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> |
a43cac0d DY |
22 | #include <crypto/hash.h> |
23 | #include <crypto/sha.h> | |
24 | #include <linux/syscalls.h> | |
25 | #include <linux/vmalloc.h> | |
26 | #include "kexec_internal.h" | |
27 | ||
28 | /* | |
29 | * Declare these symbols weak so that if architecture provides a purgatory, | |
30 | * these will be overridden. | |
31 | */ | |
32 | char __weak kexec_purgatory[0]; | |
33 | size_t __weak kexec_purgatory_size = 0; | |
34 | ||
35 | static int kexec_calculate_store_digests(struct kimage *image); | |
36 | ||
a43cac0d DY |
37 | /* Architectures can provide this probe function */ |
38 | int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf, | |
39 | unsigned long buf_len) | |
40 | { | |
41 | return -ENOEXEC; | |
42 | } | |
43 | ||
44 | void * __weak arch_kexec_kernel_image_load(struct kimage *image) | |
45 | { | |
46 | return ERR_PTR(-ENOEXEC); | |
47 | } | |
48 | ||
49 | int __weak arch_kimage_file_post_load_cleanup(struct kimage *image) | |
50 | { | |
51 | return -EINVAL; | |
52 | } | |
53 | ||
978e30c9 | 54 | #ifdef CONFIG_KEXEC_VERIFY_SIG |
a43cac0d DY |
55 | int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf, |
56 | unsigned long buf_len) | |
57 | { | |
58 | return -EKEYREJECTED; | |
59 | } | |
978e30c9 | 60 | #endif |
a43cac0d DY |
61 | |
62 | /* Apply relocations of type RELA */ | |
63 | int __weak | |
64 | arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs, | |
65 | unsigned int relsec) | |
66 | { | |
67 | pr_err("RELA relocation unsupported.\n"); | |
68 | return -ENOEXEC; | |
69 | } | |
70 | ||
71 | /* Apply relocations of type REL */ | |
72 | int __weak | |
73 | arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs, | |
74 | unsigned int relsec) | |
75 | { | |
76 | pr_err("REL relocation unsupported.\n"); | |
77 | return -ENOEXEC; | |
78 | } | |
79 | ||
80 | /* | |
81 | * Free up memory used by kernel, initrd, and command line. This is temporary | |
82 | * memory allocation which is not needed any more after these buffers have | |
83 | * been loaded into separate segments and have been copied elsewhere. | |
84 | */ | |
85 | void kimage_file_post_load_cleanup(struct kimage *image) | |
86 | { | |
87 | struct purgatory_info *pi = &image->purgatory_info; | |
88 | ||
89 | vfree(image->kernel_buf); | |
90 | image->kernel_buf = NULL; | |
91 | ||
92 | vfree(image->initrd_buf); | |
93 | image->initrd_buf = NULL; | |
94 | ||
95 | kfree(image->cmdline_buf); | |
96 | image->cmdline_buf = NULL; | |
97 | ||
98 | vfree(pi->purgatory_buf); | |
99 | pi->purgatory_buf = NULL; | |
100 | ||
101 | vfree(pi->sechdrs); | |
102 | pi->sechdrs = NULL; | |
103 | ||
104 | /* See if architecture has anything to cleanup post load */ | |
105 | arch_kimage_file_post_load_cleanup(image); | |
106 | ||
107 | /* | |
108 | * Above call should have called into bootloader to free up | |
109 | * any data stored in kimage->image_loader_data. It should | |
110 | * be ok now to free it up. | |
111 | */ | |
112 | kfree(image->image_loader_data); | |
113 | image->image_loader_data = NULL; | |
114 | } | |
115 | ||
116 | /* | |
117 | * In file mode list of segments is prepared by kernel. Copy relevant | |
118 | * data from user space, do error checking, prepare segment list | |
119 | */ | |
120 | static int | |
121 | kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, | |
122 | const char __user *cmdline_ptr, | |
123 | unsigned long cmdline_len, unsigned flags) | |
124 | { | |
125 | int ret = 0; | |
126 | void *ldata; | |
b804defe | 127 | loff_t size; |
a43cac0d | 128 | |
b804defe MZ |
129 | ret = kernel_read_file_from_fd(kernel_fd, &image->kernel_buf, |
130 | &size, INT_MAX, READING_KEXEC_IMAGE); | |
a43cac0d DY |
131 | if (ret) |
132 | return ret; | |
b804defe | 133 | image->kernel_buf_len = size; |
a43cac0d DY |
134 | |
135 | /* Call arch image probe handlers */ | |
136 | ret = arch_kexec_kernel_image_probe(image, image->kernel_buf, | |
137 | image->kernel_buf_len); | |
a43cac0d DY |
138 | if (ret) |
139 | goto out; | |
140 | ||
141 | #ifdef CONFIG_KEXEC_VERIFY_SIG | |
142 | ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf, | |
143 | image->kernel_buf_len); | |
144 | if (ret) { | |
145 | pr_debug("kernel signature verification failed.\n"); | |
146 | goto out; | |
147 | } | |
148 | pr_debug("kernel signature verification successful.\n"); | |
149 | #endif | |
150 | /* It is possible that there no initramfs is being loaded */ | |
151 | if (!(flags & KEXEC_FILE_NO_INITRAMFS)) { | |
b804defe MZ |
152 | ret = kernel_read_file_from_fd(initrd_fd, &image->initrd_buf, |
153 | &size, INT_MAX, | |
154 | READING_KEXEC_INITRAMFS); | |
a43cac0d DY |
155 | if (ret) |
156 | goto out; | |
b804defe | 157 | image->initrd_buf_len = size; |
a43cac0d DY |
158 | } |
159 | ||
160 | if (cmdline_len) { | |
161 | image->cmdline_buf = kzalloc(cmdline_len, GFP_KERNEL); | |
162 | if (!image->cmdline_buf) { | |
163 | ret = -ENOMEM; | |
164 | goto out; | |
165 | } | |
166 | ||
167 | ret = copy_from_user(image->cmdline_buf, cmdline_ptr, | |
168 | cmdline_len); | |
169 | if (ret) { | |
170 | ret = -EFAULT; | |
171 | goto out; | |
172 | } | |
173 | ||
174 | image->cmdline_buf_len = cmdline_len; | |
175 | ||
176 | /* command line should be a string with last byte null */ | |
177 | if (image->cmdline_buf[cmdline_len - 1] != '\0') { | |
178 | ret = -EINVAL; | |
179 | goto out; | |
180 | } | |
181 | } | |
182 | ||
183 | /* Call arch image load handlers */ | |
184 | ldata = arch_kexec_kernel_image_load(image); | |
185 | ||
186 | if (IS_ERR(ldata)) { | |
187 | ret = PTR_ERR(ldata); | |
188 | goto out; | |
189 | } | |
190 | ||
191 | image->image_loader_data = ldata; | |
192 | out: | |
193 | /* In case of error, free up all allocated memory in this function */ | |
194 | if (ret) | |
195 | kimage_file_post_load_cleanup(image); | |
196 | return ret; | |
197 | } | |
198 | ||
199 | static int | |
200 | kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, | |
201 | int initrd_fd, const char __user *cmdline_ptr, | |
202 | unsigned long cmdline_len, unsigned long flags) | |
203 | { | |
204 | int ret; | |
205 | struct kimage *image; | |
206 | bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; | |
207 | ||
208 | image = do_kimage_alloc_init(); | |
209 | if (!image) | |
210 | return -ENOMEM; | |
211 | ||
212 | image->file_mode = 1; | |
213 | ||
214 | if (kexec_on_panic) { | |
215 | /* Enable special crash kernel control page alloc policy. */ | |
216 | image->control_page = crashk_res.start; | |
217 | image->type = KEXEC_TYPE_CRASH; | |
218 | } | |
219 | ||
220 | ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, | |
221 | cmdline_ptr, cmdline_len, flags); | |
222 | if (ret) | |
223 | goto out_free_image; | |
224 | ||
225 | ret = sanity_check_segment_list(image); | |
226 | if (ret) | |
227 | goto out_free_post_load_bufs; | |
228 | ||
229 | ret = -ENOMEM; | |
230 | image->control_code_page = kimage_alloc_control_pages(image, | |
231 | get_order(KEXEC_CONTROL_PAGE_SIZE)); | |
232 | if (!image->control_code_page) { | |
233 | pr_err("Could not allocate control_code_buffer\n"); | |
234 | goto out_free_post_load_bufs; | |
235 | } | |
236 | ||
237 | if (!kexec_on_panic) { | |
238 | image->swap_page = kimage_alloc_control_pages(image, 0); | |
239 | if (!image->swap_page) { | |
240 | pr_err("Could not allocate swap buffer\n"); | |
241 | goto out_free_control_pages; | |
242 | } | |
243 | } | |
244 | ||
245 | *rimage = image; | |
246 | return 0; | |
247 | out_free_control_pages: | |
248 | kimage_free_page_list(&image->control_pages); | |
249 | out_free_post_load_bufs: | |
250 | kimage_file_post_load_cleanup(image); | |
251 | out_free_image: | |
252 | kfree(image); | |
253 | return ret; | |
254 | } | |
255 | ||
256 | SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, | |
257 | unsigned long, cmdline_len, const char __user *, cmdline_ptr, | |
258 | unsigned long, flags) | |
259 | { | |
260 | int ret = 0, i; | |
261 | struct kimage **dest_image, *image; | |
262 | ||
263 | /* We only trust the superuser with rebooting the system. */ | |
264 | if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) | |
265 | return -EPERM; | |
266 | ||
267 | /* Make sure we have a legal set of flags */ | |
268 | if (flags != (flags & KEXEC_FILE_FLAGS)) | |
269 | return -EINVAL; | |
270 | ||
271 | image = NULL; | |
272 | ||
273 | if (!mutex_trylock(&kexec_mutex)) | |
274 | return -EBUSY; | |
275 | ||
276 | dest_image = &kexec_image; | |
277 | if (flags & KEXEC_FILE_ON_CRASH) | |
278 | dest_image = &kexec_crash_image; | |
279 | ||
280 | if (flags & KEXEC_FILE_UNLOAD) | |
281 | goto exchange; | |
282 | ||
283 | /* | |
284 | * In case of crash, new kernel gets loaded in reserved region. It is | |
285 | * same memory where old crash kernel might be loaded. Free any | |
286 | * current crash dump kernel before we corrupt it. | |
287 | */ | |
288 | if (flags & KEXEC_FILE_ON_CRASH) | |
289 | kimage_free(xchg(&kexec_crash_image, NULL)); | |
290 | ||
291 | ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, | |
292 | cmdline_len, flags); | |
293 | if (ret) | |
294 | goto out; | |
295 | ||
296 | ret = machine_kexec_prepare(image); | |
297 | if (ret) | |
298 | goto out; | |
299 | ||
300 | ret = kexec_calculate_store_digests(image); | |
301 | if (ret) | |
302 | goto out; | |
303 | ||
304 | for (i = 0; i < image->nr_segments; i++) { | |
305 | struct kexec_segment *ksegment; | |
306 | ||
307 | ksegment = &image->segment[i]; | |
308 | pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", | |
309 | i, ksegment->buf, ksegment->bufsz, ksegment->mem, | |
310 | ksegment->memsz); | |
311 | ||
312 | ret = kimage_load_segment(image, &image->segment[i]); | |
313 | if (ret) | |
314 | goto out; | |
315 | } | |
316 | ||
317 | kimage_terminate(image); | |
318 | ||
319 | /* | |
320 | * Free up any temporary buffers allocated which are not needed | |
321 | * after image has been loaded | |
322 | */ | |
323 | kimage_file_post_load_cleanup(image); | |
324 | exchange: | |
325 | image = xchg(dest_image, image); | |
326 | out: | |
327 | mutex_unlock(&kexec_mutex); | |
328 | kimage_free(image); | |
329 | return ret; | |
330 | } | |
331 | ||
332 | static int locate_mem_hole_top_down(unsigned long start, unsigned long end, | |
333 | struct kexec_buf *kbuf) | |
334 | { | |
335 | struct kimage *image = kbuf->image; | |
336 | unsigned long temp_start, temp_end; | |
337 | ||
338 | temp_end = min(end, kbuf->buf_max); | |
339 | temp_start = temp_end - kbuf->memsz; | |
340 | ||
341 | do { | |
342 | /* align down start */ | |
343 | temp_start = temp_start & (~(kbuf->buf_align - 1)); | |
344 | ||
345 | if (temp_start < start || temp_start < kbuf->buf_min) | |
346 | return 0; | |
347 | ||
348 | temp_end = temp_start + kbuf->memsz - 1; | |
349 | ||
350 | /* | |
351 | * Make sure this does not conflict with any of existing | |
352 | * segments | |
353 | */ | |
354 | if (kimage_is_destination_range(image, temp_start, temp_end)) { | |
355 | temp_start = temp_start - PAGE_SIZE; | |
356 | continue; | |
357 | } | |
358 | ||
359 | /* We found a suitable memory range */ | |
360 | break; | |
361 | } while (1); | |
362 | ||
363 | /* If we are here, we found a suitable memory range */ | |
364 | kbuf->mem = temp_start; | |
365 | ||
366 | /* Success, stop navigating through remaining System RAM ranges */ | |
367 | return 1; | |
368 | } | |
369 | ||
370 | static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, | |
371 | struct kexec_buf *kbuf) | |
372 | { | |
373 | struct kimage *image = kbuf->image; | |
374 | unsigned long temp_start, temp_end; | |
375 | ||
376 | temp_start = max(start, kbuf->buf_min); | |
377 | ||
378 | do { | |
379 | temp_start = ALIGN(temp_start, kbuf->buf_align); | |
380 | temp_end = temp_start + kbuf->memsz - 1; | |
381 | ||
382 | if (temp_end > end || temp_end > kbuf->buf_max) | |
383 | return 0; | |
384 | /* | |
385 | * Make sure this does not conflict with any of existing | |
386 | * segments | |
387 | */ | |
388 | if (kimage_is_destination_range(image, temp_start, temp_end)) { | |
389 | temp_start = temp_start + PAGE_SIZE; | |
390 | continue; | |
391 | } | |
392 | ||
393 | /* We found a suitable memory range */ | |
394 | break; | |
395 | } while (1); | |
396 | ||
397 | /* If we are here, we found a suitable memory range */ | |
398 | kbuf->mem = temp_start; | |
399 | ||
400 | /* Success, stop navigating through remaining System RAM ranges */ | |
401 | return 1; | |
402 | } | |
403 | ||
404 | static int locate_mem_hole_callback(u64 start, u64 end, void *arg) | |
405 | { | |
406 | struct kexec_buf *kbuf = (struct kexec_buf *)arg; | |
407 | unsigned long sz = end - start + 1; | |
408 | ||
409 | /* Returning 0 will take to next memory range */ | |
410 | if (sz < kbuf->memsz) | |
411 | return 0; | |
412 | ||
413 | if (end < kbuf->buf_min || start > kbuf->buf_max) | |
414 | return 0; | |
415 | ||
416 | /* | |
417 | * Allocate memory top down with-in ram range. Otherwise bottom up | |
418 | * allocation. | |
419 | */ | |
420 | if (kbuf->top_down) | |
421 | return locate_mem_hole_top_down(start, end, kbuf); | |
422 | return locate_mem_hole_bottom_up(start, end, kbuf); | |
423 | } | |
424 | ||
425 | /* | |
426 | * Helper function for placing a buffer in a kexec segment. This assumes | |
427 | * that kexec_mutex is held. | |
428 | */ | |
429 | int kexec_add_buffer(struct kimage *image, char *buffer, unsigned long bufsz, | |
430 | unsigned long memsz, unsigned long buf_align, | |
431 | unsigned long buf_min, unsigned long buf_max, | |
432 | bool top_down, unsigned long *load_addr) | |
433 | { | |
434 | ||
435 | struct kexec_segment *ksegment; | |
436 | struct kexec_buf buf, *kbuf; | |
437 | int ret; | |
438 | ||
439 | /* Currently adding segment this way is allowed only in file mode */ | |
440 | if (!image->file_mode) | |
441 | return -EINVAL; | |
442 | ||
443 | if (image->nr_segments >= KEXEC_SEGMENT_MAX) | |
444 | return -EINVAL; | |
445 | ||
446 | /* | |
447 | * Make sure we are not trying to add buffer after allocating | |
448 | * control pages. All segments need to be placed first before | |
449 | * any control pages are allocated. As control page allocation | |
450 | * logic goes through list of segments to make sure there are | |
451 | * no destination overlaps. | |
452 | */ | |
453 | if (!list_empty(&image->control_pages)) { | |
454 | WARN_ON(1); | |
455 | return -EINVAL; | |
456 | } | |
457 | ||
458 | memset(&buf, 0, sizeof(struct kexec_buf)); | |
459 | kbuf = &buf; | |
460 | kbuf->image = image; | |
461 | kbuf->buffer = buffer; | |
462 | kbuf->bufsz = bufsz; | |
463 | ||
464 | kbuf->memsz = ALIGN(memsz, PAGE_SIZE); | |
465 | kbuf->buf_align = max(buf_align, PAGE_SIZE); | |
466 | kbuf->buf_min = buf_min; | |
467 | kbuf->buf_max = buf_max; | |
468 | kbuf->top_down = top_down; | |
469 | ||
470 | /* Walk the RAM ranges and allocate a suitable range for the buffer */ | |
471 | if (image->type == KEXEC_TYPE_CRASH) | |
f0f4711a TK |
472 | ret = walk_iomem_res_desc(crashk_res.desc, |
473 | IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, | |
474 | crashk_res.start, crashk_res.end, kbuf, | |
475 | locate_mem_hole_callback); | |
a43cac0d DY |
476 | else |
477 | ret = walk_system_ram_res(0, -1, kbuf, | |
478 | locate_mem_hole_callback); | |
479 | if (ret != 1) { | |
480 | /* A suitable memory range could not be found for buffer */ | |
481 | return -EADDRNOTAVAIL; | |
482 | } | |
483 | ||
484 | /* Found a suitable memory range */ | |
485 | ksegment = &image->segment[image->nr_segments]; | |
486 | ksegment->kbuf = kbuf->buffer; | |
487 | ksegment->bufsz = kbuf->bufsz; | |
488 | ksegment->mem = kbuf->mem; | |
489 | ksegment->memsz = kbuf->memsz; | |
490 | image->nr_segments++; | |
491 | *load_addr = ksegment->mem; | |
492 | return 0; | |
493 | } | |
494 | ||
495 | /* Calculate and store the digest of segments */ | |
496 | static int kexec_calculate_store_digests(struct kimage *image) | |
497 | { | |
498 | struct crypto_shash *tfm; | |
499 | struct shash_desc *desc; | |
500 | int ret = 0, i, j, zero_buf_sz, sha_region_sz; | |
501 | size_t desc_size, nullsz; | |
502 | char *digest; | |
503 | void *zero_buf; | |
504 | struct kexec_sha_region *sha_regions; | |
505 | struct purgatory_info *pi = &image->purgatory_info; | |
506 | ||
507 | zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); | |
508 | zero_buf_sz = PAGE_SIZE; | |
509 | ||
510 | tfm = crypto_alloc_shash("sha256", 0, 0); | |
511 | if (IS_ERR(tfm)) { | |
512 | ret = PTR_ERR(tfm); | |
513 | goto out; | |
514 | } | |
515 | ||
516 | desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); | |
517 | desc = kzalloc(desc_size, GFP_KERNEL); | |
518 | if (!desc) { | |
519 | ret = -ENOMEM; | |
520 | goto out_free_tfm; | |
521 | } | |
522 | ||
523 | sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); | |
524 | sha_regions = vzalloc(sha_region_sz); | |
525 | if (!sha_regions) | |
526 | goto out_free_desc; | |
527 | ||
528 | desc->tfm = tfm; | |
529 | desc->flags = 0; | |
530 | ||
531 | ret = crypto_shash_init(desc); | |
532 | if (ret < 0) | |
533 | goto out_free_sha_regions; | |
534 | ||
535 | digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); | |
536 | if (!digest) { | |
537 | ret = -ENOMEM; | |
538 | goto out_free_sha_regions; | |
539 | } | |
540 | ||
541 | for (j = i = 0; i < image->nr_segments; i++) { | |
542 | struct kexec_segment *ksegment; | |
543 | ||
544 | ksegment = &image->segment[i]; | |
545 | /* | |
546 | * Skip purgatory as it will be modified once we put digest | |
547 | * info in purgatory. | |
548 | */ | |
549 | if (ksegment->kbuf == pi->purgatory_buf) | |
550 | continue; | |
551 | ||
552 | ret = crypto_shash_update(desc, ksegment->kbuf, | |
553 | ksegment->bufsz); | |
554 | if (ret) | |
555 | break; | |
556 | ||
557 | /* | |
558 | * Assume rest of the buffer is filled with zero and | |
559 | * update digest accordingly. | |
560 | */ | |
561 | nullsz = ksegment->memsz - ksegment->bufsz; | |
562 | while (nullsz) { | |
563 | unsigned long bytes = nullsz; | |
564 | ||
565 | if (bytes > zero_buf_sz) | |
566 | bytes = zero_buf_sz; | |
567 | ret = crypto_shash_update(desc, zero_buf, bytes); | |
568 | if (ret) | |
569 | break; | |
570 | nullsz -= bytes; | |
571 | } | |
572 | ||
573 | if (ret) | |
574 | break; | |
575 | ||
576 | sha_regions[j].start = ksegment->mem; | |
577 | sha_regions[j].len = ksegment->memsz; | |
578 | j++; | |
579 | } | |
580 | ||
581 | if (!ret) { | |
582 | ret = crypto_shash_final(desc, digest); | |
583 | if (ret) | |
584 | goto out_free_digest; | |
585 | ret = kexec_purgatory_get_set_symbol(image, "sha_regions", | |
586 | sha_regions, sha_region_sz, 0); | |
587 | if (ret) | |
588 | goto out_free_digest; | |
589 | ||
590 | ret = kexec_purgatory_get_set_symbol(image, "sha256_digest", | |
591 | digest, SHA256_DIGEST_SIZE, 0); | |
592 | if (ret) | |
593 | goto out_free_digest; | |
594 | } | |
595 | ||
596 | out_free_digest: | |
597 | kfree(digest); | |
598 | out_free_sha_regions: | |
599 | vfree(sha_regions); | |
600 | out_free_desc: | |
601 | kfree(desc); | |
602 | out_free_tfm: | |
603 | kfree(tfm); | |
604 | out: | |
605 | return ret; | |
606 | } | |
607 | ||
608 | /* Actually load purgatory. Lot of code taken from kexec-tools */ | |
609 | static int __kexec_load_purgatory(struct kimage *image, unsigned long min, | |
610 | unsigned long max, int top_down) | |
611 | { | |
612 | struct purgatory_info *pi = &image->purgatory_info; | |
613 | unsigned long align, buf_align, bss_align, buf_sz, bss_sz, bss_pad; | |
614 | unsigned long memsz, entry, load_addr, curr_load_addr, bss_addr, offset; | |
615 | unsigned char *buf_addr, *src; | |
616 | int i, ret = 0, entry_sidx = -1; | |
617 | const Elf_Shdr *sechdrs_c; | |
618 | Elf_Shdr *sechdrs = NULL; | |
619 | void *purgatory_buf = NULL; | |
620 | ||
621 | /* | |
622 | * sechdrs_c points to section headers in purgatory and are read | |
623 | * only. No modifications allowed. | |
624 | */ | |
625 | sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff; | |
626 | ||
627 | /* | |
628 | * We can not modify sechdrs_c[] and its fields. It is read only. | |
629 | * Copy it over to a local copy where one can store some temporary | |
630 | * data and free it at the end. We need to modify ->sh_addr and | |
631 | * ->sh_offset fields to keep track of permanent and temporary | |
632 | * locations of sections. | |
633 | */ | |
634 | sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr)); | |
635 | if (!sechdrs) | |
636 | return -ENOMEM; | |
637 | ||
638 | memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr)); | |
639 | ||
640 | /* | |
641 | * We seem to have multiple copies of sections. First copy is which | |
642 | * is embedded in kernel in read only section. Some of these sections | |
643 | * will be copied to a temporary buffer and relocated. And these | |
644 | * sections will finally be copied to their final destination at | |
645 | * segment load time. | |
646 | * | |
647 | * Use ->sh_offset to reflect section address in memory. It will | |
648 | * point to original read only copy if section is not allocatable. | |
649 | * Otherwise it will point to temporary copy which will be relocated. | |
650 | * | |
651 | * Use ->sh_addr to contain final address of the section where it | |
652 | * will go during execution time. | |
653 | */ | |
654 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
655 | if (sechdrs[i].sh_type == SHT_NOBITS) | |
656 | continue; | |
657 | ||
658 | sechdrs[i].sh_offset = (unsigned long)pi->ehdr + | |
659 | sechdrs[i].sh_offset; | |
660 | } | |
661 | ||
662 | /* | |
663 | * Identify entry point section and make entry relative to section | |
664 | * start. | |
665 | */ | |
666 | entry = pi->ehdr->e_entry; | |
667 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
668 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
669 | continue; | |
670 | ||
671 | if (!(sechdrs[i].sh_flags & SHF_EXECINSTR)) | |
672 | continue; | |
673 | ||
674 | /* Make entry section relative */ | |
675 | if (sechdrs[i].sh_addr <= pi->ehdr->e_entry && | |
676 | ((sechdrs[i].sh_addr + sechdrs[i].sh_size) > | |
677 | pi->ehdr->e_entry)) { | |
678 | entry_sidx = i; | |
679 | entry -= sechdrs[i].sh_addr; | |
680 | break; | |
681 | } | |
682 | } | |
683 | ||
684 | /* Determine how much memory is needed to load relocatable object. */ | |
685 | buf_align = 1; | |
686 | bss_align = 1; | |
687 | buf_sz = 0; | |
688 | bss_sz = 0; | |
689 | ||
690 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
691 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
692 | continue; | |
693 | ||
694 | align = sechdrs[i].sh_addralign; | |
695 | if (sechdrs[i].sh_type != SHT_NOBITS) { | |
696 | if (buf_align < align) | |
697 | buf_align = align; | |
698 | buf_sz = ALIGN(buf_sz, align); | |
699 | buf_sz += sechdrs[i].sh_size; | |
700 | } else { | |
701 | /* bss section */ | |
702 | if (bss_align < align) | |
703 | bss_align = align; | |
704 | bss_sz = ALIGN(bss_sz, align); | |
705 | bss_sz += sechdrs[i].sh_size; | |
706 | } | |
707 | } | |
708 | ||
709 | /* Determine the bss padding required to align bss properly */ | |
710 | bss_pad = 0; | |
711 | if (buf_sz & (bss_align - 1)) | |
712 | bss_pad = bss_align - (buf_sz & (bss_align - 1)); | |
713 | ||
714 | memsz = buf_sz + bss_pad + bss_sz; | |
715 | ||
716 | /* Allocate buffer for purgatory */ | |
717 | purgatory_buf = vzalloc(buf_sz); | |
718 | if (!purgatory_buf) { | |
719 | ret = -ENOMEM; | |
720 | goto out; | |
721 | } | |
722 | ||
723 | if (buf_align < bss_align) | |
724 | buf_align = bss_align; | |
725 | ||
726 | /* Add buffer to segment list */ | |
727 | ret = kexec_add_buffer(image, purgatory_buf, buf_sz, memsz, | |
728 | buf_align, min, max, top_down, | |
729 | &pi->purgatory_load_addr); | |
730 | if (ret) | |
731 | goto out; | |
732 | ||
733 | /* Load SHF_ALLOC sections */ | |
734 | buf_addr = purgatory_buf; | |
735 | load_addr = curr_load_addr = pi->purgatory_load_addr; | |
736 | bss_addr = load_addr + buf_sz + bss_pad; | |
737 | ||
738 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
739 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
740 | continue; | |
741 | ||
742 | align = sechdrs[i].sh_addralign; | |
743 | if (sechdrs[i].sh_type != SHT_NOBITS) { | |
744 | curr_load_addr = ALIGN(curr_load_addr, align); | |
745 | offset = curr_load_addr - load_addr; | |
746 | /* We already modifed ->sh_offset to keep src addr */ | |
747 | src = (char *) sechdrs[i].sh_offset; | |
748 | memcpy(buf_addr + offset, src, sechdrs[i].sh_size); | |
749 | ||
750 | /* Store load address and source address of section */ | |
751 | sechdrs[i].sh_addr = curr_load_addr; | |
752 | ||
753 | /* | |
754 | * This section got copied to temporary buffer. Update | |
755 | * ->sh_offset accordingly. | |
756 | */ | |
757 | sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset); | |
758 | ||
759 | /* Advance to the next address */ | |
760 | curr_load_addr += sechdrs[i].sh_size; | |
761 | } else { | |
762 | bss_addr = ALIGN(bss_addr, align); | |
763 | sechdrs[i].sh_addr = bss_addr; | |
764 | bss_addr += sechdrs[i].sh_size; | |
765 | } | |
766 | } | |
767 | ||
768 | /* Update entry point based on load address of text section */ | |
769 | if (entry_sidx >= 0) | |
770 | entry += sechdrs[entry_sidx].sh_addr; | |
771 | ||
772 | /* Make kernel jump to purgatory after shutdown */ | |
773 | image->start = entry; | |
774 | ||
775 | /* Used later to get/set symbol values */ | |
776 | pi->sechdrs = sechdrs; | |
777 | ||
778 | /* | |
779 | * Used later to identify which section is purgatory and skip it | |
780 | * from checksumming. | |
781 | */ | |
782 | pi->purgatory_buf = purgatory_buf; | |
783 | return ret; | |
784 | out: | |
785 | vfree(sechdrs); | |
786 | vfree(purgatory_buf); | |
787 | return ret; | |
788 | } | |
789 | ||
790 | static int kexec_apply_relocations(struct kimage *image) | |
791 | { | |
792 | int i, ret; | |
793 | struct purgatory_info *pi = &image->purgatory_info; | |
794 | Elf_Shdr *sechdrs = pi->sechdrs; | |
795 | ||
796 | /* Apply relocations */ | |
797 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
798 | Elf_Shdr *section, *symtab; | |
799 | ||
800 | if (sechdrs[i].sh_type != SHT_RELA && | |
801 | sechdrs[i].sh_type != SHT_REL) | |
802 | continue; | |
803 | ||
804 | /* | |
805 | * For section of type SHT_RELA/SHT_REL, | |
806 | * ->sh_link contains section header index of associated | |
807 | * symbol table. And ->sh_info contains section header | |
808 | * index of section to which relocations apply. | |
809 | */ | |
810 | if (sechdrs[i].sh_info >= pi->ehdr->e_shnum || | |
811 | sechdrs[i].sh_link >= pi->ehdr->e_shnum) | |
812 | return -ENOEXEC; | |
813 | ||
814 | section = &sechdrs[sechdrs[i].sh_info]; | |
815 | symtab = &sechdrs[sechdrs[i].sh_link]; | |
816 | ||
817 | if (!(section->sh_flags & SHF_ALLOC)) | |
818 | continue; | |
819 | ||
820 | /* | |
821 | * symtab->sh_link contain section header index of associated | |
822 | * string table. | |
823 | */ | |
824 | if (symtab->sh_link >= pi->ehdr->e_shnum) | |
825 | /* Invalid section number? */ | |
826 | continue; | |
827 | ||
828 | /* | |
829 | * Respective architecture needs to provide support for applying | |
830 | * relocations of type SHT_RELA/SHT_REL. | |
831 | */ | |
832 | if (sechdrs[i].sh_type == SHT_RELA) | |
833 | ret = arch_kexec_apply_relocations_add(pi->ehdr, | |
834 | sechdrs, i); | |
835 | else if (sechdrs[i].sh_type == SHT_REL) | |
836 | ret = arch_kexec_apply_relocations(pi->ehdr, | |
837 | sechdrs, i); | |
838 | if (ret) | |
839 | return ret; | |
840 | } | |
841 | ||
842 | return 0; | |
843 | } | |
844 | ||
845 | /* Load relocatable purgatory object and relocate it appropriately */ | |
846 | int kexec_load_purgatory(struct kimage *image, unsigned long min, | |
847 | unsigned long max, int top_down, | |
848 | unsigned long *load_addr) | |
849 | { | |
850 | struct purgatory_info *pi = &image->purgatory_info; | |
851 | int ret; | |
852 | ||
853 | if (kexec_purgatory_size <= 0) | |
854 | return -EINVAL; | |
855 | ||
856 | if (kexec_purgatory_size < sizeof(Elf_Ehdr)) | |
857 | return -ENOEXEC; | |
858 | ||
859 | pi->ehdr = (Elf_Ehdr *)kexec_purgatory; | |
860 | ||
861 | if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0 | |
862 | || pi->ehdr->e_type != ET_REL | |
863 | || !elf_check_arch(pi->ehdr) | |
864 | || pi->ehdr->e_shentsize != sizeof(Elf_Shdr)) | |
865 | return -ENOEXEC; | |
866 | ||
867 | if (pi->ehdr->e_shoff >= kexec_purgatory_size | |
868 | || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) > | |
869 | kexec_purgatory_size - pi->ehdr->e_shoff)) | |
870 | return -ENOEXEC; | |
871 | ||
872 | ret = __kexec_load_purgatory(image, min, max, top_down); | |
873 | if (ret) | |
874 | return ret; | |
875 | ||
876 | ret = kexec_apply_relocations(image); | |
877 | if (ret) | |
878 | goto out; | |
879 | ||
880 | *load_addr = pi->purgatory_load_addr; | |
881 | return 0; | |
882 | out: | |
883 | vfree(pi->sechdrs); | |
884 | vfree(pi->purgatory_buf); | |
885 | return ret; | |
886 | } | |
887 | ||
888 | static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, | |
889 | const char *name) | |
890 | { | |
891 | Elf_Sym *syms; | |
892 | Elf_Shdr *sechdrs; | |
893 | Elf_Ehdr *ehdr; | |
894 | int i, k; | |
895 | const char *strtab; | |
896 | ||
897 | if (!pi->sechdrs || !pi->ehdr) | |
898 | return NULL; | |
899 | ||
900 | sechdrs = pi->sechdrs; | |
901 | ehdr = pi->ehdr; | |
902 | ||
903 | for (i = 0; i < ehdr->e_shnum; i++) { | |
904 | if (sechdrs[i].sh_type != SHT_SYMTAB) | |
905 | continue; | |
906 | ||
907 | if (sechdrs[i].sh_link >= ehdr->e_shnum) | |
908 | /* Invalid strtab section number */ | |
909 | continue; | |
910 | strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset; | |
911 | syms = (Elf_Sym *)sechdrs[i].sh_offset; | |
912 | ||
913 | /* Go through symbols for a match */ | |
914 | for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { | |
915 | if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) | |
916 | continue; | |
917 | ||
918 | if (strcmp(strtab + syms[k].st_name, name) != 0) | |
919 | continue; | |
920 | ||
921 | if (syms[k].st_shndx == SHN_UNDEF || | |
922 | syms[k].st_shndx >= ehdr->e_shnum) { | |
923 | pr_debug("Symbol: %s has bad section index %d.\n", | |
924 | name, syms[k].st_shndx); | |
925 | return NULL; | |
926 | } | |
927 | ||
928 | /* Found the symbol we are looking for */ | |
929 | return &syms[k]; | |
930 | } | |
931 | } | |
932 | ||
933 | return NULL; | |
934 | } | |
935 | ||
936 | void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) | |
937 | { | |
938 | struct purgatory_info *pi = &image->purgatory_info; | |
939 | Elf_Sym *sym; | |
940 | Elf_Shdr *sechdr; | |
941 | ||
942 | sym = kexec_purgatory_find_symbol(pi, name); | |
943 | if (!sym) | |
944 | return ERR_PTR(-EINVAL); | |
945 | ||
946 | sechdr = &pi->sechdrs[sym->st_shndx]; | |
947 | ||
948 | /* | |
949 | * Returns the address where symbol will finally be loaded after | |
950 | * kexec_load_segment() | |
951 | */ | |
952 | return (void *)(sechdr->sh_addr + sym->st_value); | |
953 | } | |
954 | ||
955 | /* | |
956 | * Get or set value of a symbol. If "get_value" is true, symbol value is | |
957 | * returned in buf otherwise symbol value is set based on value in buf. | |
958 | */ | |
959 | int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, | |
960 | void *buf, unsigned int size, bool get_value) | |
961 | { | |
962 | Elf_Sym *sym; | |
963 | Elf_Shdr *sechdrs; | |
964 | struct purgatory_info *pi = &image->purgatory_info; | |
965 | char *sym_buf; | |
966 | ||
967 | sym = kexec_purgatory_find_symbol(pi, name); | |
968 | if (!sym) | |
969 | return -EINVAL; | |
970 | ||
971 | if (sym->st_size != size) { | |
972 | pr_err("symbol %s size mismatch: expected %lu actual %u\n", | |
973 | name, (unsigned long)sym->st_size, size); | |
974 | return -EINVAL; | |
975 | } | |
976 | ||
977 | sechdrs = pi->sechdrs; | |
978 | ||
979 | if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) { | |
980 | pr_err("symbol %s is in a bss section. Cannot %s\n", name, | |
981 | get_value ? "get" : "set"); | |
982 | return -EINVAL; | |
983 | } | |
984 | ||
985 | sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset + | |
986 | sym->st_value; | |
987 | ||
988 | if (get_value) | |
989 | memcpy((void *)buf, sym_buf, size); | |
990 | else | |
991 | memcpy((void *)sym_buf, buf, size); | |
992 | ||
993 | return 0; | |
994 | } |