xref: /linux/arch/riscv/kernel/vdso.c (revision a251c17aa558d8e3128a528af5cf8b9d7caae4fd)
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 
18 #ifdef CONFIG_GENERIC_TIME_VSYSCALL
19 #include <vdso/datapage.h>
20 #else
21 struct vdso_data {
22 };
23 #endif
24 
25 extern char vdso_start[], vdso_end[];
26 #ifdef CONFIG_COMPAT
27 extern char compat_vdso_start[], compat_vdso_end[];
28 #endif
29 
30 enum vvar_pages {
31 	VVAR_DATA_PAGE_OFFSET,
32 	VVAR_TIMENS_PAGE_OFFSET,
33 	VVAR_NR_PAGES,
34 };
35 
36 enum rv_vdso_map {
37 	RV_VDSO_MAP_VVAR,
38 	RV_VDSO_MAP_VDSO,
39 };
40 
41 #define VVAR_SIZE  (VVAR_NR_PAGES << PAGE_SHIFT)
42 
43 /*
44  * The vDSO data page.
45  */
46 static union {
47 	struct vdso_data	data;
48 	u8			page[PAGE_SIZE];
49 } vdso_data_store __page_aligned_data;
50 struct vdso_data *vdso_data = &vdso_data_store.data;
51 
52 struct __vdso_info {
53 	const char *name;
54 	const char *vdso_code_start;
55 	const char *vdso_code_end;
56 	unsigned long vdso_pages;
57 	/* Data Mapping */
58 	struct vm_special_mapping *dm;
59 	/* Code Mapping */
60 	struct vm_special_mapping *cm;
61 };
62 
63 static int vdso_mremap(const struct vm_special_mapping *sm,
64 		       struct vm_area_struct *new_vma)
65 {
66 	current->mm->context.vdso = (void *)new_vma->vm_start;
67 
68 	return 0;
69 }
70 
71 static void __init __vdso_init(struct __vdso_info *vdso_info)
72 {
73 	unsigned int i;
74 	struct page **vdso_pagelist;
75 	unsigned long pfn;
76 
77 	if (memcmp(vdso_info->vdso_code_start, "\177ELF", 4))
78 		panic("vDSO is not a valid ELF object!\n");
79 
80 	vdso_info->vdso_pages = (
81 		vdso_info->vdso_code_end -
82 		vdso_info->vdso_code_start) >>
83 		PAGE_SHIFT;
84 
85 	vdso_pagelist = kcalloc(vdso_info->vdso_pages,
86 				sizeof(struct page *),
87 				GFP_KERNEL);
88 	if (vdso_pagelist == NULL)
89 		panic("vDSO kcalloc failed!\n");
90 
91 	/* Grab the vDSO code pages. */
92 	pfn = sym_to_pfn(vdso_info->vdso_code_start);
93 
94 	for (i = 0; i < vdso_info->vdso_pages; i++)
95 		vdso_pagelist[i] = pfn_to_page(pfn + i);
96 
97 	vdso_info->cm->pages = vdso_pagelist;
98 }
99 
100 #ifdef CONFIG_TIME_NS
101 struct vdso_data *arch_get_vdso_data(void *vvar_page)
102 {
103 	return (struct vdso_data *)(vvar_page);
104 }
105 
106 /*
107  * The vvar mapping contains data for a specific time namespace, so when a task
108  * changes namespace we must unmap its vvar data for the old namespace.
109  * Subsequent faults will map in data for the new namespace.
110  *
111  * For more details see timens_setup_vdso_data().
112  */
113 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
114 {
115 	struct mm_struct *mm = task->mm;
116 	struct vm_area_struct *vma;
117 	VMA_ITERATOR(vmi, mm, 0);
118 	struct __vdso_info *vdso_info = mm->context.vdso_info;
119 
120 	mmap_read_lock(mm);
121 
122 	for_each_vma(vmi, vma) {
123 		unsigned long size = vma->vm_end - vma->vm_start;
124 
125 		if (vma_is_special_mapping(vma, vdso_info->dm))
126 			zap_page_range(vma, vma->vm_start, size);
127 	}
128 
129 	mmap_read_unlock(mm);
130 	return 0;
131 }
132 
133 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
134 {
135 	if (likely(vma->vm_mm == current->mm))
136 		return current->nsproxy->time_ns->vvar_page;
137 
138 	/*
139 	 * VM_PFNMAP | VM_IO protect .fault() handler from being called
140 	 * through interfaces like /proc/$pid/mem or
141 	 * process_vm_{readv,writev}() as long as there's no .access()
142 	 * in special_mapping_vmops.
143 	 * For more details check_vma_flags() and __access_remote_vm()
144 	 */
145 	WARN(1, "vvar_page accessed remotely");
146 
147 	return NULL;
148 }
149 #else
150 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
151 {
152 	return NULL;
153 }
154 #endif
155 
156 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
157 			     struct vm_area_struct *vma, struct vm_fault *vmf)
158 {
159 	struct page *timens_page = find_timens_vvar_page(vma);
160 	unsigned long pfn;
161 
162 	switch (vmf->pgoff) {
163 	case VVAR_DATA_PAGE_OFFSET:
164 		if (timens_page)
165 			pfn = page_to_pfn(timens_page);
166 		else
167 			pfn = sym_to_pfn(vdso_data);
168 		break;
169 #ifdef CONFIG_TIME_NS
170 	case VVAR_TIMENS_PAGE_OFFSET:
171 		/*
172 		 * If a task belongs to a time namespace then a namespace
173 		 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
174 		 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
175 		 * offset.
176 		 * See also the comment near timens_setup_vdso_data().
177 		 */
178 		if (!timens_page)
179 			return VM_FAULT_SIGBUS;
180 		pfn = sym_to_pfn(vdso_data);
181 		break;
182 #endif /* CONFIG_TIME_NS */
183 	default:
184 		return VM_FAULT_SIGBUS;
185 	}
186 
187 	return vmf_insert_pfn(vma, vmf->address, pfn);
188 }
189 
190 static struct vm_special_mapping rv_vdso_maps[] __ro_after_init = {
191 	[RV_VDSO_MAP_VVAR] = {
192 		.name   = "[vvar]",
193 		.fault = vvar_fault,
194 	},
195 	[RV_VDSO_MAP_VDSO] = {
196 		.name   = "[vdso]",
197 		.mremap = vdso_mremap,
198 	},
199 };
200 
201 static struct __vdso_info vdso_info __ro_after_init = {
202 	.name = "vdso",
203 	.vdso_code_start = vdso_start,
204 	.vdso_code_end = vdso_end,
205 	.dm = &rv_vdso_maps[RV_VDSO_MAP_VVAR],
206 	.cm = &rv_vdso_maps[RV_VDSO_MAP_VDSO],
207 };
208 
209 #ifdef CONFIG_COMPAT
210 static struct vm_special_mapping rv_compat_vdso_maps[] __ro_after_init = {
211 	[RV_VDSO_MAP_VVAR] = {
212 		.name   = "[vvar]",
213 		.fault = vvar_fault,
214 	},
215 	[RV_VDSO_MAP_VDSO] = {
216 		.name   = "[vdso]",
217 		.mremap = vdso_mremap,
218 	},
219 };
220 
221 static struct __vdso_info compat_vdso_info __ro_after_init = {
222 	.name = "compat_vdso",
223 	.vdso_code_start = compat_vdso_start,
224 	.vdso_code_end = compat_vdso_end,
225 	.dm = &rv_compat_vdso_maps[RV_VDSO_MAP_VVAR],
226 	.cm = &rv_compat_vdso_maps[RV_VDSO_MAP_VDSO],
227 };
228 #endif
229 
230 static int __init vdso_init(void)
231 {
232 	__vdso_init(&vdso_info);
233 #ifdef CONFIG_COMPAT
234 	__vdso_init(&compat_vdso_info);
235 #endif
236 
237 	return 0;
238 }
239 arch_initcall(vdso_init);
240 
241 static int __setup_additional_pages(struct mm_struct *mm,
242 				    struct linux_binprm *bprm,
243 				    int uses_interp,
244 				    struct __vdso_info *vdso_info)
245 {
246 	unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
247 	void *ret;
248 
249 	BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
250 
251 	vdso_text_len = vdso_info->vdso_pages << PAGE_SHIFT;
252 	/* Be sure to map the data page */
253 	vdso_mapping_len = vdso_text_len + VVAR_SIZE;
254 
255 	vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
256 	if (IS_ERR_VALUE(vdso_base)) {
257 		ret = ERR_PTR(vdso_base);
258 		goto up_fail;
259 	}
260 
261 	ret = _install_special_mapping(mm, vdso_base, VVAR_SIZE,
262 		(VM_READ | VM_MAYREAD | VM_PFNMAP), vdso_info->dm);
263 	if (IS_ERR(ret))
264 		goto up_fail;
265 
266 	vdso_base += VVAR_SIZE;
267 	mm->context.vdso = (void *)vdso_base;
268 	mm->context.vdso_info = (void *)vdso_info;
269 
270 	ret =
271 	   _install_special_mapping(mm, vdso_base, vdso_text_len,
272 		(VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC),
273 		vdso_info->cm);
274 
275 	if (IS_ERR(ret))
276 		goto up_fail;
277 
278 	return 0;
279 
280 up_fail:
281 	mm->context.vdso = NULL;
282 	return PTR_ERR(ret);
283 }
284 
285 #ifdef CONFIG_COMPAT
286 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
287 				       int uses_interp)
288 {
289 	struct mm_struct *mm = current->mm;
290 	int ret;
291 
292 	if (mmap_write_lock_killable(mm))
293 		return -EINTR;
294 
295 	ret = __setup_additional_pages(mm, bprm, uses_interp,
296 							&compat_vdso_info);
297 	mmap_write_unlock(mm);
298 
299 	return ret;
300 }
301 #endif
302 
303 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
304 {
305 	struct mm_struct *mm = current->mm;
306 	int ret;
307 
308 	if (mmap_write_lock_killable(mm))
309 		return -EINTR;
310 
311 	ret = __setup_additional_pages(mm, bprm, uses_interp, &vdso_info);
312 	mmap_write_unlock(mm);
313 
314 	return ret;
315 }
316