Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kexec: kexec_file_load system call
4 *
5 * Copyright (C) 2014 Red Hat Inc.
6 * Authors:
7 * Vivek Goyal <vgoyal@redhat.com>
8 */
9
10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12#include <linux/capability.h>
13#include <linux/mm.h>
14#include <linux/file.h>
15#include <linux/slab.h>
16#include <linux/kexec.h>
17#include <linux/memblock.h>
18#include <linux/mutex.h>
19#include <linux/list.h>
20#include <linux/fs.h>
21#include <linux/ima.h>
22#include <crypto/hash.h>
23#include <crypto/sha.h>
24#include <linux/elf.h>
25#include <linux/elfcore.h>
26#include <linux/kernel.h>
27#include <linux/syscalls.h>
28#include <linux/vmalloc.h>
29#include "kexec_internal.h"
30
31static int kexec_calculate_store_digests(struct kimage *image);
32
33/*
34 * Currently this is the only default function that is exported as some
35 * architectures need it to do additional handlings.
36 * In the future, other default functions may be exported too if required.
37 */
38int kexec_image_probe_default(struct kimage *image, void *buf,
39 unsigned long buf_len)
40{
41 const struct kexec_file_ops * const *fops;
42 int ret = -ENOEXEC;
43
44 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
45 ret = (*fops)->probe(buf, buf_len);
46 if (!ret) {
47 image->fops = *fops;
48 return ret;
49 }
50 }
51
52 return ret;
53}
54
55/* Architectures can provide this probe function */
56int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
57 unsigned long buf_len)
58{
59 return kexec_image_probe_default(image, buf, buf_len);
60}
61
62static void *kexec_image_load_default(struct kimage *image)
63{
64 if (!image->fops || !image->fops->load)
65 return ERR_PTR(-ENOEXEC);
66
67 return image->fops->load(image, image->kernel_buf,
68 image->kernel_buf_len, image->initrd_buf,
69 image->initrd_buf_len, image->cmdline_buf,
70 image->cmdline_buf_len);
71}
72
73void * __weak arch_kexec_kernel_image_load(struct kimage *image)
74{
75 return kexec_image_load_default(image);
76}
77
78int kexec_image_post_load_cleanup_default(struct kimage *image)
79{
80 if (!image->fops || !image->fops->cleanup)
81 return 0;
82
83 return image->fops->cleanup(image->image_loader_data);
84}
85
86int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
87{
88 return kexec_image_post_load_cleanup_default(image);
89}
90
91#ifdef CONFIG_KEXEC_SIG
92static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
93 unsigned long buf_len)
94{
95 if (!image->fops || !image->fops->verify_sig) {
96 pr_debug("kernel loader does not support signature verification.\n");
97 return -EKEYREJECTED;
98 }
99
100 return image->fops->verify_sig(buf, buf_len);
101}
102
103int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
104 unsigned long buf_len)
105{
106 return kexec_image_verify_sig_default(image, buf, buf_len);
107}
108#endif
109
110/*
111 * arch_kexec_apply_relocations_add - apply relocations of type RELA
112 * @pi: Purgatory to be relocated.
113 * @section: Section relocations applying to.
114 * @relsec: Section containing RELAs.
115 * @symtab: Corresponding symtab.
116 *
117 * Return: 0 on success, negative errno on error.
118 */
119int __weak
120arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section,
121 const Elf_Shdr *relsec, const Elf_Shdr *symtab)
122{
123 pr_err("RELA relocation unsupported.\n");
124 return -ENOEXEC;
125}
126
127/*
128 * arch_kexec_apply_relocations - apply relocations of type REL
129 * @pi: Purgatory to be relocated.
130 * @section: Section relocations applying to.
131 * @relsec: Section containing RELs.
132 * @symtab: Corresponding symtab.
133 *
134 * Return: 0 on success, negative errno on error.
135 */
136int __weak
137arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section,
138 const Elf_Shdr *relsec, const Elf_Shdr *symtab)
139{
140 pr_err("REL relocation unsupported.\n");
141 return -ENOEXEC;
142}
143
144/*
145 * Free up memory used by kernel, initrd, and command line. This is temporary
146 * memory allocation which is not needed any more after these buffers have
147 * been loaded into separate segments and have been copied elsewhere.
148 */
149void kimage_file_post_load_cleanup(struct kimage *image)
150{
151 struct purgatory_info *pi = &image->purgatory_info;
152
153 vfree(image->kernel_buf);
154 image->kernel_buf = NULL;
155
156 vfree(image->initrd_buf);
157 image->initrd_buf = NULL;
158
159 kfree(image->cmdline_buf);
160 image->cmdline_buf = NULL;
161
162 vfree(pi->purgatory_buf);
163 pi->purgatory_buf = NULL;
164
165 vfree(pi->sechdrs);
166 pi->sechdrs = NULL;
167
168 /* See if architecture has anything to cleanup post load */
169 arch_kimage_file_post_load_cleanup(image);
170
171 /*
172 * Above call should have called into bootloader to free up
173 * any data stored in kimage->image_loader_data. It should
174 * be ok now to free it up.
175 */
176 kfree(image->image_loader_data);
177 image->image_loader_data = NULL;
178}
179
180#ifdef CONFIG_KEXEC_SIG
181static int
182kimage_validate_signature(struct kimage *image)
183{
184 const char *reason;
185 int ret;
186
187 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
188 image->kernel_buf_len);
189 switch (ret) {
190 case 0:
191 break;
192
193 /* Certain verification errors are non-fatal if we're not
194 * checking errors, provided we aren't mandating that there
195 * must be a valid signature.
196 */
197 case -ENODATA:
198 reason = "kexec of unsigned image";
199 goto decide;
200 case -ENOPKG:
201 reason = "kexec of image with unsupported crypto";
202 goto decide;
203 case -ENOKEY:
204 reason = "kexec of image with unavailable key";
205 decide:
206 if (IS_ENABLED(CONFIG_KEXEC_SIG_FORCE)) {
207 pr_notice("%s rejected\n", reason);
208 return ret;
209 }
210
211 /* If IMA is guaranteed to appraise a signature on the kexec
212 * image, permit it even if the kernel is otherwise locked
213 * down.
214 */
215 if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
216 security_locked_down(LOCKDOWN_KEXEC))
217 return -EPERM;
218
219 return 0;
220
221 /* All other errors are fatal, including nomem, unparseable
222 * signatures and signature check failures - even if signatures
223 * aren't required.
224 */
225 default:
226 pr_notice("kernel signature verification failed (%d).\n", ret);
227 }
228
229 return ret;
230}
231#endif
232
233/*
234 * In file mode list of segments is prepared by kernel. Copy relevant
235 * data from user space, do error checking, prepare segment list
236 */
237static int
238kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
239 const char __user *cmdline_ptr,
240 unsigned long cmdline_len, unsigned flags)
241{
242 int ret;
243 void *ldata;
244 loff_t size;
245
246 ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
247 &size, INT_MAX, READING_KEXEC_IMAGE);
248 if (ret)
249 return ret;
250 image->kernel_buf_len = size;
251
252 /* Call arch image probe handlers */
253 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
254 image->kernel_buf_len);
255 if (ret)
256 goto out;
257
258#ifdef CONFIG_KEXEC_SIG
259 ret = kimage_validate_signature(image);
260
261 if (ret)
262 goto out;
263#endif
264 /* It is possible that there no initramfs is being loaded */
265 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
266 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
267 &size, INT_MAX,
268 READING_KEXEC_INITRAMFS);
269 if (ret)
270 goto out;
271 image->initrd_buf_len = size;
272 }
273
274 if (cmdline_len) {
275 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
276 if (IS_ERR(image->cmdline_buf)) {
277 ret = PTR_ERR(image->cmdline_buf);
278 image->cmdline_buf = NULL;
279 goto out;
280 }
281
282 image->cmdline_buf_len = cmdline_len;
283
284 /* command line should be a string with last byte null */
285 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
286 ret = -EINVAL;
287 goto out;
288 }
289
290 ima_kexec_cmdline(image->cmdline_buf,
291 image->cmdline_buf_len - 1);
292 }
293
294 /* IMA needs to pass the measurement list to the next kernel. */
295 ima_add_kexec_buffer(image);
296
297 /* Call arch image load handlers */
298 ldata = arch_kexec_kernel_image_load(image);
299
300 if (IS_ERR(ldata)) {
301 ret = PTR_ERR(ldata);
302 goto out;
303 }
304
305 image->image_loader_data = ldata;
306out:
307 /* In case of error, free up all allocated memory in this function */
308 if (ret)
309 kimage_file_post_load_cleanup(image);
310 return ret;
311}
312
313static int
314kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
315 int initrd_fd, const char __user *cmdline_ptr,
316 unsigned long cmdline_len, unsigned long flags)
317{
318 int ret;
319 struct kimage *image;
320 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
321
322 image = do_kimage_alloc_init();
323 if (!image)
324 return -ENOMEM;
325
326 image->file_mode = 1;
327
328 if (kexec_on_panic) {
329 /* Enable special crash kernel control page alloc policy. */
330 image->control_page = crashk_res.start;
331 image->type = KEXEC_TYPE_CRASH;
332 }
333
334 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
335 cmdline_ptr, cmdline_len, flags);
336 if (ret)
337 goto out_free_image;
338
339 ret = sanity_check_segment_list(image);
340 if (ret)
341 goto out_free_post_load_bufs;
342
343 ret = -ENOMEM;
344 image->control_code_page = kimage_alloc_control_pages(image,
345 get_order(KEXEC_CONTROL_PAGE_SIZE));
346 if (!image->control_code_page) {
347 pr_err("Could not allocate control_code_buffer\n");
348 goto out_free_post_load_bufs;
349 }
350
351 if (!kexec_on_panic) {
352 image->swap_page = kimage_alloc_control_pages(image, 0);
353 if (!image->swap_page) {
354 pr_err("Could not allocate swap buffer\n");
355 goto out_free_control_pages;
356 }
357 }
358
359 *rimage = image;
360 return 0;
361out_free_control_pages:
362 kimage_free_page_list(&image->control_pages);
363out_free_post_load_bufs:
364 kimage_file_post_load_cleanup(image);
365out_free_image:
366 kfree(image);
367 return ret;
368}
369
370SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
371 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
372 unsigned long, flags)
373{
374 int ret = 0, i;
375 struct kimage **dest_image, *image;
376
377 /* We only trust the superuser with rebooting the system. */
378 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
379 return -EPERM;
380
381 /* Make sure we have a legal set of flags */
382 if (flags != (flags & KEXEC_FILE_FLAGS))
383 return -EINVAL;
384
385 image = NULL;
386
387 if (!mutex_trylock(&kexec_mutex))
388 return -EBUSY;
389
390 dest_image = &kexec_image;
391 if (flags & KEXEC_FILE_ON_CRASH) {
392 dest_image = &kexec_crash_image;
393 if (kexec_crash_image)
394 arch_kexec_unprotect_crashkres();
395 }
396
397 if (flags & KEXEC_FILE_UNLOAD)
398 goto exchange;
399
400 /*
401 * In case of crash, new kernel gets loaded in reserved region. It is
402 * same memory where old crash kernel might be loaded. Free any
403 * current crash dump kernel before we corrupt it.
404 */
405 if (flags & KEXEC_FILE_ON_CRASH)
406 kimage_free(xchg(&kexec_crash_image, NULL));
407
408 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
409 cmdline_len, flags);
410 if (ret)
411 goto out;
412
413 ret = machine_kexec_prepare(image);
414 if (ret)
415 goto out;
416
417 /*
418 * Some architecture(like S390) may touch the crash memory before
419 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
420 */
421 ret = kimage_crash_copy_vmcoreinfo(image);
422 if (ret)
423 goto out;
424
425 ret = kexec_calculate_store_digests(image);
426 if (ret)
427 goto out;
428
429 for (i = 0; i < image->nr_segments; i++) {
430 struct kexec_segment *ksegment;
431
432 ksegment = &image->segment[i];
433 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
434 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
435 ksegment->memsz);
436
437 ret = kimage_load_segment(image, &image->segment[i]);
438 if (ret)
439 goto out;
440 }
441
442 kimage_terminate(image);
443
444 /*
445 * Free up any temporary buffers allocated which are not needed
446 * after image has been loaded
447 */
448 kimage_file_post_load_cleanup(image);
449exchange:
450 image = xchg(dest_image, image);
451out:
452 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
453 arch_kexec_protect_crashkres();
454
455 mutex_unlock(&kexec_mutex);
456 kimage_free(image);
457 return ret;
458}
459
460static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
461 struct kexec_buf *kbuf)
462{
463 struct kimage *image = kbuf->image;
464 unsigned long temp_start, temp_end;
465
466 temp_end = min(end, kbuf->buf_max);
467 temp_start = temp_end - kbuf->memsz;
468
469 do {
470 /* align down start */
471 temp_start = temp_start & (~(kbuf->buf_align - 1));
472
473 if (temp_start < start || temp_start < kbuf->buf_min)
474 return 0;
475
476 temp_end = temp_start + kbuf->memsz - 1;
477
478 /*
479 * Make sure this does not conflict with any of existing
480 * segments
481 */
482 if (kimage_is_destination_range(image, temp_start, temp_end)) {
483 temp_start = temp_start - PAGE_SIZE;
484 continue;
485 }
486
487 /* We found a suitable memory range */
488 break;
489 } while (1);
490
491 /* If we are here, we found a suitable memory range */
492 kbuf->mem = temp_start;
493
494 /* Success, stop navigating through remaining System RAM ranges */
495 return 1;
496}
497
498static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
499 struct kexec_buf *kbuf)
500{
501 struct kimage *image = kbuf->image;
502 unsigned long temp_start, temp_end;
503
504 temp_start = max(start, kbuf->buf_min);
505
506 do {
507 temp_start = ALIGN(temp_start, kbuf->buf_align);
508 temp_end = temp_start + kbuf->memsz - 1;
509
510 if (temp_end > end || temp_end > kbuf->buf_max)
511 return 0;
512 /*
513 * Make sure this does not conflict with any of existing
514 * segments
515 */
516 if (kimage_is_destination_range(image, temp_start, temp_end)) {
517 temp_start = temp_start + PAGE_SIZE;
518 continue;
519 }
520
521 /* We found a suitable memory range */
522 break;
523 } while (1);
524
525 /* If we are here, we found a suitable memory range */
526 kbuf->mem = temp_start;
527
528 /* Success, stop navigating through remaining System RAM ranges */
529 return 1;
530}
531
532static int locate_mem_hole_callback(struct resource *res, void *arg)
533{
534 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
535 u64 start = res->start, end = res->end;
536 unsigned long sz = end - start + 1;
537
538 /* Returning 0 will take to next memory range */
539 if (sz < kbuf->memsz)
540 return 0;
541
542 if (end < kbuf->buf_min || start > kbuf->buf_max)
543 return 0;
544
545 /*
546 * Allocate memory top down with-in ram range. Otherwise bottom up
547 * allocation.
548 */
549 if (kbuf->top_down)
550 return locate_mem_hole_top_down(start, end, kbuf);
551 return locate_mem_hole_bottom_up(start, end, kbuf);
552}
553
554#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
555static int kexec_walk_memblock(struct kexec_buf *kbuf,
556 int (*func)(struct resource *, void *))
557{
558 int ret = 0;
559 u64 i;
560 phys_addr_t mstart, mend;
561 struct resource res = { };
562
563 if (kbuf->image->type == KEXEC_TYPE_CRASH)
564 return func(&crashk_res, kbuf);
565
566 if (kbuf->top_down) {
567 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
568 &mstart, &mend, NULL) {
569 /*
570 * In memblock, end points to the first byte after the
571 * range while in kexec, end points to the last byte
572 * in the range.
573 */
574 res.start = mstart;
575 res.end = mend - 1;
576 ret = func(&res, kbuf);
577 if (ret)
578 break;
579 }
580 } else {
581 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
582 &mstart, &mend, NULL) {
583 /*
584 * In memblock, end points to the first byte after the
585 * range while in kexec, end points to the last byte
586 * in the range.
587 */
588 res.start = mstart;
589 res.end = mend - 1;
590 ret = func(&res, kbuf);
591 if (ret)
592 break;
593 }
594 }
595
596 return ret;
597}
598#else
599static int kexec_walk_memblock(struct kexec_buf *kbuf,
600 int (*func)(struct resource *, void *))
601{
602 return 0;
603}
604#endif
605
606/**
607 * kexec_walk_resources - call func(data) on free memory regions
608 * @kbuf: Context info for the search. Also passed to @func.
609 * @func: Function to call for each memory region.
610 *
611 * Return: The memory walk will stop when func returns a non-zero value
612 * and that value will be returned. If all free regions are visited without
613 * func returning non-zero, then zero will be returned.
614 */
615static int kexec_walk_resources(struct kexec_buf *kbuf,
616 int (*func)(struct resource *, void *))
617{
618 if (kbuf->image->type == KEXEC_TYPE_CRASH)
619 return walk_iomem_res_desc(crashk_res.desc,
620 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
621 crashk_res.start, crashk_res.end,
622 kbuf, func);
623 else
624 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
625}
626
627/**
628 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
629 * @kbuf: Parameters for the memory search.
630 *
631 * On success, kbuf->mem will have the start address of the memory region found.
632 *
633 * Return: 0 on success, negative errno on error.
634 */
635int kexec_locate_mem_hole(struct kexec_buf *kbuf)
636{
637 int ret;
638
639 /* Arch knows where to place */
640 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
641 return 0;
642
643 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
644 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
645 else
646 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
647
648 return ret == 1 ? 0 : -EADDRNOTAVAIL;
649}
650
651/**
652 * kexec_add_buffer - place a buffer in a kexec segment
653 * @kbuf: Buffer contents and memory parameters.
654 *
655 * This function assumes that kexec_mutex is held.
656 * On successful return, @kbuf->mem will have the physical address of
657 * the buffer in memory.
658 *
659 * Return: 0 on success, negative errno on error.
660 */
661int kexec_add_buffer(struct kexec_buf *kbuf)
662{
663
664 struct kexec_segment *ksegment;
665 int ret;
666
667 /* Currently adding segment this way is allowed only in file mode */
668 if (!kbuf->image->file_mode)
669 return -EINVAL;
670
671 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
672 return -EINVAL;
673
674 /*
675 * Make sure we are not trying to add buffer after allocating
676 * control pages. All segments need to be placed first before
677 * any control pages are allocated. As control page allocation
678 * logic goes through list of segments to make sure there are
679 * no destination overlaps.
680 */
681 if (!list_empty(&kbuf->image->control_pages)) {
682 WARN_ON(1);
683 return -EINVAL;
684 }
685
686 /* Ensure minimum alignment needed for segments. */
687 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
688 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
689
690 /* Walk the RAM ranges and allocate a suitable range for the buffer */
691 ret = kexec_locate_mem_hole(kbuf);
692 if (ret)
693 return ret;
694
695 /* Found a suitable memory range */
696 ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
697 ksegment->kbuf = kbuf->buffer;
698 ksegment->bufsz = kbuf->bufsz;
699 ksegment->mem = kbuf->mem;
700 ksegment->memsz = kbuf->memsz;
701 kbuf->image->nr_segments++;
702 return 0;
703}
704
705/* Calculate and store the digest of segments */
706static int kexec_calculate_store_digests(struct kimage *image)
707{
708 struct crypto_shash *tfm;
709 struct shash_desc *desc;
710 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
711 size_t desc_size, nullsz;
712 char *digest;
713 void *zero_buf;
714 struct kexec_sha_region *sha_regions;
715 struct purgatory_info *pi = &image->purgatory_info;
716
717 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
718 return 0;
719
720 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
721 zero_buf_sz = PAGE_SIZE;
722
723 tfm = crypto_alloc_shash("sha256", 0, 0);
724 if (IS_ERR(tfm)) {
725 ret = PTR_ERR(tfm);
726 goto out;
727 }
728
729 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
730 desc = kzalloc(desc_size, GFP_KERNEL);
731 if (!desc) {
732 ret = -ENOMEM;
733 goto out_free_tfm;
734 }
735
736 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
737 sha_regions = vzalloc(sha_region_sz);
738 if (!sha_regions)
739 goto out_free_desc;
740
741 desc->tfm = tfm;
742
743 ret = crypto_shash_init(desc);
744 if (ret < 0)
745 goto out_free_sha_regions;
746
747 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
748 if (!digest) {
749 ret = -ENOMEM;
750 goto out_free_sha_regions;
751 }
752
753 for (j = i = 0; i < image->nr_segments; i++) {
754 struct kexec_segment *ksegment;
755
756 ksegment = &image->segment[i];
757 /*
758 * Skip purgatory as it will be modified once we put digest
759 * info in purgatory.
760 */
761 if (ksegment->kbuf == pi->purgatory_buf)
762 continue;
763
764 ret = crypto_shash_update(desc, ksegment->kbuf,
765 ksegment->bufsz);
766 if (ret)
767 break;
768
769 /*
770 * Assume rest of the buffer is filled with zero and
771 * update digest accordingly.
772 */
773 nullsz = ksegment->memsz - ksegment->bufsz;
774 while (nullsz) {
775 unsigned long bytes = nullsz;
776
777 if (bytes > zero_buf_sz)
778 bytes = zero_buf_sz;
779 ret = crypto_shash_update(desc, zero_buf, bytes);
780 if (ret)
781 break;
782 nullsz -= bytes;
783 }
784
785 if (ret)
786 break;
787
788 sha_regions[j].start = ksegment->mem;
789 sha_regions[j].len = ksegment->memsz;
790 j++;
791 }
792
793 if (!ret) {
794 ret = crypto_shash_final(desc, digest);
795 if (ret)
796 goto out_free_digest;
797 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
798 sha_regions, sha_region_sz, 0);
799 if (ret)
800 goto out_free_digest;
801
802 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
803 digest, SHA256_DIGEST_SIZE, 0);
804 if (ret)
805 goto out_free_digest;
806 }
807
808out_free_digest:
809 kfree(digest);
810out_free_sha_regions:
811 vfree(sha_regions);
812out_free_desc:
813 kfree(desc);
814out_free_tfm:
815 kfree(tfm);
816out:
817 return ret;
818}
819
820#ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
821/*
822 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
823 * @pi: Purgatory to be loaded.
824 * @kbuf: Buffer to setup.
825 *
826 * Allocates the memory needed for the buffer. Caller is responsible to free
827 * the memory after use.
828 *
829 * Return: 0 on success, negative errno on error.
830 */
831static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
832 struct kexec_buf *kbuf)
833{
834 const Elf_Shdr *sechdrs;
835 unsigned long bss_align;
836 unsigned long bss_sz;
837 unsigned long align;
838 int i, ret;
839
840 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
841 kbuf->buf_align = bss_align = 1;
842 kbuf->bufsz = bss_sz = 0;
843
844 for (i = 0; i < pi->ehdr->e_shnum; i++) {
845 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
846 continue;
847
848 align = sechdrs[i].sh_addralign;
849 if (sechdrs[i].sh_type != SHT_NOBITS) {
850 if (kbuf->buf_align < align)
851 kbuf->buf_align = align;
852 kbuf->bufsz = ALIGN(kbuf->bufsz, align);
853 kbuf->bufsz += sechdrs[i].sh_size;
854 } else {
855 if (bss_align < align)
856 bss_align = align;
857 bss_sz = ALIGN(bss_sz, align);
858 bss_sz += sechdrs[i].sh_size;
859 }
860 }
861 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
862 kbuf->memsz = kbuf->bufsz + bss_sz;
863 if (kbuf->buf_align < bss_align)
864 kbuf->buf_align = bss_align;
865
866 kbuf->buffer = vzalloc(kbuf->bufsz);
867 if (!kbuf->buffer)
868 return -ENOMEM;
869 pi->purgatory_buf = kbuf->buffer;
870
871 ret = kexec_add_buffer(kbuf);
872 if (ret)
873 goto out;
874
875 return 0;
876out:
877 vfree(pi->purgatory_buf);
878 pi->purgatory_buf = NULL;
879 return ret;
880}
881
882/*
883 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
884 * @pi: Purgatory to be loaded.
885 * @kbuf: Buffer prepared to store purgatory.
886 *
887 * Allocates the memory needed for the buffer. Caller is responsible to free
888 * the memory after use.
889 *
890 * Return: 0 on success, negative errno on error.
891 */
892static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
893 struct kexec_buf *kbuf)
894{
895 unsigned long bss_addr;
896 unsigned long offset;
897 Elf_Shdr *sechdrs;
898 int i;
899
900 /*
901 * The section headers in kexec_purgatory are read-only. In order to
902 * have them modifiable make a temporary copy.
903 */
904 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
905 if (!sechdrs)
906 return -ENOMEM;
907 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
908 pi->ehdr->e_shnum * sizeof(Elf_Shdr));
909 pi->sechdrs = sechdrs;
910
911 offset = 0;
912 bss_addr = kbuf->mem + kbuf->bufsz;
913 kbuf->image->start = pi->ehdr->e_entry;
914
915 for (i = 0; i < pi->ehdr->e_shnum; i++) {
916 unsigned long align;
917 void *src, *dst;
918
919 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
920 continue;
921
922 align = sechdrs[i].sh_addralign;
923 if (sechdrs[i].sh_type == SHT_NOBITS) {
924 bss_addr = ALIGN(bss_addr, align);
925 sechdrs[i].sh_addr = bss_addr;
926 bss_addr += sechdrs[i].sh_size;
927 continue;
928 }
929
930 offset = ALIGN(offset, align);
931 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
932 pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
933 pi->ehdr->e_entry < (sechdrs[i].sh_addr
934 + sechdrs[i].sh_size)) {
935 kbuf->image->start -= sechdrs[i].sh_addr;
936 kbuf->image->start += kbuf->mem + offset;
937 }
938
939 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
940 dst = pi->purgatory_buf + offset;
941 memcpy(dst, src, sechdrs[i].sh_size);
942
943 sechdrs[i].sh_addr = kbuf->mem + offset;
944 sechdrs[i].sh_offset = offset;
945 offset += sechdrs[i].sh_size;
946 }
947
948 return 0;
949}
950
951static int kexec_apply_relocations(struct kimage *image)
952{
953 int i, ret;
954 struct purgatory_info *pi = &image->purgatory_info;
955 const Elf_Shdr *sechdrs;
956
957 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
958
959 for (i = 0; i < pi->ehdr->e_shnum; i++) {
960 const Elf_Shdr *relsec;
961 const Elf_Shdr *symtab;
962 Elf_Shdr *section;
963
964 relsec = sechdrs + i;
965
966 if (relsec->sh_type != SHT_RELA &&
967 relsec->sh_type != SHT_REL)
968 continue;
969
970 /*
971 * For section of type SHT_RELA/SHT_REL,
972 * ->sh_link contains section header index of associated
973 * symbol table. And ->sh_info contains section header
974 * index of section to which relocations apply.
975 */
976 if (relsec->sh_info >= pi->ehdr->e_shnum ||
977 relsec->sh_link >= pi->ehdr->e_shnum)
978 return -ENOEXEC;
979
980 section = pi->sechdrs + relsec->sh_info;
981 symtab = sechdrs + relsec->sh_link;
982
983 if (!(section->sh_flags & SHF_ALLOC))
984 continue;
985
986 /*
987 * symtab->sh_link contain section header index of associated
988 * string table.
989 */
990 if (symtab->sh_link >= pi->ehdr->e_shnum)
991 /* Invalid section number? */
992 continue;
993
994 /*
995 * Respective architecture needs to provide support for applying
996 * relocations of type SHT_RELA/SHT_REL.
997 */
998 if (relsec->sh_type == SHT_RELA)
999 ret = arch_kexec_apply_relocations_add(pi, section,
1000 relsec, symtab);
1001 else if (relsec->sh_type == SHT_REL)
1002 ret = arch_kexec_apply_relocations(pi, section,
1003 relsec, symtab);
1004 if (ret)
1005 return ret;
1006 }
1007
1008 return 0;
1009}
1010
1011/*
1012 * kexec_load_purgatory - Load and relocate the purgatory object.
1013 * @image: Image to add the purgatory to.
1014 * @kbuf: Memory parameters to use.
1015 *
1016 * Allocates the memory needed for image->purgatory_info.sechdrs and
1017 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1018 * to free the memory after use.
1019 *
1020 * Return: 0 on success, negative errno on error.
1021 */
1022int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1023{
1024 struct purgatory_info *pi = &image->purgatory_info;
1025 int ret;
1026
1027 if (kexec_purgatory_size <= 0)
1028 return -EINVAL;
1029
1030 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1031
1032 ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1033 if (ret)
1034 return ret;
1035
1036 ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1037 if (ret)
1038 goto out_free_kbuf;
1039
1040 ret = kexec_apply_relocations(image);
1041 if (ret)
1042 goto out;
1043
1044 return 0;
1045out:
1046 vfree(pi->sechdrs);
1047 pi->sechdrs = NULL;
1048out_free_kbuf:
1049 vfree(pi->purgatory_buf);
1050 pi->purgatory_buf = NULL;
1051 return ret;
1052}
1053
1054/*
1055 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1056 * @pi: Purgatory to search in.
1057 * @name: Name of the symbol.
1058 *
1059 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1060 */
1061static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1062 const char *name)
1063{
1064 const Elf_Shdr *sechdrs;
1065 const Elf_Ehdr *ehdr;
1066 const Elf_Sym *syms;
1067 const char *strtab;
1068 int i, k;
1069
1070 if (!pi->ehdr)
1071 return NULL;
1072
1073 ehdr = pi->ehdr;
1074 sechdrs = (void *)ehdr + ehdr->e_shoff;
1075
1076 for (i = 0; i < ehdr->e_shnum; i++) {
1077 if (sechdrs[i].sh_type != SHT_SYMTAB)
1078 continue;
1079
1080 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1081 /* Invalid strtab section number */
1082 continue;
1083 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1084 syms = (void *)ehdr + sechdrs[i].sh_offset;
1085
1086 /* Go through symbols for a match */
1087 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1088 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1089 continue;
1090
1091 if (strcmp(strtab + syms[k].st_name, name) != 0)
1092 continue;
1093
1094 if (syms[k].st_shndx == SHN_UNDEF ||
1095 syms[k].st_shndx >= ehdr->e_shnum) {
1096 pr_debug("Symbol: %s has bad section index %d.\n",
1097 name, syms[k].st_shndx);
1098 return NULL;
1099 }
1100
1101 /* Found the symbol we are looking for */
1102 return &syms[k];
1103 }
1104 }
1105
1106 return NULL;
1107}
1108
1109void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1110{
1111 struct purgatory_info *pi = &image->purgatory_info;
1112 const Elf_Sym *sym;
1113 Elf_Shdr *sechdr;
1114
1115 sym = kexec_purgatory_find_symbol(pi, name);
1116 if (!sym)
1117 return ERR_PTR(-EINVAL);
1118
1119 sechdr = &pi->sechdrs[sym->st_shndx];
1120
1121 /*
1122 * Returns the address where symbol will finally be loaded after
1123 * kexec_load_segment()
1124 */
1125 return (void *)(sechdr->sh_addr + sym->st_value);
1126}
1127
1128/*
1129 * Get or set value of a symbol. If "get_value" is true, symbol value is
1130 * returned in buf otherwise symbol value is set based on value in buf.
1131 */
1132int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1133 void *buf, unsigned int size, bool get_value)
1134{
1135 struct purgatory_info *pi = &image->purgatory_info;
1136 const Elf_Sym *sym;
1137 Elf_Shdr *sec;
1138 char *sym_buf;
1139
1140 sym = kexec_purgatory_find_symbol(pi, name);
1141 if (!sym)
1142 return -EINVAL;
1143
1144 if (sym->st_size != size) {
1145 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1146 name, (unsigned long)sym->st_size, size);
1147 return -EINVAL;
1148 }
1149
1150 sec = pi->sechdrs + sym->st_shndx;
1151
1152 if (sec->sh_type == SHT_NOBITS) {
1153 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1154 get_value ? "get" : "set");
1155 return -EINVAL;
1156 }
1157
1158 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1159
1160 if (get_value)
1161 memcpy((void *)buf, sym_buf, size);
1162 else
1163 memcpy((void *)sym_buf, buf, size);
1164
1165 return 0;
1166}
1167#endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1168
1169int crash_exclude_mem_range(struct crash_mem *mem,
1170 unsigned long long mstart, unsigned long long mend)
1171{
1172 int i, j;
1173 unsigned long long start, end;
1174 struct crash_mem_range temp_range = {0, 0};
1175
1176 for (i = 0; i < mem->nr_ranges; i++) {
1177 start = mem->ranges[i].start;
1178 end = mem->ranges[i].end;
1179
1180 if (mstart > end || mend < start)
1181 continue;
1182
1183 /* Truncate any area outside of range */
1184 if (mstart < start)
1185 mstart = start;
1186 if (mend > end)
1187 mend = end;
1188
1189 /* Found completely overlapping range */
1190 if (mstart == start && mend == end) {
1191 mem->ranges[i].start = 0;
1192 mem->ranges[i].end = 0;
1193 if (i < mem->nr_ranges - 1) {
1194 /* Shift rest of the ranges to left */
1195 for (j = i; j < mem->nr_ranges - 1; j++) {
1196 mem->ranges[j].start =
1197 mem->ranges[j+1].start;
1198 mem->ranges[j].end =
1199 mem->ranges[j+1].end;
1200 }
1201 }
1202 mem->nr_ranges--;
1203 return 0;
1204 }
1205
1206 if (mstart > start && mend < end) {
1207 /* Split original range */
1208 mem->ranges[i].end = mstart - 1;
1209 temp_range.start = mend + 1;
1210 temp_range.end = end;
1211 } else if (mstart != start)
1212 mem->ranges[i].end = mstart - 1;
1213 else
1214 mem->ranges[i].start = mend + 1;
1215 break;
1216 }
1217
1218 /* If a split happened, add the split to array */
1219 if (!temp_range.end)
1220 return 0;
1221
1222 /* Split happened */
1223 if (i == mem->max_nr_ranges - 1)
1224 return -ENOMEM;
1225
1226 /* Location where new range should go */
1227 j = i + 1;
1228 if (j < mem->nr_ranges) {
1229 /* Move over all ranges one slot towards the end */
1230 for (i = mem->nr_ranges - 1; i >= j; i--)
1231 mem->ranges[i + 1] = mem->ranges[i];
1232 }
1233
1234 mem->ranges[j].start = temp_range.start;
1235 mem->ranges[j].end = temp_range.end;
1236 mem->nr_ranges++;
1237 return 0;
1238}
1239
1240int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1241 void **addr, unsigned long *sz)
1242{
1243 Elf64_Ehdr *ehdr;
1244 Elf64_Phdr *phdr;
1245 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1246 unsigned char *buf;
1247 unsigned int cpu, i;
1248 unsigned long long notes_addr;
1249 unsigned long mstart, mend;
1250
1251 /* extra phdr for vmcoreinfo elf note */
1252 nr_phdr = nr_cpus + 1;
1253 nr_phdr += mem->nr_ranges;
1254
1255 /*
1256 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1257 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1258 * I think this is required by tools like gdb. So same physical
1259 * memory will be mapped in two elf headers. One will contain kernel
1260 * text virtual addresses and other will have __va(physical) addresses.
1261 */
1262
1263 nr_phdr++;
1264 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1265 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1266
1267 buf = vzalloc(elf_sz);
1268 if (!buf)
1269 return -ENOMEM;
1270
1271 ehdr = (Elf64_Ehdr *)buf;
1272 phdr = (Elf64_Phdr *)(ehdr + 1);
1273 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1274 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1275 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1276 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1277 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1278 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1279 ehdr->e_type = ET_CORE;
1280 ehdr->e_machine = ELF_ARCH;
1281 ehdr->e_version = EV_CURRENT;
1282 ehdr->e_phoff = sizeof(Elf64_Ehdr);
1283 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1284 ehdr->e_phentsize = sizeof(Elf64_Phdr);
1285
1286 /* Prepare one phdr of type PT_NOTE for each present cpu */
1287 for_each_present_cpu(cpu) {
1288 phdr->p_type = PT_NOTE;
1289 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1290 phdr->p_offset = phdr->p_paddr = notes_addr;
1291 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1292 (ehdr->e_phnum)++;
1293 phdr++;
1294 }
1295
1296 /* Prepare one PT_NOTE header for vmcoreinfo */
1297 phdr->p_type = PT_NOTE;
1298 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1299 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1300 (ehdr->e_phnum)++;
1301 phdr++;
1302
1303 /* Prepare PT_LOAD type program header for kernel text region */
1304 if (kernel_map) {
1305 phdr->p_type = PT_LOAD;
1306 phdr->p_flags = PF_R|PF_W|PF_X;
1307 phdr->p_vaddr = (Elf64_Addr)_text;
1308 phdr->p_filesz = phdr->p_memsz = _end - _text;
1309 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1310 ehdr->e_phnum++;
1311 phdr++;
1312 }
1313
1314 /* Go through all the ranges in mem->ranges[] and prepare phdr */
1315 for (i = 0; i < mem->nr_ranges; i++) {
1316 mstart = mem->ranges[i].start;
1317 mend = mem->ranges[i].end;
1318
1319 phdr->p_type = PT_LOAD;
1320 phdr->p_flags = PF_R|PF_W|PF_X;
1321 phdr->p_offset = mstart;
1322
1323 phdr->p_paddr = mstart;
1324 phdr->p_vaddr = (unsigned long long) __va(mstart);
1325 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1326 phdr->p_align = 0;
1327 ehdr->e_phnum++;
1328 phdr++;
1329 pr_debug("Crash PT_LOAD elf header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n",
1330 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1331 ehdr->e_phnum, phdr->p_offset);
1332 }
1333
1334 *addr = buf;
1335 *sz = elf_sz;
1336 return 0;
1337}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kexec: kexec_file_load system call
4 *
5 * Copyright (C) 2014 Red Hat Inc.
6 * Authors:
7 * Vivek Goyal <vgoyal@redhat.com>
8 */
9
10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12#include <linux/capability.h>
13#include <linux/mm.h>
14#include <linux/file.h>
15#include <linux/slab.h>
16#include <linux/kexec.h>
17#include <linux/memblock.h>
18#include <linux/mutex.h>
19#include <linux/list.h>
20#include <linux/fs.h>
21#include <linux/ima.h>
22#include <crypto/hash.h>
23#include <crypto/sha2.h>
24#include <linux/elf.h>
25#include <linux/elfcore.h>
26#include <linux/kernel.h>
27#include <linux/kernel_read_file.h>
28#include <linux/syscalls.h>
29#include <linux/vmalloc.h>
30#include "kexec_internal.h"
31
32#ifdef CONFIG_KEXEC_SIG
33static bool sig_enforce = IS_ENABLED(CONFIG_KEXEC_SIG_FORCE);
34
35void set_kexec_sig_enforced(void)
36{
37 sig_enforce = true;
38}
39#endif
40
41static int kexec_calculate_store_digests(struct kimage *image);
42
43/* Maximum size in bytes for kernel/initrd files. */
44#define KEXEC_FILE_SIZE_MAX min_t(s64, 4LL << 30, SSIZE_MAX)
45
46/*
47 * Currently this is the only default function that is exported as some
48 * architectures need it to do additional handlings.
49 * In the future, other default functions may be exported too if required.
50 */
51int kexec_image_probe_default(struct kimage *image, void *buf,
52 unsigned long buf_len)
53{
54 const struct kexec_file_ops * const *fops;
55 int ret = -ENOEXEC;
56
57 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
58 ret = (*fops)->probe(buf, buf_len);
59 if (!ret) {
60 image->fops = *fops;
61 return ret;
62 }
63 }
64
65 return ret;
66}
67
68void *kexec_image_load_default(struct kimage *image)
69{
70 if (!image->fops || !image->fops->load)
71 return ERR_PTR(-ENOEXEC);
72
73 return image->fops->load(image, image->kernel_buf,
74 image->kernel_buf_len, image->initrd_buf,
75 image->initrd_buf_len, image->cmdline_buf,
76 image->cmdline_buf_len);
77}
78
79int kexec_image_post_load_cleanup_default(struct kimage *image)
80{
81 if (!image->fops || !image->fops->cleanup)
82 return 0;
83
84 return image->fops->cleanup(image->image_loader_data);
85}
86
87/*
88 * Free up memory used by kernel, initrd, and command line. This is temporary
89 * memory allocation which is not needed any more after these buffers have
90 * been loaded into separate segments and have been copied elsewhere.
91 */
92void kimage_file_post_load_cleanup(struct kimage *image)
93{
94 struct purgatory_info *pi = &image->purgatory_info;
95
96 vfree(image->kernel_buf);
97 image->kernel_buf = NULL;
98
99 vfree(image->initrd_buf);
100 image->initrd_buf = NULL;
101
102 kfree(image->cmdline_buf);
103 image->cmdline_buf = NULL;
104
105 vfree(pi->purgatory_buf);
106 pi->purgatory_buf = NULL;
107
108 vfree(pi->sechdrs);
109 pi->sechdrs = NULL;
110
111#ifdef CONFIG_IMA_KEXEC
112 vfree(image->ima_buffer);
113 image->ima_buffer = NULL;
114#endif /* CONFIG_IMA_KEXEC */
115
116 /* See if architecture has anything to cleanup post load */
117 arch_kimage_file_post_load_cleanup(image);
118
119 /*
120 * Above call should have called into bootloader to free up
121 * any data stored in kimage->image_loader_data. It should
122 * be ok now to free it up.
123 */
124 kfree(image->image_loader_data);
125 image->image_loader_data = NULL;
126}
127
128#ifdef CONFIG_KEXEC_SIG
129#ifdef CONFIG_SIGNED_PE_FILE_VERIFICATION
130int kexec_kernel_verify_pe_sig(const char *kernel, unsigned long kernel_len)
131{
132 int ret;
133
134 ret = verify_pefile_signature(kernel, kernel_len,
135 VERIFY_USE_SECONDARY_KEYRING,
136 VERIFYING_KEXEC_PE_SIGNATURE);
137 if (ret == -ENOKEY && IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING)) {
138 ret = verify_pefile_signature(kernel, kernel_len,
139 VERIFY_USE_PLATFORM_KEYRING,
140 VERIFYING_KEXEC_PE_SIGNATURE);
141 }
142 return ret;
143}
144#endif
145
146static int kexec_image_verify_sig(struct kimage *image, void *buf,
147 unsigned long buf_len)
148{
149 if (!image->fops || !image->fops->verify_sig) {
150 pr_debug("kernel loader does not support signature verification.\n");
151 return -EKEYREJECTED;
152 }
153
154 return image->fops->verify_sig(buf, buf_len);
155}
156
157static int
158kimage_validate_signature(struct kimage *image)
159{
160 int ret;
161
162 ret = kexec_image_verify_sig(image, image->kernel_buf,
163 image->kernel_buf_len);
164 if (ret) {
165
166 if (sig_enforce) {
167 pr_notice("Enforced kernel signature verification failed (%d).\n", ret);
168 return ret;
169 }
170
171 /*
172 * If IMA is guaranteed to appraise a signature on the kexec
173 * image, permit it even if the kernel is otherwise locked
174 * down.
175 */
176 if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
177 security_locked_down(LOCKDOWN_KEXEC))
178 return -EPERM;
179
180 pr_debug("kernel signature verification failed (%d).\n", ret);
181 }
182
183 return 0;
184}
185#endif
186
187/*
188 * In file mode list of segments is prepared by kernel. Copy relevant
189 * data from user space, do error checking, prepare segment list
190 */
191static int
192kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
193 const char __user *cmdline_ptr,
194 unsigned long cmdline_len, unsigned flags)
195{
196 ssize_t ret;
197 void *ldata;
198
199 ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf,
200 KEXEC_FILE_SIZE_MAX, NULL,
201 READING_KEXEC_IMAGE);
202 if (ret < 0)
203 return ret;
204 image->kernel_buf_len = ret;
205
206 /* Call arch image probe handlers */
207 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
208 image->kernel_buf_len);
209 if (ret)
210 goto out;
211
212#ifdef CONFIG_KEXEC_SIG
213 ret = kimage_validate_signature(image);
214
215 if (ret)
216 goto out;
217#endif
218 /* It is possible that there no initramfs is being loaded */
219 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
220 ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf,
221 KEXEC_FILE_SIZE_MAX, NULL,
222 READING_KEXEC_INITRAMFS);
223 if (ret < 0)
224 goto out;
225 image->initrd_buf_len = ret;
226 ret = 0;
227 }
228
229 if (cmdline_len) {
230 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
231 if (IS_ERR(image->cmdline_buf)) {
232 ret = PTR_ERR(image->cmdline_buf);
233 image->cmdline_buf = NULL;
234 goto out;
235 }
236
237 image->cmdline_buf_len = cmdline_len;
238
239 /* command line should be a string with last byte null */
240 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
241 ret = -EINVAL;
242 goto out;
243 }
244
245 ima_kexec_cmdline(kernel_fd, image->cmdline_buf,
246 image->cmdline_buf_len - 1);
247 }
248
249 /* IMA needs to pass the measurement list to the next kernel. */
250 ima_add_kexec_buffer(image);
251
252 /* Call arch image load handlers */
253 ldata = arch_kexec_kernel_image_load(image);
254
255 if (IS_ERR(ldata)) {
256 ret = PTR_ERR(ldata);
257 goto out;
258 }
259
260 image->image_loader_data = ldata;
261out:
262 /* In case of error, free up all allocated memory in this function */
263 if (ret)
264 kimage_file_post_load_cleanup(image);
265 return ret;
266}
267
268static int
269kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
270 int initrd_fd, const char __user *cmdline_ptr,
271 unsigned long cmdline_len, unsigned long flags)
272{
273 int ret;
274 struct kimage *image;
275 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
276
277 image = do_kimage_alloc_init();
278 if (!image)
279 return -ENOMEM;
280
281 image->file_mode = 1;
282
283 if (kexec_on_panic) {
284 /* Enable special crash kernel control page alloc policy. */
285 image->control_page = crashk_res.start;
286 image->type = KEXEC_TYPE_CRASH;
287 }
288
289 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
290 cmdline_ptr, cmdline_len, flags);
291 if (ret)
292 goto out_free_image;
293
294 ret = sanity_check_segment_list(image);
295 if (ret)
296 goto out_free_post_load_bufs;
297
298 ret = -ENOMEM;
299 image->control_code_page = kimage_alloc_control_pages(image,
300 get_order(KEXEC_CONTROL_PAGE_SIZE));
301 if (!image->control_code_page) {
302 pr_err("Could not allocate control_code_buffer\n");
303 goto out_free_post_load_bufs;
304 }
305
306 if (!kexec_on_panic) {
307 image->swap_page = kimage_alloc_control_pages(image, 0);
308 if (!image->swap_page) {
309 pr_err("Could not allocate swap buffer\n");
310 goto out_free_control_pages;
311 }
312 }
313
314 *rimage = image;
315 return 0;
316out_free_control_pages:
317 kimage_free_page_list(&image->control_pages);
318out_free_post_load_bufs:
319 kimage_file_post_load_cleanup(image);
320out_free_image:
321 kfree(image);
322 return ret;
323}
324
325SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
326 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
327 unsigned long, flags)
328{
329 int ret = 0, i;
330 struct kimage **dest_image, *image;
331
332 /* We only trust the superuser with rebooting the system. */
333 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
334 return -EPERM;
335
336 /* Make sure we have a legal set of flags */
337 if (flags != (flags & KEXEC_FILE_FLAGS))
338 return -EINVAL;
339
340 image = NULL;
341
342 if (!kexec_trylock())
343 return -EBUSY;
344
345 dest_image = &kexec_image;
346 if (flags & KEXEC_FILE_ON_CRASH) {
347 dest_image = &kexec_crash_image;
348 if (kexec_crash_image)
349 arch_kexec_unprotect_crashkres();
350 }
351
352 if (flags & KEXEC_FILE_UNLOAD)
353 goto exchange;
354
355 /*
356 * In case of crash, new kernel gets loaded in reserved region. It is
357 * same memory where old crash kernel might be loaded. Free any
358 * current crash dump kernel before we corrupt it.
359 */
360 if (flags & KEXEC_FILE_ON_CRASH)
361 kimage_free(xchg(&kexec_crash_image, NULL));
362
363 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
364 cmdline_len, flags);
365 if (ret)
366 goto out;
367
368 ret = machine_kexec_prepare(image);
369 if (ret)
370 goto out;
371
372 /*
373 * Some architecture(like S390) may touch the crash memory before
374 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
375 */
376 ret = kimage_crash_copy_vmcoreinfo(image);
377 if (ret)
378 goto out;
379
380 ret = kexec_calculate_store_digests(image);
381 if (ret)
382 goto out;
383
384 for (i = 0; i < image->nr_segments; i++) {
385 struct kexec_segment *ksegment;
386
387 ksegment = &image->segment[i];
388 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
389 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
390 ksegment->memsz);
391
392 ret = kimage_load_segment(image, &image->segment[i]);
393 if (ret)
394 goto out;
395 }
396
397 kimage_terminate(image);
398
399 ret = machine_kexec_post_load(image);
400 if (ret)
401 goto out;
402
403 /*
404 * Free up any temporary buffers allocated which are not needed
405 * after image has been loaded
406 */
407 kimage_file_post_load_cleanup(image);
408exchange:
409 image = xchg(dest_image, image);
410out:
411 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
412 arch_kexec_protect_crashkres();
413
414 kexec_unlock();
415 kimage_free(image);
416 return ret;
417}
418
419static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
420 struct kexec_buf *kbuf)
421{
422 struct kimage *image = kbuf->image;
423 unsigned long temp_start, temp_end;
424
425 temp_end = min(end, kbuf->buf_max);
426 temp_start = temp_end - kbuf->memsz;
427
428 do {
429 /* align down start */
430 temp_start = temp_start & (~(kbuf->buf_align - 1));
431
432 if (temp_start < start || temp_start < kbuf->buf_min)
433 return 0;
434
435 temp_end = temp_start + kbuf->memsz - 1;
436
437 /*
438 * Make sure this does not conflict with any of existing
439 * segments
440 */
441 if (kimage_is_destination_range(image, temp_start, temp_end)) {
442 temp_start = temp_start - PAGE_SIZE;
443 continue;
444 }
445
446 /* We found a suitable memory range */
447 break;
448 } while (1);
449
450 /* If we are here, we found a suitable memory range */
451 kbuf->mem = temp_start;
452
453 /* Success, stop navigating through remaining System RAM ranges */
454 return 1;
455}
456
457static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
458 struct kexec_buf *kbuf)
459{
460 struct kimage *image = kbuf->image;
461 unsigned long temp_start, temp_end;
462
463 temp_start = max(start, kbuf->buf_min);
464
465 do {
466 temp_start = ALIGN(temp_start, kbuf->buf_align);
467 temp_end = temp_start + kbuf->memsz - 1;
468
469 if (temp_end > end || temp_end > kbuf->buf_max)
470 return 0;
471 /*
472 * Make sure this does not conflict with any of existing
473 * segments
474 */
475 if (kimage_is_destination_range(image, temp_start, temp_end)) {
476 temp_start = temp_start + PAGE_SIZE;
477 continue;
478 }
479
480 /* We found a suitable memory range */
481 break;
482 } while (1);
483
484 /* If we are here, we found a suitable memory range */
485 kbuf->mem = temp_start;
486
487 /* Success, stop navigating through remaining System RAM ranges */
488 return 1;
489}
490
491static int locate_mem_hole_callback(struct resource *res, void *arg)
492{
493 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
494 u64 start = res->start, end = res->end;
495 unsigned long sz = end - start + 1;
496
497 /* Returning 0 will take to next memory range */
498
499 /* Don't use memory that will be detected and handled by a driver. */
500 if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
501 return 0;
502
503 if (sz < kbuf->memsz)
504 return 0;
505
506 if (end < kbuf->buf_min || start > kbuf->buf_max)
507 return 0;
508
509 /*
510 * Allocate memory top down with-in ram range. Otherwise bottom up
511 * allocation.
512 */
513 if (kbuf->top_down)
514 return locate_mem_hole_top_down(start, end, kbuf);
515 return locate_mem_hole_bottom_up(start, end, kbuf);
516}
517
518#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
519static int kexec_walk_memblock(struct kexec_buf *kbuf,
520 int (*func)(struct resource *, void *))
521{
522 int ret = 0;
523 u64 i;
524 phys_addr_t mstart, mend;
525 struct resource res = { };
526
527 if (kbuf->image->type == KEXEC_TYPE_CRASH)
528 return func(&crashk_res, kbuf);
529
530 /*
531 * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
532 * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
533 * locate_mem_hole_callback().
534 */
535 if (kbuf->top_down) {
536 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
537 &mstart, &mend, NULL) {
538 /*
539 * In memblock, end points to the first byte after the
540 * range while in kexec, end points to the last byte
541 * in the range.
542 */
543 res.start = mstart;
544 res.end = mend - 1;
545 ret = func(&res, kbuf);
546 if (ret)
547 break;
548 }
549 } else {
550 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
551 &mstart, &mend, NULL) {
552 /*
553 * In memblock, end points to the first byte after the
554 * range while in kexec, end points to the last byte
555 * in the range.
556 */
557 res.start = mstart;
558 res.end = mend - 1;
559 ret = func(&res, kbuf);
560 if (ret)
561 break;
562 }
563 }
564
565 return ret;
566}
567#else
568static int kexec_walk_memblock(struct kexec_buf *kbuf,
569 int (*func)(struct resource *, void *))
570{
571 return 0;
572}
573#endif
574
575/**
576 * kexec_walk_resources - call func(data) on free memory regions
577 * @kbuf: Context info for the search. Also passed to @func.
578 * @func: Function to call for each memory region.
579 *
580 * Return: The memory walk will stop when func returns a non-zero value
581 * and that value will be returned. If all free regions are visited without
582 * func returning non-zero, then zero will be returned.
583 */
584static int kexec_walk_resources(struct kexec_buf *kbuf,
585 int (*func)(struct resource *, void *))
586{
587 if (kbuf->image->type == KEXEC_TYPE_CRASH)
588 return walk_iomem_res_desc(crashk_res.desc,
589 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
590 crashk_res.start, crashk_res.end,
591 kbuf, func);
592 else
593 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
594}
595
596/**
597 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
598 * @kbuf: Parameters for the memory search.
599 *
600 * On success, kbuf->mem will have the start address of the memory region found.
601 *
602 * Return: 0 on success, negative errno on error.
603 */
604int kexec_locate_mem_hole(struct kexec_buf *kbuf)
605{
606 int ret;
607
608 /* Arch knows where to place */
609 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
610 return 0;
611
612 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
613 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
614 else
615 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
616
617 return ret == 1 ? 0 : -EADDRNOTAVAIL;
618}
619
620/**
621 * kexec_add_buffer - place a buffer in a kexec segment
622 * @kbuf: Buffer contents and memory parameters.
623 *
624 * This function assumes that kexec_mutex is held.
625 * On successful return, @kbuf->mem will have the physical address of
626 * the buffer in memory.
627 *
628 * Return: 0 on success, negative errno on error.
629 */
630int kexec_add_buffer(struct kexec_buf *kbuf)
631{
632 struct kexec_segment *ksegment;
633 int ret;
634
635 /* Currently adding segment this way is allowed only in file mode */
636 if (!kbuf->image->file_mode)
637 return -EINVAL;
638
639 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
640 return -EINVAL;
641
642 /*
643 * Make sure we are not trying to add buffer after allocating
644 * control pages. All segments need to be placed first before
645 * any control pages are allocated. As control page allocation
646 * logic goes through list of segments to make sure there are
647 * no destination overlaps.
648 */
649 if (!list_empty(&kbuf->image->control_pages)) {
650 WARN_ON(1);
651 return -EINVAL;
652 }
653
654 /* Ensure minimum alignment needed for segments. */
655 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
656 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
657
658 /* Walk the RAM ranges and allocate a suitable range for the buffer */
659 ret = arch_kexec_locate_mem_hole(kbuf);
660 if (ret)
661 return ret;
662
663 /* Found a suitable memory range */
664 ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
665 ksegment->kbuf = kbuf->buffer;
666 ksegment->bufsz = kbuf->bufsz;
667 ksegment->mem = kbuf->mem;
668 ksegment->memsz = kbuf->memsz;
669 kbuf->image->nr_segments++;
670 return 0;
671}
672
673/* Calculate and store the digest of segments */
674static int kexec_calculate_store_digests(struct kimage *image)
675{
676 struct crypto_shash *tfm;
677 struct shash_desc *desc;
678 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
679 size_t desc_size, nullsz;
680 char *digest;
681 void *zero_buf;
682 struct kexec_sha_region *sha_regions;
683 struct purgatory_info *pi = &image->purgatory_info;
684
685 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
686 return 0;
687
688 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
689 zero_buf_sz = PAGE_SIZE;
690
691 tfm = crypto_alloc_shash("sha256", 0, 0);
692 if (IS_ERR(tfm)) {
693 ret = PTR_ERR(tfm);
694 goto out;
695 }
696
697 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
698 desc = kzalloc(desc_size, GFP_KERNEL);
699 if (!desc) {
700 ret = -ENOMEM;
701 goto out_free_tfm;
702 }
703
704 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
705 sha_regions = vzalloc(sha_region_sz);
706 if (!sha_regions) {
707 ret = -ENOMEM;
708 goto out_free_desc;
709 }
710
711 desc->tfm = tfm;
712
713 ret = crypto_shash_init(desc);
714 if (ret < 0)
715 goto out_free_sha_regions;
716
717 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
718 if (!digest) {
719 ret = -ENOMEM;
720 goto out_free_sha_regions;
721 }
722
723 for (j = i = 0; i < image->nr_segments; i++) {
724 struct kexec_segment *ksegment;
725
726 ksegment = &image->segment[i];
727 /*
728 * Skip purgatory as it will be modified once we put digest
729 * info in purgatory.
730 */
731 if (ksegment->kbuf == pi->purgatory_buf)
732 continue;
733
734 ret = crypto_shash_update(desc, ksegment->kbuf,
735 ksegment->bufsz);
736 if (ret)
737 break;
738
739 /*
740 * Assume rest of the buffer is filled with zero and
741 * update digest accordingly.
742 */
743 nullsz = ksegment->memsz - ksegment->bufsz;
744 while (nullsz) {
745 unsigned long bytes = nullsz;
746
747 if (bytes > zero_buf_sz)
748 bytes = zero_buf_sz;
749 ret = crypto_shash_update(desc, zero_buf, bytes);
750 if (ret)
751 break;
752 nullsz -= bytes;
753 }
754
755 if (ret)
756 break;
757
758 sha_regions[j].start = ksegment->mem;
759 sha_regions[j].len = ksegment->memsz;
760 j++;
761 }
762
763 if (!ret) {
764 ret = crypto_shash_final(desc, digest);
765 if (ret)
766 goto out_free_digest;
767 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
768 sha_regions, sha_region_sz, 0);
769 if (ret)
770 goto out_free_digest;
771
772 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
773 digest, SHA256_DIGEST_SIZE, 0);
774 if (ret)
775 goto out_free_digest;
776 }
777
778out_free_digest:
779 kfree(digest);
780out_free_sha_regions:
781 vfree(sha_regions);
782out_free_desc:
783 kfree(desc);
784out_free_tfm:
785 kfree(tfm);
786out:
787 return ret;
788}
789
790#ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
791/*
792 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
793 * @pi: Purgatory to be loaded.
794 * @kbuf: Buffer to setup.
795 *
796 * Allocates the memory needed for the buffer. Caller is responsible to free
797 * the memory after use.
798 *
799 * Return: 0 on success, negative errno on error.
800 */
801static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
802 struct kexec_buf *kbuf)
803{
804 const Elf_Shdr *sechdrs;
805 unsigned long bss_align;
806 unsigned long bss_sz;
807 unsigned long align;
808 int i, ret;
809
810 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
811 kbuf->buf_align = bss_align = 1;
812 kbuf->bufsz = bss_sz = 0;
813
814 for (i = 0; i < pi->ehdr->e_shnum; i++) {
815 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
816 continue;
817
818 align = sechdrs[i].sh_addralign;
819 if (sechdrs[i].sh_type != SHT_NOBITS) {
820 if (kbuf->buf_align < align)
821 kbuf->buf_align = align;
822 kbuf->bufsz = ALIGN(kbuf->bufsz, align);
823 kbuf->bufsz += sechdrs[i].sh_size;
824 } else {
825 if (bss_align < align)
826 bss_align = align;
827 bss_sz = ALIGN(bss_sz, align);
828 bss_sz += sechdrs[i].sh_size;
829 }
830 }
831 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
832 kbuf->memsz = kbuf->bufsz + bss_sz;
833 if (kbuf->buf_align < bss_align)
834 kbuf->buf_align = bss_align;
835
836 kbuf->buffer = vzalloc(kbuf->bufsz);
837 if (!kbuf->buffer)
838 return -ENOMEM;
839 pi->purgatory_buf = kbuf->buffer;
840
841 ret = kexec_add_buffer(kbuf);
842 if (ret)
843 goto out;
844
845 return 0;
846out:
847 vfree(pi->purgatory_buf);
848 pi->purgatory_buf = NULL;
849 return ret;
850}
851
852/*
853 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
854 * @pi: Purgatory to be loaded.
855 * @kbuf: Buffer prepared to store purgatory.
856 *
857 * Allocates the memory needed for the buffer. Caller is responsible to free
858 * the memory after use.
859 *
860 * Return: 0 on success, negative errno on error.
861 */
862static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
863 struct kexec_buf *kbuf)
864{
865 unsigned long bss_addr;
866 unsigned long offset;
867 Elf_Shdr *sechdrs;
868 int i;
869
870 /*
871 * The section headers in kexec_purgatory are read-only. In order to
872 * have them modifiable make a temporary copy.
873 */
874 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
875 if (!sechdrs)
876 return -ENOMEM;
877 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
878 pi->ehdr->e_shnum * sizeof(Elf_Shdr));
879 pi->sechdrs = sechdrs;
880
881 offset = 0;
882 bss_addr = kbuf->mem + kbuf->bufsz;
883 kbuf->image->start = pi->ehdr->e_entry;
884
885 for (i = 0; i < pi->ehdr->e_shnum; i++) {
886 unsigned long align;
887 void *src, *dst;
888
889 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
890 continue;
891
892 align = sechdrs[i].sh_addralign;
893 if (sechdrs[i].sh_type == SHT_NOBITS) {
894 bss_addr = ALIGN(bss_addr, align);
895 sechdrs[i].sh_addr = bss_addr;
896 bss_addr += sechdrs[i].sh_size;
897 continue;
898 }
899
900 offset = ALIGN(offset, align);
901 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
902 pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
903 pi->ehdr->e_entry < (sechdrs[i].sh_addr
904 + sechdrs[i].sh_size)) {
905 kbuf->image->start -= sechdrs[i].sh_addr;
906 kbuf->image->start += kbuf->mem + offset;
907 }
908
909 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
910 dst = pi->purgatory_buf + offset;
911 memcpy(dst, src, sechdrs[i].sh_size);
912
913 sechdrs[i].sh_addr = kbuf->mem + offset;
914 sechdrs[i].sh_offset = offset;
915 offset += sechdrs[i].sh_size;
916 }
917
918 return 0;
919}
920
921static int kexec_apply_relocations(struct kimage *image)
922{
923 int i, ret;
924 struct purgatory_info *pi = &image->purgatory_info;
925 const Elf_Shdr *sechdrs;
926
927 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
928
929 for (i = 0; i < pi->ehdr->e_shnum; i++) {
930 const Elf_Shdr *relsec;
931 const Elf_Shdr *symtab;
932 Elf_Shdr *section;
933
934 relsec = sechdrs + i;
935
936 if (relsec->sh_type != SHT_RELA &&
937 relsec->sh_type != SHT_REL)
938 continue;
939
940 /*
941 * For section of type SHT_RELA/SHT_REL,
942 * ->sh_link contains section header index of associated
943 * symbol table. And ->sh_info contains section header
944 * index of section to which relocations apply.
945 */
946 if (relsec->sh_info >= pi->ehdr->e_shnum ||
947 relsec->sh_link >= pi->ehdr->e_shnum)
948 return -ENOEXEC;
949
950 section = pi->sechdrs + relsec->sh_info;
951 symtab = sechdrs + relsec->sh_link;
952
953 if (!(section->sh_flags & SHF_ALLOC))
954 continue;
955
956 /*
957 * symtab->sh_link contain section header index of associated
958 * string table.
959 */
960 if (symtab->sh_link >= pi->ehdr->e_shnum)
961 /* Invalid section number? */
962 continue;
963
964 /*
965 * Respective architecture needs to provide support for applying
966 * relocations of type SHT_RELA/SHT_REL.
967 */
968 if (relsec->sh_type == SHT_RELA)
969 ret = arch_kexec_apply_relocations_add(pi, section,
970 relsec, symtab);
971 else if (relsec->sh_type == SHT_REL)
972 ret = arch_kexec_apply_relocations(pi, section,
973 relsec, symtab);
974 if (ret)
975 return ret;
976 }
977
978 return 0;
979}
980
981/*
982 * kexec_load_purgatory - Load and relocate the purgatory object.
983 * @image: Image to add the purgatory to.
984 * @kbuf: Memory parameters to use.
985 *
986 * Allocates the memory needed for image->purgatory_info.sechdrs and
987 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
988 * to free the memory after use.
989 *
990 * Return: 0 on success, negative errno on error.
991 */
992int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
993{
994 struct purgatory_info *pi = &image->purgatory_info;
995 int ret;
996
997 if (kexec_purgatory_size <= 0)
998 return -EINVAL;
999
1000 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1001
1002 ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1003 if (ret)
1004 return ret;
1005
1006 ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1007 if (ret)
1008 goto out_free_kbuf;
1009
1010 ret = kexec_apply_relocations(image);
1011 if (ret)
1012 goto out;
1013
1014 return 0;
1015out:
1016 vfree(pi->sechdrs);
1017 pi->sechdrs = NULL;
1018out_free_kbuf:
1019 vfree(pi->purgatory_buf);
1020 pi->purgatory_buf = NULL;
1021 return ret;
1022}
1023
1024/*
1025 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1026 * @pi: Purgatory to search in.
1027 * @name: Name of the symbol.
1028 *
1029 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1030 */
1031static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1032 const char *name)
1033{
1034 const Elf_Shdr *sechdrs;
1035 const Elf_Ehdr *ehdr;
1036 const Elf_Sym *syms;
1037 const char *strtab;
1038 int i, k;
1039
1040 if (!pi->ehdr)
1041 return NULL;
1042
1043 ehdr = pi->ehdr;
1044 sechdrs = (void *)ehdr + ehdr->e_shoff;
1045
1046 for (i = 0; i < ehdr->e_shnum; i++) {
1047 if (sechdrs[i].sh_type != SHT_SYMTAB)
1048 continue;
1049
1050 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1051 /* Invalid strtab section number */
1052 continue;
1053 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1054 syms = (void *)ehdr + sechdrs[i].sh_offset;
1055
1056 /* Go through symbols for a match */
1057 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1058 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1059 continue;
1060
1061 if (strcmp(strtab + syms[k].st_name, name) != 0)
1062 continue;
1063
1064 if (syms[k].st_shndx == SHN_UNDEF ||
1065 syms[k].st_shndx >= ehdr->e_shnum) {
1066 pr_debug("Symbol: %s has bad section index %d.\n",
1067 name, syms[k].st_shndx);
1068 return NULL;
1069 }
1070
1071 /* Found the symbol we are looking for */
1072 return &syms[k];
1073 }
1074 }
1075
1076 return NULL;
1077}
1078
1079void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1080{
1081 struct purgatory_info *pi = &image->purgatory_info;
1082 const Elf_Sym *sym;
1083 Elf_Shdr *sechdr;
1084
1085 sym = kexec_purgatory_find_symbol(pi, name);
1086 if (!sym)
1087 return ERR_PTR(-EINVAL);
1088
1089 sechdr = &pi->sechdrs[sym->st_shndx];
1090
1091 /*
1092 * Returns the address where symbol will finally be loaded after
1093 * kexec_load_segment()
1094 */
1095 return (void *)(sechdr->sh_addr + sym->st_value);
1096}
1097
1098/*
1099 * Get or set value of a symbol. If "get_value" is true, symbol value is
1100 * returned in buf otherwise symbol value is set based on value in buf.
1101 */
1102int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1103 void *buf, unsigned int size, bool get_value)
1104{
1105 struct purgatory_info *pi = &image->purgatory_info;
1106 const Elf_Sym *sym;
1107 Elf_Shdr *sec;
1108 char *sym_buf;
1109
1110 sym = kexec_purgatory_find_symbol(pi, name);
1111 if (!sym)
1112 return -EINVAL;
1113
1114 if (sym->st_size != size) {
1115 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1116 name, (unsigned long)sym->st_size, size);
1117 return -EINVAL;
1118 }
1119
1120 sec = pi->sechdrs + sym->st_shndx;
1121
1122 if (sec->sh_type == SHT_NOBITS) {
1123 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1124 get_value ? "get" : "set");
1125 return -EINVAL;
1126 }
1127
1128 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1129
1130 if (get_value)
1131 memcpy((void *)buf, sym_buf, size);
1132 else
1133 memcpy((void *)sym_buf, buf, size);
1134
1135 return 0;
1136}
1137#endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1138
1139int crash_exclude_mem_range(struct crash_mem *mem,
1140 unsigned long long mstart, unsigned long long mend)
1141{
1142 int i, j;
1143 unsigned long long start, end, p_start, p_end;
1144 struct range temp_range = {0, 0};
1145
1146 for (i = 0; i < mem->nr_ranges; i++) {
1147 start = mem->ranges[i].start;
1148 end = mem->ranges[i].end;
1149 p_start = mstart;
1150 p_end = mend;
1151
1152 if (mstart > end || mend < start)
1153 continue;
1154
1155 /* Truncate any area outside of range */
1156 if (mstart < start)
1157 p_start = start;
1158 if (mend > end)
1159 p_end = end;
1160
1161 /* Found completely overlapping range */
1162 if (p_start == start && p_end == end) {
1163 mem->ranges[i].start = 0;
1164 mem->ranges[i].end = 0;
1165 if (i < mem->nr_ranges - 1) {
1166 /* Shift rest of the ranges to left */
1167 for (j = i; j < mem->nr_ranges - 1; j++) {
1168 mem->ranges[j].start =
1169 mem->ranges[j+1].start;
1170 mem->ranges[j].end =
1171 mem->ranges[j+1].end;
1172 }
1173
1174 /*
1175 * Continue to check if there are another overlapping ranges
1176 * from the current position because of shifting the above
1177 * mem ranges.
1178 */
1179 i--;
1180 mem->nr_ranges--;
1181 continue;
1182 }
1183 mem->nr_ranges--;
1184 return 0;
1185 }
1186
1187 if (p_start > start && p_end < end) {
1188 /* Split original range */
1189 mem->ranges[i].end = p_start - 1;
1190 temp_range.start = p_end + 1;
1191 temp_range.end = end;
1192 } else if (p_start != start)
1193 mem->ranges[i].end = p_start - 1;
1194 else
1195 mem->ranges[i].start = p_end + 1;
1196 break;
1197 }
1198
1199 /* If a split happened, add the split to array */
1200 if (!temp_range.end)
1201 return 0;
1202
1203 /* Split happened */
1204 if (i == mem->max_nr_ranges - 1)
1205 return -ENOMEM;
1206
1207 /* Location where new range should go */
1208 j = i + 1;
1209 if (j < mem->nr_ranges) {
1210 /* Move over all ranges one slot towards the end */
1211 for (i = mem->nr_ranges - 1; i >= j; i--)
1212 mem->ranges[i + 1] = mem->ranges[i];
1213 }
1214
1215 mem->ranges[j].start = temp_range.start;
1216 mem->ranges[j].end = temp_range.end;
1217 mem->nr_ranges++;
1218 return 0;
1219}
1220
1221int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
1222 void **addr, unsigned long *sz)
1223{
1224 Elf64_Ehdr *ehdr;
1225 Elf64_Phdr *phdr;
1226 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1227 unsigned char *buf;
1228 unsigned int cpu, i;
1229 unsigned long long notes_addr;
1230 unsigned long mstart, mend;
1231
1232 /* extra phdr for vmcoreinfo ELF note */
1233 nr_phdr = nr_cpus + 1;
1234 nr_phdr += mem->nr_ranges;
1235
1236 /*
1237 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1238 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1239 * I think this is required by tools like gdb. So same physical
1240 * memory will be mapped in two ELF headers. One will contain kernel
1241 * text virtual addresses and other will have __va(physical) addresses.
1242 */
1243
1244 nr_phdr++;
1245 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1246 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1247
1248 buf = vzalloc(elf_sz);
1249 if (!buf)
1250 return -ENOMEM;
1251
1252 ehdr = (Elf64_Ehdr *)buf;
1253 phdr = (Elf64_Phdr *)(ehdr + 1);
1254 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1255 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1256 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1257 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1258 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1259 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1260 ehdr->e_type = ET_CORE;
1261 ehdr->e_machine = ELF_ARCH;
1262 ehdr->e_version = EV_CURRENT;
1263 ehdr->e_phoff = sizeof(Elf64_Ehdr);
1264 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1265 ehdr->e_phentsize = sizeof(Elf64_Phdr);
1266
1267 /* Prepare one phdr of type PT_NOTE for each present CPU */
1268 for_each_present_cpu(cpu) {
1269 phdr->p_type = PT_NOTE;
1270 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1271 phdr->p_offset = phdr->p_paddr = notes_addr;
1272 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1273 (ehdr->e_phnum)++;
1274 phdr++;
1275 }
1276
1277 /* Prepare one PT_NOTE header for vmcoreinfo */
1278 phdr->p_type = PT_NOTE;
1279 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1280 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1281 (ehdr->e_phnum)++;
1282 phdr++;
1283
1284 /* Prepare PT_LOAD type program header for kernel text region */
1285 if (need_kernel_map) {
1286 phdr->p_type = PT_LOAD;
1287 phdr->p_flags = PF_R|PF_W|PF_X;
1288 phdr->p_vaddr = (unsigned long) _text;
1289 phdr->p_filesz = phdr->p_memsz = _end - _text;
1290 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1291 ehdr->e_phnum++;
1292 phdr++;
1293 }
1294
1295 /* Go through all the ranges in mem->ranges[] and prepare phdr */
1296 for (i = 0; i < mem->nr_ranges; i++) {
1297 mstart = mem->ranges[i].start;
1298 mend = mem->ranges[i].end;
1299
1300 phdr->p_type = PT_LOAD;
1301 phdr->p_flags = PF_R|PF_W|PF_X;
1302 phdr->p_offset = mstart;
1303
1304 phdr->p_paddr = mstart;
1305 phdr->p_vaddr = (unsigned long) __va(mstart);
1306 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1307 phdr->p_align = 0;
1308 ehdr->e_phnum++;
1309 pr_debug("Crash PT_LOAD ELF header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n",
1310 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1311 ehdr->e_phnum, phdr->p_offset);
1312 phdr++;
1313 }
1314
1315 *addr = buf;
1316 *sz = elf_sz;
1317 return 0;
1318}