1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * kexec: kexec_file_load system call |
4 | * |
5 | * Copyright (C) 2014 Red Hat Inc. |
6 | * Authors: |
7 | * Vivek Goyal <vgoyal@redhat.com> |
8 | */ |
9 | |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | |
12 | #include <linux/capability.h> |
13 | #include <linux/mm.h> |
14 | #include <linux/file.h> |
15 | #include <linux/slab.h> |
16 | #include <linux/kexec.h> |
17 | #include <linux/memblock.h> |
18 | #include <linux/mutex.h> |
19 | #include <linux/list.h> |
20 | #include <linux/fs.h> |
21 | #include <linux/ima.h> |
22 | #include <crypto/hash.h> |
23 | #include <crypto/sha2.h> |
24 | #include <linux/elf.h> |
25 | #include <linux/elfcore.h> |
26 | #include <linux/kernel.h> |
27 | #include <linux/kernel_read_file.h> |
28 | #include <linux/syscalls.h> |
29 | #include <linux/vmalloc.h> |
30 | #include "kexec_internal.h" |
31 | |
32 | #ifdef CONFIG_KEXEC_SIG |
33 | static bool sig_enforce = IS_ENABLED(CONFIG_KEXEC_SIG_FORCE); |
34 | |
35 | void set_kexec_sig_enforced(void) |
36 | { |
37 | sig_enforce = true; |
38 | } |
39 | #endif |
40 | |
41 | static int kexec_calculate_store_digests(struct kimage *image); |
42 | |
43 | /* Maximum size in bytes for kernel/initrd files. */ |
44 | #define KEXEC_FILE_SIZE_MAX min_t(s64, 4LL << 30, SSIZE_MAX) |
45 | |
46 | /* |
47 | * Currently this is the only default function that is exported as some |
48 | * architectures need it to do additional handlings. |
49 | * In the future, other default functions may be exported too if required. |
50 | */ |
51 | int kexec_image_probe_default(struct kimage *image, void *buf, |
52 | unsigned long buf_len) |
53 | { |
54 | const struct kexec_file_ops * const *fops; |
55 | int ret = -ENOEXEC; |
56 | |
57 | for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) { |
58 | ret = (*fops)->probe(buf, buf_len); |
59 | if (!ret) { |
60 | image->fops = *fops; |
61 | return ret; |
62 | } |
63 | } |
64 | |
65 | return ret; |
66 | } |
67 | |
68 | static void *kexec_image_load_default(struct kimage *image) |
69 | { |
70 | if (!image->fops || !image->fops->load) |
71 | return ERR_PTR(error: -ENOEXEC); |
72 | |
73 | return image->fops->load(image, image->kernel_buf, |
74 | image->kernel_buf_len, image->initrd_buf, |
75 | image->initrd_buf_len, image->cmdline_buf, |
76 | image->cmdline_buf_len); |
77 | } |
78 | |
79 | int kexec_image_post_load_cleanup_default(struct kimage *image) |
80 | { |
81 | if (!image->fops || !image->fops->cleanup) |
82 | return 0; |
83 | |
84 | return image->fops->cleanup(image->image_loader_data); |
85 | } |
86 | |
87 | /* |
88 | * Free up memory used by kernel, initrd, and command line. This is temporary |
89 | * memory allocation which is not needed any more after these buffers have |
90 | * been loaded into separate segments and have been copied elsewhere. |
91 | */ |
92 | void kimage_file_post_load_cleanup(struct kimage *image) |
93 | { |
94 | struct purgatory_info *pi = &image->purgatory_info; |
95 | |
96 | vfree(addr: image->kernel_buf); |
97 | image->kernel_buf = NULL; |
98 | |
99 | vfree(addr: image->initrd_buf); |
100 | image->initrd_buf = NULL; |
101 | |
102 | kfree(objp: image->cmdline_buf); |
103 | image->cmdline_buf = NULL; |
104 | |
105 | vfree(addr: pi->purgatory_buf); |
106 | pi->purgatory_buf = NULL; |
107 | |
108 | vfree(addr: pi->sechdrs); |
109 | pi->sechdrs = NULL; |
110 | |
111 | #ifdef CONFIG_IMA_KEXEC |
112 | vfree(addr: image->ima_buffer); |
113 | image->ima_buffer = NULL; |
114 | #endif /* CONFIG_IMA_KEXEC */ |
115 | |
116 | /* See if architecture has anything to cleanup post load */ |
117 | arch_kimage_file_post_load_cleanup(image); |
118 | |
119 | /* |
120 | * Above call should have called into bootloader to free up |
121 | * any data stored in kimage->image_loader_data. It should |
122 | * be ok now to free it up. |
123 | */ |
124 | kfree(objp: image->image_loader_data); |
125 | image->image_loader_data = NULL; |
126 | } |
127 | |
128 | #ifdef CONFIG_KEXEC_SIG |
129 | #ifdef CONFIG_SIGNED_PE_FILE_VERIFICATION |
130 | int kexec_kernel_verify_pe_sig(const char *kernel, unsigned long kernel_len) |
131 | { |
132 | int ret; |
133 | |
134 | ret = verify_pefile_signature(pebuf: kernel, pelen: kernel_len, |
135 | VERIFY_USE_SECONDARY_KEYRING, |
136 | usage: VERIFYING_KEXEC_PE_SIGNATURE); |
137 | if (ret == -ENOKEY && IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING)) { |
138 | ret = verify_pefile_signature(pebuf: kernel, pelen: kernel_len, |
139 | VERIFY_USE_PLATFORM_KEYRING, |
140 | usage: VERIFYING_KEXEC_PE_SIGNATURE); |
141 | } |
142 | return ret; |
143 | } |
144 | #endif |
145 | |
146 | static int kexec_image_verify_sig(struct kimage *image, void *buf, |
147 | unsigned long buf_len) |
148 | { |
149 | if (!image->fops || !image->fops->verify_sig) { |
150 | pr_debug("kernel loader does not support signature verification.\n" ); |
151 | return -EKEYREJECTED; |
152 | } |
153 | |
154 | return image->fops->verify_sig(buf, buf_len); |
155 | } |
156 | |
157 | static int |
158 | kimage_validate_signature(struct kimage *image) |
159 | { |
160 | int ret; |
161 | |
162 | ret = kexec_image_verify_sig(image, buf: image->kernel_buf, |
163 | buf_len: image->kernel_buf_len); |
164 | if (ret) { |
165 | |
166 | if (sig_enforce) { |
167 | pr_notice("Enforced kernel signature verification failed (%d).\n" , ret); |
168 | return ret; |
169 | } |
170 | |
171 | /* |
172 | * If IMA is guaranteed to appraise a signature on the kexec |
173 | * image, permit it even if the kernel is otherwise locked |
174 | * down. |
175 | */ |
176 | if (!ima_appraise_signature(func: READING_KEXEC_IMAGE) && |
177 | security_locked_down(what: LOCKDOWN_KEXEC)) |
178 | return -EPERM; |
179 | |
180 | pr_debug("kernel signature verification failed (%d).\n" , ret); |
181 | } |
182 | |
183 | return 0; |
184 | } |
185 | #endif |
186 | |
187 | /* |
188 | * In file mode list of segments is prepared by kernel. Copy relevant |
189 | * data from user space, do error checking, prepare segment list |
190 | */ |
191 | static int |
192 | kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, |
193 | const char __user *cmdline_ptr, |
194 | unsigned long cmdline_len, unsigned flags) |
195 | { |
196 | ssize_t ret; |
197 | void *ldata; |
198 | |
199 | ret = kernel_read_file_from_fd(fd: kernel_fd, offset: 0, buf: &image->kernel_buf, |
200 | KEXEC_FILE_SIZE_MAX, NULL, |
201 | id: READING_KEXEC_IMAGE); |
202 | if (ret < 0) |
203 | return ret; |
204 | image->kernel_buf_len = ret; |
205 | |
206 | /* Call arch image probe handlers */ |
207 | ret = arch_kexec_kernel_image_probe(image, buf: image->kernel_buf, |
208 | buf_len: image->kernel_buf_len); |
209 | if (ret) |
210 | goto out; |
211 | |
212 | #ifdef CONFIG_KEXEC_SIG |
213 | ret = kimage_validate_signature(image); |
214 | |
215 | if (ret) |
216 | goto out; |
217 | #endif |
218 | /* It is possible that there no initramfs is being loaded */ |
219 | if (!(flags & KEXEC_FILE_NO_INITRAMFS)) { |
220 | ret = kernel_read_file_from_fd(fd: initrd_fd, offset: 0, buf: &image->initrd_buf, |
221 | KEXEC_FILE_SIZE_MAX, NULL, |
222 | id: READING_KEXEC_INITRAMFS); |
223 | if (ret < 0) |
224 | goto out; |
225 | image->initrd_buf_len = ret; |
226 | ret = 0; |
227 | } |
228 | |
229 | if (cmdline_len) { |
230 | image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len); |
231 | if (IS_ERR(ptr: image->cmdline_buf)) { |
232 | ret = PTR_ERR(ptr: image->cmdline_buf); |
233 | image->cmdline_buf = NULL; |
234 | goto out; |
235 | } |
236 | |
237 | image->cmdline_buf_len = cmdline_len; |
238 | |
239 | /* command line should be a string with last byte null */ |
240 | if (image->cmdline_buf[cmdline_len - 1] != '\0') { |
241 | ret = -EINVAL; |
242 | goto out; |
243 | } |
244 | |
245 | ima_kexec_cmdline(kernel_fd, buf: image->cmdline_buf, |
246 | size: image->cmdline_buf_len - 1); |
247 | } |
248 | |
249 | /* IMA needs to pass the measurement list to the next kernel. */ |
250 | ima_add_kexec_buffer(image); |
251 | |
252 | /* Call image load handler */ |
253 | ldata = kexec_image_load_default(image); |
254 | |
255 | if (IS_ERR(ptr: ldata)) { |
256 | ret = PTR_ERR(ptr: ldata); |
257 | goto out; |
258 | } |
259 | |
260 | image->image_loader_data = ldata; |
261 | out: |
262 | /* In case of error, free up all allocated memory in this function */ |
263 | if (ret) |
264 | kimage_file_post_load_cleanup(image); |
265 | return ret; |
266 | } |
267 | |
268 | static int |
269 | kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, |
270 | int initrd_fd, const char __user *cmdline_ptr, |
271 | unsigned long cmdline_len, unsigned long flags) |
272 | { |
273 | int ret; |
274 | struct kimage *image; |
275 | bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; |
276 | |
277 | image = do_kimage_alloc_init(); |
278 | if (!image) |
279 | return -ENOMEM; |
280 | |
281 | image->file_mode = 1; |
282 | |
283 | if (kexec_on_panic) { |
284 | /* Enable special crash kernel control page alloc policy. */ |
285 | image->control_page = crashk_res.start; |
286 | image->type = KEXEC_TYPE_CRASH; |
287 | } |
288 | |
289 | ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, |
290 | cmdline_ptr, cmdline_len, flags); |
291 | if (ret) |
292 | goto out_free_image; |
293 | |
294 | ret = sanity_check_segment_list(image); |
295 | if (ret) |
296 | goto out_free_post_load_bufs; |
297 | |
298 | ret = -ENOMEM; |
299 | image->control_code_page = kimage_alloc_control_pages(image, |
300 | order: get_order(KEXEC_CONTROL_PAGE_SIZE)); |
301 | if (!image->control_code_page) { |
302 | pr_err("Could not allocate control_code_buffer\n" ); |
303 | goto out_free_post_load_bufs; |
304 | } |
305 | |
306 | if (!kexec_on_panic) { |
307 | image->swap_page = kimage_alloc_control_pages(image, order: 0); |
308 | if (!image->swap_page) { |
309 | pr_err("Could not allocate swap buffer\n" ); |
310 | goto out_free_control_pages; |
311 | } |
312 | } |
313 | |
314 | *rimage = image; |
315 | return 0; |
316 | out_free_control_pages: |
317 | kimage_free_page_list(list: &image->control_pages); |
318 | out_free_post_load_bufs: |
319 | kimage_file_post_load_cleanup(image); |
320 | out_free_image: |
321 | kfree(objp: image); |
322 | return ret; |
323 | } |
324 | |
325 | SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, |
326 | unsigned long, cmdline_len, const char __user *, cmdline_ptr, |
327 | unsigned long, flags) |
328 | { |
329 | int image_type = (flags & KEXEC_FILE_ON_CRASH) ? |
330 | KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT; |
331 | struct kimage **dest_image, *image; |
332 | int ret = 0, i; |
333 | |
334 | /* We only trust the superuser with rebooting the system. */ |
335 | if (!kexec_load_permitted(kexec_image_type: image_type)) |
336 | return -EPERM; |
337 | |
338 | /* Make sure we have a legal set of flags */ |
339 | if (flags != (flags & KEXEC_FILE_FLAGS)) |
340 | return -EINVAL; |
341 | |
342 | image = NULL; |
343 | |
344 | if (!kexec_trylock()) |
345 | return -EBUSY; |
346 | |
347 | if (image_type == KEXEC_TYPE_CRASH) { |
348 | dest_image = &kexec_crash_image; |
349 | if (kexec_crash_image) |
350 | arch_kexec_unprotect_crashkres(); |
351 | } else { |
352 | dest_image = &kexec_image; |
353 | } |
354 | |
355 | if (flags & KEXEC_FILE_UNLOAD) |
356 | goto exchange; |
357 | |
358 | /* |
359 | * In case of crash, new kernel gets loaded in reserved region. It is |
360 | * same memory where old crash kernel might be loaded. Free any |
361 | * current crash dump kernel before we corrupt it. |
362 | */ |
363 | if (flags & KEXEC_FILE_ON_CRASH) |
364 | kimage_free(xchg(&kexec_crash_image, NULL)); |
365 | |
366 | ret = kimage_file_alloc_init(rimage: &image, kernel_fd, initrd_fd, cmdline_ptr, |
367 | cmdline_len, flags); |
368 | if (ret) |
369 | goto out; |
370 | |
371 | ret = machine_kexec_prepare(image); |
372 | if (ret) |
373 | goto out; |
374 | |
375 | /* |
376 | * Some architecture(like S390) may touch the crash memory before |
377 | * machine_kexec_prepare(), we must copy vmcoreinfo data after it. |
378 | */ |
379 | ret = kimage_crash_copy_vmcoreinfo(image); |
380 | if (ret) |
381 | goto out; |
382 | |
383 | ret = kexec_calculate_store_digests(image); |
384 | if (ret) |
385 | goto out; |
386 | |
387 | for (i = 0; i < image->nr_segments; i++) { |
388 | struct kexec_segment *ksegment; |
389 | |
390 | ksegment = &image->segment[i]; |
391 | pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n" , |
392 | i, ksegment->buf, ksegment->bufsz, ksegment->mem, |
393 | ksegment->memsz); |
394 | |
395 | ret = kimage_load_segment(image, segment: &image->segment[i]); |
396 | if (ret) |
397 | goto out; |
398 | } |
399 | |
400 | kimage_terminate(image); |
401 | |
402 | ret = machine_kexec_post_load(image); |
403 | if (ret) |
404 | goto out; |
405 | |
406 | /* |
407 | * Free up any temporary buffers allocated which are not needed |
408 | * after image has been loaded |
409 | */ |
410 | kimage_file_post_load_cleanup(image); |
411 | exchange: |
412 | image = xchg(dest_image, image); |
413 | out: |
414 | if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image) |
415 | arch_kexec_protect_crashkres(); |
416 | |
417 | kexec_unlock(); |
418 | kimage_free(image); |
419 | return ret; |
420 | } |
421 | |
422 | static int locate_mem_hole_top_down(unsigned long start, unsigned long end, |
423 | struct kexec_buf *kbuf) |
424 | { |
425 | struct kimage *image = kbuf->image; |
426 | unsigned long temp_start, temp_end; |
427 | |
428 | temp_end = min(end, kbuf->buf_max); |
429 | temp_start = temp_end - kbuf->memsz; |
430 | |
431 | do { |
432 | /* align down start */ |
433 | temp_start = temp_start & (~(kbuf->buf_align - 1)); |
434 | |
435 | if (temp_start < start || temp_start < kbuf->buf_min) |
436 | return 0; |
437 | |
438 | temp_end = temp_start + kbuf->memsz - 1; |
439 | |
440 | /* |
441 | * Make sure this does not conflict with any of existing |
442 | * segments |
443 | */ |
444 | if (kimage_is_destination_range(image, start: temp_start, end: temp_end)) { |
445 | temp_start = temp_start - PAGE_SIZE; |
446 | continue; |
447 | } |
448 | |
449 | /* We found a suitable memory range */ |
450 | break; |
451 | } while (1); |
452 | |
453 | /* If we are here, we found a suitable memory range */ |
454 | kbuf->mem = temp_start; |
455 | |
456 | /* Success, stop navigating through remaining System RAM ranges */ |
457 | return 1; |
458 | } |
459 | |
460 | static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, |
461 | struct kexec_buf *kbuf) |
462 | { |
463 | struct kimage *image = kbuf->image; |
464 | unsigned long temp_start, temp_end; |
465 | |
466 | temp_start = max(start, kbuf->buf_min); |
467 | |
468 | do { |
469 | temp_start = ALIGN(temp_start, kbuf->buf_align); |
470 | temp_end = temp_start + kbuf->memsz - 1; |
471 | |
472 | if (temp_end > end || temp_end > kbuf->buf_max) |
473 | return 0; |
474 | /* |
475 | * Make sure this does not conflict with any of existing |
476 | * segments |
477 | */ |
478 | if (kimage_is_destination_range(image, start: temp_start, end: temp_end)) { |
479 | temp_start = temp_start + PAGE_SIZE; |
480 | continue; |
481 | } |
482 | |
483 | /* We found a suitable memory range */ |
484 | break; |
485 | } while (1); |
486 | |
487 | /* If we are here, we found a suitable memory range */ |
488 | kbuf->mem = temp_start; |
489 | |
490 | /* Success, stop navigating through remaining System RAM ranges */ |
491 | return 1; |
492 | } |
493 | |
494 | static int locate_mem_hole_callback(struct resource *res, void *arg) |
495 | { |
496 | struct kexec_buf *kbuf = (struct kexec_buf *)arg; |
497 | u64 start = res->start, end = res->end; |
498 | unsigned long sz = end - start + 1; |
499 | |
500 | /* Returning 0 will take to next memory range */ |
501 | |
502 | /* Don't use memory that will be detected and handled by a driver. */ |
503 | if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED) |
504 | return 0; |
505 | |
506 | if (sz < kbuf->memsz) |
507 | return 0; |
508 | |
509 | if (end < kbuf->buf_min || start > kbuf->buf_max) |
510 | return 0; |
511 | |
512 | /* |
513 | * Allocate memory top down with-in ram range. Otherwise bottom up |
514 | * allocation. |
515 | */ |
516 | if (kbuf->top_down) |
517 | return locate_mem_hole_top_down(start, end, kbuf); |
518 | return locate_mem_hole_bottom_up(start, end, kbuf); |
519 | } |
520 | |
521 | #ifdef CONFIG_ARCH_KEEP_MEMBLOCK |
522 | static int kexec_walk_memblock(struct kexec_buf *kbuf, |
523 | int (*func)(struct resource *, void *)) |
524 | { |
525 | int ret = 0; |
526 | u64 i; |
527 | phys_addr_t mstart, mend; |
528 | struct resource res = { }; |
529 | |
530 | if (kbuf->image->type == KEXEC_TYPE_CRASH) |
531 | return func(&crashk_res, kbuf); |
532 | |
533 | /* |
534 | * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See |
535 | * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in |
536 | * locate_mem_hole_callback(). |
537 | */ |
538 | if (kbuf->top_down) { |
539 | for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE, |
540 | &mstart, &mend, NULL) { |
541 | /* |
542 | * In memblock, end points to the first byte after the |
543 | * range while in kexec, end points to the last byte |
544 | * in the range. |
545 | */ |
546 | res.start = mstart; |
547 | res.end = mend - 1; |
548 | ret = func(&res, kbuf); |
549 | if (ret) |
550 | break; |
551 | } |
552 | } else { |
553 | for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, |
554 | &mstart, &mend, NULL) { |
555 | /* |
556 | * In memblock, end points to the first byte after the |
557 | * range while in kexec, end points to the last byte |
558 | * in the range. |
559 | */ |
560 | res.start = mstart; |
561 | res.end = mend - 1; |
562 | ret = func(&res, kbuf); |
563 | if (ret) |
564 | break; |
565 | } |
566 | } |
567 | |
568 | return ret; |
569 | } |
570 | #else |
571 | static int kexec_walk_memblock(struct kexec_buf *kbuf, |
572 | int (*func)(struct resource *, void *)) |
573 | { |
574 | return 0; |
575 | } |
576 | #endif |
577 | |
578 | /** |
579 | * kexec_walk_resources - call func(data) on free memory regions |
580 | * @kbuf: Context info for the search. Also passed to @func. |
581 | * @func: Function to call for each memory region. |
582 | * |
583 | * Return: The memory walk will stop when func returns a non-zero value |
584 | * and that value will be returned. If all free regions are visited without |
585 | * func returning non-zero, then zero will be returned. |
586 | */ |
587 | static int kexec_walk_resources(struct kexec_buf *kbuf, |
588 | int (*func)(struct resource *, void *)) |
589 | { |
590 | if (kbuf->image->type == KEXEC_TYPE_CRASH) |
591 | return walk_iomem_res_desc(desc: crashk_res.desc, |
592 | IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY, |
593 | start: crashk_res.start, end: crashk_res.end, |
594 | arg: kbuf, func); |
595 | else |
596 | return walk_system_ram_res(start: 0, ULONG_MAX, arg: kbuf, func); |
597 | } |
598 | |
599 | /** |
600 | * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel |
601 | * @kbuf: Parameters for the memory search. |
602 | * |
603 | * On success, kbuf->mem will have the start address of the memory region found. |
604 | * |
605 | * Return: 0 on success, negative errno on error. |
606 | */ |
607 | int kexec_locate_mem_hole(struct kexec_buf *kbuf) |
608 | { |
609 | int ret; |
610 | |
611 | /* Arch knows where to place */ |
612 | if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN) |
613 | return 0; |
614 | |
615 | if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK)) |
616 | ret = kexec_walk_resources(kbuf, func: locate_mem_hole_callback); |
617 | else |
618 | ret = kexec_walk_memblock(kbuf, func: locate_mem_hole_callback); |
619 | |
620 | return ret == 1 ? 0 : -EADDRNOTAVAIL; |
621 | } |
622 | |
623 | /** |
624 | * kexec_add_buffer - place a buffer in a kexec segment |
625 | * @kbuf: Buffer contents and memory parameters. |
626 | * |
627 | * This function assumes that kexec_lock is held. |
628 | * On successful return, @kbuf->mem will have the physical address of |
629 | * the buffer in memory. |
630 | * |
631 | * Return: 0 on success, negative errno on error. |
632 | */ |
633 | int kexec_add_buffer(struct kexec_buf *kbuf) |
634 | { |
635 | struct kexec_segment *ksegment; |
636 | int ret; |
637 | |
638 | /* Currently adding segment this way is allowed only in file mode */ |
639 | if (!kbuf->image->file_mode) |
640 | return -EINVAL; |
641 | |
642 | if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX) |
643 | return -EINVAL; |
644 | |
645 | /* |
646 | * Make sure we are not trying to add buffer after allocating |
647 | * control pages. All segments need to be placed first before |
648 | * any control pages are allocated. As control page allocation |
649 | * logic goes through list of segments to make sure there are |
650 | * no destination overlaps. |
651 | */ |
652 | if (!list_empty(head: &kbuf->image->control_pages)) { |
653 | WARN_ON(1); |
654 | return -EINVAL; |
655 | } |
656 | |
657 | /* Ensure minimum alignment needed for segments. */ |
658 | kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE); |
659 | kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE); |
660 | |
661 | /* Walk the RAM ranges and allocate a suitable range for the buffer */ |
662 | ret = arch_kexec_locate_mem_hole(kbuf); |
663 | if (ret) |
664 | return ret; |
665 | |
666 | /* Found a suitable memory range */ |
667 | ksegment = &kbuf->image->segment[kbuf->image->nr_segments]; |
668 | ksegment->kbuf = kbuf->buffer; |
669 | ksegment->bufsz = kbuf->bufsz; |
670 | ksegment->mem = kbuf->mem; |
671 | ksegment->memsz = kbuf->memsz; |
672 | kbuf->image->nr_segments++; |
673 | return 0; |
674 | } |
675 | |
676 | /* Calculate and store the digest of segments */ |
677 | static int kexec_calculate_store_digests(struct kimage *image) |
678 | { |
679 | struct crypto_shash *tfm; |
680 | struct shash_desc *desc; |
681 | int ret = 0, i, j, zero_buf_sz, sha_region_sz; |
682 | size_t desc_size, nullsz; |
683 | char *digest; |
684 | void *zero_buf; |
685 | struct kexec_sha_region *sha_regions; |
686 | struct purgatory_info *pi = &image->purgatory_info; |
687 | |
688 | if (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY)) |
689 | return 0; |
690 | |
691 | zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); |
692 | zero_buf_sz = PAGE_SIZE; |
693 | |
694 | tfm = crypto_alloc_shash(alg_name: "sha256" , type: 0, mask: 0); |
695 | if (IS_ERR(ptr: tfm)) { |
696 | ret = PTR_ERR(ptr: tfm); |
697 | goto out; |
698 | } |
699 | |
700 | desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); |
701 | desc = kzalloc(size: desc_size, GFP_KERNEL); |
702 | if (!desc) { |
703 | ret = -ENOMEM; |
704 | goto out_free_tfm; |
705 | } |
706 | |
707 | sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); |
708 | sha_regions = vzalloc(size: sha_region_sz); |
709 | if (!sha_regions) { |
710 | ret = -ENOMEM; |
711 | goto out_free_desc; |
712 | } |
713 | |
714 | desc->tfm = tfm; |
715 | |
716 | ret = crypto_shash_init(desc); |
717 | if (ret < 0) |
718 | goto out_free_sha_regions; |
719 | |
720 | digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); |
721 | if (!digest) { |
722 | ret = -ENOMEM; |
723 | goto out_free_sha_regions; |
724 | } |
725 | |
726 | for (j = i = 0; i < image->nr_segments; i++) { |
727 | struct kexec_segment *ksegment; |
728 | |
729 | #ifdef CONFIG_CRASH_HOTPLUG |
730 | /* Exclude elfcorehdr segment to allow future changes via hotplug */ |
731 | if (j == image->elfcorehdr_index) |
732 | continue; |
733 | #endif |
734 | |
735 | ksegment = &image->segment[i]; |
736 | /* |
737 | * Skip purgatory as it will be modified once we put digest |
738 | * info in purgatory. |
739 | */ |
740 | if (ksegment->kbuf == pi->purgatory_buf) |
741 | continue; |
742 | |
743 | ret = crypto_shash_update(desc, data: ksegment->kbuf, |
744 | len: ksegment->bufsz); |
745 | if (ret) |
746 | break; |
747 | |
748 | /* |
749 | * Assume rest of the buffer is filled with zero and |
750 | * update digest accordingly. |
751 | */ |
752 | nullsz = ksegment->memsz - ksegment->bufsz; |
753 | while (nullsz) { |
754 | unsigned long bytes = nullsz; |
755 | |
756 | if (bytes > zero_buf_sz) |
757 | bytes = zero_buf_sz; |
758 | ret = crypto_shash_update(desc, data: zero_buf, len: bytes); |
759 | if (ret) |
760 | break; |
761 | nullsz -= bytes; |
762 | } |
763 | |
764 | if (ret) |
765 | break; |
766 | |
767 | sha_regions[j].start = ksegment->mem; |
768 | sha_regions[j].len = ksegment->memsz; |
769 | j++; |
770 | } |
771 | |
772 | if (!ret) { |
773 | ret = crypto_shash_final(desc, out: digest); |
774 | if (ret) |
775 | goto out_free_digest; |
776 | ret = kexec_purgatory_get_set_symbol(image, name: "purgatory_sha_regions" , |
777 | buf: sha_regions, size: sha_region_sz, get_value: 0); |
778 | if (ret) |
779 | goto out_free_digest; |
780 | |
781 | ret = kexec_purgatory_get_set_symbol(image, name: "purgatory_sha256_digest" , |
782 | buf: digest, SHA256_DIGEST_SIZE, get_value: 0); |
783 | if (ret) |
784 | goto out_free_digest; |
785 | } |
786 | |
787 | out_free_digest: |
788 | kfree(objp: digest); |
789 | out_free_sha_regions: |
790 | vfree(addr: sha_regions); |
791 | out_free_desc: |
792 | kfree(objp: desc); |
793 | out_free_tfm: |
794 | kfree(objp: tfm); |
795 | out: |
796 | return ret; |
797 | } |
798 | |
799 | #ifdef CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY |
800 | /* |
801 | * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory. |
802 | * @pi: Purgatory to be loaded. |
803 | * @kbuf: Buffer to setup. |
804 | * |
805 | * Allocates the memory needed for the buffer. Caller is responsible to free |
806 | * the memory after use. |
807 | * |
808 | * Return: 0 on success, negative errno on error. |
809 | */ |
810 | static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi, |
811 | struct kexec_buf *kbuf) |
812 | { |
813 | const Elf_Shdr *sechdrs; |
814 | unsigned long bss_align; |
815 | unsigned long bss_sz; |
816 | unsigned long align; |
817 | int i, ret; |
818 | |
819 | sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; |
820 | kbuf->buf_align = bss_align = 1; |
821 | kbuf->bufsz = bss_sz = 0; |
822 | |
823 | for (i = 0; i < pi->ehdr->e_shnum; i++) { |
824 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) |
825 | continue; |
826 | |
827 | align = sechdrs[i].sh_addralign; |
828 | if (sechdrs[i].sh_type != SHT_NOBITS) { |
829 | if (kbuf->buf_align < align) |
830 | kbuf->buf_align = align; |
831 | kbuf->bufsz = ALIGN(kbuf->bufsz, align); |
832 | kbuf->bufsz += sechdrs[i].sh_size; |
833 | } else { |
834 | if (bss_align < align) |
835 | bss_align = align; |
836 | bss_sz = ALIGN(bss_sz, align); |
837 | bss_sz += sechdrs[i].sh_size; |
838 | } |
839 | } |
840 | kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align); |
841 | kbuf->memsz = kbuf->bufsz + bss_sz; |
842 | if (kbuf->buf_align < bss_align) |
843 | kbuf->buf_align = bss_align; |
844 | |
845 | kbuf->buffer = vzalloc(size: kbuf->bufsz); |
846 | if (!kbuf->buffer) |
847 | return -ENOMEM; |
848 | pi->purgatory_buf = kbuf->buffer; |
849 | |
850 | ret = kexec_add_buffer(kbuf); |
851 | if (ret) |
852 | goto out; |
853 | |
854 | return 0; |
855 | out: |
856 | vfree(addr: pi->purgatory_buf); |
857 | pi->purgatory_buf = NULL; |
858 | return ret; |
859 | } |
860 | |
861 | /* |
862 | * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer. |
863 | * @pi: Purgatory to be loaded. |
864 | * @kbuf: Buffer prepared to store purgatory. |
865 | * |
866 | * Allocates the memory needed for the buffer. Caller is responsible to free |
867 | * the memory after use. |
868 | * |
869 | * Return: 0 on success, negative errno on error. |
870 | */ |
871 | static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi, |
872 | struct kexec_buf *kbuf) |
873 | { |
874 | unsigned long bss_addr; |
875 | unsigned long offset; |
876 | size_t sechdrs_size; |
877 | Elf_Shdr *sechdrs; |
878 | int i; |
879 | |
880 | /* |
881 | * The section headers in kexec_purgatory are read-only. In order to |
882 | * have them modifiable make a temporary copy. |
883 | */ |
884 | sechdrs_size = array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum); |
885 | sechdrs = vzalloc(size: sechdrs_size); |
886 | if (!sechdrs) |
887 | return -ENOMEM; |
888 | memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, sechdrs_size); |
889 | pi->sechdrs = sechdrs; |
890 | |
891 | offset = 0; |
892 | bss_addr = kbuf->mem + kbuf->bufsz; |
893 | kbuf->image->start = pi->ehdr->e_entry; |
894 | |
895 | for (i = 0; i < pi->ehdr->e_shnum; i++) { |
896 | unsigned long align; |
897 | void *src, *dst; |
898 | |
899 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) |
900 | continue; |
901 | |
902 | align = sechdrs[i].sh_addralign; |
903 | if (sechdrs[i].sh_type == SHT_NOBITS) { |
904 | bss_addr = ALIGN(bss_addr, align); |
905 | sechdrs[i].sh_addr = bss_addr; |
906 | bss_addr += sechdrs[i].sh_size; |
907 | continue; |
908 | } |
909 | |
910 | offset = ALIGN(offset, align); |
911 | |
912 | /* |
913 | * Check if the segment contains the entry point, if so, |
914 | * calculate the value of image->start based on it. |
915 | * If the compiler has produced more than one .text section |
916 | * (Eg: .text.hot), they are generally after the main .text |
917 | * section, and they shall not be used to calculate |
918 | * image->start. So do not re-calculate image->start if it |
919 | * is not set to the initial value, and warn the user so they |
920 | * have a chance to fix their purgatory's linker script. |
921 | */ |
922 | if (sechdrs[i].sh_flags & SHF_EXECINSTR && |
923 | pi->ehdr->e_entry >= sechdrs[i].sh_addr && |
924 | pi->ehdr->e_entry < (sechdrs[i].sh_addr |
925 | + sechdrs[i].sh_size) && |
926 | !WARN_ON(kbuf->image->start != pi->ehdr->e_entry)) { |
927 | kbuf->image->start -= sechdrs[i].sh_addr; |
928 | kbuf->image->start += kbuf->mem + offset; |
929 | } |
930 | |
931 | src = (void *)pi->ehdr + sechdrs[i].sh_offset; |
932 | dst = pi->purgatory_buf + offset; |
933 | memcpy(dst, src, sechdrs[i].sh_size); |
934 | |
935 | sechdrs[i].sh_addr = kbuf->mem + offset; |
936 | sechdrs[i].sh_offset = offset; |
937 | offset += sechdrs[i].sh_size; |
938 | } |
939 | |
940 | return 0; |
941 | } |
942 | |
943 | static int kexec_apply_relocations(struct kimage *image) |
944 | { |
945 | int i, ret; |
946 | struct purgatory_info *pi = &image->purgatory_info; |
947 | const Elf_Shdr *sechdrs; |
948 | |
949 | sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; |
950 | |
951 | for (i = 0; i < pi->ehdr->e_shnum; i++) { |
952 | const Elf_Shdr *relsec; |
953 | const Elf_Shdr *symtab; |
954 | Elf_Shdr *section; |
955 | |
956 | relsec = sechdrs + i; |
957 | |
958 | if (relsec->sh_type != SHT_RELA && |
959 | relsec->sh_type != SHT_REL) |
960 | continue; |
961 | |
962 | /* |
963 | * For section of type SHT_RELA/SHT_REL, |
964 | * ->sh_link contains section header index of associated |
965 | * symbol table. And ->sh_info contains section header |
966 | * index of section to which relocations apply. |
967 | */ |
968 | if (relsec->sh_info >= pi->ehdr->e_shnum || |
969 | relsec->sh_link >= pi->ehdr->e_shnum) |
970 | return -ENOEXEC; |
971 | |
972 | section = pi->sechdrs + relsec->sh_info; |
973 | symtab = sechdrs + relsec->sh_link; |
974 | |
975 | if (!(section->sh_flags & SHF_ALLOC)) |
976 | continue; |
977 | |
978 | /* |
979 | * symtab->sh_link contain section header index of associated |
980 | * string table. |
981 | */ |
982 | if (symtab->sh_link >= pi->ehdr->e_shnum) |
983 | /* Invalid section number? */ |
984 | continue; |
985 | |
986 | /* |
987 | * Respective architecture needs to provide support for applying |
988 | * relocations of type SHT_RELA/SHT_REL. |
989 | */ |
990 | if (relsec->sh_type == SHT_RELA) |
991 | ret = arch_kexec_apply_relocations_add(pi, section, |
992 | relsec, symtab); |
993 | else if (relsec->sh_type == SHT_REL) |
994 | ret = arch_kexec_apply_relocations(pi, section, |
995 | relsec, symtab); |
996 | if (ret) |
997 | return ret; |
998 | } |
999 | |
1000 | return 0; |
1001 | } |
1002 | |
1003 | /* |
1004 | * kexec_load_purgatory - Load and relocate the purgatory object. |
1005 | * @image: Image to add the purgatory to. |
1006 | * @kbuf: Memory parameters to use. |
1007 | * |
1008 | * Allocates the memory needed for image->purgatory_info.sechdrs and |
1009 | * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible |
1010 | * to free the memory after use. |
1011 | * |
1012 | * Return: 0 on success, negative errno on error. |
1013 | */ |
1014 | int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf) |
1015 | { |
1016 | struct purgatory_info *pi = &image->purgatory_info; |
1017 | int ret; |
1018 | |
1019 | if (kexec_purgatory_size <= 0) |
1020 | return -EINVAL; |
1021 | |
1022 | pi->ehdr = (const Elf_Ehdr *)kexec_purgatory; |
1023 | |
1024 | ret = kexec_purgatory_setup_kbuf(pi, kbuf); |
1025 | if (ret) |
1026 | return ret; |
1027 | |
1028 | ret = kexec_purgatory_setup_sechdrs(pi, kbuf); |
1029 | if (ret) |
1030 | goto out_free_kbuf; |
1031 | |
1032 | ret = kexec_apply_relocations(image); |
1033 | if (ret) |
1034 | goto out; |
1035 | |
1036 | return 0; |
1037 | out: |
1038 | vfree(addr: pi->sechdrs); |
1039 | pi->sechdrs = NULL; |
1040 | out_free_kbuf: |
1041 | vfree(addr: pi->purgatory_buf); |
1042 | pi->purgatory_buf = NULL; |
1043 | return ret; |
1044 | } |
1045 | |
1046 | /* |
1047 | * kexec_purgatory_find_symbol - find a symbol in the purgatory |
1048 | * @pi: Purgatory to search in. |
1049 | * @name: Name of the symbol. |
1050 | * |
1051 | * Return: pointer to symbol in read-only symtab on success, NULL on error. |
1052 | */ |
1053 | static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, |
1054 | const char *name) |
1055 | { |
1056 | const Elf_Shdr *sechdrs; |
1057 | const Elf_Ehdr *ehdr; |
1058 | const Elf_Sym *syms; |
1059 | const char *strtab; |
1060 | int i, k; |
1061 | |
1062 | if (!pi->ehdr) |
1063 | return NULL; |
1064 | |
1065 | ehdr = pi->ehdr; |
1066 | sechdrs = (void *)ehdr + ehdr->e_shoff; |
1067 | |
1068 | for (i = 0; i < ehdr->e_shnum; i++) { |
1069 | if (sechdrs[i].sh_type != SHT_SYMTAB) |
1070 | continue; |
1071 | |
1072 | if (sechdrs[i].sh_link >= ehdr->e_shnum) |
1073 | /* Invalid strtab section number */ |
1074 | continue; |
1075 | strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset; |
1076 | syms = (void *)ehdr + sechdrs[i].sh_offset; |
1077 | |
1078 | /* Go through symbols for a match */ |
1079 | for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { |
1080 | if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) |
1081 | continue; |
1082 | |
1083 | if (strcmp(strtab + syms[k].st_name, name) != 0) |
1084 | continue; |
1085 | |
1086 | if (syms[k].st_shndx == SHN_UNDEF || |
1087 | syms[k].st_shndx >= ehdr->e_shnum) { |
1088 | pr_debug("Symbol: %s has bad section index %d.\n" , |
1089 | name, syms[k].st_shndx); |
1090 | return NULL; |
1091 | } |
1092 | |
1093 | /* Found the symbol we are looking for */ |
1094 | return &syms[k]; |
1095 | } |
1096 | } |
1097 | |
1098 | return NULL; |
1099 | } |
1100 | |
1101 | void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) |
1102 | { |
1103 | struct purgatory_info *pi = &image->purgatory_info; |
1104 | const Elf_Sym *sym; |
1105 | Elf_Shdr *sechdr; |
1106 | |
1107 | sym = kexec_purgatory_find_symbol(pi, name); |
1108 | if (!sym) |
1109 | return ERR_PTR(error: -EINVAL); |
1110 | |
1111 | sechdr = &pi->sechdrs[sym->st_shndx]; |
1112 | |
1113 | /* |
1114 | * Returns the address where symbol will finally be loaded after |
1115 | * kexec_load_segment() |
1116 | */ |
1117 | return (void *)(sechdr->sh_addr + sym->st_value); |
1118 | } |
1119 | |
1120 | /* |
1121 | * Get or set value of a symbol. If "get_value" is true, symbol value is |
1122 | * returned in buf otherwise symbol value is set based on value in buf. |
1123 | */ |
1124 | int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, |
1125 | void *buf, unsigned int size, bool get_value) |
1126 | { |
1127 | struct purgatory_info *pi = &image->purgatory_info; |
1128 | const Elf_Sym *sym; |
1129 | Elf_Shdr *sec; |
1130 | char *sym_buf; |
1131 | |
1132 | sym = kexec_purgatory_find_symbol(pi, name); |
1133 | if (!sym) |
1134 | return -EINVAL; |
1135 | |
1136 | if (sym->st_size != size) { |
1137 | pr_err("symbol %s size mismatch: expected %lu actual %u\n" , |
1138 | name, (unsigned long)sym->st_size, size); |
1139 | return -EINVAL; |
1140 | } |
1141 | |
1142 | sec = pi->sechdrs + sym->st_shndx; |
1143 | |
1144 | if (sec->sh_type == SHT_NOBITS) { |
1145 | pr_err("symbol %s is in a bss section. Cannot %s\n" , name, |
1146 | get_value ? "get" : "set" ); |
1147 | return -EINVAL; |
1148 | } |
1149 | |
1150 | sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value; |
1151 | |
1152 | if (get_value) |
1153 | memcpy((void *)buf, sym_buf, size); |
1154 | else |
1155 | memcpy((void *)sym_buf, buf, size); |
1156 | |
1157 | return 0; |
1158 | } |
1159 | #endif /* CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY */ |
1160 | |