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/sha2.h>
24 #include <linux/elf.h>
25 #include <linux/elfcore.h>
26 #include <linux/kernel.h>
27 #include <linux/kernel_read_file.h>
28 #include <linux/syscalls.h>
29 #include <linux/vmalloc.h>
30 #include "kexec_internal.h"
32 static int kexec_calculate_store_digests(struct kimage *image);
35 * Currently this is the only default function that is exported as some
36 * architectures need it to do additional handlings.
37 * In the future, other default functions may be exported too if required.
39 int kexec_image_probe_default(struct kimage *image, void *buf,
40 unsigned long buf_len)
42 const struct kexec_file_ops * const *fops;
45 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
46 ret = (*fops)->probe(buf, buf_len);
56 /* Architectures can provide this probe function */
57 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
58 unsigned long buf_len)
60 return kexec_image_probe_default(image, buf, buf_len);
63 static void *kexec_image_load_default(struct kimage *image)
65 if (!image->fops || !image->fops->load)
66 return ERR_PTR(-ENOEXEC);
68 return image->fops->load(image, image->kernel_buf,
69 image->kernel_buf_len, image->initrd_buf,
70 image->initrd_buf_len, image->cmdline_buf,
71 image->cmdline_buf_len);
74 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
76 return kexec_image_load_default(image);
79 int kexec_image_post_load_cleanup_default(struct kimage *image)
81 if (!image->fops || !image->fops->cleanup)
84 return image->fops->cleanup(image->image_loader_data);
87 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
89 return kexec_image_post_load_cleanup_default(image);
92 #ifdef CONFIG_KEXEC_SIG
93 static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
94 unsigned long buf_len)
96 if (!image->fops || !image->fops->verify_sig) {
97 pr_debug("kernel loader does not support signature verification.\n");
101 return image->fops->verify_sig(buf, buf_len);
104 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
105 unsigned long buf_len)
107 return kexec_image_verify_sig_default(image, buf, buf_len);
112 * arch_kexec_apply_relocations_add - apply relocations of type RELA
113 * @pi: Purgatory to be relocated.
114 * @section: Section relocations applying to.
115 * @relsec: Section containing RELAs.
116 * @symtab: Corresponding symtab.
118 * Return: 0 on success, negative errno on error.
121 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section,
122 const Elf_Shdr *relsec, const Elf_Shdr *symtab)
124 pr_err("RELA relocation unsupported.\n");
129 * arch_kexec_apply_relocations - apply relocations of type REL
130 * @pi: Purgatory to be relocated.
131 * @section: Section relocations applying to.
132 * @relsec: Section containing RELs.
133 * @symtab: Corresponding symtab.
135 * Return: 0 on success, negative errno on error.
138 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section,
139 const Elf_Shdr *relsec, const Elf_Shdr *symtab)
141 pr_err("REL relocation unsupported.\n");
146 * Free up memory used by kernel, initrd, and command line. This is temporary
147 * memory allocation which is not needed any more after these buffers have
148 * been loaded into separate segments and have been copied elsewhere.
150 void kimage_file_post_load_cleanup(struct kimage *image)
152 struct purgatory_info *pi = &image->purgatory_info;
154 vfree(image->kernel_buf);
155 image->kernel_buf = NULL;
157 vfree(image->initrd_buf);
158 image->initrd_buf = NULL;
160 kfree(image->cmdline_buf);
161 image->cmdline_buf = NULL;
163 vfree(pi->purgatory_buf);
164 pi->purgatory_buf = NULL;
169 #ifdef CONFIG_IMA_KEXEC
170 vfree(image->ima_buffer);
171 image->ima_buffer = NULL;
172 #endif /* CONFIG_IMA_KEXEC */
174 /* See if architecture has anything to cleanup post load */
175 arch_kimage_file_post_load_cleanup(image);
178 * Above call should have called into bootloader to free up
179 * any data stored in kimage->image_loader_data. It should
180 * be ok now to free it up.
182 kfree(image->image_loader_data);
183 image->image_loader_data = NULL;
186 #ifdef CONFIG_KEXEC_SIG
188 kimage_validate_signature(struct kimage *image)
192 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
193 image->kernel_buf_len);
196 if (IS_ENABLED(CONFIG_KEXEC_SIG_FORCE)) {
197 pr_notice("Enforced kernel signature verification failed (%d).\n", ret);
202 * If IMA is guaranteed to appraise a signature on the kexec
203 * image, permit it even if the kernel is otherwise locked
206 if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
207 security_locked_down(LOCKDOWN_KEXEC))
210 pr_debug("kernel signature verification failed (%d).\n", ret);
218 * In file mode list of segments is prepared by kernel. Copy relevant
219 * data from user space, do error checking, prepare segment list
222 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
223 const char __user *cmdline_ptr,
224 unsigned long cmdline_len, unsigned flags)
229 ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf,
230 INT_MAX, NULL, READING_KEXEC_IMAGE);
233 image->kernel_buf_len = ret;
235 /* Call arch image probe handlers */
236 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
237 image->kernel_buf_len);
241 #ifdef CONFIG_KEXEC_SIG
242 ret = kimage_validate_signature(image);
247 /* It is possible that there no initramfs is being loaded */
248 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
249 ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf,
251 READING_KEXEC_INITRAMFS);
254 image->initrd_buf_len = ret;
259 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
260 if (IS_ERR(image->cmdline_buf)) {
261 ret = PTR_ERR(image->cmdline_buf);
262 image->cmdline_buf = NULL;
266 image->cmdline_buf_len = cmdline_len;
268 /* command line should be a string with last byte null */
269 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
274 ima_kexec_cmdline(kernel_fd, image->cmdline_buf,
275 image->cmdline_buf_len - 1);
278 /* IMA needs to pass the measurement list to the next kernel. */
279 ima_add_kexec_buffer(image);
281 /* Call arch image load handlers */
282 ldata = arch_kexec_kernel_image_load(image);
285 ret = PTR_ERR(ldata);
289 image->image_loader_data = ldata;
291 /* In case of error, free up all allocated memory in this function */
293 kimage_file_post_load_cleanup(image);
298 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
299 int initrd_fd, const char __user *cmdline_ptr,
300 unsigned long cmdline_len, unsigned long flags)
303 struct kimage *image;
304 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
306 image = do_kimage_alloc_init();
310 image->file_mode = 1;
312 if (kexec_on_panic) {
313 /* Enable special crash kernel control page alloc policy. */
314 image->control_page = crashk_res.start;
315 image->type = KEXEC_TYPE_CRASH;
318 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
319 cmdline_ptr, cmdline_len, flags);
323 ret = sanity_check_segment_list(image);
325 goto out_free_post_load_bufs;
328 image->control_code_page = kimage_alloc_control_pages(image,
329 get_order(KEXEC_CONTROL_PAGE_SIZE));
330 if (!image->control_code_page) {
331 pr_err("Could not allocate control_code_buffer\n");
332 goto out_free_post_load_bufs;
335 if (!kexec_on_panic) {
336 image->swap_page = kimage_alloc_control_pages(image, 0);
337 if (!image->swap_page) {
338 pr_err("Could not allocate swap buffer\n");
339 goto out_free_control_pages;
345 out_free_control_pages:
346 kimage_free_page_list(&image->control_pages);
347 out_free_post_load_bufs:
348 kimage_file_post_load_cleanup(image);
354 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
355 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
356 unsigned long, flags)
359 struct kimage **dest_image, *image;
361 /* We only trust the superuser with rebooting the system. */
362 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
365 /* Make sure we have a legal set of flags */
366 if (flags != (flags & KEXEC_FILE_FLAGS))
371 if (!mutex_trylock(&kexec_mutex))
374 dest_image = &kexec_image;
375 if (flags & KEXEC_FILE_ON_CRASH) {
376 dest_image = &kexec_crash_image;
377 if (kexec_crash_image)
378 arch_kexec_unprotect_crashkres();
381 if (flags & KEXEC_FILE_UNLOAD)
385 * In case of crash, new kernel gets loaded in reserved region. It is
386 * same memory where old crash kernel might be loaded. Free any
387 * current crash dump kernel before we corrupt it.
389 if (flags & KEXEC_FILE_ON_CRASH)
390 kimage_free(xchg(&kexec_crash_image, NULL));
392 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
397 ret = machine_kexec_prepare(image);
402 * Some architecture(like S390) may touch the crash memory before
403 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
405 ret = kimage_crash_copy_vmcoreinfo(image);
409 ret = kexec_calculate_store_digests(image);
413 for (i = 0; i < image->nr_segments; i++) {
414 struct kexec_segment *ksegment;
416 ksegment = &image->segment[i];
417 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
418 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
421 ret = kimage_load_segment(image, &image->segment[i]);
426 kimage_terminate(image);
428 ret = machine_kexec_post_load(image);
433 * Free up any temporary buffers allocated which are not needed
434 * after image has been loaded
436 kimage_file_post_load_cleanup(image);
438 image = xchg(dest_image, image);
440 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
441 arch_kexec_protect_crashkres();
443 mutex_unlock(&kexec_mutex);
448 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
449 struct kexec_buf *kbuf)
451 struct kimage *image = kbuf->image;
452 unsigned long temp_start, temp_end;
454 temp_end = min(end, kbuf->buf_max);
455 temp_start = temp_end - kbuf->memsz;
458 /* align down start */
459 temp_start = temp_start & (~(kbuf->buf_align - 1));
461 if (temp_start < start || temp_start < kbuf->buf_min)
464 temp_end = temp_start + kbuf->memsz - 1;
467 * Make sure this does not conflict with any of existing
470 if (kimage_is_destination_range(image, temp_start, temp_end)) {
471 temp_start = temp_start - PAGE_SIZE;
475 /* We found a suitable memory range */
479 /* If we are here, we found a suitable memory range */
480 kbuf->mem = temp_start;
482 /* Success, stop navigating through remaining System RAM ranges */
486 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
487 struct kexec_buf *kbuf)
489 struct kimage *image = kbuf->image;
490 unsigned long temp_start, temp_end;
492 temp_start = max(start, kbuf->buf_min);
495 temp_start = ALIGN(temp_start, kbuf->buf_align);
496 temp_end = temp_start + kbuf->memsz - 1;
498 if (temp_end > end || temp_end > kbuf->buf_max)
501 * Make sure this does not conflict with any of existing
504 if (kimage_is_destination_range(image, temp_start, temp_end)) {
505 temp_start = temp_start + PAGE_SIZE;
509 /* We found a suitable memory range */
513 /* If we are here, we found a suitable memory range */
514 kbuf->mem = temp_start;
516 /* Success, stop navigating through remaining System RAM ranges */
520 static int locate_mem_hole_callback(struct resource *res, void *arg)
522 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
523 u64 start = res->start, end = res->end;
524 unsigned long sz = end - start + 1;
526 /* Returning 0 will take to next memory range */
528 /* Don't use memory that will be detected and handled by a driver. */
529 if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
532 if (sz < kbuf->memsz)
535 if (end < kbuf->buf_min || start > kbuf->buf_max)
539 * Allocate memory top down with-in ram range. Otherwise bottom up
543 return locate_mem_hole_top_down(start, end, kbuf);
544 return locate_mem_hole_bottom_up(start, end, kbuf);
547 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
548 static int kexec_walk_memblock(struct kexec_buf *kbuf,
549 int (*func)(struct resource *, void *))
553 phys_addr_t mstart, mend;
554 struct resource res = { };
556 if (kbuf->image->type == KEXEC_TYPE_CRASH)
557 return func(&crashk_res, kbuf);
560 * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
561 * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
562 * locate_mem_hole_callback().
564 if (kbuf->top_down) {
565 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
566 &mstart, &mend, NULL) {
568 * In memblock, end points to the first byte after the
569 * range while in kexec, end points to the last byte
574 ret = func(&res, kbuf);
579 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
580 &mstart, &mend, NULL) {
582 * In memblock, end points to the first byte after the
583 * range while in kexec, end points to the last byte
588 ret = func(&res, kbuf);
597 static int kexec_walk_memblock(struct kexec_buf *kbuf,
598 int (*func)(struct resource *, void *))
605 * kexec_walk_resources - call func(data) on free memory regions
606 * @kbuf: Context info for the search. Also passed to @func.
607 * @func: Function to call for each memory region.
609 * Return: The memory walk will stop when func returns a non-zero value
610 * and that value will be returned. If all free regions are visited without
611 * func returning non-zero, then zero will be returned.
613 static int kexec_walk_resources(struct kexec_buf *kbuf,
614 int (*func)(struct resource *, void *))
616 if (kbuf->image->type == KEXEC_TYPE_CRASH)
617 return walk_iomem_res_desc(crashk_res.desc,
618 IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
619 crashk_res.start, crashk_res.end,
622 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
626 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
627 * @kbuf: Parameters for the memory search.
629 * On success, kbuf->mem will have the start address of the memory region found.
631 * Return: 0 on success, negative errno on error.
633 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
637 /* Arch knows where to place */
638 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
641 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
642 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
644 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
646 return ret == 1 ? 0 : -EADDRNOTAVAIL;
650 * arch_kexec_locate_mem_hole - Find free memory to place the segments.
651 * @kbuf: Parameters for the memory search.
653 * On success, kbuf->mem will have the start address of the memory region found.
655 * Return: 0 on success, negative errno on error.
657 int __weak arch_kexec_locate_mem_hole(struct kexec_buf *kbuf)
659 return kexec_locate_mem_hole(kbuf);
663 * kexec_add_buffer - place a buffer in a kexec segment
664 * @kbuf: Buffer contents and memory parameters.
666 * This function assumes that kexec_mutex is held.
667 * On successful return, @kbuf->mem will have the physical address of
668 * the buffer in memory.
670 * Return: 0 on success, negative errno on error.
672 int kexec_add_buffer(struct kexec_buf *kbuf)
674 struct kexec_segment *ksegment;
677 /* Currently adding segment this way is allowed only in file mode */
678 if (!kbuf->image->file_mode)
681 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
685 * Make sure we are not trying to add buffer after allocating
686 * control pages. All segments need to be placed first before
687 * any control pages are allocated. As control page allocation
688 * logic goes through list of segments to make sure there are
689 * no destination overlaps.
691 if (!list_empty(&kbuf->image->control_pages)) {
696 /* Ensure minimum alignment needed for segments. */
697 kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
698 kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
700 /* Walk the RAM ranges and allocate a suitable range for the buffer */
701 ret = arch_kexec_locate_mem_hole(kbuf);
705 /* Found a suitable memory range */
706 ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
707 ksegment->kbuf = kbuf->buffer;
708 ksegment->bufsz = kbuf->bufsz;
709 ksegment->mem = kbuf->mem;
710 ksegment->memsz = kbuf->memsz;
711 kbuf->image->nr_segments++;
715 /* Calculate and store the digest of segments */
716 static int kexec_calculate_store_digests(struct kimage *image)
718 struct crypto_shash *tfm;
719 struct shash_desc *desc;
720 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
721 size_t desc_size, nullsz;
724 struct kexec_sha_region *sha_regions;
725 struct purgatory_info *pi = &image->purgatory_info;
727 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
730 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
731 zero_buf_sz = PAGE_SIZE;
733 tfm = crypto_alloc_shash("sha256", 0, 0);
739 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
740 desc = kzalloc(desc_size, GFP_KERNEL);
746 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
747 sha_regions = vzalloc(sha_region_sz);
755 ret = crypto_shash_init(desc);
757 goto out_free_sha_regions;
759 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
762 goto out_free_sha_regions;
765 for (j = i = 0; i < image->nr_segments; i++) {
766 struct kexec_segment *ksegment;
768 ksegment = &image->segment[i];
770 * Skip purgatory as it will be modified once we put digest
773 if (ksegment->kbuf == pi->purgatory_buf)
776 ret = crypto_shash_update(desc, ksegment->kbuf,
782 * Assume rest of the buffer is filled with zero and
783 * update digest accordingly.
785 nullsz = ksegment->memsz - ksegment->bufsz;
787 unsigned long bytes = nullsz;
789 if (bytes > zero_buf_sz)
791 ret = crypto_shash_update(desc, zero_buf, bytes);
800 sha_regions[j].start = ksegment->mem;
801 sha_regions[j].len = ksegment->memsz;
806 ret = crypto_shash_final(desc, digest);
808 goto out_free_digest;
809 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
810 sha_regions, sha_region_sz, 0);
812 goto out_free_digest;
814 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
815 digest, SHA256_DIGEST_SIZE, 0);
817 goto out_free_digest;
822 out_free_sha_regions:
832 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
834 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
835 * @pi: Purgatory to be loaded.
836 * @kbuf: Buffer to setup.
838 * Allocates the memory needed for the buffer. Caller is responsible to free
839 * the memory after use.
841 * Return: 0 on success, negative errno on error.
843 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
844 struct kexec_buf *kbuf)
846 const Elf_Shdr *sechdrs;
847 unsigned long bss_align;
848 unsigned long bss_sz;
852 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
853 kbuf->buf_align = bss_align = 1;
854 kbuf->bufsz = bss_sz = 0;
856 for (i = 0; i < pi->ehdr->e_shnum; i++) {
857 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
860 align = sechdrs[i].sh_addralign;
861 if (sechdrs[i].sh_type != SHT_NOBITS) {
862 if (kbuf->buf_align < align)
863 kbuf->buf_align = align;
864 kbuf->bufsz = ALIGN(kbuf->bufsz, align);
865 kbuf->bufsz += sechdrs[i].sh_size;
867 if (bss_align < align)
869 bss_sz = ALIGN(bss_sz, align);
870 bss_sz += sechdrs[i].sh_size;
873 kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
874 kbuf->memsz = kbuf->bufsz + bss_sz;
875 if (kbuf->buf_align < bss_align)
876 kbuf->buf_align = bss_align;
878 kbuf->buffer = vzalloc(kbuf->bufsz);
881 pi->purgatory_buf = kbuf->buffer;
883 ret = kexec_add_buffer(kbuf);
889 vfree(pi->purgatory_buf);
890 pi->purgatory_buf = NULL;
895 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
896 * @pi: Purgatory to be loaded.
897 * @kbuf: Buffer prepared to store purgatory.
899 * Allocates the memory needed for the buffer. Caller is responsible to free
900 * the memory after use.
902 * Return: 0 on success, negative errno on error.
904 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
905 struct kexec_buf *kbuf)
907 unsigned long bss_addr;
908 unsigned long offset;
913 * The section headers in kexec_purgatory are read-only. In order to
914 * have them modifiable make a temporary copy.
916 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
919 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
920 pi->ehdr->e_shnum * sizeof(Elf_Shdr));
921 pi->sechdrs = sechdrs;
924 bss_addr = kbuf->mem + kbuf->bufsz;
925 kbuf->image->start = pi->ehdr->e_entry;
927 for (i = 0; i < pi->ehdr->e_shnum; i++) {
931 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
934 align = sechdrs[i].sh_addralign;
935 if (sechdrs[i].sh_type == SHT_NOBITS) {
936 bss_addr = ALIGN(bss_addr, align);
937 sechdrs[i].sh_addr = bss_addr;
938 bss_addr += sechdrs[i].sh_size;
942 offset = ALIGN(offset, align);
943 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
944 pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
945 pi->ehdr->e_entry < (sechdrs[i].sh_addr
946 + sechdrs[i].sh_size)) {
947 kbuf->image->start -= sechdrs[i].sh_addr;
948 kbuf->image->start += kbuf->mem + offset;
951 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
952 dst = pi->purgatory_buf + offset;
953 memcpy(dst, src, sechdrs[i].sh_size);
955 sechdrs[i].sh_addr = kbuf->mem + offset;
956 sechdrs[i].sh_offset = offset;
957 offset += sechdrs[i].sh_size;
963 static int kexec_apply_relocations(struct kimage *image)
966 struct purgatory_info *pi = &image->purgatory_info;
967 const Elf_Shdr *sechdrs;
969 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
971 for (i = 0; i < pi->ehdr->e_shnum; i++) {
972 const Elf_Shdr *relsec;
973 const Elf_Shdr *symtab;
976 relsec = sechdrs + i;
978 if (relsec->sh_type != SHT_RELA &&
979 relsec->sh_type != SHT_REL)
983 * For section of type SHT_RELA/SHT_REL,
984 * ->sh_link contains section header index of associated
985 * symbol table. And ->sh_info contains section header
986 * index of section to which relocations apply.
988 if (relsec->sh_info >= pi->ehdr->e_shnum ||
989 relsec->sh_link >= pi->ehdr->e_shnum)
992 section = pi->sechdrs + relsec->sh_info;
993 symtab = sechdrs + relsec->sh_link;
995 if (!(section->sh_flags & SHF_ALLOC))
999 * symtab->sh_link contain section header index of associated
1002 if (symtab->sh_link >= pi->ehdr->e_shnum)
1003 /* Invalid section number? */
1007 * Respective architecture needs to provide support for applying
1008 * relocations of type SHT_RELA/SHT_REL.
1010 if (relsec->sh_type == SHT_RELA)
1011 ret = arch_kexec_apply_relocations_add(pi, section,
1013 else if (relsec->sh_type == SHT_REL)
1014 ret = arch_kexec_apply_relocations(pi, section,
1024 * kexec_load_purgatory - Load and relocate the purgatory object.
1025 * @image: Image to add the purgatory to.
1026 * @kbuf: Memory parameters to use.
1028 * Allocates the memory needed for image->purgatory_info.sechdrs and
1029 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1030 * to free the memory after use.
1032 * Return: 0 on success, negative errno on error.
1034 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1036 struct purgatory_info *pi = &image->purgatory_info;
1039 if (kexec_purgatory_size <= 0)
1042 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1044 ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1048 ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1052 ret = kexec_apply_relocations(image);
1061 vfree(pi->purgatory_buf);
1062 pi->purgatory_buf = NULL;
1067 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1068 * @pi: Purgatory to search in.
1069 * @name: Name of the symbol.
1071 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1073 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1076 const Elf_Shdr *sechdrs;
1077 const Elf_Ehdr *ehdr;
1078 const Elf_Sym *syms;
1086 sechdrs = (void *)ehdr + ehdr->e_shoff;
1088 for (i = 0; i < ehdr->e_shnum; i++) {
1089 if (sechdrs[i].sh_type != SHT_SYMTAB)
1092 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1093 /* Invalid strtab section number */
1095 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1096 syms = (void *)ehdr + sechdrs[i].sh_offset;
1098 /* Go through symbols for a match */
1099 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1100 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1103 if (strcmp(strtab + syms[k].st_name, name) != 0)
1106 if (syms[k].st_shndx == SHN_UNDEF ||
1107 syms[k].st_shndx >= ehdr->e_shnum) {
1108 pr_debug("Symbol: %s has bad section index %d.\n",
1109 name, syms[k].st_shndx);
1113 /* Found the symbol we are looking for */
1121 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1123 struct purgatory_info *pi = &image->purgatory_info;
1127 sym = kexec_purgatory_find_symbol(pi, name);
1129 return ERR_PTR(-EINVAL);
1131 sechdr = &pi->sechdrs[sym->st_shndx];
1134 * Returns the address where symbol will finally be loaded after
1135 * kexec_load_segment()
1137 return (void *)(sechdr->sh_addr + sym->st_value);
1141 * Get or set value of a symbol. If "get_value" is true, symbol value is
1142 * returned in buf otherwise symbol value is set based on value in buf.
1144 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1145 void *buf, unsigned int size, bool get_value)
1147 struct purgatory_info *pi = &image->purgatory_info;
1152 sym = kexec_purgatory_find_symbol(pi, name);
1156 if (sym->st_size != size) {
1157 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1158 name, (unsigned long)sym->st_size, size);
1162 sec = pi->sechdrs + sym->st_shndx;
1164 if (sec->sh_type == SHT_NOBITS) {
1165 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1166 get_value ? "get" : "set");
1170 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1173 memcpy((void *)buf, sym_buf, size);
1175 memcpy((void *)sym_buf, buf, size);
1179 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1181 int crash_exclude_mem_range(struct crash_mem *mem,
1182 unsigned long long mstart, unsigned long long mend)
1185 unsigned long long start, end, p_start, p_end;
1186 struct crash_mem_range temp_range = {0, 0};
1188 for (i = 0; i < mem->nr_ranges; i++) {
1189 start = mem->ranges[i].start;
1190 end = mem->ranges[i].end;
1194 if (mstart > end || mend < start)
1197 /* Truncate any area outside of range */
1203 /* Found completely overlapping range */
1204 if (p_start == start && p_end == end) {
1205 mem->ranges[i].start = 0;
1206 mem->ranges[i].end = 0;
1207 if (i < mem->nr_ranges - 1) {
1208 /* Shift rest of the ranges to left */
1209 for (j = i; j < mem->nr_ranges - 1; j++) {
1210 mem->ranges[j].start =
1211 mem->ranges[j+1].start;
1212 mem->ranges[j].end =
1213 mem->ranges[j+1].end;
1217 * Continue to check if there are another overlapping ranges
1218 * from the current position because of shifting the above
1229 if (p_start > start && p_end < end) {
1230 /* Split original range */
1231 mem->ranges[i].end = p_start - 1;
1232 temp_range.start = p_end + 1;
1233 temp_range.end = end;
1234 } else if (p_start != start)
1235 mem->ranges[i].end = p_start - 1;
1237 mem->ranges[i].start = p_end + 1;
1241 /* If a split happened, add the split to array */
1242 if (!temp_range.end)
1245 /* Split happened */
1246 if (i == mem->max_nr_ranges - 1)
1249 /* Location where new range should go */
1251 if (j < mem->nr_ranges) {
1252 /* Move over all ranges one slot towards the end */
1253 for (i = mem->nr_ranges - 1; i >= j; i--)
1254 mem->ranges[i + 1] = mem->ranges[i];
1257 mem->ranges[j].start = temp_range.start;
1258 mem->ranges[j].end = temp_range.end;
1263 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1264 void **addr, unsigned long *sz)
1268 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1270 unsigned int cpu, i;
1271 unsigned long long notes_addr;
1272 unsigned long mstart, mend;
1274 /* extra phdr for vmcoreinfo ELF note */
1275 nr_phdr = nr_cpus + 1;
1276 nr_phdr += mem->nr_ranges;
1279 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1280 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1281 * I think this is required by tools like gdb. So same physical
1282 * memory will be mapped in two ELF headers. One will contain kernel
1283 * text virtual addresses and other will have __va(physical) addresses.
1287 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1288 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1290 buf = vzalloc(elf_sz);
1294 ehdr = (Elf64_Ehdr *)buf;
1295 phdr = (Elf64_Phdr *)(ehdr + 1);
1296 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1297 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1298 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1299 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1300 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1301 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1302 ehdr->e_type = ET_CORE;
1303 ehdr->e_machine = ELF_ARCH;
1304 ehdr->e_version = EV_CURRENT;
1305 ehdr->e_phoff = sizeof(Elf64_Ehdr);
1306 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1307 ehdr->e_phentsize = sizeof(Elf64_Phdr);
1309 /* Prepare one phdr of type PT_NOTE for each present CPU */
1310 for_each_present_cpu(cpu) {
1311 phdr->p_type = PT_NOTE;
1312 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1313 phdr->p_offset = phdr->p_paddr = notes_addr;
1314 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1319 /* Prepare one PT_NOTE header for vmcoreinfo */
1320 phdr->p_type = PT_NOTE;
1321 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1322 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1326 /* Prepare PT_LOAD type program header for kernel text region */
1328 phdr->p_type = PT_LOAD;
1329 phdr->p_flags = PF_R|PF_W|PF_X;
1330 phdr->p_vaddr = (unsigned long) _text;
1331 phdr->p_filesz = phdr->p_memsz = _end - _text;
1332 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1337 /* Go through all the ranges in mem->ranges[] and prepare phdr */
1338 for (i = 0; i < mem->nr_ranges; i++) {
1339 mstart = mem->ranges[i].start;
1340 mend = mem->ranges[i].end;
1342 phdr->p_type = PT_LOAD;
1343 phdr->p_flags = PF_R|PF_W|PF_X;
1344 phdr->p_offset = mstart;
1346 phdr->p_paddr = mstart;
1347 phdr->p_vaddr = (unsigned long) __va(mstart);
1348 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1351 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",
1352 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1353 ehdr->e_phnum, phdr->p_offset);