xref: /openbmc/linux/arch/x86/boot/compressed/kaslr.c (revision 9a6b55ac)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * kaslr.c
4  *
5  * This contains the routines needed to generate a reasonable level of
6  * entropy to choose a randomized kernel base address offset in support
7  * of Kernel Address Space Layout Randomization (KASLR). Additionally
8  * handles walking the physical memory maps (and tracking memory regions
9  * to avoid) in order to select a physical memory location that can
10  * contain the entire properly aligned running kernel image.
11  *
12  */
13 
14 /*
15  * isspace() in linux/ctype.h is expected by next_args() to filter
16  * out "space/lf/tab". While boot/ctype.h conflicts with linux/ctype.h,
17  * since isdigit() is implemented in both of them. Hence disable it
18  * here.
19  */
20 #define BOOT_CTYPE_H
21 
22 /*
23  * _ctype[] in lib/ctype.c is needed by isspace() of linux/ctype.h.
24  * While both lib/ctype.c and lib/cmdline.c will bring EXPORT_SYMBOL
25  * which is meaningless and will cause compiling error in some cases.
26  */
27 #define __DISABLE_EXPORTS
28 
29 #include "misc.h"
30 #include "error.h"
31 #include "../string.h"
32 
33 #include <generated/compile.h>
34 #include <linux/module.h>
35 #include <linux/uts.h>
36 #include <linux/utsname.h>
37 #include <linux/ctype.h>
38 #include <linux/efi.h>
39 #include <generated/utsrelease.h>
40 #include <asm/efi.h>
41 
42 /* Macros used by the included decompressor code below. */
43 #define STATIC
44 #include <linux/decompress/mm.h>
45 
46 #ifdef CONFIG_X86_5LEVEL
47 unsigned int __pgtable_l5_enabled;
48 unsigned int pgdir_shift __ro_after_init = 39;
49 unsigned int ptrs_per_p4d __ro_after_init = 1;
50 #endif
51 
52 extern unsigned long get_cmd_line_ptr(void);
53 
54 /* Used by PAGE_KERN* macros: */
55 pteval_t __default_kernel_pte_mask __read_mostly = ~0;
56 
57 /* Simplified build-specific string for starting entropy. */
58 static const char build_str[] = UTS_RELEASE " (" LINUX_COMPILE_BY "@"
59 		LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION;
60 
61 static unsigned long rotate_xor(unsigned long hash, const void *area,
62 				size_t size)
63 {
64 	size_t i;
65 	unsigned long *ptr = (unsigned long *)area;
66 
67 	for (i = 0; i < size / sizeof(hash); i++) {
68 		/* Rotate by odd number of bits and XOR. */
69 		hash = (hash << ((sizeof(hash) * 8) - 7)) | (hash >> 7);
70 		hash ^= ptr[i];
71 	}
72 
73 	return hash;
74 }
75 
76 /* Attempt to create a simple but unpredictable starting entropy. */
77 static unsigned long get_boot_seed(void)
78 {
79 	unsigned long hash = 0;
80 
81 	hash = rotate_xor(hash, build_str, sizeof(build_str));
82 	hash = rotate_xor(hash, boot_params, sizeof(*boot_params));
83 
84 	return hash;
85 }
86 
87 #define KASLR_COMPRESSED_BOOT
88 #include "../../lib/kaslr.c"
89 
90 
91 /* Only supporting at most 4 unusable memmap regions with kaslr */
92 #define MAX_MEMMAP_REGIONS	4
93 
94 static bool memmap_too_large;
95 
96 
97 /* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
98 static unsigned long long mem_limit = ULLONG_MAX;
99 
100 /* Number of immovable memory regions */
101 static int num_immovable_mem;
102 
103 enum mem_avoid_index {
104 	MEM_AVOID_ZO_RANGE = 0,
105 	MEM_AVOID_INITRD,
106 	MEM_AVOID_CMDLINE,
107 	MEM_AVOID_BOOTPARAMS,
108 	MEM_AVOID_MEMMAP_BEGIN,
109 	MEM_AVOID_MEMMAP_END = MEM_AVOID_MEMMAP_BEGIN + MAX_MEMMAP_REGIONS - 1,
110 	MEM_AVOID_MAX,
111 };
112 
113 static struct mem_vector mem_avoid[MEM_AVOID_MAX];
114 
115 static bool mem_overlaps(struct mem_vector *one, struct mem_vector *two)
116 {
117 	/* Item one is entirely before item two. */
118 	if (one->start + one->size <= two->start)
119 		return false;
120 	/* Item one is entirely after item two. */
121 	if (one->start >= two->start + two->size)
122 		return false;
123 	return true;
124 }
125 
126 char *skip_spaces(const char *str)
127 {
128 	while (isspace(*str))
129 		++str;
130 	return (char *)str;
131 }
132 #include "../../../../lib/ctype.c"
133 #include "../../../../lib/cmdline.c"
134 
135 enum parse_mode {
136 	PARSE_MEMMAP,
137 	PARSE_EFI,
138 };
139 
140 static int
141 parse_memmap(char *p, unsigned long long *start, unsigned long long *size,
142 		enum parse_mode mode)
143 {
144 	char *oldp;
145 
146 	if (!p)
147 		return -EINVAL;
148 
149 	/* We don't care about this option here */
150 	if (!strncmp(p, "exactmap", 8))
151 		return -EINVAL;
152 
153 	oldp = p;
154 	*size = memparse(p, &p);
155 	if (p == oldp)
156 		return -EINVAL;
157 
158 	switch (*p) {
159 	case '#':
160 	case '$':
161 	case '!':
162 		*start = memparse(p + 1, &p);
163 		return 0;
164 	case '@':
165 		if (mode == PARSE_MEMMAP) {
166 			/*
167 			 * memmap=nn@ss specifies usable region, should
168 			 * be skipped
169 			 */
170 			*size = 0;
171 		} else {
172 			unsigned long long flags;
173 
174 			/*
175 			 * efi_fake_mem=nn@ss:attr the attr specifies
176 			 * flags that might imply a soft-reservation.
177 			 */
178 			*start = memparse(p + 1, &p);
179 			if (p && *p == ':') {
180 				p++;
181 				if (kstrtoull(p, 0, &flags) < 0)
182 					*size = 0;
183 				else if (flags & EFI_MEMORY_SP)
184 					return 0;
185 			}
186 			*size = 0;
187 		}
188 		/* Fall through */
189 	default:
190 		/*
191 		 * If w/o offset, only size specified, memmap=nn[KMG] has the
192 		 * same behaviour as mem=nn[KMG]. It limits the max address
193 		 * system can use. Region above the limit should be avoided.
194 		 */
195 		*start = 0;
196 		return 0;
197 	}
198 
199 	return -EINVAL;
200 }
201 
202 static void mem_avoid_memmap(enum parse_mode mode, char *str)
203 {
204 	static int i;
205 
206 	if (i >= MAX_MEMMAP_REGIONS)
207 		return;
208 
209 	while (str && (i < MAX_MEMMAP_REGIONS)) {
210 		int rc;
211 		unsigned long long start, size;
212 		char *k = strchr(str, ',');
213 
214 		if (k)
215 			*k++ = 0;
216 
217 		rc = parse_memmap(str, &start, &size, mode);
218 		if (rc < 0)
219 			break;
220 		str = k;
221 
222 		if (start == 0) {
223 			/* Store the specified memory limit if size > 0 */
224 			if (size > 0)
225 				mem_limit = size;
226 
227 			continue;
228 		}
229 
230 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
231 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
232 		i++;
233 	}
234 
235 	/* More than 4 memmaps, fail kaslr */
236 	if ((i >= MAX_MEMMAP_REGIONS) && str)
237 		memmap_too_large = true;
238 }
239 
240 /* Store the number of 1GB huge pages which users specified: */
241 static unsigned long max_gb_huge_pages;
242 
243 static void parse_gb_huge_pages(char *param, char *val)
244 {
245 	static bool gbpage_sz;
246 	char *p;
247 
248 	if (!strcmp(param, "hugepagesz")) {
249 		p = val;
250 		if (memparse(p, &p) != PUD_SIZE) {
251 			gbpage_sz = false;
252 			return;
253 		}
254 
255 		if (gbpage_sz)
256 			warn("Repeatedly set hugeTLB page size of 1G!\n");
257 		gbpage_sz = true;
258 		return;
259 	}
260 
261 	if (!strcmp(param, "hugepages") && gbpage_sz) {
262 		p = val;
263 		max_gb_huge_pages = simple_strtoull(p, &p, 0);
264 		return;
265 	}
266 }
267 
268 static void handle_mem_options(void)
269 {
270 	char *args = (char *)get_cmd_line_ptr();
271 	size_t len = strlen((char *)args);
272 	char *tmp_cmdline;
273 	char *param, *val;
274 	u64 mem_size;
275 
276 	if (!strstr(args, "memmap=") && !strstr(args, "mem=") &&
277 		!strstr(args, "hugepages"))
278 		return;
279 
280 	tmp_cmdline = malloc(len + 1);
281 	if (!tmp_cmdline)
282 		error("Failed to allocate space for tmp_cmdline");
283 
284 	memcpy(tmp_cmdline, args, len);
285 	tmp_cmdline[len] = 0;
286 	args = tmp_cmdline;
287 
288 	/* Chew leading spaces */
289 	args = skip_spaces(args);
290 
291 	while (*args) {
292 		args = next_arg(args, &param, &val);
293 		/* Stop at -- */
294 		if (!val && strcmp(param, "--") == 0) {
295 			warn("Only '--' specified in cmdline");
296 			goto out;
297 		}
298 
299 		if (!strcmp(param, "memmap")) {
300 			mem_avoid_memmap(PARSE_MEMMAP, val);
301 		} else if (strstr(param, "hugepages")) {
302 			parse_gb_huge_pages(param, val);
303 		} else if (!strcmp(param, "mem")) {
304 			char *p = val;
305 
306 			if (!strcmp(p, "nopentium"))
307 				continue;
308 			mem_size = memparse(p, &p);
309 			if (mem_size == 0)
310 				goto out;
311 
312 			mem_limit = mem_size;
313 		} else if (!strcmp(param, "efi_fake_mem")) {
314 			mem_avoid_memmap(PARSE_EFI, val);
315 		}
316 	}
317 
318 out:
319 	free(tmp_cmdline);
320 	return;
321 }
322 
323 /*
324  * In theory, KASLR can put the kernel anywhere in the range of [16M, 64T).
325  * The mem_avoid array is used to store the ranges that need to be avoided
326  * when KASLR searches for an appropriate random address. We must avoid any
327  * regions that are unsafe to overlap with during decompression, and other
328  * things like the initrd, cmdline and boot_params. This comment seeks to
329  * explain mem_avoid as clearly as possible since incorrect mem_avoid
330  * memory ranges lead to really hard to debug boot failures.
331  *
332  * The initrd, cmdline, and boot_params are trivial to identify for
333  * avoiding. They are MEM_AVOID_INITRD, MEM_AVOID_CMDLINE, and
334  * MEM_AVOID_BOOTPARAMS respectively below.
335  *
336  * What is not obvious how to avoid is the range of memory that is used
337  * during decompression (MEM_AVOID_ZO_RANGE below). This range must cover
338  * the compressed kernel (ZO) and its run space, which is used to extract
339  * the uncompressed kernel (VO) and relocs.
340  *
341  * ZO's full run size sits against the end of the decompression buffer, so
342  * we can calculate where text, data, bss, etc of ZO are positioned more
343  * easily.
344  *
345  * For additional background, the decompression calculations can be found
346  * in header.S, and the memory diagram is based on the one found in misc.c.
347  *
348  * The following conditions are already enforced by the image layouts and
349  * associated code:
350  *  - input + input_size >= output + output_size
351  *  - kernel_total_size <= init_size
352  *  - kernel_total_size <= output_size (see Note below)
353  *  - output + init_size >= output + output_size
354  *
355  * (Note that kernel_total_size and output_size have no fundamental
356  * relationship, but output_size is passed to choose_random_location
357  * as a maximum of the two. The diagram is showing a case where
358  * kernel_total_size is larger than output_size, but this case is
359  * handled by bumping output_size.)
360  *
361  * The above conditions can be illustrated by a diagram:
362  *
363  * 0   output            input            input+input_size    output+init_size
364  * |     |                 |                             |             |
365  * |     |                 |                             |             |
366  * |-----|--------|--------|--------------|-----------|--|-------------|
367  *                |                       |           |
368  *                |                       |           |
369  * output+init_size-ZO_INIT_SIZE  output+output_size  output+kernel_total_size
370  *
371  * [output, output+init_size) is the entire memory range used for
372  * extracting the compressed image.
373  *
374  * [output, output+kernel_total_size) is the range needed for the
375  * uncompressed kernel (VO) and its run size (bss, brk, etc).
376  *
377  * [output, output+output_size) is VO plus relocs (i.e. the entire
378  * uncompressed payload contained by ZO). This is the area of the buffer
379  * written to during decompression.
380  *
381  * [output+init_size-ZO_INIT_SIZE, output+init_size) is the worst-case
382  * range of the copied ZO and decompression code. (i.e. the range
383  * covered backwards of size ZO_INIT_SIZE, starting from output+init_size.)
384  *
385  * [input, input+input_size) is the original copied compressed image (ZO)
386  * (i.e. it does not include its run size). This range must be avoided
387  * because it contains the data used for decompression.
388  *
389  * [input+input_size, output+init_size) is [_text, _end) for ZO. This
390  * range includes ZO's heap and stack, and must be avoided since it
391  * performs the decompression.
392  *
393  * Since the above two ranges need to be avoided and they are adjacent,
394  * they can be merged, resulting in: [input, output+init_size) which
395  * becomes the MEM_AVOID_ZO_RANGE below.
396  */
397 static void mem_avoid_init(unsigned long input, unsigned long input_size,
398 			   unsigned long output)
399 {
400 	unsigned long init_size = boot_params->hdr.init_size;
401 	u64 initrd_start, initrd_size;
402 	u64 cmd_line, cmd_line_size;
403 	char *ptr;
404 
405 	/*
406 	 * Avoid the region that is unsafe to overlap during
407 	 * decompression.
408 	 */
409 	mem_avoid[MEM_AVOID_ZO_RANGE].start = input;
410 	mem_avoid[MEM_AVOID_ZO_RANGE].size = (output + init_size) - input;
411 	add_identity_map(mem_avoid[MEM_AVOID_ZO_RANGE].start,
412 			 mem_avoid[MEM_AVOID_ZO_RANGE].size);
413 
414 	/* Avoid initrd. */
415 	initrd_start  = (u64)boot_params->ext_ramdisk_image << 32;
416 	initrd_start |= boot_params->hdr.ramdisk_image;
417 	initrd_size  = (u64)boot_params->ext_ramdisk_size << 32;
418 	initrd_size |= boot_params->hdr.ramdisk_size;
419 	mem_avoid[MEM_AVOID_INITRD].start = initrd_start;
420 	mem_avoid[MEM_AVOID_INITRD].size = initrd_size;
421 	/* No need to set mapping for initrd, it will be handled in VO. */
422 
423 	/* Avoid kernel command line. */
424 	cmd_line  = (u64)boot_params->ext_cmd_line_ptr << 32;
425 	cmd_line |= boot_params->hdr.cmd_line_ptr;
426 	/* Calculate size of cmd_line. */
427 	ptr = (char *)(unsigned long)cmd_line;
428 	for (cmd_line_size = 0; ptr[cmd_line_size++];)
429 		;
430 	mem_avoid[MEM_AVOID_CMDLINE].start = cmd_line;
431 	mem_avoid[MEM_AVOID_CMDLINE].size = cmd_line_size;
432 	add_identity_map(mem_avoid[MEM_AVOID_CMDLINE].start,
433 			 mem_avoid[MEM_AVOID_CMDLINE].size);
434 
435 	/* Avoid boot parameters. */
436 	mem_avoid[MEM_AVOID_BOOTPARAMS].start = (unsigned long)boot_params;
437 	mem_avoid[MEM_AVOID_BOOTPARAMS].size = sizeof(*boot_params);
438 	add_identity_map(mem_avoid[MEM_AVOID_BOOTPARAMS].start,
439 			 mem_avoid[MEM_AVOID_BOOTPARAMS].size);
440 
441 	/* We don't need to set a mapping for setup_data. */
442 
443 	/* Mark the memmap regions we need to avoid */
444 	handle_mem_options();
445 
446 	/* Enumerate the immovable memory regions */
447 	num_immovable_mem = count_immovable_mem_regions();
448 
449 #ifdef CONFIG_X86_VERBOSE_BOOTUP
450 	/* Make sure video RAM can be used. */
451 	add_identity_map(0, PMD_SIZE);
452 #endif
453 }
454 
455 /*
456  * Does this memory vector overlap a known avoided area? If so, record the
457  * overlap region with the lowest address.
458  */
459 static bool mem_avoid_overlap(struct mem_vector *img,
460 			      struct mem_vector *overlap)
461 {
462 	int i;
463 	struct setup_data *ptr;
464 	unsigned long earliest = img->start + img->size;
465 	bool is_overlapping = false;
466 
467 	for (i = 0; i < MEM_AVOID_MAX; i++) {
468 		if (mem_overlaps(img, &mem_avoid[i]) &&
469 		    mem_avoid[i].start < earliest) {
470 			*overlap = mem_avoid[i];
471 			earliest = overlap->start;
472 			is_overlapping = true;
473 		}
474 	}
475 
476 	/* Avoid all entries in the setup_data linked list. */
477 	ptr = (struct setup_data *)(unsigned long)boot_params->hdr.setup_data;
478 	while (ptr) {
479 		struct mem_vector avoid;
480 
481 		avoid.start = (unsigned long)ptr;
482 		avoid.size = sizeof(*ptr) + ptr->len;
483 
484 		if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
485 			*overlap = avoid;
486 			earliest = overlap->start;
487 			is_overlapping = true;
488 		}
489 
490 		if (ptr->type == SETUP_INDIRECT &&
491 		    ((struct setup_indirect *)ptr->data)->type != SETUP_INDIRECT) {
492 			avoid.start = ((struct setup_indirect *)ptr->data)->addr;
493 			avoid.size = ((struct setup_indirect *)ptr->data)->len;
494 
495 			if (mem_overlaps(img, &avoid) && (avoid.start < earliest)) {
496 				*overlap = avoid;
497 				earliest = overlap->start;
498 				is_overlapping = true;
499 			}
500 		}
501 
502 		ptr = (struct setup_data *)(unsigned long)ptr->next;
503 	}
504 
505 	return is_overlapping;
506 }
507 
508 struct slot_area {
509 	unsigned long addr;
510 	int num;
511 };
512 
513 #define MAX_SLOT_AREA 100
514 
515 static struct slot_area slot_areas[MAX_SLOT_AREA];
516 
517 static unsigned long slot_max;
518 
519 static unsigned long slot_area_index;
520 
521 static void store_slot_info(struct mem_vector *region, unsigned long image_size)
522 {
523 	struct slot_area slot_area;
524 
525 	if (slot_area_index == MAX_SLOT_AREA)
526 		return;
527 
528 	slot_area.addr = region->start;
529 	slot_area.num = (region->size - image_size) /
530 			CONFIG_PHYSICAL_ALIGN + 1;
531 
532 	if (slot_area.num > 0) {
533 		slot_areas[slot_area_index++] = slot_area;
534 		slot_max += slot_area.num;
535 	}
536 }
537 
538 /*
539  * Skip as many 1GB huge pages as possible in the passed region
540  * according to the number which users specified:
541  */
542 static void
543 process_gb_huge_pages(struct mem_vector *region, unsigned long image_size)
544 {
545 	unsigned long addr, size = 0;
546 	struct mem_vector tmp;
547 	int i = 0;
548 
549 	if (!max_gb_huge_pages) {
550 		store_slot_info(region, image_size);
551 		return;
552 	}
553 
554 	addr = ALIGN(region->start, PUD_SIZE);
555 	/* Did we raise the address above the passed in memory entry? */
556 	if (addr < region->start + region->size)
557 		size = region->size - (addr - region->start);
558 
559 	/* Check how many 1GB huge pages can be filtered out: */
560 	while (size > PUD_SIZE && max_gb_huge_pages) {
561 		size -= PUD_SIZE;
562 		max_gb_huge_pages--;
563 		i++;
564 	}
565 
566 	/* No good 1GB huge pages found: */
567 	if (!i) {
568 		store_slot_info(region, image_size);
569 		return;
570 	}
571 
572 	/*
573 	 * Skip those 'i'*1GB good huge pages, and continue checking and
574 	 * processing the remaining head or tail part of the passed region
575 	 * if available.
576 	 */
577 
578 	if (addr >= region->start + image_size) {
579 		tmp.start = region->start;
580 		tmp.size = addr - region->start;
581 		store_slot_info(&tmp, image_size);
582 	}
583 
584 	size  = region->size - (addr - region->start) - i * PUD_SIZE;
585 	if (size >= image_size) {
586 		tmp.start = addr + i * PUD_SIZE;
587 		tmp.size = size;
588 		store_slot_info(&tmp, image_size);
589 	}
590 }
591 
592 static unsigned long slots_fetch_random(void)
593 {
594 	unsigned long slot;
595 	int i;
596 
597 	/* Handle case of no slots stored. */
598 	if (slot_max == 0)
599 		return 0;
600 
601 	slot = kaslr_get_random_long("Physical") % slot_max;
602 
603 	for (i = 0; i < slot_area_index; i++) {
604 		if (slot >= slot_areas[i].num) {
605 			slot -= slot_areas[i].num;
606 			continue;
607 		}
608 		return slot_areas[i].addr + slot * CONFIG_PHYSICAL_ALIGN;
609 	}
610 
611 	if (i == slot_area_index)
612 		debug_putstr("slots_fetch_random() failed!?\n");
613 	return 0;
614 }
615 
616 static void __process_mem_region(struct mem_vector *entry,
617 				 unsigned long minimum,
618 				 unsigned long image_size)
619 {
620 	struct mem_vector region, overlap;
621 	unsigned long start_orig, end;
622 	struct mem_vector cur_entry;
623 
624 	/* On 32-bit, ignore entries entirely above our maximum. */
625 	if (IS_ENABLED(CONFIG_X86_32) && entry->start >= KERNEL_IMAGE_SIZE)
626 		return;
627 
628 	/* Ignore entries entirely below our minimum. */
629 	if (entry->start + entry->size < minimum)
630 		return;
631 
632 	/* Ignore entries above memory limit */
633 	end = min(entry->size + entry->start, mem_limit);
634 	if (entry->start >= end)
635 		return;
636 	cur_entry.start = entry->start;
637 	cur_entry.size = end - entry->start;
638 
639 	region.start = cur_entry.start;
640 	region.size = cur_entry.size;
641 
642 	/* Give up if slot area array is full. */
643 	while (slot_area_index < MAX_SLOT_AREA) {
644 		start_orig = region.start;
645 
646 		/* Potentially raise address to minimum location. */
647 		if (region.start < minimum)
648 			region.start = minimum;
649 
650 		/* Potentially raise address to meet alignment needs. */
651 		region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
652 
653 		/* Did we raise the address above the passed in memory entry? */
654 		if (region.start > cur_entry.start + cur_entry.size)
655 			return;
656 
657 		/* Reduce size by any delta from the original address. */
658 		region.size -= region.start - start_orig;
659 
660 		/* On 32-bit, reduce region size to fit within max size. */
661 		if (IS_ENABLED(CONFIG_X86_32) &&
662 		    region.start + region.size > KERNEL_IMAGE_SIZE)
663 			region.size = KERNEL_IMAGE_SIZE - region.start;
664 
665 		/* Return if region can't contain decompressed kernel */
666 		if (region.size < image_size)
667 			return;
668 
669 		/* If nothing overlaps, store the region and return. */
670 		if (!mem_avoid_overlap(&region, &overlap)) {
671 			process_gb_huge_pages(&region, image_size);
672 			return;
673 		}
674 
675 		/* Store beginning of region if holds at least image_size. */
676 		if (overlap.start > region.start + image_size) {
677 			struct mem_vector beginning;
678 
679 			beginning.start = region.start;
680 			beginning.size = overlap.start - region.start;
681 			process_gb_huge_pages(&beginning, image_size);
682 		}
683 
684 		/* Return if overlap extends to or past end of region. */
685 		if (overlap.start + overlap.size >= region.start + region.size)
686 			return;
687 
688 		/* Clip off the overlapping region and start over. */
689 		region.size -= overlap.start - region.start + overlap.size;
690 		region.start = overlap.start + overlap.size;
691 	}
692 }
693 
694 static bool process_mem_region(struct mem_vector *region,
695 			       unsigned long long minimum,
696 			       unsigned long long image_size)
697 {
698 	int i;
699 	/*
700 	 * If no immovable memory found, or MEMORY_HOTREMOVE disabled,
701 	 * use @region directly.
702 	 */
703 	if (!num_immovable_mem) {
704 		__process_mem_region(region, minimum, image_size);
705 
706 		if (slot_area_index == MAX_SLOT_AREA) {
707 			debug_putstr("Aborted e820/efi memmap scan (slot_areas full)!\n");
708 			return 1;
709 		}
710 		return 0;
711 	}
712 
713 #if defined(CONFIG_MEMORY_HOTREMOVE) && defined(CONFIG_ACPI)
714 	/*
715 	 * If immovable memory found, filter the intersection between
716 	 * immovable memory and @region.
717 	 */
718 	for (i = 0; i < num_immovable_mem; i++) {
719 		unsigned long long start, end, entry_end, region_end;
720 		struct mem_vector entry;
721 
722 		if (!mem_overlaps(region, &immovable_mem[i]))
723 			continue;
724 
725 		start = immovable_mem[i].start;
726 		end = start + immovable_mem[i].size;
727 		region_end = region->start + region->size;
728 
729 		entry.start = clamp(region->start, start, end);
730 		entry_end = clamp(region_end, start, end);
731 		entry.size = entry_end - entry.start;
732 
733 		__process_mem_region(&entry, minimum, image_size);
734 
735 		if (slot_area_index == MAX_SLOT_AREA) {
736 			debug_putstr("Aborted e820/efi memmap scan when walking immovable regions(slot_areas full)!\n");
737 			return 1;
738 		}
739 	}
740 #endif
741 	return 0;
742 }
743 
744 #ifdef CONFIG_EFI
745 /*
746  * Returns true if mirror region found (and must have been processed
747  * for slots adding)
748  */
749 static bool
750 process_efi_entries(unsigned long minimum, unsigned long image_size)
751 {
752 	struct efi_info *e = &boot_params->efi_info;
753 	bool efi_mirror_found = false;
754 	struct mem_vector region;
755 	efi_memory_desc_t *md;
756 	unsigned long pmap;
757 	char *signature;
758 	u32 nr_desc;
759 	int i;
760 
761 	signature = (char *)&e->efi_loader_signature;
762 	if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) &&
763 	    strncmp(signature, EFI64_LOADER_SIGNATURE, 4))
764 		return false;
765 
766 #ifdef CONFIG_X86_32
767 	/* Can't handle data above 4GB at this time */
768 	if (e->efi_memmap_hi) {
769 		warn("EFI memmap is above 4GB, can't be handled now on x86_32. EFI should be disabled.\n");
770 		return false;
771 	}
772 	pmap =  e->efi_memmap;
773 #else
774 	pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
775 #endif
776 
777 	nr_desc = e->efi_memmap_size / e->efi_memdesc_size;
778 	for (i = 0; i < nr_desc; i++) {
779 		md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
780 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
781 			efi_mirror_found = true;
782 			break;
783 		}
784 	}
785 
786 	for (i = 0; i < nr_desc; i++) {
787 		md = efi_early_memdesc_ptr(pmap, e->efi_memdesc_size, i);
788 
789 		/*
790 		 * Here we are more conservative in picking free memory than
791 		 * the EFI spec allows:
792 		 *
793 		 * According to the spec, EFI_BOOT_SERVICES_{CODE|DATA} are also
794 		 * free memory and thus available to place the kernel image into,
795 		 * but in practice there's firmware where using that memory leads
796 		 * to crashes.
797 		 *
798 		 * Only EFI_CONVENTIONAL_MEMORY is guaranteed to be free.
799 		 */
800 		if (md->type != EFI_CONVENTIONAL_MEMORY)
801 			continue;
802 
803 		if (efi_soft_reserve_enabled() &&
804 		    (md->attribute & EFI_MEMORY_SP))
805 			continue;
806 
807 		if (efi_mirror_found &&
808 		    !(md->attribute & EFI_MEMORY_MORE_RELIABLE))
809 			continue;
810 
811 		region.start = md->phys_addr;
812 		region.size = md->num_pages << EFI_PAGE_SHIFT;
813 		if (process_mem_region(&region, minimum, image_size))
814 			break;
815 	}
816 	return true;
817 }
818 #else
819 static inline bool
820 process_efi_entries(unsigned long minimum, unsigned long image_size)
821 {
822 	return false;
823 }
824 #endif
825 
826 static void process_e820_entries(unsigned long minimum,
827 				 unsigned long image_size)
828 {
829 	int i;
830 	struct mem_vector region;
831 	struct boot_e820_entry *entry;
832 
833 	/* Verify potential e820 positions, appending to slots list. */
834 	for (i = 0; i < boot_params->e820_entries; i++) {
835 		entry = &boot_params->e820_table[i];
836 		/* Skip non-RAM entries. */
837 		if (entry->type != E820_TYPE_RAM)
838 			continue;
839 		region.start = entry->addr;
840 		region.size = entry->size;
841 		if (process_mem_region(&region, minimum, image_size))
842 			break;
843 	}
844 }
845 
846 static unsigned long find_random_phys_addr(unsigned long minimum,
847 					   unsigned long image_size)
848 {
849 	/* Check if we had too many memmaps. */
850 	if (memmap_too_large) {
851 		debug_putstr("Aborted memory entries scan (more than 4 memmap= args)!\n");
852 		return 0;
853 	}
854 
855 	/* Make sure minimum is aligned. */
856 	minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN);
857 
858 	if (process_efi_entries(minimum, image_size))
859 		return slots_fetch_random();
860 
861 	process_e820_entries(minimum, image_size);
862 	return slots_fetch_random();
863 }
864 
865 static unsigned long find_random_virt_addr(unsigned long minimum,
866 					   unsigned long image_size)
867 {
868 	unsigned long slots, random_addr;
869 
870 	/* Make sure minimum is aligned. */
871 	minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN);
872 	/* Align image_size for easy slot calculations. */
873 	image_size = ALIGN(image_size, CONFIG_PHYSICAL_ALIGN);
874 
875 	/*
876 	 * There are how many CONFIG_PHYSICAL_ALIGN-sized slots
877 	 * that can hold image_size within the range of minimum to
878 	 * KERNEL_IMAGE_SIZE?
879 	 */
880 	slots = (KERNEL_IMAGE_SIZE - minimum - image_size) /
881 		 CONFIG_PHYSICAL_ALIGN + 1;
882 
883 	random_addr = kaslr_get_random_long("Virtual") % slots;
884 
885 	return random_addr * CONFIG_PHYSICAL_ALIGN + minimum;
886 }
887 
888 /*
889  * Since this function examines addresses much more numerically,
890  * it takes the input and output pointers as 'unsigned long'.
891  */
892 void choose_random_location(unsigned long input,
893 			    unsigned long input_size,
894 			    unsigned long *output,
895 			    unsigned long output_size,
896 			    unsigned long *virt_addr)
897 {
898 	unsigned long random_addr, min_addr;
899 
900 	if (cmdline_find_option_bool("nokaslr")) {
901 		warn("KASLR disabled: 'nokaslr' on cmdline.");
902 		return;
903 	}
904 
905 #ifdef CONFIG_X86_5LEVEL
906 	if (__read_cr4() & X86_CR4_LA57) {
907 		__pgtable_l5_enabled = 1;
908 		pgdir_shift = 48;
909 		ptrs_per_p4d = 512;
910 	}
911 #endif
912 
913 	boot_params->hdr.loadflags |= KASLR_FLAG;
914 
915 	/* Prepare to add new identity pagetables on demand. */
916 	initialize_identity_maps();
917 
918 	/* Record the various known unsafe memory ranges. */
919 	mem_avoid_init(input, input_size, *output);
920 
921 	/*
922 	 * Low end of the randomization range should be the
923 	 * smaller of 512M or the initial kernel image
924 	 * location:
925 	 */
926 	min_addr = min(*output, 512UL << 20);
927 
928 	/* Walk available memory entries to find a random address. */
929 	random_addr = find_random_phys_addr(min_addr, output_size);
930 	if (!random_addr) {
931 		warn("Physical KASLR disabled: no suitable memory region!");
932 	} else {
933 		/* Update the new physical address location. */
934 		if (*output != random_addr) {
935 			add_identity_map(random_addr, output_size);
936 			*output = random_addr;
937 		}
938 
939 		/*
940 		 * This loads the identity mapping page table.
941 		 * This should only be done if a new physical address
942 		 * is found for the kernel, otherwise we should keep
943 		 * the old page table to make it be like the "nokaslr"
944 		 * case.
945 		 */
946 		finalize_identity_maps();
947 	}
948 
949 
950 	/* Pick random virtual address starting from LOAD_PHYSICAL_ADDR. */
951 	if (IS_ENABLED(CONFIG_X86_64))
952 		random_addr = find_random_virt_addr(LOAD_PHYSICAL_ADDR, output_size);
953 	*virt_addr = random_addr;
954 }
955