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