1 // SPDX-License-Identifier: GPL-2.0-only
3 * kexec: kexec_file_load system call
5 * Copyright (C) 2014 Red Hat Inc.
7 * Vivek Goyal <vgoyal@redhat.com>
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/capability.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>
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;
44 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
45 ret = (*fops)->probe(buf, buf_len);
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)
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");
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);
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.
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");
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.
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");
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;
168 #ifdef CONFIG_IMA_KEXEC
169 vfree(image->ima_buffer);
170 image->ima_buffer = NULL;
171 #endif /* CONFIG_IMA_KEXEC */
173 /* See if architecture has anything to cleanup post load */
174 arch_kimage_file_post_load_cleanup(image);
177 * Above call should have called into bootloader to free up
178 * any data stored in kimage->image_loader_data. It should
179 * be ok now to free it up.
181 kfree(image->image_loader_data);
182 image->image_loader_data = NULL;
185 #ifdef CONFIG_KEXEC_SIG
187 kimage_validate_signature(struct kimage *image)
192 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
193 image->kernel_buf_len);
198 /* Certain verification errors are non-fatal if we're not
199 * checking errors, provided we aren't mandating that there
200 * must be a valid signature.
203 reason = "kexec of unsigned image";
206 reason = "kexec of image with unsupported crypto";
209 reason = "kexec of image with unavailable key";
211 if (IS_ENABLED(CONFIG_KEXEC_SIG_FORCE)) {
212 pr_notice("%s rejected\n", reason);
216 /* If IMA is guaranteed to appraise a signature on the kexec
217 * image, permit it even if the kernel is otherwise locked
220 if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
221 security_locked_down(LOCKDOWN_KEXEC))
226 /* All other errors are fatal, including nomem, unparseable
227 * signatures and signature check failures - even if signatures
231 pr_notice("kernel signature verification failed (%d).\n", ret);
239 * In file mode list of segments is prepared by kernel. Copy relevant
240 * data from user space, do error checking, prepare segment list
243 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
244 const char __user *cmdline_ptr,
245 unsigned long cmdline_len, unsigned flags)
251 ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf,
252 &size, INT_MAX, READING_KEXEC_IMAGE);
255 image->kernel_buf_len = size;
257 /* Call arch image probe handlers */
258 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
259 image->kernel_buf_len);
263 #ifdef CONFIG_KEXEC_SIG
264 ret = kimage_validate_signature(image);
269 /* It is possible that there no initramfs is being loaded */
270 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
271 ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf,
273 READING_KEXEC_INITRAMFS);
276 image->initrd_buf_len = size;
280 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
281 if (IS_ERR(image->cmdline_buf)) {
282 ret = PTR_ERR(image->cmdline_buf);
283 image->cmdline_buf = NULL;
287 image->cmdline_buf_len = cmdline_len;
289 /* command line should be a string with last byte null */
290 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
295 ima_kexec_cmdline(image->cmdline_buf,
296 image->cmdline_buf_len - 1);
299 /* IMA needs to pass the measurement list to the next kernel. */
300 ima_add_kexec_buffer(image);
302 /* Call arch image load handlers */
303 ldata = arch_kexec_kernel_image_load(image);
306 ret = PTR_ERR(ldata);
310 image->image_loader_data = ldata;
312 /* In case of error, free up all allocated memory in this function */
314 kimage_file_post_load_cleanup(image);
319 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
320 int initrd_fd, const char __user *cmdline_ptr,
321 unsigned long cmdline_len, unsigned long flags)
324 struct kimage *image;
325 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
327 image = do_kimage_alloc_init();
331 image->file_mode = 1;
333 if (kexec_on_panic) {
334 /* Enable special crash kernel control page alloc policy. */
335 image->control_page = crashk_res.start;
336 image->type = KEXEC_TYPE_CRASH;
339 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
340 cmdline_ptr, cmdline_len, flags);
344 ret = sanity_check_segment_list(image);
346 goto out_free_post_load_bufs;
349 image->control_code_page = kimage_alloc_control_pages(image,
350 get_order(KEXEC_CONTROL_PAGE_SIZE));
351 if (!image->control_code_page) {
352 pr_err("Could not allocate control_code_buffer\n");
353 goto out_free_post_load_bufs;
356 if (!kexec_on_panic) {
357 image->swap_page = kimage_alloc_control_pages(image, 0);
358 if (!image->swap_page) {
359 pr_err("Could not allocate swap buffer\n");
360 goto out_free_control_pages;
366 out_free_control_pages:
367 kimage_free_page_list(&image->control_pages);
368 out_free_post_load_bufs:
369 kimage_file_post_load_cleanup(image);
375 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
376 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
377 unsigned long, flags)
380 struct kimage **dest_image, *image;
382 /* We only trust the superuser with rebooting the system. */
383 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
386 /* Make sure we have a legal set of flags */
387 if (flags != (flags & KEXEC_FILE_FLAGS))
392 if (!mutex_trylock(&kexec_mutex))
395 dest_image = &kexec_image;
396 if (flags & KEXEC_FILE_ON_CRASH) {
397 dest_image = &kexec_crash_image;
398 if (kexec_crash_image)
399 arch_kexec_unprotect_crashkres();
402 if (flags & KEXEC_FILE_UNLOAD)
406 * In case of crash, new kernel gets loaded in reserved region. It is
407 * same memory where old crash kernel might be loaded. Free any
408 * current crash dump kernel before we corrupt it.
410 if (flags & KEXEC_FILE_ON_CRASH)
411 kimage_free(xchg(&kexec_crash_image, NULL));
413 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
418 ret = machine_kexec_prepare(image);
423 * Some architecture(like S390) may touch the crash memory before
424 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
426 ret = kimage_crash_copy_vmcoreinfo(image);
430 ret = kexec_calculate_store_digests(image);
434 for (i = 0; i < image->nr_segments; i++) {
435 struct kexec_segment *ksegment;
437 ksegment = &image->segment[i];
438 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
439 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
442 ret = kimage_load_segment(image, &image->segment[i]);
447 kimage_terminate(image);
450 * Free up any temporary buffers allocated which are not needed
451 * after image has been loaded
453 kimage_file_post_load_cleanup(image);
455 image = xchg(dest_image, image);
457 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
458 arch_kexec_protect_crashkres();
460 mutex_unlock(&kexec_mutex);
465 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
466 struct kexec_buf *kbuf)
468 struct kimage *image = kbuf->image;
469 unsigned long temp_start, temp_end;
471 temp_end = min(end, kbuf->buf_max);
472 temp_start = temp_end - kbuf->memsz;
475 /* align down start */
476 temp_start = temp_start & (~(kbuf->buf_align - 1));
478 if (temp_start < start || temp_start < kbuf->buf_min)
481 temp_end = temp_start + kbuf->memsz - 1;
484 * Make sure this does not conflict with any of existing
487 if (kimage_is_destination_range(image, temp_start, temp_end)) {
488 temp_start = temp_start - PAGE_SIZE;
492 /* We found a suitable memory range */
496 /* If we are here, we found a suitable memory range */
497 kbuf->mem = temp_start;
499 /* Success, stop navigating through remaining System RAM ranges */
503 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
504 struct kexec_buf *kbuf)
506 struct kimage *image = kbuf->image;
507 unsigned long temp_start, temp_end;
509 temp_start = max(start, kbuf->buf_min);
512 temp_start = ALIGN(temp_start, kbuf->buf_align);
513 temp_end = temp_start + kbuf->memsz - 1;
515 if (temp_end > end || temp_end > kbuf->buf_max)
518 * Make sure this does not conflict with any of existing
521 if (kimage_is_destination_range(image, temp_start, temp_end)) {
522 temp_start = temp_start + PAGE_SIZE;
526 /* We found a suitable memory range */
530 /* If we are here, we found a suitable memory range */
531 kbuf->mem = temp_start;
533 /* Success, stop navigating through remaining System RAM ranges */
537 static int locate_mem_hole_callback(struct resource *res, void *arg)
539 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
540 u64 start = res->start, end = res->end;
541 unsigned long sz = end - start + 1;
543 /* Returning 0 will take to next memory range */
544 if (sz < kbuf->memsz)
547 if (end < kbuf->buf_min || start > kbuf->buf_max)
551 * Allocate memory top down with-in ram range. Otherwise bottom up
555 return locate_mem_hole_top_down(start, end, kbuf);
556 return locate_mem_hole_bottom_up(start, end, kbuf);
559 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
560 static int kexec_walk_memblock(struct kexec_buf *kbuf,
561 int (*func)(struct resource *, void *))
565 phys_addr_t mstart, mend;
566 struct resource res = { };
568 if (kbuf->image->type == KEXEC_TYPE_CRASH)
569 return func(&crashk_res, kbuf);
571 if (kbuf->top_down) {
572 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
573 &mstart, &mend, NULL) {
575 * In memblock, end points to the first byte after the
576 * range while in kexec, end points to the last byte
581 ret = func(&res, kbuf);
586 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
587 &mstart, &mend, NULL) {
589 * In memblock, end points to the first byte after the
590 * range while in kexec, end points to the last byte
595 ret = func(&res, kbuf);
604 static int kexec_walk_memblock(struct kexec_buf *kbuf,
605 int (*func)(struct resource *, void *))
612 * kexec_walk_resources - call func(data) on free memory regions
613 * @kbuf: Context info for the search. Also passed to @func.
614 * @func: Function to call for each memory region.
616 * Return: The memory walk will stop when func returns a non-zero value
617 * and that value will be returned. If all free regions are visited without
618 * func returning non-zero, then zero will be returned.
620 static int kexec_walk_resources(struct kexec_buf *kbuf,
621 int (*func)(struct resource *, void *))
623 if (kbuf->image->type == KEXEC_TYPE_CRASH)
624 return walk_iomem_res_desc(crashk_res.desc,
625 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
626 crashk_res.start, crashk_res.end,
629 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
633 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
634 * @kbuf: Parameters for the memory search.
636 * On success, kbuf->mem will have the start address of the memory region found.
638 * Return: 0 on success, negative errno on error.
640 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
644 /* Arch knows where to place */
645 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
648 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
649 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
651 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
653 return ret == 1 ? 0 : -EADDRNOTAVAIL;
657 * kexec_add_buffer - place a buffer in a kexec segment
658 * @kbuf: Buffer contents and memory parameters.
660 * This function assumes that kexec_mutex is held.
661 * On successful return, @kbuf->mem will have the physical address of
662 * the buffer in memory.
664 * Return: 0 on success, negative errno on error.
666 int kexec_add_buffer(struct kexec_buf *kbuf)
669 struct kexec_segment *ksegment;
672 /* Currently adding segment this way is allowed only in file mode */
673 if (!kbuf->image->file_mode)
676 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
680 * Make sure we are not trying to add buffer after allocating
681 * control pages. All segments need to be placed first before
682 * any control pages are allocated. As control page allocation
683 * logic goes through list of segments to make sure there are
684 * no destination overlaps.
686 if (!list_empty(&kbuf->image->control_pages)) {
691 /* Ensure minimum alignment needed for segments. */
692 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
693 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
695 /* Walk the RAM ranges and allocate a suitable range for the buffer */
696 ret = kexec_locate_mem_hole(kbuf);
700 /* Found a suitable memory range */
701 ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
702 ksegment->kbuf = kbuf->buffer;
703 ksegment->bufsz = kbuf->bufsz;
704 ksegment->mem = kbuf->mem;
705 ksegment->memsz = kbuf->memsz;
706 kbuf->image->nr_segments++;
710 /* Calculate and store the digest of segments */
711 static int kexec_calculate_store_digests(struct kimage *image)
713 struct crypto_shash *tfm;
714 struct shash_desc *desc;
715 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
716 size_t desc_size, nullsz;
719 struct kexec_sha_region *sha_regions;
720 struct purgatory_info *pi = &image->purgatory_info;
722 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
725 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
726 zero_buf_sz = PAGE_SIZE;
728 tfm = crypto_alloc_shash("sha256", 0, 0);
734 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
735 desc = kzalloc(desc_size, GFP_KERNEL);
741 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
742 sha_regions = vzalloc(sha_region_sz);
750 ret = crypto_shash_init(desc);
752 goto out_free_sha_regions;
754 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
757 goto out_free_sha_regions;
760 for (j = i = 0; i < image->nr_segments; i++) {
761 struct kexec_segment *ksegment;
763 ksegment = &image->segment[i];
765 * Skip purgatory as it will be modified once we put digest
768 if (ksegment->kbuf == pi->purgatory_buf)
771 ret = crypto_shash_update(desc, ksegment->kbuf,
777 * Assume rest of the buffer is filled with zero and
778 * update digest accordingly.
780 nullsz = ksegment->memsz - ksegment->bufsz;
782 unsigned long bytes = nullsz;
784 if (bytes > zero_buf_sz)
786 ret = crypto_shash_update(desc, zero_buf, bytes);
795 sha_regions[j].start = ksegment->mem;
796 sha_regions[j].len = ksegment->memsz;
801 ret = crypto_shash_final(desc, digest);
803 goto out_free_digest;
804 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
805 sha_regions, sha_region_sz, 0);
807 goto out_free_digest;
809 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
810 digest, SHA256_DIGEST_SIZE, 0);
812 goto out_free_digest;
817 out_free_sha_regions:
827 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
829 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
830 * @pi: Purgatory to be loaded.
831 * @kbuf: Buffer to setup.
833 * Allocates the memory needed for the buffer. Caller is responsible to free
834 * the memory after use.
836 * Return: 0 on success, negative errno on error.
838 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
839 struct kexec_buf *kbuf)
841 const Elf_Shdr *sechdrs;
842 unsigned long bss_align;
843 unsigned long bss_sz;
847 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
848 kbuf->buf_align = bss_align = 1;
849 kbuf->bufsz = bss_sz = 0;
851 for (i = 0; i < pi->ehdr->e_shnum; i++) {
852 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
855 align = sechdrs[i].sh_addralign;
856 if (sechdrs[i].sh_type != SHT_NOBITS) {
857 if (kbuf->buf_align < align)
858 kbuf->buf_align = align;
859 kbuf->bufsz = ALIGN(kbuf->bufsz, align);
860 kbuf->bufsz += sechdrs[i].sh_size;
862 if (bss_align < align)
864 bss_sz = ALIGN(bss_sz, align);
865 bss_sz += sechdrs[i].sh_size;
868 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
869 kbuf->memsz = kbuf->bufsz + bss_sz;
870 if (kbuf->buf_align < bss_align)
871 kbuf->buf_align = bss_align;
873 kbuf->buffer = vzalloc(kbuf->bufsz);
876 pi->purgatory_buf = kbuf->buffer;
878 ret = kexec_add_buffer(kbuf);
884 vfree(pi->purgatory_buf);
885 pi->purgatory_buf = NULL;
890 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
891 * @pi: Purgatory to be loaded.
892 * @kbuf: Buffer prepared to store purgatory.
894 * Allocates the memory needed for the buffer. Caller is responsible to free
895 * the memory after use.
897 * Return: 0 on success, negative errno on error.
899 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
900 struct kexec_buf *kbuf)
902 unsigned long bss_addr;
903 unsigned long offset;
908 * The section headers in kexec_purgatory are read-only. In order to
909 * have them modifiable make a temporary copy.
911 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
914 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
915 pi->ehdr->e_shnum * sizeof(Elf_Shdr));
916 pi->sechdrs = sechdrs;
919 bss_addr = kbuf->mem + kbuf->bufsz;
920 kbuf->image->start = pi->ehdr->e_entry;
922 for (i = 0; i < pi->ehdr->e_shnum; i++) {
926 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
929 align = sechdrs[i].sh_addralign;
930 if (sechdrs[i].sh_type == SHT_NOBITS) {
931 bss_addr = ALIGN(bss_addr, align);
932 sechdrs[i].sh_addr = bss_addr;
933 bss_addr += sechdrs[i].sh_size;
937 offset = ALIGN(offset, align);
938 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
939 pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
940 pi->ehdr->e_entry < (sechdrs[i].sh_addr
941 + sechdrs[i].sh_size)) {
942 kbuf->image->start -= sechdrs[i].sh_addr;
943 kbuf->image->start += kbuf->mem + offset;
946 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
947 dst = pi->purgatory_buf + offset;
948 memcpy(dst, src, sechdrs[i].sh_size);
950 sechdrs[i].sh_addr = kbuf->mem + offset;
951 sechdrs[i].sh_offset = offset;
952 offset += sechdrs[i].sh_size;
958 static int kexec_apply_relocations(struct kimage *image)
961 struct purgatory_info *pi = &image->purgatory_info;
962 const Elf_Shdr *sechdrs;
964 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
966 for (i = 0; i < pi->ehdr->e_shnum; i++) {
967 const Elf_Shdr *relsec;
968 const Elf_Shdr *symtab;
971 relsec = sechdrs + i;
973 if (relsec->sh_type != SHT_RELA &&
974 relsec->sh_type != SHT_REL)
978 * For section of type SHT_RELA/SHT_REL,
979 * ->sh_link contains section header index of associated
980 * symbol table. And ->sh_info contains section header
981 * index of section to which relocations apply.
983 if (relsec->sh_info >= pi->ehdr->e_shnum ||
984 relsec->sh_link >= pi->ehdr->e_shnum)
987 section = pi->sechdrs + relsec->sh_info;
988 symtab = sechdrs + relsec->sh_link;
990 if (!(section->sh_flags & SHF_ALLOC))
994 * symtab->sh_link contain section header index of associated
997 if (symtab->sh_link >= pi->ehdr->e_shnum)
998 /* Invalid section number? */
1002 * Respective architecture needs to provide support for applying
1003 * relocations of type SHT_RELA/SHT_REL.
1005 if (relsec->sh_type == SHT_RELA)
1006 ret = arch_kexec_apply_relocations_add(pi, section,
1008 else if (relsec->sh_type == SHT_REL)
1009 ret = arch_kexec_apply_relocations(pi, section,
1019 * kexec_load_purgatory - Load and relocate the purgatory object.
1020 * @image: Image to add the purgatory to.
1021 * @kbuf: Memory parameters to use.
1023 * Allocates the memory needed for image->purgatory_info.sechdrs and
1024 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1025 * to free the memory after use.
1027 * Return: 0 on success, negative errno on error.
1029 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1031 struct purgatory_info *pi = &image->purgatory_info;
1034 if (kexec_purgatory_size <= 0)
1037 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1039 ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1043 ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1047 ret = kexec_apply_relocations(image);
1056 vfree(pi->purgatory_buf);
1057 pi->purgatory_buf = NULL;
1062 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1063 * @pi: Purgatory to search in.
1064 * @name: Name of the symbol.
1066 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1068 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1071 const Elf_Shdr *sechdrs;
1072 const Elf_Ehdr *ehdr;
1073 const Elf_Sym *syms;
1081 sechdrs = (void *)ehdr + ehdr->e_shoff;
1083 for (i = 0; i < ehdr->e_shnum; i++) {
1084 if (sechdrs[i].sh_type != SHT_SYMTAB)
1087 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1088 /* Invalid strtab section number */
1090 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1091 syms = (void *)ehdr + sechdrs[i].sh_offset;
1093 /* Go through symbols for a match */
1094 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1095 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1098 if (strcmp(strtab + syms[k].st_name, name) != 0)
1101 if (syms[k].st_shndx == SHN_UNDEF ||
1102 syms[k].st_shndx >= ehdr->e_shnum) {
1103 pr_debug("Symbol: %s has bad section index %d.\n",
1104 name, syms[k].st_shndx);
1108 /* Found the symbol we are looking for */
1116 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1118 struct purgatory_info *pi = &image->purgatory_info;
1122 sym = kexec_purgatory_find_symbol(pi, name);
1124 return ERR_PTR(-EINVAL);
1126 sechdr = &pi->sechdrs[sym->st_shndx];
1129 * Returns the address where symbol will finally be loaded after
1130 * kexec_load_segment()
1132 return (void *)(sechdr->sh_addr + sym->st_value);
1136 * Get or set value of a symbol. If "get_value" is true, symbol value is
1137 * returned in buf otherwise symbol value is set based on value in buf.
1139 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1140 void *buf, unsigned int size, bool get_value)
1142 struct purgatory_info *pi = &image->purgatory_info;
1147 sym = kexec_purgatory_find_symbol(pi, name);
1151 if (sym->st_size != size) {
1152 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1153 name, (unsigned long)sym->st_size, size);
1157 sec = pi->sechdrs + sym->st_shndx;
1159 if (sec->sh_type == SHT_NOBITS) {
1160 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1161 get_value ? "get" : "set");
1165 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1168 memcpy((void *)buf, sym_buf, size);
1170 memcpy((void *)sym_buf, buf, size);
1174 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1176 int crash_exclude_mem_range(struct crash_mem *mem,
1177 unsigned long long mstart, unsigned long long mend)
1180 unsigned long long start, end;
1181 struct crash_mem_range temp_range = {0, 0};
1183 for (i = 0; i < mem->nr_ranges; i++) {
1184 start = mem->ranges[i].start;
1185 end = mem->ranges[i].end;
1187 if (mstart > end || mend < start)
1190 /* Truncate any area outside of range */
1196 /* Found completely overlapping range */
1197 if (mstart == start && mend == end) {
1198 mem->ranges[i].start = 0;
1199 mem->ranges[i].end = 0;
1200 if (i < mem->nr_ranges - 1) {
1201 /* Shift rest of the ranges to left */
1202 for (j = i; j < mem->nr_ranges - 1; j++) {
1203 mem->ranges[j].start =
1204 mem->ranges[j+1].start;
1205 mem->ranges[j].end =
1206 mem->ranges[j+1].end;
1213 if (mstart > start && mend < end) {
1214 /* Split original range */
1215 mem->ranges[i].end = mstart - 1;
1216 temp_range.start = mend + 1;
1217 temp_range.end = end;
1218 } else if (mstart != start)
1219 mem->ranges[i].end = mstart - 1;
1221 mem->ranges[i].start = mend + 1;
1225 /* If a split happened, add the split to array */
1226 if (!temp_range.end)
1229 /* Split happened */
1230 if (i == mem->max_nr_ranges - 1)
1233 /* Location where new range should go */
1235 if (j < mem->nr_ranges) {
1236 /* Move over all ranges one slot towards the end */
1237 for (i = mem->nr_ranges - 1; i >= j; i--)
1238 mem->ranges[i + 1] = mem->ranges[i];
1241 mem->ranges[j].start = temp_range.start;
1242 mem->ranges[j].end = temp_range.end;
1247 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1248 void **addr, unsigned long *sz)
1252 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1254 unsigned int cpu, i;
1255 unsigned long long notes_addr;
1256 unsigned long mstart, mend;
1258 /* extra phdr for vmcoreinfo elf note */
1259 nr_phdr = nr_cpus + 1;
1260 nr_phdr += mem->nr_ranges;
1263 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1264 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1265 * I think this is required by tools like gdb. So same physical
1266 * memory will be mapped in two elf headers. One will contain kernel
1267 * text virtual addresses and other will have __va(physical) addresses.
1271 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1272 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1274 buf = vzalloc(elf_sz);
1278 ehdr = (Elf64_Ehdr *)buf;
1279 phdr = (Elf64_Phdr *)(ehdr + 1);
1280 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1281 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1282 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1283 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1284 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1285 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1286 ehdr->e_type = ET_CORE;
1287 ehdr->e_machine = ELF_ARCH;
1288 ehdr->e_version = EV_CURRENT;
1289 ehdr->e_phoff = sizeof(Elf64_Ehdr);
1290 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1291 ehdr->e_phentsize = sizeof(Elf64_Phdr);
1293 /* Prepare one phdr of type PT_NOTE for each present cpu */
1294 for_each_present_cpu(cpu) {
1295 phdr->p_type = PT_NOTE;
1296 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1297 phdr->p_offset = phdr->p_paddr = notes_addr;
1298 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1303 /* Prepare one PT_NOTE header for vmcoreinfo */
1304 phdr->p_type = PT_NOTE;
1305 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1306 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1310 /* Prepare PT_LOAD type program header for kernel text region */
1312 phdr->p_type = PT_LOAD;
1313 phdr->p_flags = PF_R|PF_W|PF_X;
1314 phdr->p_vaddr = (Elf64_Addr)_text;
1315 phdr->p_filesz = phdr->p_memsz = _end - _text;
1316 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1321 /* Go through all the ranges in mem->ranges[] and prepare phdr */
1322 for (i = 0; i < mem->nr_ranges; i++) {
1323 mstart = mem->ranges[i].start;
1324 mend = mem->ranges[i].end;
1326 phdr->p_type = PT_LOAD;
1327 phdr->p_flags = PF_R|PF_W|PF_X;
1328 phdr->p_offset = mstart;
1330 phdr->p_paddr = mstart;
1331 phdr->p_vaddr = (unsigned long long) __va(mstart);
1332 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1336 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",
1337 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1338 ehdr->e_phnum, phdr->p_offset);