4 * This contains the routines needed to generate a reasonable level of
5 * entropy to choose a randomized kernel base address offset in support
6 * of Kernel Address Space Layout Randomization (KASLR). Additionally
7 * handles walking the physical memory maps (and tracking memory regions
8 * to avoid) in order to select a physical memory location that can
9 * contain the entire properly aligned running kernel image.
16 #include <asm/archrandom.h>
19 #include <generated/compile.h>
20 #include <linux/module.h>
21 #include <linux/uts.h>
22 #include <linux/utsname.h>
23 #include <generated/utsrelease.h>
25 /* Simplified build-specific string for starting entropy. */
26 static const char build_str[] = UTS_RELEASE " (" LINUX_COMPILE_BY "@"
27 LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION;
29 #define I8254_PORT_CONTROL 0x43
30 #define I8254_PORT_COUNTER0 0x40
31 #define I8254_CMD_READBACK 0xC0
32 #define I8254_SELECT_COUNTER0 0x02
33 #define I8254_STATUS_NOTREADY 0x40
34 static inline u16 i8254(void)
39 outb(I8254_PORT_CONTROL,
40 I8254_CMD_READBACK | I8254_SELECT_COUNTER0);
41 status = inb(I8254_PORT_COUNTER0);
42 timer = inb(I8254_PORT_COUNTER0);
43 timer |= inb(I8254_PORT_COUNTER0) << 8;
44 } while (status & I8254_STATUS_NOTREADY);
49 static unsigned long rotate_xor(unsigned long hash, const void *area,
53 unsigned long *ptr = (unsigned long *)area;
55 for (i = 0; i < size / sizeof(hash); i++) {
56 /* Rotate by odd number of bits and XOR. */
57 hash = (hash << ((sizeof(hash) * 8) - 7)) | (hash >> 7);
64 /* Attempt to create a simple but unpredictable starting entropy. */
65 static unsigned long get_random_boot(void)
67 unsigned long hash = 0;
69 hash = rotate_xor(hash, build_str, sizeof(build_str));
70 hash = rotate_xor(hash, boot_params, sizeof(*boot_params));
75 static unsigned long get_random_long(void)
78 const unsigned long mix_const = 0x5d6008cbf3848dd3UL;
80 const unsigned long mix_const = 0x3f39e593UL;
82 unsigned long raw, random = get_random_boot();
83 bool use_i8254 = true;
85 debug_putstr("KASLR using");
87 if (has_cpuflag(X86_FEATURE_RDRAND)) {
88 debug_putstr(" RDRAND");
89 if (rdrand_long(&raw)) {
95 if (has_cpuflag(X86_FEATURE_TSC)) {
96 debug_putstr(" RDTSC");
104 debug_putstr(" i8254");
108 /* Circular multiply for better bit diffusion */
110 : "=a" (random), "=d" (raw)
111 : "a" (random), "rm" (mix_const));
114 debug_putstr("...\n");
124 #define MEM_AVOID_MAX 4
125 static struct mem_vector mem_avoid[MEM_AVOID_MAX];
127 static bool mem_contains(struct mem_vector *region, struct mem_vector *item)
129 /* Item at least partially before region. */
130 if (item->start < region->start)
132 /* Item at least partially after region. */
133 if (item->start + item->size > region->start + region->size)
138 static bool mem_overlaps(struct mem_vector *one, struct mem_vector *two)
140 /* Item one is entirely before item two. */
141 if (one->start + one->size <= two->start)
143 /* Item one is entirely after item two. */
144 if (one->start >= two->start + two->size)
150 * In theroy, KASLR can put the kernel anywhere in area of [16M, 64T). The
151 * mem_avoid array is used to store the ranges that need to be avoided when
152 * KASLR searches for a an appropriate random address. We must avoid any
153 * regions that are unsafe to overlap with during decompression, and other
154 * things like the initrd, cmdline and boot_params.
156 * How to calculate the unsafe areas is detailed here, and is informed by
157 * the decompression calculations in header.S, and the diagram in misc.c.
159 * The compressed vmlinux (ZO) plus relocs and the run space of ZO can't be
160 * overwritten by decompression output.
162 * ZO sits against the end of the decompression buffer, so we can calculate
163 * where text, data, bss, etc of ZO are positioned.
165 * The follow are already enforced by the code:
166 * - init_size >= kernel_total_size
167 * - input + input_len >= output + output_len
168 * - kernel_total_size could be >= or < output_len
170 * From this, we can make several observations, illustrated by a diagram:
171 * - init_size >= kernel_total_size
172 * - input + input_len > output + output_len
173 * - kernel_total_size >= output_len
175 * 0 output input input+input_len output+init_size
178 * |-----|--------|--------|------------------|----|------------|----------|
181 * output+init_size-ZO_INIT_SIZE output+output_len output+kernel_total_size
183 * [output, output+init_size) is for the buffer for decompressing the
184 * compressed kernel (ZO).
186 * [output, output+kernel_total_size) is for the uncompressed kernel (VO)
187 * and its bss, brk, etc.
188 * [output, output+output_len) is VO plus relocs
190 * [output+init_size-ZO_INIT_SIZE, output+init_size) is the copied ZO.
191 * [input, input+input_len) is the copied compressed (VO (vmlinux after
192 * objcopy) plus relocs), not the ZO.
194 * [input+input_len, output+init_size) is [_text, _end) for ZO. That was the
195 * first range in mem_avoid, which included ZO's heap and stack. Also
196 * [input, input+input_size) need be put in mem_avoid array, but since it
197 * is adjacent to the first entry, they can be merged. This is how we get
198 * the first entry in mem_avoid[].
200 static void mem_avoid_init(unsigned long input, unsigned long input_size,
201 unsigned long output)
203 unsigned long init_size = boot_params->hdr.init_size;
204 u64 initrd_start, initrd_size;
205 u64 cmd_line, cmd_line_size;
209 * Avoid the region that is unsafe to overlap during
212 mem_avoid[0].start = input;
213 mem_avoid[0].size = (output + init_size) - input;
216 initrd_start = (u64)boot_params->ext_ramdisk_image << 32;
217 initrd_start |= boot_params->hdr.ramdisk_image;
218 initrd_size = (u64)boot_params->ext_ramdisk_size << 32;
219 initrd_size |= boot_params->hdr.ramdisk_size;
220 mem_avoid[1].start = initrd_start;
221 mem_avoid[1].size = initrd_size;
223 /* Avoid kernel command line. */
224 cmd_line = (u64)boot_params->ext_cmd_line_ptr << 32;
225 cmd_line |= boot_params->hdr.cmd_line_ptr;
226 /* Calculate size of cmd_line. */
227 ptr = (char *)(unsigned long)cmd_line;
228 for (cmd_line_size = 0; ptr[cmd_line_size++]; )
230 mem_avoid[2].start = cmd_line;
231 mem_avoid[2].size = cmd_line_size;
234 mem_avoid[3].start = (unsigned long)boot_params;
235 mem_avoid[3].size = sizeof(*boot_params);
238 /* Does this memory vector overlap a known avoided area? */
239 static bool mem_avoid_overlap(struct mem_vector *img)
242 struct setup_data *ptr;
244 for (i = 0; i < MEM_AVOID_MAX; i++) {
245 if (mem_overlaps(img, &mem_avoid[i]))
249 /* Avoid all entries in the setup_data linked list. */
250 ptr = (struct setup_data *)(unsigned long)boot_params->hdr.setup_data;
252 struct mem_vector avoid;
254 avoid.start = (unsigned long)ptr;
255 avoid.size = sizeof(*ptr) + ptr->len;
257 if (mem_overlaps(img, &avoid))
260 ptr = (struct setup_data *)(unsigned long)ptr->next;
266 static unsigned long slots[KERNEL_IMAGE_SIZE / CONFIG_PHYSICAL_ALIGN];
267 static unsigned long slot_max;
269 static void slots_append(unsigned long addr)
271 /* Overflowing the slots list should be impossible. */
272 if (slot_max >= KERNEL_IMAGE_SIZE / CONFIG_PHYSICAL_ALIGN)
275 slots[slot_max++] = addr;
278 static unsigned long slots_fetch_random(void)
280 /* Handle case of no slots stored. */
284 return slots[get_random_long() % slot_max];
287 static void process_e820_entry(struct e820entry *entry,
288 unsigned long minimum,
289 unsigned long image_size)
291 struct mem_vector region, img;
293 /* Skip non-RAM entries. */
294 if (entry->type != E820_RAM)
297 /* Ignore entries entirely above our maximum. */
298 if (entry->addr >= KERNEL_IMAGE_SIZE)
301 /* Ignore entries entirely below our minimum. */
302 if (entry->addr + entry->size < minimum)
305 region.start = entry->addr;
306 region.size = entry->size;
308 /* Potentially raise address to minimum location. */
309 if (region.start < minimum)
310 region.start = minimum;
312 /* Potentially raise address to meet alignment requirements. */
313 region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
315 /* Did we raise the address above the bounds of this e820 region? */
316 if (region.start > entry->addr + entry->size)
319 /* Reduce size by any delta from the original address. */
320 region.size -= region.start - entry->addr;
322 /* Reduce maximum size to fit end of image within maximum limit. */
323 if (region.start + region.size > KERNEL_IMAGE_SIZE)
324 region.size = KERNEL_IMAGE_SIZE - region.start;
326 /* Walk each aligned slot and check for avoided areas. */
327 for (img.start = region.start, img.size = image_size ;
328 mem_contains(®ion, &img) ;
329 img.start += CONFIG_PHYSICAL_ALIGN) {
330 if (mem_avoid_overlap(&img))
332 slots_append(img.start);
336 static unsigned long find_random_addr(unsigned long minimum,
342 /* Make sure minimum is aligned. */
343 minimum = ALIGN(minimum, CONFIG_PHYSICAL_ALIGN);
345 /* Verify potential e820 positions, appending to slots list. */
346 for (i = 0; i < boot_params->e820_entries; i++) {
347 process_e820_entry(&boot_params->e820_map[i], minimum, size);
350 return slots_fetch_random();
353 unsigned char *choose_random_location(unsigned char *input_ptr,
354 unsigned long input_size,
355 unsigned char *output_ptr,
356 unsigned long output_size)
359 * The caller of choose_random_location() uses unsigned char * for
360 * buffer pointers since it performs decompression, elf parsing, etc.
361 * Since this code examines addresses much more numerically,
362 * unsigned long is used internally here. Instead of sprinkling
363 * more casts into extract_kernel, do them here and at return.
365 unsigned long input = (unsigned long)input_ptr;
366 unsigned long output = (unsigned long)output_ptr;
367 unsigned long choice = output;
368 unsigned long random_addr;
370 #ifdef CONFIG_HIBERNATION
371 if (!cmdline_find_option_bool("kaslr")) {
372 warn("KASLR disabled: 'kaslr' not on cmdline (hibernation selected).");
376 if (cmdline_find_option_bool("nokaslr")) {
377 warn("KASLR disabled: 'nokaslr' on cmdline.");
382 boot_params->hdr.loadflags |= KASLR_FLAG;
384 /* Record the various known unsafe memory ranges. */
385 mem_avoid_init(input, input_size, output);
387 /* Walk e820 and find a random address. */
388 random_addr = find_random_addr(output, output_size);
390 warn("KASLR disabled: could not find suitable E820 region!");
394 /* Always enforce the minimum. */
395 if (random_addr < choice)
398 choice = random_addr;
400 return (unsigned char *)choice;