Loading...
1/*
2 * kexec.c - kexec system call core code.
3 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
4 *
5 * This source code is licensed under the GNU General Public License,
6 * Version 2. See the file COPYING for more details.
7 */
8
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11#include <linux/capability.h>
12#include <linux/mm.h>
13#include <linux/file.h>
14#include <linux/slab.h>
15#include <linux/fs.h>
16#include <linux/kexec.h>
17#include <linux/mutex.h>
18#include <linux/list.h>
19#include <linux/highmem.h>
20#include <linux/syscalls.h>
21#include <linux/reboot.h>
22#include <linux/ioport.h>
23#include <linux/hardirq.h>
24#include <linux/elf.h>
25#include <linux/elfcore.h>
26#include <linux/utsname.h>
27#include <linux/numa.h>
28#include <linux/suspend.h>
29#include <linux/device.h>
30#include <linux/freezer.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
42#include <asm/page.h>
43#include <asm/sections.h>
44
45#include <crypto/hash.h>
46#include <crypto/sha.h>
47#include "kexec_internal.h"
48
49DEFINE_MUTEX(kexec_mutex);
50
51/* Per cpu memory for storing cpu states in case of system crash. */
52note_buf_t __percpu *crash_notes;
53
54/* vmcoreinfo stuff */
55static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
56u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
57size_t vmcoreinfo_size;
58size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
59
60/* Flag to indicate we are going to kexec a new kernel */
61bool kexec_in_progress = false;
62
63
64/* Location of the reserved area for the crash kernel */
65struct resource crashk_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};
72struct resource crashk_low_res = {
73 .name = "Crash kernel",
74 .start = 0,
75 .end = 0,
76 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
77 .desc = IORES_DESC_CRASH_KERNEL
78};
79
80int kexec_should_crash(struct task_struct *p)
81{
82 /*
83 * If crash_kexec_post_notifiers is enabled, don't run
84 * crash_kexec() here yet, which must be run after panic
85 * notifiers in panic().
86 */
87 if (crash_kexec_post_notifiers)
88 return 0;
89 /*
90 * There are 4 panic() calls in do_exit() path, each of which
91 * corresponds to each of these 4 conditions.
92 */
93 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
94 return 1;
95 return 0;
96}
97
98int kexec_crash_loaded(void)
99{
100 return !!kexec_crash_image;
101}
102EXPORT_SYMBOL_GPL(kexec_crash_loaded);
103
104/*
105 * When kexec transitions to the new kernel there is a one-to-one
106 * mapping between physical and virtual addresses. On processors
107 * where you can disable the MMU this is trivial, and easy. For
108 * others it is still a simple predictable page table to setup.
109 *
110 * In that environment kexec copies the new kernel to its final
111 * resting place. This means I can only support memory whose
112 * physical address can fit in an unsigned long. In particular
113 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
114 * If the assembly stub has more restrictive requirements
115 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
116 * defined more restrictively in <asm/kexec.h>.
117 *
118 * The code for the transition from the current kernel to the
119 * the new kernel is placed in the control_code_buffer, whose size
120 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
121 * page of memory is necessary, but some architectures require more.
122 * Because this memory must be identity mapped in the transition from
123 * virtual to physical addresses it must live in the range
124 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
125 * modifiable.
126 *
127 * The assembly stub in the control code buffer is passed a linked list
128 * of descriptor pages detailing the source pages of the new kernel,
129 * and the destination addresses of those source pages. As this data
130 * structure is not used in the context of the current OS, it must
131 * be self-contained.
132 *
133 * The code has been made to work with highmem pages and will use a
134 * destination page in its final resting place (if it happens
135 * to allocate it). The end product of this is that most of the
136 * physical address space, and most of RAM can be used.
137 *
138 * Future directions include:
139 * - allocating a page table with the control code buffer identity
140 * mapped, to simplify machine_kexec and make kexec_on_panic more
141 * reliable.
142 */
143
144/*
145 * KIMAGE_NO_DEST is an impossible destination address..., for
146 * allocating pages whose destination address we do not care about.
147 */
148#define KIMAGE_NO_DEST (-1UL)
149#define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
150
151static struct page *kimage_alloc_page(struct kimage *image,
152 gfp_t gfp_mask,
153 unsigned long dest);
154
155int sanity_check_segment_list(struct kimage *image)
156{
157 int i;
158 unsigned long nr_segments = image->nr_segments;
159 unsigned long total_pages = 0;
160
161 /*
162 * Verify we have good destination addresses. The caller is
163 * responsible for making certain we don't attempt to load
164 * the new image into invalid or reserved areas of RAM. This
165 * just verifies it is an address we can use.
166 *
167 * Since the kernel does everything in page size chunks ensure
168 * the destination addresses are page aligned. Too many
169 * special cases crop of when we don't do this. The most
170 * insidious is getting overlapping destination addresses
171 * simply because addresses are changed to page size
172 * granularity.
173 */
174 for (i = 0; i < nr_segments; i++) {
175 unsigned long mstart, mend;
176
177 mstart = image->segment[i].mem;
178 mend = mstart + image->segment[i].memsz;
179 if (mstart > mend)
180 return -EADDRNOTAVAIL;
181 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
182 return -EADDRNOTAVAIL;
183 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
184 return -EADDRNOTAVAIL;
185 }
186
187 /* Verify our destination addresses do not overlap.
188 * If we alloed overlapping destination addresses
189 * through very weird things can happen with no
190 * easy explanation as one segment stops on another.
191 */
192 for (i = 0; i < nr_segments; i++) {
193 unsigned long mstart, mend;
194 unsigned long j;
195
196 mstart = image->segment[i].mem;
197 mend = mstart + image->segment[i].memsz;
198 for (j = 0; j < i; j++) {
199 unsigned long pstart, pend;
200
201 pstart = image->segment[j].mem;
202 pend = pstart + image->segment[j].memsz;
203 /* Do the segments overlap ? */
204 if ((mend > pstart) && (mstart < pend))
205 return -EINVAL;
206 }
207 }
208
209 /* Ensure our buffer sizes are strictly less than
210 * our memory sizes. This should always be the case,
211 * and it is easier to check up front than to be surprised
212 * later on.
213 */
214 for (i = 0; i < nr_segments; i++) {
215 if (image->segment[i].bufsz > image->segment[i].memsz)
216 return -EINVAL;
217 }
218
219 /*
220 * Verify that no more than half of memory will be consumed. If the
221 * request from userspace is too large, a large amount of time will be
222 * wasted allocating pages, which can cause a soft lockup.
223 */
224 for (i = 0; i < nr_segments; i++) {
225 if (PAGE_COUNT(image->segment[i].memsz) > totalram_pages / 2)
226 return -EINVAL;
227
228 total_pages += PAGE_COUNT(image->segment[i].memsz);
229 }
230
231 if (total_pages > totalram_pages / 2)
232 return -EINVAL;
233
234 /*
235 * Verify we have good destination addresses. Normally
236 * the caller is responsible for making certain we don't
237 * attempt to load the new image into invalid or reserved
238 * areas of RAM. But crash kernels are preloaded into a
239 * reserved area of ram. We must ensure the addresses
240 * are in the reserved area otherwise preloading the
241 * kernel could corrupt things.
242 */
243
244 if (image->type == KEXEC_TYPE_CRASH) {
245 for (i = 0; i < nr_segments; i++) {
246 unsigned long mstart, mend;
247
248 mstart = image->segment[i].mem;
249 mend = mstart + image->segment[i].memsz - 1;
250 /* Ensure we are within the crash kernel limits */
251 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
252 (mend > phys_to_boot_phys(crashk_res.end)))
253 return -EADDRNOTAVAIL;
254 }
255 }
256
257 return 0;
258}
259
260struct kimage *do_kimage_alloc_init(void)
261{
262 struct kimage *image;
263
264 /* Allocate a controlling structure */
265 image = kzalloc(sizeof(*image), GFP_KERNEL);
266 if (!image)
267 return NULL;
268
269 image->head = 0;
270 image->entry = &image->head;
271 image->last_entry = &image->head;
272 image->control_page = ~0; /* By default this does not apply */
273 image->type = KEXEC_TYPE_DEFAULT;
274
275 /* Initialize the list of control pages */
276 INIT_LIST_HEAD(&image->control_pages);
277
278 /* Initialize the list of destination pages */
279 INIT_LIST_HEAD(&image->dest_pages);
280
281 /* Initialize the list of unusable pages */
282 INIT_LIST_HEAD(&image->unusable_pages);
283
284 return image;
285}
286
287int kimage_is_destination_range(struct kimage *image,
288 unsigned long start,
289 unsigned long end)
290{
291 unsigned long i;
292
293 for (i = 0; i < image->nr_segments; i++) {
294 unsigned long mstart, mend;
295
296 mstart = image->segment[i].mem;
297 mend = mstart + image->segment[i].memsz;
298 if ((end > mstart) && (start < mend))
299 return 1;
300 }
301
302 return 0;
303}
304
305static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
306{
307 struct page *pages;
308
309 pages = alloc_pages(gfp_mask, order);
310 if (pages) {
311 unsigned int count, i;
312
313 pages->mapping = NULL;
314 set_page_private(pages, order);
315 count = 1 << order;
316 for (i = 0; i < count; i++)
317 SetPageReserved(pages + i);
318 }
319
320 return pages;
321}
322
323static void kimage_free_pages(struct page *page)
324{
325 unsigned int order, count, i;
326
327 order = page_private(page);
328 count = 1 << order;
329 for (i = 0; i < count; i++)
330 ClearPageReserved(page + i);
331 __free_pages(page, order);
332}
333
334void kimage_free_page_list(struct list_head *list)
335{
336 struct page *page, *next;
337
338 list_for_each_entry_safe(page, next, list, lru) {
339 list_del(&page->lru);
340 kimage_free_pages(page);
341 }
342}
343
344static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
345 unsigned int order)
346{
347 /* Control pages are special, they are the intermediaries
348 * that are needed while we copy the rest of the pages
349 * to their final resting place. As such they must
350 * not conflict with either the destination addresses
351 * or memory the kernel is already using.
352 *
353 * The only case where we really need more than one of
354 * these are for architectures where we cannot disable
355 * the MMU and must instead generate an identity mapped
356 * page table for all of the memory.
357 *
358 * At worst this runs in O(N) of the image size.
359 */
360 struct list_head extra_pages;
361 struct page *pages;
362 unsigned int count;
363
364 count = 1 << order;
365 INIT_LIST_HEAD(&extra_pages);
366
367 /* Loop while I can allocate a page and the page allocated
368 * is a destination page.
369 */
370 do {
371 unsigned long pfn, epfn, addr, eaddr;
372
373 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
374 if (!pages)
375 break;
376 pfn = page_to_boot_pfn(pages);
377 epfn = pfn + count;
378 addr = pfn << PAGE_SHIFT;
379 eaddr = epfn << PAGE_SHIFT;
380 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
381 kimage_is_destination_range(image, addr, eaddr)) {
382 list_add(&pages->lru, &extra_pages);
383 pages = NULL;
384 }
385 } while (!pages);
386
387 if (pages) {
388 /* Remember the allocated page... */
389 list_add(&pages->lru, &image->control_pages);
390
391 /* Because the page is already in it's destination
392 * location we will never allocate another page at
393 * that address. Therefore kimage_alloc_pages
394 * will not return it (again) and we don't need
395 * to give it an entry in image->segment[].
396 */
397 }
398 /* Deal with the destination pages I have inadvertently allocated.
399 *
400 * Ideally I would convert multi-page allocations into single
401 * page allocations, and add everything to image->dest_pages.
402 *
403 * For now it is simpler to just free the pages.
404 */
405 kimage_free_page_list(&extra_pages);
406
407 return pages;
408}
409
410static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
411 unsigned int order)
412{
413 /* Control pages are special, they are the intermediaries
414 * that are needed while we copy the rest of the pages
415 * to their final resting place. As such they must
416 * not conflict with either the destination addresses
417 * or memory the kernel is already using.
418 *
419 * Control pages are also the only pags we must allocate
420 * when loading a crash kernel. All of the other pages
421 * are specified by the segments and we just memcpy
422 * into them directly.
423 *
424 * The only case where we really need more than one of
425 * these are for architectures where we cannot disable
426 * the MMU and must instead generate an identity mapped
427 * page table for all of the memory.
428 *
429 * Given the low demand this implements a very simple
430 * allocator that finds the first hole of the appropriate
431 * size in the reserved memory region, and allocates all
432 * of the memory up to and including the hole.
433 */
434 unsigned long hole_start, hole_end, size;
435 struct page *pages;
436
437 pages = NULL;
438 size = (1 << order) << PAGE_SHIFT;
439 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
440 hole_end = hole_start + size - 1;
441 while (hole_end <= crashk_res.end) {
442 unsigned long i;
443
444 cond_resched();
445
446 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
447 break;
448 /* See if I overlap any of the segments */
449 for (i = 0; i < image->nr_segments; i++) {
450 unsigned long mstart, mend;
451
452 mstart = image->segment[i].mem;
453 mend = mstart + image->segment[i].memsz - 1;
454 if ((hole_end >= mstart) && (hole_start <= mend)) {
455 /* Advance the hole to the end of the segment */
456 hole_start = (mend + (size - 1)) & ~(size - 1);
457 hole_end = hole_start + size - 1;
458 break;
459 }
460 }
461 /* If I don't overlap any segments I have found my hole! */
462 if (i == image->nr_segments) {
463 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
464 image->control_page = hole_end;
465 break;
466 }
467 }
468
469 return pages;
470}
471
472
473struct page *kimage_alloc_control_pages(struct kimage *image,
474 unsigned int order)
475{
476 struct page *pages = NULL;
477
478 switch (image->type) {
479 case KEXEC_TYPE_DEFAULT:
480 pages = kimage_alloc_normal_control_pages(image, order);
481 break;
482 case KEXEC_TYPE_CRASH:
483 pages = kimage_alloc_crash_control_pages(image, order);
484 break;
485 }
486
487 return pages;
488}
489
490static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
491{
492 if (*image->entry != 0)
493 image->entry++;
494
495 if (image->entry == image->last_entry) {
496 kimage_entry_t *ind_page;
497 struct page *page;
498
499 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
500 if (!page)
501 return -ENOMEM;
502
503 ind_page = page_address(page);
504 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
505 image->entry = ind_page;
506 image->last_entry = ind_page +
507 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
508 }
509 *image->entry = entry;
510 image->entry++;
511 *image->entry = 0;
512
513 return 0;
514}
515
516static int kimage_set_destination(struct kimage *image,
517 unsigned long destination)
518{
519 int result;
520
521 destination &= PAGE_MASK;
522 result = kimage_add_entry(image, destination | IND_DESTINATION);
523
524 return result;
525}
526
527
528static int kimage_add_page(struct kimage *image, unsigned long page)
529{
530 int result;
531
532 page &= PAGE_MASK;
533 result = kimage_add_entry(image, page | IND_SOURCE);
534
535 return result;
536}
537
538
539static void kimage_free_extra_pages(struct kimage *image)
540{
541 /* Walk through and free any extra destination pages I may have */
542 kimage_free_page_list(&image->dest_pages);
543
544 /* Walk through and free any unusable pages I have cached */
545 kimage_free_page_list(&image->unusable_pages);
546
547}
548void kimage_terminate(struct kimage *image)
549{
550 if (*image->entry != 0)
551 image->entry++;
552
553 *image->entry = IND_DONE;
554}
555
556#define for_each_kimage_entry(image, ptr, entry) \
557 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
558 ptr = (entry & IND_INDIRECTION) ? \
559 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
560
561static void kimage_free_entry(kimage_entry_t entry)
562{
563 struct page *page;
564
565 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
566 kimage_free_pages(page);
567}
568
569void kimage_free(struct kimage *image)
570{
571 kimage_entry_t *ptr, entry;
572 kimage_entry_t ind = 0;
573
574 if (!image)
575 return;
576
577 kimage_free_extra_pages(image);
578 for_each_kimage_entry(image, ptr, entry) {
579 if (entry & IND_INDIRECTION) {
580 /* Free the previous indirection page */
581 if (ind & IND_INDIRECTION)
582 kimage_free_entry(ind);
583 /* Save this indirection page until we are
584 * done with it.
585 */
586 ind = entry;
587 } else if (entry & IND_SOURCE)
588 kimage_free_entry(entry);
589 }
590 /* Free the final indirection page */
591 if (ind & IND_INDIRECTION)
592 kimage_free_entry(ind);
593
594 /* Handle any machine specific cleanup */
595 machine_kexec_cleanup(image);
596
597 /* Free the kexec control pages... */
598 kimage_free_page_list(&image->control_pages);
599
600 /*
601 * Free up any temporary buffers allocated. This might hit if
602 * error occurred much later after buffer allocation.
603 */
604 if (image->file_mode)
605 kimage_file_post_load_cleanup(image);
606
607 kfree(image);
608}
609
610static kimage_entry_t *kimage_dst_used(struct kimage *image,
611 unsigned long page)
612{
613 kimage_entry_t *ptr, entry;
614 unsigned long destination = 0;
615
616 for_each_kimage_entry(image, ptr, entry) {
617 if (entry & IND_DESTINATION)
618 destination = entry & PAGE_MASK;
619 else if (entry & IND_SOURCE) {
620 if (page == destination)
621 return ptr;
622 destination += PAGE_SIZE;
623 }
624 }
625
626 return NULL;
627}
628
629static struct page *kimage_alloc_page(struct kimage *image,
630 gfp_t gfp_mask,
631 unsigned long destination)
632{
633 /*
634 * Here we implement safeguards to ensure that a source page
635 * is not copied to its destination page before the data on
636 * the destination page is no longer useful.
637 *
638 * To do this we maintain the invariant that a source page is
639 * either its own destination page, or it is not a
640 * destination page at all.
641 *
642 * That is slightly stronger than required, but the proof
643 * that no problems will not occur is trivial, and the
644 * implementation is simply to verify.
645 *
646 * When allocating all pages normally this algorithm will run
647 * in O(N) time, but in the worst case it will run in O(N^2)
648 * time. If the runtime is a problem the data structures can
649 * be fixed.
650 */
651 struct page *page;
652 unsigned long addr;
653
654 /*
655 * Walk through the list of destination pages, and see if I
656 * have a match.
657 */
658 list_for_each_entry(page, &image->dest_pages, lru) {
659 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
660 if (addr == destination) {
661 list_del(&page->lru);
662 return page;
663 }
664 }
665 page = NULL;
666 while (1) {
667 kimage_entry_t *old;
668
669 /* Allocate a page, if we run out of memory give up */
670 page = kimage_alloc_pages(gfp_mask, 0);
671 if (!page)
672 return NULL;
673 /* If the page cannot be used file it away */
674 if (page_to_boot_pfn(page) >
675 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
676 list_add(&page->lru, &image->unusable_pages);
677 continue;
678 }
679 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
680
681 /* If it is the destination page we want use it */
682 if (addr == destination)
683 break;
684
685 /* If the page is not a destination page use it */
686 if (!kimage_is_destination_range(image, addr,
687 addr + PAGE_SIZE))
688 break;
689
690 /*
691 * I know that the page is someones destination page.
692 * See if there is already a source page for this
693 * destination page. And if so swap the source pages.
694 */
695 old = kimage_dst_used(image, addr);
696 if (old) {
697 /* If so move it */
698 unsigned long old_addr;
699 struct page *old_page;
700
701 old_addr = *old & PAGE_MASK;
702 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
703 copy_highpage(page, old_page);
704 *old = addr | (*old & ~PAGE_MASK);
705
706 /* The old page I have found cannot be a
707 * destination page, so return it if it's
708 * gfp_flags honor the ones passed in.
709 */
710 if (!(gfp_mask & __GFP_HIGHMEM) &&
711 PageHighMem(old_page)) {
712 kimage_free_pages(old_page);
713 continue;
714 }
715 addr = old_addr;
716 page = old_page;
717 break;
718 }
719 /* Place the page on the destination list, to be used later */
720 list_add(&page->lru, &image->dest_pages);
721 }
722
723 return page;
724}
725
726static int kimage_load_normal_segment(struct kimage *image,
727 struct kexec_segment *segment)
728{
729 unsigned long maddr;
730 size_t ubytes, mbytes;
731 int result;
732 unsigned char __user *buf = NULL;
733 unsigned char *kbuf = NULL;
734
735 result = 0;
736 if (image->file_mode)
737 kbuf = segment->kbuf;
738 else
739 buf = segment->buf;
740 ubytes = segment->bufsz;
741 mbytes = segment->memsz;
742 maddr = segment->mem;
743
744 result = kimage_set_destination(image, maddr);
745 if (result < 0)
746 goto out;
747
748 while (mbytes) {
749 struct page *page;
750 char *ptr;
751 size_t uchunk, mchunk;
752
753 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
754 if (!page) {
755 result = -ENOMEM;
756 goto out;
757 }
758 result = kimage_add_page(image, page_to_boot_pfn(page)
759 << PAGE_SHIFT);
760 if (result < 0)
761 goto out;
762
763 ptr = kmap(page);
764 /* Start with a clear page */
765 clear_page(ptr);
766 ptr += maddr & ~PAGE_MASK;
767 mchunk = min_t(size_t, mbytes,
768 PAGE_SIZE - (maddr & ~PAGE_MASK));
769 uchunk = min(ubytes, mchunk);
770
771 /* For file based kexec, source pages are in kernel memory */
772 if (image->file_mode)
773 memcpy(ptr, kbuf, uchunk);
774 else
775 result = copy_from_user(ptr, buf, uchunk);
776 kunmap(page);
777 if (result) {
778 result = -EFAULT;
779 goto out;
780 }
781 ubytes -= uchunk;
782 maddr += mchunk;
783 if (image->file_mode)
784 kbuf += mchunk;
785 else
786 buf += mchunk;
787 mbytes -= mchunk;
788 }
789out:
790 return result;
791}
792
793static int kimage_load_crash_segment(struct kimage *image,
794 struct kexec_segment *segment)
795{
796 /* For crash dumps kernels we simply copy the data from
797 * user space to it's destination.
798 * We do things a page at a time for the sake of kmap.
799 */
800 unsigned long maddr;
801 size_t ubytes, mbytes;
802 int result;
803 unsigned char __user *buf = NULL;
804 unsigned char *kbuf = NULL;
805
806 result = 0;
807 if (image->file_mode)
808 kbuf = segment->kbuf;
809 else
810 buf = segment->buf;
811 ubytes = segment->bufsz;
812 mbytes = segment->memsz;
813 maddr = segment->mem;
814 while (mbytes) {
815 struct page *page;
816 char *ptr;
817 size_t uchunk, mchunk;
818
819 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
820 if (!page) {
821 result = -ENOMEM;
822 goto out;
823 }
824 ptr = kmap(page);
825 ptr += maddr & ~PAGE_MASK;
826 mchunk = min_t(size_t, mbytes,
827 PAGE_SIZE - (maddr & ~PAGE_MASK));
828 uchunk = min(ubytes, mchunk);
829 if (mchunk > uchunk) {
830 /* Zero the trailing part of the page */
831 memset(ptr + uchunk, 0, mchunk - uchunk);
832 }
833
834 /* For file based kexec, source pages are in kernel memory */
835 if (image->file_mode)
836 memcpy(ptr, kbuf, uchunk);
837 else
838 result = copy_from_user(ptr, buf, uchunk);
839 kexec_flush_icache_page(page);
840 kunmap(page);
841 if (result) {
842 result = -EFAULT;
843 goto out;
844 }
845 ubytes -= uchunk;
846 maddr += mchunk;
847 if (image->file_mode)
848 kbuf += mchunk;
849 else
850 buf += mchunk;
851 mbytes -= mchunk;
852 }
853out:
854 return result;
855}
856
857int kimage_load_segment(struct kimage *image,
858 struct kexec_segment *segment)
859{
860 int result = -ENOMEM;
861
862 switch (image->type) {
863 case KEXEC_TYPE_DEFAULT:
864 result = kimage_load_normal_segment(image, segment);
865 break;
866 case KEXEC_TYPE_CRASH:
867 result = kimage_load_crash_segment(image, segment);
868 break;
869 }
870
871 return result;
872}
873
874struct kimage *kexec_image;
875struct kimage *kexec_crash_image;
876int kexec_load_disabled;
877
878/*
879 * No panic_cpu check version of crash_kexec(). This function is called
880 * only when panic_cpu holds the current CPU number; this is the only CPU
881 * which processes crash_kexec routines.
882 */
883void __crash_kexec(struct pt_regs *regs)
884{
885 /* Take the kexec_mutex here to prevent sys_kexec_load
886 * running on one cpu from replacing the crash kernel
887 * we are using after a panic on a different cpu.
888 *
889 * If the crash kernel was not located in a fixed area
890 * of memory the xchg(&kexec_crash_image) would be
891 * sufficient. But since I reuse the memory...
892 */
893 if (mutex_trylock(&kexec_mutex)) {
894 if (kexec_crash_image) {
895 struct pt_regs fixed_regs;
896
897 crash_setup_regs(&fixed_regs, regs);
898 crash_save_vmcoreinfo();
899 machine_crash_shutdown(&fixed_regs);
900 machine_kexec(kexec_crash_image);
901 }
902 mutex_unlock(&kexec_mutex);
903 }
904}
905
906void crash_kexec(struct pt_regs *regs)
907{
908 int old_cpu, this_cpu;
909
910 /*
911 * Only one CPU is allowed to execute the crash_kexec() code as with
912 * panic(). Otherwise parallel calls of panic() and crash_kexec()
913 * may stop each other. To exclude them, we use panic_cpu here too.
914 */
915 this_cpu = raw_smp_processor_id();
916 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
917 if (old_cpu == PANIC_CPU_INVALID) {
918 /* This is the 1st CPU which comes here, so go ahead. */
919 printk_nmi_flush_on_panic();
920 __crash_kexec(regs);
921
922 /*
923 * Reset panic_cpu to allow another panic()/crash_kexec()
924 * call.
925 */
926 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
927 }
928}
929
930size_t crash_get_memory_size(void)
931{
932 size_t size = 0;
933
934 mutex_lock(&kexec_mutex);
935 if (crashk_res.end != crashk_res.start)
936 size = resource_size(&crashk_res);
937 mutex_unlock(&kexec_mutex);
938 return size;
939}
940
941void __weak crash_free_reserved_phys_range(unsigned long begin,
942 unsigned long end)
943{
944 unsigned long addr;
945
946 for (addr = begin; addr < end; addr += PAGE_SIZE)
947 free_reserved_page(boot_pfn_to_page(addr >> PAGE_SHIFT));
948}
949
950int crash_shrink_memory(unsigned long new_size)
951{
952 int ret = 0;
953 unsigned long start, end;
954 unsigned long old_size;
955 struct resource *ram_res;
956
957 mutex_lock(&kexec_mutex);
958
959 if (kexec_crash_image) {
960 ret = -ENOENT;
961 goto unlock;
962 }
963 start = crashk_res.start;
964 end = crashk_res.end;
965 old_size = (end == 0) ? 0 : end - start + 1;
966 if (new_size >= old_size) {
967 ret = (new_size == old_size) ? 0 : -EINVAL;
968 goto unlock;
969 }
970
971 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
972 if (!ram_res) {
973 ret = -ENOMEM;
974 goto unlock;
975 }
976
977 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
978 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
979
980 crash_free_reserved_phys_range(end, crashk_res.end);
981
982 if ((start == end) && (crashk_res.parent != NULL))
983 release_resource(&crashk_res);
984
985 ram_res->start = end;
986 ram_res->end = crashk_res.end;
987 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
988 ram_res->name = "System RAM";
989
990 crashk_res.end = end - 1;
991
992 insert_resource(&iomem_resource, ram_res);
993
994unlock:
995 mutex_unlock(&kexec_mutex);
996 return ret;
997}
998
999static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1000 size_t data_len)
1001{
1002 struct elf_note note;
1003
1004 note.n_namesz = strlen(name) + 1;
1005 note.n_descsz = data_len;
1006 note.n_type = type;
1007 memcpy(buf, ¬e, sizeof(note));
1008 buf += (sizeof(note) + 3)/4;
1009 memcpy(buf, name, note.n_namesz);
1010 buf += (note.n_namesz + 3)/4;
1011 memcpy(buf, data, note.n_descsz);
1012 buf += (note.n_descsz + 3)/4;
1013
1014 return buf;
1015}
1016
1017static void final_note(u32 *buf)
1018{
1019 struct elf_note note;
1020
1021 note.n_namesz = 0;
1022 note.n_descsz = 0;
1023 note.n_type = 0;
1024 memcpy(buf, ¬e, sizeof(note));
1025}
1026
1027void crash_save_cpu(struct pt_regs *regs, int cpu)
1028{
1029 struct elf_prstatus prstatus;
1030 u32 *buf;
1031
1032 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1033 return;
1034
1035 /* Using ELF notes here is opportunistic.
1036 * I need a well defined structure format
1037 * for the data I pass, and I need tags
1038 * on the data to indicate what information I have
1039 * squirrelled away. ELF notes happen to provide
1040 * all of that, so there is no need to invent something new.
1041 */
1042 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1043 if (!buf)
1044 return;
1045 memset(&prstatus, 0, sizeof(prstatus));
1046 prstatus.pr_pid = current->pid;
1047 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1048 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1049 &prstatus, sizeof(prstatus));
1050 final_note(buf);
1051}
1052
1053static int __init crash_notes_memory_init(void)
1054{
1055 /* Allocate memory for saving cpu registers. */
1056 size_t size, align;
1057
1058 /*
1059 * crash_notes could be allocated across 2 vmalloc pages when percpu
1060 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1061 * pages are also on 2 continuous physical pages. In this case the
1062 * 2nd part of crash_notes in 2nd page could be lost since only the
1063 * starting address and size of crash_notes are exported through sysfs.
1064 * Here round up the size of crash_notes to the nearest power of two
1065 * and pass it to __alloc_percpu as align value. This can make sure
1066 * crash_notes is allocated inside one physical page.
1067 */
1068 size = sizeof(note_buf_t);
1069 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1070
1071 /*
1072 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1073 * definitely will be in 2 pages with that.
1074 */
1075 BUILD_BUG_ON(size > PAGE_SIZE);
1076
1077 crash_notes = __alloc_percpu(size, align);
1078 if (!crash_notes) {
1079 pr_warn("Memory allocation for saving cpu register states failed\n");
1080 return -ENOMEM;
1081 }
1082 return 0;
1083}
1084subsys_initcall(crash_notes_memory_init);
1085
1086
1087/*
1088 * parsing the "crashkernel" commandline
1089 *
1090 * this code is intended to be called from architecture specific code
1091 */
1092
1093
1094/*
1095 * This function parses command lines in the format
1096 *
1097 * crashkernel=ramsize-range:size[,...][@offset]
1098 *
1099 * The function returns 0 on success and -EINVAL on failure.
1100 */
1101static int __init parse_crashkernel_mem(char *cmdline,
1102 unsigned long long system_ram,
1103 unsigned long long *crash_size,
1104 unsigned long long *crash_base)
1105{
1106 char *cur = cmdline, *tmp;
1107
1108 /* for each entry of the comma-separated list */
1109 do {
1110 unsigned long long start, end = ULLONG_MAX, size;
1111
1112 /* get the start of the range */
1113 start = memparse(cur, &tmp);
1114 if (cur == tmp) {
1115 pr_warn("crashkernel: Memory value expected\n");
1116 return -EINVAL;
1117 }
1118 cur = tmp;
1119 if (*cur != '-') {
1120 pr_warn("crashkernel: '-' expected\n");
1121 return -EINVAL;
1122 }
1123 cur++;
1124
1125 /* if no ':' is here, than we read the end */
1126 if (*cur != ':') {
1127 end = memparse(cur, &tmp);
1128 if (cur == tmp) {
1129 pr_warn("crashkernel: Memory value expected\n");
1130 return -EINVAL;
1131 }
1132 cur = tmp;
1133 if (end <= start) {
1134 pr_warn("crashkernel: end <= start\n");
1135 return -EINVAL;
1136 }
1137 }
1138
1139 if (*cur != ':') {
1140 pr_warn("crashkernel: ':' expected\n");
1141 return -EINVAL;
1142 }
1143 cur++;
1144
1145 size = memparse(cur, &tmp);
1146 if (cur == tmp) {
1147 pr_warn("Memory value expected\n");
1148 return -EINVAL;
1149 }
1150 cur = tmp;
1151 if (size >= system_ram) {
1152 pr_warn("crashkernel: invalid size\n");
1153 return -EINVAL;
1154 }
1155
1156 /* match ? */
1157 if (system_ram >= start && system_ram < end) {
1158 *crash_size = size;
1159 break;
1160 }
1161 } while (*cur++ == ',');
1162
1163 if (*crash_size > 0) {
1164 while (*cur && *cur != ' ' && *cur != '@')
1165 cur++;
1166 if (*cur == '@') {
1167 cur++;
1168 *crash_base = memparse(cur, &tmp);
1169 if (cur == tmp) {
1170 pr_warn("Memory value expected after '@'\n");
1171 return -EINVAL;
1172 }
1173 }
1174 }
1175
1176 return 0;
1177}
1178
1179/*
1180 * That function parses "simple" (old) crashkernel command lines like
1181 *
1182 * crashkernel=size[@offset]
1183 *
1184 * It returns 0 on success and -EINVAL on failure.
1185 */
1186static int __init parse_crashkernel_simple(char *cmdline,
1187 unsigned long long *crash_size,
1188 unsigned long long *crash_base)
1189{
1190 char *cur = cmdline;
1191
1192 *crash_size = memparse(cmdline, &cur);
1193 if (cmdline == cur) {
1194 pr_warn("crashkernel: memory value expected\n");
1195 return -EINVAL;
1196 }
1197
1198 if (*cur == '@')
1199 *crash_base = memparse(cur+1, &cur);
1200 else if (*cur != ' ' && *cur != '\0') {
1201 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1202 return -EINVAL;
1203 }
1204
1205 return 0;
1206}
1207
1208#define SUFFIX_HIGH 0
1209#define SUFFIX_LOW 1
1210#define SUFFIX_NULL 2
1211static __initdata char *suffix_tbl[] = {
1212 [SUFFIX_HIGH] = ",high",
1213 [SUFFIX_LOW] = ",low",
1214 [SUFFIX_NULL] = NULL,
1215};
1216
1217/*
1218 * That function parses "suffix" crashkernel command lines like
1219 *
1220 * crashkernel=size,[high|low]
1221 *
1222 * It returns 0 on success and -EINVAL on failure.
1223 */
1224static int __init parse_crashkernel_suffix(char *cmdline,
1225 unsigned long long *crash_size,
1226 const char *suffix)
1227{
1228 char *cur = cmdline;
1229
1230 *crash_size = memparse(cmdline, &cur);
1231 if (cmdline == cur) {
1232 pr_warn("crashkernel: memory value expected\n");
1233 return -EINVAL;
1234 }
1235
1236 /* check with suffix */
1237 if (strncmp(cur, suffix, strlen(suffix))) {
1238 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1239 return -EINVAL;
1240 }
1241 cur += strlen(suffix);
1242 if (*cur != ' ' && *cur != '\0') {
1243 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1244 return -EINVAL;
1245 }
1246
1247 return 0;
1248}
1249
1250static __init char *get_last_crashkernel(char *cmdline,
1251 const char *name,
1252 const char *suffix)
1253{
1254 char *p = cmdline, *ck_cmdline = NULL;
1255
1256 /* find crashkernel and use the last one if there are more */
1257 p = strstr(p, name);
1258 while (p) {
1259 char *end_p = strchr(p, ' ');
1260 char *q;
1261
1262 if (!end_p)
1263 end_p = p + strlen(p);
1264
1265 if (!suffix) {
1266 int i;
1267
1268 /* skip the one with any known suffix */
1269 for (i = 0; suffix_tbl[i]; i++) {
1270 q = end_p - strlen(suffix_tbl[i]);
1271 if (!strncmp(q, suffix_tbl[i],
1272 strlen(suffix_tbl[i])))
1273 goto next;
1274 }
1275 ck_cmdline = p;
1276 } else {
1277 q = end_p - strlen(suffix);
1278 if (!strncmp(q, suffix, strlen(suffix)))
1279 ck_cmdline = p;
1280 }
1281next:
1282 p = strstr(p+1, name);
1283 }
1284
1285 if (!ck_cmdline)
1286 return NULL;
1287
1288 return ck_cmdline;
1289}
1290
1291static int __init __parse_crashkernel(char *cmdline,
1292 unsigned long long system_ram,
1293 unsigned long long *crash_size,
1294 unsigned long long *crash_base,
1295 const char *name,
1296 const char *suffix)
1297{
1298 char *first_colon, *first_space;
1299 char *ck_cmdline;
1300
1301 BUG_ON(!crash_size || !crash_base);
1302 *crash_size = 0;
1303 *crash_base = 0;
1304
1305 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1306
1307 if (!ck_cmdline)
1308 return -EINVAL;
1309
1310 ck_cmdline += strlen(name);
1311
1312 if (suffix)
1313 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1314 suffix);
1315 /*
1316 * if the commandline contains a ':', then that's the extended
1317 * syntax -- if not, it must be the classic syntax
1318 */
1319 first_colon = strchr(ck_cmdline, ':');
1320 first_space = strchr(ck_cmdline, ' ');
1321 if (first_colon && (!first_space || first_colon < first_space))
1322 return parse_crashkernel_mem(ck_cmdline, system_ram,
1323 crash_size, crash_base);
1324
1325 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1326}
1327
1328/*
1329 * That function is the entry point for command line parsing and should be
1330 * called from the arch-specific code.
1331 */
1332int __init parse_crashkernel(char *cmdline,
1333 unsigned long long system_ram,
1334 unsigned long long *crash_size,
1335 unsigned long long *crash_base)
1336{
1337 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1338 "crashkernel=", NULL);
1339}
1340
1341int __init parse_crashkernel_high(char *cmdline,
1342 unsigned long long system_ram,
1343 unsigned long long *crash_size,
1344 unsigned long long *crash_base)
1345{
1346 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1347 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1348}
1349
1350int __init parse_crashkernel_low(char *cmdline,
1351 unsigned long long system_ram,
1352 unsigned long long *crash_size,
1353 unsigned long long *crash_base)
1354{
1355 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1356 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1357}
1358
1359static void update_vmcoreinfo_note(void)
1360{
1361 u32 *buf = vmcoreinfo_note;
1362
1363 if (!vmcoreinfo_size)
1364 return;
1365 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1366 vmcoreinfo_size);
1367 final_note(buf);
1368}
1369
1370void crash_save_vmcoreinfo(void)
1371{
1372 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1373 update_vmcoreinfo_note();
1374}
1375
1376void vmcoreinfo_append_str(const char *fmt, ...)
1377{
1378 va_list args;
1379 char buf[0x50];
1380 size_t r;
1381
1382 va_start(args, fmt);
1383 r = vscnprintf(buf, sizeof(buf), fmt, args);
1384 va_end(args);
1385
1386 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1387
1388 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1389
1390 vmcoreinfo_size += r;
1391}
1392
1393/*
1394 * provide an empty default implementation here -- architecture
1395 * code may override this
1396 */
1397void __weak arch_crash_save_vmcoreinfo(void)
1398{}
1399
1400phys_addr_t __weak paddr_vmcoreinfo_note(void)
1401{
1402 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1403}
1404
1405static int __init crash_save_vmcoreinfo_init(void)
1406{
1407 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1408 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1409
1410 VMCOREINFO_SYMBOL(init_uts_ns);
1411 VMCOREINFO_SYMBOL(node_online_map);
1412#ifdef CONFIG_MMU
1413 VMCOREINFO_SYMBOL(swapper_pg_dir);
1414#endif
1415 VMCOREINFO_SYMBOL(_stext);
1416 VMCOREINFO_SYMBOL(vmap_area_list);
1417
1418#ifndef CONFIG_NEED_MULTIPLE_NODES
1419 VMCOREINFO_SYMBOL(mem_map);
1420 VMCOREINFO_SYMBOL(contig_page_data);
1421#endif
1422#ifdef CONFIG_SPARSEMEM
1423 VMCOREINFO_SYMBOL(mem_section);
1424 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1425 VMCOREINFO_STRUCT_SIZE(mem_section);
1426 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1427#endif
1428 VMCOREINFO_STRUCT_SIZE(page);
1429 VMCOREINFO_STRUCT_SIZE(pglist_data);
1430 VMCOREINFO_STRUCT_SIZE(zone);
1431 VMCOREINFO_STRUCT_SIZE(free_area);
1432 VMCOREINFO_STRUCT_SIZE(list_head);
1433 VMCOREINFO_SIZE(nodemask_t);
1434 VMCOREINFO_OFFSET(page, flags);
1435 VMCOREINFO_OFFSET(page, _refcount);
1436 VMCOREINFO_OFFSET(page, mapping);
1437 VMCOREINFO_OFFSET(page, lru);
1438 VMCOREINFO_OFFSET(page, _mapcount);
1439 VMCOREINFO_OFFSET(page, private);
1440 VMCOREINFO_OFFSET(page, compound_dtor);
1441 VMCOREINFO_OFFSET(page, compound_order);
1442 VMCOREINFO_OFFSET(page, compound_head);
1443 VMCOREINFO_OFFSET(pglist_data, node_zones);
1444 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1445#ifdef CONFIG_FLAT_NODE_MEM_MAP
1446 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1447#endif
1448 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1449 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1450 VMCOREINFO_OFFSET(pglist_data, node_id);
1451 VMCOREINFO_OFFSET(zone, free_area);
1452 VMCOREINFO_OFFSET(zone, vm_stat);
1453 VMCOREINFO_OFFSET(zone, spanned_pages);
1454 VMCOREINFO_OFFSET(free_area, free_list);
1455 VMCOREINFO_OFFSET(list_head, next);
1456 VMCOREINFO_OFFSET(list_head, prev);
1457 VMCOREINFO_OFFSET(vmap_area, va_start);
1458 VMCOREINFO_OFFSET(vmap_area, list);
1459 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1460 log_buf_kexec_setup();
1461 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1462 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1463 VMCOREINFO_NUMBER(PG_lru);
1464 VMCOREINFO_NUMBER(PG_private);
1465 VMCOREINFO_NUMBER(PG_swapcache);
1466 VMCOREINFO_NUMBER(PG_slab);
1467#ifdef CONFIG_MEMORY_FAILURE
1468 VMCOREINFO_NUMBER(PG_hwpoison);
1469#endif
1470 VMCOREINFO_NUMBER(PG_head_mask);
1471 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1472#ifdef CONFIG_HUGETLB_PAGE
1473 VMCOREINFO_NUMBER(HUGETLB_PAGE_DTOR);
1474#endif
1475
1476 arch_crash_save_vmcoreinfo();
1477 update_vmcoreinfo_note();
1478
1479 return 0;
1480}
1481
1482subsys_initcall(crash_save_vmcoreinfo_init);
1483
1484/*
1485 * Move into place and start executing a preloaded standalone
1486 * executable. If nothing was preloaded return an error.
1487 */
1488int kernel_kexec(void)
1489{
1490 int error = 0;
1491
1492 if (!mutex_trylock(&kexec_mutex))
1493 return -EBUSY;
1494 if (!kexec_image) {
1495 error = -EINVAL;
1496 goto Unlock;
1497 }
1498
1499#ifdef CONFIG_KEXEC_JUMP
1500 if (kexec_image->preserve_context) {
1501 lock_system_sleep();
1502 pm_prepare_console();
1503 error = freeze_processes();
1504 if (error) {
1505 error = -EBUSY;
1506 goto Restore_console;
1507 }
1508 suspend_console();
1509 error = dpm_suspend_start(PMSG_FREEZE);
1510 if (error)
1511 goto Resume_console;
1512 /* At this point, dpm_suspend_start() has been called,
1513 * but *not* dpm_suspend_end(). We *must* call
1514 * dpm_suspend_end() now. Otherwise, drivers for
1515 * some devices (e.g. interrupt controllers) become
1516 * desynchronized with the actual state of the
1517 * hardware at resume time, and evil weirdness ensues.
1518 */
1519 error = dpm_suspend_end(PMSG_FREEZE);
1520 if (error)
1521 goto Resume_devices;
1522 error = disable_nonboot_cpus();
1523 if (error)
1524 goto Enable_cpus;
1525 local_irq_disable();
1526 error = syscore_suspend();
1527 if (error)
1528 goto Enable_irqs;
1529 } else
1530#endif
1531 {
1532 kexec_in_progress = true;
1533 kernel_restart_prepare(NULL);
1534 migrate_to_reboot_cpu();
1535
1536 /*
1537 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1538 * no further code needs to use CPU hotplug (which is true in
1539 * the reboot case). However, the kexec path depends on using
1540 * CPU hotplug again; so re-enable it here.
1541 */
1542 cpu_hotplug_enable();
1543 pr_emerg("Starting new kernel\n");
1544 machine_shutdown();
1545 }
1546
1547 machine_kexec(kexec_image);
1548
1549#ifdef CONFIG_KEXEC_JUMP
1550 if (kexec_image->preserve_context) {
1551 syscore_resume();
1552 Enable_irqs:
1553 local_irq_enable();
1554 Enable_cpus:
1555 enable_nonboot_cpus();
1556 dpm_resume_start(PMSG_RESTORE);
1557 Resume_devices:
1558 dpm_resume_end(PMSG_RESTORE);
1559 Resume_console:
1560 resume_console();
1561 thaw_processes();
1562 Restore_console:
1563 pm_restore_console();
1564 unlock_system_sleep();
1565 }
1566#endif
1567
1568 Unlock:
1569 mutex_unlock(&kexec_mutex);
1570 return error;
1571}
1572
1573/*
1574 * Protection mechanism for crashkernel reserved memory after
1575 * the kdump kernel is loaded.
1576 *
1577 * Provide an empty default implementation here -- architecture
1578 * code may override this
1579 */
1580void __weak arch_kexec_protect_crashkres(void)
1581{}
1582
1583void __weak arch_kexec_unprotect_crashkres(void)
1584{}
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
57int kexec_should_crash(struct task_struct *p)
58{
59 /*
60 * If crash_kexec_post_notifiers is enabled, don't run
61 * crash_kexec() here yet, which must be run after panic
62 * notifiers in panic().
63 */
64 if (crash_kexec_post_notifiers)
65 return 0;
66 /*
67 * There are 4 panic() calls in make_task_dead() path, each of which
68 * corresponds to each of these 4 conditions.
69 */
70 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
71 return 1;
72 return 0;
73}
74
75int kexec_crash_loaded(void)
76{
77 return !!kexec_crash_image;
78}
79EXPORT_SYMBOL_GPL(kexec_crash_loaded);
80
81/*
82 * When kexec transitions to the new kernel there is a one-to-one
83 * mapping between physical and virtual addresses. On processors
84 * where you can disable the MMU this is trivial, and easy. For
85 * others it is still a simple predictable page table to setup.
86 *
87 * In that environment kexec copies the new kernel to its final
88 * resting place. This means I can only support memory whose
89 * physical address can fit in an unsigned long. In particular
90 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
91 * If the assembly stub has more restrictive requirements
92 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
93 * defined more restrictively in <asm/kexec.h>.
94 *
95 * The code for the transition from the current kernel to the
96 * new kernel is placed in the control_code_buffer, whose size
97 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
98 * page of memory is necessary, but some architectures require more.
99 * Because this memory must be identity mapped in the transition from
100 * virtual to physical addresses it must live in the range
101 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
102 * modifiable.
103 *
104 * The assembly stub in the control code buffer is passed a linked list
105 * of descriptor pages detailing the source pages of the new kernel,
106 * and the destination addresses of those source pages. As this data
107 * structure is not used in the context of the current OS, it must
108 * be self-contained.
109 *
110 * The code has been made to work with highmem pages and will use a
111 * destination page in its final resting place (if it happens
112 * to allocate it). The end product of this is that most of the
113 * physical address space, and most of RAM can be used.
114 *
115 * Future directions include:
116 * - allocating a page table with the control code buffer identity
117 * mapped, to simplify machine_kexec and make kexec_on_panic more
118 * reliable.
119 */
120
121/*
122 * KIMAGE_NO_DEST is an impossible destination address..., for
123 * allocating pages whose destination address we do not care about.
124 */
125#define KIMAGE_NO_DEST (-1UL)
126#define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
127
128static struct page *kimage_alloc_page(struct kimage *image,
129 gfp_t gfp_mask,
130 unsigned long dest);
131
132int sanity_check_segment_list(struct kimage *image)
133{
134 int i;
135 unsigned long nr_segments = image->nr_segments;
136 unsigned long total_pages = 0;
137 unsigned long nr_pages = totalram_pages();
138
139 /*
140 * Verify we have good destination addresses. The caller is
141 * responsible for making certain we don't attempt to load
142 * the new image into invalid or reserved areas of RAM. This
143 * just verifies it is an address we can use.
144 *
145 * Since the kernel does everything in page size chunks ensure
146 * the destination addresses are page aligned. Too many
147 * special cases crop of when we don't do this. The most
148 * insidious is getting overlapping destination addresses
149 * simply because addresses are changed to page size
150 * granularity.
151 */
152 for (i = 0; i < nr_segments; i++) {
153 unsigned long mstart, mend;
154
155 mstart = image->segment[i].mem;
156 mend = mstart + image->segment[i].memsz;
157 if (mstart > mend)
158 return -EADDRNOTAVAIL;
159 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
160 return -EADDRNOTAVAIL;
161 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
162 return -EADDRNOTAVAIL;
163 }
164
165 /* Verify our destination addresses do not overlap.
166 * If we alloed overlapping destination addresses
167 * through very weird things can happen with no
168 * easy explanation as one segment stops on another.
169 */
170 for (i = 0; i < nr_segments; i++) {
171 unsigned long mstart, mend;
172 unsigned long j;
173
174 mstart = image->segment[i].mem;
175 mend = mstart + image->segment[i].memsz;
176 for (j = 0; j < i; j++) {
177 unsigned long pstart, pend;
178
179 pstart = image->segment[j].mem;
180 pend = pstart + image->segment[j].memsz;
181 /* Do the segments overlap ? */
182 if ((mend > pstart) && (mstart < pend))
183 return -EINVAL;
184 }
185 }
186
187 /* Ensure our buffer sizes are strictly less than
188 * our memory sizes. This should always be the case,
189 * and it is easier to check up front than to be surprised
190 * later on.
191 */
192 for (i = 0; i < nr_segments; i++) {
193 if (image->segment[i].bufsz > image->segment[i].memsz)
194 return -EINVAL;
195 }
196
197 /*
198 * Verify that no more than half of memory will be consumed. If the
199 * request from userspace is too large, a large amount of time will be
200 * wasted allocating pages, which can cause a soft lockup.
201 */
202 for (i = 0; i < nr_segments; i++) {
203 if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
204 return -EINVAL;
205
206 total_pages += PAGE_COUNT(image->segment[i].memsz);
207 }
208
209 if (total_pages > nr_pages / 2)
210 return -EINVAL;
211
212 /*
213 * Verify we have good destination addresses. Normally
214 * the caller is responsible for making certain we don't
215 * attempt to load the new image into invalid or reserved
216 * areas of RAM. But crash kernels are preloaded into a
217 * reserved area of ram. We must ensure the addresses
218 * are in the reserved area otherwise preloading the
219 * kernel could corrupt things.
220 */
221
222 if (image->type == KEXEC_TYPE_CRASH) {
223 for (i = 0; i < nr_segments; i++) {
224 unsigned long mstart, mend;
225
226 mstart = image->segment[i].mem;
227 mend = mstart + image->segment[i].memsz - 1;
228 /* Ensure we are within the crash kernel limits */
229 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
230 (mend > phys_to_boot_phys(crashk_res.end)))
231 return -EADDRNOTAVAIL;
232 }
233 }
234
235 return 0;
236}
237
238struct kimage *do_kimage_alloc_init(void)
239{
240 struct kimage *image;
241
242 /* Allocate a controlling structure */
243 image = kzalloc(sizeof(*image), GFP_KERNEL);
244 if (!image)
245 return NULL;
246
247 image->head = 0;
248 image->entry = &image->head;
249 image->last_entry = &image->head;
250 image->control_page = ~0; /* By default this does not apply */
251 image->type = KEXEC_TYPE_DEFAULT;
252
253 /* Initialize the list of control pages */
254 INIT_LIST_HEAD(&image->control_pages);
255
256 /* Initialize the list of destination pages */
257 INIT_LIST_HEAD(&image->dest_pages);
258
259 /* Initialize the list of unusable pages */
260 INIT_LIST_HEAD(&image->unusable_pages);
261
262#ifdef CONFIG_CRASH_HOTPLUG
263 image->hp_action = KEXEC_CRASH_HP_NONE;
264 image->elfcorehdr_index = -1;
265 image->elfcorehdr_updated = false;
266#endif
267
268 return image;
269}
270
271int kimage_is_destination_range(struct kimage *image,
272 unsigned long start,
273 unsigned long end)
274{
275 unsigned long i;
276
277 for (i = 0; i < image->nr_segments; i++) {
278 unsigned long mstart, mend;
279
280 mstart = image->segment[i].mem;
281 mend = mstart + image->segment[i].memsz - 1;
282 if ((end >= mstart) && (start <= mend))
283 return 1;
284 }
285
286 return 0;
287}
288
289static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
290{
291 struct page *pages;
292
293 if (fatal_signal_pending(current))
294 return NULL;
295 pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
296 if (pages) {
297 unsigned int count, i;
298
299 pages->mapping = NULL;
300 set_page_private(pages, order);
301 count = 1 << order;
302 for (i = 0; i < count; i++)
303 SetPageReserved(pages + i);
304
305 arch_kexec_post_alloc_pages(page_address(pages), count,
306 gfp_mask);
307
308 if (gfp_mask & __GFP_ZERO)
309 for (i = 0; i < count; i++)
310 clear_highpage(pages + i);
311 }
312
313 return pages;
314}
315
316static void kimage_free_pages(struct page *page)
317{
318 unsigned int order, count, i;
319
320 order = page_private(page);
321 count = 1 << order;
322
323 arch_kexec_pre_free_pages(page_address(page), count);
324
325 for (i = 0; i < count; i++)
326 ClearPageReserved(page + i);
327 __free_pages(page, order);
328}
329
330void kimage_free_page_list(struct list_head *list)
331{
332 struct page *page, *next;
333
334 list_for_each_entry_safe(page, next, list, lru) {
335 list_del(&page->lru);
336 kimage_free_pages(page);
337 }
338}
339
340static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
341 unsigned int order)
342{
343 /* Control pages are special, they are the intermediaries
344 * that are needed while we copy the rest of the pages
345 * to their final resting place. As such they must
346 * not conflict with either the destination addresses
347 * or memory the kernel is already using.
348 *
349 * The only case where we really need more than one of
350 * these are for architectures where we cannot disable
351 * the MMU and must instead generate an identity mapped
352 * page table for all of the memory.
353 *
354 * At worst this runs in O(N) of the image size.
355 */
356 struct list_head extra_pages;
357 struct page *pages;
358 unsigned int count;
359
360 count = 1 << order;
361 INIT_LIST_HEAD(&extra_pages);
362
363 /* Loop while I can allocate a page and the page allocated
364 * is a destination page.
365 */
366 do {
367 unsigned long pfn, epfn, addr, eaddr;
368
369 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
370 if (!pages)
371 break;
372 pfn = page_to_boot_pfn(pages);
373 epfn = pfn + count;
374 addr = pfn << PAGE_SHIFT;
375 eaddr = (epfn << PAGE_SHIFT) - 1;
376 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
377 kimage_is_destination_range(image, addr, eaddr)) {
378 list_add(&pages->lru, &extra_pages);
379 pages = NULL;
380 }
381 } while (!pages);
382
383 if (pages) {
384 /* Remember the allocated page... */
385 list_add(&pages->lru, &image->control_pages);
386
387 /* Because the page is already in it's destination
388 * location we will never allocate another page at
389 * that address. Therefore kimage_alloc_pages
390 * will not return it (again) and we don't need
391 * to give it an entry in image->segment[].
392 */
393 }
394 /* Deal with the destination pages I have inadvertently allocated.
395 *
396 * Ideally I would convert multi-page allocations into single
397 * page allocations, and add everything to image->dest_pages.
398 *
399 * For now it is simpler to just free the pages.
400 */
401 kimage_free_page_list(&extra_pages);
402
403 return pages;
404}
405
406static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
407 unsigned int order)
408{
409 /* Control pages are special, they are the intermediaries
410 * that are needed while we copy the rest of the pages
411 * to their final resting place. As such they must
412 * not conflict with either the destination addresses
413 * or memory the kernel is already using.
414 *
415 * Control pages are also the only pags we must allocate
416 * when loading a crash kernel. All of the other pages
417 * are specified by the segments and we just memcpy
418 * into them directly.
419 *
420 * The only case where we really need more than one of
421 * these are for architectures where we cannot disable
422 * the MMU and must instead generate an identity mapped
423 * page table for all of the memory.
424 *
425 * Given the low demand this implements a very simple
426 * allocator that finds the first hole of the appropriate
427 * size in the reserved memory region, and allocates all
428 * of the memory up to and including the hole.
429 */
430 unsigned long hole_start, hole_end, size;
431 struct page *pages;
432
433 pages = NULL;
434 size = (1 << order) << PAGE_SHIFT;
435 hole_start = ALIGN(image->control_page, size);
436 hole_end = hole_start + size - 1;
437 while (hole_end <= crashk_res.end) {
438 unsigned long i;
439
440 cond_resched();
441
442 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
443 break;
444 /* See if I overlap any of the segments */
445 for (i = 0; i < image->nr_segments; i++) {
446 unsigned long mstart, mend;
447
448 mstart = image->segment[i].mem;
449 mend = mstart + image->segment[i].memsz - 1;
450 if ((hole_end >= mstart) && (hole_start <= mend)) {
451 /* Advance the hole to the end of the segment */
452 hole_start = ALIGN(mend, size);
453 hole_end = hole_start + size - 1;
454 break;
455 }
456 }
457 /* If I don't overlap any segments I have found my hole! */
458 if (i == image->nr_segments) {
459 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
460 image->control_page = hole_end + 1;
461 break;
462 }
463 }
464
465 /* Ensure that these pages are decrypted if SME is enabled. */
466 if (pages)
467 arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
468
469 return pages;
470}
471
472
473struct page *kimage_alloc_control_pages(struct kimage *image,
474 unsigned int order)
475{
476 struct page *pages = NULL;
477
478 switch (image->type) {
479 case KEXEC_TYPE_DEFAULT:
480 pages = kimage_alloc_normal_control_pages(image, order);
481 break;
482 case KEXEC_TYPE_CRASH:
483 pages = kimage_alloc_crash_control_pages(image, order);
484 break;
485 }
486
487 return pages;
488}
489
490int kimage_crash_copy_vmcoreinfo(struct kimage *image)
491{
492 struct page *vmcoreinfo_page;
493 void *safecopy;
494
495 if (image->type != KEXEC_TYPE_CRASH)
496 return 0;
497
498 /*
499 * For kdump, allocate one vmcoreinfo safe copy from the
500 * crash memory. as we have arch_kexec_protect_crashkres()
501 * after kexec syscall, we naturally protect it from write
502 * (even read) access under kernel direct mapping. But on
503 * the other hand, we still need to operate it when crash
504 * happens to generate vmcoreinfo note, hereby we rely on
505 * vmap for this purpose.
506 */
507 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
508 if (!vmcoreinfo_page) {
509 pr_warn("Could not allocate vmcoreinfo buffer\n");
510 return -ENOMEM;
511 }
512 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
513 if (!safecopy) {
514 pr_warn("Could not vmap vmcoreinfo buffer\n");
515 return -ENOMEM;
516 }
517
518 image->vmcoreinfo_data_copy = safecopy;
519 crash_update_vmcoreinfo_safecopy(safecopy);
520
521 return 0;
522}
523
524static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
525{
526 if (*image->entry != 0)
527 image->entry++;
528
529 if (image->entry == image->last_entry) {
530 kimage_entry_t *ind_page;
531 struct page *page;
532
533 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
534 if (!page)
535 return -ENOMEM;
536
537 ind_page = page_address(page);
538 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
539 image->entry = ind_page;
540 image->last_entry = ind_page +
541 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
542 }
543 *image->entry = entry;
544 image->entry++;
545 *image->entry = 0;
546
547 return 0;
548}
549
550static int kimage_set_destination(struct kimage *image,
551 unsigned long destination)
552{
553 destination &= PAGE_MASK;
554
555 return kimage_add_entry(image, destination | IND_DESTINATION);
556}
557
558
559static int kimage_add_page(struct kimage *image, unsigned long page)
560{
561 page &= PAGE_MASK;
562
563 return kimage_add_entry(image, page | IND_SOURCE);
564}
565
566
567static void kimage_free_extra_pages(struct kimage *image)
568{
569 /* Walk through and free any extra destination pages I may have */
570 kimage_free_page_list(&image->dest_pages);
571
572 /* Walk through and free any unusable pages I have cached */
573 kimage_free_page_list(&image->unusable_pages);
574
575}
576
577void kimage_terminate(struct kimage *image)
578{
579 if (*image->entry != 0)
580 image->entry++;
581
582 *image->entry = IND_DONE;
583}
584
585#define for_each_kimage_entry(image, ptr, entry) \
586 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
587 ptr = (entry & IND_INDIRECTION) ? \
588 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
589
590static void kimage_free_entry(kimage_entry_t entry)
591{
592 struct page *page;
593
594 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
595 kimage_free_pages(page);
596}
597
598void kimage_free(struct kimage *image)
599{
600 kimage_entry_t *ptr, entry;
601 kimage_entry_t ind = 0;
602
603 if (!image)
604 return;
605
606 if (image->vmcoreinfo_data_copy) {
607 crash_update_vmcoreinfo_safecopy(NULL);
608 vunmap(image->vmcoreinfo_data_copy);
609 }
610
611 kimage_free_extra_pages(image);
612 for_each_kimage_entry(image, ptr, entry) {
613 if (entry & IND_INDIRECTION) {
614 /* Free the previous indirection page */
615 if (ind & IND_INDIRECTION)
616 kimage_free_entry(ind);
617 /* Save this indirection page until we are
618 * done with it.
619 */
620 ind = entry;
621 } else if (entry & IND_SOURCE)
622 kimage_free_entry(entry);
623 }
624 /* Free the final indirection page */
625 if (ind & IND_INDIRECTION)
626 kimage_free_entry(ind);
627
628 /* Handle any machine specific cleanup */
629 machine_kexec_cleanup(image);
630
631 /* Free the kexec control pages... */
632 kimage_free_page_list(&image->control_pages);
633
634 /*
635 * Free up any temporary buffers allocated. This might hit if
636 * error occurred much later after buffer allocation.
637 */
638 if (image->file_mode)
639 kimage_file_post_load_cleanup(image);
640
641 kfree(image);
642}
643
644static kimage_entry_t *kimage_dst_used(struct kimage *image,
645 unsigned long page)
646{
647 kimage_entry_t *ptr, entry;
648 unsigned long destination = 0;
649
650 for_each_kimage_entry(image, ptr, entry) {
651 if (entry & IND_DESTINATION)
652 destination = entry & PAGE_MASK;
653 else if (entry & IND_SOURCE) {
654 if (page == destination)
655 return ptr;
656 destination += PAGE_SIZE;
657 }
658 }
659
660 return NULL;
661}
662
663static struct page *kimage_alloc_page(struct kimage *image,
664 gfp_t gfp_mask,
665 unsigned long destination)
666{
667 /*
668 * Here we implement safeguards to ensure that a source page
669 * is not copied to its destination page before the data on
670 * the destination page is no longer useful.
671 *
672 * To do this we maintain the invariant that a source page is
673 * either its own destination page, or it is not a
674 * destination page at all.
675 *
676 * That is slightly stronger than required, but the proof
677 * that no problems will not occur is trivial, and the
678 * implementation is simply to verify.
679 *
680 * When allocating all pages normally this algorithm will run
681 * in O(N) time, but in the worst case it will run in O(N^2)
682 * time. If the runtime is a problem the data structures can
683 * be fixed.
684 */
685 struct page *page;
686 unsigned long addr;
687
688 /*
689 * Walk through the list of destination pages, and see if I
690 * have a match.
691 */
692 list_for_each_entry(page, &image->dest_pages, lru) {
693 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
694 if (addr == destination) {
695 list_del(&page->lru);
696 return page;
697 }
698 }
699 page = NULL;
700 while (1) {
701 kimage_entry_t *old;
702
703 /* Allocate a page, if we run out of memory give up */
704 page = kimage_alloc_pages(gfp_mask, 0);
705 if (!page)
706 return NULL;
707 /* If the page cannot be used file it away */
708 if (page_to_boot_pfn(page) >
709 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
710 list_add(&page->lru, &image->unusable_pages);
711 continue;
712 }
713 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
714
715 /* If it is the destination page we want use it */
716 if (addr == destination)
717 break;
718
719 /* If the page is not a destination page use it */
720 if (!kimage_is_destination_range(image, addr,
721 addr + PAGE_SIZE - 1))
722 break;
723
724 /*
725 * I know that the page is someones destination page.
726 * See if there is already a source page for this
727 * destination page. And if so swap the source pages.
728 */
729 old = kimage_dst_used(image, addr);
730 if (old) {
731 /* If so move it */
732 unsigned long old_addr;
733 struct page *old_page;
734
735 old_addr = *old & PAGE_MASK;
736 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
737 copy_highpage(page, old_page);
738 *old = addr | (*old & ~PAGE_MASK);
739
740 /* The old page I have found cannot be a
741 * destination page, so return it if it's
742 * gfp_flags honor the ones passed in.
743 */
744 if (!(gfp_mask & __GFP_HIGHMEM) &&
745 PageHighMem(old_page)) {
746 kimage_free_pages(old_page);
747 continue;
748 }
749 page = old_page;
750 break;
751 }
752 /* Place the page on the destination list, to be used later */
753 list_add(&page->lru, &image->dest_pages);
754 }
755
756 return page;
757}
758
759static int kimage_load_normal_segment(struct kimage *image,
760 struct kexec_segment *segment)
761{
762 unsigned long maddr;
763 size_t ubytes, mbytes;
764 int result;
765 unsigned char __user *buf = NULL;
766 unsigned char *kbuf = NULL;
767
768 if (image->file_mode)
769 kbuf = segment->kbuf;
770 else
771 buf = segment->buf;
772 ubytes = segment->bufsz;
773 mbytes = segment->memsz;
774 maddr = segment->mem;
775
776 result = kimage_set_destination(image, maddr);
777 if (result < 0)
778 goto out;
779
780 while (mbytes) {
781 struct page *page;
782 char *ptr;
783 size_t uchunk, mchunk;
784
785 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
786 if (!page) {
787 result = -ENOMEM;
788 goto out;
789 }
790 result = kimage_add_page(image, page_to_boot_pfn(page)
791 << PAGE_SHIFT);
792 if (result < 0)
793 goto out;
794
795 ptr = kmap_local_page(page);
796 /* Start with a clear page */
797 clear_page(ptr);
798 ptr += maddr & ~PAGE_MASK;
799 mchunk = min_t(size_t, mbytes,
800 PAGE_SIZE - (maddr & ~PAGE_MASK));
801 uchunk = min(ubytes, mchunk);
802
803 /* For file based kexec, source pages are in kernel memory */
804 if (image->file_mode)
805 memcpy(ptr, kbuf, uchunk);
806 else
807 result = copy_from_user(ptr, buf, uchunk);
808 kunmap_local(ptr);
809 if (result) {
810 result = -EFAULT;
811 goto out;
812 }
813 ubytes -= uchunk;
814 maddr += mchunk;
815 if (image->file_mode)
816 kbuf += mchunk;
817 else
818 buf += mchunk;
819 mbytes -= mchunk;
820
821 cond_resched();
822 }
823out:
824 return result;
825}
826
827static int kimage_load_crash_segment(struct kimage *image,
828 struct kexec_segment *segment)
829{
830 /* For crash dumps kernels we simply copy the data from
831 * user space to it's destination.
832 * We do things a page at a time for the sake of kmap.
833 */
834 unsigned long maddr;
835 size_t ubytes, mbytes;
836 int result;
837 unsigned char __user *buf = NULL;
838 unsigned char *kbuf = NULL;
839
840 result = 0;
841 if (image->file_mode)
842 kbuf = segment->kbuf;
843 else
844 buf = segment->buf;
845 ubytes = segment->bufsz;
846 mbytes = segment->memsz;
847 maddr = segment->mem;
848 while (mbytes) {
849 struct page *page;
850 char *ptr;
851 size_t uchunk, mchunk;
852
853 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
854 if (!page) {
855 result = -ENOMEM;
856 goto out;
857 }
858 arch_kexec_post_alloc_pages(page_address(page), 1, 0);
859 ptr = kmap_local_page(page);
860 ptr += maddr & ~PAGE_MASK;
861 mchunk = min_t(size_t, mbytes,
862 PAGE_SIZE - (maddr & ~PAGE_MASK));
863 uchunk = min(ubytes, mchunk);
864 if (mchunk > uchunk) {
865 /* Zero the trailing part of the page */
866 memset(ptr + uchunk, 0, mchunk - uchunk);
867 }
868
869 /* For file based kexec, source pages are in kernel memory */
870 if (image->file_mode)
871 memcpy(ptr, kbuf, uchunk);
872 else
873 result = copy_from_user(ptr, buf, uchunk);
874 kexec_flush_icache_page(page);
875 kunmap_local(ptr);
876 arch_kexec_pre_free_pages(page_address(page), 1);
877 if (result) {
878 result = -EFAULT;
879 goto out;
880 }
881 ubytes -= uchunk;
882 maddr += mchunk;
883 if (image->file_mode)
884 kbuf += mchunk;
885 else
886 buf += mchunk;
887 mbytes -= mchunk;
888
889 cond_resched();
890 }
891out:
892 return result;
893}
894
895int kimage_load_segment(struct kimage *image,
896 struct kexec_segment *segment)
897{
898 int result = -ENOMEM;
899
900 switch (image->type) {
901 case KEXEC_TYPE_DEFAULT:
902 result = kimage_load_normal_segment(image, segment);
903 break;
904 case KEXEC_TYPE_CRASH:
905 result = kimage_load_crash_segment(image, segment);
906 break;
907 }
908
909 return result;
910}
911
912struct kexec_load_limit {
913 /* Mutex protects the limit count. */
914 struct mutex mutex;
915 int limit;
916};
917
918static struct kexec_load_limit load_limit_reboot = {
919 .mutex = __MUTEX_INITIALIZER(load_limit_reboot.mutex),
920 .limit = -1,
921};
922
923static struct kexec_load_limit load_limit_panic = {
924 .mutex = __MUTEX_INITIALIZER(load_limit_panic.mutex),
925 .limit = -1,
926};
927
928struct kimage *kexec_image;
929struct kimage *kexec_crash_image;
930static int kexec_load_disabled;
931
932#ifdef CONFIG_SYSCTL
933static int kexec_limit_handler(struct ctl_table *table, int write,
934 void *buffer, size_t *lenp, loff_t *ppos)
935{
936 struct kexec_load_limit *limit = table->data;
937 int val;
938 struct ctl_table tmp = {
939 .data = &val,
940 .maxlen = sizeof(val),
941 .mode = table->mode,
942 };
943 int ret;
944
945 if (write) {
946 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
947 if (ret)
948 return ret;
949
950 if (val < 0)
951 return -EINVAL;
952
953 mutex_lock(&limit->mutex);
954 if (limit->limit != -1 && val >= limit->limit)
955 ret = -EINVAL;
956 else
957 limit->limit = val;
958 mutex_unlock(&limit->mutex);
959
960 return ret;
961 }
962
963 mutex_lock(&limit->mutex);
964 val = limit->limit;
965 mutex_unlock(&limit->mutex);
966
967 return proc_dointvec(&tmp, write, buffer, lenp, ppos);
968}
969
970static struct ctl_table kexec_core_sysctls[] = {
971 {
972 .procname = "kexec_load_disabled",
973 .data = &kexec_load_disabled,
974 .maxlen = sizeof(int),
975 .mode = 0644,
976 /* only handle a transition from default "0" to "1" */
977 .proc_handler = proc_dointvec_minmax,
978 .extra1 = SYSCTL_ONE,
979 .extra2 = SYSCTL_ONE,
980 },
981 {
982 .procname = "kexec_load_limit_panic",
983 .data = &load_limit_panic,
984 .mode = 0644,
985 .proc_handler = kexec_limit_handler,
986 },
987 {
988 .procname = "kexec_load_limit_reboot",
989 .data = &load_limit_reboot,
990 .mode = 0644,
991 .proc_handler = kexec_limit_handler,
992 },
993 { }
994};
995
996static int __init kexec_core_sysctl_init(void)
997{
998 register_sysctl_init("kernel", kexec_core_sysctls);
999 return 0;
1000}
1001late_initcall(kexec_core_sysctl_init);
1002#endif
1003
1004bool kexec_load_permitted(int kexec_image_type)
1005{
1006 struct kexec_load_limit *limit;
1007
1008 /*
1009 * Only the superuser can use the kexec syscall and if it has not
1010 * been disabled.
1011 */
1012 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1013 return false;
1014
1015 /* Check limit counter and decrease it.*/
1016 limit = (kexec_image_type == KEXEC_TYPE_CRASH) ?
1017 &load_limit_panic : &load_limit_reboot;
1018 mutex_lock(&limit->mutex);
1019 if (!limit->limit) {
1020 mutex_unlock(&limit->mutex);
1021 return false;
1022 }
1023 if (limit->limit != -1)
1024 limit->limit--;
1025 mutex_unlock(&limit->mutex);
1026
1027 return true;
1028}
1029
1030/*
1031 * No panic_cpu check version of crash_kexec(). This function is called
1032 * only when panic_cpu holds the current CPU number; this is the only CPU
1033 * which processes crash_kexec routines.
1034 */
1035void __noclone __crash_kexec(struct pt_regs *regs)
1036{
1037 /* Take the kexec_lock here to prevent sys_kexec_load
1038 * running on one cpu from replacing the crash kernel
1039 * we are using after a panic on a different cpu.
1040 *
1041 * If the crash kernel was not located in a fixed area
1042 * of memory the xchg(&kexec_crash_image) would be
1043 * sufficient. But since I reuse the memory...
1044 */
1045 if (kexec_trylock()) {
1046 if (kexec_crash_image) {
1047 struct pt_regs fixed_regs;
1048
1049 crash_setup_regs(&fixed_regs, regs);
1050 crash_save_vmcoreinfo();
1051 machine_crash_shutdown(&fixed_regs);
1052 machine_kexec(kexec_crash_image);
1053 }
1054 kexec_unlock();
1055 }
1056}
1057STACK_FRAME_NON_STANDARD(__crash_kexec);
1058
1059__bpf_kfunc void crash_kexec(struct pt_regs *regs)
1060{
1061 int old_cpu, this_cpu;
1062
1063 /*
1064 * Only one CPU is allowed to execute the crash_kexec() code as with
1065 * panic(). Otherwise parallel calls of panic() and crash_kexec()
1066 * may stop each other. To exclude them, we use panic_cpu here too.
1067 */
1068 old_cpu = PANIC_CPU_INVALID;
1069 this_cpu = raw_smp_processor_id();
1070
1071 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
1072 /* This is the 1st CPU which comes here, so go ahead. */
1073 __crash_kexec(regs);
1074
1075 /*
1076 * Reset panic_cpu to allow another panic()/crash_kexec()
1077 * call.
1078 */
1079 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
1080 }
1081}
1082
1083static inline resource_size_t crash_resource_size(const struct resource *res)
1084{
1085 return !res->end ? 0 : resource_size(res);
1086}
1087
1088ssize_t crash_get_memory_size(void)
1089{
1090 ssize_t size = 0;
1091
1092 if (!kexec_trylock())
1093 return -EBUSY;
1094
1095 size += crash_resource_size(&crashk_res);
1096 size += crash_resource_size(&crashk_low_res);
1097
1098 kexec_unlock();
1099 return size;
1100}
1101
1102static int __crash_shrink_memory(struct resource *old_res,
1103 unsigned long new_size)
1104{
1105 struct resource *ram_res;
1106
1107 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1108 if (!ram_res)
1109 return -ENOMEM;
1110
1111 ram_res->start = old_res->start + new_size;
1112 ram_res->end = old_res->end;
1113 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1114 ram_res->name = "System RAM";
1115
1116 if (!new_size) {
1117 release_resource(old_res);
1118 old_res->start = 0;
1119 old_res->end = 0;
1120 } else {
1121 crashk_res.end = ram_res->start - 1;
1122 }
1123
1124 crash_free_reserved_phys_range(ram_res->start, ram_res->end);
1125 insert_resource(&iomem_resource, ram_res);
1126
1127 return 0;
1128}
1129
1130int crash_shrink_memory(unsigned long new_size)
1131{
1132 int ret = 0;
1133 unsigned long old_size, low_size;
1134
1135 if (!kexec_trylock())
1136 return -EBUSY;
1137
1138 if (kexec_crash_image) {
1139 ret = -ENOENT;
1140 goto unlock;
1141 }
1142
1143 low_size = crash_resource_size(&crashk_low_res);
1144 old_size = crash_resource_size(&crashk_res) + low_size;
1145 new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
1146 if (new_size >= old_size) {
1147 ret = (new_size == old_size) ? 0 : -EINVAL;
1148 goto unlock;
1149 }
1150
1151 /*
1152 * (low_size > new_size) implies that low_size is greater than zero.
1153 * This also means that if low_size is zero, the else branch is taken.
1154 *
1155 * If low_size is greater than 0, (low_size > new_size) indicates that
1156 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
1157 * needs to be shrunken.
1158 */
1159 if (low_size > new_size) {
1160 ret = __crash_shrink_memory(&crashk_res, 0);
1161 if (ret)
1162 goto unlock;
1163
1164 ret = __crash_shrink_memory(&crashk_low_res, new_size);
1165 } else {
1166 ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
1167 }
1168
1169 /* Swap crashk_res and crashk_low_res if needed */
1170 if (!crashk_res.end && crashk_low_res.end) {
1171 crashk_res.start = crashk_low_res.start;
1172 crashk_res.end = crashk_low_res.end;
1173 release_resource(&crashk_low_res);
1174 crashk_low_res.start = 0;
1175 crashk_low_res.end = 0;
1176 insert_resource(&iomem_resource, &crashk_res);
1177 }
1178
1179unlock:
1180 kexec_unlock();
1181 return ret;
1182}
1183
1184void crash_save_cpu(struct pt_regs *regs, int cpu)
1185{
1186 struct elf_prstatus prstatus;
1187 u32 *buf;
1188
1189 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1190 return;
1191
1192 /* Using ELF notes here is opportunistic.
1193 * I need a well defined structure format
1194 * for the data I pass, and I need tags
1195 * on the data to indicate what information I have
1196 * squirrelled away. ELF notes happen to provide
1197 * all of that, so there is no need to invent something new.
1198 */
1199 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1200 if (!buf)
1201 return;
1202 memset(&prstatus, 0, sizeof(prstatus));
1203 prstatus.common.pr_pid = current->pid;
1204 elf_core_copy_regs(&prstatus.pr_reg, regs);
1205 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1206 &prstatus, sizeof(prstatus));
1207 final_note(buf);
1208}
1209
1210/*
1211 * Move into place and start executing a preloaded standalone
1212 * executable. If nothing was preloaded return an error.
1213 */
1214int kernel_kexec(void)
1215{
1216 int error = 0;
1217
1218 if (!kexec_trylock())
1219 return -EBUSY;
1220 if (!kexec_image) {
1221 error = -EINVAL;
1222 goto Unlock;
1223 }
1224
1225#ifdef CONFIG_KEXEC_JUMP
1226 if (kexec_image->preserve_context) {
1227 pm_prepare_console();
1228 error = freeze_processes();
1229 if (error) {
1230 error = -EBUSY;
1231 goto Restore_console;
1232 }
1233 suspend_console();
1234 error = dpm_suspend_start(PMSG_FREEZE);
1235 if (error)
1236 goto Resume_console;
1237 /* At this point, dpm_suspend_start() has been called,
1238 * but *not* dpm_suspend_end(). We *must* call
1239 * dpm_suspend_end() now. Otherwise, drivers for
1240 * some devices (e.g. interrupt controllers) become
1241 * desynchronized with the actual state of the
1242 * hardware at resume time, and evil weirdness ensues.
1243 */
1244 error = dpm_suspend_end(PMSG_FREEZE);
1245 if (error)
1246 goto Resume_devices;
1247 error = suspend_disable_secondary_cpus();
1248 if (error)
1249 goto Enable_cpus;
1250 local_irq_disable();
1251 error = syscore_suspend();
1252 if (error)
1253 goto Enable_irqs;
1254 } else
1255#endif
1256 {
1257 kexec_in_progress = true;
1258 kernel_restart_prepare("kexec reboot");
1259 migrate_to_reboot_cpu();
1260 syscore_shutdown();
1261
1262 /*
1263 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1264 * no further code needs to use CPU hotplug (which is true in
1265 * the reboot case). However, the kexec path depends on using
1266 * CPU hotplug again; so re-enable it here.
1267 */
1268 cpu_hotplug_enable();
1269 pr_notice("Starting new kernel\n");
1270 machine_shutdown();
1271 }
1272
1273 kmsg_dump(KMSG_DUMP_SHUTDOWN);
1274 machine_kexec(kexec_image);
1275
1276#ifdef CONFIG_KEXEC_JUMP
1277 if (kexec_image->preserve_context) {
1278 syscore_resume();
1279 Enable_irqs:
1280 local_irq_enable();
1281 Enable_cpus:
1282 suspend_enable_secondary_cpus();
1283 dpm_resume_start(PMSG_RESTORE);
1284 Resume_devices:
1285 dpm_resume_end(PMSG_RESTORE);
1286 Resume_console:
1287 resume_console();
1288 thaw_processes();
1289 Restore_console:
1290 pm_restore_console();
1291 }
1292#endif
1293
1294 Unlock:
1295 kexec_unlock();
1296 return error;
1297}