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