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
);
43 /* Maximum size in bytes for kernel/initrd files. */
44 #define KEXEC_FILE_SIZE_MAX min_t(s64, 4LL << 30, SSIZE_MAX)
47 * Currently this is the only default function that is exported as some
48 * architectures need it to do additional handlings.
49 * In the future, other default functions may be exported too if required.
51 int kexec_image_probe_default(struct kimage
*image
, void *buf
,
52 unsigned long buf_len
)
54 const struct kexec_file_ops
* const *fops
;
57 for (fops
= &kexec_file_loaders
[0]; *fops
&& (*fops
)->probe
; ++fops
) {
58 ret
= (*fops
)->probe(buf
, buf_len
);
68 void *kexec_image_load_default(struct kimage
*image
)
70 if (!image
->fops
|| !image
->fops
->load
)
71 return ERR_PTR(-ENOEXEC
);
73 return image
->fops
->load(image
, image
->kernel_buf
,
74 image
->kernel_buf_len
, image
->initrd_buf
,
75 image
->initrd_buf_len
, image
->cmdline_buf
,
76 image
->cmdline_buf_len
);
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
);
88 * Free up memory used by kernel, initrd, and command line. This is temporary
89 * memory allocation which is not needed any more after these buffers have
90 * been loaded into separate segments and have been copied elsewhere.
92 void kimage_file_post_load_cleanup(struct kimage
*image
)
94 struct purgatory_info
*pi
= &image
->purgatory_info
;
96 vfree(image
->kernel_buf
);
97 image
->kernel_buf
= NULL
;
99 vfree(image
->initrd_buf
);
100 image
->initrd_buf
= NULL
;
102 kfree(image
->cmdline_buf
);
103 image
->cmdline_buf
= NULL
;
105 vfree(pi
->purgatory_buf
);
106 pi
->purgatory_buf
= NULL
;
111 #ifdef CONFIG_IMA_KEXEC
112 vfree(image
->ima_buffer
);
113 image
->ima_buffer
= NULL
;
114 #endif /* CONFIG_IMA_KEXEC */
116 /* See if architecture has anything to cleanup post load */
117 arch_kimage_file_post_load_cleanup(image
);
120 * Above call should have called into bootloader to free up
121 * any data stored in kimage->image_loader_data. It should
122 * be ok now to free it up.
124 kfree(image
->image_loader_data
);
125 image
->image_loader_data
= NULL
;
128 #ifdef CONFIG_KEXEC_SIG
129 #ifdef CONFIG_SIGNED_PE_FILE_VERIFICATION
130 int kexec_kernel_verify_pe_sig(const char *kernel
, unsigned long kernel_len
)
134 ret
= verify_pefile_signature(kernel
, kernel_len
,
135 VERIFY_USE_SECONDARY_KEYRING
,
136 VERIFYING_KEXEC_PE_SIGNATURE
);
137 if (ret
== -ENOKEY
&& IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING
)) {
138 ret
= verify_pefile_signature(kernel
, kernel_len
,
139 VERIFY_USE_PLATFORM_KEYRING
,
140 VERIFYING_KEXEC_PE_SIGNATURE
);
146 static int kexec_image_verify_sig(struct kimage
*image
, void *buf
,
147 unsigned long buf_len
)
149 if (!image
->fops
|| !image
->fops
->verify_sig
) {
150 pr_debug("kernel loader does not support signature verification.\n");
151 return -EKEYREJECTED
;
154 return image
->fops
->verify_sig(buf
, buf_len
);
158 kimage_validate_signature(struct kimage
*image
)
162 ret
= kexec_image_verify_sig(image
, image
->kernel_buf
,
163 image
->kernel_buf_len
);
167 pr_notice("Enforced kernel signature verification failed (%d).\n", ret
);
172 * If IMA is guaranteed to appraise a signature on the kexec
173 * image, permit it even if the kernel is otherwise locked
176 if (!ima_appraise_signature(READING_KEXEC_IMAGE
) &&
177 security_locked_down(LOCKDOWN_KEXEC
))
180 pr_debug("kernel signature verification failed (%d).\n", ret
);
188 * In file mode list of segments is prepared by kernel. Copy relevant
189 * data from user space, do error checking, prepare segment list
192 kimage_file_prepare_segments(struct kimage
*image
, int kernel_fd
, int initrd_fd
,
193 const char __user
*cmdline_ptr
,
194 unsigned long cmdline_len
, unsigned flags
)
199 ret
= kernel_read_file_from_fd(kernel_fd
, 0, &image
->kernel_buf
,
200 KEXEC_FILE_SIZE_MAX
, NULL
,
201 READING_KEXEC_IMAGE
);
204 image
->kernel_buf_len
= ret
;
206 /* Call arch image probe handlers */
207 ret
= arch_kexec_kernel_image_probe(image
, image
->kernel_buf
,
208 image
->kernel_buf_len
);
212 #ifdef CONFIG_KEXEC_SIG
213 ret
= kimage_validate_signature(image
);
218 /* It is possible that there no initramfs is being loaded */
219 if (!(flags
& KEXEC_FILE_NO_INITRAMFS
)) {
220 ret
= kernel_read_file_from_fd(initrd_fd
, 0, &image
->initrd_buf
,
221 KEXEC_FILE_SIZE_MAX
, NULL
,
222 READING_KEXEC_INITRAMFS
);
225 image
->initrd_buf_len
= ret
;
230 image
->cmdline_buf
= memdup_user(cmdline_ptr
, cmdline_len
);
231 if (IS_ERR(image
->cmdline_buf
)) {
232 ret
= PTR_ERR(image
->cmdline_buf
);
233 image
->cmdline_buf
= NULL
;
237 image
->cmdline_buf_len
= cmdline_len
;
239 /* command line should be a string with last byte null */
240 if (image
->cmdline_buf
[cmdline_len
- 1] != '\0') {
245 ima_kexec_cmdline(kernel_fd
, image
->cmdline_buf
,
246 image
->cmdline_buf_len
- 1);
249 /* IMA needs to pass the measurement list to the next kernel. */
250 ima_add_kexec_buffer(image
);
252 /* Call arch image load handlers */
253 ldata
= arch_kexec_kernel_image_load(image
);
256 ret
= PTR_ERR(ldata
);
260 image
->image_loader_data
= ldata
;
262 /* In case of error, free up all allocated memory in this function */
264 kimage_file_post_load_cleanup(image
);
269 kimage_file_alloc_init(struct kimage
**rimage
, int kernel_fd
,
270 int initrd_fd
, const char __user
*cmdline_ptr
,
271 unsigned long cmdline_len
, unsigned long flags
)
274 struct kimage
*image
;
275 bool kexec_on_panic
= flags
& KEXEC_FILE_ON_CRASH
;
277 image
= do_kimage_alloc_init();
281 image
->file_mode
= 1;
283 if (kexec_on_panic
) {
284 /* Enable special crash kernel control page alloc policy. */
285 image
->control_page
= crashk_res
.start
;
286 image
->type
= KEXEC_TYPE_CRASH
;
289 ret
= kimage_file_prepare_segments(image
, kernel_fd
, initrd_fd
,
290 cmdline_ptr
, cmdline_len
, flags
);
294 ret
= sanity_check_segment_list(image
);
296 goto out_free_post_load_bufs
;
299 image
->control_code_page
= kimage_alloc_control_pages(image
,
300 get_order(KEXEC_CONTROL_PAGE_SIZE
));
301 if (!image
->control_code_page
) {
302 pr_err("Could not allocate control_code_buffer\n");
303 goto out_free_post_load_bufs
;
306 if (!kexec_on_panic
) {
307 image
->swap_page
= kimage_alloc_control_pages(image
, 0);
308 if (!image
->swap_page
) {
309 pr_err("Could not allocate swap buffer\n");
310 goto out_free_control_pages
;
316 out_free_control_pages
:
317 kimage_free_page_list(&image
->control_pages
);
318 out_free_post_load_bufs
:
319 kimage_file_post_load_cleanup(image
);
325 SYSCALL_DEFINE5(kexec_file_load
, int, kernel_fd
, int, initrd_fd
,
326 unsigned long, cmdline_len
, const char __user
*, cmdline_ptr
,
327 unsigned long, flags
)
330 struct kimage
**dest_image
, *image
;
332 /* We only trust the superuser with rebooting the system. */
333 if (!capable(CAP_SYS_BOOT
) || kexec_load_disabled
)
336 /* Make sure we have a legal set of flags */
337 if (flags
!= (flags
& KEXEC_FILE_FLAGS
))
342 if (!kexec_trylock())
345 dest_image
= &kexec_image
;
346 if (flags
& KEXEC_FILE_ON_CRASH
) {
347 dest_image
= &kexec_crash_image
;
348 if (kexec_crash_image
)
349 arch_kexec_unprotect_crashkres();
352 if (flags
& KEXEC_FILE_UNLOAD
)
356 * In case of crash, new kernel gets loaded in reserved region. It is
357 * same memory where old crash kernel might be loaded. Free any
358 * current crash dump kernel before we corrupt it.
360 if (flags
& KEXEC_FILE_ON_CRASH
)
361 kimage_free(xchg(&kexec_crash_image
, NULL
));
363 ret
= kimage_file_alloc_init(&image
, kernel_fd
, initrd_fd
, cmdline_ptr
,
368 ret
= machine_kexec_prepare(image
);
373 * Some architecture(like S390) may touch the crash memory before
374 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
376 ret
= kimage_crash_copy_vmcoreinfo(image
);
380 ret
= kexec_calculate_store_digests(image
);
384 for (i
= 0; i
< image
->nr_segments
; i
++) {
385 struct kexec_segment
*ksegment
;
387 ksegment
= &image
->segment
[i
];
388 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
389 i
, ksegment
->buf
, ksegment
->bufsz
, ksegment
->mem
,
392 ret
= kimage_load_segment(image
, &image
->segment
[i
]);
397 kimage_terminate(image
);
399 ret
= machine_kexec_post_load(image
);
404 * Free up any temporary buffers allocated which are not needed
405 * after image has been loaded
407 kimage_file_post_load_cleanup(image
);
409 image
= xchg(dest_image
, image
);
411 if ((flags
& KEXEC_FILE_ON_CRASH
) && kexec_crash_image
)
412 arch_kexec_protect_crashkres();
419 static int locate_mem_hole_top_down(unsigned long start
, unsigned long end
,
420 struct kexec_buf
*kbuf
)
422 struct kimage
*image
= kbuf
->image
;
423 unsigned long temp_start
, temp_end
;
425 temp_end
= min(end
, kbuf
->buf_max
);
426 temp_start
= temp_end
- kbuf
->memsz
;
429 /* align down start */
430 temp_start
= temp_start
& (~(kbuf
->buf_align
- 1));
432 if (temp_start
< start
|| temp_start
< kbuf
->buf_min
)
435 temp_end
= temp_start
+ kbuf
->memsz
- 1;
438 * Make sure this does not conflict with any of existing
441 if (kimage_is_destination_range(image
, temp_start
, temp_end
)) {
442 temp_start
= temp_start
- PAGE_SIZE
;
446 /* We found a suitable memory range */
450 /* If we are here, we found a suitable memory range */
451 kbuf
->mem
= temp_start
;
453 /* Success, stop navigating through remaining System RAM ranges */
457 static int locate_mem_hole_bottom_up(unsigned long start
, unsigned long end
,
458 struct kexec_buf
*kbuf
)
460 struct kimage
*image
= kbuf
->image
;
461 unsigned long temp_start
, temp_end
;
463 temp_start
= max(start
, kbuf
->buf_min
);
466 temp_start
= ALIGN(temp_start
, kbuf
->buf_align
);
467 temp_end
= temp_start
+ kbuf
->memsz
- 1;
469 if (temp_end
> end
|| temp_end
> kbuf
->buf_max
)
472 * Make sure this does not conflict with any of existing
475 if (kimage_is_destination_range(image
, temp_start
, temp_end
)) {
476 temp_start
= temp_start
+ PAGE_SIZE
;
480 /* We found a suitable memory range */
484 /* If we are here, we found a suitable memory range */
485 kbuf
->mem
= temp_start
;
487 /* Success, stop navigating through remaining System RAM ranges */
491 static int locate_mem_hole_callback(struct resource
*res
, void *arg
)
493 struct kexec_buf
*kbuf
= (struct kexec_buf
*)arg
;
494 u64 start
= res
->start
, end
= res
->end
;
495 unsigned long sz
= end
- start
+ 1;
497 /* Returning 0 will take to next memory range */
499 /* Don't use memory that will be detected and handled by a driver. */
500 if (res
->flags
& IORESOURCE_SYSRAM_DRIVER_MANAGED
)
503 if (sz
< kbuf
->memsz
)
506 if (end
< kbuf
->buf_min
|| start
> kbuf
->buf_max
)
510 * Allocate memory top down with-in ram range. Otherwise bottom up
514 return locate_mem_hole_top_down(start
, end
, kbuf
);
515 return locate_mem_hole_bottom_up(start
, end
, kbuf
);
518 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
519 static int kexec_walk_memblock(struct kexec_buf
*kbuf
,
520 int (*func
)(struct resource
*, void *))
524 phys_addr_t mstart
, mend
;
525 struct resource res
= { };
527 if (kbuf
->image
->type
== KEXEC_TYPE_CRASH
)
528 return func(&crashk_res
, kbuf
);
531 * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
532 * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
533 * locate_mem_hole_callback().
535 if (kbuf
->top_down
) {
536 for_each_free_mem_range_reverse(i
, NUMA_NO_NODE
, MEMBLOCK_NONE
,
537 &mstart
, &mend
, NULL
) {
539 * In memblock, end points to the first byte after the
540 * range while in kexec, end points to the last byte
545 ret
= func(&res
, kbuf
);
550 for_each_free_mem_range(i
, NUMA_NO_NODE
, MEMBLOCK_NONE
,
551 &mstart
, &mend
, NULL
) {
553 * In memblock, end points to the first byte after the
554 * range while in kexec, end points to the last byte
559 ret
= func(&res
, kbuf
);
568 static int kexec_walk_memblock(struct kexec_buf
*kbuf
,
569 int (*func
)(struct resource
*, void *))
576 * kexec_walk_resources - call func(data) on free memory regions
577 * @kbuf: Context info for the search. Also passed to @func.
578 * @func: Function to call for each memory region.
580 * Return: The memory walk will stop when func returns a non-zero value
581 * and that value will be returned. If all free regions are visited without
582 * func returning non-zero, then zero will be returned.
584 static int kexec_walk_resources(struct kexec_buf
*kbuf
,
585 int (*func
)(struct resource
*, void *))
587 if (kbuf
->image
->type
== KEXEC_TYPE_CRASH
)
588 return walk_iomem_res_desc(crashk_res
.desc
,
589 IORESOURCE_SYSTEM_RAM
| IORESOURCE_BUSY
,
590 crashk_res
.start
, crashk_res
.end
,
593 return walk_system_ram_res(0, ULONG_MAX
, kbuf
, func
);
597 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
598 * @kbuf: Parameters for the memory search.
600 * On success, kbuf->mem will have the start address of the memory region found.
602 * Return: 0 on success, negative errno on error.
604 int kexec_locate_mem_hole(struct kexec_buf
*kbuf
)
608 /* Arch knows where to place */
609 if (kbuf
->mem
!= KEXEC_BUF_MEM_UNKNOWN
)
612 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK
))
613 ret
= kexec_walk_resources(kbuf
, locate_mem_hole_callback
);
615 ret
= kexec_walk_memblock(kbuf
, locate_mem_hole_callback
);
617 return ret
== 1 ? 0 : -EADDRNOTAVAIL
;
621 * kexec_add_buffer - place a buffer in a kexec segment
622 * @kbuf: Buffer contents and memory parameters.
624 * This function assumes that kexec_mutex is held.
625 * On successful return, @kbuf->mem will have the physical address of
626 * the buffer in memory.
628 * Return: 0 on success, negative errno on error.
630 int kexec_add_buffer(struct kexec_buf
*kbuf
)
632 struct kexec_segment
*ksegment
;
635 /* Currently adding segment this way is allowed only in file mode */
636 if (!kbuf
->image
->file_mode
)
639 if (kbuf
->image
->nr_segments
>= KEXEC_SEGMENT_MAX
)
643 * Make sure we are not trying to add buffer after allocating
644 * control pages. All segments need to be placed first before
645 * any control pages are allocated. As control page allocation
646 * logic goes through list of segments to make sure there are
647 * no destination overlaps.
649 if (!list_empty(&kbuf
->image
->control_pages
)) {
654 /* Ensure minimum alignment needed for segments. */
655 kbuf
->memsz
= ALIGN(kbuf
->memsz
, PAGE_SIZE
);
656 kbuf
->buf_align
= max(kbuf
->buf_align
, PAGE_SIZE
);
658 /* Walk the RAM ranges and allocate a suitable range for the buffer */
659 ret
= arch_kexec_locate_mem_hole(kbuf
);
663 /* Found a suitable memory range */
664 ksegment
= &kbuf
->image
->segment
[kbuf
->image
->nr_segments
];
665 ksegment
->kbuf
= kbuf
->buffer
;
666 ksegment
->bufsz
= kbuf
->bufsz
;
667 ksegment
->mem
= kbuf
->mem
;
668 ksegment
->memsz
= kbuf
->memsz
;
669 kbuf
->image
->nr_segments
++;
673 /* Calculate and store the digest of segments */
674 static int kexec_calculate_store_digests(struct kimage
*image
)
676 struct crypto_shash
*tfm
;
677 struct shash_desc
*desc
;
678 int ret
= 0, i
, j
, zero_buf_sz
, sha_region_sz
;
679 size_t desc_size
, nullsz
;
682 struct kexec_sha_region
*sha_regions
;
683 struct purgatory_info
*pi
= &image
->purgatory_info
;
685 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY
))
688 zero_buf
= __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT
);
689 zero_buf_sz
= PAGE_SIZE
;
691 tfm
= crypto_alloc_shash("sha256", 0, 0);
697 desc_size
= crypto_shash_descsize(tfm
) + sizeof(*desc
);
698 desc
= kzalloc(desc_size
, GFP_KERNEL
);
704 sha_region_sz
= KEXEC_SEGMENT_MAX
* sizeof(struct kexec_sha_region
);
705 sha_regions
= vzalloc(sha_region_sz
);
713 ret
= crypto_shash_init(desc
);
715 goto out_free_sha_regions
;
717 digest
= kzalloc(SHA256_DIGEST_SIZE
, GFP_KERNEL
);
720 goto out_free_sha_regions
;
723 for (j
= i
= 0; i
< image
->nr_segments
; i
++) {
724 struct kexec_segment
*ksegment
;
726 ksegment
= &image
->segment
[i
];
728 * Skip purgatory as it will be modified once we put digest
731 if (ksegment
->kbuf
== pi
->purgatory_buf
)
734 ret
= crypto_shash_update(desc
, ksegment
->kbuf
,
740 * Assume rest of the buffer is filled with zero and
741 * update digest accordingly.
743 nullsz
= ksegment
->memsz
- ksegment
->bufsz
;
745 unsigned long bytes
= nullsz
;
747 if (bytes
> zero_buf_sz
)
749 ret
= crypto_shash_update(desc
, zero_buf
, bytes
);
758 sha_regions
[j
].start
= ksegment
->mem
;
759 sha_regions
[j
].len
= ksegment
->memsz
;
764 ret
= crypto_shash_final(desc
, digest
);
766 goto out_free_digest
;
767 ret
= kexec_purgatory_get_set_symbol(image
, "purgatory_sha_regions",
768 sha_regions
, sha_region_sz
, 0);
770 goto out_free_digest
;
772 ret
= kexec_purgatory_get_set_symbol(image
, "purgatory_sha256_digest",
773 digest
, SHA256_DIGEST_SIZE
, 0);
775 goto out_free_digest
;
780 out_free_sha_regions
:
790 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
792 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
793 * @pi: Purgatory to be loaded.
794 * @kbuf: Buffer to setup.
796 * Allocates the memory needed for the buffer. Caller is responsible to free
797 * the memory after use.
799 * Return: 0 on success, negative errno on error.
801 static int kexec_purgatory_setup_kbuf(struct purgatory_info
*pi
,
802 struct kexec_buf
*kbuf
)
804 const Elf_Shdr
*sechdrs
;
805 unsigned long bss_align
;
806 unsigned long bss_sz
;
810 sechdrs
= (void *)pi
->ehdr
+ pi
->ehdr
->e_shoff
;
811 kbuf
->buf_align
= bss_align
= 1;
812 kbuf
->bufsz
= bss_sz
= 0;
814 for (i
= 0; i
< pi
->ehdr
->e_shnum
; i
++) {
815 if (!(sechdrs
[i
].sh_flags
& SHF_ALLOC
))
818 align
= sechdrs
[i
].sh_addralign
;
819 if (sechdrs
[i
].sh_type
!= SHT_NOBITS
) {
820 if (kbuf
->buf_align
< align
)
821 kbuf
->buf_align
= align
;
822 kbuf
->bufsz
= ALIGN(kbuf
->bufsz
, align
);
823 kbuf
->bufsz
+= sechdrs
[i
].sh_size
;
825 if (bss_align
< align
)
827 bss_sz
= ALIGN(bss_sz
, align
);
828 bss_sz
+= sechdrs
[i
].sh_size
;
831 kbuf
->bufsz
= ALIGN(kbuf
->bufsz
, bss_align
);
832 kbuf
->memsz
= kbuf
->bufsz
+ bss_sz
;
833 if (kbuf
->buf_align
< bss_align
)
834 kbuf
->buf_align
= bss_align
;
836 kbuf
->buffer
= vzalloc(kbuf
->bufsz
);
839 pi
->purgatory_buf
= kbuf
->buffer
;
841 ret
= kexec_add_buffer(kbuf
);
847 vfree(pi
->purgatory_buf
);
848 pi
->purgatory_buf
= NULL
;
853 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
854 * @pi: Purgatory to be loaded.
855 * @kbuf: Buffer prepared to store purgatory.
857 * Allocates the memory needed for the buffer. Caller is responsible to free
858 * the memory after use.
860 * Return: 0 on success, negative errno on error.
862 static int kexec_purgatory_setup_sechdrs(struct purgatory_info
*pi
,
863 struct kexec_buf
*kbuf
)
865 unsigned long bss_addr
;
866 unsigned long offset
;
871 * The section headers in kexec_purgatory are read-only. In order to
872 * have them modifiable make a temporary copy.
874 sechdrs
= vzalloc(array_size(sizeof(Elf_Shdr
), pi
->ehdr
->e_shnum
));
877 memcpy(sechdrs
, (void *)pi
->ehdr
+ pi
->ehdr
->e_shoff
,
878 pi
->ehdr
->e_shnum
* sizeof(Elf_Shdr
));
879 pi
->sechdrs
= sechdrs
;
882 bss_addr
= kbuf
->mem
+ kbuf
->bufsz
;
883 kbuf
->image
->start
= pi
->ehdr
->e_entry
;
885 for (i
= 0; i
< pi
->ehdr
->e_shnum
; i
++) {
889 if (!(sechdrs
[i
].sh_flags
& SHF_ALLOC
))
892 align
= sechdrs
[i
].sh_addralign
;
893 if (sechdrs
[i
].sh_type
== SHT_NOBITS
) {
894 bss_addr
= ALIGN(bss_addr
, align
);
895 sechdrs
[i
].sh_addr
= bss_addr
;
896 bss_addr
+= sechdrs
[i
].sh_size
;
900 offset
= ALIGN(offset
, align
);
901 if (sechdrs
[i
].sh_flags
& SHF_EXECINSTR
&&
902 pi
->ehdr
->e_entry
>= sechdrs
[i
].sh_addr
&&
903 pi
->ehdr
->e_entry
< (sechdrs
[i
].sh_addr
904 + sechdrs
[i
].sh_size
)) {
905 kbuf
->image
->start
-= sechdrs
[i
].sh_addr
;
906 kbuf
->image
->start
+= kbuf
->mem
+ offset
;
909 src
= (void *)pi
->ehdr
+ sechdrs
[i
].sh_offset
;
910 dst
= pi
->purgatory_buf
+ offset
;
911 memcpy(dst
, src
, sechdrs
[i
].sh_size
);
913 sechdrs
[i
].sh_addr
= kbuf
->mem
+ offset
;
914 sechdrs
[i
].sh_offset
= offset
;
915 offset
+= sechdrs
[i
].sh_size
;
921 static int kexec_apply_relocations(struct kimage
*image
)
924 struct purgatory_info
*pi
= &image
->purgatory_info
;
925 const Elf_Shdr
*sechdrs
;
927 sechdrs
= (void *)pi
->ehdr
+ pi
->ehdr
->e_shoff
;
929 for (i
= 0; i
< pi
->ehdr
->e_shnum
; i
++) {
930 const Elf_Shdr
*relsec
;
931 const Elf_Shdr
*symtab
;
934 relsec
= sechdrs
+ i
;
936 if (relsec
->sh_type
!= SHT_RELA
&&
937 relsec
->sh_type
!= SHT_REL
)
941 * For section of type SHT_RELA/SHT_REL,
942 * ->sh_link contains section header index of associated
943 * symbol table. And ->sh_info contains section header
944 * index of section to which relocations apply.
946 if (relsec
->sh_info
>= pi
->ehdr
->e_shnum
||
947 relsec
->sh_link
>= pi
->ehdr
->e_shnum
)
950 section
= pi
->sechdrs
+ relsec
->sh_info
;
951 symtab
= sechdrs
+ relsec
->sh_link
;
953 if (!(section
->sh_flags
& SHF_ALLOC
))
957 * symtab->sh_link contain section header index of associated
960 if (symtab
->sh_link
>= pi
->ehdr
->e_shnum
)
961 /* Invalid section number? */
965 * Respective architecture needs to provide support for applying
966 * relocations of type SHT_RELA/SHT_REL.
968 if (relsec
->sh_type
== SHT_RELA
)
969 ret
= arch_kexec_apply_relocations_add(pi
, section
,
971 else if (relsec
->sh_type
== SHT_REL
)
972 ret
= arch_kexec_apply_relocations(pi
, section
,
982 * kexec_load_purgatory - Load and relocate the purgatory object.
983 * @image: Image to add the purgatory to.
984 * @kbuf: Memory parameters to use.
986 * Allocates the memory needed for image->purgatory_info.sechdrs and
987 * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
988 * to free the memory after use.
990 * Return: 0 on success, negative errno on error.
992 int kexec_load_purgatory(struct kimage
*image
, struct kexec_buf
*kbuf
)
994 struct purgatory_info
*pi
= &image
->purgatory_info
;
997 if (kexec_purgatory_size
<= 0)
1000 pi
->ehdr
= (const Elf_Ehdr
*)kexec_purgatory
;
1002 ret
= kexec_purgatory_setup_kbuf(pi
, kbuf
);
1006 ret
= kexec_purgatory_setup_sechdrs(pi
, kbuf
);
1010 ret
= kexec_apply_relocations(image
);
1019 vfree(pi
->purgatory_buf
);
1020 pi
->purgatory_buf
= NULL
;
1025 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1026 * @pi: Purgatory to search in.
1027 * @name: Name of the symbol.
1029 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1031 static const Elf_Sym
*kexec_purgatory_find_symbol(struct purgatory_info
*pi
,
1034 const Elf_Shdr
*sechdrs
;
1035 const Elf_Ehdr
*ehdr
;
1036 const Elf_Sym
*syms
;
1044 sechdrs
= (void *)ehdr
+ ehdr
->e_shoff
;
1046 for (i
= 0; i
< ehdr
->e_shnum
; i
++) {
1047 if (sechdrs
[i
].sh_type
!= SHT_SYMTAB
)
1050 if (sechdrs
[i
].sh_link
>= ehdr
->e_shnum
)
1051 /* Invalid strtab section number */
1053 strtab
= (void *)ehdr
+ sechdrs
[sechdrs
[i
].sh_link
].sh_offset
;
1054 syms
= (void *)ehdr
+ sechdrs
[i
].sh_offset
;
1056 /* Go through symbols for a match */
1057 for (k
= 0; k
< sechdrs
[i
].sh_size
/sizeof(Elf_Sym
); k
++) {
1058 if (ELF_ST_BIND(syms
[k
].st_info
) != STB_GLOBAL
)
1061 if (strcmp(strtab
+ syms
[k
].st_name
, name
) != 0)
1064 if (syms
[k
].st_shndx
== SHN_UNDEF
||
1065 syms
[k
].st_shndx
>= ehdr
->e_shnum
) {
1066 pr_debug("Symbol: %s has bad section index %d.\n",
1067 name
, syms
[k
].st_shndx
);
1071 /* Found the symbol we are looking for */
1079 void *kexec_purgatory_get_symbol_addr(struct kimage
*image
, const char *name
)
1081 struct purgatory_info
*pi
= &image
->purgatory_info
;
1085 sym
= kexec_purgatory_find_symbol(pi
, name
);
1087 return ERR_PTR(-EINVAL
);
1089 sechdr
= &pi
->sechdrs
[sym
->st_shndx
];
1092 * Returns the address where symbol will finally be loaded after
1093 * kexec_load_segment()
1095 return (void *)(sechdr
->sh_addr
+ sym
->st_value
);
1099 * Get or set value of a symbol. If "get_value" is true, symbol value is
1100 * returned in buf otherwise symbol value is set based on value in buf.
1102 int kexec_purgatory_get_set_symbol(struct kimage
*image
, const char *name
,
1103 void *buf
, unsigned int size
, bool get_value
)
1105 struct purgatory_info
*pi
= &image
->purgatory_info
;
1110 sym
= kexec_purgatory_find_symbol(pi
, name
);
1114 if (sym
->st_size
!= size
) {
1115 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1116 name
, (unsigned long)sym
->st_size
, size
);
1120 sec
= pi
->sechdrs
+ sym
->st_shndx
;
1122 if (sec
->sh_type
== SHT_NOBITS
) {
1123 pr_err("symbol %s is in a bss section. Cannot %s\n", name
,
1124 get_value
? "get" : "set");
1128 sym_buf
= (char *)pi
->purgatory_buf
+ sec
->sh_offset
+ sym
->st_value
;
1131 memcpy((void *)buf
, sym_buf
, size
);
1133 memcpy((void *)sym_buf
, buf
, size
);
1137 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1139 int crash_exclude_mem_range(struct crash_mem
*mem
,
1140 unsigned long long mstart
, unsigned long long mend
)
1143 unsigned long long start
, end
, p_start
, p_end
;
1144 struct range temp_range
= {0, 0};
1146 for (i
= 0; i
< mem
->nr_ranges
; i
++) {
1147 start
= mem
->ranges
[i
].start
;
1148 end
= mem
->ranges
[i
].end
;
1152 if (mstart
> end
|| mend
< start
)
1155 /* Truncate any area outside of range */
1161 /* Found completely overlapping range */
1162 if (p_start
== start
&& p_end
== end
) {
1163 mem
->ranges
[i
].start
= 0;
1164 mem
->ranges
[i
].end
= 0;
1165 if (i
< mem
->nr_ranges
- 1) {
1166 /* Shift rest of the ranges to left */
1167 for (j
= i
; j
< mem
->nr_ranges
- 1; j
++) {
1168 mem
->ranges
[j
].start
=
1169 mem
->ranges
[j
+1].start
;
1170 mem
->ranges
[j
].end
=
1171 mem
->ranges
[j
+1].end
;
1175 * Continue to check if there are another overlapping ranges
1176 * from the current position because of shifting the above
1187 if (p_start
> start
&& p_end
< end
) {
1188 /* Split original range */
1189 mem
->ranges
[i
].end
= p_start
- 1;
1190 temp_range
.start
= p_end
+ 1;
1191 temp_range
.end
= end
;
1192 } else if (p_start
!= start
)
1193 mem
->ranges
[i
].end
= p_start
- 1;
1195 mem
->ranges
[i
].start
= p_end
+ 1;
1199 /* If a split happened, add the split to array */
1200 if (!temp_range
.end
)
1203 /* Split happened */
1204 if (i
== mem
->max_nr_ranges
- 1)
1207 /* Location where new range should go */
1209 if (j
< mem
->nr_ranges
) {
1210 /* Move over all ranges one slot towards the end */
1211 for (i
= mem
->nr_ranges
- 1; i
>= j
; i
--)
1212 mem
->ranges
[i
+ 1] = mem
->ranges
[i
];
1215 mem
->ranges
[j
].start
= temp_range
.start
;
1216 mem
->ranges
[j
].end
= temp_range
.end
;
1221 int crash_prepare_elf64_headers(struct crash_mem
*mem
, int need_kernel_map
,
1222 void **addr
, unsigned long *sz
)
1226 unsigned long nr_cpus
= num_possible_cpus(), nr_phdr
, elf_sz
;
1228 unsigned int cpu
, i
;
1229 unsigned long long notes_addr
;
1230 unsigned long mstart
, mend
;
1232 /* extra phdr for vmcoreinfo ELF note */
1233 nr_phdr
= nr_cpus
+ 1;
1234 nr_phdr
+= mem
->nr_ranges
;
1237 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1238 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1239 * I think this is required by tools like gdb. So same physical
1240 * memory will be mapped in two ELF headers. One will contain kernel
1241 * text virtual addresses and other will have __va(physical) addresses.
1245 elf_sz
= sizeof(Elf64_Ehdr
) + nr_phdr
* sizeof(Elf64_Phdr
);
1246 elf_sz
= ALIGN(elf_sz
, ELF_CORE_HEADER_ALIGN
);
1248 buf
= vzalloc(elf_sz
);
1252 ehdr
= (Elf64_Ehdr
*)buf
;
1253 phdr
= (Elf64_Phdr
*)(ehdr
+ 1);
1254 memcpy(ehdr
->e_ident
, ELFMAG
, SELFMAG
);
1255 ehdr
->e_ident
[EI_CLASS
] = ELFCLASS64
;
1256 ehdr
->e_ident
[EI_DATA
] = ELFDATA2LSB
;
1257 ehdr
->e_ident
[EI_VERSION
] = EV_CURRENT
;
1258 ehdr
->e_ident
[EI_OSABI
] = ELF_OSABI
;
1259 memset(ehdr
->e_ident
+ EI_PAD
, 0, EI_NIDENT
- EI_PAD
);
1260 ehdr
->e_type
= ET_CORE
;
1261 ehdr
->e_machine
= ELF_ARCH
;
1262 ehdr
->e_version
= EV_CURRENT
;
1263 ehdr
->e_phoff
= sizeof(Elf64_Ehdr
);
1264 ehdr
->e_ehsize
= sizeof(Elf64_Ehdr
);
1265 ehdr
->e_phentsize
= sizeof(Elf64_Phdr
);
1267 /* Prepare one phdr of type PT_NOTE for each present CPU */
1268 for_each_present_cpu(cpu
) {
1269 phdr
->p_type
= PT_NOTE
;
1270 notes_addr
= per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes
, cpu
));
1271 phdr
->p_offset
= phdr
->p_paddr
= notes_addr
;
1272 phdr
->p_filesz
= phdr
->p_memsz
= sizeof(note_buf_t
);
1277 /* Prepare one PT_NOTE header for vmcoreinfo */
1278 phdr
->p_type
= PT_NOTE
;
1279 phdr
->p_offset
= phdr
->p_paddr
= paddr_vmcoreinfo_note();
1280 phdr
->p_filesz
= phdr
->p_memsz
= VMCOREINFO_NOTE_SIZE
;
1284 /* Prepare PT_LOAD type program header for kernel text region */
1285 if (need_kernel_map
) {
1286 phdr
->p_type
= PT_LOAD
;
1287 phdr
->p_flags
= PF_R
|PF_W
|PF_X
;
1288 phdr
->p_vaddr
= (unsigned long) _text
;
1289 phdr
->p_filesz
= phdr
->p_memsz
= _end
- _text
;
1290 phdr
->p_offset
= phdr
->p_paddr
= __pa_symbol(_text
);
1295 /* Go through all the ranges in mem->ranges[] and prepare phdr */
1296 for (i
= 0; i
< mem
->nr_ranges
; i
++) {
1297 mstart
= mem
->ranges
[i
].start
;
1298 mend
= mem
->ranges
[i
].end
;
1300 phdr
->p_type
= PT_LOAD
;
1301 phdr
->p_flags
= PF_R
|PF_W
|PF_X
;
1302 phdr
->p_offset
= mstart
;
1304 phdr
->p_paddr
= mstart
;
1305 phdr
->p_vaddr
= (unsigned long) __va(mstart
);
1306 phdr
->p_filesz
= phdr
->p_memsz
= mend
- mstart
+ 1;
1309 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",
1310 phdr
, phdr
->p_vaddr
, phdr
->p_paddr
, phdr
->p_filesz
,
1311 ehdr
->e_phnum
, phdr
->p_offset
);