xref: /linux-6.15/arch/arm64/kernel/vdso.c (revision de2b84d2) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VDSO implementations.
4  *
5  * Copyright (C) 2012 ARM Limited
6  *
7  * Author: Will Deacon <[email protected]>
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/timekeeper_internal.h>
23 #include <linux/vmalloc.h>
24 #include <vdso/datapage.h>
25 #include <vdso/helpers.h>
26 #include <vdso/vsyscall.h>
27 
28 #include <asm/cacheflush.h>
29 #include <asm/signal32.h>
30 #include <asm/vdso.h>
31 
32 extern char vdso_start[], vdso_end[];
33 extern char vdso32_start[], vdso32_end[];
34 
35 enum vdso_abi {
36 	VDSO_ABI_AA64,
37 	VDSO_ABI_AA32,
38 };
39 
40 enum vvar_pages {
41 	VVAR_DATA_PAGE_OFFSET,
42 	VVAR_TIMENS_PAGE_OFFSET,
43 	VVAR_NR_PAGES,
44 };
45 
46 struct vdso_abi_info {
47 	const char *name;
48 	const char *vdso_code_start;
49 	const char *vdso_code_end;
50 	unsigned long vdso_pages;
51 	/* Data Mapping */
52 	struct vm_special_mapping *dm;
53 	/* Code Mapping */
54 	struct vm_special_mapping *cm;
55 };
56 
57 static struct vdso_abi_info vdso_info[] __ro_after_init = {
58 	[VDSO_ABI_AA64] = {
59 		.name = "vdso",
60 		.vdso_code_start = vdso_start,
61 		.vdso_code_end = vdso_end,
62 	},
63 #ifdef CONFIG_COMPAT_VDSO
64 	[VDSO_ABI_AA32] = {
65 		.name = "vdso32",
66 		.vdso_code_start = vdso32_start,
67 		.vdso_code_end = vdso32_end,
68 	},
69 #endif /* CONFIG_COMPAT_VDSO */
70 };
71 
72 /*
73  * The vDSO data page.
74  */
75 static union {
76 	struct vdso_data	data[CS_BASES];
77 	u8			page[PAGE_SIZE];
78 } vdso_data_store __page_aligned_data;
79 struct vdso_data *vdso_data = vdso_data_store.data;
80 
81 static int vdso_mremap(const struct vm_special_mapping *sm,
82 		struct vm_area_struct *new_vma)
83 {
84 	current->mm->context.vdso = (void *)new_vma->vm_start;
85 
86 	return 0;
87 }
88 
89 static int __init __vdso_init(enum vdso_abi abi)
90 {
91 	int i;
92 	struct page **vdso_pagelist;
93 	unsigned long pfn;
94 
95 	if (memcmp(vdso_info[abi].vdso_code_start, "\177ELF", 4)) {
96 		pr_err("vDSO is not a valid ELF object!\n");
97 		return -EINVAL;
98 	}
99 
100 	vdso_info[abi].vdso_pages = (
101 			vdso_info[abi].vdso_code_end -
102 			vdso_info[abi].vdso_code_start) >>
103 			PAGE_SHIFT;
104 
105 	vdso_pagelist = kcalloc(vdso_info[abi].vdso_pages,
106 				sizeof(struct page *),
107 				GFP_KERNEL);
108 	if (vdso_pagelist == NULL)
109 		return -ENOMEM;
110 
111 	/* Grab the vDSO code pages. */
112 	pfn = sym_to_pfn(vdso_info[abi].vdso_code_start);
113 
114 	for (i = 0; i < vdso_info[abi].vdso_pages; i++)
115 		vdso_pagelist[i] = pfn_to_page(pfn + i);
116 
117 	vdso_info[abi].cm->pages = vdso_pagelist;
118 
119 	return 0;
120 }
121 
122 #ifdef CONFIG_TIME_NS
123 struct vdso_data *arch_get_vdso_data(void *vvar_page)
124 {
125 	return (struct vdso_data *)(vvar_page);
126 }
127 
128 /*
129  * The vvar mapping contains data for a specific time namespace, so when a task
130  * changes namespace we must unmap its vvar data for the old namespace.
131  * Subsequent faults will map in data for the new namespace.
132  *
133  * For more details see timens_setup_vdso_data().
134  */
135 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
136 {
137 	struct mm_struct *mm = task->mm;
138 	struct vm_area_struct *vma;
139 	VMA_ITERATOR(vmi, mm, 0);
140 
141 	mmap_read_lock(mm);
142 
143 	for_each_vma(vmi, vma) {
144 		unsigned long size = vma->vm_end - vma->vm_start;
145 
146 		if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA64].dm))
147 			zap_page_range(vma, vma->vm_start, size);
148 #ifdef CONFIG_COMPAT_VDSO
149 		if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA32].dm))
150 			zap_page_range(vma, vma->vm_start, size);
151 #endif
152 	}
153 
154 	mmap_read_unlock(mm);
155 	return 0;
156 }
157 
158 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
159 {
160 	if (likely(vma->vm_mm == current->mm))
161 		return current->nsproxy->time_ns->vvar_page;
162 
163 	/*
164 	 * VM_PFNMAP | VM_IO protect .fault() handler from being called
165 	 * through interfaces like /proc/$pid/mem or
166 	 * process_vm_{readv,writev}() as long as there's no .access()
167 	 * in special_mapping_vmops.
168 	 * For more details check_vma_flags() and __access_remote_vm()
169 	 */
170 	WARN(1, "vvar_page accessed remotely");
171 
172 	return NULL;
173 }
174 #else
175 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
176 {
177 	return NULL;
178 }
179 #endif
180 
181 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
182 			     struct vm_area_struct *vma, struct vm_fault *vmf)
183 {
184 	struct page *timens_page = find_timens_vvar_page(vma);
185 	unsigned long pfn;
186 
187 	switch (vmf->pgoff) {
188 	case VVAR_DATA_PAGE_OFFSET:
189 		if (timens_page)
190 			pfn = page_to_pfn(timens_page);
191 		else
192 			pfn = sym_to_pfn(vdso_data);
193 		break;
194 #ifdef CONFIG_TIME_NS
195 	case VVAR_TIMENS_PAGE_OFFSET:
196 		/*
197 		 * If a task belongs to a time namespace then a namespace
198 		 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
199 		 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
200 		 * offset.
201 		 * See also the comment near timens_setup_vdso_data().
202 		 */
203 		if (!timens_page)
204 			return VM_FAULT_SIGBUS;
205 		pfn = sym_to_pfn(vdso_data);
206 		break;
207 #endif /* CONFIG_TIME_NS */
208 	default:
209 		return VM_FAULT_SIGBUS;
210 	}
211 
212 	return vmf_insert_pfn(vma, vmf->address, pfn);
213 }
214 
215 static int __setup_additional_pages(enum vdso_abi abi,
216 				    struct mm_struct *mm,
217 				    struct linux_binprm *bprm,
218 				    int uses_interp)
219 {
220 	unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
221 	unsigned long gp_flags = 0;
222 	void *ret;
223 
224 	BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
225 
226 	vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT;
227 	/* Be sure to map the data page */
228 	vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
229 
230 	vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
231 	if (IS_ERR_VALUE(vdso_base)) {
232 		ret = ERR_PTR(vdso_base);
233 		goto up_fail;
234 	}
235 
236 	ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE,
237 				       VM_READ|VM_MAYREAD|VM_PFNMAP,
238 				       vdso_info[abi].dm);
239 	if (IS_ERR(ret))
240 		goto up_fail;
241 
242 	if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) && system_supports_bti())
243 		gp_flags = VM_ARM64_BTI;
244 
245 	vdso_base += VVAR_NR_PAGES * PAGE_SIZE;
246 	mm->context.vdso = (void *)vdso_base;
247 	ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
248 				       VM_READ|VM_EXEC|gp_flags|
249 				       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
250 				       vdso_info[abi].cm);
251 	if (IS_ERR(ret))
252 		goto up_fail;
253 
254 	return 0;
255 
256 up_fail:
257 	mm->context.vdso = NULL;
258 	return PTR_ERR(ret);
259 }
260 
261 #ifdef CONFIG_COMPAT
262 /*
263  * Create and map the vectors page for AArch32 tasks.
264  */
265 enum aarch32_map {
266 	AA32_MAP_VECTORS, /* kuser helpers */
267 	AA32_MAP_SIGPAGE,
268 	AA32_MAP_VVAR,
269 	AA32_MAP_VDSO,
270 };
271 
272 static struct page *aarch32_vectors_page __ro_after_init;
273 static struct page *aarch32_sig_page __ro_after_init;
274 
275 static int aarch32_sigpage_mremap(const struct vm_special_mapping *sm,
276 				  struct vm_area_struct *new_vma)
277 {
278 	current->mm->context.sigpage = (void *)new_vma->vm_start;
279 
280 	return 0;
281 }
282 
283 static struct vm_special_mapping aarch32_vdso_maps[] = {
284 	[AA32_MAP_VECTORS] = {
285 		.name	= "[vectors]", /* ABI */
286 		.pages	= &aarch32_vectors_page,
287 	},
288 	[AA32_MAP_SIGPAGE] = {
289 		.name	= "[sigpage]", /* ABI */
290 		.pages	= &aarch32_sig_page,
291 		.mremap	= aarch32_sigpage_mremap,
292 	},
293 	[AA32_MAP_VVAR] = {
294 		.name = "[vvar]",
295 		.fault = vvar_fault,
296 	},
297 	[AA32_MAP_VDSO] = {
298 		.name = "[vdso]",
299 		.mremap = vdso_mremap,
300 	},
301 };
302 
303 static int aarch32_alloc_kuser_vdso_page(void)
304 {
305 	extern char __kuser_helper_start[], __kuser_helper_end[];
306 	int kuser_sz = __kuser_helper_end - __kuser_helper_start;
307 	unsigned long vdso_page;
308 
309 	if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
310 		return 0;
311 
312 	vdso_page = get_zeroed_page(GFP_KERNEL);
313 	if (!vdso_page)
314 		return -ENOMEM;
315 
316 	memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
317 	       kuser_sz);
318 	aarch32_vectors_page = virt_to_page(vdso_page);
319 	return 0;
320 }
321 
322 #define COMPAT_SIGPAGE_POISON_WORD	0xe7fddef1
323 static int aarch32_alloc_sigpage(void)
324 {
325 	extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
326 	int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
327 	__le32 poison = cpu_to_le32(COMPAT_SIGPAGE_POISON_WORD);
328 	void *sigpage;
329 
330 	sigpage = (void *)__get_free_page(GFP_KERNEL);
331 	if (!sigpage)
332 		return -ENOMEM;
333 
334 	memset32(sigpage, (__force u32)poison, PAGE_SIZE / sizeof(poison));
335 	memcpy(sigpage, __aarch32_sigret_code_start, sigret_sz);
336 	aarch32_sig_page = virt_to_page(sigpage);
337 	return 0;
338 }
339 
340 static int __init __aarch32_alloc_vdso_pages(void)
341 {
342 
343 	if (!IS_ENABLED(CONFIG_COMPAT_VDSO))
344 		return 0;
345 
346 	vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR];
347 	vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO];
348 
349 	return __vdso_init(VDSO_ABI_AA32);
350 }
351 
352 static int __init aarch32_alloc_vdso_pages(void)
353 {
354 	int ret;
355 
356 	ret = __aarch32_alloc_vdso_pages();
357 	if (ret)
358 		return ret;
359 
360 	ret = aarch32_alloc_sigpage();
361 	if (ret)
362 		return ret;
363 
364 	return aarch32_alloc_kuser_vdso_page();
365 }
366 arch_initcall(aarch32_alloc_vdso_pages);
367 
368 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
369 {
370 	void *ret;
371 
372 	if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
373 		return 0;
374 
375 	/*
376 	 * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
377 	 * not safe to CoW the page containing the CPU exception vectors.
378 	 */
379 	ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
380 				       VM_READ | VM_EXEC |
381 				       VM_MAYREAD | VM_MAYEXEC,
382 				       &aarch32_vdso_maps[AA32_MAP_VECTORS]);
383 
384 	return PTR_ERR_OR_ZERO(ret);
385 }
386 
387 static int aarch32_sigreturn_setup(struct mm_struct *mm)
388 {
389 	unsigned long addr;
390 	void *ret;
391 
392 	addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
393 	if (IS_ERR_VALUE(addr)) {
394 		ret = ERR_PTR(addr);
395 		goto out;
396 	}
397 
398 	/*
399 	 * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
400 	 * set breakpoints.
401 	 */
402 	ret = _install_special_mapping(mm, addr, PAGE_SIZE,
403 				       VM_READ | VM_EXEC | VM_MAYREAD |
404 				       VM_MAYWRITE | VM_MAYEXEC,
405 				       &aarch32_vdso_maps[AA32_MAP_SIGPAGE]);
406 	if (IS_ERR(ret))
407 		goto out;
408 
409 	mm->context.sigpage = (void *)addr;
410 
411 out:
412 	return PTR_ERR_OR_ZERO(ret);
413 }
414 
415 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
416 {
417 	struct mm_struct *mm = current->mm;
418 	int ret;
419 
420 	if (mmap_write_lock_killable(mm))
421 		return -EINTR;
422 
423 	ret = aarch32_kuser_helpers_setup(mm);
424 	if (ret)
425 		goto out;
426 
427 	if (IS_ENABLED(CONFIG_COMPAT_VDSO)) {
428 		ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm,
429 					       uses_interp);
430 		if (ret)
431 			goto out;
432 	}
433 
434 	ret = aarch32_sigreturn_setup(mm);
435 out:
436 	mmap_write_unlock(mm);
437 	return ret;
438 }
439 #endif /* CONFIG_COMPAT */
440 
441 enum aarch64_map {
442 	AA64_MAP_VVAR,
443 	AA64_MAP_VDSO,
444 };
445 
446 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = {
447 	[AA64_MAP_VVAR] = {
448 		.name	= "[vvar]",
449 		.fault = vvar_fault,
450 	},
451 	[AA64_MAP_VDSO] = {
452 		.name	= "[vdso]",
453 		.mremap = vdso_mremap,
454 	},
455 };
456 
457 static int __init vdso_init(void)
458 {
459 	vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR];
460 	vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO];
461 
462 	return __vdso_init(VDSO_ABI_AA64);
463 }
464 arch_initcall(vdso_init);
465 
466 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
467 {
468 	struct mm_struct *mm = current->mm;
469 	int ret;
470 
471 	if (mmap_write_lock_killable(mm))
472 		return -EINTR;
473 
474 	ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp);
475 	mmap_write_unlock(mm);
476 
477 	return ret;
478 }
479