xref: /linux/arch/x86/mm/mmap.c (revision b68fc09be48edbc47de1a0f3d42ef8adf6c0ac55)
1 /*
2  * Flexible mmap layout support
3  *
4  * Based on code by Ingo Molnar and Andi Kleen, copyrighted
5  * as follows:
6  *
7  * Copyright 2003-2009 Red Hat Inc.
8  * All Rights Reserved.
9  * Copyright 2005 Andi Kleen, SUSE Labs.
10  * Copyright 2007 Jiri Kosina, SUSE Labs.
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License as published by
14  * the Free Software Foundation; either version 2 of the License, or
15  * (at your option) any later version.
16  *
17  * This program is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20  * GNU General Public License for more details.
21  *
22  * You should have received a copy of the GNU General Public License
23  * along with this program; if not, write to the Free Software
24  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
25  */
26 
27 #include <linux/personality.h>
28 #include <linux/mm.h>
29 #include <linux/random.h>
30 #include <linux/limits.h>
31 #include <linux/sched/signal.h>
32 #include <linux/sched/mm.h>
33 #include <linux/compat.h>
34 #include <asm/elf.h>
35 
36 #include "physaddr.h"
37 
38 struct va_alignment __read_mostly va_align = {
39 	.flags = -1,
40 };
41 
42 unsigned long task_size_32bit(void)
43 {
44 	return IA32_PAGE_OFFSET;
45 }
46 
47 unsigned long task_size_64bit(int full_addr_space)
48 {
49 	return full_addr_space ? TASK_SIZE_MAX : DEFAULT_MAP_WINDOW;
50 }
51 
52 static unsigned long stack_maxrandom_size(unsigned long task_size)
53 {
54 	unsigned long max = 0;
55 	if (current->flags & PF_RANDOMIZE) {
56 		max = (-1UL) & __STACK_RND_MASK(task_size == task_size_32bit());
57 		max <<= PAGE_SHIFT;
58 	}
59 
60 	return max;
61 }
62 
63 #ifdef CONFIG_COMPAT
64 # define mmap32_rnd_bits  mmap_rnd_compat_bits
65 # define mmap64_rnd_bits  mmap_rnd_bits
66 #else
67 # define mmap32_rnd_bits  mmap_rnd_bits
68 # define mmap64_rnd_bits  mmap_rnd_bits
69 #endif
70 
71 #define SIZE_128M    (128 * 1024 * 1024UL)
72 
73 static int mmap_is_legacy(void)
74 {
75 	if (current->personality & ADDR_COMPAT_LAYOUT)
76 		return 1;
77 
78 	return sysctl_legacy_va_layout;
79 }
80 
81 static unsigned long arch_rnd(unsigned int rndbits)
82 {
83 	if (!(current->flags & PF_RANDOMIZE))
84 		return 0;
85 	return (get_random_long() & ((1UL << rndbits) - 1)) << PAGE_SHIFT;
86 }
87 
88 unsigned long arch_mmap_rnd(void)
89 {
90 	return arch_rnd(mmap_is_ia32() ? mmap32_rnd_bits : mmap64_rnd_bits);
91 }
92 
93 static unsigned long mmap_base(unsigned long rnd, unsigned long task_size,
94 			       struct rlimit *rlim_stack)
95 {
96 	unsigned long gap = rlim_stack->rlim_cur;
97 	unsigned long pad = stack_maxrandom_size(task_size) + stack_guard_gap;
98 	unsigned long gap_min, gap_max;
99 
100 	/* Values close to RLIM_INFINITY can overflow. */
101 	if (gap + pad > gap)
102 		gap += pad;
103 
104 	/*
105 	 * Top of mmap area (just below the process stack).
106 	 * Leave an at least ~128 MB hole with possible stack randomization.
107 	 */
108 	gap_min = SIZE_128M;
109 	gap_max = (task_size / 6) * 5;
110 
111 	if (gap < gap_min)
112 		gap = gap_min;
113 	else if (gap > gap_max)
114 		gap = gap_max;
115 
116 	return PAGE_ALIGN(task_size - gap - rnd);
117 }
118 
119 static unsigned long mmap_legacy_base(unsigned long rnd,
120 				      unsigned long task_size)
121 {
122 	return __TASK_UNMAPPED_BASE(task_size) + rnd;
123 }
124 
125 /*
126  * This function, called very early during the creation of a new
127  * process VM image, sets up which VM layout function to use:
128  */
129 static void arch_pick_mmap_base(unsigned long *base, unsigned long *legacy_base,
130 		unsigned long random_factor, unsigned long task_size,
131 		struct rlimit *rlim_stack)
132 {
133 	*legacy_base = mmap_legacy_base(random_factor, task_size);
134 	if (mmap_is_legacy())
135 		*base = *legacy_base;
136 	else
137 		*base = mmap_base(random_factor, task_size, rlim_stack);
138 }
139 
140 void arch_pick_mmap_layout(struct mm_struct *mm, struct rlimit *rlim_stack)
141 {
142 	if (mmap_is_legacy())
143 		mm->get_unmapped_area = arch_get_unmapped_area;
144 	else
145 		mm->get_unmapped_area = arch_get_unmapped_area_topdown;
146 
147 	arch_pick_mmap_base(&mm->mmap_base, &mm->mmap_legacy_base,
148 			arch_rnd(mmap64_rnd_bits), task_size_64bit(0),
149 			rlim_stack);
150 
151 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
152 	/*
153 	 * The mmap syscall mapping base decision depends solely on the
154 	 * syscall type (64-bit or compat). This applies for 64bit
155 	 * applications and 32bit applications. The 64bit syscall uses
156 	 * mmap_base, the compat syscall uses mmap_compat_base.
157 	 */
158 	arch_pick_mmap_base(&mm->mmap_compat_base, &mm->mmap_compat_legacy_base,
159 			arch_rnd(mmap32_rnd_bits), task_size_32bit(),
160 			rlim_stack);
161 #endif
162 }
163 
164 unsigned long get_mmap_base(int is_legacy)
165 {
166 	struct mm_struct *mm = current->mm;
167 
168 #ifdef CONFIG_HAVE_ARCH_COMPAT_MMAP_BASES
169 	if (in_compat_syscall()) {
170 		return is_legacy ? mm->mmap_compat_legacy_base
171 				 : mm->mmap_compat_base;
172 	}
173 #endif
174 	return is_legacy ? mm->mmap_legacy_base : mm->mmap_base;
175 }
176 
177 const char *arch_vma_name(struct vm_area_struct *vma)
178 {
179 	if (vma->vm_flags & VM_MPX)
180 		return "[mpx]";
181 	return NULL;
182 }
183 
184 /**
185  * mmap_address_hint_valid - Validate the address hint of mmap
186  * @addr:	Address hint
187  * @len:	Mapping length
188  *
189  * Check whether @addr and @addr + @len result in a valid mapping.
190  *
191  * On 32bit this only checks whether @addr + @len is <= TASK_SIZE.
192  *
193  * On 64bit with 5-level page tables another sanity check is required
194  * because mappings requested by mmap(@addr, 0) which cross the 47-bit
195  * virtual address boundary can cause the following theoretical issue:
196  *
197  *  An application calls mmap(addr, 0), i.e. without MAP_FIXED, where @addr
198  *  is below the border of the 47-bit address space and @addr + @len is
199  *  above the border.
200  *
201  *  With 4-level paging this request succeeds, but the resulting mapping
202  *  address will always be within the 47-bit virtual address space, because
203  *  the hint address does not result in a valid mapping and is
204  *  ignored. Hence applications which are not prepared to handle virtual
205  *  addresses above 47-bit work correctly.
206  *
207  *  With 5-level paging this request would be granted and result in a
208  *  mapping which crosses the border of the 47-bit virtual address
209  *  space. If the application cannot handle addresses above 47-bit this
210  *  will lead to misbehaviour and hard to diagnose failures.
211  *
212  * Therefore ignore address hints which would result in a mapping crossing
213  * the 47-bit virtual address boundary.
214  *
215  * Note, that in the same scenario with MAP_FIXED the behaviour is
216  * different. The request with @addr < 47-bit and @addr + @len > 47-bit
217  * fails on a 4-level paging machine but succeeds on a 5-level paging
218  * machine. It is reasonable to expect that an application does not rely on
219  * the failure of such a fixed mapping request, so the restriction is not
220  * applied.
221  */
222 bool mmap_address_hint_valid(unsigned long addr, unsigned long len)
223 {
224 	if (TASK_SIZE - len < addr)
225 		return false;
226 
227 	return (addr > DEFAULT_MAP_WINDOW) == (addr + len > DEFAULT_MAP_WINDOW);
228 }
229 
230 /* Can we access it for direct reading/writing? Must be RAM: */
231 int valid_phys_addr_range(phys_addr_t addr, size_t count)
232 {
233 	return addr + count <= __pa(high_memory);
234 }
235 
236 /* Can we access it through mmap? Must be a valid physical address: */
237 int valid_mmap_phys_addr_range(unsigned long pfn, size_t count)
238 {
239 	phys_addr_t addr = (phys_addr_t)pfn << PAGE_SHIFT;
240 
241 	return phys_addr_valid(addr + count - 1);
242 }
243 
244 /*
245  * Only allow root to set high MMIO mappings to PROT_NONE.
246  * This prevents an unpriv. user to set them to PROT_NONE and invert
247  * them, then pointing to valid memory for L1TF speculation.
248  *
249  * Note: for locked down kernels may want to disable the root override.
250  */
251 bool pfn_modify_allowed(unsigned long pfn, pgprot_t prot)
252 {
253 	if (!boot_cpu_has_bug(X86_BUG_L1TF))
254 		return true;
255 	if (!__pte_needs_invert(pgprot_val(prot)))
256 		return true;
257 	/* If it's real memory always allow */
258 	if (pfn_valid(pfn))
259 		return true;
260 	if (pfn >= l1tf_pfn_limit() && !capable(CAP_SYS_ADMIN))
261 		return false;
262 	return true;
263 }
264