1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * kexec: kexec_file_load system call
4 *
5 * Copyright (C) 2014 Red Hat Inc.
6 * Authors:
7 * Vivek Goyal <vgoyal@redhat.com>
8 */
9
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12 #include <linux/capability.h>
13 #include <linux/mm.h>
14 #include <linux/file.h>
15 #include <linux/slab.h>
16 #include <linux/kexec.h>
17 #include <linux/memblock.h>
18 #include <linux/mutex.h>
19 #include <linux/list.h>
20 #include <linux/fs.h>
21 #include <linux/ima.h>
22 #include <crypto/hash.h>
23 #include <crypto/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"
31
32 #ifdef CONFIG_KEXEC_SIG
33 static bool sig_enforce = IS_ENABLED(CONFIG_KEXEC_SIG_FORCE);
34
set_kexec_sig_enforced(void)35 void set_kexec_sig_enforced(void)
36 {
37 sig_enforce = true;
38 }
39 #endif
40
41 static int kexec_calculate_store_digests(struct kimage *image);
42
43 /* Maximum size in bytes for kernel/initrd files. */
44 #define KEXEC_FILE_SIZE_MAX min_t(s64, 4LL << 30, SSIZE_MAX)
45
46 /*
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.
50 */
kexec_image_probe_default(struct kimage * image,void * buf,unsigned long buf_len)51 int kexec_image_probe_default(struct kimage *image, void *buf,
52 unsigned long buf_len)
53 {
54 const struct kexec_file_ops * const *fops;
55 int ret = -ENOEXEC;
56
57 for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
58 ret = (*fops)->probe(buf, buf_len);
59 if (!ret) {
60 image->fops = *fops;
61 return ret;
62 }
63 }
64
65 return ret;
66 }
67
kexec_image_load_default(struct kimage * image)68 void *kexec_image_load_default(struct kimage *image)
69 {
70 if (!image->fops || !image->fops->load)
71 return ERR_PTR(-ENOEXEC);
72
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);
77 }
78
kexec_image_post_load_cleanup_default(struct kimage * image)79 int kexec_image_post_load_cleanup_default(struct kimage *image)
80 {
81 if (!image->fops || !image->fops->cleanup)
82 return 0;
83
84 return image->fops->cleanup(image->image_loader_data);
85 }
86
87 /*
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.
91 */
kimage_file_post_load_cleanup(struct kimage * image)92 void kimage_file_post_load_cleanup(struct kimage *image)
93 {
94 struct purgatory_info *pi = &image->purgatory_info;
95
96 vfree(image->kernel_buf);
97 image->kernel_buf = NULL;
98
99 vfree(image->initrd_buf);
100 image->initrd_buf = NULL;
101
102 kfree(image->cmdline_buf);
103 image->cmdline_buf = NULL;
104
105 vfree(pi->purgatory_buf);
106 pi->purgatory_buf = NULL;
107
108 vfree(pi->sechdrs);
109 pi->sechdrs = NULL;
110
111 #ifdef CONFIG_IMA_KEXEC
112 vfree(image->ima_buffer);
113 image->ima_buffer = NULL;
114 #endif /* CONFIG_IMA_KEXEC */
115
116 /* See if architecture has anything to cleanup post load */
117 arch_kimage_file_post_load_cleanup(image);
118
119 /*
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.
123 */
124 kfree(image->image_loader_data);
125 image->image_loader_data = NULL;
126 }
127
128 #ifdef CONFIG_KEXEC_SIG
129 #ifdef CONFIG_SIGNED_PE_FILE_VERIFICATION
kexec_kernel_verify_pe_sig(const char * kernel,unsigned long kernel_len)130 int kexec_kernel_verify_pe_sig(const char *kernel, unsigned long kernel_len)
131 {
132 int ret;
133
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);
141 }
142 return ret;
143 }
144 #endif
145
kexec_image_verify_sig(struct kimage * image,void * buf,unsigned long buf_len)146 static int kexec_image_verify_sig(struct kimage *image, void *buf,
147 unsigned long buf_len)
148 {
149 if (!image->fops || !image->fops->verify_sig) {
150 pr_debug("kernel loader does not support signature verification.\n");
151 return -EKEYREJECTED;
152 }
153
154 return image->fops->verify_sig(buf, buf_len);
155 }
156
157 static int
kimage_validate_signature(struct kimage * image)158 kimage_validate_signature(struct kimage *image)
159 {
160 int ret;
161
162 ret = kexec_image_verify_sig(image, image->kernel_buf,
163 image->kernel_buf_len);
164 if (ret) {
165
166 if (sig_enforce) {
167 pr_notice("Enforced kernel signature verification failed (%d).\n", ret);
168 return ret;
169 }
170
171 /*
172 * If IMA is guaranteed to appraise a signature on the kexec
173 * image, permit it even if the kernel is otherwise locked
174 * down.
175 */
176 if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
177 security_locked_down(LOCKDOWN_KEXEC))
178 return -EPERM;
179
180 pr_debug("kernel signature verification failed (%d).\n", ret);
181 }
182
183 return 0;
184 }
185 #endif
186
187 /*
188 * In file mode list of segments is prepared by kernel. Copy relevant
189 * data from user space, do error checking, prepare segment list
190 */
191 static int
kimage_file_prepare_segments(struct kimage * image,int kernel_fd,int initrd_fd,const char __user * cmdline_ptr,unsigned long cmdline_len,unsigned flags)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)
195 {
196 ssize_t ret;
197 void *ldata;
198
199 ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf,
200 KEXEC_FILE_SIZE_MAX, NULL,
201 READING_KEXEC_IMAGE);
202 if (ret < 0)
203 return ret;
204 image->kernel_buf_len = ret;
205
206 /* Call arch image probe handlers */
207 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
208 image->kernel_buf_len);
209 if (ret)
210 goto out;
211
212 #ifdef CONFIG_KEXEC_SIG
213 ret = kimage_validate_signature(image);
214
215 if (ret)
216 goto out;
217 #endif
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);
223 if (ret < 0)
224 goto out;
225 image->initrd_buf_len = ret;
226 ret = 0;
227 }
228
229 if (cmdline_len) {
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;
234 goto out;
235 }
236
237 image->cmdline_buf_len = cmdline_len;
238
239 /* command line should be a string with last byte null */
240 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
241 ret = -EINVAL;
242 goto out;
243 }
244
245 ima_kexec_cmdline(kernel_fd, image->cmdline_buf,
246 image->cmdline_buf_len - 1);
247 }
248
249 /* IMA needs to pass the measurement list to the next kernel. */
250 ima_add_kexec_buffer(image);
251
252 /* Call arch image load handlers */
253 ldata = arch_kexec_kernel_image_load(image);
254
255 if (IS_ERR(ldata)) {
256 ret = PTR_ERR(ldata);
257 goto out;
258 }
259
260 image->image_loader_data = ldata;
261 out:
262 /* In case of error, free up all allocated memory in this function */
263 if (ret)
264 kimage_file_post_load_cleanup(image);
265 return ret;
266 }
267
268 static int
kimage_file_alloc_init(struct kimage ** rimage,int kernel_fd,int initrd_fd,const char __user * cmdline_ptr,unsigned long cmdline_len,unsigned long flags)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)
272 {
273 int ret;
274 struct kimage *image;
275 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
276
277 image = do_kimage_alloc_init();
278 if (!image)
279 return -ENOMEM;
280
281 image->file_mode = 1;
282
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;
287 }
288
289 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
290 cmdline_ptr, cmdline_len, flags);
291 if (ret)
292 goto out_free_image;
293
294 ret = sanity_check_segment_list(image);
295 if (ret)
296 goto out_free_post_load_bufs;
297
298 ret = -ENOMEM;
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;
304 }
305
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;
311 }
312 }
313
314 *rimage = image;
315 return 0;
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);
320 out_free_image:
321 kfree(image);
322 return ret;
323 }
324
SYSCALL_DEFINE5(kexec_file_load,int,kernel_fd,int,initrd_fd,unsigned long,cmdline_len,const char __user *,cmdline_ptr,unsigned long,flags)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)
328 {
329 int ret = 0, i;
330 struct kimage **dest_image, *image;
331
332 /* We only trust the superuser with rebooting the system. */
333 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
334 return -EPERM;
335
336 /* Make sure we have a legal set of flags */
337 if (flags != (flags & KEXEC_FILE_FLAGS))
338 return -EINVAL;
339
340 image = NULL;
341
342 if (!kexec_trylock())
343 return -EBUSY;
344
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();
350 }
351
352 if (flags & KEXEC_FILE_UNLOAD)
353 goto exchange;
354
355 /*
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.
359 */
360 if (flags & KEXEC_FILE_ON_CRASH)
361 kimage_free(xchg(&kexec_crash_image, NULL));
362
363 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
364 cmdline_len, flags);
365 if (ret)
366 goto out;
367
368 ret = machine_kexec_prepare(image);
369 if (ret)
370 goto out;
371
372 /*
373 * Some architecture(like S390) may touch the crash memory before
374 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
375 */
376 ret = kimage_crash_copy_vmcoreinfo(image);
377 if (ret)
378 goto out;
379
380 ret = kexec_calculate_store_digests(image);
381 if (ret)
382 goto out;
383
384 for (i = 0; i < image->nr_segments; i++) {
385 struct kexec_segment *ksegment;
386
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,
390 ksegment->memsz);
391
392 ret = kimage_load_segment(image, &image->segment[i]);
393 if (ret)
394 goto out;
395 }
396
397 kimage_terminate(image);
398
399 ret = machine_kexec_post_load(image);
400 if (ret)
401 goto out;
402
403 /*
404 * Free up any temporary buffers allocated which are not needed
405 * after image has been loaded
406 */
407 kimage_file_post_load_cleanup(image);
408 exchange:
409 image = xchg(dest_image, image);
410 out:
411 if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
412 arch_kexec_protect_crashkres();
413
414 kexec_unlock();
415 kimage_free(image);
416 return ret;
417 }
418
locate_mem_hole_top_down(unsigned long start,unsigned long end,struct kexec_buf * kbuf)419 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
420 struct kexec_buf *kbuf)
421 {
422 struct kimage *image = kbuf->image;
423 unsigned long temp_start, temp_end;
424
425 temp_end = min(end, kbuf->buf_max);
426 temp_start = temp_end - kbuf->memsz;
427
428 do {
429 /* align down start */
430 temp_start = temp_start & (~(kbuf->buf_align - 1));
431
432 if (temp_start < start || temp_start < kbuf->buf_min)
433 return 0;
434
435 temp_end = temp_start + kbuf->memsz - 1;
436
437 /*
438 * Make sure this does not conflict with any of existing
439 * segments
440 */
441 if (kimage_is_destination_range(image, temp_start, temp_end)) {
442 temp_start = temp_start - PAGE_SIZE;
443 continue;
444 }
445
446 /* We found a suitable memory range */
447 break;
448 } while (1);
449
450 /* If we are here, we found a suitable memory range */
451 kbuf->mem = temp_start;
452
453 /* Success, stop navigating through remaining System RAM ranges */
454 return 1;
455 }
456
locate_mem_hole_bottom_up(unsigned long start,unsigned long end,struct kexec_buf * kbuf)457 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
458 struct kexec_buf *kbuf)
459 {
460 struct kimage *image = kbuf->image;
461 unsigned long temp_start, temp_end;
462
463 temp_start = max(start, kbuf->buf_min);
464
465 do {
466 temp_start = ALIGN(temp_start, kbuf->buf_align);
467 temp_end = temp_start + kbuf->memsz - 1;
468
469 if (temp_end > end || temp_end > kbuf->buf_max)
470 return 0;
471 /*
472 * Make sure this does not conflict with any of existing
473 * segments
474 */
475 if (kimage_is_destination_range(image, temp_start, temp_end)) {
476 temp_start = temp_start + PAGE_SIZE;
477 continue;
478 }
479
480 /* We found a suitable memory range */
481 break;
482 } while (1);
483
484 /* If we are here, we found a suitable memory range */
485 kbuf->mem = temp_start;
486
487 /* Success, stop navigating through remaining System RAM ranges */
488 return 1;
489 }
490
locate_mem_hole_callback(struct resource * res,void * arg)491 static int locate_mem_hole_callback(struct resource *res, void *arg)
492 {
493 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
494 u64 start = res->start, end = res->end;
495 unsigned long sz = end - start + 1;
496
497 /* Returning 0 will take to next memory range */
498
499 /* Don't use memory that will be detected and handled by a driver. */
500 if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
501 return 0;
502
503 if (sz < kbuf->memsz)
504 return 0;
505
506 if (end < kbuf->buf_min || start > kbuf->buf_max)
507 return 0;
508
509 /*
510 * Allocate memory top down with-in ram range. Otherwise bottom up
511 * allocation.
512 */
513 if (kbuf->top_down)
514 return locate_mem_hole_top_down(start, end, kbuf);
515 return locate_mem_hole_bottom_up(start, end, kbuf);
516 }
517
518 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
kexec_walk_memblock(struct kexec_buf * kbuf,int (* func)(struct resource *,void *))519 static int kexec_walk_memblock(struct kexec_buf *kbuf,
520 int (*func)(struct resource *, void *))
521 {
522 int ret = 0;
523 u64 i;
524 phys_addr_t mstart, mend;
525 struct resource res = { };
526
527 if (kbuf->image->type == KEXEC_TYPE_CRASH)
528 return func(&crashk_res, kbuf);
529
530 /*
531 * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
532 * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
533 * locate_mem_hole_callback().
534 */
535 if (kbuf->top_down) {
536 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
537 &mstart, &mend, NULL) {
538 /*
539 * In memblock, end points to the first byte after the
540 * range while in kexec, end points to the last byte
541 * in the range.
542 */
543 res.start = mstart;
544 res.end = mend - 1;
545 ret = func(&res, kbuf);
546 if (ret)
547 break;
548 }
549 } else {
550 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
551 &mstart, &mend, NULL) {
552 /*
553 * In memblock, end points to the first byte after the
554 * range while in kexec, end points to the last byte
555 * in the range.
556 */
557 res.start = mstart;
558 res.end = mend - 1;
559 ret = func(&res, kbuf);
560 if (ret)
561 break;
562 }
563 }
564
565 return ret;
566 }
567 #else
kexec_walk_memblock(struct kexec_buf * kbuf,int (* func)(struct resource *,void *))568 static int kexec_walk_memblock(struct kexec_buf *kbuf,
569 int (*func)(struct resource *, void *))
570 {
571 return 0;
572 }
573 #endif
574
575 /**
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.
579 *
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.
583 */
kexec_walk_resources(struct kexec_buf * kbuf,int (* func)(struct resource *,void *))584 static int kexec_walk_resources(struct kexec_buf *kbuf,
585 int (*func)(struct resource *, void *))
586 {
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,
591 kbuf, func);
592 else
593 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
594 }
595
596 /**
597 * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
598 * @kbuf: Parameters for the memory search.
599 *
600 * On success, kbuf->mem will have the start address of the memory region found.
601 *
602 * Return: 0 on success, negative errno on error.
603 */
kexec_locate_mem_hole(struct kexec_buf * kbuf)604 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
605 {
606 int ret;
607
608 /* Arch knows where to place */
609 if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
610 return 0;
611
612 if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
613 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
614 else
615 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
616
617 return ret == 1 ? 0 : -EADDRNOTAVAIL;
618 }
619
620 /**
621 * kexec_add_buffer - place a buffer in a kexec segment
622 * @kbuf: Buffer contents and memory parameters.
623 *
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.
627 *
628 * Return: 0 on success, negative errno on error.
629 */
kexec_add_buffer(struct kexec_buf * kbuf)630 int kexec_add_buffer(struct kexec_buf *kbuf)
631 {
632 struct kexec_segment *ksegment;
633 int ret;
634
635 /* Currently adding segment this way is allowed only in file mode */
636 if (!kbuf->image->file_mode)
637 return -EINVAL;
638
639 if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
640 return -EINVAL;
641
642 /*
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.
648 */
649 if (!list_empty(&kbuf->image->control_pages)) {
650 WARN_ON(1);
651 return -EINVAL;
652 }
653
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);
657
658 /* Walk the RAM ranges and allocate a suitable range for the buffer */
659 ret = arch_kexec_locate_mem_hole(kbuf);
660 if (ret)
661 return ret;
662
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++;
670 return 0;
671 }
672
673 /* Calculate and store the digest of segments */
kexec_calculate_store_digests(struct kimage * image)674 static int kexec_calculate_store_digests(struct kimage *image)
675 {
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;
680 char *digest;
681 void *zero_buf;
682 struct kexec_sha_region *sha_regions;
683 struct purgatory_info *pi = &image->purgatory_info;
684
685 if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
686 return 0;
687
688 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
689 zero_buf_sz = PAGE_SIZE;
690
691 tfm = crypto_alloc_shash("sha256", 0, 0);
692 if (IS_ERR(tfm)) {
693 ret = PTR_ERR(tfm);
694 goto out;
695 }
696
697 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
698 desc = kzalloc(desc_size, GFP_KERNEL);
699 if (!desc) {
700 ret = -ENOMEM;
701 goto out_free_tfm;
702 }
703
704 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
705 sha_regions = vzalloc(sha_region_sz);
706 if (!sha_regions) {
707 ret = -ENOMEM;
708 goto out_free_desc;
709 }
710
711 desc->tfm = tfm;
712
713 ret = crypto_shash_init(desc);
714 if (ret < 0)
715 goto out_free_sha_regions;
716
717 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
718 if (!digest) {
719 ret = -ENOMEM;
720 goto out_free_sha_regions;
721 }
722
723 for (j = i = 0; i < image->nr_segments; i++) {
724 struct kexec_segment *ksegment;
725
726 ksegment = &image->segment[i];
727 /*
728 * Skip purgatory as it will be modified once we put digest
729 * info in purgatory.
730 */
731 if (ksegment->kbuf == pi->purgatory_buf)
732 continue;
733
734 ret = crypto_shash_update(desc, ksegment->kbuf,
735 ksegment->bufsz);
736 if (ret)
737 break;
738
739 /*
740 * Assume rest of the buffer is filled with zero and
741 * update digest accordingly.
742 */
743 nullsz = ksegment->memsz - ksegment->bufsz;
744 while (nullsz) {
745 unsigned long bytes = nullsz;
746
747 if (bytes > zero_buf_sz)
748 bytes = zero_buf_sz;
749 ret = crypto_shash_update(desc, zero_buf, bytes);
750 if (ret)
751 break;
752 nullsz -= bytes;
753 }
754
755 if (ret)
756 break;
757
758 sha_regions[j].start = ksegment->mem;
759 sha_regions[j].len = ksegment->memsz;
760 j++;
761 }
762
763 if (!ret) {
764 ret = crypto_shash_final(desc, digest);
765 if (ret)
766 goto out_free_digest;
767 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
768 sha_regions, sha_region_sz, 0);
769 if (ret)
770 goto out_free_digest;
771
772 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
773 digest, SHA256_DIGEST_SIZE, 0);
774 if (ret)
775 goto out_free_digest;
776 }
777
778 out_free_digest:
779 kfree(digest);
780 out_free_sha_regions:
781 vfree(sha_regions);
782 out_free_desc:
783 kfree(desc);
784 out_free_tfm:
785 kfree(tfm);
786 out:
787 return ret;
788 }
789
790 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
791 /*
792 * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
793 * @pi: Purgatory to be loaded.
794 * @kbuf: Buffer to setup.
795 *
796 * Allocates the memory needed for the buffer. Caller is responsible to free
797 * the memory after use.
798 *
799 * Return: 0 on success, negative errno on error.
800 */
kexec_purgatory_setup_kbuf(struct purgatory_info * pi,struct kexec_buf * kbuf)801 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
802 struct kexec_buf *kbuf)
803 {
804 const Elf_Shdr *sechdrs;
805 unsigned long bss_align;
806 unsigned long bss_sz;
807 unsigned long align;
808 int i, ret;
809
810 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
811 kbuf->buf_align = bss_align = 1;
812 kbuf->bufsz = bss_sz = 0;
813
814 for (i = 0; i < pi->ehdr->e_shnum; i++) {
815 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
816 continue;
817
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;
824 } else {
825 if (bss_align < align)
826 bss_align = align;
827 bss_sz = ALIGN(bss_sz, align);
828 bss_sz += sechdrs[i].sh_size;
829 }
830 }
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;
835
836 kbuf->buffer = vzalloc(kbuf->bufsz);
837 if (!kbuf->buffer)
838 return -ENOMEM;
839 pi->purgatory_buf = kbuf->buffer;
840
841 ret = kexec_add_buffer(kbuf);
842 if (ret)
843 goto out;
844
845 return 0;
846 out:
847 vfree(pi->purgatory_buf);
848 pi->purgatory_buf = NULL;
849 return ret;
850 }
851
852 /*
853 * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
854 * @pi: Purgatory to be loaded.
855 * @kbuf: Buffer prepared to store purgatory.
856 *
857 * Allocates the memory needed for the buffer. Caller is responsible to free
858 * the memory after use.
859 *
860 * Return: 0 on success, negative errno on error.
861 */
kexec_purgatory_setup_sechdrs(struct purgatory_info * pi,struct kexec_buf * kbuf)862 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
863 struct kexec_buf *kbuf)
864 {
865 unsigned long bss_addr;
866 unsigned long offset;
867 Elf_Shdr *sechdrs;
868 int i;
869
870 /*
871 * The section headers in kexec_purgatory are read-only. In order to
872 * have them modifiable make a temporary copy.
873 */
874 sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
875 if (!sechdrs)
876 return -ENOMEM;
877 memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
878 pi->ehdr->e_shnum * sizeof(Elf_Shdr));
879 pi->sechdrs = sechdrs;
880
881 offset = 0;
882 bss_addr = kbuf->mem + kbuf->bufsz;
883 kbuf->image->start = pi->ehdr->e_entry;
884
885 for (i = 0; i < pi->ehdr->e_shnum; i++) {
886 unsigned long align;
887 void *src, *dst;
888
889 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
890 continue;
891
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;
897 continue;
898 }
899
900 offset = ALIGN(offset, align);
901
902 /*
903 * Check if the segment contains the entry point, if so,
904 * calculate the value of image->start based on it.
905 * If the compiler has produced more than one .text section
906 * (Eg: .text.hot), they are generally after the main .text
907 * section, and they shall not be used to calculate
908 * image->start. So do not re-calculate image->start if it
909 * is not set to the initial value, and warn the user so they
910 * have a chance to fix their purgatory's linker script.
911 */
912 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
913 pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
914 pi->ehdr->e_entry < (sechdrs[i].sh_addr
915 + sechdrs[i].sh_size) &&
916 !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) {
917 kbuf->image->start -= sechdrs[i].sh_addr;
918 kbuf->image->start += kbuf->mem + offset;
919 }
920
921 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
922 dst = pi->purgatory_buf + offset;
923 memcpy(dst, src, sechdrs[i].sh_size);
924
925 sechdrs[i].sh_addr = kbuf->mem + offset;
926 sechdrs[i].sh_offset = offset;
927 offset += sechdrs[i].sh_size;
928 }
929
930 return 0;
931 }
932
kexec_apply_relocations(struct kimage * image)933 static int kexec_apply_relocations(struct kimage *image)
934 {
935 int i, ret;
936 struct purgatory_info *pi = &image->purgatory_info;
937 const Elf_Shdr *sechdrs;
938
939 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
940
941 for (i = 0; i < pi->ehdr->e_shnum; i++) {
942 const Elf_Shdr *relsec;
943 const Elf_Shdr *symtab;
944 Elf_Shdr *section;
945
946 relsec = sechdrs + i;
947
948 if (relsec->sh_type != SHT_RELA &&
949 relsec->sh_type != SHT_REL)
950 continue;
951
952 /*
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.
957 */
958 if (relsec->sh_info >= pi->ehdr->e_shnum ||
959 relsec->sh_link >= pi->ehdr->e_shnum)
960 return -ENOEXEC;
961
962 section = pi->sechdrs + relsec->sh_info;
963 symtab = sechdrs + relsec->sh_link;
964
965 if (!(section->sh_flags & SHF_ALLOC))
966 continue;
967
968 /*
969 * symtab->sh_link contain section header index of associated
970 * string table.
971 */
972 if (symtab->sh_link >= pi->ehdr->e_shnum)
973 /* Invalid section number? */
974 continue;
975
976 /*
977 * Respective architecture needs to provide support for applying
978 * relocations of type SHT_RELA/SHT_REL.
979 */
980 if (relsec->sh_type == SHT_RELA)
981 ret = arch_kexec_apply_relocations_add(pi, section,
982 relsec, symtab);
983 else if (relsec->sh_type == SHT_REL)
984 ret = arch_kexec_apply_relocations(pi, section,
985 relsec, symtab);
986 if (ret)
987 return ret;
988 }
989
990 return 0;
991 }
992
993 /*
994 * kexec_load_purgatory - Load and relocate the purgatory object.
995 * @image: Image to add the purgatory to.
996 * @kbuf: Memory parameters to use.
997 *
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.
1001 *
1002 * Return: 0 on success, negative errno on error.
1003 */
kexec_load_purgatory(struct kimage * image,struct kexec_buf * kbuf)1004 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1005 {
1006 struct purgatory_info *pi = &image->purgatory_info;
1007 int ret;
1008
1009 if (kexec_purgatory_size <= 0)
1010 return -EINVAL;
1011
1012 pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1013
1014 ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1015 if (ret)
1016 return ret;
1017
1018 ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1019 if (ret)
1020 goto out_free_kbuf;
1021
1022 ret = kexec_apply_relocations(image);
1023 if (ret)
1024 goto out;
1025
1026 return 0;
1027 out:
1028 vfree(pi->sechdrs);
1029 pi->sechdrs = NULL;
1030 out_free_kbuf:
1031 vfree(pi->purgatory_buf);
1032 pi->purgatory_buf = NULL;
1033 return ret;
1034 }
1035
1036 /*
1037 * kexec_purgatory_find_symbol - find a symbol in the purgatory
1038 * @pi: Purgatory to search in.
1039 * @name: Name of the symbol.
1040 *
1041 * Return: pointer to symbol in read-only symtab on success, NULL on error.
1042 */
kexec_purgatory_find_symbol(struct purgatory_info * pi,const char * name)1043 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1044 const char *name)
1045 {
1046 const Elf_Shdr *sechdrs;
1047 const Elf_Ehdr *ehdr;
1048 const Elf_Sym *syms;
1049 const char *strtab;
1050 int i, k;
1051
1052 if (!pi->ehdr)
1053 return NULL;
1054
1055 ehdr = pi->ehdr;
1056 sechdrs = (void *)ehdr + ehdr->e_shoff;
1057
1058 for (i = 0; i < ehdr->e_shnum; i++) {
1059 if (sechdrs[i].sh_type != SHT_SYMTAB)
1060 continue;
1061
1062 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1063 /* Invalid strtab section number */
1064 continue;
1065 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1066 syms = (void *)ehdr + sechdrs[i].sh_offset;
1067
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)
1071 continue;
1072
1073 if (strcmp(strtab + syms[k].st_name, name) != 0)
1074 continue;
1075
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);
1080 return NULL;
1081 }
1082
1083 /* Found the symbol we are looking for */
1084 return &syms[k];
1085 }
1086 }
1087
1088 return NULL;
1089 }
1090
kexec_purgatory_get_symbol_addr(struct kimage * image,const char * name)1091 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1092 {
1093 struct purgatory_info *pi = &image->purgatory_info;
1094 const Elf_Sym *sym;
1095 Elf_Shdr *sechdr;
1096
1097 sym = kexec_purgatory_find_symbol(pi, name);
1098 if (!sym)
1099 return ERR_PTR(-EINVAL);
1100
1101 sechdr = &pi->sechdrs[sym->st_shndx];
1102
1103 /*
1104 * Returns the address where symbol will finally be loaded after
1105 * kexec_load_segment()
1106 */
1107 return (void *)(sechdr->sh_addr + sym->st_value);
1108 }
1109
1110 /*
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.
1113 */
kexec_purgatory_get_set_symbol(struct kimage * image,const char * name,void * buf,unsigned int size,bool get_value)1114 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1115 void *buf, unsigned int size, bool get_value)
1116 {
1117 struct purgatory_info *pi = &image->purgatory_info;
1118 const Elf_Sym *sym;
1119 Elf_Shdr *sec;
1120 char *sym_buf;
1121
1122 sym = kexec_purgatory_find_symbol(pi, name);
1123 if (!sym)
1124 return -EINVAL;
1125
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);
1129 return -EINVAL;
1130 }
1131
1132 sec = pi->sechdrs + sym->st_shndx;
1133
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");
1137 return -EINVAL;
1138 }
1139
1140 sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1141
1142 if (get_value)
1143 memcpy((void *)buf, sym_buf, size);
1144 else
1145 memcpy((void *)sym_buf, buf, size);
1146
1147 return 0;
1148 }
1149 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1150
crash_exclude_mem_range(struct crash_mem * mem,unsigned long long mstart,unsigned long long mend)1151 int crash_exclude_mem_range(struct crash_mem *mem,
1152 unsigned long long mstart, unsigned long long mend)
1153 {
1154 int i, j;
1155 unsigned long long start, end, p_start, p_end;
1156 struct crash_mem_range temp_range = {0, 0};
1157
1158 for (i = 0; i < mem->nr_ranges; i++) {
1159 start = mem->ranges[i].start;
1160 end = mem->ranges[i].end;
1161 p_start = mstart;
1162 p_end = mend;
1163
1164 if (mstart > end || mend < start)
1165 continue;
1166
1167 /* Truncate any area outside of range */
1168 if (mstart < start)
1169 p_start = start;
1170 if (mend > end)
1171 p_end = end;
1172
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;
1184 }
1185
1186 /*
1187 * Continue to check if there are another overlapping ranges
1188 * from the current position because of shifting the above
1189 * mem ranges.
1190 */
1191 i--;
1192 mem->nr_ranges--;
1193 continue;
1194 }
1195 mem->nr_ranges--;
1196 return 0;
1197 }
1198
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;
1206 else
1207 mem->ranges[i].start = p_end + 1;
1208 break;
1209 }
1210
1211 /* If a split happened, add the split to array */
1212 if (!temp_range.end)
1213 return 0;
1214
1215 /* Split happened */
1216 if (i == mem->max_nr_ranges - 1)
1217 return -ENOMEM;
1218
1219 /* Location where new range should go */
1220 j = i + 1;
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];
1225 }
1226
1227 mem->ranges[j].start = temp_range.start;
1228 mem->ranges[j].end = temp_range.end;
1229 mem->nr_ranges++;
1230 return 0;
1231 }
1232
crash_prepare_elf64_headers(struct crash_mem * mem,int need_kernel_map,void ** addr,unsigned long * sz)1233 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
1234 void **addr, unsigned long *sz)
1235 {
1236 Elf64_Ehdr *ehdr;
1237 Elf64_Phdr *phdr;
1238 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1239 unsigned char *buf;
1240 unsigned int cpu, i;
1241 unsigned long long notes_addr;
1242 unsigned long mstart, mend;
1243
1244 /* extra phdr for vmcoreinfo ELF note */
1245 nr_phdr = nr_cpus + 1;
1246 nr_phdr += mem->nr_ranges;
1247
1248 /*
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.
1254 */
1255
1256 nr_phdr++;
1257 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1258 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1259
1260 buf = vzalloc(elf_sz);
1261 if (!buf)
1262 return -ENOMEM;
1263
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);
1278
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);
1285 (ehdr->e_phnum)++;
1286 phdr++;
1287 }
1288
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;
1293 (ehdr->e_phnum)++;
1294 phdr++;
1295
1296 /* Prepare PT_LOAD type program header for kernel text region */
1297 if (need_kernel_map) {
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);
1303 ehdr->e_phnum++;
1304 phdr++;
1305 }
1306
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;
1311
1312 phdr->p_type = PT_LOAD;
1313 phdr->p_flags = PF_R|PF_W|PF_X;
1314 phdr->p_offset = mstart;
1315
1316 phdr->p_paddr = mstart;
1317 phdr->p_vaddr = (unsigned long) __va(mstart);
1318 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1319 phdr->p_align = 0;
1320 ehdr->e_phnum++;
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);
1324 phdr++;
1325 }
1326
1327 *addr = buf;
1328 *sz = elf_sz;
1329 return 0;
1330 }
1331