Loading...
1/*
2 * kexec: kexec_file_load system call
3 *
4 * Copyright (C) 2014 Red Hat Inc.
5 * Authors:
6 * Vivek Goyal <vgoyal@redhat.com>
7 *
8 * This source code is licensed under the GNU General Public License,
9 * Version 2. See the file COPYING for more details.
10 */
11
12#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
14#include <linux/capability.h>
15#include <linux/mm.h>
16#include <linux/file.h>
17#include <linux/slab.h>
18#include <linux/kexec.h>
19#include <linux/mutex.h>
20#include <linux/list.h>
21#include <linux/fs.h>
22#include <crypto/hash.h>
23#include <crypto/sha.h>
24#include <linux/syscalls.h>
25#include <linux/vmalloc.h>
26#include "kexec_internal.h"
27
28/*
29 * Declare these symbols weak so that if architecture provides a purgatory,
30 * these will be overridden.
31 */
32char __weak kexec_purgatory[0];
33size_t __weak kexec_purgatory_size = 0;
34
35static int kexec_calculate_store_digests(struct kimage *image);
36
37/* Architectures can provide this probe function */
38int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
39 unsigned long buf_len)
40{
41 return -ENOEXEC;
42}
43
44void * __weak arch_kexec_kernel_image_load(struct kimage *image)
45{
46 return ERR_PTR(-ENOEXEC);
47}
48
49int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
50{
51 return -EINVAL;
52}
53
54#ifdef CONFIG_KEXEC_VERIFY_SIG
55int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
56 unsigned long buf_len)
57{
58 return -EKEYREJECTED;
59}
60#endif
61
62/* Apply relocations of type RELA */
63int __weak
64arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
65 unsigned int relsec)
66{
67 pr_err("RELA relocation unsupported.\n");
68 return -ENOEXEC;
69}
70
71/* Apply relocations of type REL */
72int __weak
73arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
74 unsigned int relsec)
75{
76 pr_err("REL relocation unsupported.\n");
77 return -ENOEXEC;
78}
79
80/*
81 * Free up memory used by kernel, initrd, and command line. This is temporary
82 * memory allocation which is not needed any more after these buffers have
83 * been loaded into separate segments and have been copied elsewhere.
84 */
85void kimage_file_post_load_cleanup(struct kimage *image)
86{
87 struct purgatory_info *pi = &image->purgatory_info;
88
89 vfree(image->kernel_buf);
90 image->kernel_buf = NULL;
91
92 vfree(image->initrd_buf);
93 image->initrd_buf = NULL;
94
95 kfree(image->cmdline_buf);
96 image->cmdline_buf = NULL;
97
98 vfree(pi->purgatory_buf);
99 pi->purgatory_buf = NULL;
100
101 vfree(pi->sechdrs);
102 pi->sechdrs = NULL;
103
104 /* See if architecture has anything to cleanup post load */
105 arch_kimage_file_post_load_cleanup(image);
106
107 /*
108 * Above call should have called into bootloader to free up
109 * any data stored in kimage->image_loader_data. It should
110 * be ok now to free it up.
111 */
112 kfree(image->image_loader_data);
113 image->image_loader_data = NULL;
114}
115
116/*
117 * In file mode list of segments is prepared by kernel. Copy relevant
118 * data from user space, do error checking, prepare segment list
119 */
120static int
121kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
122 const char __user *cmdline_ptr,
123 unsigned long cmdline_len, unsigned flags)
124{
125 int ret = 0;
126 void *ldata;
127 loff_t size;
128
129 ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
130 &size, INT_MAX, READING_KEXEC_IMAGE);
131 if (ret)
132 return ret;
133 image->kernel_buf_len = size;
134
135 /* Call arch image probe handlers */
136 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
137 image->kernel_buf_len);
138 if (ret)
139 goto out;
140
141#ifdef CONFIG_KEXEC_VERIFY_SIG
142 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
143 image->kernel_buf_len);
144 if (ret) {
145 pr_debug("kernel signature verification failed.\n");
146 goto out;
147 }
148 pr_debug("kernel signature verification successful.\n");
149#endif
150 /* It is possible that there no initramfs is being loaded */
151 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
152 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
153 &size, INT_MAX,
154 READING_KEXEC_INITRAMFS);
155 if (ret)
156 goto out;
157 image->initrd_buf_len = size;
158 }
159
160 if (cmdline_len) {
161 image->cmdline_buf = kzalloc(cmdline_len, GFP_KERNEL);
162 if (!image->cmdline_buf) {
163 ret = -ENOMEM;
164 goto out;
165 }
166
167 ret = copy_from_user(image->cmdline_buf, cmdline_ptr,
168 cmdline_len);
169 if (ret) {
170 ret = -EFAULT;
171 goto out;
172 }
173
174 image->cmdline_buf_len = cmdline_len;
175
176 /* command line should be a string with last byte null */
177 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
178 ret = -EINVAL;
179 goto out;
180 }
181 }
182
183 /* Call arch image load handlers */
184 ldata = arch_kexec_kernel_image_load(image);
185
186 if (IS_ERR(ldata)) {
187 ret = PTR_ERR(ldata);
188 goto out;
189 }
190
191 image->image_loader_data = ldata;
192out:
193 /* In case of error, free up all allocated memory in this function */
194 if (ret)
195 kimage_file_post_load_cleanup(image);
196 return ret;
197}
198
199static int
200kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
201 int initrd_fd, const char __user *cmdline_ptr,
202 unsigned long cmdline_len, unsigned long flags)
203{
204 int ret;
205 struct kimage *image;
206 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
207
208 image = do_kimage_alloc_init();
209 if (!image)
210 return -ENOMEM;
211
212 image->file_mode = 1;
213
214 if (kexec_on_panic) {
215 /* Enable special crash kernel control page alloc policy. */
216 image->control_page = crashk_res.start;
217 image->type = KEXEC_TYPE_CRASH;
218 }
219
220 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
221 cmdline_ptr, cmdline_len, flags);
222 if (ret)
223 goto out_free_image;
224
225 ret = sanity_check_segment_list(image);
226 if (ret)
227 goto out_free_post_load_bufs;
228
229 ret = -ENOMEM;
230 image->control_code_page = kimage_alloc_control_pages(image,
231 get_order(KEXEC_CONTROL_PAGE_SIZE));
232 if (!image->control_code_page) {
233 pr_err("Could not allocate control_code_buffer\n");
234 goto out_free_post_load_bufs;
235 }
236
237 if (!kexec_on_panic) {
238 image->swap_page = kimage_alloc_control_pages(image, 0);
239 if (!image->swap_page) {
240 pr_err("Could not allocate swap buffer\n");
241 goto out_free_control_pages;
242 }
243 }
244
245 *rimage = image;
246 return 0;
247out_free_control_pages:
248 kimage_free_page_list(&image->control_pages);
249out_free_post_load_bufs:
250 kimage_file_post_load_cleanup(image);
251out_free_image:
252 kfree(image);
253 return ret;
254}
255
256SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
257 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
258 unsigned long, flags)
259{
260 int ret = 0, i;
261 struct kimage **dest_image, *image;
262
263 /* We only trust the superuser with rebooting the system. */
264 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
265 return -EPERM;
266
267 /* Make sure we have a legal set of flags */
268 if (flags != (flags & KEXEC_FILE_FLAGS))
269 return -EINVAL;
270
271 image = NULL;
272
273 if (!mutex_trylock(&kexec_mutex))
274 return -EBUSY;
275
276 dest_image = &kexec_image;
277 if (flags & KEXEC_FILE_ON_CRASH)
278 dest_image = &kexec_crash_image;
279
280 if (flags & KEXEC_FILE_UNLOAD)
281 goto exchange;
282
283 /*
284 * In case of crash, new kernel gets loaded in reserved region. It is
285 * same memory where old crash kernel might be loaded. Free any
286 * current crash dump kernel before we corrupt it.
287 */
288 if (flags & KEXEC_FILE_ON_CRASH)
289 kimage_free(xchg(&kexec_crash_image, NULL));
290
291 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
292 cmdline_len, flags);
293 if (ret)
294 goto out;
295
296 ret = machine_kexec_prepare(image);
297 if (ret)
298 goto out;
299
300 ret = kexec_calculate_store_digests(image);
301 if (ret)
302 goto out;
303
304 for (i = 0; i < image->nr_segments; i++) {
305 struct kexec_segment *ksegment;
306
307 ksegment = &image->segment[i];
308 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
309 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
310 ksegment->memsz);
311
312 ret = kimage_load_segment(image, &image->segment[i]);
313 if (ret)
314 goto out;
315 }
316
317 kimage_terminate(image);
318
319 /*
320 * Free up any temporary buffers allocated which are not needed
321 * after image has been loaded
322 */
323 kimage_file_post_load_cleanup(image);
324exchange:
325 image = xchg(dest_image, image);
326out:
327 mutex_unlock(&kexec_mutex);
328 kimage_free(image);
329 return ret;
330}
331
332static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
333 struct kexec_buf *kbuf)
334{
335 struct kimage *image = kbuf->image;
336 unsigned long temp_start, temp_end;
337
338 temp_end = min(end, kbuf->buf_max);
339 temp_start = temp_end - kbuf->memsz;
340
341 do {
342 /* align down start */
343 temp_start = temp_start & (~(kbuf->buf_align - 1));
344
345 if (temp_start < start || temp_start < kbuf->buf_min)
346 return 0;
347
348 temp_end = temp_start + kbuf->memsz - 1;
349
350 /*
351 * Make sure this does not conflict with any of existing
352 * segments
353 */
354 if (kimage_is_destination_range(image, temp_start, temp_end)) {
355 temp_start = temp_start - PAGE_SIZE;
356 continue;
357 }
358
359 /* We found a suitable memory range */
360 break;
361 } while (1);
362
363 /* If we are here, we found a suitable memory range */
364 kbuf->mem = temp_start;
365
366 /* Success, stop navigating through remaining System RAM ranges */
367 return 1;
368}
369
370static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
371 struct kexec_buf *kbuf)
372{
373 struct kimage *image = kbuf->image;
374 unsigned long temp_start, temp_end;
375
376 temp_start = max(start, kbuf->buf_min);
377
378 do {
379 temp_start = ALIGN(temp_start, kbuf->buf_align);
380 temp_end = temp_start + kbuf->memsz - 1;
381
382 if (temp_end > end || temp_end > kbuf->buf_max)
383 return 0;
384 /*
385 * Make sure this does not conflict with any of existing
386 * segments
387 */
388 if (kimage_is_destination_range(image, temp_start, temp_end)) {
389 temp_start = temp_start + PAGE_SIZE;
390 continue;
391 }
392
393 /* We found a suitable memory range */
394 break;
395 } while (1);
396
397 /* If we are here, we found a suitable memory range */
398 kbuf->mem = temp_start;
399
400 /* Success, stop navigating through remaining System RAM ranges */
401 return 1;
402}
403
404static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
405{
406 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
407 unsigned long sz = end - start + 1;
408
409 /* Returning 0 will take to next memory range */
410 if (sz < kbuf->memsz)
411 return 0;
412
413 if (end < kbuf->buf_min || start > kbuf->buf_max)
414 return 0;
415
416 /*
417 * Allocate memory top down with-in ram range. Otherwise bottom up
418 * allocation.
419 */
420 if (kbuf->top_down)
421 return locate_mem_hole_top_down(start, end, kbuf);
422 return locate_mem_hole_bottom_up(start, end, kbuf);
423}
424
425/*
426 * Helper function for placing a buffer in a kexec segment. This assumes
427 * that kexec_mutex is held.
428 */
429int kexec_add_buffer(struct kimage *image, char *buffer, unsigned long bufsz,
430 unsigned long memsz, unsigned long buf_align,
431 unsigned long buf_min, unsigned long buf_max,
432 bool top_down, unsigned long *load_addr)
433{
434
435 struct kexec_segment *ksegment;
436 struct kexec_buf buf, *kbuf;
437 int ret;
438
439 /* Currently adding segment this way is allowed only in file mode */
440 if (!image->file_mode)
441 return -EINVAL;
442
443 if (image->nr_segments >= KEXEC_SEGMENT_MAX)
444 return -EINVAL;
445
446 /*
447 * Make sure we are not trying to add buffer after allocating
448 * control pages. All segments need to be placed first before
449 * any control pages are allocated. As control page allocation
450 * logic goes through list of segments to make sure there are
451 * no destination overlaps.
452 */
453 if (!list_empty(&image->control_pages)) {
454 WARN_ON(1);
455 return -EINVAL;
456 }
457
458 memset(&buf, 0, sizeof(struct kexec_buf));
459 kbuf = &buf;
460 kbuf->image = image;
461 kbuf->buffer = buffer;
462 kbuf->bufsz = bufsz;
463
464 kbuf->memsz = ALIGN(memsz, PAGE_SIZE);
465 kbuf->buf_align = max(buf_align, PAGE_SIZE);
466 kbuf->buf_min = buf_min;
467 kbuf->buf_max = buf_max;
468 kbuf->top_down = top_down;
469
470 /* Walk the RAM ranges and allocate a suitable range for the buffer */
471 if (image->type == KEXEC_TYPE_CRASH)
472 ret = walk_iomem_res_desc(crashk_res.desc,
473 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
474 crashk_res.start, crashk_res.end, kbuf,
475 locate_mem_hole_callback);
476 else
477 ret = walk_system_ram_res(0, -1, kbuf,
478 locate_mem_hole_callback);
479 if (ret != 1) {
480 /* A suitable memory range could not be found for buffer */
481 return -EADDRNOTAVAIL;
482 }
483
484 /* Found a suitable memory range */
485 ksegment = &image->segment[image->nr_segments];
486 ksegment->kbuf = kbuf->buffer;
487 ksegment->bufsz = kbuf->bufsz;
488 ksegment->mem = kbuf->mem;
489 ksegment->memsz = kbuf->memsz;
490 image->nr_segments++;
491 *load_addr = ksegment->mem;
492 return 0;
493}
494
495/* Calculate and store the digest of segments */
496static int kexec_calculate_store_digests(struct kimage *image)
497{
498 struct crypto_shash *tfm;
499 struct shash_desc *desc;
500 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
501 size_t desc_size, nullsz;
502 char *digest;
503 void *zero_buf;
504 struct kexec_sha_region *sha_regions;
505 struct purgatory_info *pi = &image->purgatory_info;
506
507 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
508 zero_buf_sz = PAGE_SIZE;
509
510 tfm = crypto_alloc_shash("sha256", 0, 0);
511 if (IS_ERR(tfm)) {
512 ret = PTR_ERR(tfm);
513 goto out;
514 }
515
516 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
517 desc = kzalloc(desc_size, GFP_KERNEL);
518 if (!desc) {
519 ret = -ENOMEM;
520 goto out_free_tfm;
521 }
522
523 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
524 sha_regions = vzalloc(sha_region_sz);
525 if (!sha_regions)
526 goto out_free_desc;
527
528 desc->tfm = tfm;
529 desc->flags = 0;
530
531 ret = crypto_shash_init(desc);
532 if (ret < 0)
533 goto out_free_sha_regions;
534
535 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
536 if (!digest) {
537 ret = -ENOMEM;
538 goto out_free_sha_regions;
539 }
540
541 for (j = i = 0; i < image->nr_segments; i++) {
542 struct kexec_segment *ksegment;
543
544 ksegment = &image->segment[i];
545 /*
546 * Skip purgatory as it will be modified once we put digest
547 * info in purgatory.
548 */
549 if (ksegment->kbuf == pi->purgatory_buf)
550 continue;
551
552 ret = crypto_shash_update(desc, ksegment->kbuf,
553 ksegment->bufsz);
554 if (ret)
555 break;
556
557 /*
558 * Assume rest of the buffer is filled with zero and
559 * update digest accordingly.
560 */
561 nullsz = ksegment->memsz - ksegment->bufsz;
562 while (nullsz) {
563 unsigned long bytes = nullsz;
564
565 if (bytes > zero_buf_sz)
566 bytes = zero_buf_sz;
567 ret = crypto_shash_update(desc, zero_buf, bytes);
568 if (ret)
569 break;
570 nullsz -= bytes;
571 }
572
573 if (ret)
574 break;
575
576 sha_regions[j].start = ksegment->mem;
577 sha_regions[j].len = ksegment->memsz;
578 j++;
579 }
580
581 if (!ret) {
582 ret = crypto_shash_final(desc, digest);
583 if (ret)
584 goto out_free_digest;
585 ret = kexec_purgatory_get_set_symbol(image, "sha_regions",
586 sha_regions, sha_region_sz, 0);
587 if (ret)
588 goto out_free_digest;
589
590 ret = kexec_purgatory_get_set_symbol(image, "sha256_digest",
591 digest, SHA256_DIGEST_SIZE, 0);
592 if (ret)
593 goto out_free_digest;
594 }
595
596out_free_digest:
597 kfree(digest);
598out_free_sha_regions:
599 vfree(sha_regions);
600out_free_desc:
601 kfree(desc);
602out_free_tfm:
603 kfree(tfm);
604out:
605 return ret;
606}
607
608/* Actually load purgatory. Lot of code taken from kexec-tools */
609static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
610 unsigned long max, int top_down)
611{
612 struct purgatory_info *pi = &image->purgatory_info;
613 unsigned long align, buf_align, bss_align, buf_sz, bss_sz, bss_pad;
614 unsigned long memsz, entry, load_addr, curr_load_addr, bss_addr, offset;
615 unsigned char *buf_addr, *src;
616 int i, ret = 0, entry_sidx = -1;
617 const Elf_Shdr *sechdrs_c;
618 Elf_Shdr *sechdrs = NULL;
619 void *purgatory_buf = NULL;
620
621 /*
622 * sechdrs_c points to section headers in purgatory and are read
623 * only. No modifications allowed.
624 */
625 sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
626
627 /*
628 * We can not modify sechdrs_c[] and its fields. It is read only.
629 * Copy it over to a local copy where one can store some temporary
630 * data and free it at the end. We need to modify ->sh_addr and
631 * ->sh_offset fields to keep track of permanent and temporary
632 * locations of sections.
633 */
634 sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
635 if (!sechdrs)
636 return -ENOMEM;
637
638 memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
639
640 /*
641 * We seem to have multiple copies of sections. First copy is which
642 * is embedded in kernel in read only section. Some of these sections
643 * will be copied to a temporary buffer and relocated. And these
644 * sections will finally be copied to their final destination at
645 * segment load time.
646 *
647 * Use ->sh_offset to reflect section address in memory. It will
648 * point to original read only copy if section is not allocatable.
649 * Otherwise it will point to temporary copy which will be relocated.
650 *
651 * Use ->sh_addr to contain final address of the section where it
652 * will go during execution time.
653 */
654 for (i = 0; i < pi->ehdr->e_shnum; i++) {
655 if (sechdrs[i].sh_type == SHT_NOBITS)
656 continue;
657
658 sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
659 sechdrs[i].sh_offset;
660 }
661
662 /*
663 * Identify entry point section and make entry relative to section
664 * start.
665 */
666 entry = pi->ehdr->e_entry;
667 for (i = 0; i < pi->ehdr->e_shnum; i++) {
668 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
669 continue;
670
671 if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
672 continue;
673
674 /* Make entry section relative */
675 if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
676 ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
677 pi->ehdr->e_entry)) {
678 entry_sidx = i;
679 entry -= sechdrs[i].sh_addr;
680 break;
681 }
682 }
683
684 /* Determine how much memory is needed to load relocatable object. */
685 buf_align = 1;
686 bss_align = 1;
687 buf_sz = 0;
688 bss_sz = 0;
689
690 for (i = 0; i < pi->ehdr->e_shnum; i++) {
691 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
692 continue;
693
694 align = sechdrs[i].sh_addralign;
695 if (sechdrs[i].sh_type != SHT_NOBITS) {
696 if (buf_align < align)
697 buf_align = align;
698 buf_sz = ALIGN(buf_sz, align);
699 buf_sz += sechdrs[i].sh_size;
700 } else {
701 /* bss section */
702 if (bss_align < align)
703 bss_align = align;
704 bss_sz = ALIGN(bss_sz, align);
705 bss_sz += sechdrs[i].sh_size;
706 }
707 }
708
709 /* Determine the bss padding required to align bss properly */
710 bss_pad = 0;
711 if (buf_sz & (bss_align - 1))
712 bss_pad = bss_align - (buf_sz & (bss_align - 1));
713
714 memsz = buf_sz + bss_pad + bss_sz;
715
716 /* Allocate buffer for purgatory */
717 purgatory_buf = vzalloc(buf_sz);
718 if (!purgatory_buf) {
719 ret = -ENOMEM;
720 goto out;
721 }
722
723 if (buf_align < bss_align)
724 buf_align = bss_align;
725
726 /* Add buffer to segment list */
727 ret = kexec_add_buffer(image, purgatory_buf, buf_sz, memsz,
728 buf_align, min, max, top_down,
729 &pi->purgatory_load_addr);
730 if (ret)
731 goto out;
732
733 /* Load SHF_ALLOC sections */
734 buf_addr = purgatory_buf;
735 load_addr = curr_load_addr = pi->purgatory_load_addr;
736 bss_addr = load_addr + buf_sz + bss_pad;
737
738 for (i = 0; i < pi->ehdr->e_shnum; i++) {
739 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
740 continue;
741
742 align = sechdrs[i].sh_addralign;
743 if (sechdrs[i].sh_type != SHT_NOBITS) {
744 curr_load_addr = ALIGN(curr_load_addr, align);
745 offset = curr_load_addr - load_addr;
746 /* We already modifed ->sh_offset to keep src addr */
747 src = (char *) sechdrs[i].sh_offset;
748 memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
749
750 /* Store load address and source address of section */
751 sechdrs[i].sh_addr = curr_load_addr;
752
753 /*
754 * This section got copied to temporary buffer. Update
755 * ->sh_offset accordingly.
756 */
757 sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
758
759 /* Advance to the next address */
760 curr_load_addr += sechdrs[i].sh_size;
761 } else {
762 bss_addr = ALIGN(bss_addr, align);
763 sechdrs[i].sh_addr = bss_addr;
764 bss_addr += sechdrs[i].sh_size;
765 }
766 }
767
768 /* Update entry point based on load address of text section */
769 if (entry_sidx >= 0)
770 entry += sechdrs[entry_sidx].sh_addr;
771
772 /* Make kernel jump to purgatory after shutdown */
773 image->start = entry;
774
775 /* Used later to get/set symbol values */
776 pi->sechdrs = sechdrs;
777
778 /*
779 * Used later to identify which section is purgatory and skip it
780 * from checksumming.
781 */
782 pi->purgatory_buf = purgatory_buf;
783 return ret;
784out:
785 vfree(sechdrs);
786 vfree(purgatory_buf);
787 return ret;
788}
789
790static int kexec_apply_relocations(struct kimage *image)
791{
792 int i, ret;
793 struct purgatory_info *pi = &image->purgatory_info;
794 Elf_Shdr *sechdrs = pi->sechdrs;
795
796 /* Apply relocations */
797 for (i = 0; i < pi->ehdr->e_shnum; i++) {
798 Elf_Shdr *section, *symtab;
799
800 if (sechdrs[i].sh_type != SHT_RELA &&
801 sechdrs[i].sh_type != SHT_REL)
802 continue;
803
804 /*
805 * For section of type SHT_RELA/SHT_REL,
806 * ->sh_link contains section header index of associated
807 * symbol table. And ->sh_info contains section header
808 * index of section to which relocations apply.
809 */
810 if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
811 sechdrs[i].sh_link >= pi->ehdr->e_shnum)
812 return -ENOEXEC;
813
814 section = &sechdrs[sechdrs[i].sh_info];
815 symtab = &sechdrs[sechdrs[i].sh_link];
816
817 if (!(section->sh_flags & SHF_ALLOC))
818 continue;
819
820 /*
821 * symtab->sh_link contain section header index of associated
822 * string table.
823 */
824 if (symtab->sh_link >= pi->ehdr->e_shnum)
825 /* Invalid section number? */
826 continue;
827
828 /*
829 * Respective architecture needs to provide support for applying
830 * relocations of type SHT_RELA/SHT_REL.
831 */
832 if (sechdrs[i].sh_type == SHT_RELA)
833 ret = arch_kexec_apply_relocations_add(pi->ehdr,
834 sechdrs, i);
835 else if (sechdrs[i].sh_type == SHT_REL)
836 ret = arch_kexec_apply_relocations(pi->ehdr,
837 sechdrs, i);
838 if (ret)
839 return ret;
840 }
841
842 return 0;
843}
844
845/* Load relocatable purgatory object and relocate it appropriately */
846int kexec_load_purgatory(struct kimage *image, unsigned long min,
847 unsigned long max, int top_down,
848 unsigned long *load_addr)
849{
850 struct purgatory_info *pi = &image->purgatory_info;
851 int ret;
852
853 if (kexec_purgatory_size <= 0)
854 return -EINVAL;
855
856 if (kexec_purgatory_size < sizeof(Elf_Ehdr))
857 return -ENOEXEC;
858
859 pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
860
861 if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
862 || pi->ehdr->e_type != ET_REL
863 || !elf_check_arch(pi->ehdr)
864 || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
865 return -ENOEXEC;
866
867 if (pi->ehdr->e_shoff >= kexec_purgatory_size
868 || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
869 kexec_purgatory_size - pi->ehdr->e_shoff))
870 return -ENOEXEC;
871
872 ret = __kexec_load_purgatory(image, min, max, top_down);
873 if (ret)
874 return ret;
875
876 ret = kexec_apply_relocations(image);
877 if (ret)
878 goto out;
879
880 *load_addr = pi->purgatory_load_addr;
881 return 0;
882out:
883 vfree(pi->sechdrs);
884 vfree(pi->purgatory_buf);
885 return ret;
886}
887
888static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
889 const char *name)
890{
891 Elf_Sym *syms;
892 Elf_Shdr *sechdrs;
893 Elf_Ehdr *ehdr;
894 int i, k;
895 const char *strtab;
896
897 if (!pi->sechdrs || !pi->ehdr)
898 return NULL;
899
900 sechdrs = pi->sechdrs;
901 ehdr = pi->ehdr;
902
903 for (i = 0; i < ehdr->e_shnum; i++) {
904 if (sechdrs[i].sh_type != SHT_SYMTAB)
905 continue;
906
907 if (sechdrs[i].sh_link >= ehdr->e_shnum)
908 /* Invalid strtab section number */
909 continue;
910 strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
911 syms = (Elf_Sym *)sechdrs[i].sh_offset;
912
913 /* Go through symbols for a match */
914 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
915 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
916 continue;
917
918 if (strcmp(strtab + syms[k].st_name, name) != 0)
919 continue;
920
921 if (syms[k].st_shndx == SHN_UNDEF ||
922 syms[k].st_shndx >= ehdr->e_shnum) {
923 pr_debug("Symbol: %s has bad section index %d.\n",
924 name, syms[k].st_shndx);
925 return NULL;
926 }
927
928 /* Found the symbol we are looking for */
929 return &syms[k];
930 }
931 }
932
933 return NULL;
934}
935
936void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
937{
938 struct purgatory_info *pi = &image->purgatory_info;
939 Elf_Sym *sym;
940 Elf_Shdr *sechdr;
941
942 sym = kexec_purgatory_find_symbol(pi, name);
943 if (!sym)
944 return ERR_PTR(-EINVAL);
945
946 sechdr = &pi->sechdrs[sym->st_shndx];
947
948 /*
949 * Returns the address where symbol will finally be loaded after
950 * kexec_load_segment()
951 */
952 return (void *)(sechdr->sh_addr + sym->st_value);
953}
954
955/*
956 * Get or set value of a symbol. If "get_value" is true, symbol value is
957 * returned in buf otherwise symbol value is set based on value in buf.
958 */
959int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
960 void *buf, unsigned int size, bool get_value)
961{
962 Elf_Sym *sym;
963 Elf_Shdr *sechdrs;
964 struct purgatory_info *pi = &image->purgatory_info;
965 char *sym_buf;
966
967 sym = kexec_purgatory_find_symbol(pi, name);
968 if (!sym)
969 return -EINVAL;
970
971 if (sym->st_size != size) {
972 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
973 name, (unsigned long)sym->st_size, size);
974 return -EINVAL;
975 }
976
977 sechdrs = pi->sechdrs;
978
979 if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
980 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
981 get_value ? "get" : "set");
982 return -EINVAL;
983 }
984
985 sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
986 sym->st_value;
987
988 if (get_value)
989 memcpy((void *)buf, sym_buf, size);
990 else
991 memcpy((void *)sym_buf, buf, size);
992
993 return 0;
994}
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
68static void *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 kexec_file_dbg_print = false;
128}
129
130#ifdef CONFIG_KEXEC_SIG
131#ifdef CONFIG_SIGNED_PE_FILE_VERIFICATION
132int kexec_kernel_verify_pe_sig(const char *kernel, unsigned long kernel_len)
133{
134 int ret;
135
136 ret = verify_pefile_signature(kernel, kernel_len,
137 VERIFY_USE_SECONDARY_KEYRING,
138 VERIFYING_KEXEC_PE_SIGNATURE);
139 if (ret == -ENOKEY && IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING)) {
140 ret = verify_pefile_signature(kernel, kernel_len,
141 VERIFY_USE_PLATFORM_KEYRING,
142 VERIFYING_KEXEC_PE_SIGNATURE);
143 }
144 return ret;
145}
146#endif
147
148static int kexec_image_verify_sig(struct kimage *image, void *buf,
149 unsigned long buf_len)
150{
151 if (!image->fops || !image->fops->verify_sig) {
152 pr_debug("kernel loader does not support signature verification.\n");
153 return -EKEYREJECTED;
154 }
155
156 return image->fops->verify_sig(buf, buf_len);
157}
158
159static int
160kimage_validate_signature(struct kimage *image)
161{
162 int ret;
163
164 ret = kexec_image_verify_sig(image, image->kernel_buf,
165 image->kernel_buf_len);
166 if (ret) {
167
168 if (sig_enforce) {
169 pr_notice("Enforced kernel signature verification failed (%d).\n", ret);
170 return ret;
171 }
172
173 /*
174 * If IMA is guaranteed to appraise a signature on the kexec
175 * image, permit it even if the kernel is otherwise locked
176 * down.
177 */
178 if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
179 security_locked_down(LOCKDOWN_KEXEC))
180 return -EPERM;
181
182 pr_debug("kernel signature verification failed (%d).\n", ret);
183 }
184
185 return 0;
186}
187#endif
188
189/*
190 * In file mode list of segments is prepared by kernel. Copy relevant
191 * data from user space, do error checking, prepare segment list
192 */
193static int
194kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
195 const char __user *cmdline_ptr,
196 unsigned long cmdline_len, unsigned flags)
197{
198 ssize_t ret;
199 void *ldata;
200
201 ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf,
202 KEXEC_FILE_SIZE_MAX, NULL,
203 READING_KEXEC_IMAGE);
204 if (ret < 0)
205 return ret;
206 image->kernel_buf_len = ret;
207 kexec_dprintk("kernel: %p kernel_size: %#lx\n",
208 image->kernel_buf, image->kernel_buf_len);
209
210 /* Call arch image probe handlers */
211 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
212 image->kernel_buf_len);
213 if (ret)
214 goto out;
215
216#ifdef CONFIG_KEXEC_SIG
217 ret = kimage_validate_signature(image);
218
219 if (ret)
220 goto out;
221#endif
222 /* It is possible that there no initramfs is being loaded */
223 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
224 ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf,
225 KEXEC_FILE_SIZE_MAX, NULL,
226 READING_KEXEC_INITRAMFS);
227 if (ret < 0)
228 goto out;
229 image->initrd_buf_len = ret;
230 ret = 0;
231 }
232
233 if (cmdline_len) {
234 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
235 if (IS_ERR(image->cmdline_buf)) {
236 ret = PTR_ERR(image->cmdline_buf);
237 image->cmdline_buf = NULL;
238 goto out;
239 }
240
241 image->cmdline_buf_len = cmdline_len;
242
243 /* command line should be a string with last byte null */
244 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
245 ret = -EINVAL;
246 goto out;
247 }
248
249 ima_kexec_cmdline(kernel_fd, image->cmdline_buf,
250 image->cmdline_buf_len - 1);
251 }
252
253 /* IMA needs to pass the measurement list to the next kernel. */
254 ima_add_kexec_buffer(image);
255
256 /* Call image load handler */
257 ldata = kexec_image_load_default(image);
258
259 if (IS_ERR(ldata)) {
260 ret = PTR_ERR(ldata);
261 goto out;
262 }
263
264 image->image_loader_data = ldata;
265out:
266 /* In case of error, free up all allocated memory in this function */
267 if (ret)
268 kimage_file_post_load_cleanup(image);
269 return ret;
270}
271
272static int
273kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
274 int initrd_fd, const char __user *cmdline_ptr,
275 unsigned long cmdline_len, unsigned long flags)
276{
277 int ret;
278 struct kimage *image;
279 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
280
281 image = do_kimage_alloc_init();
282 if (!image)
283 return -ENOMEM;
284
285 kexec_file_dbg_print = !!(flags & KEXEC_FILE_DEBUG);
286 image->file_mode = 1;
287
288#ifdef CONFIG_CRASH_DUMP
289 if (kexec_on_panic) {
290 /* Enable special crash kernel control page alloc policy. */
291 image->control_page = crashk_res.start;
292 image->type = KEXEC_TYPE_CRASH;
293 }
294#endif
295
296 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
297 cmdline_ptr, cmdline_len, flags);
298 if (ret)
299 goto out_free_image;
300
301 ret = sanity_check_segment_list(image);
302 if (ret)
303 goto out_free_post_load_bufs;
304
305 ret = -ENOMEM;
306 image->control_code_page = kimage_alloc_control_pages(image,
307 get_order(KEXEC_CONTROL_PAGE_SIZE));
308 if (!image->control_code_page) {
309 pr_err("Could not allocate control_code_buffer\n");
310 goto out_free_post_load_bufs;
311 }
312
313 if (!kexec_on_panic) {
314 image->swap_page = kimage_alloc_control_pages(image, 0);
315 if (!image->swap_page) {
316 pr_err("Could not allocate swap buffer\n");
317 goto out_free_control_pages;
318 }
319 }
320
321 *rimage = image;
322 return 0;
323out_free_control_pages:
324 kimage_free_page_list(&image->control_pages);
325out_free_post_load_bufs:
326 kimage_file_post_load_cleanup(image);
327out_free_image:
328 kfree(image);
329 return ret;
330}
331
332SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
333 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
334 unsigned long, flags)
335{
336 int image_type = (flags & KEXEC_FILE_ON_CRASH) ?
337 KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT;
338 struct kimage **dest_image, *image;
339 int ret = 0, i;
340
341 /* We only trust the superuser with rebooting the system. */
342 if (!kexec_load_permitted(image_type))
343 return -EPERM;
344
345 /* Make sure we have a legal set of flags */
346 if (flags != (flags & KEXEC_FILE_FLAGS))
347 return -EINVAL;
348
349 image = NULL;
350
351 if (!kexec_trylock())
352 return -EBUSY;
353
354#ifdef CONFIG_CRASH_DUMP
355 if (image_type == KEXEC_TYPE_CRASH) {
356 dest_image = &kexec_crash_image;
357 if (kexec_crash_image)
358 arch_kexec_unprotect_crashkres();
359 } else
360#endif
361 dest_image = &kexec_image;
362
363 if (flags & KEXEC_FILE_UNLOAD)
364 goto exchange;
365
366 /*
367 * In case of crash, new kernel gets loaded in reserved region. It is
368 * same memory where old crash kernel might be loaded. Free any
369 * current crash dump kernel before we corrupt it.
370 */
371 if (flags & KEXEC_FILE_ON_CRASH)
372 kimage_free(xchg(&kexec_crash_image, NULL));
373
374 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
375 cmdline_len, flags);
376 if (ret)
377 goto out;
378
379#ifdef CONFIG_CRASH_HOTPLUG
380 if ((flags & KEXEC_FILE_ON_CRASH) && arch_crash_hotplug_support(image, flags))
381 image->hotplug_support = 1;
382#endif
383
384 ret = machine_kexec_prepare(image);
385 if (ret)
386 goto out;
387
388 /*
389 * Some architecture(like S390) may touch the crash memory before
390 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
391 */
392 ret = kimage_crash_copy_vmcoreinfo(image);
393 if (ret)
394 goto out;
395
396 ret = kexec_calculate_store_digests(image);
397 if (ret)
398 goto out;
399
400 kexec_dprintk("nr_segments = %lu\n", image->nr_segments);
401 for (i = 0; i < image->nr_segments; i++) {
402 struct kexec_segment *ksegment;
403
404 ksegment = &image->segment[i];
405 kexec_dprintk("segment[%d]: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
406 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
407 ksegment->memsz);
408
409 ret = kimage_load_segment(image, &image->segment[i]);
410 if (ret)
411 goto out;
412 }
413
414 kimage_terminate(image);
415
416 ret = machine_kexec_post_load(image);
417 if (ret)
418 goto out;
419
420 kexec_dprintk("kexec_file_load: type:%u, start:0x%lx head:0x%lx flags:0x%lx\n",
421 image->type, image->start, image->head, flags);
422 /*
423 * Free up any temporary buffers allocated which are not needed
424 * after image has been loaded
425 */
426 kimage_file_post_load_cleanup(image);
427exchange:
428 image = xchg(dest_image, image);
429out:
430#ifdef CONFIG_CRASH_DUMP
431 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
432 arch_kexec_protect_crashkres();
433#endif
434
435 kexec_unlock();
436 kimage_free(image);
437 return ret;
438}
439
440static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
441 struct kexec_buf *kbuf)
442{
443 struct kimage *image = kbuf->image;
444 unsigned long temp_start, temp_end;
445
446 temp_end = min(end, kbuf->buf_max);
447 temp_start = temp_end - kbuf->memsz + 1;
448
449 do {
450 /* align down start */
451 temp_start = ALIGN_DOWN(temp_start, kbuf->buf_align);
452
453 if (temp_start < start || temp_start < kbuf->buf_min)
454 return 0;
455
456 temp_end = temp_start + kbuf->memsz - 1;
457
458 /*
459 * Make sure this does not conflict with any of existing
460 * segments
461 */
462 if (kimage_is_destination_range(image, temp_start, temp_end)) {
463 temp_start = temp_start - PAGE_SIZE;
464 continue;
465 }
466
467 /* We found a suitable memory range */
468 break;
469 } while (1);
470
471 /* If we are here, we found a suitable memory range */
472 kbuf->mem = temp_start;
473
474 /* Success, stop navigating through remaining System RAM ranges */
475 return 1;
476}
477
478static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
479 struct kexec_buf *kbuf)
480{
481 struct kimage *image = kbuf->image;
482 unsigned long temp_start, temp_end;
483
484 temp_start = max(start, kbuf->buf_min);
485
486 do {
487 temp_start = ALIGN(temp_start, kbuf->buf_align);
488 temp_end = temp_start + kbuf->memsz - 1;
489
490 if (temp_end > end || temp_end > kbuf->buf_max)
491 return 0;
492 /*
493 * Make sure this does not conflict with any of existing
494 * segments
495 */
496 if (kimage_is_destination_range(image, temp_start, temp_end)) {
497 temp_start = temp_start + PAGE_SIZE;
498 continue;
499 }
500
501 /* We found a suitable memory range */
502 break;
503 } while (1);
504
505 /* If we are here, we found a suitable memory range */
506 kbuf->mem = temp_start;
507
508 /* Success, stop navigating through remaining System RAM ranges */
509 return 1;
510}
511
512static int locate_mem_hole_callback(struct resource *res, void *arg)
513{
514 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
515 u64 start = res->start, end = res->end;
516 unsigned long sz = end - start + 1;
517
518 /* Returning 0 will take to next memory range */
519
520 /* Don't use memory that will be detected and handled by a driver. */
521 if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
522 return 0;
523
524 if (sz < kbuf->memsz)
525 return 0;
526
527 if (end < kbuf->buf_min || start > kbuf->buf_max)
528 return 0;
529
530 /*
531 * Allocate memory top down with-in ram range. Otherwise bottom up
532 * allocation.
533 */
534 if (kbuf->top_down)
535 return locate_mem_hole_top_down(start, end, kbuf);
536 return locate_mem_hole_bottom_up(start, end, kbuf);
537}
538
539#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
540static int kexec_walk_memblock(struct kexec_buf *kbuf,
541 int (*func)(struct resource *, void *))
542{
543 int ret = 0;
544 u64 i;
545 phys_addr_t mstart, mend;
546 struct resource res = { };
547
548#ifdef CONFIG_CRASH_DUMP
549 if (kbuf->image->type == KEXEC_TYPE_CRASH)
550 return func(&crashk_res, kbuf);
551#endif
552
553 /*
554 * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
555 * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
556 * locate_mem_hole_callback().
557 */
558 if (kbuf->top_down) {
559 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
560 &mstart, &mend, NULL) {
561 /*
562 * In memblock, end points to the first byte after the
563 * range while in kexec, end points to the last byte
564 * in the range.
565 */
566 res.start = mstart;
567 res.end = mend - 1;
568 ret = func(&res, kbuf);
569 if (ret)
570 break;
571 }
572 } else {
573 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
574 &mstart, &mend, NULL) {
575 /*
576 * In memblock, end points to the first byte after the
577 * range while in kexec, end points to the last byte
578 * in the range.
579 */
580 res.start = mstart;
581 res.end = mend - 1;
582 ret = func(&res, kbuf);
583 if (ret)
584 break;
585 }
586 }
587
588 return ret;
589}
590#else
591static int kexec_walk_memblock(struct kexec_buf *kbuf,
592 int (*func)(struct resource *, void *))
593{
594 return 0;
595}
596#endif
597
598/**
599 * kexec_walk_resources - call func(data) on free memory regions
600 * @kbuf: Context info for the search. Also passed to @func.
601 * @func: Function to call for each memory region.
602 *
603 * Return: The memory walk will stop when func returns a non-zero value
604 * and that value will be returned. If all free regions are visited without
605 * func returning non-zero, then zero will be returned.
606 */
607static int kexec_walk_resources(struct kexec_buf *kbuf,
608 int (*func)(struct resource *, void *))
609{
610#ifdef CONFIG_CRASH_DUMP
611 if (kbuf->image->type == KEXEC_TYPE_CRASH)
612 return walk_iomem_res_desc(crashk_res.desc,
613 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
614 crashk_res.start, crashk_res.end,
615 kbuf, func);
616#endif
617 if (kbuf->top_down)
618 return walk_system_ram_res_rev(0, ULONG_MAX, kbuf, func);
619 else
620 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
621}
622
623/**
624 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
625 * @kbuf: Parameters for the memory search.
626 *
627 * On success, kbuf->mem will have the start address of the memory region found.
628 *
629 * Return: 0 on success, negative errno on error.
630 */
631int kexec_locate_mem_hole(struct kexec_buf *kbuf)
632{
633 int ret;
634
635 /* Arch knows where to place */
636 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
637 return 0;
638
639 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
640 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
641 else
642 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
643
644 return ret == 1 ? 0 : -EADDRNOTAVAIL;
645}
646
647/**
648 * kexec_add_buffer - place a buffer in a kexec segment
649 * @kbuf: Buffer contents and memory parameters.
650 *
651 * This function assumes that kexec_lock is held.
652 * On successful return, @kbuf->mem will have the physical address of
653 * the buffer in memory.
654 *
655 * Return: 0 on success, negative errno on error.
656 */
657int kexec_add_buffer(struct kexec_buf *kbuf)
658{
659 struct kexec_segment *ksegment;
660 int ret;
661
662 /* Currently adding segment this way is allowed only in file mode */
663 if (!kbuf->image->file_mode)
664 return -EINVAL;
665
666 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
667 return -EINVAL;
668
669 /*
670 * Make sure we are not trying to add buffer after allocating
671 * control pages. All segments need to be placed first before
672 * any control pages are allocated. As control page allocation
673 * logic goes through list of segments to make sure there are
674 * no destination overlaps.
675 */
676 if (!list_empty(&kbuf->image->control_pages)) {
677 WARN_ON(1);
678 return -EINVAL;
679 }
680
681 /* Ensure minimum alignment needed for segments. */
682 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
683 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
684
685 /* Walk the RAM ranges and allocate a suitable range for the buffer */
686 ret = arch_kexec_locate_mem_hole(kbuf);
687 if (ret)
688 return ret;
689
690 /* Found a suitable memory range */
691 ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
692 ksegment->kbuf = kbuf->buffer;
693 ksegment->bufsz = kbuf->bufsz;
694 ksegment->mem = kbuf->mem;
695 ksegment->memsz = kbuf->memsz;
696 kbuf->image->nr_segments++;
697 return 0;
698}
699
700/* Calculate and store the digest of segments */
701static int kexec_calculate_store_digests(struct kimage *image)
702{
703 struct crypto_shash *tfm;
704 struct shash_desc *desc;
705 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
706 size_t desc_size, nullsz;
707 char *digest;
708 void *zero_buf;
709 struct kexec_sha_region *sha_regions;
710 struct purgatory_info *pi = &image->purgatory_info;
711
712 if (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY))
713 return 0;
714
715 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
716 zero_buf_sz = PAGE_SIZE;
717
718 tfm = crypto_alloc_shash("sha256", 0, 0);
719 if (IS_ERR(tfm)) {
720 ret = PTR_ERR(tfm);
721 goto out;
722 }
723
724 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
725 desc = kzalloc(desc_size, GFP_KERNEL);
726 if (!desc) {
727 ret = -ENOMEM;
728 goto out_free_tfm;
729 }
730
731 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
732 sha_regions = vzalloc(sha_region_sz);
733 if (!sha_regions) {
734 ret = -ENOMEM;
735 goto out_free_desc;
736 }
737
738 desc->tfm = tfm;
739
740 ret = crypto_shash_init(desc);
741 if (ret < 0)
742 goto out_free_sha_regions;
743
744 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
745 if (!digest) {
746 ret = -ENOMEM;
747 goto out_free_sha_regions;
748 }
749
750 for (j = i = 0; i < image->nr_segments; i++) {
751 struct kexec_segment *ksegment;
752
753#ifdef CONFIG_CRASH_HOTPLUG
754 /* Exclude elfcorehdr segment to allow future changes via hotplug */
755 if (i == image->elfcorehdr_index)
756 continue;
757#endif
758
759 ksegment = &image->segment[i];
760 /*
761 * Skip purgatory as it will be modified once we put digest
762 * info in purgatory.
763 */
764 if (ksegment->kbuf == pi->purgatory_buf)
765 continue;
766
767 ret = crypto_shash_update(desc, ksegment->kbuf,
768 ksegment->bufsz);
769 if (ret)
770 break;
771
772 /*
773 * Assume rest of the buffer is filled with zero and
774 * update digest accordingly.
775 */
776 nullsz = ksegment->memsz - ksegment->bufsz;
777 while (nullsz) {
778 unsigned long bytes = nullsz;
779
780 if (bytes > zero_buf_sz)
781 bytes = zero_buf_sz;
782 ret = crypto_shash_update(desc, zero_buf, bytes);
783 if (ret)
784 break;
785 nullsz -= bytes;
786 }
787
788 if (ret)
789 break;
790
791 sha_regions[j].start = ksegment->mem;
792 sha_regions[j].len = ksegment->memsz;
793 j++;
794 }
795
796 if (!ret) {
797 ret = crypto_shash_final(desc, digest);
798 if (ret)
799 goto out_free_digest;
800 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
801 sha_regions, sha_region_sz, 0);
802 if (ret)
803 goto out_free_digest;
804
805 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
806 digest, SHA256_DIGEST_SIZE, 0);
807 if (ret)
808 goto out_free_digest;
809 }
810
811out_free_digest:
812 kfree(digest);
813out_free_sha_regions:
814 vfree(sha_regions);
815out_free_desc:
816 kfree(desc);
817out_free_tfm:
818 kfree(tfm);
819out:
820 return ret;
821}
822
823#ifdef CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY
824/*
825 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
826 * @pi: Purgatory to be loaded.
827 * @kbuf: Buffer to setup.
828 *
829 * Allocates the memory needed for the buffer. Caller is responsible to free
830 * the memory after use.
831 *
832 * Return: 0 on success, negative errno on error.
833 */
834static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
835 struct kexec_buf *kbuf)
836{
837 const Elf_Shdr *sechdrs;
838 unsigned long bss_align;
839 unsigned long bss_sz;
840 unsigned long align;
841 int i, ret;
842
843 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
844 kbuf->buf_align = bss_align = 1;
845 kbuf->bufsz = bss_sz = 0;
846
847 for (i = 0; i < pi->ehdr->e_shnum; i++) {
848 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
849 continue;
850
851 align = sechdrs[i].sh_addralign;
852 if (sechdrs[i].sh_type != SHT_NOBITS) {
853 if (kbuf->buf_align < align)
854 kbuf->buf_align = align;
855 kbuf->bufsz = ALIGN(kbuf->bufsz, align);
856 kbuf->bufsz += sechdrs[i].sh_size;
857 } else {
858 if (bss_align < align)
859 bss_align = align;
860 bss_sz = ALIGN(bss_sz, align);
861 bss_sz += sechdrs[i].sh_size;
862 }
863 }
864 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
865 kbuf->memsz = kbuf->bufsz + bss_sz;
866 if (kbuf->buf_align < bss_align)
867 kbuf->buf_align = bss_align;
868
869 kbuf->buffer = vzalloc(kbuf->bufsz);
870 if (!kbuf->buffer)
871 return -ENOMEM;
872 pi->purgatory_buf = kbuf->buffer;
873
874 ret = kexec_add_buffer(kbuf);
875 if (ret)
876 goto out;
877
878 return 0;
879out:
880 vfree(pi->purgatory_buf);
881 pi->purgatory_buf = NULL;
882 return ret;
883}
884
885/*
886 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
887 * @pi: Purgatory to be loaded.
888 * @kbuf: Buffer prepared to store purgatory.
889 *
890 * Allocates the memory needed for the buffer. Caller is responsible to free
891 * the memory after use.
892 *
893 * Return: 0 on success, negative errno on error.
894 */
895static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
896 struct kexec_buf *kbuf)
897{
898 unsigned long bss_addr;
899 unsigned long offset;
900 size_t sechdrs_size;
901 Elf_Shdr *sechdrs;
902 int i;
903
904 /*
905 * The section headers in kexec_purgatory are read-only. In order to
906 * have them modifiable make a temporary copy.
907 */
908 sechdrs_size = array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum);
909 sechdrs = vzalloc(sechdrs_size);
910 if (!sechdrs)
911 return -ENOMEM;
912 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, sechdrs_size);
913 pi->sechdrs = sechdrs;
914
915 offset = 0;
916 bss_addr = kbuf->mem + kbuf->bufsz;
917 kbuf->image->start = pi->ehdr->e_entry;
918
919 for (i = 0; i < pi->ehdr->e_shnum; i++) {
920 unsigned long align;
921 void *src, *dst;
922
923 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
924 continue;
925
926 align = sechdrs[i].sh_addralign;
927 if (sechdrs[i].sh_type == SHT_NOBITS) {
928 bss_addr = ALIGN(bss_addr, align);
929 sechdrs[i].sh_addr = bss_addr;
930 bss_addr += sechdrs[i].sh_size;
931 continue;
932 }
933
934 offset = ALIGN(offset, align);
935
936 /*
937 * Check if the segment contains the entry point, if so,
938 * calculate the value of image->start based on it.
939 * If the compiler has produced more than one .text section
940 * (Eg: .text.hot), they are generally after the main .text
941 * section, and they shall not be used to calculate
942 * image->start. So do not re-calculate image->start if it
943 * is not set to the initial value, and warn the user so they
944 * have a chance to fix their purgatory's linker script.
945 */
946 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
947 pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
948 pi->ehdr->e_entry < (sechdrs[i].sh_addr
949 + sechdrs[i].sh_size) &&
950 !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) {
951 kbuf->image->start -= sechdrs[i].sh_addr;
952 kbuf->image->start += kbuf->mem + offset;
953 }
954
955 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
956 dst = pi->purgatory_buf + offset;
957 memcpy(dst, src, sechdrs[i].sh_size);
958
959 sechdrs[i].sh_addr = kbuf->mem + offset;
960 sechdrs[i].sh_offset = offset;
961 offset += sechdrs[i].sh_size;
962 }
963
964 return 0;
965}
966
967static int kexec_apply_relocations(struct kimage *image)
968{
969 int i, ret;
970 struct purgatory_info *pi = &image->purgatory_info;
971 const Elf_Shdr *sechdrs;
972
973 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
974
975 for (i = 0; i < pi->ehdr->e_shnum; i++) {
976 const Elf_Shdr *relsec;
977 const Elf_Shdr *symtab;
978 Elf_Shdr *section;
979
980 relsec = sechdrs + i;
981
982 if (relsec->sh_type != SHT_RELA &&
983 relsec->sh_type != SHT_REL)
984 continue;
985
986 /*
987 * For section of type SHT_RELA/SHT_REL,
988 * ->sh_link contains section header index of associated
989 * symbol table. And ->sh_info contains section header
990 * index of section to which relocations apply.
991 */
992 if (relsec->sh_info >= pi->ehdr->e_shnum ||
993 relsec->sh_link >= pi->ehdr->e_shnum)
994 return -ENOEXEC;
995
996 section = pi->sechdrs + relsec->sh_info;
997 symtab = sechdrs + relsec->sh_link;
998
999 if (!(section->sh_flags & SHF_ALLOC))
1000 continue;
1001
1002 /*
1003 * symtab->sh_link contain section header index of associated
1004 * string table.
1005 */
1006 if (symtab->sh_link >= pi->ehdr->e_shnum)
1007 /* Invalid section number? */
1008 continue;
1009
1010 /*
1011 * Respective architecture needs to provide support for applying
1012 * relocations of type SHT_RELA/SHT_REL.
1013 */
1014 if (relsec->sh_type == SHT_RELA)
1015 ret = arch_kexec_apply_relocations_add(pi, section,
1016 relsec, symtab);
1017 else if (relsec->sh_type == SHT_REL)
1018 ret = arch_kexec_apply_relocations(pi, section,
1019 relsec, symtab);
1020 if (ret)
1021 return ret;
1022 }
1023
1024 return 0;
1025}
1026
1027/*
1028 * kexec_load_purgatory - Load and relocate the purgatory object.
1029 * @image: Image to add the purgatory to.
1030 * @kbuf: Memory parameters to use.
1031 *
1032 * Allocates the memory needed for image->purgatory_info.sechdrs and
1033 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1034 * to free the memory after use.
1035 *
1036 * Return: 0 on success, negative errno on error.
1037 */
1038int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1039{
1040 struct purgatory_info *pi = &image->purgatory_info;
1041 int ret;
1042
1043 if (kexec_purgatory_size <= 0)
1044 return -EINVAL;
1045
1046 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1047
1048 ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1049 if (ret)
1050 return ret;
1051
1052 ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1053 if (ret)
1054 goto out_free_kbuf;
1055
1056 ret = kexec_apply_relocations(image);
1057 if (ret)
1058 goto out;
1059
1060 return 0;
1061out:
1062 vfree(pi->sechdrs);
1063 pi->sechdrs = NULL;
1064out_free_kbuf:
1065 vfree(pi->purgatory_buf);
1066 pi->purgatory_buf = NULL;
1067 return ret;
1068}
1069
1070/*
1071 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1072 * @pi: Purgatory to search in.
1073 * @name: Name of the symbol.
1074 *
1075 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1076 */
1077static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1078 const char *name)
1079{
1080 const Elf_Shdr *sechdrs;
1081 const Elf_Ehdr *ehdr;
1082 const Elf_Sym *syms;
1083 const char *strtab;
1084 int i, k;
1085
1086 if (!pi->ehdr)
1087 return NULL;
1088
1089 ehdr = pi->ehdr;
1090 sechdrs = (void *)ehdr + ehdr->e_shoff;
1091
1092 for (i = 0; i < ehdr->e_shnum; i++) {
1093 if (sechdrs[i].sh_type != SHT_SYMTAB)
1094 continue;
1095
1096 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1097 /* Invalid strtab section number */
1098 continue;
1099 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1100 syms = (void *)ehdr + sechdrs[i].sh_offset;
1101
1102 /* Go through symbols for a match */
1103 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1104 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1105 continue;
1106
1107 if (strcmp(strtab + syms[k].st_name, name) != 0)
1108 continue;
1109
1110 if (syms[k].st_shndx == SHN_UNDEF ||
1111 syms[k].st_shndx >= ehdr->e_shnum) {
1112 pr_debug("Symbol: %s has bad section index %d.\n",
1113 name, syms[k].st_shndx);
1114 return NULL;
1115 }
1116
1117 /* Found the symbol we are looking for */
1118 return &syms[k];
1119 }
1120 }
1121
1122 return NULL;
1123}
1124
1125void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1126{
1127 struct purgatory_info *pi = &image->purgatory_info;
1128 const Elf_Sym *sym;
1129 Elf_Shdr *sechdr;
1130
1131 sym = kexec_purgatory_find_symbol(pi, name);
1132 if (!sym)
1133 return ERR_PTR(-EINVAL);
1134
1135 sechdr = &pi->sechdrs[sym->st_shndx];
1136
1137 /*
1138 * Returns the address where symbol will finally be loaded after
1139 * kexec_load_segment()
1140 */
1141 return (void *)(sechdr->sh_addr + sym->st_value);
1142}
1143
1144/*
1145 * Get or set value of a symbol. If "get_value" is true, symbol value is
1146 * returned in buf otherwise symbol value is set based on value in buf.
1147 */
1148int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1149 void *buf, unsigned int size, bool get_value)
1150{
1151 struct purgatory_info *pi = &image->purgatory_info;
1152 const Elf_Sym *sym;
1153 Elf_Shdr *sec;
1154 char *sym_buf;
1155
1156 sym = kexec_purgatory_find_symbol(pi, name);
1157 if (!sym)
1158 return -EINVAL;
1159
1160 if (sym->st_size != size) {
1161 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1162 name, (unsigned long)sym->st_size, size);
1163 return -EINVAL;
1164 }
1165
1166 sec = pi->sechdrs + sym->st_shndx;
1167
1168 if (sec->sh_type == SHT_NOBITS) {
1169 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1170 get_value ? "get" : "set");
1171 return -EINVAL;
1172 }
1173
1174 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1175
1176 if (get_value)
1177 memcpy((void *)buf, sym_buf, size);
1178 else
1179 memcpy((void *)sym_buf, buf, size);
1180
1181 return 0;
1182}
1183#endif /* CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY */