]>
Commit | Line | Data |
---|---|---|
a43cac0d DY |
1 | /* |
2 | * kexec: kexec_file_load system call | |
3 | * | |
4 | * Copyright (C) 2014 Red Hat Inc. | |
5 | * Authors: | |
6 | * Vivek Goyal <vgoyal@redhat.com> | |
7 | * | |
8 | * This source code is licensed under the GNU General Public License, | |
9 | * Version 2. See the file COPYING for more details. | |
10 | */ | |
11 | ||
de90a6bc MH |
12 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
13 | ||
a43cac0d DY |
14 | #include <linux/capability.h> |
15 | #include <linux/mm.h> | |
16 | #include <linux/file.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/kexec.h> | |
19 | #include <linux/mutex.h> | |
20 | #include <linux/list.h> | |
b804defe | 21 | #include <linux/fs.h> |
7b8589cc | 22 | #include <linux/ima.h> |
a43cac0d DY |
23 | #include <crypto/hash.h> |
24 | #include <crypto/sha.h> | |
25 | #include <linux/syscalls.h> | |
26 | #include <linux/vmalloc.h> | |
27 | #include "kexec_internal.h" | |
28 | ||
a43cac0d DY |
29 | static int kexec_calculate_store_digests(struct kimage *image); |
30 | ||
a43cac0d DY |
31 | /* Architectures can provide this probe function */ |
32 | int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf, | |
33 | unsigned long buf_len) | |
34 | { | |
35 | return -ENOEXEC; | |
36 | } | |
37 | ||
38 | void * __weak arch_kexec_kernel_image_load(struct kimage *image) | |
39 | { | |
40 | return ERR_PTR(-ENOEXEC); | |
41 | } | |
42 | ||
43 | int __weak arch_kimage_file_post_load_cleanup(struct kimage *image) | |
44 | { | |
45 | return -EINVAL; | |
46 | } | |
47 | ||
978e30c9 | 48 | #ifdef CONFIG_KEXEC_VERIFY_SIG |
a43cac0d DY |
49 | int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf, |
50 | unsigned long buf_len) | |
51 | { | |
52 | return -EKEYREJECTED; | |
53 | } | |
978e30c9 | 54 | #endif |
a43cac0d DY |
55 | |
56 | /* Apply relocations of type RELA */ | |
57 | int __weak | |
58 | arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs, | |
59 | unsigned int relsec) | |
60 | { | |
61 | pr_err("RELA relocation unsupported.\n"); | |
62 | return -ENOEXEC; | |
63 | } | |
64 | ||
65 | /* Apply relocations of type REL */ | |
66 | int __weak | |
67 | arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs, | |
68 | unsigned int relsec) | |
69 | { | |
70 | pr_err("REL relocation unsupported.\n"); | |
71 | return -ENOEXEC; | |
72 | } | |
73 | ||
74 | /* | |
75 | * Free up memory used by kernel, initrd, and command line. This is temporary | |
76 | * memory allocation which is not needed any more after these buffers have | |
77 | * been loaded into separate segments and have been copied elsewhere. | |
78 | */ | |
79 | void kimage_file_post_load_cleanup(struct kimage *image) | |
80 | { | |
81 | struct purgatory_info *pi = &image->purgatory_info; | |
82 | ||
83 | vfree(image->kernel_buf); | |
84 | image->kernel_buf = NULL; | |
85 | ||
86 | vfree(image->initrd_buf); | |
87 | image->initrd_buf = NULL; | |
88 | ||
89 | kfree(image->cmdline_buf); | |
90 | image->cmdline_buf = NULL; | |
91 | ||
92 | vfree(pi->purgatory_buf); | |
93 | pi->purgatory_buf = NULL; | |
94 | ||
95 | vfree(pi->sechdrs); | |
96 | pi->sechdrs = NULL; | |
97 | ||
98 | /* See if architecture has anything to cleanup post load */ | |
99 | arch_kimage_file_post_load_cleanup(image); | |
100 | ||
101 | /* | |
102 | * Above call should have called into bootloader to free up | |
103 | * any data stored in kimage->image_loader_data. It should | |
104 | * be ok now to free it up. | |
105 | */ | |
106 | kfree(image->image_loader_data); | |
107 | image->image_loader_data = NULL; | |
108 | } | |
109 | ||
110 | /* | |
111 | * In file mode list of segments is prepared by kernel. Copy relevant | |
112 | * data from user space, do error checking, prepare segment list | |
113 | */ | |
114 | static int | |
115 | kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, | |
116 | const char __user *cmdline_ptr, | |
117 | unsigned long cmdline_len, unsigned flags) | |
118 | { | |
119 | int ret = 0; | |
120 | void *ldata; | |
b804defe | 121 | loff_t size; |
a43cac0d | 122 | |
b804defe MZ |
123 | ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf, |
124 | &size, INT_MAX, READING_KEXEC_IMAGE); | |
a43cac0d DY |
125 | if (ret) |
126 | return ret; | |
b804defe | 127 | image->kernel_buf_len = size; |
a43cac0d | 128 | |
7b8589cc MZ |
129 | /* IMA needs to pass the measurement list to the next kernel. */ |
130 | ima_add_kexec_buffer(image); | |
131 | ||
a43cac0d DY |
132 | /* Call arch image probe handlers */ |
133 | ret = arch_kexec_kernel_image_probe(image, image->kernel_buf, | |
134 | image->kernel_buf_len); | |
a43cac0d DY |
135 | if (ret) |
136 | goto out; | |
137 | ||
138 | #ifdef CONFIG_KEXEC_VERIFY_SIG | |
139 | ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf, | |
140 | image->kernel_buf_len); | |
141 | if (ret) { | |
142 | pr_debug("kernel signature verification failed.\n"); | |
143 | goto out; | |
144 | } | |
145 | pr_debug("kernel signature verification successful.\n"); | |
146 | #endif | |
147 | /* It is possible that there no initramfs is being loaded */ | |
148 | if (!(flags & KEXEC_FILE_NO_INITRAMFS)) { | |
b804defe MZ |
149 | ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf, |
150 | &size, INT_MAX, | |
151 | READING_KEXEC_INITRAMFS); | |
a43cac0d DY |
152 | if (ret) |
153 | goto out; | |
b804defe | 154 | image->initrd_buf_len = size; |
a43cac0d DY |
155 | } |
156 | ||
157 | if (cmdline_len) { | |
a9bd8dfa AV |
158 | image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len); |
159 | if (IS_ERR(image->cmdline_buf)) { | |
160 | ret = PTR_ERR(image->cmdline_buf); | |
161 | image->cmdline_buf = NULL; | |
a43cac0d DY |
162 | goto out; |
163 | } | |
164 | ||
165 | image->cmdline_buf_len = cmdline_len; | |
166 | ||
167 | /* command line should be a string with last byte null */ | |
168 | if (image->cmdline_buf[cmdline_len - 1] != '\0') { | |
169 | ret = -EINVAL; | |
170 | goto out; | |
171 | } | |
172 | } | |
173 | ||
174 | /* Call arch image load handlers */ | |
175 | ldata = arch_kexec_kernel_image_load(image); | |
176 | ||
177 | if (IS_ERR(ldata)) { | |
178 | ret = PTR_ERR(ldata); | |
179 | goto out; | |
180 | } | |
181 | ||
182 | image->image_loader_data = ldata; | |
183 | out: | |
184 | /* In case of error, free up all allocated memory in this function */ | |
185 | if (ret) | |
186 | kimage_file_post_load_cleanup(image); | |
187 | return ret; | |
188 | } | |
189 | ||
190 | static int | |
191 | kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, | |
192 | int initrd_fd, const char __user *cmdline_ptr, | |
193 | unsigned long cmdline_len, unsigned long flags) | |
194 | { | |
195 | int ret; | |
196 | struct kimage *image; | |
197 | bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; | |
198 | ||
199 | image = do_kimage_alloc_init(); | |
200 | if (!image) | |
201 | return -ENOMEM; | |
202 | ||
203 | image->file_mode = 1; | |
204 | ||
205 | if (kexec_on_panic) { | |
206 | /* Enable special crash kernel control page alloc policy. */ | |
207 | image->control_page = crashk_res.start; | |
208 | image->type = KEXEC_TYPE_CRASH; | |
209 | } | |
210 | ||
211 | ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, | |
212 | cmdline_ptr, cmdline_len, flags); | |
213 | if (ret) | |
214 | goto out_free_image; | |
215 | ||
216 | ret = sanity_check_segment_list(image); | |
217 | if (ret) | |
218 | goto out_free_post_load_bufs; | |
219 | ||
220 | ret = -ENOMEM; | |
221 | image->control_code_page = kimage_alloc_control_pages(image, | |
222 | get_order(KEXEC_CONTROL_PAGE_SIZE)); | |
223 | if (!image->control_code_page) { | |
224 | pr_err("Could not allocate control_code_buffer\n"); | |
225 | goto out_free_post_load_bufs; | |
226 | } | |
227 | ||
228 | if (!kexec_on_panic) { | |
229 | image->swap_page = kimage_alloc_control_pages(image, 0); | |
230 | if (!image->swap_page) { | |
231 | pr_err("Could not allocate swap buffer\n"); | |
232 | goto out_free_control_pages; | |
233 | } | |
234 | } | |
235 | ||
236 | *rimage = image; | |
237 | return 0; | |
238 | out_free_control_pages: | |
239 | kimage_free_page_list(&image->control_pages); | |
240 | out_free_post_load_bufs: | |
241 | kimage_file_post_load_cleanup(image); | |
242 | out_free_image: | |
243 | kfree(image); | |
244 | return ret; | |
245 | } | |
246 | ||
247 | SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, | |
248 | unsigned long, cmdline_len, const char __user *, cmdline_ptr, | |
249 | unsigned long, flags) | |
250 | { | |
251 | int ret = 0, i; | |
252 | struct kimage **dest_image, *image; | |
253 | ||
254 | /* We only trust the superuser with rebooting the system. */ | |
255 | if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) | |
256 | return -EPERM; | |
257 | ||
258 | /* Make sure we have a legal set of flags */ | |
259 | if (flags != (flags & KEXEC_FILE_FLAGS)) | |
260 | return -EINVAL; | |
261 | ||
262 | image = NULL; | |
263 | ||
264 | if (!mutex_trylock(&kexec_mutex)) | |
265 | return -EBUSY; | |
266 | ||
267 | dest_image = &kexec_image; | |
9b492cf5 | 268 | if (flags & KEXEC_FILE_ON_CRASH) { |
a43cac0d | 269 | dest_image = &kexec_crash_image; |
9b492cf5 XP |
270 | if (kexec_crash_image) |
271 | arch_kexec_unprotect_crashkres(); | |
272 | } | |
a43cac0d DY |
273 | |
274 | if (flags & KEXEC_FILE_UNLOAD) | |
275 | goto exchange; | |
276 | ||
277 | /* | |
278 | * In case of crash, new kernel gets loaded in reserved region. It is | |
279 | * same memory where old crash kernel might be loaded. Free any | |
280 | * current crash dump kernel before we corrupt it. | |
281 | */ | |
282 | if (flags & KEXEC_FILE_ON_CRASH) | |
283 | kimage_free(xchg(&kexec_crash_image, NULL)); | |
284 | ||
285 | ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, | |
286 | cmdline_len, flags); | |
287 | if (ret) | |
288 | goto out; | |
289 | ||
290 | ret = machine_kexec_prepare(image); | |
291 | if (ret) | |
292 | goto out; | |
293 | ||
1229384f XP |
294 | /* |
295 | * Some architecture(like S390) may touch the crash memory before | |
296 | * machine_kexec_prepare(), we must copy vmcoreinfo data after it. | |
297 | */ | |
298 | ret = kimage_crash_copy_vmcoreinfo(image); | |
299 | if (ret) | |
300 | goto out; | |
301 | ||
a43cac0d DY |
302 | ret = kexec_calculate_store_digests(image); |
303 | if (ret) | |
304 | goto out; | |
305 | ||
306 | for (i = 0; i < image->nr_segments; i++) { | |
307 | struct kexec_segment *ksegment; | |
308 | ||
309 | ksegment = &image->segment[i]; | |
310 | pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", | |
311 | i, ksegment->buf, ksegment->bufsz, ksegment->mem, | |
312 | ksegment->memsz); | |
313 | ||
314 | ret = kimage_load_segment(image, &image->segment[i]); | |
315 | if (ret) | |
316 | goto out; | |
317 | } | |
318 | ||
319 | kimage_terminate(image); | |
320 | ||
321 | /* | |
322 | * Free up any temporary buffers allocated which are not needed | |
323 | * after image has been loaded | |
324 | */ | |
325 | kimage_file_post_load_cleanup(image); | |
326 | exchange: | |
327 | image = xchg(dest_image, image); | |
328 | out: | |
9b492cf5 XP |
329 | if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) |
330 | arch_kexec_protect_crashkres(); | |
331 | ||
a43cac0d DY |
332 | mutex_unlock(&kexec_mutex); |
333 | kimage_free(image); | |
334 | return ret; | |
335 | } | |
336 | ||
337 | static int locate_mem_hole_top_down(unsigned long start, unsigned long end, | |
338 | struct kexec_buf *kbuf) | |
339 | { | |
340 | struct kimage *image = kbuf->image; | |
341 | unsigned long temp_start, temp_end; | |
342 | ||
343 | temp_end = min(end, kbuf->buf_max); | |
344 | temp_start = temp_end - kbuf->memsz; | |
345 | ||
346 | do { | |
347 | /* align down start */ | |
348 | temp_start = temp_start & (~(kbuf->buf_align - 1)); | |
349 | ||
350 | if (temp_start < start || temp_start < kbuf->buf_min) | |
351 | return 0; | |
352 | ||
353 | temp_end = temp_start + kbuf->memsz - 1; | |
354 | ||
355 | /* | |
356 | * Make sure this does not conflict with any of existing | |
357 | * segments | |
358 | */ | |
359 | if (kimage_is_destination_range(image, temp_start, temp_end)) { | |
360 | temp_start = temp_start - PAGE_SIZE; | |
361 | continue; | |
362 | } | |
363 | ||
364 | /* We found a suitable memory range */ | |
365 | break; | |
366 | } while (1); | |
367 | ||
368 | /* If we are here, we found a suitable memory range */ | |
369 | kbuf->mem = temp_start; | |
370 | ||
371 | /* Success, stop navigating through remaining System RAM ranges */ | |
372 | return 1; | |
373 | } | |
374 | ||
375 | static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, | |
376 | struct kexec_buf *kbuf) | |
377 | { | |
378 | struct kimage *image = kbuf->image; | |
379 | unsigned long temp_start, temp_end; | |
380 | ||
381 | temp_start = max(start, kbuf->buf_min); | |
382 | ||
383 | do { | |
384 | temp_start = ALIGN(temp_start, kbuf->buf_align); | |
385 | temp_end = temp_start + kbuf->memsz - 1; | |
386 | ||
387 | if (temp_end > end || temp_end > kbuf->buf_max) | |
388 | return 0; | |
389 | /* | |
390 | * Make sure this does not conflict with any of existing | |
391 | * segments | |
392 | */ | |
393 | if (kimage_is_destination_range(image, temp_start, temp_end)) { | |
394 | temp_start = temp_start + PAGE_SIZE; | |
395 | continue; | |
396 | } | |
397 | ||
398 | /* We found a suitable memory range */ | |
399 | break; | |
400 | } while (1); | |
401 | ||
402 | /* If we are here, we found a suitable memory range */ | |
403 | kbuf->mem = temp_start; | |
404 | ||
405 | /* Success, stop navigating through remaining System RAM ranges */ | |
406 | return 1; | |
407 | } | |
408 | ||
1d2e733b | 409 | static int locate_mem_hole_callback(struct resource *res, void *arg) |
a43cac0d DY |
410 | { |
411 | struct kexec_buf *kbuf = (struct kexec_buf *)arg; | |
1d2e733b | 412 | u64 start = res->start, end = res->end; |
a43cac0d DY |
413 | unsigned long sz = end - start + 1; |
414 | ||
415 | /* Returning 0 will take to next memory range */ | |
416 | if (sz < kbuf->memsz) | |
417 | return 0; | |
418 | ||
419 | if (end < kbuf->buf_min || start > kbuf->buf_max) | |
420 | return 0; | |
421 | ||
422 | /* | |
423 | * Allocate memory top down with-in ram range. Otherwise bottom up | |
424 | * allocation. | |
425 | */ | |
426 | if (kbuf->top_down) | |
427 | return locate_mem_hole_top_down(start, end, kbuf); | |
428 | return locate_mem_hole_bottom_up(start, end, kbuf); | |
429 | } | |
430 | ||
60fe3910 TJB |
431 | /** |
432 | * arch_kexec_walk_mem - call func(data) on free memory regions | |
433 | * @kbuf: Context info for the search. Also passed to @func. | |
434 | * @func: Function to call for each memory region. | |
435 | * | |
436 | * Return: The memory walk will stop when func returns a non-zero value | |
437 | * and that value will be returned. If all free regions are visited without | |
438 | * func returning non-zero, then zero will be returned. | |
439 | */ | |
440 | int __weak arch_kexec_walk_mem(struct kexec_buf *kbuf, | |
1d2e733b | 441 | int (*func)(struct resource *, void *)) |
60fe3910 TJB |
442 | { |
443 | if (kbuf->image->type == KEXEC_TYPE_CRASH) | |
444 | return walk_iomem_res_desc(crashk_res.desc, | |
445 | IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, | |
446 | crashk_res.start, crashk_res.end, | |
447 | kbuf, func); | |
448 | else | |
449 | return walk_system_ram_res(0, ULONG_MAX, kbuf, func); | |
450 | } | |
451 | ||
e2e806f9 TJB |
452 | /** |
453 | * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel | |
454 | * @kbuf: Parameters for the memory search. | |
455 | * | |
456 | * On success, kbuf->mem will have the start address of the memory region found. | |
457 | * | |
458 | * Return: 0 on success, negative errno on error. | |
459 | */ | |
460 | int kexec_locate_mem_hole(struct kexec_buf *kbuf) | |
461 | { | |
462 | int ret; | |
463 | ||
464 | ret = arch_kexec_walk_mem(kbuf, locate_mem_hole_callback); | |
465 | ||
466 | return ret == 1 ? 0 : -EADDRNOTAVAIL; | |
467 | } | |
468 | ||
ec2b9bfa TJB |
469 | /** |
470 | * kexec_add_buffer - place a buffer in a kexec segment | |
471 | * @kbuf: Buffer contents and memory parameters. | |
472 | * | |
473 | * This function assumes that kexec_mutex is held. | |
474 | * On successful return, @kbuf->mem will have the physical address of | |
475 | * the buffer in memory. | |
476 | * | |
477 | * Return: 0 on success, negative errno on error. | |
a43cac0d | 478 | */ |
ec2b9bfa | 479 | int kexec_add_buffer(struct kexec_buf *kbuf) |
a43cac0d DY |
480 | { |
481 | ||
482 | struct kexec_segment *ksegment; | |
a43cac0d DY |
483 | int ret; |
484 | ||
485 | /* Currently adding segment this way is allowed only in file mode */ | |
ec2b9bfa | 486 | if (!kbuf->image->file_mode) |
a43cac0d DY |
487 | return -EINVAL; |
488 | ||
ec2b9bfa | 489 | if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) |
a43cac0d DY |
490 | return -EINVAL; |
491 | ||
492 | /* | |
493 | * Make sure we are not trying to add buffer after allocating | |
494 | * control pages. All segments need to be placed first before | |
495 | * any control pages are allocated. As control page allocation | |
496 | * logic goes through list of segments to make sure there are | |
497 | * no destination overlaps. | |
498 | */ | |
ec2b9bfa | 499 | if (!list_empty(&kbuf->image->control_pages)) { |
a43cac0d DY |
500 | WARN_ON(1); |
501 | return -EINVAL; | |
502 | } | |
503 | ||
ec2b9bfa TJB |
504 | /* Ensure minimum alignment needed for segments. */ |
505 | kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); | |
506 | kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); | |
a43cac0d DY |
507 | |
508 | /* Walk the RAM ranges and allocate a suitable range for the buffer */ | |
e2e806f9 TJB |
509 | ret = kexec_locate_mem_hole(kbuf); |
510 | if (ret) | |
511 | return ret; | |
a43cac0d DY |
512 | |
513 | /* Found a suitable memory range */ | |
ec2b9bfa | 514 | ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; |
a43cac0d DY |
515 | ksegment->kbuf = kbuf->buffer; |
516 | ksegment->bufsz = kbuf->bufsz; | |
517 | ksegment->mem = kbuf->mem; | |
518 | ksegment->memsz = kbuf->memsz; | |
ec2b9bfa | 519 | kbuf->image->nr_segments++; |
a43cac0d DY |
520 | return 0; |
521 | } | |
522 | ||
523 | /* Calculate and store the digest of segments */ | |
524 | static int kexec_calculate_store_digests(struct kimage *image) | |
525 | { | |
526 | struct crypto_shash *tfm; | |
527 | struct shash_desc *desc; | |
528 | int ret = 0, i, j, zero_buf_sz, sha_region_sz; | |
529 | size_t desc_size, nullsz; | |
530 | char *digest; | |
531 | void *zero_buf; | |
532 | struct kexec_sha_region *sha_regions; | |
533 | struct purgatory_info *pi = &image->purgatory_info; | |
534 | ||
535 | zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); | |
536 | zero_buf_sz = PAGE_SIZE; | |
537 | ||
538 | tfm = crypto_alloc_shash("sha256", 0, 0); | |
539 | if (IS_ERR(tfm)) { | |
540 | ret = PTR_ERR(tfm); | |
541 | goto out; | |
542 | } | |
543 | ||
544 | desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); | |
545 | desc = kzalloc(desc_size, GFP_KERNEL); | |
546 | if (!desc) { | |
547 | ret = -ENOMEM; | |
548 | goto out_free_tfm; | |
549 | } | |
550 | ||
551 | sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); | |
552 | sha_regions = vzalloc(sha_region_sz); | |
553 | if (!sha_regions) | |
554 | goto out_free_desc; | |
555 | ||
556 | desc->tfm = tfm; | |
557 | desc->flags = 0; | |
558 | ||
559 | ret = crypto_shash_init(desc); | |
560 | if (ret < 0) | |
561 | goto out_free_sha_regions; | |
562 | ||
563 | digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); | |
564 | if (!digest) { | |
565 | ret = -ENOMEM; | |
566 | goto out_free_sha_regions; | |
567 | } | |
568 | ||
569 | for (j = i = 0; i < image->nr_segments; i++) { | |
570 | struct kexec_segment *ksegment; | |
571 | ||
572 | ksegment = &image->segment[i]; | |
573 | /* | |
574 | * Skip purgatory as it will be modified once we put digest | |
575 | * info in purgatory. | |
576 | */ | |
577 | if (ksegment->kbuf == pi->purgatory_buf) | |
578 | continue; | |
579 | ||
580 | ret = crypto_shash_update(desc, ksegment->kbuf, | |
581 | ksegment->bufsz); | |
582 | if (ret) | |
583 | break; | |
584 | ||
585 | /* | |
586 | * Assume rest of the buffer is filled with zero and | |
587 | * update digest accordingly. | |
588 | */ | |
589 | nullsz = ksegment->memsz - ksegment->bufsz; | |
590 | while (nullsz) { | |
591 | unsigned long bytes = nullsz; | |
592 | ||
593 | if (bytes > zero_buf_sz) | |
594 | bytes = zero_buf_sz; | |
595 | ret = crypto_shash_update(desc, zero_buf, bytes); | |
596 | if (ret) | |
597 | break; | |
598 | nullsz -= bytes; | |
599 | } | |
600 | ||
601 | if (ret) | |
602 | break; | |
603 | ||
604 | sha_regions[j].start = ksegment->mem; | |
605 | sha_regions[j].len = ksegment->memsz; | |
606 | j++; | |
607 | } | |
608 | ||
609 | if (!ret) { | |
610 | ret = crypto_shash_final(desc, digest); | |
611 | if (ret) | |
612 | goto out_free_digest; | |
40c50c1f TG |
613 | ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions", |
614 | sha_regions, sha_region_sz, 0); | |
a43cac0d DY |
615 | if (ret) |
616 | goto out_free_digest; | |
617 | ||
40c50c1f TG |
618 | ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest", |
619 | digest, SHA256_DIGEST_SIZE, 0); | |
a43cac0d DY |
620 | if (ret) |
621 | goto out_free_digest; | |
622 | } | |
623 | ||
624 | out_free_digest: | |
625 | kfree(digest); | |
626 | out_free_sha_regions: | |
627 | vfree(sha_regions); | |
628 | out_free_desc: | |
629 | kfree(desc); | |
630 | out_free_tfm: | |
631 | kfree(tfm); | |
632 | out: | |
633 | return ret; | |
634 | } | |
635 | ||
636 | /* Actually load purgatory. Lot of code taken from kexec-tools */ | |
637 | static int __kexec_load_purgatory(struct kimage *image, unsigned long min, | |
638 | unsigned long max, int top_down) | |
639 | { | |
640 | struct purgatory_info *pi = &image->purgatory_info; | |
ec2b9bfa TJB |
641 | unsigned long align, bss_align, bss_sz, bss_pad; |
642 | unsigned long entry, load_addr, curr_load_addr, bss_addr, offset; | |
a43cac0d DY |
643 | unsigned char *buf_addr, *src; |
644 | int i, ret = 0, entry_sidx = -1; | |
645 | const Elf_Shdr *sechdrs_c; | |
646 | Elf_Shdr *sechdrs = NULL; | |
ec2b9bfa TJB |
647 | struct kexec_buf kbuf = { .image = image, .bufsz = 0, .buf_align = 1, |
648 | .buf_min = min, .buf_max = max, | |
649 | .top_down = top_down }; | |
a43cac0d DY |
650 | |
651 | /* | |
652 | * sechdrs_c points to section headers in purgatory and are read | |
653 | * only. No modifications allowed. | |
654 | */ | |
655 | sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff; | |
656 | ||
657 | /* | |
658 | * We can not modify sechdrs_c[] and its fields. It is read only. | |
659 | * Copy it over to a local copy where one can store some temporary | |
660 | * data and free it at the end. We need to modify ->sh_addr and | |
661 | * ->sh_offset fields to keep track of permanent and temporary | |
662 | * locations of sections. | |
663 | */ | |
664 | sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr)); | |
665 | if (!sechdrs) | |
666 | return -ENOMEM; | |
667 | ||
668 | memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr)); | |
669 | ||
670 | /* | |
671 | * We seem to have multiple copies of sections. First copy is which | |
672 | * is embedded in kernel in read only section. Some of these sections | |
673 | * will be copied to a temporary buffer and relocated. And these | |
674 | * sections will finally be copied to their final destination at | |
675 | * segment load time. | |
676 | * | |
677 | * Use ->sh_offset to reflect section address in memory. It will | |
678 | * point to original read only copy if section is not allocatable. | |
679 | * Otherwise it will point to temporary copy which will be relocated. | |
680 | * | |
681 | * Use ->sh_addr to contain final address of the section where it | |
682 | * will go during execution time. | |
683 | */ | |
684 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
685 | if (sechdrs[i].sh_type == SHT_NOBITS) | |
686 | continue; | |
687 | ||
688 | sechdrs[i].sh_offset = (unsigned long)pi->ehdr + | |
689 | sechdrs[i].sh_offset; | |
690 | } | |
691 | ||
692 | /* | |
693 | * Identify entry point section and make entry relative to section | |
694 | * start. | |
695 | */ | |
696 | entry = pi->ehdr->e_entry; | |
697 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
698 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
699 | continue; | |
700 | ||
701 | if (!(sechdrs[i].sh_flags & SHF_EXECINSTR)) | |
702 | continue; | |
703 | ||
704 | /* Make entry section relative */ | |
705 | if (sechdrs[i].sh_addr <= pi->ehdr->e_entry && | |
706 | ((sechdrs[i].sh_addr + sechdrs[i].sh_size) > | |
707 | pi->ehdr->e_entry)) { | |
708 | entry_sidx = i; | |
709 | entry -= sechdrs[i].sh_addr; | |
710 | break; | |
711 | } | |
712 | } | |
713 | ||
714 | /* Determine how much memory is needed to load relocatable object. */ | |
a43cac0d | 715 | bss_align = 1; |
a43cac0d DY |
716 | bss_sz = 0; |
717 | ||
718 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
719 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
720 | continue; | |
721 | ||
722 | align = sechdrs[i].sh_addralign; | |
723 | if (sechdrs[i].sh_type != SHT_NOBITS) { | |
ec2b9bfa TJB |
724 | if (kbuf.buf_align < align) |
725 | kbuf.buf_align = align; | |
726 | kbuf.bufsz = ALIGN(kbuf.bufsz, align); | |
727 | kbuf.bufsz += sechdrs[i].sh_size; | |
a43cac0d DY |
728 | } else { |
729 | /* bss section */ | |
730 | if (bss_align < align) | |
731 | bss_align = align; | |
732 | bss_sz = ALIGN(bss_sz, align); | |
733 | bss_sz += sechdrs[i].sh_size; | |
734 | } | |
735 | } | |
736 | ||
737 | /* Determine the bss padding required to align bss properly */ | |
738 | bss_pad = 0; | |
ec2b9bfa TJB |
739 | if (kbuf.bufsz & (bss_align - 1)) |
740 | bss_pad = bss_align - (kbuf.bufsz & (bss_align - 1)); | |
a43cac0d | 741 | |
ec2b9bfa | 742 | kbuf.memsz = kbuf.bufsz + bss_pad + bss_sz; |
a43cac0d DY |
743 | |
744 | /* Allocate buffer for purgatory */ | |
ec2b9bfa TJB |
745 | kbuf.buffer = vzalloc(kbuf.bufsz); |
746 | if (!kbuf.buffer) { | |
a43cac0d DY |
747 | ret = -ENOMEM; |
748 | goto out; | |
749 | } | |
750 | ||
ec2b9bfa TJB |
751 | if (kbuf.buf_align < bss_align) |
752 | kbuf.buf_align = bss_align; | |
a43cac0d DY |
753 | |
754 | /* Add buffer to segment list */ | |
ec2b9bfa | 755 | ret = kexec_add_buffer(&kbuf); |
a43cac0d DY |
756 | if (ret) |
757 | goto out; | |
ec2b9bfa | 758 | pi->purgatory_load_addr = kbuf.mem; |
a43cac0d DY |
759 | |
760 | /* Load SHF_ALLOC sections */ | |
ec2b9bfa | 761 | buf_addr = kbuf.buffer; |
a43cac0d | 762 | load_addr = curr_load_addr = pi->purgatory_load_addr; |
ec2b9bfa | 763 | bss_addr = load_addr + kbuf.bufsz + bss_pad; |
a43cac0d DY |
764 | |
765 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
766 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
767 | continue; | |
768 | ||
769 | align = sechdrs[i].sh_addralign; | |
770 | if (sechdrs[i].sh_type != SHT_NOBITS) { | |
771 | curr_load_addr = ALIGN(curr_load_addr, align); | |
772 | offset = curr_load_addr - load_addr; | |
773 | /* We already modifed ->sh_offset to keep src addr */ | |
774 | src = (char *) sechdrs[i].sh_offset; | |
775 | memcpy(buf_addr + offset, src, sechdrs[i].sh_size); | |
776 | ||
777 | /* Store load address and source address of section */ | |
778 | sechdrs[i].sh_addr = curr_load_addr; | |
779 | ||
780 | /* | |
781 | * This section got copied to temporary buffer. Update | |
782 | * ->sh_offset accordingly. | |
783 | */ | |
784 | sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset); | |
785 | ||
786 | /* Advance to the next address */ | |
787 | curr_load_addr += sechdrs[i].sh_size; | |
788 | } else { | |
789 | bss_addr = ALIGN(bss_addr, align); | |
790 | sechdrs[i].sh_addr = bss_addr; | |
791 | bss_addr += sechdrs[i].sh_size; | |
792 | } | |
793 | } | |
794 | ||
795 | /* Update entry point based on load address of text section */ | |
796 | if (entry_sidx >= 0) | |
797 | entry += sechdrs[entry_sidx].sh_addr; | |
798 | ||
799 | /* Make kernel jump to purgatory after shutdown */ | |
800 | image->start = entry; | |
801 | ||
802 | /* Used later to get/set symbol values */ | |
803 | pi->sechdrs = sechdrs; | |
804 | ||
805 | /* | |
806 | * Used later to identify which section is purgatory and skip it | |
807 | * from checksumming. | |
808 | */ | |
ec2b9bfa | 809 | pi->purgatory_buf = kbuf.buffer; |
a43cac0d DY |
810 | return ret; |
811 | out: | |
812 | vfree(sechdrs); | |
ec2b9bfa | 813 | vfree(kbuf.buffer); |
a43cac0d DY |
814 | return ret; |
815 | } | |
816 | ||
817 | static int kexec_apply_relocations(struct kimage *image) | |
818 | { | |
819 | int i, ret; | |
820 | struct purgatory_info *pi = &image->purgatory_info; | |
821 | Elf_Shdr *sechdrs = pi->sechdrs; | |
822 | ||
823 | /* Apply relocations */ | |
824 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
825 | Elf_Shdr *section, *symtab; | |
826 | ||
827 | if (sechdrs[i].sh_type != SHT_RELA && | |
828 | sechdrs[i].sh_type != SHT_REL) | |
829 | continue; | |
830 | ||
831 | /* | |
832 | * For section of type SHT_RELA/SHT_REL, | |
833 | * ->sh_link contains section header index of associated | |
834 | * symbol table. And ->sh_info contains section header | |
835 | * index of section to which relocations apply. | |
836 | */ | |
837 | if (sechdrs[i].sh_info >= pi->ehdr->e_shnum || | |
838 | sechdrs[i].sh_link >= pi->ehdr->e_shnum) | |
839 | return -ENOEXEC; | |
840 | ||
841 | section = &sechdrs[sechdrs[i].sh_info]; | |
842 | symtab = &sechdrs[sechdrs[i].sh_link]; | |
843 | ||
844 | if (!(section->sh_flags & SHF_ALLOC)) | |
845 | continue; | |
846 | ||
847 | /* | |
848 | * symtab->sh_link contain section header index of associated | |
849 | * string table. | |
850 | */ | |
851 | if (symtab->sh_link >= pi->ehdr->e_shnum) | |
852 | /* Invalid section number? */ | |
853 | continue; | |
854 | ||
855 | /* | |
856 | * Respective architecture needs to provide support for applying | |
857 | * relocations of type SHT_RELA/SHT_REL. | |
858 | */ | |
859 | if (sechdrs[i].sh_type == SHT_RELA) | |
860 | ret = arch_kexec_apply_relocations_add(pi->ehdr, | |
861 | sechdrs, i); | |
862 | else if (sechdrs[i].sh_type == SHT_REL) | |
863 | ret = arch_kexec_apply_relocations(pi->ehdr, | |
864 | sechdrs, i); | |
865 | if (ret) | |
866 | return ret; | |
867 | } | |
868 | ||
869 | return 0; | |
870 | } | |
871 | ||
872 | /* Load relocatable purgatory object and relocate it appropriately */ | |
873 | int kexec_load_purgatory(struct kimage *image, unsigned long min, | |
874 | unsigned long max, int top_down, | |
875 | unsigned long *load_addr) | |
876 | { | |
877 | struct purgatory_info *pi = &image->purgatory_info; | |
878 | int ret; | |
879 | ||
880 | if (kexec_purgatory_size <= 0) | |
881 | return -EINVAL; | |
882 | ||
883 | if (kexec_purgatory_size < sizeof(Elf_Ehdr)) | |
884 | return -ENOEXEC; | |
885 | ||
886 | pi->ehdr = (Elf_Ehdr *)kexec_purgatory; | |
887 | ||
888 | if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0 | |
889 | || pi->ehdr->e_type != ET_REL | |
890 | || !elf_check_arch(pi->ehdr) | |
891 | || pi->ehdr->e_shentsize != sizeof(Elf_Shdr)) | |
892 | return -ENOEXEC; | |
893 | ||
894 | if (pi->ehdr->e_shoff >= kexec_purgatory_size | |
895 | || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) > | |
896 | kexec_purgatory_size - pi->ehdr->e_shoff)) | |
897 | return -ENOEXEC; | |
898 | ||
899 | ret = __kexec_load_purgatory(image, min, max, top_down); | |
900 | if (ret) | |
901 | return ret; | |
902 | ||
903 | ret = kexec_apply_relocations(image); | |
904 | if (ret) | |
905 | goto out; | |
906 | ||
907 | *load_addr = pi->purgatory_load_addr; | |
908 | return 0; | |
909 | out: | |
910 | vfree(pi->sechdrs); | |
070c43ee TJB |
911 | pi->sechdrs = NULL; |
912 | ||
a43cac0d | 913 | vfree(pi->purgatory_buf); |
070c43ee | 914 | pi->purgatory_buf = NULL; |
a43cac0d DY |
915 | return ret; |
916 | } | |
917 | ||
918 | static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, | |
919 | const char *name) | |
920 | { | |
921 | Elf_Sym *syms; | |
922 | Elf_Shdr *sechdrs; | |
923 | Elf_Ehdr *ehdr; | |
924 | int i, k; | |
925 | const char *strtab; | |
926 | ||
927 | if (!pi->sechdrs || !pi->ehdr) | |
928 | return NULL; | |
929 | ||
930 | sechdrs = pi->sechdrs; | |
931 | ehdr = pi->ehdr; | |
932 | ||
933 | for (i = 0; i < ehdr->e_shnum; i++) { | |
934 | if (sechdrs[i].sh_type != SHT_SYMTAB) | |
935 | continue; | |
936 | ||
937 | if (sechdrs[i].sh_link >= ehdr->e_shnum) | |
938 | /* Invalid strtab section number */ | |
939 | continue; | |
940 | strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset; | |
941 | syms = (Elf_Sym *)sechdrs[i].sh_offset; | |
942 | ||
943 | /* Go through symbols for a match */ | |
944 | for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { | |
945 | if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) | |
946 | continue; | |
947 | ||
948 | if (strcmp(strtab + syms[k].st_name, name) != 0) | |
949 | continue; | |
950 | ||
951 | if (syms[k].st_shndx == SHN_UNDEF || | |
952 | syms[k].st_shndx >= ehdr->e_shnum) { | |
953 | pr_debug("Symbol: %s has bad section index %d.\n", | |
954 | name, syms[k].st_shndx); | |
955 | return NULL; | |
956 | } | |
957 | ||
958 | /* Found the symbol we are looking for */ | |
959 | return &syms[k]; | |
960 | } | |
961 | } | |
962 | ||
963 | return NULL; | |
964 | } | |
965 | ||
966 | void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) | |
967 | { | |
968 | struct purgatory_info *pi = &image->purgatory_info; | |
969 | Elf_Sym *sym; | |
970 | Elf_Shdr *sechdr; | |
971 | ||
972 | sym = kexec_purgatory_find_symbol(pi, name); | |
973 | if (!sym) | |
974 | return ERR_PTR(-EINVAL); | |
975 | ||
976 | sechdr = &pi->sechdrs[sym->st_shndx]; | |
977 | ||
978 | /* | |
979 | * Returns the address where symbol will finally be loaded after | |
980 | * kexec_load_segment() | |
981 | */ | |
982 | return (void *)(sechdr->sh_addr + sym->st_value); | |
983 | } | |
984 | ||
985 | /* | |
986 | * Get or set value of a symbol. If "get_value" is true, symbol value is | |
987 | * returned in buf otherwise symbol value is set based on value in buf. | |
988 | */ | |
989 | int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, | |
990 | void *buf, unsigned int size, bool get_value) | |
991 | { | |
992 | Elf_Sym *sym; | |
993 | Elf_Shdr *sechdrs; | |
994 | struct purgatory_info *pi = &image->purgatory_info; | |
995 | char *sym_buf; | |
996 | ||
997 | sym = kexec_purgatory_find_symbol(pi, name); | |
998 | if (!sym) | |
999 | return -EINVAL; | |
1000 | ||
1001 | if (sym->st_size != size) { | |
1002 | pr_err("symbol %s size mismatch: expected %lu actual %u\n", | |
1003 | name, (unsigned long)sym->st_size, size); | |
1004 | return -EINVAL; | |
1005 | } | |
1006 | ||
1007 | sechdrs = pi->sechdrs; | |
1008 | ||
1009 | if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) { | |
1010 | pr_err("symbol %s is in a bss section. Cannot %s\n", name, | |
1011 | get_value ? "get" : "set"); | |
1012 | return -EINVAL; | |
1013 | } | |
1014 | ||
1015 | sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset + | |
1016 | sym->st_value; | |
1017 | ||
1018 | if (get_value) | |
1019 | memcpy((void *)buf, sym_buf, size); | |
1020 | else | |
1021 | memcpy((void *)sym_buf, buf, size); | |
1022 | ||
1023 | return 0; | |
1024 | } |