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