tools headers UAPI: Sync openat2.h with the kernel sources
[linux-2.6-microblaze.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 static int kexec_calculate_store_digests(struct kimage *image);
33
34 /*
35  * Currently this is the only default function that is exported as some
36  * architectures need it to do additional handlings.
37  * In the future, other default functions may be exported too if required.
38  */
39 int kexec_image_probe_default(struct kimage *image, void *buf,
40                               unsigned long buf_len)
41 {
42         const struct kexec_file_ops * const *fops;
43         int ret = -ENOEXEC;
44
45         for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
46                 ret = (*fops)->probe(buf, buf_len);
47                 if (!ret) {
48                         image->fops = *fops;
49                         return ret;
50                 }
51         }
52
53         return ret;
54 }
55
56 /* Architectures can provide this probe function */
57 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
58                                          unsigned long buf_len)
59 {
60         return kexec_image_probe_default(image, buf, buf_len);
61 }
62
63 static void *kexec_image_load_default(struct kimage *image)
64 {
65         if (!image->fops || !image->fops->load)
66                 return ERR_PTR(-ENOEXEC);
67
68         return image->fops->load(image, image->kernel_buf,
69                                  image->kernel_buf_len, image->initrd_buf,
70                                  image->initrd_buf_len, image->cmdline_buf,
71                                  image->cmdline_buf_len);
72 }
73
74 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
75 {
76         return kexec_image_load_default(image);
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 int __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
88 {
89         return kexec_image_post_load_cleanup_default(image);
90 }
91
92 #ifdef CONFIG_KEXEC_SIG
93 static int kexec_image_verify_sig_default(struct kimage *image, void *buf,
94                                           unsigned long buf_len)
95 {
96         if (!image->fops || !image->fops->verify_sig) {
97                 pr_debug("kernel loader does not support signature verification.\n");
98                 return -EKEYREJECTED;
99         }
100
101         return image->fops->verify_sig(buf, buf_len);
102 }
103
104 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
105                                         unsigned long buf_len)
106 {
107         return kexec_image_verify_sig_default(image, buf, buf_len);
108 }
109 #endif
110
111 /*
112  * arch_kexec_apply_relocations_add - apply relocations of type RELA
113  * @pi:         Purgatory to be relocated.
114  * @section:    Section relocations applying to.
115  * @relsec:     Section containing RELAs.
116  * @symtab:     Corresponding symtab.
117  *
118  * Return: 0 on success, negative errno on error.
119  */
120 int __weak
121 arch_kexec_apply_relocations_add(struct purgatory_info *pi, Elf_Shdr *section,
122                                  const Elf_Shdr *relsec, const Elf_Shdr *symtab)
123 {
124         pr_err("RELA relocation unsupported.\n");
125         return -ENOEXEC;
126 }
127
128 /*
129  * arch_kexec_apply_relocations - apply relocations of type REL
130  * @pi:         Purgatory to be relocated.
131  * @section:    Section relocations applying to.
132  * @relsec:     Section containing RELs.
133  * @symtab:     Corresponding symtab.
134  *
135  * Return: 0 on success, negative errno on error.
136  */
137 int __weak
138 arch_kexec_apply_relocations(struct purgatory_info *pi, Elf_Shdr *section,
139                              const Elf_Shdr *relsec, const Elf_Shdr *symtab)
140 {
141         pr_err("REL relocation unsupported.\n");
142         return -ENOEXEC;
143 }
144
145 /*
146  * Free up memory used by kernel, initrd, and command line. This is temporary
147  * memory allocation which is not needed any more after these buffers have
148  * been loaded into separate segments and have been copied elsewhere.
149  */
150 void kimage_file_post_load_cleanup(struct kimage *image)
151 {
152         struct purgatory_info *pi = &image->purgatory_info;
153
154         vfree(image->kernel_buf);
155         image->kernel_buf = NULL;
156
157         vfree(image->initrd_buf);
158         image->initrd_buf = NULL;
159
160         kfree(image->cmdline_buf);
161         image->cmdline_buf = NULL;
162
163         vfree(pi->purgatory_buf);
164         pi->purgatory_buf = NULL;
165
166         vfree(pi->sechdrs);
167         pi->sechdrs = NULL;
168
169 #ifdef CONFIG_IMA_KEXEC
170         vfree(image->ima_buffer);
171         image->ima_buffer = NULL;
172 #endif /* CONFIG_IMA_KEXEC */
173
174         /* See if architecture has anything to cleanup post load */
175         arch_kimage_file_post_load_cleanup(image);
176
177         /*
178          * Above call should have called into bootloader to free up
179          * any data stored in kimage->image_loader_data. It should
180          * be ok now to free it up.
181          */
182         kfree(image->image_loader_data);
183         image->image_loader_data = NULL;
184 }
185
186 #ifdef CONFIG_KEXEC_SIG
187 static int
188 kimage_validate_signature(struct kimage *image)
189 {
190         int ret;
191
192         ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
193                                            image->kernel_buf_len);
194         if (ret) {
195
196                 if (IS_ENABLED(CONFIG_KEXEC_SIG_FORCE)) {
197                         pr_notice("Enforced kernel signature verification failed (%d).\n", ret);
198                         return ret;
199                 }
200
201                 /*
202                  * If IMA is guaranteed to appraise a signature on the kexec
203                  * image, permit it even if the kernel is otherwise locked
204                  * down.
205                  */
206                 if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
207                     security_locked_down(LOCKDOWN_KEXEC))
208                         return -EPERM;
209
210                 pr_debug("kernel signature verification failed (%d).\n", ret);
211         }
212
213         return 0;
214 }
215 #endif
216
217 /*
218  * In file mode list of segments is prepared by kernel. Copy relevant
219  * data from user space, do error checking, prepare segment list
220  */
221 static int
222 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
223                              const char __user *cmdline_ptr,
224                              unsigned long cmdline_len, unsigned flags)
225 {
226         int ret;
227         void *ldata;
228
229         ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf,
230                                        INT_MAX, NULL, READING_KEXEC_IMAGE);
231         if (ret < 0)
232                 return ret;
233         image->kernel_buf_len = ret;
234
235         /* Call arch image probe handlers */
236         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
237                                             image->kernel_buf_len);
238         if (ret)
239                 goto out;
240
241 #ifdef CONFIG_KEXEC_SIG
242         ret = kimage_validate_signature(image);
243
244         if (ret)
245                 goto out;
246 #endif
247         /* It is possible that there no initramfs is being loaded */
248         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
249                 ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf,
250                                                INT_MAX, NULL,
251                                                READING_KEXEC_INITRAMFS);
252                 if (ret < 0)
253                         goto out;
254                 image->initrd_buf_len = ret;
255                 ret = 0;
256         }
257
258         if (cmdline_len) {
259                 image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
260                 if (IS_ERR(image->cmdline_buf)) {
261                         ret = PTR_ERR(image->cmdline_buf);
262                         image->cmdline_buf = NULL;
263                         goto out;
264                 }
265
266                 image->cmdline_buf_len = cmdline_len;
267
268                 /* command line should be a string with last byte null */
269                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
270                         ret = -EINVAL;
271                         goto out;
272                 }
273
274                 ima_kexec_cmdline(kernel_fd, image->cmdline_buf,
275                                   image->cmdline_buf_len - 1);
276         }
277
278         /* IMA needs to pass the measurement list to the next kernel. */
279         ima_add_kexec_buffer(image);
280
281         /* Call arch image load handlers */
282         ldata = arch_kexec_kernel_image_load(image);
283
284         if (IS_ERR(ldata)) {
285                 ret = PTR_ERR(ldata);
286                 goto out;
287         }
288
289         image->image_loader_data = ldata;
290 out:
291         /* In case of error, free up all allocated memory in this function */
292         if (ret)
293                 kimage_file_post_load_cleanup(image);
294         return ret;
295 }
296
297 static int
298 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
299                        int initrd_fd, const char __user *cmdline_ptr,
300                        unsigned long cmdline_len, unsigned long flags)
301 {
302         int ret;
303         struct kimage *image;
304         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
305
306         image = do_kimage_alloc_init();
307         if (!image)
308                 return -ENOMEM;
309
310         image->file_mode = 1;
311
312         if (kexec_on_panic) {
313                 /* Enable special crash kernel control page alloc policy. */
314                 image->control_page = crashk_res.start;
315                 image->type = KEXEC_TYPE_CRASH;
316         }
317
318         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
319                                            cmdline_ptr, cmdline_len, flags);
320         if (ret)
321                 goto out_free_image;
322
323         ret = sanity_check_segment_list(image);
324         if (ret)
325                 goto out_free_post_load_bufs;
326
327         ret = -ENOMEM;
328         image->control_code_page = kimage_alloc_control_pages(image,
329                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
330         if (!image->control_code_page) {
331                 pr_err("Could not allocate control_code_buffer\n");
332                 goto out_free_post_load_bufs;
333         }
334
335         if (!kexec_on_panic) {
336                 image->swap_page = kimage_alloc_control_pages(image, 0);
337                 if (!image->swap_page) {
338                         pr_err("Could not allocate swap buffer\n");
339                         goto out_free_control_pages;
340                 }
341         }
342
343         *rimage = image;
344         return 0;
345 out_free_control_pages:
346         kimage_free_page_list(&image->control_pages);
347 out_free_post_load_bufs:
348         kimage_file_post_load_cleanup(image);
349 out_free_image:
350         kfree(image);
351         return ret;
352 }
353
354 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
355                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
356                 unsigned long, flags)
357 {
358         int ret = 0, i;
359         struct kimage **dest_image, *image;
360
361         /* We only trust the superuser with rebooting the system. */
362         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
363                 return -EPERM;
364
365         /* Make sure we have a legal set of flags */
366         if (flags != (flags & KEXEC_FILE_FLAGS))
367                 return -EINVAL;
368
369         image = NULL;
370
371         if (!mutex_trylock(&kexec_mutex))
372                 return -EBUSY;
373
374         dest_image = &kexec_image;
375         if (flags & KEXEC_FILE_ON_CRASH) {
376                 dest_image = &kexec_crash_image;
377                 if (kexec_crash_image)
378                         arch_kexec_unprotect_crashkres();
379         }
380
381         if (flags & KEXEC_FILE_UNLOAD)
382                 goto exchange;
383
384         /*
385          * In case of crash, new kernel gets loaded in reserved region. It is
386          * same memory where old crash kernel might be loaded. Free any
387          * current crash dump kernel before we corrupt it.
388          */
389         if (flags & KEXEC_FILE_ON_CRASH)
390                 kimage_free(xchg(&kexec_crash_image, NULL));
391
392         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
393                                      cmdline_len, flags);
394         if (ret)
395                 goto out;
396
397         ret = machine_kexec_prepare(image);
398         if (ret)
399                 goto out;
400
401         /*
402          * Some architecture(like S390) may touch the crash memory before
403          * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
404          */
405         ret = kimage_crash_copy_vmcoreinfo(image);
406         if (ret)
407                 goto out;
408
409         ret = kexec_calculate_store_digests(image);
410         if (ret)
411                 goto out;
412
413         for (i = 0; i < image->nr_segments; i++) {
414                 struct kexec_segment *ksegment;
415
416                 ksegment = &image->segment[i];
417                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
418                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
419                          ksegment->memsz);
420
421                 ret = kimage_load_segment(image, &image->segment[i]);
422                 if (ret)
423                         goto out;
424         }
425
426         kimage_terminate(image);
427
428         ret = machine_kexec_post_load(image);
429         if (ret)
430                 goto out;
431
432         /*
433          * Free up any temporary buffers allocated which are not needed
434          * after image has been loaded
435          */
436         kimage_file_post_load_cleanup(image);
437 exchange:
438         image = xchg(dest_image, image);
439 out:
440         if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
441                 arch_kexec_protect_crashkres();
442
443         mutex_unlock(&kexec_mutex);
444         kimage_free(image);
445         return ret;
446 }
447
448 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
449                                     struct kexec_buf *kbuf)
450 {
451         struct kimage *image = kbuf->image;
452         unsigned long temp_start, temp_end;
453
454         temp_end = min(end, kbuf->buf_max);
455         temp_start = temp_end - kbuf->memsz;
456
457         do {
458                 /* align down start */
459                 temp_start = temp_start & (~(kbuf->buf_align - 1));
460
461                 if (temp_start < start || temp_start < kbuf->buf_min)
462                         return 0;
463
464                 temp_end = temp_start + kbuf->memsz - 1;
465
466                 /*
467                  * Make sure this does not conflict with any of existing
468                  * segments
469                  */
470                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
471                         temp_start = temp_start - PAGE_SIZE;
472                         continue;
473                 }
474
475                 /* We found a suitable memory range */
476                 break;
477         } while (1);
478
479         /* If we are here, we found a suitable memory range */
480         kbuf->mem = temp_start;
481
482         /* Success, stop navigating through remaining System RAM ranges */
483         return 1;
484 }
485
486 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
487                                      struct kexec_buf *kbuf)
488 {
489         struct kimage *image = kbuf->image;
490         unsigned long temp_start, temp_end;
491
492         temp_start = max(start, kbuf->buf_min);
493
494         do {
495                 temp_start = ALIGN(temp_start, kbuf->buf_align);
496                 temp_end = temp_start + kbuf->memsz - 1;
497
498                 if (temp_end > end || temp_end > kbuf->buf_max)
499                         return 0;
500                 /*
501                  * Make sure this does not conflict with any of existing
502                  * segments
503                  */
504                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
505                         temp_start = temp_start + PAGE_SIZE;
506                         continue;
507                 }
508
509                 /* We found a suitable memory range */
510                 break;
511         } while (1);
512
513         /* If we are here, we found a suitable memory range */
514         kbuf->mem = temp_start;
515
516         /* Success, stop navigating through remaining System RAM ranges */
517         return 1;
518 }
519
520 static int locate_mem_hole_callback(struct resource *res, void *arg)
521 {
522         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
523         u64 start = res->start, end = res->end;
524         unsigned long sz = end - start + 1;
525
526         /* Returning 0 will take to next memory range */
527
528         /* Don't use memory that will be detected and handled by a driver. */
529         if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
530                 return 0;
531
532         if (sz < kbuf->memsz)
533                 return 0;
534
535         if (end < kbuf->buf_min || start > kbuf->buf_max)
536                 return 0;
537
538         /*
539          * Allocate memory top down with-in ram range. Otherwise bottom up
540          * allocation.
541          */
542         if (kbuf->top_down)
543                 return locate_mem_hole_top_down(start, end, kbuf);
544         return locate_mem_hole_bottom_up(start, end, kbuf);
545 }
546
547 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
548 static int kexec_walk_memblock(struct kexec_buf *kbuf,
549                                int (*func)(struct resource *, void *))
550 {
551         int ret = 0;
552         u64 i;
553         phys_addr_t mstart, mend;
554         struct resource res = { };
555
556         if (kbuf->image->type == KEXEC_TYPE_CRASH)
557                 return func(&crashk_res, kbuf);
558
559         if (kbuf->top_down) {
560                 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
561                                                 &mstart, &mend, NULL) {
562                         /*
563                          * In memblock, end points to the first byte after the
564                          * range while in kexec, end points to the last byte
565                          * in the range.
566                          */
567                         res.start = mstart;
568                         res.end = mend - 1;
569                         ret = func(&res, kbuf);
570                         if (ret)
571                                 break;
572                 }
573         } else {
574                 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
575                                         &mstart, &mend, NULL) {
576                         /*
577                          * In memblock, end points to the first byte after the
578                          * range while in kexec, end points to the last byte
579                          * in the range.
580                          */
581                         res.start = mstart;
582                         res.end = mend - 1;
583                         ret = func(&res, kbuf);
584                         if (ret)
585                                 break;
586                 }
587         }
588
589         return ret;
590 }
591 #else
592 static int kexec_walk_memblock(struct kexec_buf *kbuf,
593                                int (*func)(struct resource *, void *))
594 {
595         return 0;
596 }
597 #endif
598
599 /**
600  * kexec_walk_resources - call func(data) on free memory regions
601  * @kbuf:       Context info for the search. Also passed to @func.
602  * @func:       Function to call for each memory region.
603  *
604  * Return: The memory walk will stop when func returns a non-zero value
605  * and that value will be returned. If all free regions are visited without
606  * func returning non-zero, then zero will be returned.
607  */
608 static int kexec_walk_resources(struct kexec_buf *kbuf,
609                                 int (*func)(struct resource *, void *))
610 {
611         if (kbuf->image->type == KEXEC_TYPE_CRASH)
612                 return walk_iomem_res_desc(crashk_res.desc,
613                                            IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
614                                            crashk_res.start, crashk_res.end,
615                                            kbuf, func);
616         else
617                 return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
618 }
619
620 /**
621  * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
622  * @kbuf:       Parameters for the memory search.
623  *
624  * On success, kbuf->mem will have the start address of the memory region found.
625  *
626  * Return: 0 on success, negative errno on error.
627  */
628 int kexec_locate_mem_hole(struct kexec_buf *kbuf)
629 {
630         int ret;
631
632         /* Arch knows where to place */
633         if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
634                 return 0;
635
636         if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
637                 ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
638         else
639                 ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
640
641         return ret == 1 ? 0 : -EADDRNOTAVAIL;
642 }
643
644 /**
645  * arch_kexec_locate_mem_hole - Find free memory to place the segments.
646  * @kbuf:                       Parameters for the memory search.
647  *
648  * On success, kbuf->mem will have the start address of the memory region found.
649  *
650  * Return: 0 on success, negative errno on error.
651  */
652 int __weak arch_kexec_locate_mem_hole(struct kexec_buf *kbuf)
653 {
654         return kexec_locate_mem_hole(kbuf);
655 }
656
657 /**
658  * kexec_add_buffer - place a buffer in a kexec segment
659  * @kbuf:       Buffer contents and memory parameters.
660  *
661  * This function assumes that kexec_mutex is held.
662  * On successful return, @kbuf->mem will have the physical address of
663  * the buffer in memory.
664  *
665  * Return: 0 on success, negative errno on error.
666  */
667 int kexec_add_buffer(struct kexec_buf *kbuf)
668 {
669         struct kexec_segment *ksegment;
670         int ret;
671
672         /* Currently adding segment this way is allowed only in file mode */
673         if (!kbuf->image->file_mode)
674                 return -EINVAL;
675
676         if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
677                 return -EINVAL;
678
679         /*
680          * Make sure we are not trying to add buffer after allocating
681          * control pages. All segments need to be placed first before
682          * any control pages are allocated. As control page allocation
683          * logic goes through list of segments to make sure there are
684          * no destination overlaps.
685          */
686         if (!list_empty(&kbuf->image->control_pages)) {
687                 WARN_ON(1);
688                 return -EINVAL;
689         }
690
691         /* Ensure minimum alignment needed for segments. */
692         kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
693         kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
694
695         /* Walk the RAM ranges and allocate a suitable range for the buffer */
696         ret = arch_kexec_locate_mem_hole(kbuf);
697         if (ret)
698                 return ret;
699
700         /* Found a suitable memory range */
701         ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
702         ksegment->kbuf = kbuf->buffer;
703         ksegment->bufsz = kbuf->bufsz;
704         ksegment->mem = kbuf->mem;
705         ksegment->memsz = kbuf->memsz;
706         kbuf->image->nr_segments++;
707         return 0;
708 }
709
710 /* Calculate and store the digest of segments */
711 static int kexec_calculate_store_digests(struct kimage *image)
712 {
713         struct crypto_shash *tfm;
714         struct shash_desc *desc;
715         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
716         size_t desc_size, nullsz;
717         char *digest;
718         void *zero_buf;
719         struct kexec_sha_region *sha_regions;
720         struct purgatory_info *pi = &image->purgatory_info;
721
722         if (!IS_ENABLED(CONFIG_ARCH_HAS_KEXEC_PURGATORY))
723                 return 0;
724
725         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
726         zero_buf_sz = PAGE_SIZE;
727
728         tfm = crypto_alloc_shash("sha256", 0, 0);
729         if (IS_ERR(tfm)) {
730                 ret = PTR_ERR(tfm);
731                 goto out;
732         }
733
734         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
735         desc = kzalloc(desc_size, GFP_KERNEL);
736         if (!desc) {
737                 ret = -ENOMEM;
738                 goto out_free_tfm;
739         }
740
741         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
742         sha_regions = vzalloc(sha_region_sz);
743         if (!sha_regions)
744                 goto out_free_desc;
745
746         desc->tfm   = tfm;
747
748         ret = crypto_shash_init(desc);
749         if (ret < 0)
750                 goto out_free_sha_regions;
751
752         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
753         if (!digest) {
754                 ret = -ENOMEM;
755                 goto out_free_sha_regions;
756         }
757
758         for (j = i = 0; i < image->nr_segments; i++) {
759                 struct kexec_segment *ksegment;
760
761                 ksegment = &image->segment[i];
762                 /*
763                  * Skip purgatory as it will be modified once we put digest
764                  * info in purgatory.
765                  */
766                 if (ksegment->kbuf == pi->purgatory_buf)
767                         continue;
768
769                 ret = crypto_shash_update(desc, ksegment->kbuf,
770                                           ksegment->bufsz);
771                 if (ret)
772                         break;
773
774                 /*
775                  * Assume rest of the buffer is filled with zero and
776                  * update digest accordingly.
777                  */
778                 nullsz = ksegment->memsz - ksegment->bufsz;
779                 while (nullsz) {
780                         unsigned long bytes = nullsz;
781
782                         if (bytes > zero_buf_sz)
783                                 bytes = zero_buf_sz;
784                         ret = crypto_shash_update(desc, zero_buf, bytes);
785                         if (ret)
786                                 break;
787                         nullsz -= bytes;
788                 }
789
790                 if (ret)
791                         break;
792
793                 sha_regions[j].start = ksegment->mem;
794                 sha_regions[j].len = ksegment->memsz;
795                 j++;
796         }
797
798         if (!ret) {
799                 ret = crypto_shash_final(desc, digest);
800                 if (ret)
801                         goto out_free_digest;
802                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
803                                                      sha_regions, sha_region_sz, 0);
804                 if (ret)
805                         goto out_free_digest;
806
807                 ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
808                                                      digest, SHA256_DIGEST_SIZE, 0);
809                 if (ret)
810                         goto out_free_digest;
811         }
812
813 out_free_digest:
814         kfree(digest);
815 out_free_sha_regions:
816         vfree(sha_regions);
817 out_free_desc:
818         kfree(desc);
819 out_free_tfm:
820         kfree(tfm);
821 out:
822         return ret;
823 }
824
825 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
826 /*
827  * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
828  * @pi:         Purgatory to be loaded.
829  * @kbuf:       Buffer to setup.
830  *
831  * Allocates the memory needed for the buffer. Caller is responsible to free
832  * the memory after use.
833  *
834  * Return: 0 on success, negative errno on error.
835  */
836 static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
837                                       struct kexec_buf *kbuf)
838 {
839         const Elf_Shdr *sechdrs;
840         unsigned long bss_align;
841         unsigned long bss_sz;
842         unsigned long align;
843         int i, ret;
844
845         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
846         kbuf->buf_align = bss_align = 1;
847         kbuf->bufsz = bss_sz = 0;
848
849         for (i = 0; i < pi->ehdr->e_shnum; i++) {
850                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
851                         continue;
852
853                 align = sechdrs[i].sh_addralign;
854                 if (sechdrs[i].sh_type != SHT_NOBITS) {
855                         if (kbuf->buf_align < align)
856                                 kbuf->buf_align = align;
857                         kbuf->bufsz = ALIGN(kbuf->bufsz, align);
858                         kbuf->bufsz += sechdrs[i].sh_size;
859                 } else {
860                         if (bss_align < align)
861                                 bss_align = align;
862                         bss_sz = ALIGN(bss_sz, align);
863                         bss_sz += sechdrs[i].sh_size;
864                 }
865         }
866         kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
867         kbuf->memsz = kbuf->bufsz + bss_sz;
868         if (kbuf->buf_align < bss_align)
869                 kbuf->buf_align = bss_align;
870
871         kbuf->buffer = vzalloc(kbuf->bufsz);
872         if (!kbuf->buffer)
873                 return -ENOMEM;
874         pi->purgatory_buf = kbuf->buffer;
875
876         ret = kexec_add_buffer(kbuf);
877         if (ret)
878                 goto out;
879
880         return 0;
881 out:
882         vfree(pi->purgatory_buf);
883         pi->purgatory_buf = NULL;
884         return ret;
885 }
886
887 /*
888  * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
889  * @pi:         Purgatory to be loaded.
890  * @kbuf:       Buffer prepared to store purgatory.
891  *
892  * Allocates the memory needed for the buffer. Caller is responsible to free
893  * the memory after use.
894  *
895  * Return: 0 on success, negative errno on error.
896  */
897 static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
898                                          struct kexec_buf *kbuf)
899 {
900         unsigned long bss_addr;
901         unsigned long offset;
902         Elf_Shdr *sechdrs;
903         int i;
904
905         /*
906          * The section headers in kexec_purgatory are read-only. In order to
907          * have them modifiable make a temporary copy.
908          */
909         sechdrs = vzalloc(array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum));
910         if (!sechdrs)
911                 return -ENOMEM;
912         memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff,
913                pi->ehdr->e_shnum * sizeof(Elf_Shdr));
914         pi->sechdrs = sechdrs;
915
916         offset = 0;
917         bss_addr = kbuf->mem + kbuf->bufsz;
918         kbuf->image->start = pi->ehdr->e_entry;
919
920         for (i = 0; i < pi->ehdr->e_shnum; i++) {
921                 unsigned long align;
922                 void *src, *dst;
923
924                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
925                         continue;
926
927                 align = sechdrs[i].sh_addralign;
928                 if (sechdrs[i].sh_type == SHT_NOBITS) {
929                         bss_addr = ALIGN(bss_addr, align);
930                         sechdrs[i].sh_addr = bss_addr;
931                         bss_addr += sechdrs[i].sh_size;
932                         continue;
933                 }
934
935                 offset = ALIGN(offset, align);
936                 if (sechdrs[i].sh_flags & SHF_EXECINSTR &&
937                     pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
938                     pi->ehdr->e_entry < (sechdrs[i].sh_addr
939                                          + sechdrs[i].sh_size)) {
940                         kbuf->image->start -= sechdrs[i].sh_addr;
941                         kbuf->image->start += kbuf->mem + offset;
942                 }
943
944                 src = (void *)pi->ehdr + sechdrs[i].sh_offset;
945                 dst = pi->purgatory_buf + offset;
946                 memcpy(dst, src, sechdrs[i].sh_size);
947
948                 sechdrs[i].sh_addr = kbuf->mem + offset;
949                 sechdrs[i].sh_offset = offset;
950                 offset += sechdrs[i].sh_size;
951         }
952
953         return 0;
954 }
955
956 static int kexec_apply_relocations(struct kimage *image)
957 {
958         int i, ret;
959         struct purgatory_info *pi = &image->purgatory_info;
960         const Elf_Shdr *sechdrs;
961
962         sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
963
964         for (i = 0; i < pi->ehdr->e_shnum; i++) {
965                 const Elf_Shdr *relsec;
966                 const Elf_Shdr *symtab;
967                 Elf_Shdr *section;
968
969                 relsec = sechdrs + i;
970
971                 if (relsec->sh_type != SHT_RELA &&
972                     relsec->sh_type != SHT_REL)
973                         continue;
974
975                 /*
976                  * For section of type SHT_RELA/SHT_REL,
977                  * ->sh_link contains section header index of associated
978                  * symbol table. And ->sh_info contains section header
979                  * index of section to which relocations apply.
980                  */
981                 if (relsec->sh_info >= pi->ehdr->e_shnum ||
982                     relsec->sh_link >= pi->ehdr->e_shnum)
983                         return -ENOEXEC;
984
985                 section = pi->sechdrs + relsec->sh_info;
986                 symtab = sechdrs + relsec->sh_link;
987
988                 if (!(section->sh_flags & SHF_ALLOC))
989                         continue;
990
991                 /*
992                  * symtab->sh_link contain section header index of associated
993                  * string table.
994                  */
995                 if (symtab->sh_link >= pi->ehdr->e_shnum)
996                         /* Invalid section number? */
997                         continue;
998
999                 /*
1000                  * Respective architecture needs to provide support for applying
1001                  * relocations of type SHT_RELA/SHT_REL.
1002                  */
1003                 if (relsec->sh_type == SHT_RELA)
1004                         ret = arch_kexec_apply_relocations_add(pi, section,
1005                                                                relsec, symtab);
1006                 else if (relsec->sh_type == SHT_REL)
1007                         ret = arch_kexec_apply_relocations(pi, section,
1008                                                            relsec, symtab);
1009                 if (ret)
1010                         return ret;
1011         }
1012
1013         return 0;
1014 }
1015
1016 /*
1017  * kexec_load_purgatory - Load and relocate the purgatory object.
1018  * @image:      Image to add the purgatory to.
1019  * @kbuf:       Memory parameters to use.
1020  *
1021  * Allocates the memory needed for image->purgatory_info.sechdrs and
1022  * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
1023  * to free the memory after use.
1024  *
1025  * Return: 0 on success, negative errno on error.
1026  */
1027 int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
1028 {
1029         struct purgatory_info *pi = &image->purgatory_info;
1030         int ret;
1031
1032         if (kexec_purgatory_size <= 0)
1033                 return -EINVAL;
1034
1035         pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
1036
1037         ret = kexec_purgatory_setup_kbuf(pi, kbuf);
1038         if (ret)
1039                 return ret;
1040
1041         ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
1042         if (ret)
1043                 goto out_free_kbuf;
1044
1045         ret = kexec_apply_relocations(image);
1046         if (ret)
1047                 goto out;
1048
1049         return 0;
1050 out:
1051         vfree(pi->sechdrs);
1052         pi->sechdrs = NULL;
1053 out_free_kbuf:
1054         vfree(pi->purgatory_buf);
1055         pi->purgatory_buf = NULL;
1056         return ret;
1057 }
1058
1059 /*
1060  * kexec_purgatory_find_symbol - find a symbol in the purgatory
1061  * @pi:         Purgatory to search in.
1062  * @name:       Name of the symbol.
1063  *
1064  * Return: pointer to symbol in read-only symtab on success, NULL on error.
1065  */
1066 static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
1067                                                   const char *name)
1068 {
1069         const Elf_Shdr *sechdrs;
1070         const Elf_Ehdr *ehdr;
1071         const Elf_Sym *syms;
1072         const char *strtab;
1073         int i, k;
1074
1075         if (!pi->ehdr)
1076                 return NULL;
1077
1078         ehdr = pi->ehdr;
1079         sechdrs = (void *)ehdr + ehdr->e_shoff;
1080
1081         for (i = 0; i < ehdr->e_shnum; i++) {
1082                 if (sechdrs[i].sh_type != SHT_SYMTAB)
1083                         continue;
1084
1085                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
1086                         /* Invalid strtab section number */
1087                         continue;
1088                 strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
1089                 syms = (void *)ehdr + sechdrs[i].sh_offset;
1090
1091                 /* Go through symbols for a match */
1092                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
1093                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
1094                                 continue;
1095
1096                         if (strcmp(strtab + syms[k].st_name, name) != 0)
1097                                 continue;
1098
1099                         if (syms[k].st_shndx == SHN_UNDEF ||
1100                             syms[k].st_shndx >= ehdr->e_shnum) {
1101                                 pr_debug("Symbol: %s has bad section index %d.\n",
1102                                                 name, syms[k].st_shndx);
1103                                 return NULL;
1104                         }
1105
1106                         /* Found the symbol we are looking for */
1107                         return &syms[k];
1108                 }
1109         }
1110
1111         return NULL;
1112 }
1113
1114 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
1115 {
1116         struct purgatory_info *pi = &image->purgatory_info;
1117         const Elf_Sym *sym;
1118         Elf_Shdr *sechdr;
1119
1120         sym = kexec_purgatory_find_symbol(pi, name);
1121         if (!sym)
1122                 return ERR_PTR(-EINVAL);
1123
1124         sechdr = &pi->sechdrs[sym->st_shndx];
1125
1126         /*
1127          * Returns the address where symbol will finally be loaded after
1128          * kexec_load_segment()
1129          */
1130         return (void *)(sechdr->sh_addr + sym->st_value);
1131 }
1132
1133 /*
1134  * Get or set value of a symbol. If "get_value" is true, symbol value is
1135  * returned in buf otherwise symbol value is set based on value in buf.
1136  */
1137 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
1138                                    void *buf, unsigned int size, bool get_value)
1139 {
1140         struct purgatory_info *pi = &image->purgatory_info;
1141         const Elf_Sym *sym;
1142         Elf_Shdr *sec;
1143         char *sym_buf;
1144
1145         sym = kexec_purgatory_find_symbol(pi, name);
1146         if (!sym)
1147                 return -EINVAL;
1148
1149         if (sym->st_size != size) {
1150                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
1151                        name, (unsigned long)sym->st_size, size);
1152                 return -EINVAL;
1153         }
1154
1155         sec = pi->sechdrs + sym->st_shndx;
1156
1157         if (sec->sh_type == SHT_NOBITS) {
1158                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
1159                        get_value ? "get" : "set");
1160                 return -EINVAL;
1161         }
1162
1163         sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
1164
1165         if (get_value)
1166                 memcpy((void *)buf, sym_buf, size);
1167         else
1168                 memcpy((void *)sym_buf, buf, size);
1169
1170         return 0;
1171 }
1172 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
1173
1174 int crash_exclude_mem_range(struct crash_mem *mem,
1175                             unsigned long long mstart, unsigned long long mend)
1176 {
1177         int i, j;
1178         unsigned long long start, end, p_start, p_end;
1179         struct crash_mem_range temp_range = {0, 0};
1180
1181         for (i = 0; i < mem->nr_ranges; i++) {
1182                 start = mem->ranges[i].start;
1183                 end = mem->ranges[i].end;
1184                 p_start = mstart;
1185                 p_end = mend;
1186
1187                 if (mstart > end || mend < start)
1188                         continue;
1189
1190                 /* Truncate any area outside of range */
1191                 if (mstart < start)
1192                         p_start = start;
1193                 if (mend > end)
1194                         p_end = end;
1195
1196                 /* Found completely overlapping range */
1197                 if (p_start == start && p_end == end) {
1198                         mem->ranges[i].start = 0;
1199                         mem->ranges[i].end = 0;
1200                         if (i < mem->nr_ranges - 1) {
1201                                 /* Shift rest of the ranges to left */
1202                                 for (j = i; j < mem->nr_ranges - 1; j++) {
1203                                         mem->ranges[j].start =
1204                                                 mem->ranges[j+1].start;
1205                                         mem->ranges[j].end =
1206                                                         mem->ranges[j+1].end;
1207                                 }
1208
1209                                 /*
1210                                  * Continue to check if there are another overlapping ranges
1211                                  * from the current position because of shifting the above
1212                                  * mem ranges.
1213                                  */
1214                                 i--;
1215                                 mem->nr_ranges--;
1216                                 continue;
1217                         }
1218                         mem->nr_ranges--;
1219                         return 0;
1220                 }
1221
1222                 if (p_start > start && p_end < end) {
1223                         /* Split original range */
1224                         mem->ranges[i].end = p_start - 1;
1225                         temp_range.start = p_end + 1;
1226                         temp_range.end = end;
1227                 } else if (p_start != start)
1228                         mem->ranges[i].end = p_start - 1;
1229                 else
1230                         mem->ranges[i].start = p_end + 1;
1231                 break;
1232         }
1233
1234         /* If a split happened, add the split to array */
1235         if (!temp_range.end)
1236                 return 0;
1237
1238         /* Split happened */
1239         if (i == mem->max_nr_ranges - 1)
1240                 return -ENOMEM;
1241
1242         /* Location where new range should go */
1243         j = i + 1;
1244         if (j < mem->nr_ranges) {
1245                 /* Move over all ranges one slot towards the end */
1246                 for (i = mem->nr_ranges - 1; i >= j; i--)
1247                         mem->ranges[i + 1] = mem->ranges[i];
1248         }
1249
1250         mem->ranges[j].start = temp_range.start;
1251         mem->ranges[j].end = temp_range.end;
1252         mem->nr_ranges++;
1253         return 0;
1254 }
1255
1256 int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
1257                           void **addr, unsigned long *sz)
1258 {
1259         Elf64_Ehdr *ehdr;
1260         Elf64_Phdr *phdr;
1261         unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
1262         unsigned char *buf;
1263         unsigned int cpu, i;
1264         unsigned long long notes_addr;
1265         unsigned long mstart, mend;
1266
1267         /* extra phdr for vmcoreinfo ELF note */
1268         nr_phdr = nr_cpus + 1;
1269         nr_phdr += mem->nr_ranges;
1270
1271         /*
1272          * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
1273          * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
1274          * I think this is required by tools like gdb. So same physical
1275          * memory will be mapped in two ELF headers. One will contain kernel
1276          * text virtual addresses and other will have __va(physical) addresses.
1277          */
1278
1279         nr_phdr++;
1280         elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
1281         elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
1282
1283         buf = vzalloc(elf_sz);
1284         if (!buf)
1285                 return -ENOMEM;
1286
1287         ehdr = (Elf64_Ehdr *)buf;
1288         phdr = (Elf64_Phdr *)(ehdr + 1);
1289         memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
1290         ehdr->e_ident[EI_CLASS] = ELFCLASS64;
1291         ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
1292         ehdr->e_ident[EI_VERSION] = EV_CURRENT;
1293         ehdr->e_ident[EI_OSABI] = ELF_OSABI;
1294         memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
1295         ehdr->e_type = ET_CORE;
1296         ehdr->e_machine = ELF_ARCH;
1297         ehdr->e_version = EV_CURRENT;
1298         ehdr->e_phoff = sizeof(Elf64_Ehdr);
1299         ehdr->e_ehsize = sizeof(Elf64_Ehdr);
1300         ehdr->e_phentsize = sizeof(Elf64_Phdr);
1301
1302         /* Prepare one phdr of type PT_NOTE for each present CPU */
1303         for_each_present_cpu(cpu) {
1304                 phdr->p_type = PT_NOTE;
1305                 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
1306                 phdr->p_offset = phdr->p_paddr = notes_addr;
1307                 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
1308                 (ehdr->e_phnum)++;
1309                 phdr++;
1310         }
1311
1312         /* Prepare one PT_NOTE header for vmcoreinfo */
1313         phdr->p_type = PT_NOTE;
1314         phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
1315         phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
1316         (ehdr->e_phnum)++;
1317         phdr++;
1318
1319         /* Prepare PT_LOAD type program header for kernel text region */
1320         if (kernel_map) {
1321                 phdr->p_type = PT_LOAD;
1322                 phdr->p_flags = PF_R|PF_W|PF_X;
1323                 phdr->p_vaddr = (unsigned long) _text;
1324                 phdr->p_filesz = phdr->p_memsz = _end - _text;
1325                 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
1326                 ehdr->e_phnum++;
1327                 phdr++;
1328         }
1329
1330         /* Go through all the ranges in mem->ranges[] and prepare phdr */
1331         for (i = 0; i < mem->nr_ranges; i++) {
1332                 mstart = mem->ranges[i].start;
1333                 mend = mem->ranges[i].end;
1334
1335                 phdr->p_type = PT_LOAD;
1336                 phdr->p_flags = PF_R|PF_W|PF_X;
1337                 phdr->p_offset  = mstart;
1338
1339                 phdr->p_paddr = mstart;
1340                 phdr->p_vaddr = (unsigned long) __va(mstart);
1341                 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
1342                 phdr->p_align = 0;
1343                 ehdr->e_phnum++;
1344                 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",
1345                         phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
1346                         ehdr->e_phnum, phdr->p_offset);
1347                 phdr++;
1348         }
1349
1350         *addr = buf;
1351         *sz = elf_sz;
1352         return 0;
1353 }