xref: /linux/arch/arm64/kernel/vdso.c (revision 0973fed6a5e58dd2515a83dd7f83ad674e91cc4f) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VDSO implementations.
4  *
5  * Copyright (C) 2012 ARM Limited
6  *
7  * Author: Will Deacon <will.deacon@arm.com>
8  */
9 
10 #include <linux/cache.h>
11 #include <linux/clocksource.h>
12 #include <linux/elf.h>
13 #include <linux/err.h>
14 #include <linux/errno.h>
15 #include <linux/gfp.h>
16 #include <linux/kernel.h>
17 #include <linux/mm.h>
18 #include <linux/sched.h>
19 #include <linux/signal.h>
20 #include <linux/slab.h>
21 #include <linux/time_namespace.h>
22 #include <linux/vmalloc.h>
23 #include <vdso/datapage.h>
24 #include <vdso/helpers.h>
25 #include <vdso/vsyscall.h>
26 
27 #include <asm/cacheflush.h>
28 #include <asm/signal32.h>
29 #include <asm/vdso.h>
30 
31 enum vdso_abi {
32 	VDSO_ABI_AA64,
33 	VDSO_ABI_AA32,
34 };
35 
36 struct vdso_abi_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 
47 static struct vdso_abi_info vdso_info[] __ro_after_init = {
48 	[VDSO_ABI_AA64] = {
49 		.name = "vdso",
50 		.vdso_code_start = vdso_start,
51 		.vdso_code_end = vdso_end,
52 	},
53 #ifdef CONFIG_COMPAT_VDSO
54 	[VDSO_ABI_AA32] = {
55 		.name = "vdso32",
56 		.vdso_code_start = vdso32_start,
57 		.vdso_code_end = vdso32_end,
58 	},
59 #endif /* CONFIG_COMPAT_VDSO */
60 };
61 
62 /*
63  * The vDSO data page.
64  */
65 static union vdso_data_store vdso_data_store __page_aligned_data;
66 struct vdso_data *vdso_data = vdso_data_store.data;
67 
68 static int vdso_mremap(const struct vm_special_mapping *sm,
69 		struct vm_area_struct *new_vma)
70 {
71 	current->mm->context.vdso = (void *)new_vma->vm_start;
72 
73 	return 0;
74 }
75 
76 static int __init __vdso_init(enum vdso_abi abi)
77 {
78 	int i;
79 	struct page **vdso_pagelist;
80 	unsigned long pfn;
81 
82 	if (memcmp(vdso_info[abi].vdso_code_start, "\177ELF", 4)) {
83 		pr_err("vDSO is not a valid ELF object!\n");
84 		return -EINVAL;
85 	}
86 
87 	vdso_info[abi].vdso_pages = (
88 			vdso_info[abi].vdso_code_end -
89 			vdso_info[abi].vdso_code_start) >>
90 			PAGE_SHIFT;
91 
92 	vdso_pagelist = kcalloc(vdso_info[abi].vdso_pages,
93 				sizeof(struct page *),
94 				GFP_KERNEL);
95 	if (vdso_pagelist == NULL)
96 		return -ENOMEM;
97 
98 	/* Grab the vDSO code pages. */
99 	pfn = sym_to_pfn(vdso_info[abi].vdso_code_start);
100 
101 	for (i = 0; i < vdso_info[abi].vdso_pages; i++)
102 		vdso_pagelist[i] = pfn_to_page(pfn + i);
103 
104 	vdso_info[abi].cm->pages = vdso_pagelist;
105 
106 	return 0;
107 }
108 
109 #ifdef CONFIG_TIME_NS
110 struct vdso_data *arch_get_vdso_data(void *vvar_page)
111 {
112 	return (struct vdso_data *)(vvar_page);
113 }
114 
115 /*
116  * The vvar mapping contains data for a specific time namespace, so when a task
117  * changes namespace we must unmap its vvar data for the old namespace.
118  * Subsequent faults will map in data for the new namespace.
119  *
120  * For more details see timens_setup_vdso_data().
121  */
122 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
123 {
124 	struct mm_struct *mm = task->mm;
125 	struct vm_area_struct *vma;
126 	VMA_ITERATOR(vmi, mm, 0);
127 
128 	mmap_read_lock(mm);
129 
130 	for_each_vma(vmi, vma) {
131 		if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA64].dm))
132 			zap_vma_pages(vma);
133 #ifdef CONFIG_COMPAT_VDSO
134 		if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA32].dm))
135 			zap_vma_pages(vma);
136 #endif
137 	}
138 
139 	mmap_read_unlock(mm);
140 	return 0;
141 }
142 #endif
143 
144 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
145 			     struct vm_area_struct *vma, struct vm_fault *vmf)
146 {
147 	struct page *timens_page = find_timens_vvar_page(vma);
148 	unsigned long pfn;
149 
150 	switch (vmf->pgoff) {
151 	case VVAR_DATA_PAGE_OFFSET:
152 		if (timens_page)
153 			pfn = page_to_pfn(timens_page);
154 		else
155 			pfn = sym_to_pfn(vdso_data);
156 		break;
157 #ifdef CONFIG_TIME_NS
158 	case VVAR_TIMENS_PAGE_OFFSET:
159 		/*
160 		 * If a task belongs to a time namespace then a namespace
161 		 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
162 		 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
163 		 * offset.
164 		 * See also the comment near timens_setup_vdso_data().
165 		 */
166 		if (!timens_page)
167 			return VM_FAULT_SIGBUS;
168 		pfn = sym_to_pfn(vdso_data);
169 		break;
170 #endif /* CONFIG_TIME_NS */
171 	default:
172 		return VM_FAULT_SIGBUS;
173 	}
174 
175 	return vmf_insert_pfn(vma, vmf->address, pfn);
176 }
177 
178 static int __setup_additional_pages(enum vdso_abi abi,
179 				    struct mm_struct *mm,
180 				    struct linux_binprm *bprm,
181 				    int uses_interp)
182 {
183 	unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
184 	unsigned long gp_flags = 0;
185 	void *ret;
186 
187 	BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
188 
189 	vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT;
190 	/* Be sure to map the data page */
191 	vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
192 
193 	vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
194 	if (IS_ERR_VALUE(vdso_base)) {
195 		ret = ERR_PTR(vdso_base);
196 		goto up_fail;
197 	}
198 
199 	ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE,
200 				       VM_READ|VM_MAYREAD|VM_PFNMAP,
201 				       vdso_info[abi].dm);
202 	if (IS_ERR(ret))
203 		goto up_fail;
204 
205 	if (system_supports_bti_kernel())
206 		gp_flags = VM_ARM64_BTI;
207 
208 	vdso_base += VVAR_NR_PAGES * PAGE_SIZE;
209 	mm->context.vdso = (void *)vdso_base;
210 	ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
211 				       VM_READ|VM_EXEC|gp_flags|
212 				       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
213 				       vdso_info[abi].cm);
214 	if (IS_ERR(ret))
215 		goto up_fail;
216 
217 	return 0;
218 
219 up_fail:
220 	mm->context.vdso = NULL;
221 	return PTR_ERR(ret);
222 }
223 
224 #ifdef CONFIG_COMPAT
225 /*
226  * Create and map the vectors page for AArch32 tasks.
227  */
228 enum aarch32_map {
229 	AA32_MAP_VECTORS, /* kuser helpers */
230 	AA32_MAP_SIGPAGE,
231 	AA32_MAP_VVAR,
232 	AA32_MAP_VDSO,
233 };
234 
235 static struct page *aarch32_vectors_page __ro_after_init;
236 static struct page *aarch32_sig_page __ro_after_init;
237 
238 static int aarch32_sigpage_mremap(const struct vm_special_mapping *sm,
239 				  struct vm_area_struct *new_vma)
240 {
241 	current->mm->context.sigpage = (void *)new_vma->vm_start;
242 
243 	return 0;
244 }
245 
246 static struct vm_special_mapping aarch32_vdso_maps[] = {
247 	[AA32_MAP_VECTORS] = {
248 		.name	= "[vectors]", /* ABI */
249 		.pages	= &aarch32_vectors_page,
250 	},
251 	[AA32_MAP_SIGPAGE] = {
252 		.name	= "[sigpage]", /* ABI */
253 		.pages	= &aarch32_sig_page,
254 		.mremap	= aarch32_sigpage_mremap,
255 	},
256 	[AA32_MAP_VVAR] = {
257 		.name = "[vvar]",
258 		.fault = vvar_fault,
259 	},
260 	[AA32_MAP_VDSO] = {
261 		.name = "[vdso]",
262 		.mremap = vdso_mremap,
263 	},
264 };
265 
266 static int aarch32_alloc_kuser_vdso_page(void)
267 {
268 	extern char __kuser_helper_start[], __kuser_helper_end[];
269 	int kuser_sz = __kuser_helper_end - __kuser_helper_start;
270 	unsigned long vdso_page;
271 
272 	if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
273 		return 0;
274 
275 	vdso_page = get_zeroed_page(GFP_KERNEL);
276 	if (!vdso_page)
277 		return -ENOMEM;
278 
279 	memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
280 	       kuser_sz);
281 	aarch32_vectors_page = virt_to_page((void *)vdso_page);
282 	return 0;
283 }
284 
285 #define COMPAT_SIGPAGE_POISON_WORD	0xe7fddef1
286 static int aarch32_alloc_sigpage(void)
287 {
288 	extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
289 	int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
290 	__le32 poison = cpu_to_le32(COMPAT_SIGPAGE_POISON_WORD);
291 	void *sigpage;
292 
293 	sigpage = (void *)__get_free_page(GFP_KERNEL);
294 	if (!sigpage)
295 		return -ENOMEM;
296 
297 	memset32(sigpage, (__force u32)poison, PAGE_SIZE / sizeof(poison));
298 	memcpy(sigpage, __aarch32_sigret_code_start, sigret_sz);
299 	aarch32_sig_page = virt_to_page(sigpage);
300 	return 0;
301 }
302 
303 static int __init __aarch32_alloc_vdso_pages(void)
304 {
305 
306 	if (!IS_ENABLED(CONFIG_COMPAT_VDSO))
307 		return 0;
308 
309 	vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR];
310 	vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO];
311 
312 	return __vdso_init(VDSO_ABI_AA32);
313 }
314 
315 static int __init aarch32_alloc_vdso_pages(void)
316 {
317 	int ret;
318 
319 	ret = __aarch32_alloc_vdso_pages();
320 	if (ret)
321 		return ret;
322 
323 	ret = aarch32_alloc_sigpage();
324 	if (ret)
325 		return ret;
326 
327 	return aarch32_alloc_kuser_vdso_page();
328 }
329 arch_initcall(aarch32_alloc_vdso_pages);
330 
331 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
332 {
333 	void *ret;
334 
335 	if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
336 		return 0;
337 
338 	/*
339 	 * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
340 	 * not safe to CoW the page containing the CPU exception vectors.
341 	 */
342 	ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
343 				       VM_READ | VM_EXEC |
344 				       VM_MAYREAD | VM_MAYEXEC,
345 				       &aarch32_vdso_maps[AA32_MAP_VECTORS]);
346 
347 	return PTR_ERR_OR_ZERO(ret);
348 }
349 
350 static int aarch32_sigreturn_setup(struct mm_struct *mm)
351 {
352 	unsigned long addr;
353 	void *ret;
354 
355 	addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
356 	if (IS_ERR_VALUE(addr)) {
357 		ret = ERR_PTR(addr);
358 		goto out;
359 	}
360 
361 	/*
362 	 * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
363 	 * set breakpoints.
364 	 */
365 	ret = _install_special_mapping(mm, addr, PAGE_SIZE,
366 				       VM_READ | VM_EXEC | VM_MAYREAD |
367 				       VM_MAYWRITE | VM_MAYEXEC,
368 				       &aarch32_vdso_maps[AA32_MAP_SIGPAGE]);
369 	if (IS_ERR(ret))
370 		goto out;
371 
372 	mm->context.sigpage = (void *)addr;
373 
374 out:
375 	return PTR_ERR_OR_ZERO(ret);
376 }
377 
378 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
379 {
380 	struct mm_struct *mm = current->mm;
381 	int ret;
382 
383 	if (mmap_write_lock_killable(mm))
384 		return -EINTR;
385 
386 	ret = aarch32_kuser_helpers_setup(mm);
387 	if (ret)
388 		goto out;
389 
390 	if (IS_ENABLED(CONFIG_COMPAT_VDSO)) {
391 		ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm,
392 					       uses_interp);
393 		if (ret)
394 			goto out;
395 	}
396 
397 	ret = aarch32_sigreturn_setup(mm);
398 out:
399 	mmap_write_unlock(mm);
400 	return ret;
401 }
402 #endif /* CONFIG_COMPAT */
403 
404 enum aarch64_map {
405 	AA64_MAP_VVAR,
406 	AA64_MAP_VDSO,
407 };
408 
409 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = {
410 	[AA64_MAP_VVAR] = {
411 		.name	= "[vvar]",
412 		.fault = vvar_fault,
413 	},
414 	[AA64_MAP_VDSO] = {
415 		.name	= "[vdso]",
416 		.mremap = vdso_mremap,
417 	},
418 };
419 
420 static int __init vdso_init(void)
421 {
422 	vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR];
423 	vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO];
424 
425 	return __vdso_init(VDSO_ABI_AA64);
426 }
427 arch_initcall(vdso_init);
428 
429 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
430 {
431 	struct mm_struct *mm = current->mm;
432 	int ret;
433 
434 	if (mmap_write_lock_killable(mm))
435 		return -EINTR;
436 
437 	ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp);
438 	mmap_write_unlock(mm);
439 
440 	return ret;
441 }
442