1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (c) 2014, The Linux Foundation. All rights reserved.
4 */
5 #include <linux/kernel.h>
6 #include <linux/mm.h>
7 #include <linux/module.h>
8 #include <linux/sched.h>
9 #include <linux/vmalloc.h>
10
11 #include <asm/cacheflush.h>
12 #include <asm/set_memory.h>
13 #include <asm/tlbflush.h>
14 #include <asm/kfence.h>
15
16 struct page_change_data {
17 pgprot_t set_mask;
18 pgprot_t clear_mask;
19 };
20
21 bool rodata_full __ro_after_init = IS_ENABLED(CONFIG_RODATA_FULL_DEFAULT_ENABLED);
22
can_set_direct_map(void)23 bool can_set_direct_map(void)
24 {
25 /*
26 * rodata_full and DEBUG_PAGEALLOC require linear map to be
27 * mapped at page granularity, so that it is possible to
28 * protect/unprotect single pages.
29 *
30 * KFENCE pool requires page-granular mapping if initialized late.
31 */
32 return (rodata_enabled && rodata_full) || debug_pagealloc_enabled() ||
33 arm64_kfence_can_set_direct_map();
34 }
35
change_page_range(pte_t * ptep,unsigned long addr,void * data)36 static int change_page_range(pte_t *ptep, unsigned long addr, void *data)
37 {
38 struct page_change_data *cdata = data;
39 pte_t pte = READ_ONCE(*ptep);
40
41 pte = clear_pte_bit(pte, cdata->clear_mask);
42 pte = set_pte_bit(pte, cdata->set_mask);
43
44 set_pte(ptep, pte);
45 return 0;
46 }
47
48 /*
49 * This function assumes that the range is mapped with PAGE_SIZE pages.
50 */
__change_memory_common(unsigned long start,unsigned long size,pgprot_t set_mask,pgprot_t clear_mask)51 static int __change_memory_common(unsigned long start, unsigned long size,
52 pgprot_t set_mask, pgprot_t clear_mask)
53 {
54 struct page_change_data data;
55 int ret;
56
57 data.set_mask = set_mask;
58 data.clear_mask = clear_mask;
59
60 ret = apply_to_page_range(&init_mm, start, size, change_page_range,
61 &data);
62
63 flush_tlb_kernel_range(start, start + size);
64 return ret;
65 }
66
change_memory_common(unsigned long addr,int numpages,pgprot_t set_mask,pgprot_t clear_mask)67 static int change_memory_common(unsigned long addr, int numpages,
68 pgprot_t set_mask, pgprot_t clear_mask)
69 {
70 unsigned long start = addr;
71 unsigned long size = PAGE_SIZE * numpages;
72 unsigned long end = start + size;
73 struct vm_struct *area;
74 int i;
75
76 if (!PAGE_ALIGNED(addr)) {
77 start &= PAGE_MASK;
78 end = start + size;
79 WARN_ON_ONCE(1);
80 }
81
82 /*
83 * Kernel VA mappings are always live, and splitting live section
84 * mappings into page mappings may cause TLB conflicts. This means
85 * we have to ensure that changing the permission bits of the range
86 * we are operating on does not result in such splitting.
87 *
88 * Let's restrict ourselves to mappings created by vmalloc (or vmap).
89 * Those are guaranteed to consist entirely of page mappings, and
90 * splitting is never needed.
91 *
92 * So check whether the [addr, addr + size) interval is entirely
93 * covered by precisely one VM area that has the VM_ALLOC flag set.
94 */
95 area = find_vm_area((void *)addr);
96 if (!area ||
97 end > (unsigned long)kasan_reset_tag(area->addr) + area->size ||
98 !(area->flags & VM_ALLOC))
99 return -EINVAL;
100
101 if (!numpages)
102 return 0;
103
104 /*
105 * If we are manipulating read-only permissions, apply the same
106 * change to the linear mapping of the pages that back this VM area.
107 */
108 if (rodata_enabled &&
109 rodata_full && (pgprot_val(set_mask) == PTE_RDONLY ||
110 pgprot_val(clear_mask) == PTE_RDONLY)) {
111 for (i = 0; i < area->nr_pages; i++) {
112 __change_memory_common((u64)page_address(area->pages[i]),
113 PAGE_SIZE, set_mask, clear_mask);
114 }
115 }
116
117 /*
118 * Get rid of potentially aliasing lazily unmapped vm areas that may
119 * have permissions set that deviate from the ones we are setting here.
120 */
121 vm_unmap_aliases();
122
123 return __change_memory_common(start, size, set_mask, clear_mask);
124 }
125
set_memory_ro(unsigned long addr,int numpages)126 int set_memory_ro(unsigned long addr, int numpages)
127 {
128 return change_memory_common(addr, numpages,
129 __pgprot(PTE_RDONLY),
130 __pgprot(PTE_WRITE));
131 }
132
set_memory_rw(unsigned long addr,int numpages)133 int set_memory_rw(unsigned long addr, int numpages)
134 {
135 return change_memory_common(addr, numpages,
136 __pgprot(PTE_WRITE),
137 __pgprot(PTE_RDONLY));
138 }
139
set_memory_nx(unsigned long addr,int numpages)140 int set_memory_nx(unsigned long addr, int numpages)
141 {
142 return change_memory_common(addr, numpages,
143 __pgprot(PTE_PXN),
144 __pgprot(PTE_MAYBE_GP));
145 }
146
set_memory_x(unsigned long addr,int numpages)147 int set_memory_x(unsigned long addr, int numpages)
148 {
149 return change_memory_common(addr, numpages,
150 __pgprot(PTE_MAYBE_GP),
151 __pgprot(PTE_PXN));
152 }
153
set_memory_valid(unsigned long addr,int numpages,int enable)154 int set_memory_valid(unsigned long addr, int numpages, int enable)
155 {
156 if (enable)
157 return __change_memory_common(addr, PAGE_SIZE * numpages,
158 __pgprot(PTE_VALID),
159 __pgprot(0));
160 else
161 return __change_memory_common(addr, PAGE_SIZE * numpages,
162 __pgprot(0),
163 __pgprot(PTE_VALID));
164 }
165
set_direct_map_invalid_noflush(struct page * page)166 int set_direct_map_invalid_noflush(struct page *page)
167 {
168 struct page_change_data data = {
169 .set_mask = __pgprot(0),
170 .clear_mask = __pgprot(PTE_VALID),
171 };
172
173 if (!can_set_direct_map())
174 return 0;
175
176 return apply_to_page_range(&init_mm,
177 (unsigned long)page_address(page),
178 PAGE_SIZE, change_page_range, &data);
179 }
180
set_direct_map_default_noflush(struct page * page)181 int set_direct_map_default_noflush(struct page *page)
182 {
183 struct page_change_data data = {
184 .set_mask = __pgprot(PTE_VALID | PTE_WRITE),
185 .clear_mask = __pgprot(PTE_RDONLY),
186 };
187
188 if (!can_set_direct_map())
189 return 0;
190
191 return apply_to_page_range(&init_mm,
192 (unsigned long)page_address(page),
193 PAGE_SIZE, change_page_range, &data);
194 }
195
196 #ifdef CONFIG_DEBUG_PAGEALLOC
__kernel_map_pages(struct page * page,int numpages,int enable)197 void __kernel_map_pages(struct page *page, int numpages, int enable)
198 {
199 if (!can_set_direct_map())
200 return;
201
202 set_memory_valid((unsigned long)page_address(page), numpages, enable);
203 }
204 #endif /* CONFIG_DEBUG_PAGEALLOC */
205
206 /*
207 * This function is used to determine if a linear map page has been marked as
208 * not-valid. Walk the page table and check the PTE_VALID bit.
209 *
210 * Because this is only called on the kernel linear map, p?d_sect() implies
211 * p?d_present(). When debug_pagealloc is enabled, sections mappings are
212 * disabled.
213 */
kernel_page_present(struct page * page)214 bool kernel_page_present(struct page *page)
215 {
216 pgd_t *pgdp;
217 p4d_t *p4dp;
218 pud_t *pudp, pud;
219 pmd_t *pmdp, pmd;
220 pte_t *ptep;
221 unsigned long addr = (unsigned long)page_address(page);
222
223 if (!can_set_direct_map())
224 return true;
225
226 pgdp = pgd_offset_k(addr);
227 if (pgd_none(READ_ONCE(*pgdp)))
228 return false;
229
230 p4dp = p4d_offset(pgdp, addr);
231 if (p4d_none(READ_ONCE(*p4dp)))
232 return false;
233
234 pudp = pud_offset(p4dp, addr);
235 pud = READ_ONCE(*pudp);
236 if (pud_none(pud))
237 return false;
238 if (pud_sect(pud))
239 return true;
240
241 pmdp = pmd_offset(pudp, addr);
242 pmd = READ_ONCE(*pmdp);
243 if (pmd_none(pmd))
244 return false;
245 if (pmd_sect(pmd))
246 return true;
247
248 ptep = pte_offset_kernel(pmdp, addr);
249 return pte_valid(READ_ONCE(*ptep));
250 }
251