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