xref: /openbmc/linux/arch/arm64/kernel/vdso.c (revision 11a163f2)
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/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_remap(enum vdso_abi abi,
82 			const struct vm_special_mapping *sm,
83 			struct vm_area_struct *new_vma)
84 {
85 	unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
86 	unsigned long vdso_size = vdso_info[abi].vdso_code_end -
87 				  vdso_info[abi].vdso_code_start;
88 
89 	if (vdso_size != new_size)
90 		return -EINVAL;
91 
92 	current->mm->context.vdso = (void *)new_vma->vm_start;
93 
94 	return 0;
95 }
96 
97 static int __vdso_init(enum vdso_abi abi)
98 {
99 	int i;
100 	struct page **vdso_pagelist;
101 	unsigned long pfn;
102 
103 	if (memcmp(vdso_info[abi].vdso_code_start, "\177ELF", 4)) {
104 		pr_err("vDSO is not a valid ELF object!\n");
105 		return -EINVAL;
106 	}
107 
108 	vdso_info[abi].vdso_pages = (
109 			vdso_info[abi].vdso_code_end -
110 			vdso_info[abi].vdso_code_start) >>
111 			PAGE_SHIFT;
112 
113 	vdso_pagelist = kcalloc(vdso_info[abi].vdso_pages,
114 				sizeof(struct page *),
115 				GFP_KERNEL);
116 	if (vdso_pagelist == NULL)
117 		return -ENOMEM;
118 
119 	/* Grab the vDSO code pages. */
120 	pfn = sym_to_pfn(vdso_info[abi].vdso_code_start);
121 
122 	for (i = 0; i < vdso_info[abi].vdso_pages; i++)
123 		vdso_pagelist[i] = pfn_to_page(pfn + i);
124 
125 	vdso_info[abi].cm->pages = vdso_pagelist;
126 
127 	return 0;
128 }
129 
130 #ifdef CONFIG_TIME_NS
131 struct vdso_data *arch_get_vdso_data(void *vvar_page)
132 {
133 	return (struct vdso_data *)(vvar_page);
134 }
135 
136 /*
137  * The vvar mapping contains data for a specific time namespace, so when a task
138  * changes namespace we must unmap its vvar data for the old namespace.
139  * Subsequent faults will map in data for the new namespace.
140  *
141  * For more details see timens_setup_vdso_data().
142  */
143 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
144 {
145 	struct mm_struct *mm = task->mm;
146 	struct vm_area_struct *vma;
147 
148 	mmap_read_lock(mm);
149 
150 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
151 		unsigned long size = vma->vm_end - vma->vm_start;
152 
153 		if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA64].dm))
154 			zap_page_range(vma, vma->vm_start, size);
155 #ifdef CONFIG_COMPAT_VDSO
156 		if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA32].dm))
157 			zap_page_range(vma, vma->vm_start, size);
158 #endif
159 	}
160 
161 	mmap_read_unlock(mm);
162 	return 0;
163 }
164 
165 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
166 {
167 	if (likely(vma->vm_mm == current->mm))
168 		return current->nsproxy->time_ns->vvar_page;
169 
170 	/*
171 	 * VM_PFNMAP | VM_IO protect .fault() handler from being called
172 	 * through interfaces like /proc/$pid/mem or
173 	 * process_vm_{readv,writev}() as long as there's no .access()
174 	 * in special_mapping_vmops.
175 	 * For more details check_vma_flags() and __access_remote_vm()
176 	 */
177 	WARN(1, "vvar_page accessed remotely");
178 
179 	return NULL;
180 }
181 #else
182 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
183 {
184 	return NULL;
185 }
186 #endif
187 
188 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
189 			     struct vm_area_struct *vma, struct vm_fault *vmf)
190 {
191 	struct page *timens_page = find_timens_vvar_page(vma);
192 	unsigned long pfn;
193 
194 	switch (vmf->pgoff) {
195 	case VVAR_DATA_PAGE_OFFSET:
196 		if (timens_page)
197 			pfn = page_to_pfn(timens_page);
198 		else
199 			pfn = sym_to_pfn(vdso_data);
200 		break;
201 #ifdef CONFIG_TIME_NS
202 	case VVAR_TIMENS_PAGE_OFFSET:
203 		/*
204 		 * If a task belongs to a time namespace then a namespace
205 		 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
206 		 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
207 		 * offset.
208 		 * See also the comment near timens_setup_vdso_data().
209 		 */
210 		if (!timens_page)
211 			return VM_FAULT_SIGBUS;
212 		pfn = sym_to_pfn(vdso_data);
213 		break;
214 #endif /* CONFIG_TIME_NS */
215 	default:
216 		return VM_FAULT_SIGBUS;
217 	}
218 
219 	return vmf_insert_pfn(vma, vmf->address, pfn);
220 }
221 
222 static int vvar_mremap(const struct vm_special_mapping *sm,
223 		       struct vm_area_struct *new_vma)
224 {
225 	unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
226 
227 	if (new_size != VVAR_NR_PAGES * PAGE_SIZE)
228 		return -EINVAL;
229 
230 	return 0;
231 }
232 
233 static int __setup_additional_pages(enum vdso_abi abi,
234 				    struct mm_struct *mm,
235 				    struct linux_binprm *bprm,
236 				    int uses_interp)
237 {
238 	unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
239 	unsigned long gp_flags = 0;
240 	void *ret;
241 
242 	BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
243 
244 	vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT;
245 	/* Be sure to map the data page */
246 	vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
247 
248 	vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
249 	if (IS_ERR_VALUE(vdso_base)) {
250 		ret = ERR_PTR(vdso_base);
251 		goto up_fail;
252 	}
253 
254 	ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE,
255 				       VM_READ|VM_MAYREAD|VM_PFNMAP,
256 				       vdso_info[abi].dm);
257 	if (IS_ERR(ret))
258 		goto up_fail;
259 
260 	if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) && system_supports_bti())
261 		gp_flags = VM_ARM64_BTI;
262 
263 	vdso_base += VVAR_NR_PAGES * PAGE_SIZE;
264 	mm->context.vdso = (void *)vdso_base;
265 	ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
266 				       VM_READ|VM_EXEC|gp_flags|
267 				       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
268 				       vdso_info[abi].cm);
269 	if (IS_ERR(ret))
270 		goto up_fail;
271 
272 	return 0;
273 
274 up_fail:
275 	mm->context.vdso = NULL;
276 	return PTR_ERR(ret);
277 }
278 
279 #ifdef CONFIG_COMPAT
280 /*
281  * Create and map the vectors page for AArch32 tasks.
282  */
283 static int aarch32_vdso_mremap(const struct vm_special_mapping *sm,
284 		struct vm_area_struct *new_vma)
285 {
286 	return __vdso_remap(VDSO_ABI_AA32, sm, new_vma);
287 }
288 
289 enum aarch32_map {
290 	AA32_MAP_VECTORS, /* kuser helpers */
291 	AA32_MAP_SIGPAGE,
292 	AA32_MAP_VVAR,
293 	AA32_MAP_VDSO,
294 };
295 
296 static struct page *aarch32_vectors_page __ro_after_init;
297 static struct page *aarch32_sig_page __ro_after_init;
298 
299 static struct vm_special_mapping aarch32_vdso_maps[] = {
300 	[AA32_MAP_VECTORS] = {
301 		.name	= "[vectors]", /* ABI */
302 		.pages	= &aarch32_vectors_page,
303 	},
304 	[AA32_MAP_SIGPAGE] = {
305 		.name	= "[sigpage]", /* ABI */
306 		.pages	= &aarch32_sig_page,
307 	},
308 	[AA32_MAP_VVAR] = {
309 		.name = "[vvar]",
310 		.fault = vvar_fault,
311 		.mremap = vvar_mremap,
312 	},
313 	[AA32_MAP_VDSO] = {
314 		.name = "[vdso]",
315 		.mremap = aarch32_vdso_mremap,
316 	},
317 };
318 
319 static int aarch32_alloc_kuser_vdso_page(void)
320 {
321 	extern char __kuser_helper_start[], __kuser_helper_end[];
322 	int kuser_sz = __kuser_helper_end - __kuser_helper_start;
323 	unsigned long vdso_page;
324 
325 	if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
326 		return 0;
327 
328 	vdso_page = get_zeroed_page(GFP_ATOMIC);
329 	if (!vdso_page)
330 		return -ENOMEM;
331 
332 	memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
333 	       kuser_sz);
334 	aarch32_vectors_page = virt_to_page(vdso_page);
335 	flush_dcache_page(aarch32_vectors_page);
336 	return 0;
337 }
338 
339 static int aarch32_alloc_sigpage(void)
340 {
341 	extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
342 	int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
343 	unsigned long sigpage;
344 
345 	sigpage = get_zeroed_page(GFP_ATOMIC);
346 	if (!sigpage)
347 		return -ENOMEM;
348 
349 	memcpy((void *)sigpage, __aarch32_sigret_code_start, sigret_sz);
350 	aarch32_sig_page = virt_to_page(sigpage);
351 	flush_dcache_page(aarch32_sig_page);
352 	return 0;
353 }
354 
355 static int __aarch32_alloc_vdso_pages(void)
356 {
357 
358 	if (!IS_ENABLED(CONFIG_COMPAT_VDSO))
359 		return 0;
360 
361 	vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR];
362 	vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO];
363 
364 	return __vdso_init(VDSO_ABI_AA32);
365 }
366 
367 static int __init aarch32_alloc_vdso_pages(void)
368 {
369 	int ret;
370 
371 	ret = __aarch32_alloc_vdso_pages();
372 	if (ret)
373 		return ret;
374 
375 	ret = aarch32_alloc_sigpage();
376 	if (ret)
377 		return ret;
378 
379 	return aarch32_alloc_kuser_vdso_page();
380 }
381 arch_initcall(aarch32_alloc_vdso_pages);
382 
383 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
384 {
385 	void *ret;
386 
387 	if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
388 		return 0;
389 
390 	/*
391 	 * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
392 	 * not safe to CoW the page containing the CPU exception vectors.
393 	 */
394 	ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
395 				       VM_READ | VM_EXEC |
396 				       VM_MAYREAD | VM_MAYEXEC,
397 				       &aarch32_vdso_maps[AA32_MAP_VECTORS]);
398 
399 	return PTR_ERR_OR_ZERO(ret);
400 }
401 
402 static int aarch32_sigreturn_setup(struct mm_struct *mm)
403 {
404 	unsigned long addr;
405 	void *ret;
406 
407 	addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
408 	if (IS_ERR_VALUE(addr)) {
409 		ret = ERR_PTR(addr);
410 		goto out;
411 	}
412 
413 	/*
414 	 * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
415 	 * set breakpoints.
416 	 */
417 	ret = _install_special_mapping(mm, addr, PAGE_SIZE,
418 				       VM_READ | VM_EXEC | VM_MAYREAD |
419 				       VM_MAYWRITE | VM_MAYEXEC,
420 				       &aarch32_vdso_maps[AA32_MAP_SIGPAGE]);
421 	if (IS_ERR(ret))
422 		goto out;
423 
424 	mm->context.sigpage = (void *)addr;
425 
426 out:
427 	return PTR_ERR_OR_ZERO(ret);
428 }
429 
430 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
431 {
432 	struct mm_struct *mm = current->mm;
433 	int ret;
434 
435 	if (mmap_write_lock_killable(mm))
436 		return -EINTR;
437 
438 	ret = aarch32_kuser_helpers_setup(mm);
439 	if (ret)
440 		goto out;
441 
442 	if (IS_ENABLED(CONFIG_COMPAT_VDSO)) {
443 		ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm,
444 					       uses_interp);
445 		if (ret)
446 			goto out;
447 	}
448 
449 	ret = aarch32_sigreturn_setup(mm);
450 out:
451 	mmap_write_unlock(mm);
452 	return ret;
453 }
454 #endif /* CONFIG_COMPAT */
455 
456 static int vdso_mremap(const struct vm_special_mapping *sm,
457 		struct vm_area_struct *new_vma)
458 {
459 	return __vdso_remap(VDSO_ABI_AA64, sm, new_vma);
460 }
461 
462 enum aarch64_map {
463 	AA64_MAP_VVAR,
464 	AA64_MAP_VDSO,
465 };
466 
467 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = {
468 	[AA64_MAP_VVAR] = {
469 		.name	= "[vvar]",
470 		.fault = vvar_fault,
471 		.mremap = vvar_mremap,
472 	},
473 	[AA64_MAP_VDSO] = {
474 		.name	= "[vdso]",
475 		.mremap = vdso_mremap,
476 	},
477 };
478 
479 static int __init vdso_init(void)
480 {
481 	vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR];
482 	vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO];
483 
484 	return __vdso_init(VDSO_ABI_AA64);
485 }
486 arch_initcall(vdso_init);
487 
488 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
489 {
490 	struct mm_struct *mm = current->mm;
491 	int ret;
492 
493 	if (mmap_write_lock_killable(mm))
494 		return -EINTR;
495 
496 	ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp);
497 	mmap_write_unlock(mm);
498 
499 	return ret;
500 }
501