Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
4 * <benh@kernel.crashing.org>
5 * Copyright (C) 2012 ARM Limited
6 * Copyright (C) 2015 Regents of the University of California
7 */
8
9#include <linux/elf.h>
10#include <linux/mm.h>
11#include <linux/slab.h>
12#include <linux/binfmts.h>
13#include <linux/err.h>
14#include <asm/page.h>
15#include <asm/vdso.h>
16#include <linux/time_namespace.h>
17#include <vdso/datapage.h>
18#include <vdso/vsyscall.h>
19
20enum vvar_pages {
21 VVAR_DATA_PAGE_OFFSET,
22 VVAR_TIMENS_PAGE_OFFSET,
23 VVAR_NR_PAGES,
24};
25
26enum rv_vdso_map {
27 RV_VDSO_MAP_VVAR,
28 RV_VDSO_MAP_VDSO,
29};
30
31#define VVAR_SIZE (VVAR_NR_PAGES << PAGE_SHIFT)
32
33/*
34 * The vDSO data page.
35 */
36static union {
37 struct vdso_data data;
38 u8 page[PAGE_SIZE];
39} vdso_data_store __page_aligned_data;
40struct vdso_data *vdso_data = &vdso_data_store.data;
41
42struct __vdso_info {
43 const char *name;
44 const char *vdso_code_start;
45 const char *vdso_code_end;
46 unsigned long vdso_pages;
47 /* Data Mapping */
48 struct vm_special_mapping *dm;
49 /* Code Mapping */
50 struct vm_special_mapping *cm;
51};
52
53static struct __vdso_info vdso_info;
54#ifdef CONFIG_COMPAT
55static struct __vdso_info compat_vdso_info;
56#endif
57
58static int vdso_mremap(const struct vm_special_mapping *sm,
59 struct vm_area_struct *new_vma)
60{
61 current->mm->context.vdso = (void *)new_vma->vm_start;
62
63 return 0;
64}
65
66static void __init __vdso_init(struct __vdso_info *vdso_info)
67{
68 unsigned int i;
69 struct page **vdso_pagelist;
70 unsigned long pfn;
71
72 if (memcmp(vdso_info->vdso_code_start, "\177ELF", 4))
73 panic("vDSO is not a valid ELF object!\n");
74
75 vdso_info->vdso_pages = (
76 vdso_info->vdso_code_end -
77 vdso_info->vdso_code_start) >>
78 PAGE_SHIFT;
79
80 vdso_pagelist = kcalloc(vdso_info->vdso_pages,
81 sizeof(struct page *),
82 GFP_KERNEL);
83 if (vdso_pagelist == NULL)
84 panic("vDSO kcalloc failed!\n");
85
86 /* Grab the vDSO code pages. */
87 pfn = sym_to_pfn(vdso_info->vdso_code_start);
88
89 for (i = 0; i < vdso_info->vdso_pages; i++)
90 vdso_pagelist[i] = pfn_to_page(pfn + i);
91
92 vdso_info->cm->pages = vdso_pagelist;
93}
94
95#ifdef CONFIG_TIME_NS
96struct vdso_data *arch_get_vdso_data(void *vvar_page)
97{
98 return (struct vdso_data *)(vvar_page);
99}
100
101/*
102 * The vvar mapping contains data for a specific time namespace, so when a task
103 * changes namespace we must unmap its vvar data for the old namespace.
104 * Subsequent faults will map in data for the new namespace.
105 *
106 * For more details see timens_setup_vdso_data().
107 */
108int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
109{
110 struct mm_struct *mm = task->mm;
111 struct vm_area_struct *vma;
112 VMA_ITERATOR(vmi, mm, 0);
113
114 mmap_read_lock(mm);
115
116 for_each_vma(vmi, vma) {
117 if (vma_is_special_mapping(vma, vdso_info.dm))
118 zap_vma_pages(vma);
119#ifdef CONFIG_COMPAT
120 if (vma_is_special_mapping(vma, compat_vdso_info.dm))
121 zap_vma_pages(vma);
122#endif
123 }
124
125 mmap_read_unlock(mm);
126 return 0;
127}
128#endif
129
130static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
131 struct vm_area_struct *vma, struct vm_fault *vmf)
132{
133 struct page *timens_page = find_timens_vvar_page(vma);
134 unsigned long pfn;
135
136 switch (vmf->pgoff) {
137 case VVAR_DATA_PAGE_OFFSET:
138 if (timens_page)
139 pfn = page_to_pfn(timens_page);
140 else
141 pfn = sym_to_pfn(vdso_data);
142 break;
143#ifdef CONFIG_TIME_NS
144 case VVAR_TIMENS_PAGE_OFFSET:
145 /*
146 * If a task belongs to a time namespace then a namespace
147 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
148 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
149 * offset.
150 * See also the comment near timens_setup_vdso_data().
151 */
152 if (!timens_page)
153 return VM_FAULT_SIGBUS;
154 pfn = sym_to_pfn(vdso_data);
155 break;
156#endif /* CONFIG_TIME_NS */
157 default:
158 return VM_FAULT_SIGBUS;
159 }
160
161 return vmf_insert_pfn(vma, vmf->address, pfn);
162}
163
164static struct vm_special_mapping rv_vdso_maps[] __ro_after_init = {
165 [RV_VDSO_MAP_VVAR] = {
166 .name = "[vvar]",
167 .fault = vvar_fault,
168 },
169 [RV_VDSO_MAP_VDSO] = {
170 .name = "[vdso]",
171 .mremap = vdso_mremap,
172 },
173};
174
175static struct __vdso_info vdso_info __ro_after_init = {
176 .name = "vdso",
177 .vdso_code_start = vdso_start,
178 .vdso_code_end = vdso_end,
179 .dm = &rv_vdso_maps[RV_VDSO_MAP_VVAR],
180 .cm = &rv_vdso_maps[RV_VDSO_MAP_VDSO],
181};
182
183#ifdef CONFIG_COMPAT
184static struct vm_special_mapping rv_compat_vdso_maps[] __ro_after_init = {
185 [RV_VDSO_MAP_VVAR] = {
186 .name = "[vvar]",
187 .fault = vvar_fault,
188 },
189 [RV_VDSO_MAP_VDSO] = {
190 .name = "[vdso]",
191 .mremap = vdso_mremap,
192 },
193};
194
195static struct __vdso_info compat_vdso_info __ro_after_init = {
196 .name = "compat_vdso",
197 .vdso_code_start = compat_vdso_start,
198 .vdso_code_end = compat_vdso_end,
199 .dm = &rv_compat_vdso_maps[RV_VDSO_MAP_VVAR],
200 .cm = &rv_compat_vdso_maps[RV_VDSO_MAP_VDSO],
201};
202#endif
203
204static int __init vdso_init(void)
205{
206 __vdso_init(&vdso_info);
207#ifdef CONFIG_COMPAT
208 __vdso_init(&compat_vdso_info);
209#endif
210
211 return 0;
212}
213arch_initcall(vdso_init);
214
215static int __setup_additional_pages(struct mm_struct *mm,
216 struct linux_binprm *bprm,
217 int uses_interp,
218 struct __vdso_info *vdso_info)
219{
220 unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
221 void *ret;
222
223 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
224
225 vdso_text_len = vdso_info->vdso_pages << PAGE_SHIFT;
226 /* Be sure to map the data page */
227 vdso_mapping_len = vdso_text_len + VVAR_SIZE;
228
229 vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
230 if (IS_ERR_VALUE(vdso_base)) {
231 ret = ERR_PTR(vdso_base);
232 goto up_fail;
233 }
234
235 ret = _install_special_mapping(mm, vdso_base, VVAR_SIZE,
236 (VM_READ | VM_MAYREAD | VM_PFNMAP), vdso_info->dm);
237 if (IS_ERR(ret))
238 goto up_fail;
239
240 vdso_base += VVAR_SIZE;
241 mm->context.vdso = (void *)vdso_base;
242
243 ret =
244 _install_special_mapping(mm, vdso_base, vdso_text_len,
245 (VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC),
246 vdso_info->cm);
247
248 if (IS_ERR(ret))
249 goto up_fail;
250
251 return 0;
252
253up_fail:
254 mm->context.vdso = NULL;
255 return PTR_ERR(ret);
256}
257
258#ifdef CONFIG_COMPAT
259int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
260 int uses_interp)
261{
262 struct mm_struct *mm = current->mm;
263 int ret;
264
265 if (mmap_write_lock_killable(mm))
266 return -EINTR;
267
268 ret = __setup_additional_pages(mm, bprm, uses_interp,
269 &compat_vdso_info);
270 mmap_write_unlock(mm);
271
272 return ret;
273}
274#endif
275
276int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
277{
278 struct mm_struct *mm = current->mm;
279 int ret;
280
281 if (mmap_write_lock_killable(mm))
282 return -EINTR;
283
284 ret = __setup_additional_pages(mm, bprm, uses_interp, &vdso_info);
285 mmap_write_unlock(mm);
286
287 return ret;
288}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
4 * <benh@kernel.crashing.org>
5 * Copyright (C) 2012 ARM Limited
6 * Copyright (C) 2015 Regents of the University of California
7 */
8
9#include <linux/elf.h>
10#include <linux/mm.h>
11#include <linux/slab.h>
12#include <linux/binfmts.h>
13#include <linux/err.h>
14#include <asm/page.h>
15#include <asm/vdso.h>
16#include <linux/time_namespace.h>
17#include <vdso/datapage.h>
18#include <vdso/vsyscall.h>
19
20enum vvar_pages {
21 VVAR_DATA_PAGE_OFFSET,
22 VVAR_TIMENS_PAGE_OFFSET,
23 VVAR_NR_PAGES,
24};
25
26enum rv_vdso_map {
27 RV_VDSO_MAP_VVAR,
28 RV_VDSO_MAP_VDSO,
29};
30
31#define VVAR_SIZE (VVAR_NR_PAGES << PAGE_SHIFT)
32
33static union vdso_data_store vdso_data_store __page_aligned_data;
34struct vdso_data *vdso_data = vdso_data_store.data;
35
36struct __vdso_info {
37 const char *name;
38 const char *vdso_code_start;
39 const char *vdso_code_end;
40 unsigned long vdso_pages;
41 /* Data Mapping */
42 struct vm_special_mapping *dm;
43 /* Code Mapping */
44 struct vm_special_mapping *cm;
45};
46
47static struct __vdso_info vdso_info;
48#ifdef CONFIG_COMPAT
49static struct __vdso_info compat_vdso_info;
50#endif
51
52static int vdso_mremap(const struct vm_special_mapping *sm,
53 struct vm_area_struct *new_vma)
54{
55 current->mm->context.vdso = (void *)new_vma->vm_start;
56
57 return 0;
58}
59
60static void __init __vdso_init(struct __vdso_info *vdso_info)
61{
62 unsigned int i;
63 struct page **vdso_pagelist;
64 unsigned long pfn;
65
66 if (memcmp(vdso_info->vdso_code_start, "\177ELF", 4))
67 panic("vDSO is not a valid ELF object!\n");
68
69 vdso_info->vdso_pages = (
70 vdso_info->vdso_code_end -
71 vdso_info->vdso_code_start) >>
72 PAGE_SHIFT;
73
74 vdso_pagelist = kcalloc(vdso_info->vdso_pages,
75 sizeof(struct page *),
76 GFP_KERNEL);
77 if (vdso_pagelist == NULL)
78 panic("vDSO kcalloc failed!\n");
79
80 /* Grab the vDSO code pages. */
81 pfn = sym_to_pfn(vdso_info->vdso_code_start);
82
83 for (i = 0; i < vdso_info->vdso_pages; i++)
84 vdso_pagelist[i] = pfn_to_page(pfn + i);
85
86 vdso_info->cm->pages = vdso_pagelist;
87}
88
89#ifdef CONFIG_TIME_NS
90struct vdso_data *arch_get_vdso_data(void *vvar_page)
91{
92 return (struct vdso_data *)(vvar_page);
93}
94
95/*
96 * The vvar mapping contains data for a specific time namespace, so when a task
97 * changes namespace we must unmap its vvar data for the old namespace.
98 * Subsequent faults will map in data for the new namespace.
99 *
100 * For more details see timens_setup_vdso_data().
101 */
102int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
103{
104 struct mm_struct *mm = task->mm;
105 struct vm_area_struct *vma;
106 VMA_ITERATOR(vmi, mm, 0);
107
108 mmap_read_lock(mm);
109
110 for_each_vma(vmi, vma) {
111 if (vma_is_special_mapping(vma, vdso_info.dm))
112 zap_vma_pages(vma);
113#ifdef CONFIG_COMPAT
114 if (vma_is_special_mapping(vma, compat_vdso_info.dm))
115 zap_vma_pages(vma);
116#endif
117 }
118
119 mmap_read_unlock(mm);
120 return 0;
121}
122#endif
123
124static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
125 struct vm_area_struct *vma, struct vm_fault *vmf)
126{
127 struct page *timens_page = find_timens_vvar_page(vma);
128 unsigned long pfn;
129
130 switch (vmf->pgoff) {
131 case VVAR_DATA_PAGE_OFFSET:
132 if (timens_page)
133 pfn = page_to_pfn(timens_page);
134 else
135 pfn = sym_to_pfn(vdso_data);
136 break;
137#ifdef CONFIG_TIME_NS
138 case VVAR_TIMENS_PAGE_OFFSET:
139 /*
140 * If a task belongs to a time namespace then a namespace
141 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
142 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
143 * offset.
144 * See also the comment near timens_setup_vdso_data().
145 */
146 if (!timens_page)
147 return VM_FAULT_SIGBUS;
148 pfn = sym_to_pfn(vdso_data);
149 break;
150#endif /* CONFIG_TIME_NS */
151 default:
152 return VM_FAULT_SIGBUS;
153 }
154
155 return vmf_insert_pfn(vma, vmf->address, pfn);
156}
157
158static struct vm_special_mapping rv_vdso_maps[] __ro_after_init = {
159 [RV_VDSO_MAP_VVAR] = {
160 .name = "[vvar]",
161 .fault = vvar_fault,
162 },
163 [RV_VDSO_MAP_VDSO] = {
164 .name = "[vdso]",
165 .mremap = vdso_mremap,
166 },
167};
168
169static struct __vdso_info vdso_info __ro_after_init = {
170 .name = "vdso",
171 .vdso_code_start = vdso_start,
172 .vdso_code_end = vdso_end,
173 .dm = &rv_vdso_maps[RV_VDSO_MAP_VVAR],
174 .cm = &rv_vdso_maps[RV_VDSO_MAP_VDSO],
175};
176
177#ifdef CONFIG_COMPAT
178static struct vm_special_mapping rv_compat_vdso_maps[] __ro_after_init = {
179 [RV_VDSO_MAP_VVAR] = {
180 .name = "[vvar]",
181 .fault = vvar_fault,
182 },
183 [RV_VDSO_MAP_VDSO] = {
184 .name = "[vdso]",
185 .mremap = vdso_mremap,
186 },
187};
188
189static struct __vdso_info compat_vdso_info __ro_after_init = {
190 .name = "compat_vdso",
191 .vdso_code_start = compat_vdso_start,
192 .vdso_code_end = compat_vdso_end,
193 .dm = &rv_compat_vdso_maps[RV_VDSO_MAP_VVAR],
194 .cm = &rv_compat_vdso_maps[RV_VDSO_MAP_VDSO],
195};
196#endif
197
198static int __init vdso_init(void)
199{
200 __vdso_init(&vdso_info);
201#ifdef CONFIG_COMPAT
202 __vdso_init(&compat_vdso_info);
203#endif
204
205 return 0;
206}
207arch_initcall(vdso_init);
208
209static int __setup_additional_pages(struct mm_struct *mm,
210 struct linux_binprm *bprm,
211 int uses_interp,
212 struct __vdso_info *vdso_info)
213{
214 unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
215 void *ret;
216
217 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
218
219 vdso_text_len = vdso_info->vdso_pages << PAGE_SHIFT;
220 /* Be sure to map the data page */
221 vdso_mapping_len = vdso_text_len + VVAR_SIZE;
222
223 vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
224 if (IS_ERR_VALUE(vdso_base)) {
225 ret = ERR_PTR(vdso_base);
226 goto up_fail;
227 }
228
229 ret = _install_special_mapping(mm, vdso_base, VVAR_SIZE,
230 (VM_READ | VM_MAYREAD | VM_PFNMAP), vdso_info->dm);
231 if (IS_ERR(ret))
232 goto up_fail;
233
234 vdso_base += VVAR_SIZE;
235 mm->context.vdso = (void *)vdso_base;
236
237 ret =
238 _install_special_mapping(mm, vdso_base, vdso_text_len,
239 (VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC),
240 vdso_info->cm);
241
242 if (IS_ERR(ret))
243 goto up_fail;
244
245 return 0;
246
247up_fail:
248 mm->context.vdso = NULL;
249 return PTR_ERR(ret);
250}
251
252#ifdef CONFIG_COMPAT
253int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
254 int uses_interp)
255{
256 struct mm_struct *mm = current->mm;
257 int ret;
258
259 if (mmap_write_lock_killable(mm))
260 return -EINTR;
261
262 ret = __setup_additional_pages(mm, bprm, uses_interp,
263 &compat_vdso_info);
264 mmap_write_unlock(mm);
265
266 return ret;
267}
268#endif
269
270int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
271{
272 struct mm_struct *mm = current->mm;
273 int ret;
274
275 if (mmap_write_lock_killable(mm))
276 return -EINTR;
277
278 ret = __setup_additional_pages(mm, bprm, uses_interp, &vdso_info);
279 mmap_write_unlock(mm);
280
281 return ret;
282}