1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Flexible mmap layout support
4 *
5 * Based on code by Ingo Molnar and Andi Kleen, copyrighted
6 * as follows:
7 *
8 * Copyright 2003-2009 Red Hat Inc.
9 * All Rights Reserved.
10 * Copyright 2005 Andi Kleen, SUSE Labs.
11 * Copyright 2007 Jiri Kosina, SUSE Labs.
12 */
13
14 #include <linux/page_size_compat.h>
15 #include <linux/personality.h>
16 #include <linux/mm.h>
17 #include <linux/random.h>
18 #include <linux/limits.h>
19 #include <linux/sched/signal.h>
20 #include <linux/sched/mm.h>
21 #include <linux/compat.h>
22 #include <linux/elf-randomize.h>
23 #include <asm/elf.h>
24 #include <asm/io.h>
25
26 #include "physaddr.h"
27
28 struct va_alignment __read_mostly va_align = {
29 .flags = -1,
30 };
31
task_size_32bit(void)32 unsigned long task_size_32bit(void)
33 {
34 return IA32_PAGE_OFFSET;
35 }
36
task_size_64bit(int full_addr_space)37 unsigned long task_size_64bit(int full_addr_space)
38 {
39 return full_addr_space ? TASK_SIZE_MAX : DEFAULT_MAP_WINDOW;
40 }
41
stack_maxrandom_size(unsigned long task_size)42 static unsigned long stack_maxrandom_size(unsigned long task_size)
43 {
44 unsigned long max = 0;
45 if (current->flags & PF_RANDOMIZE) {
46 max = (-1UL) & __STACK_RND_MASK(task_size == task_size_32bit());
47 max <<= PAGE_SHIFT;
48 }
49
50 return max;
51 }
52
53 #ifdef CONFIG_COMPAT
54 # define mmap32_rnd_bits mmap_rnd_compat_bits
55 # define mmap64_rnd_bits mmap_rnd_bits
56 #else
57 # define mmap32_rnd_bits mmap_rnd_bits
58 # define mmap64_rnd_bits mmap_rnd_bits
59 #endif
60
61 #define SIZE_128M (128 * 1024 * 1024UL)
62
mmap_is_legacy(void)63 static int mmap_is_legacy(void)
64 {
65 if (current->personality & ADDR_COMPAT_LAYOUT)
66 return 1;
67
68 return sysctl_legacy_va_layout;
69 }
70
arch_rnd(unsigned int rndbits)71 static unsigned long arch_rnd(unsigned int rndbits)
72 {
73 if (!(current->flags & PF_RANDOMIZE))
74 return 0;
75 return (get_random_long() & ((1UL << rndbits) - 1)) << __PAGE_SHIFT;
76 }
77
arch_mmap_rnd(void)78 unsigned long arch_mmap_rnd(void)
79 {
80 return arch_rnd(mmap_is_ia32() ? mmap32_rnd_bits : mmap64_rnd_bits);
81 }
82
mmap_base(unsigned long rnd,unsigned long task_size,struct rlimit * rlim_stack)83 static unsigned long mmap_base(unsigned long rnd, unsigned long task_size,
84 struct rlimit *rlim_stack)
85 {
86 unsigned long gap = rlim_stack->rlim_cur;
87 unsigned long pad = stack_maxrandom_size(task_size) + stack_guard_gap;
88 unsigned long gap_min, gap_max;
89
90 /* Values close to RLIM_INFINITY can overflow. */
91 if (gap + pad > gap)
92 gap += pad;
93
94 /*
95 * Top of mmap area (just below the process stack).
96 * Leave an at least ~128 MB hole with possible stack randomization.
97 */
98 gap_min = SIZE_128M;
99 gap_max = (task_size / 6) * 5;
100
101 if (gap < gap_min)
102 gap = gap_min;
103 else if (gap > gap_max)
104 gap = gap_max;
105
106 return PAGE_ALIGN(task_size - gap - rnd);
107 }
108
mmap_legacy_base(unsigned long rnd,unsigned long task_size)109 static unsigned long mmap_legacy_base(unsigned long rnd,
110 unsigned long task_size)
111 {
112 return __TASK_UNMAPPED_BASE(task_size) + rnd;
113 }
114
115 /*
116 * This function, called very early during the creation of a new
117 * process VM image, sets up which VM layout function to use:
118 */
arch_pick_mmap_base(unsigned long * base,unsigned long * legacy_base,unsigned long random_factor,unsigned long task_size,struct rlimit * rlim_stack)119 static void arch_pick_mmap_base(unsigned long *base, unsigned long *legacy_base,
120 unsigned long random_factor, unsigned long task_size,
121 struct rlimit *rlim_stack)
122 {
123 *legacy_base = mmap_legacy_base(random_factor, task_size);
124 if (mmap_is_legacy())
125 *base = *legacy_base;
126 else
127 *base = mmap_base(random_factor, task_size, rlim_stack);
128 }
129
arch_pick_mmap_layout(struct mm_struct * mm,struct rlimit * rlim_stack)130 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
131 {
132 if (mmap_is_legacy())
133 clear_bit(MMF_TOPDOWN, &mm->flags);
134 else
135 set_bit(MMF_TOPDOWN, &mm->flags);
136
137 arch_pick_mmap_base(&mm->mmap_base, &mm->mmap_legacy_base,
138 arch_rnd(mmap64_rnd_bits), task_size_64bit(0),
139 rlim_stack);
140
141 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
142 /*
143 * The mmap syscall mapping base decision depends solely on the
144 * syscall type (64-bit or compat). This applies for 64bit
145 * applications and 32bit applications. The 64bit syscall uses
146 * mmap_base, the compat syscall uses mmap_compat_base.
147 */
148 arch_pick_mmap_base(&mm->mmap_compat_base, &mm->mmap_compat_legacy_base,
149 arch_rnd(mmap32_rnd_bits), task_size_32bit(),
150 rlim_stack);
151 #endif
152 }
153
get_mmap_base(int is_legacy)154 unsigned long get_mmap_base(int is_legacy)
155 {
156 struct mm_struct *mm = current->mm;
157
158 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
159 if (in_32bit_syscall()) {
160 return is_legacy ? mm->mmap_compat_legacy_base
161 : mm->mmap_compat_base;
162 }
163 #endif
164 return is_legacy ? mm->mmap_legacy_base : mm->mmap_base;
165 }
166
arch_vma_name(struct vm_area_struct * vma)167 const char *arch_vma_name(struct vm_area_struct *vma)
168 {
169 return NULL;
170 }
171
172 /**
173 * mmap_address_hint_valid - Validate the address hint of mmap
174 * @addr: Address hint
175 * @len: Mapping length
176 *
177 * Check whether @addr and @addr + @len result in a valid mapping.
178 *
179 * On 32bit this only checks whether @addr + @len is <= TASK_SIZE.
180 *
181 * On 64bit with 5-level page tables another sanity check is required
182 * because mappings requested by mmap(@addr, 0) which cross the 47-bit
183 * virtual address boundary can cause the following theoretical issue:
184 *
185 * An application calls mmap(addr, 0), i.e. without MAP_FIXED, where @addr
186 * is below the border of the 47-bit address space and @addr + @len is
187 * above the border.
188 *
189 * With 4-level paging this request succeeds, but the resulting mapping
190 * address will always be within the 47-bit virtual address space, because
191 * the hint address does not result in a valid mapping and is
192 * ignored. Hence applications which are not prepared to handle virtual
193 * addresses above 47-bit work correctly.
194 *
195 * With 5-level paging this request would be granted and result in a
196 * mapping which crosses the border of the 47-bit virtual address
197 * space. If the application cannot handle addresses above 47-bit this
198 * will lead to misbehaviour and hard to diagnose failures.
199 *
200 * Therefore ignore address hints which would result in a mapping crossing
201 * the 47-bit virtual address boundary.
202 *
203 * Note, that in the same scenario with MAP_FIXED the behaviour is
204 * different. The request with @addr < 47-bit and @addr + @len > 47-bit
205 * fails on a 4-level paging machine but succeeds on a 5-level paging
206 * machine. It is reasonable to expect that an application does not rely on
207 * the failure of such a fixed mapping request, so the restriction is not
208 * applied.
209 */
mmap_address_hint_valid(unsigned long addr,unsigned long len)210 bool mmap_address_hint_valid(unsigned long addr, unsigned long len)
211 {
212 if (TASK_SIZE - len < addr)
213 return false;
214
215 return (addr > DEFAULT_MAP_WINDOW) == (addr + len > DEFAULT_MAP_WINDOW);
216 }
217
218 /* Can we access it for direct reading/writing? Must be RAM: */
valid_phys_addr_range(phys_addr_t addr,size_t count)219 int valid_phys_addr_range(phys_addr_t addr, size_t count)
220 {
221 return addr + count - 1 <= __pa(high_memory - 1);
222 }
223
224 /* Can we access it through mmap? Must be a valid physical address: */
valid_mmap_phys_addr_range(unsigned long pfn,size_t count)225 int valid_mmap_phys_addr_range(unsigned long pfn, size_t count)
226 {
227 phys_addr_t addr = (phys_addr_t)pfn << PAGE_SHIFT;
228
229 return phys_addr_valid(addr + count - 1);
230 }
231
232 /*
233 * Only allow root to set high MMIO mappings to PROT_NONE.
234 * This prevents an unpriv. user to set them to PROT_NONE and invert
235 * them, then pointing to valid memory for L1TF speculation.
236 *
237 * Note: for locked down kernels may want to disable the root override.
238 */
pfn_modify_allowed(unsigned long pfn,pgprot_t prot)239 bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
240 {
241 if (!boot_cpu_has_bug(X86_BUG_L1TF))
242 return true;
243 if (!__pte_needs_invert(pgprot_val(prot)))
244 return true;
245 /* If it's real memory always allow */
246 if (pfn_valid(pfn))
247 return true;
248 if (pfn >= l1tf_pfn_limit() && !capable(CAP_SYS_ADMIN))
249 return false;
250 return true;
251 }
252