Linux 5.8-rc4
[linux/fpc-iii.git] / kernel / kexec_file.c
blob09cc78df53c64cfbe3cb88f07f4e32d478dae6d8
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * kexec: kexec_file_load system call
5 * Copyright (C) 2014 Red Hat Inc.
6 * Authors:
7 * Vivek Goyal <vgoyal@redhat.com>
8 */
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/capability.h>
13 #include <linux/mm.h>
14 #include <linux/file.h>
15 #include <linux/slab.h>
16 #include <linux/kexec.h>
17 #include <linux/memblock.h>
18 #include <linux/mutex.h>
19 #include <linux/list.h>
20 #include <linux/fs.h>
21 #include <linux/ima.h>
22 #include <crypto/hash.h>
23 #include <crypto/sha.h>
24 #include <linux/elf.h>
25 #include <linux/elfcore.h>
26 #include <linux/kernel.h>
27 #include <linux/syscalls.h>
28 #include <linux/vmalloc.h>
29 #include "kexec_internal.h"
31 static int kexec_calculate_store_digests(struct kimage *image);
34 * Currently this is the only default function that is exported as some
35 * architectures need it to do additional handlings.
36 * In the future, other default functions may be exported too if required.
38 int kexec_image_probe_default(struct kimage *image, void *buf,
39 unsigned long buf_len)
41 const struct kexec_file_ops * const *fops;
42 int ret = -ENOEXEC;
44 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
45 ret = (*fops)->probe(buf, buf_len);
46 if (!ret) {
47 image->fops = *fops;
48 return ret;
52 return ret;
55 /* Architectures can provide this probe function */
56 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
57 unsigned long buf_len)
59 return kexec_image_probe_default(image, buf, buf_len);
62 static void *kexec_image_load_default(struct kimage *image)
64 if (!image->fops || !image->fops->load)
65 return ERR_PTR(-ENOEXEC);
67 return image->fops->load(image, image->kernel_buf,
68 image->kernel_buf_len, image->initrd_buf,
69 image->initrd_buf_len, image->cmdline_buf,
70 image->cmdline_buf_len);
73 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
75 return kexec_image_load_default(image);
78 int kexec_image_post_load_cleanup_default(struct kimage *image)
80 if (!image->fops || !image->fops->cleanup)
81 return 0;
83 return image->fops->cleanup(image->image_loader_data);
86 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
88 return kexec_image_post_load_cleanup_default(image);
91 #ifdef CONFIG_KEXEC_SIG
92 static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
93 unsigned long buf_len)
95 if (!image->fops || !image->fops->verify_sig) {
96 pr_debug("kernel loader does not support signature verification.\n");
97 return -EKEYREJECTED;
100 return image->fops->verify_sig(buf, buf_len);
103 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
104 unsigned long buf_len)
106 return kexec_image_verify_sig_default(image, buf, buf_len);
108 #endif
111 * arch_kexec_apply_relocations_add - apply relocations of type RELA
112 * @pi: Purgatory to be relocated.
113 * @section: Section relocations applying to.
114 * @relsec: Section containing RELAs.
115 * @symtab: Corresponding symtab.
117 * Return: 0 on success, negative errno on error.
119 int __weak
120 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section,
121 const Elf_Shdr *relsec, const Elf_Shdr *symtab)
123 pr_err("RELA relocation unsupported.\n");
124 return -ENOEXEC;
128 * arch_kexec_apply_relocations - apply relocations of type REL
129 * @pi: Purgatory to be relocated.
130 * @section: Section relocations applying to.
131 * @relsec: Section containing RELs.
132 * @symtab: Corresponding symtab.
134 * Return: 0 on success, negative errno on error.
136 int __weak
137 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section,
138 const Elf_Shdr *relsec, const Elf_Shdr *symtab)
140 pr_err("REL relocation unsupported.\n");
141 return -ENOEXEC;
145 * Free up memory used by kernel, initrd, and command line. This is temporary
146 * memory allocation which is not needed any more after these buffers have
147 * been loaded into separate segments and have been copied elsewhere.
149 void kimage_file_post_load_cleanup(struct kimage *image)
151 struct purgatory_info *pi = &image->purgatory_info;
153 vfree(image->kernel_buf);
154 image->kernel_buf = NULL;
156 vfree(image->initrd_buf);
157 image->initrd_buf = NULL;
159 kfree(image->cmdline_buf);
160 image->cmdline_buf = NULL;
162 vfree(pi->purgatory_buf);
163 pi->purgatory_buf = NULL;
165 vfree(pi->sechdrs);
166 pi->sechdrs = NULL;
168 /* See if architecture has anything to cleanup post load */
169 arch_kimage_file_post_load_cleanup(image);
172 * Above call should have called into bootloader to free up
173 * any data stored in kimage->image_loader_data. It should
174 * be ok now to free it up.
176 kfree(image->image_loader_data);
177 image->image_loader_data = NULL;
180 #ifdef CONFIG_KEXEC_SIG
181 static int
182 kimage_validate_signature(struct kimage *image)
184 int ret;
186 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
187 image->kernel_buf_len);
188 if (ret) {
190 if (IS_ENABLED(CONFIG_KEXEC_SIG_FORCE)) {
191 pr_notice("Enforced kernel signature verification failed (%d).\n", ret);
192 return ret;
196 * If IMA is guaranteed to appraise a signature on the kexec
197 * image, permit it even if the kernel is otherwise locked
198 * down.
200 if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
201 security_locked_down(LOCKDOWN_KEXEC))
202 return -EPERM;
204 pr_debug("kernel signature verification failed (%d).\n", ret);
207 return 0;
209 #endif
212 * In file mode list of segments is prepared by kernel. Copy relevant
213 * data from user space, do error checking, prepare segment list
215 static int
216 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
217 const char __user *cmdline_ptr,
218 unsigned long cmdline_len, unsigned flags)
220 int ret;
221 void *ldata;
222 loff_t size;
224 ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
225 &size, INT_MAX, READING_KEXEC_IMAGE);
226 if (ret)
227 return ret;
228 image->kernel_buf_len = size;
230 /* Call arch image probe handlers */
231 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
232 image->kernel_buf_len);
233 if (ret)
234 goto out;
236 #ifdef CONFIG_KEXEC_SIG
237 ret = kimage_validate_signature(image);
239 if (ret)
240 goto out;
241 #endif
242 /* It is possible that there no initramfs is being loaded */
243 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
244 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
245 &size, INT_MAX,
246 READING_KEXEC_INITRAMFS);
247 if (ret)
248 goto out;
249 image->initrd_buf_len = size;
252 if (cmdline_len) {
253 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
254 if (IS_ERR(image->cmdline_buf)) {
255 ret = PTR_ERR(image->cmdline_buf);
256 image->cmdline_buf = NULL;
257 goto out;
260 image->cmdline_buf_len = cmdline_len;
262 /* command line should be a string with last byte null */
263 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
264 ret = -EINVAL;
265 goto out;
268 ima_kexec_cmdline(image->cmdline_buf,
269 image->cmdline_buf_len - 1);
272 /* IMA needs to pass the measurement list to the next kernel. */
273 ima_add_kexec_buffer(image);
275 /* Call arch image load handlers */
276 ldata = arch_kexec_kernel_image_load(image);
278 if (IS_ERR(ldata)) {
279 ret = PTR_ERR(ldata);
280 goto out;
283 image->image_loader_data = ldata;
284 out:
285 /* In case of error, free up all allocated memory in this function */
286 if (ret)
287 kimage_file_post_load_cleanup(image);
288 return ret;
291 static int
292 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
293 int initrd_fd, const char __user *cmdline_ptr,
294 unsigned long cmdline_len, unsigned long flags)
296 int ret;
297 struct kimage *image;
298 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
300 image = do_kimage_alloc_init();
301 if (!image)
302 return -ENOMEM;
304 image->file_mode = 1;
306 if (kexec_on_panic) {
307 /* Enable special crash kernel control page alloc policy. */
308 image->control_page = crashk_res.start;
309 image->type = KEXEC_TYPE_CRASH;
312 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
313 cmdline_ptr, cmdline_len, flags);
314 if (ret)
315 goto out_free_image;
317 ret = sanity_check_segment_list(image);
318 if (ret)
319 goto out_free_post_load_bufs;
321 ret = -ENOMEM;
322 image->control_code_page = kimage_alloc_control_pages(image,
323 get_order(KEXEC_CONTROL_PAGE_SIZE));
324 if (!image->control_code_page) {
325 pr_err("Could not allocate control_code_buffer\n");
326 goto out_free_post_load_bufs;
329 if (!kexec_on_panic) {
330 image->swap_page = kimage_alloc_control_pages(image, 0);
331 if (!image->swap_page) {
332 pr_err("Could not allocate swap buffer\n");
333 goto out_free_control_pages;
337 *rimage = image;
338 return 0;
339 out_free_control_pages:
340 kimage_free_page_list(&image->control_pages);
341 out_free_post_load_bufs:
342 kimage_file_post_load_cleanup(image);
343 out_free_image:
344 kfree(image);
345 return ret;
348 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
349 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
350 unsigned long, flags)
352 int ret = 0, i;
353 struct kimage **dest_image, *image;
355 /* We only trust the superuser with rebooting the system. */
356 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
357 return -EPERM;
359 /* Make sure we have a legal set of flags */
360 if (flags != (flags & KEXEC_FILE_FLAGS))
361 return -EINVAL;
363 image = NULL;
365 if (!mutex_trylock(&kexec_mutex))
366 return -EBUSY;
368 dest_image = &kexec_image;
369 if (flags & KEXEC_FILE_ON_CRASH) {
370 dest_image = &kexec_crash_image;
371 if (kexec_crash_image)
372 arch_kexec_unprotect_crashkres();
375 if (flags & KEXEC_FILE_UNLOAD)
376 goto exchange;
379 * In case of crash, new kernel gets loaded in reserved region. It is
380 * same memory where old crash kernel might be loaded. Free any
381 * current crash dump kernel before we corrupt it.
383 if (flags & KEXEC_FILE_ON_CRASH)
384 kimage_free(xchg(&kexec_crash_image, NULL));
386 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
387 cmdline_len, flags);
388 if (ret)
389 goto out;
391 ret = machine_kexec_prepare(image);
392 if (ret)
393 goto out;
396 * Some architecture(like S390) may touch the crash memory before
397 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
399 ret = kimage_crash_copy_vmcoreinfo(image);
400 if (ret)
401 goto out;
403 ret = kexec_calculate_store_digests(image);
404 if (ret)
405 goto out;
407 for (i = 0; i < image->nr_segments; i++) {
408 struct kexec_segment *ksegment;
410 ksegment = &image->segment[i];
411 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
412 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
413 ksegment->memsz);
415 ret = kimage_load_segment(image, &image->segment[i]);
416 if (ret)
417 goto out;
420 kimage_terminate(image);
422 ret = machine_kexec_post_load(image);
423 if (ret)
424 goto out;
427 * Free up any temporary buffers allocated which are not needed
428 * after image has been loaded
430 kimage_file_post_load_cleanup(image);
431 exchange:
432 image = xchg(dest_image, image);
433 out:
434 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
435 arch_kexec_protect_crashkres();
437 mutex_unlock(&kexec_mutex);
438 kimage_free(image);
439 return ret;
442 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
443 struct kexec_buf *kbuf)
445 struct kimage *image = kbuf->image;
446 unsigned long temp_start, temp_end;
448 temp_end = min(end, kbuf->buf_max);
449 temp_start = temp_end - kbuf->memsz;
451 do {
452 /* align down start */
453 temp_start = temp_start & (~(kbuf->buf_align - 1));
455 if (temp_start < start || temp_start < kbuf->buf_min)
456 return 0;
458 temp_end = temp_start + kbuf->memsz - 1;
461 * Make sure this does not conflict with any of existing
462 * segments
464 if (kimage_is_destination_range(image, temp_start, temp_end)) {
465 temp_start = temp_start - PAGE_SIZE;
466 continue;
469 /* We found a suitable memory range */
470 break;
471 } while (1);
473 /* If we are here, we found a suitable memory range */
474 kbuf->mem = temp_start;
476 /* Success, stop navigating through remaining System RAM ranges */
477 return 1;
480 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
481 struct kexec_buf *kbuf)
483 struct kimage *image = kbuf->image;
484 unsigned long temp_start, temp_end;
486 temp_start = max(start, kbuf->buf_min);
488 do {
489 temp_start = ALIGN(temp_start, kbuf->buf_align);
490 temp_end = temp_start + kbuf->memsz - 1;
492 if (temp_end > end || temp_end > kbuf->buf_max)
493 return 0;
495 * Make sure this does not conflict with any of existing
496 * segments
498 if (kimage_is_destination_range(image, temp_start, temp_end)) {
499 temp_start = temp_start + PAGE_SIZE;
500 continue;
503 /* We found a suitable memory range */
504 break;
505 } while (1);
507 /* If we are here, we found a suitable memory range */
508 kbuf->mem = temp_start;
510 /* Success, stop navigating through remaining System RAM ranges */
511 return 1;
514 static int locate_mem_hole_callback(struct resource *res, void *arg)
516 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
517 u64 start = res->start, end = res->end;
518 unsigned long sz = end - start + 1;
520 /* Returning 0 will take to next memory range */
522 /* Don't use memory that will be detected and handled by a driver. */
523 if (res->flags & IORESOURCE_MEM_DRIVER_MANAGED)
524 return 0;
526 if (sz < kbuf->memsz)
527 return 0;
529 if (end < kbuf->buf_min || start > kbuf->buf_max)
530 return 0;
533 * Allocate memory top down with-in ram range. Otherwise bottom up
534 * allocation.
536 if (kbuf->top_down)
537 return locate_mem_hole_top_down(start, end, kbuf);
538 return locate_mem_hole_bottom_up(start, end, kbuf);
541 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
542 static int kexec_walk_memblock(struct kexec_buf *kbuf,
543 int (*func)(struct resource *, void *))
545 int ret = 0;
546 u64 i;
547 phys_addr_t mstart, mend;
548 struct resource res = { };
550 if (kbuf->image->type == KEXEC_TYPE_CRASH)
551 return func(&crashk_res, kbuf);
553 if (kbuf->top_down) {
554 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
555 &mstart, &mend, NULL) {
557 * In memblock, end points to the first byte after the
558 * range while in kexec, end points to the last byte
559 * in the range.
561 res.start = mstart;
562 res.end = mend - 1;
563 ret = func(&res, kbuf);
564 if (ret)
565 break;
567 } else {
568 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
569 &mstart, &mend, NULL) {
571 * In memblock, end points to the first byte after the
572 * range while in kexec, end points to the last byte
573 * in the range.
575 res.start = mstart;
576 res.end = mend - 1;
577 ret = func(&res, kbuf);
578 if (ret)
579 break;
583 return ret;
585 #else
586 static int kexec_walk_memblock(struct kexec_buf *kbuf,
587 int (*func)(struct resource *, void *))
589 return 0;
591 #endif
594 * kexec_walk_resources - call func(data) on free memory regions
595 * @kbuf: Context info for the search. Also passed to @func.
596 * @func: Function to call for each memory region.
598 * Return: The memory walk will stop when func returns a non-zero value
599 * and that value will be returned. If all free regions are visited without
600 * func returning non-zero, then zero will be returned.
602 static int kexec_walk_resources(struct kexec_buf *kbuf,
603 int (*func)(struct resource *, void *))
605 if (kbuf->image->type == KEXEC_TYPE_CRASH)
606 return walk_iomem_res_desc(crashk_res.desc,
607 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
608 crashk_res.start, crashk_res.end,
609 kbuf, func);
610 else
611 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
615 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
616 * @kbuf: Parameters for the memory search.
618 * On success, kbuf->mem will have the start address of the memory region found.
620 * Return: 0 on success, negative errno on error.
622 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
624 int ret;
626 /* Arch knows where to place */
627 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
628 return 0;
630 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
631 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
632 else
633 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
635 return ret == 1 ? 0 : -EADDRNOTAVAIL;
639 * kexec_add_buffer - place a buffer in a kexec segment
640 * @kbuf: Buffer contents and memory parameters.
642 * This function assumes that kexec_mutex is held.
643 * On successful return, @kbuf->mem will have the physical address of
644 * the buffer in memory.
646 * Return: 0 on success, negative errno on error.
648 int kexec_add_buffer(struct kexec_buf *kbuf)
651 struct kexec_segment *ksegment;
652 int ret;
654 /* Currently adding segment this way is allowed only in file mode */
655 if (!kbuf->image->file_mode)
656 return -EINVAL;
658 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
659 return -EINVAL;
662 * Make sure we are not trying to add buffer after allocating
663 * control pages. All segments need to be placed first before
664 * any control pages are allocated. As control page allocation
665 * logic goes through list of segments to make sure there are
666 * no destination overlaps.
668 if (!list_empty(&kbuf->image->control_pages)) {
669 WARN_ON(1);
670 return -EINVAL;
673 /* Ensure minimum alignment needed for segments. */
674 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
675 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
677 /* Walk the RAM ranges and allocate a suitable range for the buffer */
678 ret = kexec_locate_mem_hole(kbuf);
679 if (ret)
680 return ret;
682 /* Found a suitable memory range */
683 ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
684 ksegment->kbuf = kbuf->buffer;
685 ksegment->bufsz = kbuf->bufsz;
686 ksegment->mem = kbuf->mem;
687 ksegment->memsz = kbuf->memsz;
688 kbuf->image->nr_segments++;
689 return 0;
692 /* Calculate and store the digest of segments */
693 static int kexec_calculate_store_digests(struct kimage *image)
695 struct crypto_shash *tfm;
696 struct shash_desc *desc;
697 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
698 size_t desc_size, nullsz;
699 char *digest;
700 void *zero_buf;
701 struct kexec_sha_region *sha_regions;
702 struct purgatory_info *pi = &image->purgatory_info;
704 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
705 return 0;
707 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
708 zero_buf_sz = PAGE_SIZE;
710 tfm = crypto_alloc_shash("sha256", 0, 0);
711 if (IS_ERR(tfm)) {
712 ret = PTR_ERR(tfm);
713 goto out;
716 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
717 desc = kzalloc(desc_size, GFP_KERNEL);
718 if (!desc) {
719 ret = -ENOMEM;
720 goto out_free_tfm;
723 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
724 sha_regions = vzalloc(sha_region_sz);
725 if (!sha_regions)
726 goto out_free_desc;
728 desc->tfm = tfm;
730 ret = crypto_shash_init(desc);
731 if (ret < 0)
732 goto out_free_sha_regions;
734 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
735 if (!digest) {
736 ret = -ENOMEM;
737 goto out_free_sha_regions;
740 for (j = i = 0; i < image->nr_segments; i++) {
741 struct kexec_segment *ksegment;
743 ksegment = &image->segment[i];
745 * Skip purgatory as it will be modified once we put digest
746 * info in purgatory.
748 if (ksegment->kbuf == pi->purgatory_buf)
749 continue;
751 ret = crypto_shash_update(desc, ksegment->kbuf,
752 ksegment->bufsz);
753 if (ret)
754 break;
757 * Assume rest of the buffer is filled with zero and
758 * update digest accordingly.
760 nullsz = ksegment->memsz - ksegment->bufsz;
761 while (nullsz) {
762 unsigned long bytes = nullsz;
764 if (bytes > zero_buf_sz)
765 bytes = zero_buf_sz;
766 ret = crypto_shash_update(desc, zero_buf, bytes);
767 if (ret)
768 break;
769 nullsz -= bytes;
772 if (ret)
773 break;
775 sha_regions[j].start = ksegment->mem;
776 sha_regions[j].len = ksegment->memsz;
777 j++;
780 if (!ret) {
781 ret = crypto_shash_final(desc, digest);
782 if (ret)
783 goto out_free_digest;
784 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
785 sha_regions, sha_region_sz, 0);
786 if (ret)
787 goto out_free_digest;
789 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
790 digest, SHA256_DIGEST_SIZE, 0);
791 if (ret)
792 goto out_free_digest;
795 out_free_digest:
796 kfree(digest);
797 out_free_sha_regions:
798 vfree(sha_regions);
799 out_free_desc:
800 kfree(desc);
801 out_free_tfm:
802 kfree(tfm);
803 out:
804 return ret;
807 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
809 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
810 * @pi: Purgatory to be loaded.
811 * @kbuf: Buffer to setup.
813 * Allocates the memory needed for the buffer. Caller is responsible to free
814 * the memory after use.
816 * Return: 0 on success, negative errno on error.
818 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
819 struct kexec_buf *kbuf)
821 const Elf_Shdr *sechdrs;
822 unsigned long bss_align;
823 unsigned long bss_sz;
824 unsigned long align;
825 int i, ret;
827 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
828 kbuf->buf_align = bss_align = 1;
829 kbuf->bufsz = bss_sz = 0;
831 for (i = 0; i < pi->ehdr->e_shnum; i++) {
832 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
833 continue;
835 align = sechdrs[i].sh_addralign;
836 if (sechdrs[i].sh_type != SHT_NOBITS) {
837 if (kbuf->buf_align < align)
838 kbuf->buf_align = align;
839 kbuf->bufsz = ALIGN(kbuf->bufsz, align);
840 kbuf->bufsz += sechdrs[i].sh_size;
841 } else {
842 if (bss_align < align)
843 bss_align = align;
844 bss_sz = ALIGN(bss_sz, align);
845 bss_sz += sechdrs[i].sh_size;
848 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
849 kbuf->memsz = kbuf->bufsz + bss_sz;
850 if (kbuf->buf_align < bss_align)
851 kbuf->buf_align = bss_align;
853 kbuf->buffer = vzalloc(kbuf->bufsz);
854 if (!kbuf->buffer)
855 return -ENOMEM;
856 pi->purgatory_buf = kbuf->buffer;
858 ret = kexec_add_buffer(kbuf);
859 if (ret)
860 goto out;
862 return 0;
863 out:
864 vfree(pi->purgatory_buf);
865 pi->purgatory_buf = NULL;
866 return ret;
870 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
871 * @pi: Purgatory to be loaded.
872 * @kbuf: Buffer prepared to store purgatory.
874 * Allocates the memory needed for the buffer. Caller is responsible to free
875 * the memory after use.
877 * Return: 0 on success, negative errno on error.
879 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
880 struct kexec_buf *kbuf)
882 unsigned long bss_addr;
883 unsigned long offset;
884 Elf_Shdr *sechdrs;
885 int i;
888 * The section headers in kexec_purgatory are read-only. In order to
889 * have them modifiable make a temporary copy.
891 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
892 if (!sechdrs)
893 return -ENOMEM;
894 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
895 pi->ehdr->e_shnum * sizeof(Elf_Shdr));
896 pi->sechdrs = sechdrs;
898 offset = 0;
899 bss_addr = kbuf->mem + kbuf->bufsz;
900 kbuf->image->start = pi->ehdr->e_entry;
902 for (i = 0; i < pi->ehdr->e_shnum; i++) {
903 unsigned long align;
904 void *src, *dst;
906 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
907 continue;
909 align = sechdrs[i].sh_addralign;
910 if (sechdrs[i].sh_type == SHT_NOBITS) {
911 bss_addr = ALIGN(bss_addr, align);
912 sechdrs[i].sh_addr = bss_addr;
913 bss_addr += sechdrs[i].sh_size;
914 continue;
917 offset = ALIGN(offset, align);
918 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
919 pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
920 pi->ehdr->e_entry < (sechdrs[i].sh_addr
921 + sechdrs[i].sh_size)) {
922 kbuf->image->start -= sechdrs[i].sh_addr;
923 kbuf->image->start += kbuf->mem + offset;
926 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
927 dst = pi->purgatory_buf + offset;
928 memcpy(dst, src, sechdrs[i].sh_size);
930 sechdrs[i].sh_addr = kbuf->mem + offset;
931 sechdrs[i].sh_offset = offset;
932 offset += sechdrs[i].sh_size;
935 return 0;
938 static int kexec_apply_relocations(struct kimage *image)
940 int i, ret;
941 struct purgatory_info *pi = &image->purgatory_info;
942 const Elf_Shdr *sechdrs;
944 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
946 for (i = 0; i < pi->ehdr->e_shnum; i++) {
947 const Elf_Shdr *relsec;
948 const Elf_Shdr *symtab;
949 Elf_Shdr *section;
951 relsec = sechdrs + i;
953 if (relsec->sh_type != SHT_RELA &&
954 relsec->sh_type != SHT_REL)
955 continue;
958 * For section of type SHT_RELA/SHT_REL,
959 * ->sh_link contains section header index of associated
960 * symbol table. And ->sh_info contains section header
961 * index of section to which relocations apply.
963 if (relsec->sh_info >= pi->ehdr->e_shnum ||
964 relsec->sh_link >= pi->ehdr->e_shnum)
965 return -ENOEXEC;
967 section = pi->sechdrs + relsec->sh_info;
968 symtab = sechdrs + relsec->sh_link;
970 if (!(section->sh_flags & SHF_ALLOC))
971 continue;
974 * symtab->sh_link contain section header index of associated
975 * string table.
977 if (symtab->sh_link >= pi->ehdr->e_shnum)
978 /* Invalid section number? */
979 continue;
982 * Respective architecture needs to provide support for applying
983 * relocations of type SHT_RELA/SHT_REL.
985 if (relsec->sh_type == SHT_RELA)
986 ret = arch_kexec_apply_relocations_add(pi, section,
987 relsec, symtab);
988 else if (relsec->sh_type == SHT_REL)
989 ret = arch_kexec_apply_relocations(pi, section,
990 relsec, symtab);
991 if (ret)
992 return ret;
995 return 0;
999 * kexec_load_purgatory - Load and relocate the purgatory object.
1000 * @image: Image to add the purgatory to.
1001 * @kbuf: Memory parameters to use.
1003 * Allocates the memory needed for image->purgatory_info.sechdrs and
1004 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1005 * to free the memory after use.
1007 * Return: 0 on success, negative errno on error.
1009 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1011 struct purgatory_info *pi = &image->purgatory_info;
1012 int ret;
1014 if (kexec_purgatory_size <= 0)
1015 return -EINVAL;
1017 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1019 ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1020 if (ret)
1021 return ret;
1023 ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1024 if (ret)
1025 goto out_free_kbuf;
1027 ret = kexec_apply_relocations(image);
1028 if (ret)
1029 goto out;
1031 return 0;
1032 out:
1033 vfree(pi->sechdrs);
1034 pi->sechdrs = NULL;
1035 out_free_kbuf:
1036 vfree(pi->purgatory_buf);
1037 pi->purgatory_buf = NULL;
1038 return ret;
1042 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1043 * @pi: Purgatory to search in.
1044 * @name: Name of the symbol.
1046 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1048 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1049 const char *name)
1051 const Elf_Shdr *sechdrs;
1052 const Elf_Ehdr *ehdr;
1053 const Elf_Sym *syms;
1054 const char *strtab;
1055 int i, k;
1057 if (!pi->ehdr)
1058 return NULL;
1060 ehdr = pi->ehdr;
1061 sechdrs = (void *)ehdr + ehdr->e_shoff;
1063 for (i = 0; i < ehdr->e_shnum; i++) {
1064 if (sechdrs[i].sh_type != SHT_SYMTAB)
1065 continue;
1067 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1068 /* Invalid strtab section number */
1069 continue;
1070 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1071 syms = (void *)ehdr + sechdrs[i].sh_offset;
1073 /* Go through symbols for a match */
1074 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1075 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1076 continue;
1078 if (strcmp(strtab + syms[k].st_name, name) != 0)
1079 continue;
1081 if (syms[k].st_shndx == SHN_UNDEF ||
1082 syms[k].st_shndx >= ehdr->e_shnum) {
1083 pr_debug("Symbol: %s has bad section index %d.\n",
1084 name, syms[k].st_shndx);
1085 return NULL;
1088 /* Found the symbol we are looking for */
1089 return &syms[k];
1093 return NULL;
1096 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1098 struct purgatory_info *pi = &image->purgatory_info;
1099 const Elf_Sym *sym;
1100 Elf_Shdr *sechdr;
1102 sym = kexec_purgatory_find_symbol(pi, name);
1103 if (!sym)
1104 return ERR_PTR(-EINVAL);
1106 sechdr = &pi->sechdrs[sym->st_shndx];
1109 * Returns the address where symbol will finally be loaded after
1110 * kexec_load_segment()
1112 return (void *)(sechdr->sh_addr + sym->st_value);
1116 * Get or set value of a symbol. If "get_value" is true, symbol value is
1117 * returned in buf otherwise symbol value is set based on value in buf.
1119 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1120 void *buf, unsigned int size, bool get_value)
1122 struct purgatory_info *pi = &image->purgatory_info;
1123 const Elf_Sym *sym;
1124 Elf_Shdr *sec;
1125 char *sym_buf;
1127 sym = kexec_purgatory_find_symbol(pi, name);
1128 if (!sym)
1129 return -EINVAL;
1131 if (sym->st_size != size) {
1132 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1133 name, (unsigned long)sym->st_size, size);
1134 return -EINVAL;
1137 sec = pi->sechdrs + sym->st_shndx;
1139 if (sec->sh_type == SHT_NOBITS) {
1140 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1141 get_value ? "get" : "set");
1142 return -EINVAL;
1145 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1147 if (get_value)
1148 memcpy((void *)buf, sym_buf, size);
1149 else
1150 memcpy((void *)sym_buf, buf, size);
1152 return 0;
1154 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1156 int crash_exclude_mem_range(struct crash_mem *mem,
1157 unsigned long long mstart, unsigned long long mend)
1159 int i, j;
1160 unsigned long long start, end;
1161 struct crash_mem_range temp_range = {0, 0};
1163 for (i = 0; i < mem->nr_ranges; i++) {
1164 start = mem->ranges[i].start;
1165 end = mem->ranges[i].end;
1167 if (mstart > end || mend < start)
1168 continue;
1170 /* Truncate any area outside of range */
1171 if (mstart < start)
1172 mstart = start;
1173 if (mend > end)
1174 mend = end;
1176 /* Found completely overlapping range */
1177 if (mstart == start && mend == end) {
1178 mem->ranges[i].start = 0;
1179 mem->ranges[i].end = 0;
1180 if (i < mem->nr_ranges - 1) {
1181 /* Shift rest of the ranges to left */
1182 for (j = i; j < mem->nr_ranges - 1; j++) {
1183 mem->ranges[j].start =
1184 mem->ranges[j+1].start;
1185 mem->ranges[j].end =
1186 mem->ranges[j+1].end;
1189 mem->nr_ranges--;
1190 return 0;
1193 if (mstart > start && mend < end) {
1194 /* Split original range */
1195 mem->ranges[i].end = mstart - 1;
1196 temp_range.start = mend + 1;
1197 temp_range.end = end;
1198 } else if (mstart != start)
1199 mem->ranges[i].end = mstart - 1;
1200 else
1201 mem->ranges[i].start = mend + 1;
1202 break;
1205 /* If a split happened, add the split to array */
1206 if (!temp_range.end)
1207 return 0;
1209 /* Split happened */
1210 if (i == mem->max_nr_ranges - 1)
1211 return -ENOMEM;
1213 /* Location where new range should go */
1214 j = i + 1;
1215 if (j < mem->nr_ranges) {
1216 /* Move over all ranges one slot towards the end */
1217 for (i = mem->nr_ranges - 1; i >= j; i--)
1218 mem->ranges[i + 1] = mem->ranges[i];
1221 mem->ranges[j].start = temp_range.start;
1222 mem->ranges[j].end = temp_range.end;
1223 mem->nr_ranges++;
1224 return 0;
1227 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1228 void **addr, unsigned long *sz)
1230 Elf64_Ehdr *ehdr;
1231 Elf64_Phdr *phdr;
1232 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1233 unsigned char *buf;
1234 unsigned int cpu, i;
1235 unsigned long long notes_addr;
1236 unsigned long mstart, mend;
1238 /* extra phdr for vmcoreinfo elf note */
1239 nr_phdr = nr_cpus + 1;
1240 nr_phdr += mem->nr_ranges;
1243 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1244 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1245 * I think this is required by tools like gdb. So same physical
1246 * memory will be mapped in two elf headers. One will contain kernel
1247 * text virtual addresses and other will have __va(physical) addresses.
1250 nr_phdr++;
1251 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1252 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1254 buf = vzalloc(elf_sz);
1255 if (!buf)
1256 return -ENOMEM;
1258 ehdr = (Elf64_Ehdr *)buf;
1259 phdr = (Elf64_Phdr *)(ehdr + 1);
1260 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1261 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1262 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1263 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1264 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1265 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1266 ehdr->e_type = ET_CORE;
1267 ehdr->e_machine = ELF_ARCH;
1268 ehdr->e_version = EV_CURRENT;
1269 ehdr->e_phoff = sizeof(Elf64_Ehdr);
1270 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1271 ehdr->e_phentsize = sizeof(Elf64_Phdr);
1273 /* Prepare one phdr of type PT_NOTE for each present cpu */
1274 for_each_present_cpu(cpu) {
1275 phdr->p_type = PT_NOTE;
1276 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1277 phdr->p_offset = phdr->p_paddr = notes_addr;
1278 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1279 (ehdr->e_phnum)++;
1280 phdr++;
1283 /* Prepare one PT_NOTE header for vmcoreinfo */
1284 phdr->p_type = PT_NOTE;
1285 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1286 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1287 (ehdr->e_phnum)++;
1288 phdr++;
1290 /* Prepare PT_LOAD type program header for kernel text region */
1291 if (kernel_map) {
1292 phdr->p_type = PT_LOAD;
1293 phdr->p_flags = PF_R|PF_W|PF_X;
1294 phdr->p_vaddr = (unsigned long) _text;
1295 phdr->p_filesz = phdr->p_memsz = _end - _text;
1296 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1297 ehdr->e_phnum++;
1298 phdr++;
1301 /* Go through all the ranges in mem->ranges[] and prepare phdr */
1302 for (i = 0; i < mem->nr_ranges; i++) {
1303 mstart = mem->ranges[i].start;
1304 mend = mem->ranges[i].end;
1306 phdr->p_type = PT_LOAD;
1307 phdr->p_flags = PF_R|PF_W|PF_X;
1308 phdr->p_offset = mstart;
1310 phdr->p_paddr = mstart;
1311 phdr->p_vaddr = (unsigned long) __va(mstart);
1312 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1313 phdr->p_align = 0;
1314 ehdr->e_phnum++;
1315 phdr++;
1316 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",
1317 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1318 ehdr->e_phnum, phdr->p_offset);
1321 *addr = buf;
1322 *sz = elf_sz;
1323 return 0;