Merge tag 'kvm-x86-selftests-6.5' of https://github.com/kvm-x86/linux into HEAD
[linux-block.git] / kernel / kexec_file.c
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
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  */
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
68 static 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
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  */
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
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
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
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
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 image load handler */
253         ldata = kexec_image_load_default(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
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
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 image_type = (flags & KEXEC_FILE_ON_CRASH) ?
330                          KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT;
331         struct kimage **dest_image, *image;
332         int ret = 0, i;
333
334         /* We only trust the superuser with rebooting the system. */
335         if (!kexec_load_permitted(image_type))
336                 return -EPERM;
337
338         /* Make sure we have a legal set of flags */
339         if (flags != (flags & KEXEC_FILE_FLAGS))
340                 return -EINVAL;
341
342         image = NULL;
343
344         if (!kexec_trylock())
345                 return -EBUSY;
346
347         if (image_type == KEXEC_TYPE_CRASH) {
348                 dest_image = &kexec_crash_image;
349                 if (kexec_crash_image)
350                         arch_kexec_unprotect_crashkres();
351         } else {
352                 dest_image = &kexec_image;
353         }
354
355         if (flags & KEXEC_FILE_UNLOAD)
356                 goto exchange;
357
358         /*
359          * In case of crash, new kernel gets loaded in reserved region. It is
360          * same memory where old crash kernel might be loaded. Free any
361          * current crash dump kernel before we corrupt it.
362          */
363         if (flags & KEXEC_FILE_ON_CRASH)
364                 kimage_free(xchg(&kexec_crash_image, NULL));
365
366         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
367                                      cmdline_len, flags);
368         if (ret)
369                 goto out;
370
371         ret = machine_kexec_prepare(image);
372         if (ret)
373                 goto out;
374
375         /*
376          * Some architecture(like S390) may touch the crash memory before
377          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
378          */
379         ret = kimage_crash_copy_vmcoreinfo(image);
380         if (ret)
381                 goto out;
382
383         ret = kexec_calculate_store_digests(image);
384         if (ret)
385                 goto out;
386
387         for (i = 0; i < image->nr_segments; i++) {
388                 struct kexec_segment *ksegment;
389
390                 ksegment = &image->segment[i];
391                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
392                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
393                          ksegment->memsz);
394
395                 ret = kimage_load_segment(image, &image->segment[i]);
396                 if (ret)
397                         goto out;
398         }
399
400         kimage_terminate(image);
401
402         ret = machine_kexec_post_load(image);
403         if (ret)
404                 goto out;
405
406         /*
407          * Free up any temporary buffers allocated which are not needed
408          * after image has been loaded
409          */
410         kimage_file_post_load_cleanup(image);
411 exchange:
412         image = xchg(dest_image, image);
413 out:
414         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
415                 arch_kexec_protect_crashkres();
416
417         kexec_unlock();
418         kimage_free(image);
419         return ret;
420 }
421
422 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
423                                     struct kexec_buf *kbuf)
424 {
425         struct kimage *image = kbuf->image;
426         unsigned long temp_start, temp_end;
427
428         temp_end = min(end, kbuf->buf_max);
429         temp_start = temp_end - kbuf->memsz;
430
431         do {
432                 /* align down start */
433                 temp_start = temp_start & (~(kbuf->buf_align - 1));
434
435                 if (temp_start < start || temp_start < kbuf->buf_min)
436                         return 0;
437
438                 temp_end = temp_start + kbuf->memsz - 1;
439
440                 /*
441                  * Make sure this does not conflict with any of existing
442                  * segments
443                  */
444                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
445                         temp_start = temp_start - PAGE_SIZE;
446                         continue;
447                 }
448
449                 /* We found a suitable memory range */
450                 break;
451         } while (1);
452
453         /* If we are here, we found a suitable memory range */
454         kbuf->mem = temp_start;
455
456         /* Success, stop navigating through remaining System RAM ranges */
457         return 1;
458 }
459
460 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
461                                      struct kexec_buf *kbuf)
462 {
463         struct kimage *image = kbuf->image;
464         unsigned long temp_start, temp_end;
465
466         temp_start = max(start, kbuf->buf_min);
467
468         do {
469                 temp_start = ALIGN(temp_start, kbuf->buf_align);
470                 temp_end = temp_start + kbuf->memsz - 1;
471
472                 if (temp_end > end || temp_end > kbuf->buf_max)
473                         return 0;
474                 /*
475                  * Make sure this does not conflict with any of existing
476                  * segments
477                  */
478                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
479                         temp_start = temp_start + PAGE_SIZE;
480                         continue;
481                 }
482
483                 /* We found a suitable memory range */
484                 break;
485         } while (1);
486
487         /* If we are here, we found a suitable memory range */
488         kbuf->mem = temp_start;
489
490         /* Success, stop navigating through remaining System RAM ranges */
491         return 1;
492 }
493
494 static int locate_mem_hole_callback(struct resource *res, void *arg)
495 {
496         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
497         u64 start = res->start, end = res->end;
498         unsigned long sz = end - start + 1;
499
500         /* Returning 0 will take to next memory range */
501
502         /* Don't use memory that will be detected and handled by a driver. */
503         if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
504                 return 0;
505
506         if (sz < kbuf->memsz)
507                 return 0;
508
509         if (end < kbuf->buf_min || start > kbuf->buf_max)
510                 return 0;
511
512         /*
513          * Allocate memory top down with-in ram range. Otherwise bottom up
514          * allocation.
515          */
516         if (kbuf->top_down)
517                 return locate_mem_hole_top_down(start, end, kbuf);
518         return locate_mem_hole_bottom_up(start, end, kbuf);
519 }
520
521 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
522 static int kexec_walk_memblock(struct kexec_buf *kbuf,
523                                int (*func)(struct resource *, void *))
524 {
525         int ret = 0;
526         u64 i;
527         phys_addr_t mstart, mend;
528         struct resource res = { };
529
530         if (kbuf->image->type == KEXEC_TYPE_CRASH)
531                 return func(&crashk_res, kbuf);
532
533         /*
534          * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
535          * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
536          * locate_mem_hole_callback().
537          */
538         if (kbuf->top_down) {
539                 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
540                                                 &mstart, &mend, NULL) {
541                         /*
542                          * In memblock, end points to the first byte after the
543                          * range while in kexec, end points to the last byte
544                          * in the range.
545                          */
546                         res.start = mstart;
547                         res.end = mend - 1;
548                         ret = func(&res, kbuf);
549                         if (ret)
550                                 break;
551                 }
552         } else {
553                 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
554                                         &mstart, &mend, NULL) {
555                         /*
556                          * In memblock, end points to the first byte after the
557                          * range while in kexec, end points to the last byte
558                          * in the range.
559                          */
560                         res.start = mstart;
561                         res.end = mend - 1;
562                         ret = func(&res, kbuf);
563                         if (ret)
564                                 break;
565                 }
566         }
567
568         return ret;
569 }
570 #else
571 static int kexec_walk_memblock(struct kexec_buf *kbuf,
572                                int (*func)(struct resource *, void *))
573 {
574         return 0;
575 }
576 #endif
577
578 /**
579  * kexec_walk_resources - call func(data) on free memory regions
580  * @kbuf:       Context info for the search. Also passed to @func.
581  * @func:       Function to call for each memory region.
582  *
583  * Return: The memory walk will stop when func returns a non-zero value
584  * and that value will be returned. If all free regions are visited without
585  * func returning non-zero, then zero will be returned.
586  */
587 static int kexec_walk_resources(struct kexec_buf *kbuf,
588                                 int (*func)(struct resource *, void *))
589 {
590         if (kbuf->image->type == KEXEC_TYPE_CRASH)
591                 return walk_iomem_res_desc(crashk_res.desc,
592                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
593                                            crashk_res.start, crashk_res.end,
594                                            kbuf, func);
595         else
596                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
597 }
598
599 /**
600  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
601  * @kbuf:       Parameters for the memory search.
602  *
603  * On success, kbuf->mem will have the start address of the memory region found.
604  *
605  * Return: 0 on success, negative errno on error.
606  */
607 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
608 {
609         int ret;
610
611         /* Arch knows where to place */
612         if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
613                 return 0;
614
615         if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
616                 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
617         else
618                 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
619
620         return ret == 1 ? 0 : -EADDRNOTAVAIL;
621 }
622
623 /**
624  * kexec_add_buffer - place a buffer in a kexec segment
625  * @kbuf:       Buffer contents and memory parameters.
626  *
627  * This function assumes that kexec_mutex is held.
628  * On successful return, @kbuf->mem will have the physical address of
629  * the buffer in memory.
630  *
631  * Return: 0 on success, negative errno on error.
632  */
633 int kexec_add_buffer(struct kexec_buf *kbuf)
634 {
635         struct kexec_segment *ksegment;
636         int ret;
637
638         /* Currently adding segment this way is allowed only in file mode */
639         if (!kbuf->image->file_mode)
640                 return -EINVAL;
641
642         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
643                 return -EINVAL;
644
645         /*
646          * Make sure we are not trying to add buffer after allocating
647          * control pages. All segments need to be placed first before
648          * any control pages are allocated. As control page allocation
649          * logic goes through list of segments to make sure there are
650          * no destination overlaps.
651          */
652         if (!list_empty(&kbuf->image->control_pages)) {
653                 WARN_ON(1);
654                 return -EINVAL;
655         }
656
657         /* Ensure minimum alignment needed for segments. */
658         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
659         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
660
661         /* Walk the RAM ranges and allocate a suitable range for the buffer */
662         ret = arch_kexec_locate_mem_hole(kbuf);
663         if (ret)
664                 return ret;
665
666         /* Found a suitable memory range */
667         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
668         ksegment->kbuf = kbuf->buffer;
669         ksegment->bufsz = kbuf->bufsz;
670         ksegment->mem = kbuf->mem;
671         ksegment->memsz = kbuf->memsz;
672         kbuf->image->nr_segments++;
673         return 0;
674 }
675
676 /* Calculate and store the digest of segments */
677 static int kexec_calculate_store_digests(struct kimage *image)
678 {
679         struct crypto_shash *tfm;
680         struct shash_desc *desc;
681         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
682         size_t desc_size, nullsz;
683         char *digest;
684         void *zero_buf;
685         struct kexec_sha_region *sha_regions;
686         struct purgatory_info *pi = &image->purgatory_info;
687
688         if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
689                 return 0;
690
691         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
692         zero_buf_sz = PAGE_SIZE;
693
694         tfm = crypto_alloc_shash("sha256", 0, 0);
695         if (IS_ERR(tfm)) {
696                 ret = PTR_ERR(tfm);
697                 goto out;
698         }
699
700         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
701         desc = kzalloc(desc_size, GFP_KERNEL);
702         if (!desc) {
703                 ret = -ENOMEM;
704                 goto out_free_tfm;
705         }
706
707         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
708         sha_regions = vzalloc(sha_region_sz);
709         if (!sha_regions) {
710                 ret = -ENOMEM;
711                 goto out_free_desc;
712         }
713
714         desc->tfm   = tfm;
715
716         ret = crypto_shash_init(desc);
717         if (ret < 0)
718                 goto out_free_sha_regions;
719
720         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
721         if (!digest) {
722                 ret = -ENOMEM;
723                 goto out_free_sha_regions;
724         }
725
726         for (j = i = 0; i < image->nr_segments; i++) {
727                 struct kexec_segment *ksegment;
728
729                 ksegment = &image->segment[i];
730                 /*
731                  * Skip purgatory as it will be modified once we put digest
732                  * info in purgatory.
733                  */
734                 if (ksegment->kbuf == pi->purgatory_buf)
735                         continue;
736
737                 ret = crypto_shash_update(desc, ksegment->kbuf,
738                                           ksegment->bufsz);
739                 if (ret)
740                         break;
741
742                 /*
743                  * Assume rest of the buffer is filled with zero and
744                  * update digest accordingly.
745                  */
746                 nullsz = ksegment->memsz - ksegment->bufsz;
747                 while (nullsz) {
748                         unsigned long bytes = nullsz;
749
750                         if (bytes > zero_buf_sz)
751                                 bytes = zero_buf_sz;
752                         ret = crypto_shash_update(desc, zero_buf, bytes);
753                         if (ret)
754                                 break;
755                         nullsz -= bytes;
756                 }
757
758                 if (ret)
759                         break;
760
761                 sha_regions[j].start = ksegment->mem;
762                 sha_regions[j].len = ksegment->memsz;
763                 j++;
764         }
765
766         if (!ret) {
767                 ret = crypto_shash_final(desc, digest);
768                 if (ret)
769                         goto out_free_digest;
770                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
771                                                      sha_regions, sha_region_sz, 0);
772                 if (ret)
773                         goto out_free_digest;
774
775                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
776                                                      digest, SHA256_DIGEST_SIZE, 0);
777                 if (ret)
778                         goto out_free_digest;
779         }
780
781 out_free_digest:
782         kfree(digest);
783 out_free_sha_regions:
784         vfree(sha_regions);
785 out_free_desc:
786         kfree(desc);
787 out_free_tfm:
788         kfree(tfm);
789 out:
790         return ret;
791 }
792
793 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
794 /*
795  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
796  * @pi:         Purgatory to be loaded.
797  * @kbuf:       Buffer to setup.
798  *
799  * Allocates the memory needed for the buffer. Caller is responsible to free
800  * the memory after use.
801  *
802  * Return: 0 on success, negative errno on error.
803  */
804 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
805                                       struct kexec_buf *kbuf)
806 {
807         const Elf_Shdr *sechdrs;
808         unsigned long bss_align;
809         unsigned long bss_sz;
810         unsigned long align;
811         int i, ret;
812
813         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
814         kbuf->buf_align = bss_align = 1;
815         kbuf->bufsz = bss_sz = 0;
816
817         for (i = 0; i < pi->ehdr->e_shnum; i++) {
818                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
819                         continue;
820
821                 align = sechdrs[i].sh_addralign;
822                 if (sechdrs[i].sh_type != SHT_NOBITS) {
823                         if (kbuf->buf_align < align)
824                                 kbuf->buf_align = align;
825                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
826                         kbuf->bufsz += sechdrs[i].sh_size;
827                 } else {
828                         if (bss_align < align)
829                                 bss_align = align;
830                         bss_sz = ALIGN(bss_sz, align);
831                         bss_sz += sechdrs[i].sh_size;
832                 }
833         }
834         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
835         kbuf->memsz = kbuf->bufsz + bss_sz;
836         if (kbuf->buf_align < bss_align)
837                 kbuf->buf_align = bss_align;
838
839         kbuf->buffer = vzalloc(kbuf->bufsz);
840         if (!kbuf->buffer)
841                 return -ENOMEM;
842         pi->purgatory_buf = kbuf->buffer;
843
844         ret = kexec_add_buffer(kbuf);
845         if (ret)
846                 goto out;
847
848         return 0;
849 out:
850         vfree(pi->purgatory_buf);
851         pi->purgatory_buf = NULL;
852         return ret;
853 }
854
855 /*
856  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
857  * @pi:         Purgatory to be loaded.
858  * @kbuf:       Buffer prepared to store purgatory.
859  *
860  * Allocates the memory needed for the buffer. Caller is responsible to free
861  * the memory after use.
862  *
863  * Return: 0 on success, negative errno on error.
864  */
865 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
866                                          struct kexec_buf *kbuf)
867 {
868         unsigned long bss_addr;
869         unsigned long offset;
870         Elf_Shdr *sechdrs;
871         int i;
872
873         /*
874          * The section headers in kexec_purgatory are read-only. In order to
875          * have them modifiable make a temporary copy.
876          */
877         sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
878         if (!sechdrs)
879                 return -ENOMEM;
880         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
881                pi->ehdr->e_shnum * sizeof(Elf_Shdr));
882         pi->sechdrs = sechdrs;
883
884         offset = 0;
885         bss_addr = kbuf->mem + kbuf->bufsz;
886         kbuf->image->start = pi->ehdr->e_entry;
887
888         for (i = 0; i < pi->ehdr->e_shnum; i++) {
889                 unsigned long align;
890                 void *src, *dst;
891
892                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
893                         continue;
894
895                 align = sechdrs[i].sh_addralign;
896                 if (sechdrs[i].sh_type == SHT_NOBITS) {
897                         bss_addr = ALIGN(bss_addr, align);
898                         sechdrs[i].sh_addr = bss_addr;
899                         bss_addr += sechdrs[i].sh_size;
900                         continue;
901                 }
902
903                 offset = ALIGN(offset, align);
904
905                 /*
906                  * Check if the segment contains the entry point, if so,
907                  * calculate the value of image->start based on it.
908                  * If the compiler has produced more than one .text section
909                  * (Eg: .text.hot), they are generally after the main .text
910                  * section, and they shall not be used to calculate
911                  * image->start. So do not re-calculate image->start if it
912                  * is not set to the initial value, and warn the user so they
913                  * have a chance to fix their purgatory's linker script.
914                  */
915                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
916                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
917                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
918                                          + sechdrs[i].sh_size) &&
919                     !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) {
920                         kbuf->image->start -= sechdrs[i].sh_addr;
921                         kbuf->image->start += kbuf->mem + offset;
922                 }
923
924                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
925                 dst = pi->purgatory_buf + offset;
926                 memcpy(dst, src, sechdrs[i].sh_size);
927
928                 sechdrs[i].sh_addr = kbuf->mem + offset;
929                 sechdrs[i].sh_offset = offset;
930                 offset += sechdrs[i].sh_size;
931         }
932
933         return 0;
934 }
935
936 static int kexec_apply_relocations(struct kimage *image)
937 {
938         int i, ret;
939         struct purgatory_info *pi = &image->purgatory_info;
940         const Elf_Shdr *sechdrs;
941
942         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
943
944         for (i = 0; i < pi->ehdr->e_shnum; i++) {
945                 const Elf_Shdr *relsec;
946                 const Elf_Shdr *symtab;
947                 Elf_Shdr *section;
948
949                 relsec = sechdrs + i;
950
951                 if (relsec->sh_type != SHT_RELA &&
952                     relsec->sh_type != SHT_REL)
953                         continue;
954
955                 /*
956                  * For section of type SHT_RELA/SHT_REL,
957                  * ->sh_link contains section header index of associated
958                  * symbol table. And ->sh_info contains section header
959                  * index of section to which relocations apply.
960                  */
961                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
962                     relsec->sh_link >= pi->ehdr->e_shnum)
963                         return -ENOEXEC;
964
965                 section = pi->sechdrs + relsec->sh_info;
966                 symtab = sechdrs + relsec->sh_link;
967
968                 if (!(section->sh_flags & SHF_ALLOC))
969                         continue;
970
971                 /*
972                  * symtab->sh_link contain section header index of associated
973                  * string table.
974                  */
975                 if (symtab->sh_link >= pi->ehdr->e_shnum)
976                         /* Invalid section number? */
977                         continue;
978
979                 /*
980                  * Respective architecture needs to provide support for applying
981                  * relocations of type SHT_RELA/SHT_REL.
982                  */
983                 if (relsec->sh_type == SHT_RELA)
984                         ret = arch_kexec_apply_relocations_add(pi, section,
985                                                                relsec, symtab);
986                 else if (relsec->sh_type == SHT_REL)
987                         ret = arch_kexec_apply_relocations(pi, section,
988                                                            relsec, symtab);
989                 if (ret)
990                         return ret;
991         }
992
993         return 0;
994 }
995
996 /*
997  * kexec_load_purgatory - Load and relocate the purgatory object.
998  * @image:      Image to add the purgatory to.
999  * @kbuf:       Memory parameters to use.
1000  *
1001  * Allocates the memory needed for image->purgatory_info.sechdrs and
1002  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1003  * to free the memory after use.
1004  *
1005  * Return: 0 on success, negative errno on error.
1006  */
1007 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1008 {
1009         struct purgatory_info *pi = &image->purgatory_info;
1010         int ret;
1011
1012         if (kexec_purgatory_size <= 0)
1013                 return -EINVAL;
1014
1015         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1016
1017         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1018         if (ret)
1019                 return ret;
1020
1021         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1022         if (ret)
1023                 goto out_free_kbuf;
1024
1025         ret = kexec_apply_relocations(image);
1026         if (ret)
1027                 goto out;
1028
1029         return 0;
1030 out:
1031         vfree(pi->sechdrs);
1032         pi->sechdrs = NULL;
1033 out_free_kbuf:
1034         vfree(pi->purgatory_buf);
1035         pi->purgatory_buf = NULL;
1036         return ret;
1037 }
1038
1039 /*
1040  * kexec_purgatory_find_symbol - find a symbol in the purgatory
1041  * @pi:         Purgatory to search in.
1042  * @name:       Name of the symbol.
1043  *
1044  * Return: pointer to symbol in read-only symtab on success, NULL on error.
1045  */
1046 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1047                                                   const char *name)
1048 {
1049         const Elf_Shdr *sechdrs;
1050         const Elf_Ehdr *ehdr;
1051         const Elf_Sym *syms;
1052         const char *strtab;
1053         int i, k;
1054
1055         if (!pi->ehdr)
1056                 return NULL;
1057
1058         ehdr = pi->ehdr;
1059         sechdrs = (void *)ehdr + ehdr->e_shoff;
1060
1061         for (i = 0; i < ehdr->e_shnum; i++) {
1062                 if (sechdrs[i].sh_type != SHT_SYMTAB)
1063                         continue;
1064
1065                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1066                         /* Invalid strtab section number */
1067                         continue;
1068                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1069                 syms = (void *)ehdr + sechdrs[i].sh_offset;
1070
1071                 /* Go through symbols for a match */
1072                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1073                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1074                                 continue;
1075
1076                         if (strcmp(strtab + syms[k].st_name, name) != 0)
1077                                 continue;
1078
1079                         if (syms[k].st_shndx == SHN_UNDEF ||
1080                             syms[k].st_shndx >= ehdr->e_shnum) {
1081                                 pr_debug("Symbol: %s has bad section index %d.\n",
1082                                                 name, syms[k].st_shndx);
1083                                 return NULL;
1084                         }
1085
1086                         /* Found the symbol we are looking for */
1087                         return &syms[k];
1088                 }
1089         }
1090
1091         return NULL;
1092 }
1093
1094 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1095 {
1096         struct purgatory_info *pi = &image->purgatory_info;
1097         const Elf_Sym *sym;
1098         Elf_Shdr *sechdr;
1099
1100         sym = kexec_purgatory_find_symbol(pi, name);
1101         if (!sym)
1102                 return ERR_PTR(-EINVAL);
1103
1104         sechdr = &pi->sechdrs[sym->st_shndx];
1105
1106         /*
1107          * Returns the address where symbol will finally be loaded after
1108          * kexec_load_segment()
1109          */
1110         return (void *)(sechdr->sh_addr + sym->st_value);
1111 }
1112
1113 /*
1114  * Get or set value of a symbol. If "get_value" is true, symbol value is
1115  * returned in buf otherwise symbol value is set based on value in buf.
1116  */
1117 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1118                                    void *buf, unsigned int size, bool get_value)
1119 {
1120         struct purgatory_info *pi = &image->purgatory_info;
1121         const Elf_Sym *sym;
1122         Elf_Shdr *sec;
1123         char *sym_buf;
1124
1125         sym = kexec_purgatory_find_symbol(pi, name);
1126         if (!sym)
1127                 return -EINVAL;
1128
1129         if (sym->st_size != size) {
1130                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1131                        name, (unsigned long)sym->st_size, size);
1132                 return -EINVAL;
1133         }
1134
1135         sec = pi->sechdrs + sym->st_shndx;
1136
1137         if (sec->sh_type == SHT_NOBITS) {
1138                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1139                        get_value ? "get" : "set");
1140                 return -EINVAL;
1141         }
1142
1143         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1144
1145         if (get_value)
1146                 memcpy((void *)buf, sym_buf, size);
1147         else
1148                 memcpy((void *)sym_buf, buf, size);
1149
1150         return 0;
1151 }
1152 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1153
1154 int crash_exclude_mem_range(struct crash_mem *mem,
1155                             unsigned long long mstart, unsigned long long mend)
1156 {
1157         int i, j;
1158         unsigned long long start, end, p_start, p_end;
1159         struct range temp_range = {0, 0};
1160
1161         for (i = 0; i < mem->nr_ranges; i++) {
1162                 start = mem->ranges[i].start;
1163                 end = mem->ranges[i].end;
1164                 p_start = mstart;
1165                 p_end = mend;
1166
1167                 if (mstart > end || mend < start)
1168                         continue;
1169
1170                 /* Truncate any area outside of range */
1171                 if (mstart < start)
1172                         p_start = start;
1173                 if (mend > end)
1174                         p_end = end;
1175
1176                 /* Found completely overlapping range */
1177                 if (p_start == start && p_end == end) {
1178                         mem->ranges[i].start = 0;
1179                         mem->ranges[i].end = 0;
1180                         if (i < mem->nr_ranges - 1) {
1181                                 /* Shift rest of the ranges to left */
1182                                 for (j = i; j < mem->nr_ranges - 1; j++) {
1183                                         mem->ranges[j].start =
1184                                                 mem->ranges[j+1].start;
1185                                         mem->ranges[j].end =
1186                                                         mem->ranges[j+1].end;
1187                                 }
1188
1189                                 /*
1190                                  * Continue to check if there are another overlapping ranges
1191                                  * from the current position because of shifting the above
1192                                  * mem ranges.
1193                                  */
1194                                 i--;
1195                                 mem->nr_ranges--;
1196                                 continue;
1197                         }
1198                         mem->nr_ranges--;
1199                         return 0;
1200                 }
1201
1202                 if (p_start > start && p_end < end) {
1203                         /* Split original range */
1204                         mem->ranges[i].end = p_start - 1;
1205                         temp_range.start = p_end + 1;
1206                         temp_range.end = end;
1207                 } else if (p_start != start)
1208                         mem->ranges[i].end = p_start - 1;
1209                 else
1210                         mem->ranges[i].start = p_end + 1;
1211                 break;
1212         }
1213
1214         /* If a split happened, add the split to array */
1215         if (!temp_range.end)
1216                 return 0;
1217
1218         /* Split happened */
1219         if (i == mem->max_nr_ranges - 1)
1220                 return -ENOMEM;
1221
1222         /* Location where new range should go */
1223         j = i + 1;
1224         if (j < mem->nr_ranges) {
1225                 /* Move over all ranges one slot towards the end */
1226                 for (i = mem->nr_ranges - 1; i >= j; i--)
1227                         mem->ranges[i + 1] = mem->ranges[i];
1228         }
1229
1230         mem->ranges[j].start = temp_range.start;
1231         mem->ranges[j].end = temp_range.end;
1232         mem->nr_ranges++;
1233         return 0;
1234 }
1235
1236 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
1237                           void **addr, unsigned long *sz)
1238 {
1239         Elf64_Ehdr *ehdr;
1240         Elf64_Phdr *phdr;
1241         unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1242         unsigned char *buf;
1243         unsigned int cpu, i;
1244         unsigned long long notes_addr;
1245         unsigned long mstart, mend;
1246
1247         /* extra phdr for vmcoreinfo ELF note */
1248         nr_phdr = nr_cpus + 1;
1249         nr_phdr += mem->nr_ranges;
1250
1251         /*
1252          * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1253          * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1254          * I think this is required by tools like gdb. So same physical
1255          * memory will be mapped in two ELF headers. One will contain kernel
1256          * text virtual addresses and other will have __va(physical) addresses.
1257          */
1258
1259         nr_phdr++;
1260         elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1261         elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1262
1263         buf = vzalloc(elf_sz);
1264         if (!buf)
1265                 return -ENOMEM;
1266
1267         ehdr = (Elf64_Ehdr *)buf;
1268         phdr = (Elf64_Phdr *)(ehdr + 1);
1269         memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1270         ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1271         ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1272         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1273         ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1274         memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1275         ehdr->e_type = ET_CORE;
1276         ehdr->e_machine = ELF_ARCH;
1277         ehdr->e_version = EV_CURRENT;
1278         ehdr->e_phoff = sizeof(Elf64_Ehdr);
1279         ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1280         ehdr->e_phentsize = sizeof(Elf64_Phdr);
1281
1282         /* Prepare one phdr of type PT_NOTE for each present CPU */
1283         for_each_present_cpu(cpu) {
1284                 phdr->p_type = PT_NOTE;
1285                 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1286                 phdr->p_offset = phdr->p_paddr = notes_addr;
1287                 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1288                 (ehdr->e_phnum)++;
1289                 phdr++;
1290         }
1291
1292         /* Prepare one PT_NOTE header for vmcoreinfo */
1293         phdr->p_type = PT_NOTE;
1294         phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1295         phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1296         (ehdr->e_phnum)++;
1297         phdr++;
1298
1299         /* Prepare PT_LOAD type program header for kernel text region */
1300         if (need_kernel_map) {
1301                 phdr->p_type = PT_LOAD;
1302                 phdr->p_flags = PF_R|PF_W|PF_X;
1303                 phdr->p_vaddr = (unsigned long) _text;
1304                 phdr->p_filesz = phdr->p_memsz = _end - _text;
1305                 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1306                 ehdr->e_phnum++;
1307                 phdr++;
1308         }
1309
1310         /* Go through all the ranges in mem->ranges[] and prepare phdr */
1311         for (i = 0; i < mem->nr_ranges; i++) {
1312                 mstart = mem->ranges[i].start;
1313                 mend = mem->ranges[i].end;
1314
1315                 phdr->p_type = PT_LOAD;
1316                 phdr->p_flags = PF_R|PF_W|PF_X;
1317                 phdr->p_offset  = mstart;
1318
1319                 phdr->p_paddr = mstart;
1320                 phdr->p_vaddr = (unsigned long) __va(mstart);
1321                 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1322                 phdr->p_align = 0;
1323                 ehdr->e_phnum++;
1324                 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",
1325                         phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1326                         ehdr->e_phnum, phdr->p_offset);
1327                 phdr++;
1328         }
1329
1330         *addr = buf;
1331         *sz = elf_sz;
1332         return 0;
1333 }