer, whose size
113 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
114 * page of memory is necessary, but some architectures require more.
115 * Because this memory must be identity mapped in the transition from
116 * virtual to physical addresses it must live in the range
117 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
118 * modifiable.
119 *
120 * The assembly stub in the control code buffer is passed a linked list
121 * of descriptor pages detailing the source pages of the new kernel,
122 * and the destination addresses of those source pages. As this data
123 * structure is not used in the context of the current OS, it must
124 * be self-contained.
125 *
126 * The code has been made to work with highmem pages and will use a
127 * destination page in its final resting place (if it happens
128 * to allocate it). The end product of this is that most of the
129 * physical address space, and most of RAM can be used.
130 *
131 * Future directions include:
132 * - allocating a page table with the control code buffer identity
133 * mapped, to simplify machine_kexec and make kexec_on_panic more
134 * reliable.
135 */
136
137/*
138 * KIMAGE_NO_DEST is an impossible destination address..., for
139 * allocating pages whose destination address we do not care about.
140 */
141#define KIMAGE_NO_DEST (-1UL)
142#define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
143
144static struct page *kimage_alloc_page(struct kimage *image,
145 gfp_t gfp_mask,
146 unsigned long dest);
147
148int sanity_check_segment_list(struct kimage *image)
149{
150 int i;
151 unsigned long nr_segments = image->nr_segments;
152 unsigned long total_pages = 0;
153 unsigned long nr_pages = totalram_pages();
154
155 /*
156 * Verify we have good destination addresses. The caller is
157 * responsible for making certain we don't attempt to load
158 * the new image into invalid or reserved areas of RAM. This
159 * just verifies it is an address we can use.
160 *
161 * Since the kernel does everything in page size chunks ensure
162 * the destination addresses are page aligned. Too many
163 * special cases crop of when we don't do this. The most
164 * insidious is getting overlapping destination addresses
165 * simply because addresses are changed to page size
166 * granularity.
167 */
168 for (i = 0; i < nr_segments; i++) {
169 unsigned long mstart, mend;
170
171 mstart = image->segment[i].mem;
172 mend = mstart + image->segment[i].memsz;
173 if (mstart > mend)
174 return -EADDRNOTAVAIL;
175 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
176 return -EADDRNOTAVAIL;
177 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
178 return -EADDRNOTAVAIL;
179 }
180
181 /* Verify our destination addresses do not overlap.
182 * If we alloed overlapping destination addresses
183 * through very weird things can happen with no
184 * easy explanation as one segment stops on another.
185 */
186 for (i = 0; i < nr_segments; i++) {
187 unsigned long mstart, mend;
188 unsigned long j;
189
190 mstart = image->segment[i].mem;
191 mend = mstart + image->segment[i].memsz;
192 for (j = 0; j < i; j++) {
193 unsigned long pstart, pend;
194
195 pstart = image->segment[j].mem;
196 pend = pstart + image->segment[j].memsz;
197 /* Do the segments overlap ? */
198 if ((mend > pstart) && (mstart < pend))
199 return -EINVAL;
200 }
201 }
202
203 /* Ensure our buffer sizes are strictly less than
204 * our memory sizes. This should always be the case,
205 * and it is easier to check up front than to be surprised
206 * later on.
207 */
208 for (i = 0; i < nr_segments; i++) {
209 if (image->segment[i].bufsz > image->segment[i].memsz)
210 return -EINVAL;
211 }
212
213 /*
214 * Verify that no more than half of memory will be consumed. If the
215 * request from userspace is too large, a large amount of time will be
216 * wasted allocating pages, which can cause a soft lockup.
217 */
218 for (i = 0; i < nr_segments; i++) {
219 if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
220 return -EINVAL;
221
222 total_pages += PAGE_COUNT(image->segment[i].memsz);
223 }
224
225 if (total_pages > nr_pages / 2)
226 return -EINVAL;
227
228 /*
229 * Verify we have good destination addresses. Normally
230 * the caller is responsible for making certain we don't
231 * attempt to load the new image into invalid or reserved
232 * areas of RAM. But crash kernels are preloaded into a
233 * reserved area of ram. We must ensure the addresses
234 * are in the reserved area otherwise preloading the
235 * kernel could corrupt things.
236 */
237
238 if (image->type == KEXEC_TYPE_CRASH) {
239 for (i = 0; i < nr_segments; i++) {
240 unsigned long mstart, mend;
241
242 mstart = image->segment[i].mem;
243 mend = mstart + image->segment[i].memsz - 1;
244 /* Ensure we are within the crash kernel limits */
245 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
246 (mend > phys_to_boot_phys(crashk_res.end)))
247 return -EADDRNOTAVAIL;
248 }
249 }
250
251 return 0;
252}
253
254struct kimage *do_kimage_alloc_init(void)
255{
256 struct kimage *image;
257
258 /* Allocate a controlling structure */
259 image = kzalloc(sizeof(*image), GFP_KERNEL);
260 if (!image)
261 return NULL;
262
263 image->head = 0;
264 image->entry = &image->head;
265 image->last_entry = &image->head;
266 image->control_page = ~0; /* By default this does not apply */
267 image->type = KEXEC_TYPE_DEFAULT;
268
269 /* Initialize the list of control pages */
270 INIT_LIST_HEAD(&image->control_pages);
271
272 /* Initialize the list of destination pages */
273 INIT_LIST_HEAD(&image->dest_pages);
274
275 /* Initialize the list of unusable pages */
276 INIT_LIST_HEAD(&image->unusable_pages);
277
278 return image;
279}
280
281int kimage_is_destination_range(struct kimage *image,
282 unsigned long start,
283 unsigned long end)
284{
285 unsigned long i;
286
287 for (i = 0; i < image->nr_segments; i++) {
288 unsigned long mstart, mend;
289
290 mstart = image->segment[i].mem;
291 mend = mstart + image->segment[i].memsz;
292 if ((end > mstart) && (start < mend))
293 return 1;
294 }
295
296 return 0;
297}
298
299static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
300{
301 struct page *pages;
302
303 if (fatal_signal_pending(current))
304 return NULL;
305 pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
306 if (pages) {
307 unsigned int count, i;
308
309 pages->mapping = NULL;
310 set_page_private(pages, order);
311 count = 1 << order;
312 for (i = 0; i < count; i++)
313 SetPageReserved(pages + i);
314
315 arch_kexec_post_alloc_pages(page_address(pages), count,
316 gfp_mask);
317
318 if (gfp_mask & __GFP_ZERO)
319 for (i = 0; i < count; i++)
320 clear_highpage(pages + i);
321 }
322
323 return pages;
324}
325
326static void kimage_free_pages(struct page *page)
327{
328 unsigned int order, count, i;
329
330 order = page_private(page);
331 count = 1 << order;
332
333 arch_kexec_pre_free_pages(page_address(page), count);
334
335 for (i = 0; i < count; i++)
336 ClearPageReserved(page + i);
337 __free_pages(page, order);
338}
339
340void kimage_free_page_list(struct list_head *list)
341{
342 struct page *page, *next;
343
344 list_for_each_entry_safe(page, next, list, lru) {
345 list_del(&page->lru);
346 kimage_free_pages(page);
347 }
348}
349
350static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
351 unsigned int order)
352{
353 /* Control pages are special, they are the intermediaries
354 * that are needed while we copy the rest of the pages
355 * to their final resting place. As such they must
356 * not conflict with either the destination addresses
357 * or memory the kernel is already using.
358 *
359 * The only case where we really need more than one of
360 * these are for architectures where we cannot disable
361 * the MMU and must instead generate an identity mapped
362 * page table for all of the memory.
363 *
364 * At worst this runs in O(N) of the image size.
365 */
366 struct list_head extra_pages;
367 struct page *pages;
368 unsigned int count;
369
370 count = 1 << order;
371 INIT_LIST_HEAD(&extra_pages);
372
373 /* Loop while I can allocate a page and the page allocated
374 * is a destination page.
375 */
376 do {
377 unsigned long pfn, epfn, addr, eaddr;
378
379 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
380 if (!pages)
381 break;
382 pfn = page_to_boot_pfn(pages);
383 epfn = pfn + count;
384 addr = pfn << PAGE_SHIFT;
385 eaddr = epfn << PAGE_SHIFT;
386 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
387 kimage_is_destination_range(image, addr, eaddr)) {
388 list_add(&pages->lru, &extra_pages);
389 pages = NULL;
390 }
391 } while (!pages);
392
393 if (pages) {
394 /* Remember the allocated page... */
395 list_add(&pages->lru, &image->control_pages);
396
397 /* Because the page is already in it's destination
398 * location we will never allocate another page at
399 * that address. Therefore kimage_alloc_pages
400 * will not return it (again) and we don't need
401 * to give it an entry in image->segment[].
402 */
403 }
404 /* Deal with the destination pages I have inadvertently allocated.
405 *
406 * Ideally I would convert multi-page allocations into single
407 * page allocations, and add everything to image->dest_pages.
408 *
409 * For now it is simpler to just free the pages.
410 */
411 kimage_free_page_list(&extra_pages);
412
413 return pages;
414}
415
416static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
417 unsigned int order)
418{
419 /* Control pages are special, they are the intermediaries
420 * that are needed while we copy the rest of the pages
421 * to their final resting place. As such they must
422 * not conflict with either the destination addresses
423 * or memory the kernel is already using.
424 *
425 * Control pages are also the only pags we must allocate
426 * when loading a crash kernel. All of the other pages
427 * are specified by the segments and we just memcpy
428 * into them directly.
429 *
430 * The only case where we really need more than one of
431 * these are for architectures where we cannot disable
432 * the MMU and must instead generate an identity mapped
433 * page table for all of the memory.
434 *
435 * Given the low demand this implements a very simple
436 * allocator that finds the first hole of the appropriate
437 * size in the reserved memory region, and allocates all
438 * of the memory up to and including the hole.
439 */
440 unsigned long hole_start, hole_end, size;
441 struct page *pages;
442
443 pages = NULL;
444 size = (1 << order) << PAGE_SHIFT;
445 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
446 hole_end = hole_start + size - 1;
447 while (hole_end <= crashk_res.end) {
448 unsigned long i;
449
450 cond_resched();
451
452 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
453 break;
454 /* See if I overlap any of the segments */
455 for (i = 0; i < image->nr_segments; i++) {
456 unsigned long mstart, mend;
457
458 mstart = image->segment[i].mem;
459 mend = mstart + image->segment[i].memsz - 1;
460 if ((hole_end >= mstart) && (hole_start <= mend)) {
461 /* Advance the hole to the end of the segment */
462 hole_start = (mend + (size - 1)) & ~(size - 1);
463 hole_end = hole_start + size - 1;
464 break;
465 }
466 }
467 /* If I don't overlap any segments I have found my hole! */
468 if (i == image->nr_segments) {
469 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
470 image->control_page = hole_end;
471 break;
472 }
473 }
474
475 /* Ensure that these pages are decrypted if SME is enabled. */
476 if (pages)
477 arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
478
479 return pages;
480}
481
482
483struct page *kimage_alloc_control_pages(struct kimage *image,
484 unsigned int order)
485{
486 struct page *pages = NULL;
487
488 switch (image->type) {
489 case KEXEC_TYPE_DEFAULT:
490 pages = kimage_alloc_normal_control_pages(image, order);
491 break;
492 case KEXEC_TYPE_CRASH:
493 pages = kimage_alloc_crash_control_pages(image, order);
494 break;
495 }
496
497 return pages;
498}
499
500int kimage_crash_copy_vmcoreinfo(struct kimage *image)
501{
502 struct page *vmcoreinfo_page;
503 void *safecopy;
504
505 if (image->type != KEXEC_TYPE_CRASH)
506 return 0;
507
508 /*
509 * For kdump, allocate one vmcoreinfo safe copy from the
510 * crash memory. as we have arch_kexec_protect_crashkres()
511 * after kexec syscall, we naturally protect it from write
512 * (even read) access under kernel direct mapping. But on
513 * the other hand, we still need to operate it when crash
514 * happens to generate vmcoreinfo note, hereby we rely on
515 * vmap for this purpose.
516 */
517 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
518 if (!vmcoreinfo_page) {
519 pr_warn("Could not allocate vmcoreinfo buffer\n");
520 return -ENOMEM;
521 }
522 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
523 if (!safecopy) {
524 pr_warn("Could not vmap vmcoreinfo buffer\n");
525 return -ENOMEM;
526 }
527
528 image->vmcoreinfo_data_copy = safecopy;
529 crash_update_vmcoreinfo_safecopy(safecopy);
530
531 return 0;
532}
533
534static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
535{
536 if (*image->entry != 0)
537 image->entry++;
538
539 if (image->entry == image->last_entry) {
540 kimage_entry_t *ind_page;
541 struct page *page;
542
543 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
544 if (!page)
545 return -ENOMEM;
546
547 ind_page = page_address(page);
548 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
549 image->entry = ind_page;
550 image->last_entry = ind_page +
551 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
552 }
553 *image->entry = entry;
554 image->entry++;
555 *image->entry = 0;
556
557 return 0;
558}
559
560static int kimage_set_destination(struct kimage *image,
561 unsigned long destination)
562{
563 int result;
564
565 destination &= PAGE_MASK;
566 result = kimage_add_entry(image, destination | IND_DESTINATION);
567
568 return result;
569}
570
571
572static int kimage_add_page(struct kimage *image, unsigned long page)
573{
574 int result;
575
576 page &= PAGE_MASK;
577 result = kimage_add_entry(image, page | IND_SOURCE);
578
579 return result;
580}
581
582
583static void kimage_free_extra_pages(struct kimage *image)
584{
585 /* Walk through and free any extra destination pages I may have */
586 kimage_free_page_list(&image->dest_pages);
587
588 /* Walk through and free any unusable pages I have cached */
589 kimage_free_page_list(&image->unusable_pages);
590
591}
592void 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
605static 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
613void 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
659static 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
678static 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 addr = old_addr;
765 page = old_page;
766 break;
767 }
768 /* Place the page on the destination list, to be used later */
769 list_add(&page->lru, &image->dest_pages);
770 }
771
772 return page;
773}
774
775static int kimage_load_normal_segment(struct kimage *image,
776 struct kexec_segment *segment)
777{
778 unsigned long maddr;
779 size_t ubytes, mbytes;
780 int result;
781 unsigned char __user *buf = NULL;
782 unsigned char *kbuf = NULL;
783
784 result = 0;
785 if (image->file_mode)
786 kbuf = segment->kbuf;
787 else
788 buf = segment->buf;
789 ubytes = segment->bufsz;
790 mbytes = segment->memsz;
791 maddr = segment->mem;
792
793 result = kimage_set_destination(image, maddr);
794 if (result < 0)
795 goto out;
796
797 while (mbytes) {
798 struct page *page;
799 char *ptr;
800 size_t uchunk, mchunk;
801
802 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
803 if (!page) {
804 result = -ENOMEM;
805 goto out;
806 }
807 result = kimage_add_page(image, page_to_boot_pfn(page)
808 << PAGE_SHIFT);
809 if (result < 0)
810 goto out;
811
812 ptr = kmap(page);
813 /* Start with a clear page */
814 clear_page(ptr);
815 ptr += maddr & ~PAGE_MASK;
816 mchunk = min_t(size_t, mbytes,
817 PAGE_SIZE - (maddr & ~PAGE_MASK));
818 uchunk = min(ubytes, mchunk);
819
820 /* For file based kexec, source pages are in kernel memory */
821 if (image->file_mode)
822 memcpy(ptr, kbuf, uchunk);
823 else
824 result = copy_from_user(ptr, buf, uchunk);
825 kunmap(page);
826 if (result) {
827 result = -EFAULT;
828 goto out;
829 }
830 ubytes -= uchunk;
831 maddr += mchunk;
832 if (image->file_mode)
833 kbuf += mchunk;
834 else
835 buf += mchunk;
836 mbytes -= mchunk;
837
838 cond_resched();
839 }
840out:
841 return result;
842}
843
844static int kimage_load_crash_segment(struct kimage *image,
845 struct kexec_segment *segment)
846{
847 /* For crash dumps kernels we simply copy the data from
848 * user space to it's destination.
849 * We do things a page at a time for the sake of kmap.
850 */
851 unsigned long maddr;
852 size_t ubytes, mbytes;
853 int result;
854 unsigned char __user *buf = NULL;
855 unsigned char *kbuf = NULL;
856
857 result = 0;
858 if (image->file_mode)
859 kbuf = segment->kbuf;
860 else
861 buf = segment->buf;
862 ubytes = segment->bufsz;
863 mbytes = segment->memsz;
864 maddr = segment->mem;
865 while (mbytes) {
866 struct page *page;
867 char *ptr;
868 size_t uchunk, mchunk;
869
870 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
871 if (!page) {
872 result = -ENOMEM;
873 goto out;
874 }
875 arch_kexec_post_alloc_pages(page_address(page), 1, 0);
876 ptr = kmap(page);
877 ptr += maddr & ~PAGE_MASK;
878 mchunk = min_t(size_t, mbytes,
879 PAGE_SIZE - (maddr & ~PAGE_MASK));
880 uchunk = min(ubytes, mchunk);
881 if (mchunk > uchunk) {
882 /* Zero the trailing part of the page */
883 memset(ptr + uchunk, 0, mchunk - uchunk);
884 }
885
886 /* For file based kexec, source pages are in kernel memory */
887 if (image->file_mode)
888 memcpy(ptr, kbuf, uchunk);
889 else
890 result = copy_from_user(ptr, buf, uchunk);
891 kexec_flush_icache_page(page);
892 kunmap(page);
893 arch_kexec_pre_free_pages(page_address(page), 1);
894 if (result) {
895 result = -EFAULT;
896 goto out;
897 }
898 ubytes -= uchunk;
899 maddr += mchunk;
900 if (image->file_mode)
901 kbuf += mchunk;
902 else
903 buf += mchunk;
904 mbytes -= mchunk;
905
906 cond_resched();
907 }
908out:
909 return result;
910}
911
912int kimage_load_segment(struct kimage *image,
913 struct kexec_segment *segment)
914{
915 int result = -ENOMEM;
916
917 switch (image->type) {
918 case KEXEC_TYPE_DEFAULT:
919 result = kimage_load_normal_segment(image, segment);
920 break;
921 case KEXEC_TYPE_CRASH:
922 result = kimage_load_crash_segment(image, segment);
923 break;
924 }
925
926 return result;
927}
928
929struct kimage *kexec_image;
930struct kimage *kexec_crash_image;
931int kexec_load_disabled;
932
933/*
934 * No panic_cpu check version of crash_kexec(). This function is called
935 * only when panic_cpu holds the current CPU number; this is the only CPU
936 * which processes crash_kexec routines.
937 */
938void __noclone __crash_kexec(struct pt_regs *regs)
939{
940 /* Take the kexec_mutex here to prevent sys_kexec_load
941 * running on one cpu from replacing the crash kernel
942 * we are using after a panic on a different cpu.
943 *
944 * If the crash kernel was not located in a fixed area
945 * of memory the xchg(&kexec_crash_image) would be
946 * sufficient. But since I reuse the memory...
947 */
948 if (mutex_trylock(&kexec_mutex)) {
949 if (kexec_crash_image) {
950 struct pt_regs fixed_regs;
951
952 crash_setup_regs(&fixed_regs, regs);
953 crash_save_vmcoreinfo();
954 machine_crash_shutdown(&fixed_regs);
955 machine_kexec(kexec_crash_image);
956 }
957 mutex_unlock(&kexec_mutex);
958 }
959}
960STACK_FRAME_NON_STANDARD(__crash_kexec);
961
962void crash_kexec(struct pt_regs *regs)
963{
964 int old_cpu, this_cpu;
965
966 /*
967 * Only one CPU is allowed to execute the crash_kexec() code as with
968 * panic(). Otherwise parallel calls of panic() and crash_kexec()
969 * may stop each other. To exclude them, we use panic_cpu here too.
970 */
971 this_cpu = raw_smp_processor_id();
972 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
973 if (old_cpu == PANIC_CPU_INVALID) {
974 /* This is the 1st CPU which comes here, so go ahead. */
975 printk_safe_flush_on_panic();
976 __crash_kexec(regs);
977
978 /*
979 * Reset panic_cpu to allow another panic()/crash_kexec()
980 * call.
981 */
982 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
983 }
984}
985
986size_t crash_get_memory_size(void)
987{
988 size_t size = 0;
989
990 mutex_lock(&kexec_mutex);
991 if (crashk_res.end != crashk_res.start)
992 size = resource_size(&crashk_res);
993 mutex_unlock(&kexec_mutex);
994 return size;
995}
996
997void __weak crash_free_reserved_phys_range(unsigned long begin,
998 unsigned long end)
999{
1000 unsigned long addr;
1001
1002 for (addr = begin; addr < end; addr += PAGE_SIZE)
1003 free_reserved_page(boot_pfn_to_page(addr >> PAGE_SHIFT));
1004}
1005
1006int crash_shrink_memory(unsigned long new_size)
1007{
1008 int ret = 0;
1009 unsigned long start, end;
1010 unsigned long old_size;
1011 struct resource *ram_res;
1012
1013 mutex_lock(&kexec_mutex);
1014
1015 if (kexec_crash_image) {
1016 ret = -ENOENT;
1017 goto unlock;
1018 }
1019 start = crashk_res.start;
1020 end = crashk_res.end;
1021 old_size = (end == 0) ? 0 : end - start + 1;
1022 if (new_size >= old_size) {
1023 ret = (new_size == old_size) ? 0 : -EINVAL;
1024 goto unlock;
1025 }
1026
1027 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1028 if (!ram_res) {
1029 ret = -ENOMEM;
1030 goto unlock;
1031 }
1032
1033 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1034 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1035
1036 crash_free_reserved_phys_range(end, crashk_res.end);
1037
1038 if ((start == end) && (crashk_res.parent != NULL))
1039 release_resource(&crashk_res);
1040
1041 ram_res->start = end;
1042 ram_res->end = crashk_res.end;
1043 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1044 ram_res->name = "System RAM";
1045
1046 crashk_res.end = end - 1;
1047
1048 insert_resource(&iomem_resource, ram_res);
1049
1050unlock:
1051 mutex_unlock(&kexec_mutex);
1052 return ret;
1053}
1054
1055void crash_save_cpu(struct pt_regs *regs, int cpu)
1056{
1057 struct elf_prstatus prstatus;
1058 u32 *buf;
1059
1060 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1061 return;
1062
1063 /* Using ELF notes here is opportunistic.
1064 * I need a well defined structure format
1065 * for the data I pass, and I need tags
1066 * on the data to indicate what information I have
1067 * squirrelled away. ELF notes happen to provide
1068 * all of that, so there is no need to invent something new.
1069 */
1070 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1071 if (!buf)
1072 return;
1073 memset(&prstatus, 0, sizeof(prstatus));
1074 prstatus.pr_pid = current->pid;
1075 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1076 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1077 &prstatus, sizeof(prstatus));
1078 final_note(buf);
1079}
1080
1081static int __init crash_notes_memory_init(void)
1082{
1083 /* Allocate memory for saving cpu registers. */
1084 size_t size, align;
1085
1086 /*
1087 * crash_notes could be allocated across 2 vmalloc pages when percpu
1088 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1089 * pages are also on 2 continuous physical pages. In this case the
1090 * 2nd part of crash_notes in 2nd page could be lost since only the
1091 * starting address and size of crash_notes are exported through sysfs.
1092 * Here round up the size of crash_notes to the nearest power of two
1093 * and pass it to __alloc_percpu as align value. This can make sure
1094 * crash_notes is allocated inside one physical page.
1095 */
1096 size = sizeof(note_buf_t);
1097 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1098
1099 /*
1100 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1101 * definitely will be in 2 pages with that.
1102 */
1103 BUILD_BUG_ON(size > PAGE_SIZE);
1104
1105 crash_notes = __alloc_percpu(size, align);
1106 if (!crash_notes) {
1107 pr_warn("Memory allocation for saving cpu register states failed\n");
1108 return -ENOMEM;
1109 }
1110 return 0;
1111}
1112subsys_initcall(crash_notes_memory_init);
1113
1114
1115/*
1116 * Move into place and start executing a preloaded standalone
1117 * executable. If nothing was preloaded return an error.
1118 */
1119int kernel_kexec(void)
1120{
1121 int error = 0;
1122
1123 if (!mutex_trylock(&kexec_mutex))
1124 return -EBUSY;
1125 if (!kexec_image) {
1126 error = -EINVAL;
1127 goto Unlock;
1128 }
1129
1130#ifdef CONFIG_KEXEC_JUMP
1131 if (kexec_image->preserve_context) {
1132 lock_system_sleep();
1133 pm_prepare_console();
1134 error = freeze_processes();
1135 if (error) {
1136 error = -EBUSY;
1137 goto Restore_console;
1138 }
1139 suspend_console();
1140 error = dpm_suspend_start(PMSG_FREEZE);
1141 if (error)
1142 goto Resume_console;
1143 /* At this point, dpm_suspend_start() has been called,
1144 * but *not* dpm_suspend_end(). We *must* call
1145 * dpm_suspend_end() now. Otherwise, drivers for
1146 * some devices (e.g. interrupt controllers) become
1147 * desynchronized with the actual state of the
1148 * hardware at resume time, and evil weirdness ensues.
1149 */
1150 error = dpm_suspend_end(PMSG_FREEZE);
1151 if (error)
1152 goto Resume_devices;
1153 error = suspend_disable_secondary_cpus();
1154 if (error)
1155 goto Enable_cpus;
1156 local_irq_disable();
1157 error = syscore_suspend();
1158 if (error)
1159 goto Enable_irqs;
1160 } else
1161#endif
1162 {
1163 kexec_in_progress = true;
1164 kernel_restart_prepare(NULL);
1165 migrate_to_reboot_cpu();
1166
1167 /*
1168 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1169 * no further code needs to use CPU hotplug (which is true in
1170 * the reboot case). However, the kexec path depends on using
1171 * CPU hotplug again; so re-enable it here.
1172 */
1173 cpu_hotplug_enable();
1174 pr_emerg("Starting new kernel\n");
1175 machine_shutdown();
1176 }
1177
1178 machine_kexec(kexec_image);
1179
1180#ifdef CONFIG_KEXEC_JUMP
1181 if (kexec_image->preserve_context) {
1182 syscore_resume();
1183 Enable_irqs:
1184 local_irq_enable();
1185 Enable_cpus:
1186 suspend_enable_secondary_cpus();
1187 dpm_resume_start(PMSG_RESTORE);
1188 Resume_devices:
1189 dpm_resume_end(PMSG_RESTORE);
1190 Resume_console:
1191 resume_console();
1192 thaw_processes();
1193 Restore_console:
1194 pm_restore_console();
1195 unlock_system_sleep();
1196 }
1197#endif
1198
1199 Unlock:
1200 mutex_unlock(&kexec_mutex);
1201 return error;
1202}
1203
1204/*
1205 * Protection mechanism for crashkernel reserved memory after
1206 * the kdump kernel is loaded.
1207 *
1208 * Provide an empty default implementation here -- architecture
1209 * code may override this
1210 */
1211void __weak arch_kexec_protect_crashkres(void)
1212{}
1213
1214void __weak arch_kexec_unprotect_crashkres(void)
1215{}
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
50atomic_t __kexec_lock = ATOMIC_INIT(0);
51
52/* Flag to indicate we are going to kexec a new kernel */
53bool kexec_in_progress = false;
54
55bool 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
104static struct page *kimage_alloc_page(struct kimage *image,
105 gfp_t gfp_mask,
106 unsigned long dest);
107
108int 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
216struct 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
249int 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
267static 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
294static 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
308void 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
318static 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
385static 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
453struct 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
472static 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
498static 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
507static 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
515static 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
525void 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
538static 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
546void 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
594static 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
613static 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
709static 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 }
775out:
776 return result;
777}
778
779#ifdef CONFIG_CRASH_DUMP
780static 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 }
846out:
847 return result;
848}
849#endif
850
851int 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
870struct kexec_load_limit {
871 /* Mutex protects the limit count. */
872 struct mutex mutex;
873 int limit;
874};
875
876static struct kexec_load_limit load_limit_reboot = {
877 .mutex = __MUTEX_INITIALIZER(load_limit_reboot.mutex),
878 .limit = -1,
879};
880
881static struct kexec_load_limit load_limit_panic = {
882 .mutex = __MUTEX_INITIALIZER(load_limit_panic.mutex),
883 .limit = -1,
884};
885
886struct kimage *kexec_image;
887struct kimage *kexec_crash_image;
888static int kexec_load_disabled;
889
890#ifdef CONFIG_SYSCTL
891static 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
928static 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
953static int __init kexec_core_sysctl_init(void)
954{
955 register_sysctl_init("kernel", kexec_core_sysctls);
956 return 0;
957}
958late_initcall(kexec_core_sysctl_init);
959#endif
960
961bool 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 */
991int 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}