| 1 | |
| 2 | |
| 3 | * kexec: kexec_file_load system call
|
| 4 | *
|
| 5 | * Copyright (C) 2014 Red Hat Inc.
|
| 6 | * Authors:
|
| 7 | * Vivek Goyal <[email protected]>
|
| 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/sha2.h>
|
| 23 | #include <linux/elf.h>
|
| 24 | #include <linux/elfcore.h>
|
| 25 | #include <linux/kernel.h>
|
| 26 | #include <linux/kernel_read_file.h>
|
| 27 | #include <linux/syscalls.h>
|
| 28 | #include <linux/vmalloc.h>
|
| 29 | #include <linux/dma-map-ops.h>
|
| 30 | #include <linux/kexec_handover.h>
|
| 31 | #include "kexec_internal.h"
|
| 32 |
|
| 33 | #ifdef CONFIG_KEXEC_SIG
|
| 34 | static bool sig_enforce = IS_ENABLED(CONFIG_KEXEC_SIG_FORCE);
|
| 35 |
|
| 36 | void set_kexec_sig_enforced(void)
|
| 37 | {
|
| 38 | sig_enforce = true;
|
| 39 | }
|
| 40 | #endif
|
| 41 |
|
| 42 | #ifdef CONFIG_IMA_KEXEC
|
| 43 | static bool check_ima_segment_index(struct kimage *image, int i)
|
| 44 | {
|
| 45 | if (image->is_ima_segment_index_set && i == image->ima_segment_index)
|
| 46 | return true;
|
| 47 | else
|
| 48 | return false;
|
| 49 | }
|
| 50 | #else
|
| 51 | static bool check_ima_segment_index(struct kimage *image, int i)
|
| 52 | {
|
| 53 | return false;
|
| 54 | }
|
| 55 | #endif
|
| 56 |
|
| 57 | static int kexec_calculate_store_digests(struct kimage *image);
|
| 58 |
|
| 59 |
|
| 60 | #define KEXEC_FILE_SIZE_MAX min_t(s64, 4LL << 30, SSIZE_MAX)
|
| 61 |
|
| 62 | |
| 63 | * Currently this is the only default function that is exported as some
|
| 64 | * architectures need it to do additional handlings.
|
| 65 | * In the future, other default functions may be exported too if required.
|
| 66 |
|
| 67 | int kexec_image_probe_default(struct kimage *image, void *buf,
|
| 68 | unsigned long buf_len)
|
| 69 | {
|
| 70 | const struct kexec_file_ops * const *fops;
|
| 71 | int ret = -ENOEXEC;
|
| 72 |
|
| 73 | for (fops = &kexec_file_loaders[0]; *fops && (*fops)->probe; ++fops) {
|
| 74 | ret = (*fops)->probe(buf, buf_len);
|
| 75 | if (!ret) {
|
| 76 | image->fops = *fops;
|
| 77 | return ret;
|
| 78 | }
|
| 79 | }
|
| 80 |
|
| 81 | return ret;
|
| 82 | }
|
| 83 |
|
| 84 | static void *kexec_image_load_default(struct kimage *image)
|
| 85 | {
|
| 86 | if (!image->fops || !image->fops->load)
|
| 87 | return ERR_PTR(-ENOEXEC);
|
| 88 |
|
| 89 | return image->fops->load(image, image->kernel_buf,
|
| 90 | image->kernel_buf_len, image->initrd_buf,
|
| 91 | image->initrd_buf_len, image->cmdline_buf,
|
| 92 | image->cmdline_buf_len);
|
| 93 | }
|
| 94 |
|
| 95 | int kexec_image_post_load_cleanup_default(struct kimage *image)
|
| 96 | {
|
| 97 | if (!image->fops || !image->fops->cleanup)
|
| 98 | return 0;
|
| 99 |
|
| 100 | return image->fops->cleanup(image->image_loader_data);
|
| 101 | }
|
| 102 |
|
| 103 | |
| 104 | * Free up memory used by kernel, initrd, and command line. This is temporary
|
| 105 | * memory allocation which is not needed any more after these buffers have
|
| 106 | * been loaded into separate segments and have been copied elsewhere.
|
| 107 |
|
| 108 | void kimage_file_post_load_cleanup(struct kimage *image)
|
| 109 | {
|
| 110 | struct purgatory_info *pi = &image->purgatory_info;
|
| 111 |
|
| 112 | vfree(image->kernel_buf);
|
| 113 | image->kernel_buf = NULL;
|
| 114 |
|
| 115 | vfree(image->initrd_buf);
|
| 116 | image->initrd_buf = NULL;
|
| 117 |
|
| 118 | kfree(image->cmdline_buf);
|
| 119 | image->cmdline_buf = NULL;
|
| 120 |
|
| 121 | vfree(pi->purgatory_buf);
|
| 122 | pi->purgatory_buf = NULL;
|
| 123 |
|
| 124 | vfree(pi->sechdrs);
|
| 125 | pi->sechdrs = NULL;
|
| 126 |
|
| 127 | #ifdef CONFIG_IMA_KEXEC
|
| 128 | vfree(image->ima_buffer);
|
| 129 | image->ima_buffer = NULL;
|
| 130 | #endif
|
| 131 |
|
| 132 |
|
| 133 | arch_kimage_file_post_load_cleanup(image);
|
| 134 |
|
| 135 | |
| 136 | * Above call should have called into bootloader to free up
|
| 137 | * any data stored in kimage->image_loader_data. It should
|
| 138 | * be ok now to free it up.
|
| 139 |
|
| 140 | kfree(image->image_loader_data);
|
| 141 | image->image_loader_data = NULL;
|
| 142 |
|
| 143 | kexec_file_dbg_print = false;
|
| 144 | }
|
| 145 |
|
| 146 | #ifdef CONFIG_KEXEC_SIG
|
| 147 | #ifdef CONFIG_SIGNED_PE_FILE_VERIFICATION
|
| 148 | int kexec_kernel_verify_pe_sig(const char *kernel, unsigned long kernel_len)
|
| 149 | {
|
| 150 | int ret;
|
| 151 |
|
| 152 | ret = verify_pefile_signature(kernel, kernel_len,
|
| 153 | VERIFY_USE_SECONDARY_KEYRING,
|
| 154 | VERIFYING_KEXEC_PE_SIGNATURE);
|
| 155 | if (ret == -ENOKEY && IS_ENABLED(CONFIG_INTEGRITY_PLATFORM_KEYRING)) {
|
| 156 | ret = verify_pefile_signature(kernel, kernel_len,
|
| 157 | VERIFY_USE_PLATFORM_KEYRING,
|
| 158 | VERIFYING_KEXEC_PE_SIGNATURE);
|
| 159 | }
|
| 160 | return ret;
|
| 161 | }
|
| 162 | #endif
|
| 163 |
|
| 164 | static int kexec_image_verify_sig(struct kimage *image, void *buf,
|
| 165 | unsigned long buf_len)
|
| 166 | {
|
| 167 | if (!image->fops || !image->fops->verify_sig) {
|
| 168 | pr_debug("kernel loader does not support signature verification.\n");
|
| 169 | return -EKEYREJECTED;
|
| 170 | }
|
| 171 |
|
| 172 | return image->fops->verify_sig(buf, buf_len);
|
| 173 | }
|
| 174 |
|
| 175 | static int
|
| 176 | kimage_validate_signature(struct kimage *image)
|
| 177 | {
|
| 178 | int ret;
|
| 179 |
|
| 180 | ret = kexec_image_verify_sig(image, image->kernel_buf,
|
| 181 | image->kernel_buf_len);
|
| 182 | if (ret) {
|
| 183 |
|
| 184 | if (sig_enforce) {
|
| 185 | pr_notice("Enforced kernel signature verification failed (%d).\n", ret);
|
| 186 | return ret;
|
| 187 | }
|
| 188 |
|
| 189 | |
| 190 | * If IMA is guaranteed to appraise a signature on the kexec
|
| 191 | * image, permit it even if the kernel is otherwise locked
|
| 192 | * down.
|
| 193 |
|
| 194 | if (!ima_appraise_signature(READING_KEXEC_IMAGE) &&
|
| 195 | security_locked_down(LOCKDOWN_KEXEC))
|
| 196 | return -EPERM;
|
| 197 |
|
| 198 | pr_debug("kernel signature verification failed (%d).\n", ret);
|
| 199 | }
|
| 200 |
|
| 201 | return 0;
|
| 202 | }
|
| 203 | #endif
|
| 204 |
|
| 205 | static int kexec_post_load(struct kimage *image, unsigned long flags)
|
| 206 | {
|
| 207 | #ifdef CONFIG_IMA_KEXEC
|
| 208 | if (!(flags & KEXEC_FILE_ON_CRASH))
|
| 209 | ima_kexec_post_load(image);
|
| 210 | #endif
|
| 211 | return machine_kexec_post_load(image);
|
| 212 | }
|
| 213 |
|
| 214 | |
| 215 | * In file mode list of segments is prepared by kernel. Copy relevant
|
| 216 | * data from user space, do error checking, prepare segment list
|
| 217 |
|
| 218 | static int
|
| 219 | kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
|
| 220 | const char __user *cmdline_ptr,
|
| 221 | unsigned long cmdline_len, unsigned flags)
|
| 222 | {
|
| 223 | ssize_t ret;
|
| 224 | void *ldata;
|
| 225 |
|
| 226 | ret = kernel_read_file_from_fd(kernel_fd, 0, &image->kernel_buf,
|
| 227 | KEXEC_FILE_SIZE_MAX, NULL,
|
| 228 | READING_KEXEC_IMAGE);
|
| 229 | if (ret < 0)
|
| 230 | return ret;
|
| 231 | image->kernel_buf_len = ret;
|
| 232 | kexec_dprintk("kernel: %p kernel_size: %#lx\n",
|
| 233 | image->kernel_buf, image->kernel_buf_len);
|
| 234 |
|
| 235 |
|
| 236 | ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
|
| 237 | image->kernel_buf_len);
|
| 238 | if (ret)
|
| 239 | goto out;
|
| 240 |
|
| 241 | #ifdef CONFIG_KEXEC_SIG
|
| 242 | ret = kimage_validate_signature(image);
|
| 243 |
|
| 244 | if (ret)
|
| 245 | goto out;
|
| 246 | #endif
|
| 247 |
|
| 248 | if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
|
| 249 | ret = kernel_read_file_from_fd(initrd_fd, 0, &image->initrd_buf,
|
| 250 | KEXEC_FILE_SIZE_MAX, NULL,
|
| 251 | READING_KEXEC_INITRAMFS);
|
| 252 | if (ret < 0)
|
| 253 | goto out;
|
| 254 | image->initrd_buf_len = ret;
|
| 255 | ret = 0;
|
| 256 | }
|
| 257 |
|
| 258 | image->no_cma = !!(flags & KEXEC_FILE_NO_CMA);
|
| 259 | image->force_dtb = flags & KEXEC_FILE_FORCE_DTB;
|
| 260 |
|
| 261 | if (cmdline_len) {
|
| 262 | image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
|
| 263 | if (IS_ERR(image->cmdline_buf)) {
|
| 264 | ret = PTR_ERR(image->cmdline_buf);
|
| 265 | image->cmdline_buf = NULL;
|
| 266 | goto out;
|
| 267 | }
|
| 268 |
|
| 269 | image->cmdline_buf_len = cmdline_len;
|
| 270 |
|
| 271 |
|
| 272 | if (image->cmdline_buf[cmdline_len - 1] != '\0') {
|
| 273 | ret = -EINVAL;
|
| 274 | goto out;
|
| 275 | }
|
| 276 |
|
| 277 | ima_kexec_cmdline(kernel_fd, image->cmdline_buf,
|
| 278 | image->cmdline_buf_len - 1);
|
| 279 | }
|
| 280 |
|
| 281 |
|
| 282 | ima_add_kexec_buffer(image);
|
| 283 |
|
| 284 |
|
| 285 | ret = kho_fill_kimage(image);
|
| 286 | if (ret)
|
| 287 | goto out;
|
| 288 |
|
| 289 |
|
| 290 | ldata = kexec_image_load_default(image);
|
| 291 |
|
| 292 | if (IS_ERR(ldata)) {
|
| 293 | ret = PTR_ERR(ldata);
|
| 294 | goto out;
|
| 295 | }
|
| 296 |
|
| 297 | image->image_loader_data = ldata;
|
| 298 | out:
|
| 299 |
|
| 300 | if (ret)
|
| 301 | kimage_file_post_load_cleanup(image);
|
| 302 | return ret;
|
| 303 | }
|
| 304 |
|
| 305 | static int
|
| 306 | kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
|
| 307 | int initrd_fd, const char __user *cmdline_ptr,
|
| 308 | unsigned long cmdline_len, unsigned long flags)
|
| 309 | {
|
| 310 | int ret;
|
| 311 | struct kimage *image;
|
| 312 | bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
|
| 313 |
|
| 314 | image = do_kimage_alloc_init();
|
| 315 | if (!image)
|
| 316 | return -ENOMEM;
|
| 317 |
|
| 318 | kexec_file_dbg_print = !!(flags & KEXEC_FILE_DEBUG);
|
| 319 | image->file_mode = 1;
|
| 320 |
|
| 321 | #ifdef CONFIG_CRASH_DUMP
|
| 322 | if (kexec_on_panic) {
|
| 323 |
|
| 324 | image->control_page = crashk_res.start;
|
| 325 | image->type = KEXEC_TYPE_CRASH;
|
| 326 | }
|
| 327 | #endif
|
| 328 |
|
| 329 | ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
|
| 330 | cmdline_ptr, cmdline_len, flags);
|
| 331 | if (ret)
|
| 332 | goto out_free_image;
|
| 333 |
|
| 334 | ret = sanity_check_segment_list(image);
|
| 335 | if (ret)
|
| 336 | goto out_free_post_load_bufs;
|
| 337 |
|
| 338 | ret = -ENOMEM;
|
| 339 | image->control_code_page = kimage_alloc_control_pages(image,
|
| 340 | get_order(KEXEC_CONTROL_PAGE_SIZE));
|
| 341 | if (!image->control_code_page) {
|
| 342 | pr_err("Could not allocate control_code_buffer\n");
|
| 343 | goto out_free_post_load_bufs;
|
| 344 | }
|
| 345 |
|
| 346 | if (!kexec_on_panic) {
|
| 347 | image->swap_page = kimage_alloc_control_pages(image, 0);
|
| 348 | if (!image->swap_page) {
|
| 349 | pr_err("Could not allocate swap buffer\n");
|
| 350 | goto out_free_control_pages;
|
| 351 | }
|
| 352 | }
|
| 353 |
|
| 354 | *rimage = image;
|
| 355 | return 0;
|
| 356 | out_free_control_pages:
|
| 357 | kimage_free_page_list(&image->control_pages);
|
| 358 | out_free_post_load_bufs:
|
| 359 | kimage_file_post_load_cleanup(image);
|
| 360 | out_free_image:
|
| 361 | kfree(image);
|
| 362 | return ret;
|
| 363 | }
|
| 364 |
|
| 365 | SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
|
| 366 | unsigned long, cmdline_len, const char __user *, cmdline_ptr,
|
| 367 | unsigned long, flags)
|
| 368 | {
|
| 369 | int image_type = (flags & KEXEC_FILE_ON_CRASH) ?
|
| 370 | KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT;
|
| 371 | struct kimage **dest_image, *image;
|
| 372 | int ret = 0, i;
|
| 373 |
|
| 374 |
|
| 375 | if (!kexec_load_permitted(image_type))
|
| 376 | return -EPERM;
|
| 377 |
|
| 378 |
|
| 379 | if (flags != (flags & KEXEC_FILE_FLAGS))
|
| 380 | return -EINVAL;
|
| 381 |
|
| 382 | image = NULL;
|
| 383 |
|
| 384 | if (!kexec_trylock())
|
| 385 | return -EBUSY;
|
| 386 |
|
| 387 | #ifdef CONFIG_CRASH_DUMP
|
| 388 | if (image_type == KEXEC_TYPE_CRASH) {
|
| 389 | dest_image = &kexec_crash_image;
|
| 390 | if (kexec_crash_image)
|
| 391 | arch_kexec_unprotect_crashkres();
|
| 392 | } else
|
| 393 | #endif
|
| 394 | dest_image = &kexec_image;
|
| 395 |
|
| 396 | if (flags & KEXEC_FILE_UNLOAD)
|
| 397 | goto exchange;
|
| 398 |
|
| 399 | |
| 400 | * In case of crash, new kernel gets loaded in reserved region. It is
|
| 401 | * same memory where old crash kernel might be loaded. Free any
|
| 402 | * current crash dump kernel before we corrupt it.
|
| 403 |
|
| 404 | if (flags & KEXEC_FILE_ON_CRASH)
|
| 405 | kimage_free(xchg(&kexec_crash_image, NULL));
|
| 406 |
|
| 407 | ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
|
| 408 | cmdline_len, flags);
|
| 409 | if (ret)
|
| 410 | goto out;
|
| 411 |
|
| 412 | #ifdef CONFIG_CRASH_HOTPLUG
|
| 413 | if ((flags & KEXEC_FILE_ON_CRASH) && arch_crash_hotplug_support(image, flags))
|
| 414 | image->hotplug_support = 1;
|
| 415 | #endif
|
| 416 |
|
| 417 | ret = machine_kexec_prepare(image);
|
| 418 | if (ret)
|
| 419 | goto out;
|
| 420 |
|
| 421 | |
| 422 | * Some architecture(like S390) may touch the crash memory before
|
| 423 | * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
|
| 424 |
|
| 425 | ret = kimage_crash_copy_vmcoreinfo(image);
|
| 426 | if (ret)
|
| 427 | goto out;
|
| 428 |
|
| 429 | ret = kexec_calculate_store_digests(image);
|
| 430 | if (ret)
|
| 431 | goto out;
|
| 432 |
|
| 433 | kexec_dprintk("nr_segments = %lu\n", image->nr_segments);
|
| 434 | for (i = 0; i < image->nr_segments; i++) {
|
| 435 | struct kexec_segment *ksegment;
|
| 436 |
|
| 437 | ksegment = &image->segment[i];
|
| 438 | kexec_dprintk("segment[%d]: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
|
| 439 | i, ksegment->buf, ksegment->bufsz, ksegment->mem,
|
| 440 | ksegment->memsz);
|
| 441 |
|
| 442 | ret = kimage_load_segment(image, i);
|
| 443 | if (ret)
|
| 444 | goto out;
|
| 445 | }
|
| 446 |
|
| 447 | kimage_terminate(image);
|
| 448 |
|
| 449 | ret = kexec_post_load(image, flags);
|
| 450 | if (ret)
|
| 451 | goto out;
|
| 452 |
|
| 453 | kexec_dprintk("kexec_file_load: type:%u, start:0x%lx head:0x%lx flags:0x%lx\n",
|
| 454 | image->type, image->start, image->head, flags);
|
| 455 | |
| 456 | * Free up any temporary buffers allocated which are not needed
|
| 457 | * after image has been loaded
|
| 458 |
|
| 459 | kimage_file_post_load_cleanup(image);
|
| 460 | exchange:
|
| 461 | image = xchg(dest_image, image);
|
| 462 | out:
|
| 463 | #ifdef CONFIG_CRASH_DUMP
|
| 464 | if ((flags & KEXEC_FILE_ON_CRASH) && kexec_crash_image)
|
| 465 | arch_kexec_protect_crashkres();
|
| 466 | #endif
|
| 467 |
|
| 468 | kexec_unlock();
|
| 469 | kimage_free(image);
|
| 470 | return ret;
|
| 471 | }
|
| 472 |
|
| 473 | static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
|
| 474 | struct kexec_buf *kbuf)
|
| 475 | {
|
| 476 | struct kimage *image = kbuf->image;
|
| 477 | unsigned long temp_start, temp_end;
|
| 478 |
|
| 479 | temp_end = min(end, kbuf->buf_max);
|
| 480 | temp_start = temp_end - kbuf->memsz + 1;
|
| 481 | kexec_random_range_start(temp_start, temp_end, kbuf, &temp_start);
|
| 482 |
|
| 483 | do {
|
| 484 |
|
| 485 | temp_start = ALIGN_DOWN(temp_start, kbuf->buf_align);
|
| 486 |
|
| 487 | if (temp_start < start || temp_start < kbuf->buf_min)
|
| 488 | return 0;
|
| 489 |
|
| 490 | temp_end = temp_start + kbuf->memsz - 1;
|
| 491 |
|
| 492 | |
| 493 | * Make sure this does not conflict with any of existing
|
| 494 | * segments
|
| 495 |
|
| 496 | if (kimage_is_destination_range(image, temp_start, temp_end)) {
|
| 497 | temp_start = temp_start - PAGE_SIZE;
|
| 498 | continue;
|
| 499 | }
|
| 500 |
|
| 501 |
|
| 502 | if (arch_check_excluded_range(image, temp_start, temp_end)) {
|
| 503 | temp_start = temp_start - PAGE_SIZE;
|
| 504 | continue;
|
| 505 | }
|
| 506 |
|
| 507 |
|
| 508 | break;
|
| 509 | } while (1);
|
| 510 |
|
| 511 |
|
| 512 | kbuf->mem = temp_start;
|
| 513 |
|
| 514 |
|
| 515 | return 1;
|
| 516 | }
|
| 517 |
|
| 518 | static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
|
| 519 | struct kexec_buf *kbuf)
|
| 520 | {
|
| 521 | struct kimage *image = kbuf->image;
|
| 522 | unsigned long temp_start, temp_end;
|
| 523 |
|
| 524 | temp_start = max(start, kbuf->buf_min);
|
| 525 |
|
| 526 | kexec_random_range_start(temp_start, end, kbuf, &temp_start);
|
| 527 |
|
| 528 | do {
|
| 529 | temp_start = ALIGN(temp_start, kbuf->buf_align);
|
| 530 | temp_end = temp_start + kbuf->memsz - 1;
|
| 531 |
|
| 532 | if (temp_end > end || temp_end > kbuf->buf_max)
|
| 533 | return 0;
|
| 534 | |
| 535 | * Make sure this does not conflict with any of existing
|
| 536 | * segments
|
| 537 |
|
| 538 | if (kimage_is_destination_range(image, temp_start, temp_end)) {
|
| 539 | temp_start = temp_start + PAGE_SIZE;
|
| 540 | continue;
|
| 541 | }
|
| 542 |
|
| 543 |
|
| 544 | if (arch_check_excluded_range(image, temp_start, temp_end)) {
|
| 545 | temp_start = temp_start + PAGE_SIZE;
|
| 546 | continue;
|
| 547 | }
|
| 548 |
|
| 549 |
|
| 550 | break;
|
| 551 | } while (1);
|
| 552 |
|
| 553 |
|
| 554 | kbuf->mem = temp_start;
|
| 555 |
|
| 556 |
|
| 557 | return 1;
|
| 558 | }
|
| 559 |
|
| 560 | static int locate_mem_hole_callback(struct resource *res, void *arg)
|
| 561 | {
|
| 562 | struct kexec_buf *kbuf = (struct kexec_buf *)arg;
|
| 563 | u64 start = res->start, end = res->end;
|
| 564 | unsigned long sz = end - start + 1;
|
| 565 |
|
| 566 |
|
| 567 |
|
| 568 |
|
| 569 | if (res->flags & IORESOURCE_SYSRAM_DRIVER_MANAGED)
|
| 570 | return 0;
|
| 571 |
|
| 572 | if (sz < kbuf->memsz)
|
| 573 | return 0;
|
| 574 |
|
| 575 | if (end < kbuf->buf_min || start > kbuf->buf_max)
|
| 576 | return 0;
|
| 577 |
|
| 578 | |
| 579 | * Allocate memory top down with-in ram range. Otherwise bottom up
|
| 580 | * allocation.
|
| 581 |
|
| 582 | if (kbuf->top_down)
|
| 583 | return locate_mem_hole_top_down(start, end, kbuf);
|
| 584 | return locate_mem_hole_bottom_up(start, end, kbuf);
|
| 585 | }
|
| 586 |
|
| 587 | #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
|
| 588 | static int kexec_walk_memblock(struct kexec_buf *kbuf,
|
| 589 | int (*func)(struct resource *, void *))
|
| 590 | {
|
| 591 | int ret = 0;
|
| 592 | u64 i;
|
| 593 | phys_addr_t mstart, mend;
|
| 594 | struct resource res = { };
|
| 595 |
|
| 596 | #ifdef CONFIG_CRASH_DUMP
|
| 597 | if (kbuf->image->type == KEXEC_TYPE_CRASH)
|
| 598 | return func(&crashk_res, kbuf);
|
| 599 | #endif
|
| 600 |
|
| 601 | |
| 602 | * Using MEMBLOCK_NONE will properly skip MEMBLOCK_DRIVER_MANAGED. See
|
| 603 | * IORESOURCE_SYSRAM_DRIVER_MANAGED handling in
|
| 604 | * locate_mem_hole_callback().
|
| 605 |
|
| 606 | if (kbuf->top_down) {
|
| 607 | for_each_free_mem_range_reverse(i, NUMA_NO_NODE, MEMBLOCK_NONE,
|
| 608 | &mstart, &mend, NULL) {
|
| 609 | |
| 610 | * In memblock, end points to the first byte after the
|
| 611 | * range while in kexec, end points to the last byte
|
| 612 | * in the range.
|
| 613 |
|
| 614 | res.start = mstart;
|
| 615 | res.end = mend - 1;
|
| 616 | ret = func(&res, kbuf);
|
| 617 | if (ret)
|
| 618 | break;
|
| 619 | }
|
| 620 | } else {
|
| 621 | for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE,
|
| 622 | &mstart, &mend, NULL) {
|
| 623 | |
| 624 | * In memblock, end points to the first byte after the
|
| 625 | * range while in kexec, end points to the last byte
|
| 626 | * in the range.
|
| 627 |
|
| 628 | res.start = mstart;
|
| 629 | res.end = mend - 1;
|
| 630 | ret = func(&res, kbuf);
|
| 631 | if (ret)
|
| 632 | break;
|
| 633 | }
|
| 634 | }
|
| 635 |
|
| 636 | return ret;
|
| 637 | }
|
| 638 | #else
|
| 639 | static int kexec_walk_memblock(struct kexec_buf *kbuf,
|
| 640 | int (*func)(struct resource *, void *))
|
| 641 | {
|
| 642 | return 0;
|
| 643 | }
|
| 644 | #endif
|
| 645 |
|
| 646 | |
| 647 | * kexec_walk_resources - call func(data) on free memory regions
|
| 648 | * @kbuf: Context info for the search. Also passed to @func.
|
| 649 | * @func: Function to call for each memory region.
|
| 650 | *
|
| 651 | * Return: The memory walk will stop when func returns a non-zero value
|
| 652 | * and that value will be returned. If all free regions are visited without
|
| 653 | * func returning non-zero, then zero will be returned.
|
| 654 |
|
| 655 | static int kexec_walk_resources(struct kexec_buf *kbuf,
|
| 656 | int (*func)(struct resource *, void *))
|
| 657 | {
|
| 658 | #ifdef CONFIG_CRASH_DUMP
|
| 659 | if (kbuf->image->type == KEXEC_TYPE_CRASH)
|
| 660 | return walk_iomem_res_desc(crashk_res.desc,
|
| 661 | IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY,
|
| 662 | crashk_res.start, crashk_res.end,
|
| 663 | kbuf, func);
|
| 664 | #endif
|
| 665 | if (kbuf->top_down)
|
| 666 | return walk_system_ram_res_rev(0, ULONG_MAX, kbuf, func);
|
| 667 | else
|
| 668 | return walk_system_ram_res(0, ULONG_MAX, kbuf, func);
|
| 669 | }
|
| 670 |
|
| 671 | static int kexec_alloc_contig(struct kexec_buf *kbuf)
|
| 672 | {
|
| 673 | size_t nr_pages = kbuf->memsz >> PAGE_SHIFT;
|
| 674 | unsigned long mem;
|
| 675 | struct page *p;
|
| 676 |
|
| 677 |
|
| 678 | if (kbuf->image->no_cma)
|
| 679 | return -EPERM;
|
| 680 |
|
| 681 |
|
| 682 | if (kbuf->image->type == KEXEC_TYPE_CRASH)
|
| 683 | return -EPERM;
|
| 684 |
|
| 685 | p = dma_alloc_from_contiguous(NULL, nr_pages, get_order(kbuf->buf_align), true);
|
| 686 | if (!p)
|
| 687 | return -ENOMEM;
|
| 688 |
|
| 689 | pr_debug("allocated %zu DMA pages at 0x%lx", nr_pages, page_to_boot_pfn(p));
|
| 690 |
|
| 691 | mem = page_to_boot_pfn(p) << PAGE_SHIFT;
|
| 692 |
|
| 693 | if (kimage_is_destination_range(kbuf->image, mem, mem + kbuf->memsz)) {
|
| 694 |
|
| 695 | pr_debug("CMA overlaps existing mem: 0x%lx+0x%lx\n", mem, kbuf->memsz);
|
| 696 | dma_release_from_contiguous(NULL, p, nr_pages);
|
| 697 | return -EBUSY;
|
| 698 | }
|
| 699 |
|
| 700 | kbuf->mem = page_to_boot_pfn(p) << PAGE_SHIFT;
|
| 701 | kbuf->cma = p;
|
| 702 |
|
| 703 | arch_kexec_post_alloc_pages(page_address(p), (int)nr_pages, 0);
|
| 704 |
|
| 705 | return 0;
|
| 706 | }
|
| 707 |
|
| 708 | |
| 709 | * kexec_locate_mem_hole - find free memory for the purgatory or the next kernel
|
| 710 | * @kbuf: Parameters for the memory search.
|
| 711 | *
|
| 712 | * On success, kbuf->mem will have the start address of the memory region found.
|
| 713 | *
|
| 714 | * Return: 0 on success, negative errno on error.
|
| 715 |
|
| 716 | int kexec_locate_mem_hole(struct kexec_buf *kbuf)
|
| 717 | {
|
| 718 | int ret;
|
| 719 |
|
| 720 |
|
| 721 | if (kbuf->mem != KEXEC_BUF_MEM_UNKNOWN)
|
| 722 | return 0;
|
| 723 |
|
| 724 | |
| 725 | * If KHO is active, only use KHO scratch memory. All other memory
|
| 726 | * could potentially be handed over.
|
| 727 |
|
| 728 | ret = kho_locate_mem_hole(kbuf, locate_mem_hole_callback);
|
| 729 | if (ret <= 0)
|
| 730 | return ret;
|
| 731 |
|
| 732 | |
| 733 | * Try to find a free physically contiguous block of memory first. With that, we
|
| 734 | * can avoid any copying at kexec time.
|
| 735 |
|
| 736 | if (!kexec_alloc_contig(kbuf))
|
| 737 | return 0;
|
| 738 |
|
| 739 | if (!IS_ENABLED(CONFIG_ARCH_KEEP_MEMBLOCK))
|
| 740 | ret = kexec_walk_resources(kbuf, locate_mem_hole_callback);
|
| 741 | else
|
| 742 | ret = kexec_walk_memblock(kbuf, locate_mem_hole_callback);
|
| 743 |
|
| 744 | return ret == 1 ? 0 : -EADDRNOTAVAIL;
|
| 745 | }
|
| 746 |
|
| 747 | |
| 748 | * kexec_add_buffer - place a buffer in a kexec segment
|
| 749 | * @kbuf: Buffer contents and memory parameters.
|
| 750 | *
|
| 751 | * This function assumes that kexec_lock is held.
|
| 752 | * On successful return, @kbuf->mem will have the physical address of
|
| 753 | * the buffer in memory.
|
| 754 | *
|
| 755 | * Return: 0 on success, negative errno on error.
|
| 756 |
|
| 757 | int kexec_add_buffer(struct kexec_buf *kbuf)
|
| 758 | {
|
| 759 | struct kexec_segment *ksegment;
|
| 760 | int ret;
|
| 761 |
|
| 762 |
|
| 763 | if (!kbuf->image->file_mode)
|
| 764 | return -EINVAL;
|
| 765 |
|
| 766 | if (kbuf->image->nr_segments >= KEXEC_SEGMENT_MAX)
|
| 767 | return -EINVAL;
|
| 768 |
|
| 769 | |
| 770 | * Make sure we are not trying to add buffer after allocating
|
| 771 | * control pages. All segments need to be placed first before
|
| 772 | * any control pages are allocated. As control page allocation
|
| 773 | * logic goes through list of segments to make sure there are
|
| 774 | * no destination overlaps.
|
| 775 |
|
| 776 | if (!list_empty(&kbuf->image->control_pages)) {
|
| 777 | WARN_ON(1);
|
| 778 | return -EINVAL;
|
| 779 | }
|
| 780 |
|
| 781 |
|
| 782 | kbuf->memsz = ALIGN(kbuf->memsz, PAGE_SIZE);
|
| 783 | kbuf->buf_align = max(kbuf->buf_align, PAGE_SIZE);
|
| 784 | kbuf->cma = NULL;
|
| 785 |
|
| 786 |
|
| 787 | ret = arch_kexec_locate_mem_hole(kbuf);
|
| 788 | if (ret)
|
| 789 | return ret;
|
| 790 |
|
| 791 |
|
| 792 | ksegment = &kbuf->image->segment[kbuf->image->nr_segments];
|
| 793 | ksegment->kbuf = kbuf->buffer;
|
| 794 | ksegment->bufsz = kbuf->bufsz;
|
| 795 | ksegment->mem = kbuf->mem;
|
| 796 | ksegment->memsz = kbuf->memsz;
|
| 797 | kbuf->image->segment_cma[kbuf->image->nr_segments] = kbuf->cma;
|
| 798 | kbuf->image->nr_segments++;
|
| 799 | return 0;
|
| 800 | }
|
| 801 |
|
| 802 | static bool kexec_only_cma_segments(struct kimage *image)
|
| 803 | {
|
| 804 | for (int i = 0; i < image->nr_segments; i++) {
|
| 805 | if (!image->segment_cma[i])
|
| 806 | return false;
|
| 807 | }
|
| 808 |
|
| 809 | return true;
|
| 810 | }
|
| 811 |
|
| 812 |
|
| 813 | static int kexec_calculate_store_digests(struct kimage *image)
|
| 814 | {
|
| 815 | struct sha256_ctx sctx;
|
| 816 | int ret = 0, i, j, zero_buf_sz, sha_region_sz;
|
| 817 | size_t nullsz;
|
| 818 | u8 digest[SHA256_DIGEST_SIZE];
|
| 819 | void *zero_buf;
|
| 820 | struct kexec_sha_region *sha_regions;
|
| 821 | struct purgatory_info *pi = &image->purgatory_info;
|
| 822 |
|
| 823 | if (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY))
|
| 824 | return 0;
|
| 825 |
|
| 826 | zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
|
| 827 | zero_buf_sz = PAGE_SIZE;
|
| 828 |
|
| 829 | sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
|
| 830 | sha_regions = vzalloc(sha_region_sz);
|
| 831 | if (!sha_regions)
|
| 832 | return -ENOMEM;
|
| 833 |
|
| 834 | sha256_init(&sctx);
|
| 835 |
|
| 836 | |
| 837 | * If KHO is enabled, the destinations are located in KHO scratch.
|
| 838 | * KHO scratch can only contain early boot allocations and movable
|
| 839 | * allocations. That means there is no risk of memory corruption by
|
| 840 | * uncancelled DMA.
|
| 841 | *
|
| 842 | * If all segments were loaded into contiguous memory, there will be no
|
| 843 | * relocations at all, so also no risk of corruption.
|
| 844 |
|
| 845 | if (image->type != KEXEC_TYPE_CRASH &&
|
| 846 | (kho_is_enabled() || kexec_only_cma_segments(image))) {
|
| 847 | pr_debug("disabling checksum verification in purgatory\n");
|
| 848 | goto skip_checksum;
|
| 849 | }
|
| 850 |
|
| 851 | for (j = i = 0; i < image->nr_segments; i++) {
|
| 852 | struct kexec_segment *ksegment;
|
| 853 |
|
| 854 | #ifdef CONFIG_CRASH_HOTPLUG
|
| 855 |
|
| 856 | if (i == image->elfcorehdr_index)
|
| 857 | continue;
|
| 858 | #endif
|
| 859 |
|
| 860 | ksegment = &image->segment[i];
|
| 861 | |
| 862 | * Skip purgatory as it will be modified once we put digest
|
| 863 | * info in purgatory.
|
| 864 |
|
| 865 | if (ksegment->kbuf == pi->purgatory_buf)
|
| 866 | continue;
|
| 867 |
|
| 868 | |
| 869 | * Skip the segment if ima_segment_index is set and matches
|
| 870 | * the current index
|
| 871 |
|
| 872 | if (check_ima_segment_index(image, i))
|
| 873 | continue;
|
| 874 |
|
| 875 | sha256_update(&sctx, ksegment->kbuf, ksegment->bufsz);
|
| 876 |
|
| 877 | |
| 878 | * Assume rest of the buffer is filled with zero and
|
| 879 | * update digest accordingly.
|
| 880 |
|
| 881 | nullsz = ksegment->memsz - ksegment->bufsz;
|
| 882 | while (nullsz) {
|
| 883 | unsigned long bytes = nullsz;
|
| 884 |
|
| 885 | if (bytes > zero_buf_sz)
|
| 886 | bytes = zero_buf_sz;
|
| 887 | sha256_update(&sctx, zero_buf, bytes);
|
| 888 | nullsz -= bytes;
|
| 889 | }
|
| 890 |
|
| 891 | sha_regions[j].start = ksegment->mem;
|
| 892 | sha_regions[j].len = ksegment->memsz;
|
| 893 | j++;
|
| 894 | }
|
| 895 |
|
| 896 | skip_checksum:
|
| 897 | sha256_final(&sctx, digest);
|
| 898 |
|
| 899 | ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha_regions",
|
| 900 | sha_regions, sha_region_sz, 0);
|
| 901 | if (ret)
|
| 902 | goto out_free_sha_regions;
|
| 903 |
|
| 904 | ret = kexec_purgatory_get_set_symbol(image, "purgatory_sha256_digest",
|
| 905 | digest, SHA256_DIGEST_SIZE, 0);
|
| 906 | out_free_sha_regions:
|
| 907 | vfree(sha_regions);
|
| 908 | return ret;
|
| 909 | }
|
| 910 |
|
| 911 | #ifdef CONFIG_ARCH_SUPPORTS_KEXEC_PURGATORY
|
| 912 | |
| 913 | * kexec_purgatory_find_symbol - find a symbol in the purgatory
|
| 914 | * @pi: Purgatory to search in.
|
| 915 | * @name: Name of the symbol.
|
| 916 | *
|
| 917 | * Return: pointer to symbol in read-only symtab on success, NULL on error.
|
| 918 |
|
| 919 | static const Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
|
| 920 | const char *name)
|
| 921 | {
|
| 922 | const Elf_Shdr *sechdrs;
|
| 923 | const Elf_Ehdr *ehdr;
|
| 924 | const Elf_Sym *syms;
|
| 925 | const char *strtab;
|
| 926 | int i, k;
|
| 927 |
|
| 928 | if (!pi->ehdr)
|
| 929 | return NULL;
|
| 930 |
|
| 931 | ehdr = pi->ehdr;
|
| 932 | sechdrs = (void *)ehdr + ehdr->e_shoff;
|
| 933 |
|
| 934 | for (i = 0; i < ehdr->e_shnum; i++) {
|
| 935 | if (sechdrs[i].sh_type != SHT_SYMTAB)
|
| 936 | continue;
|
| 937 |
|
| 938 | if (sechdrs[i].sh_link >= ehdr->e_shnum)
|
| 939 |
|
| 940 | continue;
|
| 941 | strtab = (void *)ehdr + sechdrs[sechdrs[i].sh_link].sh_offset;
|
| 942 | syms = (void *)ehdr + sechdrs[i].sh_offset;
|
| 943 |
|
| 944 |
|
| 945 | for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
|
| 946 | if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
|
| 947 | continue;
|
| 948 |
|
| 949 | if (strcmp(strtab + syms[k].st_name, name) != 0)
|
| 950 | continue;
|
| 951 |
|
| 952 | if (syms[k].st_shndx == SHN_UNDEF ||
|
| 953 | syms[k].st_shndx >= ehdr->e_shnum) {
|
| 954 | pr_debug("Symbol: %s has bad section index %d.\n",
|
| 955 | name, syms[k].st_shndx);
|
| 956 | return NULL;
|
| 957 | }
|
| 958 |
|
| 959 |
|
| 960 | return &syms[k];
|
| 961 | }
|
| 962 | }
|
| 963 |
|
| 964 | return NULL;
|
| 965 | }
|
| 966 | |
| 967 | * kexec_purgatory_setup_kbuf - prepare buffer to load purgatory.
|
| 968 | * @pi: Purgatory to be loaded.
|
| 969 | * @kbuf: Buffer to setup.
|
| 970 | *
|
| 971 | * Allocates the memory needed for the buffer. Caller is responsible to free
|
| 972 | * the memory after use.
|
| 973 | *
|
| 974 | * Return: 0 on success, negative errno on error.
|
| 975 |
|
| 976 | static int kexec_purgatory_setup_kbuf(struct purgatory_info *pi,
|
| 977 | struct kexec_buf *kbuf)
|
| 978 | {
|
| 979 | const Elf_Shdr *sechdrs;
|
| 980 | unsigned long bss_align;
|
| 981 | unsigned long bss_sz;
|
| 982 | unsigned long align;
|
| 983 | int i, ret;
|
| 984 |
|
| 985 | sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
|
| 986 | kbuf->buf_align = bss_align = 1;
|
| 987 | kbuf->bufsz = bss_sz = 0;
|
| 988 |
|
| 989 | for (i = 0; i < pi->ehdr->e_shnum; i++) {
|
| 990 | if (!(sechdrs[i].sh_flags & SHF_ALLOC))
|
| 991 | continue;
|
| 992 |
|
| 993 | align = sechdrs[i].sh_addralign;
|
| 994 | if (sechdrs[i].sh_type != SHT_NOBITS) {
|
| 995 | if (kbuf->buf_align < align)
|
| 996 | kbuf->buf_align = align;
|
| 997 | kbuf->bufsz = ALIGN(kbuf->bufsz, align);
|
| 998 | kbuf->bufsz += sechdrs[i].sh_size;
|
| 999 | } else {
|
| 1000 | if (bss_align < align)
|
| 1001 | bss_align = align;
|
| 1002 | bss_sz = ALIGN(bss_sz, align);
|
| 1003 | bss_sz += sechdrs[i].sh_size;
|
| 1004 | }
|
| 1005 | }
|
| 1006 | kbuf->bufsz = ALIGN(kbuf->bufsz, bss_align);
|
| 1007 | kbuf->memsz = kbuf->bufsz + bss_sz;
|
| 1008 | if (kbuf->buf_align < bss_align)
|
| 1009 | kbuf->buf_align = bss_align;
|
| 1010 |
|
| 1011 | kbuf->buffer = vzalloc(kbuf->bufsz);
|
| 1012 | if (!kbuf->buffer)
|
| 1013 | return -ENOMEM;
|
| 1014 | pi->purgatory_buf = kbuf->buffer;
|
| 1015 |
|
| 1016 | ret = kexec_add_buffer(kbuf);
|
| 1017 | if (ret)
|
| 1018 | goto out;
|
| 1019 |
|
| 1020 | return 0;
|
| 1021 | out:
|
| 1022 | vfree(pi->purgatory_buf);
|
| 1023 | pi->purgatory_buf = NULL;
|
| 1024 | return ret;
|
| 1025 | }
|
| 1026 |
|
| 1027 | |
| 1028 | * kexec_purgatory_setup_sechdrs - prepares the pi->sechdrs buffer.
|
| 1029 | * @pi: Purgatory to be loaded.
|
| 1030 | * @kbuf: Buffer prepared to store purgatory.
|
| 1031 | *
|
| 1032 | * Allocates the memory needed for the buffer. Caller is responsible to free
|
| 1033 | * the memory after use.
|
| 1034 | *
|
| 1035 | * Return: 0 on success, negative errno on error.
|
| 1036 |
|
| 1037 | static int kexec_purgatory_setup_sechdrs(struct purgatory_info *pi,
|
| 1038 | struct kexec_buf *kbuf)
|
| 1039 | {
|
| 1040 | unsigned long bss_addr;
|
| 1041 | unsigned long offset;
|
| 1042 | size_t sechdrs_size;
|
| 1043 | Elf_Shdr *sechdrs;
|
| 1044 | const Elf_Sym *entry_sym;
|
| 1045 | u16 entry_shndx = 0;
|
| 1046 | unsigned long entry_off = 0;
|
| 1047 | bool start_fixed = false;
|
| 1048 | int i;
|
| 1049 |
|
| 1050 | |
| 1051 | * The section headers in kexec_purgatory are read-only. In order to
|
| 1052 | * have them modifiable make a temporary copy.
|
| 1053 |
|
| 1054 | sechdrs_size = array_size(sizeof(Elf_Shdr), pi->ehdr->e_shnum);
|
| 1055 | sechdrs = vzalloc(sechdrs_size);
|
| 1056 | if (!sechdrs)
|
| 1057 | return -ENOMEM;
|
| 1058 | memcpy(sechdrs, (void *)pi->ehdr + pi->ehdr->e_shoff, sechdrs_size);
|
| 1059 | pi->sechdrs = sechdrs;
|
| 1060 |
|
| 1061 | offset = 0;
|
| 1062 | bss_addr = kbuf->mem + kbuf->bufsz;
|
| 1063 | kbuf->image->start = pi->ehdr->e_entry;
|
| 1064 |
|
| 1065 | entry_sym = kexec_purgatory_find_symbol(pi, "purgatory_start");
|
| 1066 | if (entry_sym) {
|
| 1067 | entry_shndx = entry_sym->st_shndx;
|
| 1068 | entry_off = entry_sym->st_value;
|
| 1069 | }
|
| 1070 |
|
| 1071 | for (i = 0; i < pi->ehdr->e_shnum; i++) {
|
| 1072 | unsigned long align;
|
| 1073 | void *src, *dst;
|
| 1074 |
|
| 1075 | if (!(sechdrs[i].sh_flags & SHF_ALLOC))
|
| 1076 | continue;
|
| 1077 |
|
| 1078 | align = sechdrs[i].sh_addralign;
|
| 1079 | if (sechdrs[i].sh_type == SHT_NOBITS) {
|
| 1080 | bss_addr = ALIGN(bss_addr, align);
|
| 1081 | sechdrs[i].sh_addr = bss_addr;
|
| 1082 | bss_addr += sechdrs[i].sh_size;
|
| 1083 | continue;
|
| 1084 | }
|
| 1085 |
|
| 1086 | offset = ALIGN(offset, align);
|
| 1087 |
|
| 1088 | if (!start_fixed && entry_sym && i == entry_shndx &&
|
| 1089 | (sechdrs[i].sh_flags & SHF_EXECINSTR) &&
|
| 1090 | entry_off < sechdrs[i].sh_size) {
|
| 1091 | kbuf->image->start = kbuf->mem + offset + entry_off;
|
| 1092 | start_fixed = true;
|
| 1093 | }
|
| 1094 |
|
| 1095 | |
| 1096 | * Check if the segment contains the entry point, if so,
|
| 1097 | * calculate the value of image->start based on it.
|
| 1098 | * If the compiler has produced more than one .text section
|
| 1099 | * (Eg: .text.hot), they are generally after the main .text
|
| 1100 | * section, and they shall not be used to calculate
|
| 1101 | * image->start. So do not re-calculate image->start if it
|
| 1102 | * is not set to the initial value, and warn the user so they
|
| 1103 | * have a chance to fix their purgatory's linker script.
|
| 1104 |
|
| 1105 | if (!start_fixed && sechdrs[i].sh_flags & SHF_EXECINSTR &&
|
| 1106 | pi->ehdr->e_entry >= sechdrs[i].sh_addr &&
|
| 1107 | pi->ehdr->e_entry < (sechdrs[i].sh_addr
|
| 1108 | + sechdrs[i].sh_size) &&
|
| 1109 | kbuf->image->start == pi->ehdr->e_entry) {
|
| 1110 | kbuf->image->start -= sechdrs[i].sh_addr;
|
| 1111 | kbuf->image->start += kbuf->mem + offset;
|
| 1112 | start_fixed = true;
|
| 1113 | }
|
| 1114 |
|
| 1115 | src = (void *)pi->ehdr + sechdrs[i].sh_offset;
|
| 1116 | dst = pi->purgatory_buf + offset;
|
| 1117 | memcpy(dst, src, sechdrs[i].sh_size);
|
| 1118 |
|
| 1119 | sechdrs[i].sh_addr = kbuf->mem + offset;
|
| 1120 | sechdrs[i].sh_offset = offset;
|
| 1121 | offset += sechdrs[i].sh_size;
|
| 1122 | }
|
| 1123 |
|
| 1124 | return 0;
|
| 1125 | }
|
| 1126 |
|
| 1127 | static int kexec_apply_relocations(struct kimage *image)
|
| 1128 | {
|
| 1129 | int i, ret;
|
| 1130 | struct purgatory_info *pi = &image->purgatory_info;
|
| 1131 | const Elf_Shdr *sechdrs;
|
| 1132 |
|
| 1133 | sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
|
| 1134 |
|
| 1135 | for (i = 0; i < pi->ehdr->e_shnum; i++) {
|
| 1136 | const Elf_Shdr *relsec;
|
| 1137 | const Elf_Shdr *symtab;
|
| 1138 | Elf_Shdr *section;
|
| 1139 |
|
| 1140 | relsec = sechdrs + i;
|
| 1141 |
|
| 1142 | if (relsec->sh_type != SHT_RELA &&
|
| 1143 | relsec->sh_type != SHT_REL)
|
| 1144 | continue;
|
| 1145 |
|
| 1146 | |
| 1147 | * For section of type SHT_RELA/SHT_REL,
|
| 1148 | * ->sh_link contains section header index of associated
|
| 1149 | * symbol table. And ->sh_info contains section header
|
| 1150 | * index of section to which relocations apply.
|
| 1151 |
|
| 1152 | if (relsec->sh_info >= pi->ehdr->e_shnum ||
|
| 1153 | relsec->sh_link >= pi->ehdr->e_shnum)
|
| 1154 | return -ENOEXEC;
|
| 1155 |
|
| 1156 | section = pi->sechdrs + relsec->sh_info;
|
| 1157 | symtab = sechdrs + relsec->sh_link;
|
| 1158 |
|
| 1159 | if (!(section->sh_flags & SHF_ALLOC))
|
| 1160 | continue;
|
| 1161 |
|
| 1162 | |
| 1163 | * symtab->sh_link contain section header index of associated
|
| 1164 | * string table.
|
| 1165 |
|
| 1166 | if (symtab->sh_link >= pi->ehdr->e_shnum)
|
| 1167 |
|
| 1168 | continue;
|
| 1169 |
|
| 1170 | |
| 1171 | * Respective architecture needs to provide support for applying
|
| 1172 | * relocations of type SHT_RELA/SHT_REL.
|
| 1173 |
|
| 1174 | if (relsec->sh_type == SHT_RELA)
|
| 1175 | ret = arch_kexec_apply_relocations_add(pi, section,
|
| 1176 | relsec, symtab);
|
| 1177 | else if (relsec->sh_type == SHT_REL)
|
| 1178 | ret = arch_kexec_apply_relocations(pi, section,
|
| 1179 | relsec, symtab);
|
| 1180 | if (ret)
|
| 1181 | return ret;
|
| 1182 | }
|
| 1183 |
|
| 1184 | return 0;
|
| 1185 | }
|
| 1186 |
|
| 1187 | |
| 1188 | * kexec_load_purgatory - Load and relocate the purgatory object.
|
| 1189 | * @image: Image to add the purgatory to.
|
| 1190 | * @kbuf: Memory parameters to use.
|
| 1191 | *
|
| 1192 | * Allocates the memory needed for image->purgatory_info.sechdrs and
|
| 1193 | * image->purgatory_info.purgatory_buf/kbuf->buffer. Caller is responsible
|
| 1194 | * to free the memory after use.
|
| 1195 | *
|
| 1196 | * Return: 0 on success, negative errno on error.
|
| 1197 |
|
| 1198 | int kexec_load_purgatory(struct kimage *image, struct kexec_buf *kbuf)
|
| 1199 | {
|
| 1200 | struct purgatory_info *pi = &image->purgatory_info;
|
| 1201 | int ret;
|
| 1202 |
|
| 1203 | if (kexec_purgatory_size <= 0)
|
| 1204 | return -EINVAL;
|
| 1205 |
|
| 1206 | pi->ehdr = (const Elf_Ehdr *)kexec_purgatory;
|
| 1207 |
|
| 1208 | ret = kexec_purgatory_setup_kbuf(pi, kbuf);
|
| 1209 | if (ret)
|
| 1210 | return ret;
|
| 1211 |
|
| 1212 | ret = kexec_purgatory_setup_sechdrs(pi, kbuf);
|
| 1213 | if (ret)
|
| 1214 | goto out_free_kbuf;
|
| 1215 |
|
| 1216 | ret = kexec_apply_relocations(image);
|
| 1217 | if (ret)
|
| 1218 | goto out;
|
| 1219 |
|
| 1220 | return 0;
|
| 1221 | out:
|
| 1222 | vfree(pi->sechdrs);
|
| 1223 | pi->sechdrs = NULL;
|
| 1224 | out_free_kbuf:
|
| 1225 | vfree(pi->purgatory_buf);
|
| 1226 | pi->purgatory_buf = NULL;
|
| 1227 | return ret;
|
| 1228 | }
|
| 1229 |
|
| 1230 | void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
|
| 1231 | {
|
| 1232 | struct purgatory_info *pi = &image->purgatory_info;
|
| 1233 | const Elf_Sym *sym;
|
| 1234 | Elf_Shdr *sechdr;
|
| 1235 |
|
| 1236 | sym = kexec_purgatory_find_symbol(pi, name);
|
| 1237 | if (!sym)
|
| 1238 | return ERR_PTR(-EINVAL);
|
| 1239 |
|
| 1240 | sechdr = &pi->sechdrs[sym->st_shndx];
|
| 1241 |
|
| 1242 | |
| 1243 | * Returns the address where symbol will finally be loaded after
|
| 1244 | * kexec_load_segment()
|
| 1245 |
|
| 1246 | return (void *)(sechdr->sh_addr + sym->st_value);
|
| 1247 | }
|
| 1248 |
|
| 1249 | |
| 1250 | * Get or set value of a symbol. If "get_value" is true, symbol value is
|
| 1251 | * returned in buf otherwise symbol value is set based on value in buf.
|
| 1252 |
|
| 1253 | int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
|
| 1254 | void *buf, unsigned int size, bool get_value)
|
| 1255 | {
|
| 1256 | struct purgatory_info *pi = &image->purgatory_info;
|
| 1257 | const Elf_Sym *sym;
|
| 1258 | Elf_Shdr *sec;
|
| 1259 | char *sym_buf;
|
| 1260 |
|
| 1261 | sym = kexec_purgatory_find_symbol(pi, name);
|
| 1262 | if (!sym)
|
| 1263 | return -EINVAL;
|
| 1264 |
|
| 1265 | if (sym->st_size != size) {
|
| 1266 | pr_err("symbol %s size mismatch: expected %lu actual %u\n",
|
| 1267 | name, (unsigned long)sym->st_size, size);
|
| 1268 | return -EINVAL;
|
| 1269 | }
|
| 1270 |
|
| 1271 | sec = pi->sechdrs + sym->st_shndx;
|
| 1272 |
|
| 1273 | if (sec->sh_type == SHT_NOBITS) {
|
| 1274 | pr_err("symbol %s is in a bss section. Cannot %s\n", name,
|
| 1275 | get_value ? "get" : "set");
|
| 1276 | return -EINVAL;
|
| 1277 | }
|
| 1278 |
|
| 1279 | sym_buf = (char *)pi->purgatory_buf + sec->sh_offset + sym->st_value;
|
| 1280 |
|
| 1281 | if (get_value)
|
| 1282 | memcpy((void *)buf, sym_buf, size);
|
| 1283 | else
|
| 1284 | memcpy((void *)sym_buf, buf, size);
|
| 1285 |
|
| 1286 | return 0;
|
| 1287 | }
|
| 1288 | #endif
|