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 #ifdef CONFIG_KEXEC_SIG
33 static bool sig_enforce
= IS_ENABLED(CONFIG_KEXEC_SIG_FORCE
);
35 void set_kexec_sig_enforced(void)
41 static int kexec_calculate_store_digests(struct kimage
*image
);
44 * Currently this is the only default function that is exported as some
45 * architectures need it to do additional handlings.
46 * In the future, other default functions may be exported too if required.
48 int kexec_image_probe_default(struct kimage
*image
, void *buf
,
49 unsigned long buf_len
)
51 const struct kexec_file_ops
* const *fops
;
54 for (fops
= &kexec_file_loaders
[0]; *fops
&& (*fops
)->probe
; ++fops
) {
55 ret
= (*fops
)->probe(buf
, buf_len
);
65 /* Architectures can provide this probe function */
66 int __weak
arch_kexec_kernel_image_probe(struct kimage
*image
, void *buf
,
67 unsigned long buf_len
)
69 return kexec_image_probe_default(image
, buf
, buf_len
);
72 static void *kexec_image_load_default(struct kimage
*image
)
74 if (!image
->fops
|| !image
->fops
->load
)
75 return ERR_PTR(-ENOEXEC
);
77 return image
->fops
->load(image
, image
->kernel_buf
,
78 image
->kernel_buf_len
, image
->initrd_buf
,
79 image
->initrd_buf_len
, image
->cmdline_buf
,
80 image
->cmdline_buf_len
);
83 void * __weak
arch_kexec_kernel_image_load(struct kimage
*image
)
85 return kexec_image_load_default(image
);
88 int kexec_image_post_load_cleanup_default(struct kimage
*image
)
90 if (!image
->fops
|| !image
->fops
->cleanup
)
93 return image
->fops
->cleanup(image
->image_loader_data
);
96 int __weak
arch_kimage_file_post_load_cleanup(struct kimage
*image
)
98 return kexec_image_post_load_cleanup_default(image
);
101 #ifdef CONFIG_KEXEC_SIG
102 static int kexec_image_verify_sig_default(struct kimage
*image
, void *buf
,
103 unsigned long buf_len
)
105 if (!image
->fops
|| !image
->fops
->verify_sig
) {
106 pr_debug("kernel loader does not support signature verification.\n");
107 return -EKEYREJECTED
;
110 return image
->fops
->verify_sig(buf
, buf_len
);
113 int __weak
arch_kexec_kernel_verify_sig(struct kimage
*image
, void *buf
,
114 unsigned long buf_len
)
116 return kexec_image_verify_sig_default(image
, buf
, buf_len
);
121 * Free up memory used by kernel, initrd, and command line. This is temporary
122 * memory allocation which is not needed any more after these buffers have
123 * been loaded into separate segments and have been copied elsewhere.
125 void kimage_file_post_load_cleanup(struct kimage
*image
)
127 struct purgatory_info
*pi
= &image
->purgatory_info
;
129 vfree(image
->kernel_buf
);
130 image
->kernel_buf
= NULL
;
132 vfree(image
->initrd_buf
);
133 image
->initrd_buf
= NULL
;
135 kfree(image
->cmdline_buf
);
136 image
->cmdline_buf
= NULL
;
138 vfree(pi
->purgatory_buf
);
139 pi
->purgatory_buf
= NULL
;
144 #ifdef CONFIG_IMA_KEXEC
145 vfree(image
->ima_buffer
);
146 image
->ima_buffer
= NULL
;
147 #endif /* CONFIG_IMA_KEXEC */
149 /* See if architecture has anything to cleanup post load */
150 arch_kimage_file_post_load_cleanup(image
);
153 * Above call should have called into bootloader to free up
154 * any data stored in kimage->image_loader_data. It should
155 * be ok now to free it up.
157 kfree(image
->image_loader_data
);
158 image
->image_loader_data
= NULL
;
161 #ifdef CONFIG_KEXEC_SIG
163 kimage_validate_signature(struct kimage
*image
)
167 ret
= arch_kexec_kernel_verify_sig(image
, image
->kernel_buf
,
168 image
->kernel_buf_len
);
172 pr_notice("Enforced kernel signature verification failed (%d).\n", ret
);
177 * If IMA is guaranteed to appraise a signature on the kexec
178 * image, permit it even if the kernel is otherwise locked
181 if (!ima_appraise_signature(READING_KEXEC_IMAGE
) &&
182 security_locked_down(LOCKDOWN_KEXEC
))
185 pr_debug("kernel signature verification failed (%d).\n", ret
);
193 * In file mode list of segments is prepared by kernel. Copy relevant
194 * data from user space, do error checking, prepare segment list
197 kimage_file_prepare_segments(struct kimage
*image
, int kernel_fd
, int initrd_fd
,
198 const char __user
*cmdline_ptr
,
199 unsigned long cmdline_len
, unsigned flags
)
204 ret
= kernel_read_file_from_fd(kernel_fd
, 0, &image
->kernel_buf
,
205 INT_MAX
, NULL
, READING_KEXEC_IMAGE
);
208 image
->kernel_buf_len
= ret
;
210 /* Call arch image probe handlers */
211 ret
= arch_kexec_kernel_image_probe(image
, image
->kernel_buf
,
212 image
->kernel_buf_len
);
216 #ifdef CONFIG_KEXEC_SIG
217 ret
= kimage_validate_signature(image
);
222 /* It is possible that there no initramfs is being loaded */
223 if (!(flags
& KEXEC_FILE_NO_INITRAMFS
)) {
224 ret
= kernel_read_file_from_fd(initrd_fd
, 0, &image
->initrd_buf
,
226 READING_KEXEC_INITRAMFS
);
229 image
->initrd_buf_len
= ret
;
234 image
->cmdline_buf
= memdup_user(cmdline_ptr
, cmdline_len
);
235 if (IS_ERR(image
->cmdline_buf
)) {
236 ret
= PTR_ERR(image
->cmdline_buf
);
237 image
->cmdline_buf
= NULL
;
241 image
->cmdline_buf_len
= cmdline_len
;
243 /* command line should be a string with last byte null */
244 if (image
->cmdline_buf
[cmdline_len
- 1] != '\0') {
249 ima_kexec_cmdline(kernel_fd
, image
->cmdline_buf
,
250 image
->cmdline_buf_len
- 1);
253 /* IMA needs to pass the measurement list to the next kernel. */
254 ima_add_kexec_buffer(image
);
256 /* Call arch image load handlers */
257 ldata
= arch_kexec_kernel_image_load(image
);
260 ret
= PTR_ERR(ldata
);
264 image
->image_loader_data
= ldata
;
266 /* In case of error, free up all allocated memory in this function */
268 kimage_file_post_load_cleanup(image
);
273 kimage_file_alloc_init(struct kimage
**rimage
, int kernel_fd
,
274 int initrd_fd
, const char __user
*cmdline_ptr
,
275 unsigned long cmdline_len
, unsigned long flags
)
278 struct kimage
*image
;
279 bool kexec_on_panic
= flags
& KEXEC_FILE_ON_CRASH
;
281 image
= do_kimage_alloc_init();
285 image
->file_mode
= 1;
287 if (kexec_on_panic
) {
288 /* Enable special crash kernel control page alloc policy. */
289 image
->control_page
= crashk_res
.start
;
290 image
->type
= KEXEC_TYPE_CRASH
;
293 ret
= kimage_file_prepare_segments(image
, kernel_fd
, initrd_fd
,
294 cmdline_ptr
, cmdline_len
, flags
);
298 ret
= sanity_check_segment_list(image
);
300 goto out_free_post_load_bufs
;
303 image
->control_code_page
= kimage_alloc_control_pages(image
,
304 get_order(KEXEC_CONTROL_PAGE_SIZE
));
305 if (!image
->control_code_page
) {
306 pr_err("Could not allocate control_code_buffer\n");
307 goto out_free_post_load_bufs
;
310 if (!kexec_on_panic
) {
311 image
->swap_page
= kimage_alloc_control_pages(image
, 0);
312 if (!image
->swap_page
) {
313 pr_err("Could not allocate swap buffer\n");
314 goto out_free_control_pages
;
320 out_free_control_pages
:
321 kimage_free_page_list(&image
->control_pages
);
322 out_free_post_load_bufs
:
323 kimage_file_post_load_cleanup(image
);
329 SYSCALL_DEFINE5(kexec_file_load
, int, kernel_fd
, int, initrd_fd
,
330 unsigned long, cmdline_len
, const char __user
*, cmdline_ptr
,
331 unsigned long, flags
)
334 struct kimage
**dest_image
, *image
;
336 /* We only trust the superuser with rebooting the system. */
337 if (!capable(CAP_SYS_BOOT
) || kexec_load_disabled
)
340 /* Make sure we have a legal set of flags */
341 if (flags
!= (flags
& KEXEC_FILE_FLAGS
))
346 if (!mutex_trylock(&kexec_mutex
))
349 dest_image
= &kexec_image
;
350 if (flags
& KEXEC_FILE_ON_CRASH
) {
351 dest_image
= &kexec_crash_image
;
352 if (kexec_crash_image
)
353 arch_kexec_unprotect_crashkres();
356 if (flags
& KEXEC_FILE_UNLOAD
)
360 * In case of crash, new kernel gets loaded in reserved region. It is
361 * same memory where old crash kernel might be loaded. Free any
362 * current crash dump kernel before we corrupt it.
364 if (flags
& KEXEC_FILE_ON_CRASH
)
365 kimage_free(xchg(&kexec_crash_image
, NULL
));
367 ret
= kimage_file_alloc_init(&image
, kernel_fd
, initrd_fd
, cmdline_ptr
,
372 ret
= machine_kexec_prepare(image
);
377 * Some architecture(like S390) may touch the crash memory before
378 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
380 ret
= kimage_crash_copy_vmcoreinfo(image
);
384 ret
= kexec_calculate_store_digests(image
);
388 for (i
= 0; i
< image
->nr_segments
; i
++) {
389 struct kexec_segment
*ksegment
;
391 ksegment
= &image
->segment
[i
];
392 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
393 i
, ksegment
->buf
, ksegment
->bufsz
, ksegment
->mem
,
396 ret
= kimage_load_segment(image
, &image
->segment
[i
]);
401 kimage_terminate(image
);
403 ret
= machine_kexec_post_load(image
);
408 * Free up any temporary buffers allocated which are not needed
409 * after image has been loaded
411 kimage_file_post_load_cleanup(image
);
413 image
= xchg(dest_image
, image
);
415 if ((flags
& KEXEC_FILE_ON_CRASH
) && kexec_crash_image
)
416 arch_kexec_protect_crashkres();
418 mutex_unlock(&kexec_mutex
);
423 static int locate_mem_hole_top_down(unsigned long start
, unsigned long end
,
424 struct kexec_buf
*kbuf
)
426 struct kimage
*image
= kbuf
->image
;
427 unsigned long temp_start
, temp_end
;
429 temp_end
= min(end
, kbuf
->buf_max
);
430 temp_start
= temp_end
- kbuf
->memsz
;
433 /* align down start */
434 temp_start
= temp_start
& (~(kbuf
->buf_align
- 1));
436 if (temp_start
< start
|| temp_start
< kbuf
->buf_min
)
439 temp_end
= temp_start
+ kbuf
->memsz
- 1;
442 * Make sure this does not conflict with any of existing
445 if (kimage_is_destination_range(image
, temp_start
, temp_end
)) {
446 temp_start
= temp_start
- PAGE_SIZE
;
450 /* We found a suitable memory range */
454 /* If we are here, we found a suitable memory range */
455 kbuf
->mem
= temp_start
;
457 /* Success, stop navigating through remaining System RAM ranges */
461 static int locate_mem_hole_bottom_up(unsigned long start
, unsigned long end
,
462 struct kexec_buf
*kbuf
)
464 struct kimage
*image
= kbuf
->image
;
465 unsigned long temp_start
, temp_end
;
467 temp_start
= max(start
, kbuf
->buf_min
);
470 temp_start
= ALIGN(temp_start
, kbuf
->buf_align
);
471 temp_end
= temp_start
+ kbuf
->memsz
- 1;
473 if (temp_end
> end
|| temp_end
> kbuf
->buf_max
)
476 * Make sure this does not conflict with any of existing
479 if (kimage_is_destination_range(image
, temp_start
, temp_end
)) {
480 temp_start
= temp_start
+ PAGE_SIZE
;
484 /* We found a suitable memory range */
488 /* If we are here, we found a suitable memory range */
489 kbuf
->mem
= temp_start
;
491 /* Success, stop navigating through remaining System RAM ranges */
495 static int locate_mem_hole_callback(struct resource
*res
, void *arg
)
497 struct kexec_buf
*kbuf
= (struct kexec_buf
*)arg
;
498 u64 start
= res
->start
, end
= res
->end
;
499 unsigned long sz
= end
- start
+ 1;
501 /* Returning 0 will take to next memory range */
503 /* Don't use memory that will be detected and handled by a driver. */
504 if (res
->flags
& IORESOURCE_SYSRAM_DRIVER_MANAGED
)
507 if (sz
< kbuf
->memsz
)
510 if (end
< kbuf
->buf_min
|| start
> kbuf
->buf_max
)
514 * Allocate memory top down with-in ram range. Otherwise bottom up
518 return locate_mem_hole_top_down(start
, end
, kbuf
);
519 return locate_mem_hole_bottom_up(start
, end
, kbuf
);
522 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
523 static int kexec_walk_memblock(struct kexec_buf
*kbuf
,
524 int (*func
)(struct resource
*, void *))
528 phys_addr_t mstart
, mend
;
529 struct resource res
= { };
531 if (kbuf
->image
->type
== KEXEC_TYPE_CRASH
)
532 return func(&crashk_res
, kbuf
);
534 if (kbuf
->top_down
) {
535 for_each_free_mem_range_reverse(i
, NUMA_NO_NODE
, MEMBLOCK_NONE
,
536 &mstart
, &mend
, NULL
) {
538 * In memblock, end points to the first byte after the
539 * range while in kexec, end points to the last byte
544 ret
= func(&res
, kbuf
);
549 for_each_free_mem_range(i
, NUMA_NO_NODE
, MEMBLOCK_NONE
,
550 &mstart
, &mend
, NULL
) {
552 * In memblock, end points to the first byte after the
553 * range while in kexec, end points to the last byte
558 ret
= func(&res
, kbuf
);
567 static int kexec_walk_memblock(struct kexec_buf
*kbuf
,
568 int (*func
)(struct resource
*, void *))
575 * kexec_walk_resources - call func(data) on free memory regions
576 * @kbuf: Context info for the search. Also passed to @func.
577 * @func: Function to call for each memory region.
579 * Return: The memory walk will stop when func returns a non-zero value
580 * and that value will be returned. If all free regions are visited without
581 * func returning non-zero, then zero will be returned.
583 static int kexec_walk_resources(struct kexec_buf
*kbuf
,
584 int (*func
)(struct resource
*, void *))
586 if (kbuf
->image
->type
== KEXEC_TYPE_CRASH
)
587 return walk_iomem_res_desc(crashk_res
.desc
,
588 IORESOURCE_SYSTEM_RAM
| IORESOURCE_BUSY
,
589 crashk_res
.start
, crashk_res
.end
,
592 return walk_system_ram_res(0, ULONG_MAX
, kbuf
, func
);
596 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
597 * @kbuf: Parameters for the memory search.
599 * On success, kbuf->mem will have the start address of the memory region found.
601 * Return: 0 on success, negative errno on error.
603 int kexec_locate_mem_hole(struct kexec_buf
*kbuf
)
607 /* Arch knows where to place */
608 if (kbuf
->mem
!= KEXEC_BUF_MEM_UNKNOWN
)
611 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK
))
612 ret
= kexec_walk_resources(kbuf
, locate_mem_hole_callback
);
614 ret
= kexec_walk_memblock(kbuf
, locate_mem_hole_callback
);
616 return ret
== 1 ? 0 : -EADDRNOTAVAIL
;
620 * arch_kexec_locate_mem_hole - Find free memory to place the segments.
621 * @kbuf: Parameters for the memory search.
623 * On success, kbuf->mem will have the start address of the memory region found.
625 * Return: 0 on success, negative errno on error.
627 int __weak
arch_kexec_locate_mem_hole(struct kexec_buf
*kbuf
)
629 return kexec_locate_mem_hole(kbuf
);
633 * kexec_add_buffer - place a buffer in a kexec segment
634 * @kbuf: Buffer contents and memory parameters.
636 * This function assumes that kexec_mutex is held.
637 * On successful return, @kbuf->mem will have the physical address of
638 * the buffer in memory.
640 * Return: 0 on success, negative errno on error.
642 int kexec_add_buffer(struct kexec_buf
*kbuf
)
644 struct kexec_segment
*ksegment
;
647 /* Currently adding segment this way is allowed only in file mode */
648 if (!kbuf
->image
->file_mode
)
651 if (kbuf
->image
->nr_segments
>= KEXEC_SEGMENT_MAX
)
655 * Make sure we are not trying to add buffer after allocating
656 * control pages. All segments need to be placed first before
657 * any control pages are allocated. As control page allocation
658 * logic goes through list of segments to make sure there are
659 * no destination overlaps.
661 if (!list_empty(&kbuf
->image
->control_pages
)) {
666 /* Ensure minimum alignment needed for segments. */
667 kbuf
->memsz
= ALIGN(kbuf
->memsz
, PAGE_SIZE
);
668 kbuf
->buf_align
= max(kbuf
->buf_align
, PAGE_SIZE
);
670 /* Walk the RAM ranges and allocate a suitable range for the buffer */
671 ret
= arch_kexec_locate_mem_hole(kbuf
);
675 /* Found a suitable memory range */
676 ksegment
= &kbuf
->image
->segment
[kbuf
->image
->nr_segments
];
677 ksegment
->kbuf
= kbuf
->buffer
;
678 ksegment
->bufsz
= kbuf
->bufsz
;
679 ksegment
->mem
= kbuf
->mem
;
680 ksegment
->memsz
= kbuf
->memsz
;
681 kbuf
->image
->nr_segments
++;
685 /* Calculate and store the digest of segments */
686 static int kexec_calculate_store_digests(struct kimage
*image
)
688 struct crypto_shash
*tfm
;
689 struct shash_desc
*desc
;
690 int ret
= 0, i
, j
, zero_buf_sz
, sha_region_sz
;
691 size_t desc_size
, nullsz
;
694 struct kexec_sha_region
*sha_regions
;
695 struct purgatory_info
*pi
= &image
->purgatory_info
;
697 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY
))
700 zero_buf
= __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT
);
701 zero_buf_sz
= PAGE_SIZE
;
703 tfm
= crypto_alloc_shash("sha256", 0, 0);
709 desc_size
= crypto_shash_descsize(tfm
) + sizeof(*desc
);
710 desc
= kzalloc(desc_size
, GFP_KERNEL
);
716 sha_region_sz
= KEXEC_SEGMENT_MAX
* sizeof(struct kexec_sha_region
);
717 sha_regions
= vzalloc(sha_region_sz
);
725 ret
= crypto_shash_init(desc
);
727 goto out_free_sha_regions
;
729 digest
= kzalloc(SHA256_DIGEST_SIZE
, GFP_KERNEL
);
732 goto out_free_sha_regions
;
735 for (j
= i
= 0; i
< image
->nr_segments
; i
++) {
736 struct kexec_segment
*ksegment
;
738 ksegment
= &image
->segment
[i
];
740 * Skip purgatory as it will be modified once we put digest
743 if (ksegment
->kbuf
== pi
->purgatory_buf
)
746 ret
= crypto_shash_update(desc
, ksegment
->kbuf
,
752 * Assume rest of the buffer is filled with zero and
753 * update digest accordingly.
755 nullsz
= ksegment
->memsz
- ksegment
->bufsz
;
757 unsigned long bytes
= nullsz
;
759 if (bytes
> zero_buf_sz
)
761 ret
= crypto_shash_update(desc
, zero_buf
, bytes
);
770 sha_regions
[j
].start
= ksegment
->mem
;
771 sha_regions
[j
].len
= ksegment
->memsz
;
776 ret
= crypto_shash_final(desc
, digest
);
778 goto out_free_digest
;
779 ret
= kexec_purgatory_get_set_symbol(image
, "purgatory_sha_regions",
780 sha_regions
, sha_region_sz
, 0);
782 goto out_free_digest
;
784 ret
= kexec_purgatory_get_set_symbol(image
, "purgatory_sha256_digest",
785 digest
, SHA256_DIGEST_SIZE
, 0);
787 goto out_free_digest
;
792 out_free_sha_regions
:
802 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
804 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
805 * @pi: Purgatory to be loaded.
806 * @kbuf: Buffer to setup.
808 * Allocates the memory needed for the buffer. Caller is responsible to free
809 * the memory after use.
811 * Return: 0 on success, negative errno on error.
813 static int kexec_purgatory_setup_kbuf(struct purgatory_info
*pi
,
814 struct kexec_buf
*kbuf
)
816 const Elf_Shdr
*sechdrs
;
817 unsigned long bss_align
;
818 unsigned long bss_sz
;
822 sechdrs
= (void *)pi
->ehdr
+ pi
->ehdr
->e_shoff
;
823 kbuf
->buf_align
= bss_align
= 1;
824 kbuf
->bufsz
= bss_sz
= 0;
826 for (i
= 0; i
< pi
->ehdr
->e_shnum
; i
++) {
827 if (!(sechdrs
[i
].sh_flags
& SHF_ALLOC
))
830 align
= sechdrs
[i
].sh_addralign
;
831 if (sechdrs
[i
].sh_type
!= SHT_NOBITS
) {
832 if (kbuf
->buf_align
< align
)
833 kbuf
->buf_align
= align
;
834 kbuf
->bufsz
= ALIGN(kbuf
->bufsz
, align
);
835 kbuf
->bufsz
+= sechdrs
[i
].sh_size
;
837 if (bss_align
< align
)
839 bss_sz
= ALIGN(bss_sz
, align
);
840 bss_sz
+= sechdrs
[i
].sh_size
;
843 kbuf
->bufsz
= ALIGN(kbuf
->bufsz
, bss_align
);
844 kbuf
->memsz
= kbuf
->bufsz
+ bss_sz
;
845 if (kbuf
->buf_align
< bss_align
)
846 kbuf
->buf_align
= bss_align
;
848 kbuf
->buffer
= vzalloc(kbuf
->bufsz
);
851 pi
->purgatory_buf
= kbuf
->buffer
;
853 ret
= kexec_add_buffer(kbuf
);
859 vfree(pi
->purgatory_buf
);
860 pi
->purgatory_buf
= NULL
;
865 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
866 * @pi: Purgatory to be loaded.
867 * @kbuf: Buffer prepared to store purgatory.
869 * Allocates the memory needed for the buffer. Caller is responsible to free
870 * the memory after use.
872 * Return: 0 on success, negative errno on error.
874 static int kexec_purgatory_setup_sechdrs(struct purgatory_info
*pi
,
875 struct kexec_buf
*kbuf
)
877 unsigned long bss_addr
;
878 unsigned long offset
;
883 * The section headers in kexec_purgatory are read-only. In order to
884 * have them modifiable make a temporary copy.
886 sechdrs
= vzalloc(array_size(sizeof(Elf_Shdr
), pi
->ehdr
->e_shnum
));
889 memcpy(sechdrs
, (void *)pi
->ehdr
+ pi
->ehdr
->e_shoff
,
890 pi
->ehdr
->e_shnum
* sizeof(Elf_Shdr
));
891 pi
->sechdrs
= sechdrs
;
894 bss_addr
= kbuf
->mem
+ kbuf
->bufsz
;
895 kbuf
->image
->start
= pi
->ehdr
->e_entry
;
897 for (i
= 0; i
< pi
->ehdr
->e_shnum
; i
++) {
901 if (!(sechdrs
[i
].sh_flags
& SHF_ALLOC
))
904 align
= sechdrs
[i
].sh_addralign
;
905 if (sechdrs
[i
].sh_type
== SHT_NOBITS
) {
906 bss_addr
= ALIGN(bss_addr
, align
);
907 sechdrs
[i
].sh_addr
= bss_addr
;
908 bss_addr
+= sechdrs
[i
].sh_size
;
912 offset
= ALIGN(offset
, align
);
913 if (sechdrs
[i
].sh_flags
& SHF_EXECINSTR
&&
914 pi
->ehdr
->e_entry
>= sechdrs
[i
].sh_addr
&&
915 pi
->ehdr
->e_entry
< (sechdrs
[i
].sh_addr
916 + sechdrs
[i
].sh_size
)) {
917 kbuf
->image
->start
-= sechdrs
[i
].sh_addr
;
918 kbuf
->image
->start
+= kbuf
->mem
+ offset
;
921 src
= (void *)pi
->ehdr
+ sechdrs
[i
].sh_offset
;
922 dst
= pi
->purgatory_buf
+ offset
;
923 memcpy(dst
, src
, sechdrs
[i
].sh_size
);
925 sechdrs
[i
].sh_addr
= kbuf
->mem
+ offset
;
926 sechdrs
[i
].sh_offset
= offset
;
927 offset
+= sechdrs
[i
].sh_size
;
933 static int kexec_apply_relocations(struct kimage
*image
)
936 struct purgatory_info
*pi
= &image
->purgatory_info
;
937 const Elf_Shdr
*sechdrs
;
939 sechdrs
= (void *)pi
->ehdr
+ pi
->ehdr
->e_shoff
;
941 for (i
= 0; i
< pi
->ehdr
->e_shnum
; i
++) {
942 const Elf_Shdr
*relsec
;
943 const Elf_Shdr
*symtab
;
946 relsec
= sechdrs
+ i
;
948 if (relsec
->sh_type
!= SHT_RELA
&&
949 relsec
->sh_type
!= SHT_REL
)
953 * For section of type SHT_RELA/SHT_REL,
954 * ->sh_link contains section header index of associated
955 * symbol table. And ->sh_info contains section header
956 * index of section to which relocations apply.
958 if (relsec
->sh_info
>= pi
->ehdr
->e_shnum
||
959 relsec
->sh_link
>= pi
->ehdr
->e_shnum
)
962 section
= pi
->sechdrs
+ relsec
->sh_info
;
963 symtab
= sechdrs
+ relsec
->sh_link
;
965 if (!(section
->sh_flags
& SHF_ALLOC
))
969 * symtab->sh_link contain section header index of associated
972 if (symtab
->sh_link
>= pi
->ehdr
->e_shnum
)
973 /* Invalid section number? */
977 * Respective architecture needs to provide support for applying
978 * relocations of type SHT_RELA/SHT_REL.
980 if (relsec
->sh_type
== SHT_RELA
)
981 ret
= arch_kexec_apply_relocations_add(pi
, section
,
983 else if (relsec
->sh_type
== SHT_REL
)
984 ret
= arch_kexec_apply_relocations(pi
, section
,
994 * kexec_load_purgatory - Load and relocate the purgatory object.
995 * @image: Image to add the purgatory to.
996 * @kbuf: Memory parameters to use.
998 * Allocates the memory needed for image->purgatory_info.sechdrs and
999 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1000 * to free the memory after use.
1002 * Return: 0 on success, negative errno on error.
1004 int kexec_load_purgatory(struct kimage
*image
, struct kexec_buf
*kbuf
)
1006 struct purgatory_info
*pi
= &image
->purgatory_info
;
1009 if (kexec_purgatory_size
<= 0)
1012 pi
->ehdr
= (const Elf_Ehdr
*)kexec_purgatory
;
1014 ret
= kexec_purgatory_setup_kbuf(pi
, kbuf
);
1018 ret
= kexec_purgatory_setup_sechdrs(pi
, kbuf
);
1022 ret
= kexec_apply_relocations(image
);
1031 vfree(pi
->purgatory_buf
);
1032 pi
->purgatory_buf
= NULL
;
1037 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1038 * @pi: Purgatory to search in.
1039 * @name: Name of the symbol.
1041 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1043 static const Elf_Sym
*kexec_purgatory_find_symbol(struct purgatory_info
*pi
,
1046 const Elf_Shdr
*sechdrs
;
1047 const Elf_Ehdr
*ehdr
;
1048 const Elf_Sym
*syms
;
1056 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
1058 for (i
= 0; i
< ehdr
->e_shnum
; i
++) {
1059 if (sechdrs
[i
].sh_type
!= SHT_SYMTAB
)
1062 if (sechdrs
[i
].sh_link
>= ehdr
->e_shnum
)
1063 /* Invalid strtab section number */
1065 strtab
= (void *)ehdr
+ sechdrs
[sechdrs
[i
].sh_link
].sh_offset
;
1066 syms
= (void *)ehdr
+ sechdrs
[i
].sh_offset
;
1068 /* Go through symbols for a match */
1069 for (k
= 0; k
< sechdrs
[i
].sh_size
/sizeof(Elf_Sym
); k
++) {
1070 if (ELF_ST_BIND(syms
[k
].st_info
) != STB_GLOBAL
)
1073 if (strcmp(strtab
+ syms
[k
].st_name
, name
) != 0)
1076 if (syms
[k
].st_shndx
== SHN_UNDEF
||
1077 syms
[k
].st_shndx
>= ehdr
->e_shnum
) {
1078 pr_debug("Symbol: %s has bad section index %d.\n",
1079 name
, syms
[k
].st_shndx
);
1083 /* Found the symbol we are looking for */
1091 void *kexec_purgatory_get_symbol_addr(struct kimage
*image
, const char *name
)
1093 struct purgatory_info
*pi
= &image
->purgatory_info
;
1097 sym
= kexec_purgatory_find_symbol(pi
, name
);
1099 return ERR_PTR(-EINVAL
);
1101 sechdr
= &pi
->sechdrs
[sym
->st_shndx
];
1104 * Returns the address where symbol will finally be loaded after
1105 * kexec_load_segment()
1107 return (void *)(sechdr
->sh_addr
+ sym
->st_value
);
1111 * Get or set value of a symbol. If "get_value" is true, symbol value is
1112 * returned in buf otherwise symbol value is set based on value in buf.
1114 int kexec_purgatory_get_set_symbol(struct kimage
*image
, const char *name
,
1115 void *buf
, unsigned int size
, bool get_value
)
1117 struct purgatory_info
*pi
= &image
->purgatory_info
;
1122 sym
= kexec_purgatory_find_symbol(pi
, name
);
1126 if (sym
->st_size
!= size
) {
1127 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1128 name
, (unsigned long)sym
->st_size
, size
);
1132 sec
= pi
->sechdrs
+ sym
->st_shndx
;
1134 if (sec
->sh_type
== SHT_NOBITS
) {
1135 pr_err("symbol %s is in a bss section. Cannot %s\n", name
,
1136 get_value
? "get" : "set");
1140 sym_buf
= (char *)pi
->purgatory_buf
+ sec
->sh_offset
+ sym
->st_value
;
1143 memcpy((void *)buf
, sym_buf
, size
);
1145 memcpy((void *)sym_buf
, buf
, size
);
1149 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1151 int crash_exclude_mem_range(struct crash_mem
*mem
,
1152 unsigned long long mstart
, unsigned long long mend
)
1155 unsigned long long start
, end
, p_start
, p_end
;
1156 struct crash_mem_range temp_range
= {0, 0};
1158 for (i
= 0; i
< mem
->nr_ranges
; i
++) {
1159 start
= mem
->ranges
[i
].start
;
1160 end
= mem
->ranges
[i
].end
;
1164 if (mstart
> end
|| mend
< start
)
1167 /* Truncate any area outside of range */
1173 /* Found completely overlapping range */
1174 if (p_start
== start
&& p_end
== end
) {
1175 mem
->ranges
[i
].start
= 0;
1176 mem
->ranges
[i
].end
= 0;
1177 if (i
< mem
->nr_ranges
- 1) {
1178 /* Shift rest of the ranges to left */
1179 for (j
= i
; j
< mem
->nr_ranges
- 1; j
++) {
1180 mem
->ranges
[j
].start
=
1181 mem
->ranges
[j
+1].start
;
1182 mem
->ranges
[j
].end
=
1183 mem
->ranges
[j
+1].end
;
1187 * Continue to check if there are another overlapping ranges
1188 * from the current position because of shifting the above
1199 if (p_start
> start
&& p_end
< end
) {
1200 /* Split original range */
1201 mem
->ranges
[i
].end
= p_start
- 1;
1202 temp_range
.start
= p_end
+ 1;
1203 temp_range
.end
= end
;
1204 } else if (p_start
!= start
)
1205 mem
->ranges
[i
].end
= p_start
- 1;
1207 mem
->ranges
[i
].start
= p_end
+ 1;
1211 /* If a split happened, add the split to array */
1212 if (!temp_range
.end
)
1215 /* Split happened */
1216 if (i
== mem
->max_nr_ranges
- 1)
1219 /* Location where new range should go */
1221 if (j
< mem
->nr_ranges
) {
1222 /* Move over all ranges one slot towards the end */
1223 for (i
= mem
->nr_ranges
- 1; i
>= j
; i
--)
1224 mem
->ranges
[i
+ 1] = mem
->ranges
[i
];
1227 mem
->ranges
[j
].start
= temp_range
.start
;
1228 mem
->ranges
[j
].end
= temp_range
.end
;
1233 int crash_prepare_elf64_headers(struct crash_mem
*mem
, int kernel_map
,
1234 void **addr
, unsigned long *sz
)
1238 unsigned long nr_cpus
= num_possible_cpus(), nr_phdr
, elf_sz
;
1240 unsigned int cpu
, i
;
1241 unsigned long long notes_addr
;
1242 unsigned long mstart
, mend
;
1244 /* extra phdr for vmcoreinfo ELF note */
1245 nr_phdr
= nr_cpus
+ 1;
1246 nr_phdr
+= mem
->nr_ranges
;
1249 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1250 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1251 * I think this is required by tools like gdb. So same physical
1252 * memory will be mapped in two ELF headers. One will contain kernel
1253 * text virtual addresses and other will have __va(physical) addresses.
1257 elf_sz
= sizeof(Elf64_Ehdr
) + nr_phdr
* sizeof(Elf64_Phdr
);
1258 elf_sz
= ALIGN(elf_sz
, ELF_CORE_HEADER_ALIGN
);
1260 buf
= vzalloc(elf_sz
);
1264 ehdr
= (Elf64_Ehdr
*)buf
;
1265 phdr
= (Elf64_Phdr
*)(ehdr
+ 1);
1266 memcpy(ehdr
->e_ident
, ELFMAG
, SELFMAG
);
1267 ehdr
->e_ident
[EI_CLASS
] = ELFCLASS64
;
1268 ehdr
->e_ident
[EI_DATA
] = ELFDATA2LSB
;
1269 ehdr
->e_ident
[EI_VERSION
] = EV_CURRENT
;
1270 ehdr
->e_ident
[EI_OSABI
] = ELF_OSABI
;
1271 memset(ehdr
->e_ident
+ EI_PAD
, 0, EI_NIDENT
- EI_PAD
);
1272 ehdr
->e_type
= ET_CORE
;
1273 ehdr
->e_machine
= ELF_ARCH
;
1274 ehdr
->e_version
= EV_CURRENT
;
1275 ehdr
->e_phoff
= sizeof(Elf64_Ehdr
);
1276 ehdr
->e_ehsize
= sizeof(Elf64_Ehdr
);
1277 ehdr
->e_phentsize
= sizeof(Elf64_Phdr
);
1279 /* Prepare one phdr of type PT_NOTE for each present CPU */
1280 for_each_present_cpu(cpu
) {
1281 phdr
->p_type
= PT_NOTE
;
1282 notes_addr
= per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes
, cpu
));
1283 phdr
->p_offset
= phdr
->p_paddr
= notes_addr
;
1284 phdr
->p_filesz
= phdr
->p_memsz
= sizeof(note_buf_t
);
1289 /* Prepare one PT_NOTE header for vmcoreinfo */
1290 phdr
->p_type
= PT_NOTE
;
1291 phdr
->p_offset
= phdr
->p_paddr
= paddr_vmcoreinfo_note();
1292 phdr
->p_filesz
= phdr
->p_memsz
= VMCOREINFO_NOTE_SIZE
;
1296 /* Prepare PT_LOAD type program header for kernel text region */
1298 phdr
->p_type
= PT_LOAD
;
1299 phdr
->p_flags
= PF_R
|PF_W
|PF_X
;
1300 phdr
->p_vaddr
= (unsigned long) _text
;
1301 phdr
->p_filesz
= phdr
->p_memsz
= _end
- _text
;
1302 phdr
->p_offset
= phdr
->p_paddr
= __pa_symbol(_text
);
1307 /* Go through all the ranges in mem->ranges[] and prepare phdr */
1308 for (i
= 0; i
< mem
->nr_ranges
; i
++) {
1309 mstart
= mem
->ranges
[i
].start
;
1310 mend
= mem
->ranges
[i
].end
;
1312 phdr
->p_type
= PT_LOAD
;
1313 phdr
->p_flags
= PF_R
|PF_W
|PF_X
;
1314 phdr
->p_offset
= mstart
;
1316 phdr
->p_paddr
= mstart
;
1317 phdr
->p_vaddr
= (unsigned long) __va(mstart
);
1318 phdr
->p_filesz
= phdr
->p_memsz
= mend
- mstart
+ 1;
1321 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",
1322 phdr
, phdr
->p_vaddr
, phdr
->p_paddr
, phdr
->p_filesz
,
1323 ehdr
->e_phnum
, phdr
->p_offset
);