1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Author: Huacai Chen <chenhuacai@loongson.cn>
4  * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
5  */
6 
7 #include <linux/binfmts.h>
8 #include <linux/elf.h>
9 #include <linux/err.h>
10 #include <linux/init.h>
11 #include <linux/ioport.h>
12 #include <linux/kernel.h>
13 #include <linux/mm.h>
14 #include <linux/random.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/time_namespace.h>
18 #include <linux/timekeeper_internal.h>
19 
20 #include <asm/page.h>
21 #include <asm/vdso.h>
22 #include <vdso/helpers.h>
23 #include <vdso/vsyscall.h>
24 #include <generated/vdso-offsets.h>
25 
26 extern char vdso_start[], vdso_end[];
27 
28 /* Kernel-provided data used by the VDSO. */
29 static union {
30 	u8 page[PAGE_SIZE];
31 	struct vdso_data data[CS_BASES];
32 } generic_vdso_data __page_aligned_data;
33 
34 static union {
35 	u8 page[LOONGARCH_VDSO_DATA_SIZE];
36 	struct loongarch_vdso_data vdata;
37 } loongarch_vdso_data __page_aligned_data;
38 
39 static struct page *vdso_pages[] = { NULL };
40 struct vdso_data *vdso_data = generic_vdso_data.data;
41 struct vdso_pcpu_data *vdso_pdata = loongarch_vdso_data.vdata.pdata;
42 
vdso_mremap(const struct vm_special_mapping * sm,struct vm_area_struct * new_vma)43 static int vdso_mremap(const struct vm_special_mapping *sm, struct vm_area_struct *new_vma)
44 {
45 	current->mm->context.vdso = (void *)(new_vma->vm_start);
46 
47 	return 0;
48 }
49 
vvar_fault(const struct vm_special_mapping * sm,struct vm_area_struct * vma,struct vm_fault * vmf)50 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
51 			     struct vm_area_struct *vma, struct vm_fault *vmf)
52 {
53 	unsigned long pfn;
54 	struct page *timens_page = find_timens_vvar_page(vma);
55 
56 	switch (vmf->pgoff) {
57 	case VVAR_GENERIC_PAGE_OFFSET:
58 		if (!timens_page)
59 			pfn = sym_to_pfn(vdso_data);
60 		else
61 			pfn = page_to_pfn(timens_page);
62 		break;
63 #ifdef CONFIG_TIME_NS
64 	case VVAR_TIMENS_PAGE_OFFSET:
65 		/*
66 		 * If a task belongs to a time namespace then a namespace specific
67 		 * VVAR is mapped with the VVAR_GENERIC_PAGE_OFFSET and the real
68 		 * VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET offset.
69 		 * See also the comment near timens_setup_vdso_data().
70 		 */
71 		if (!timens_page)
72 			return VM_FAULT_SIGBUS;
73 		else
74 			pfn = sym_to_pfn(vdso_data);
75 		break;
76 #endif /* CONFIG_TIME_NS */
77 	case VVAR_LOONGARCH_PAGES_START ... VVAR_LOONGARCH_PAGES_END:
78 		pfn = sym_to_pfn(&loongarch_vdso_data) + vmf->pgoff - VVAR_LOONGARCH_PAGES_START;
79 		break;
80 	default:
81 		return VM_FAULT_SIGBUS;
82 	}
83 
84 	return vmf_insert_pfn(vma, vmf->address, pfn);
85 }
86 
87 struct loongarch_vdso_info vdso_info = {
88 	.vdso = vdso_start,
89 	.size = PAGE_SIZE,
90 	.code_mapping = {
91 		.name = "[vdso]",
92 		.pages = vdso_pages,
93 		.mremap = vdso_mremap,
94 	},
95 	.data_mapping = {
96 		.name = "[vvar]",
97 		.fault = vvar_fault,
98 	},
99 	.offset_sigreturn = vdso_offset_sigreturn,
100 };
101 
init_vdso(void)102 static int __init init_vdso(void)
103 {
104 	unsigned long i, cpu, pfn;
105 
106 	BUG_ON(!PAGE_ALIGNED(vdso_info.vdso));
107 	BUG_ON(!PAGE_ALIGNED(vdso_info.size));
108 
109 	for_each_possible_cpu(cpu)
110 		vdso_pdata[cpu].node = cpu_to_node(cpu);
111 
112 	pfn = __phys_to_pfn(__pa_symbol(vdso_info.vdso));
113 	for (i = 0; i < vdso_info.size / PAGE_SIZE; i++)
114 		vdso_info.code_mapping.pages[i] = pfn_to_page(pfn + i);
115 
116 	return 0;
117 }
118 subsys_initcall(init_vdso);
119 
120 #ifdef CONFIG_TIME_NS
arch_get_vdso_data(void * vvar_page)121 struct vdso_data *arch_get_vdso_data(void *vvar_page)
122 {
123 	return (struct vdso_data *)(vvar_page);
124 }
125 
126 /*
127  * The vvar mapping contains data for a specific time namespace, so when a
128  * task changes namespace we must unmap its vvar data for the old namespace.
129  * Subsequent faults will map in data for the new namespace.
130  *
131  * For more details see timens_setup_vdso_data().
132  */
vdso_join_timens(struct task_struct * task,struct time_namespace * ns)133 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
134 {
135 	struct mm_struct *mm = task->mm;
136 	struct vm_area_struct *vma;
137 
138 	VMA_ITERATOR(vmi, mm, 0);
139 
140 	mmap_read_lock(mm);
141 	for_each_vma(vmi, vma) {
142 		if (vma_is_special_mapping(vma, &vdso_info.data_mapping))
143 			zap_vma_pages(vma);
144 	}
145 	mmap_read_unlock(mm);
146 
147 	return 0;
148 }
149 #endif
150 
vdso_base(void)151 static unsigned long vdso_base(void)
152 {
153 	unsigned long base = STACK_TOP;
154 
155 	if (current->flags & PF_RANDOMIZE) {
156 		base += get_random_u32_below(VDSO_RANDOMIZE_SIZE);
157 		base = PAGE_ALIGN(base);
158 	}
159 
160 	return base;
161 }
162 
arch_setup_additional_pages(struct linux_binprm * bprm,int uses_interp)163 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
164 {
165 	int ret;
166 	unsigned long size, data_addr, vdso_addr;
167 	struct mm_struct *mm = current->mm;
168 	struct vm_area_struct *vma;
169 	struct loongarch_vdso_info *info = current->thread.vdso;
170 
171 	if (mmap_write_lock_killable(mm))
172 		return -EINTR;
173 
174 	/*
175 	 * Determine total area size. This includes the VDSO data itself
176 	 * and the data pages.
177 	 */
178 	size = VVAR_SIZE + info->size;
179 
180 	data_addr = get_unmapped_area(NULL, vdso_base(), size, 0, 0);
181 	if (IS_ERR_VALUE(data_addr)) {
182 		ret = data_addr;
183 		goto out;
184 	}
185 
186 	vma = _install_special_mapping(mm, data_addr, VVAR_SIZE,
187 				       VM_READ | VM_MAYREAD | VM_PFNMAP,
188 				       &info->data_mapping);
189 	if (IS_ERR(vma)) {
190 		ret = PTR_ERR(vma);
191 		goto out;
192 	}
193 
194 	vdso_addr = data_addr + VVAR_SIZE;
195 	vma = _install_special_mapping(mm, vdso_addr, info->size,
196 				       VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC,
197 				       &info->code_mapping);
198 	if (IS_ERR(vma)) {
199 		ret = PTR_ERR(vma);
200 		goto out;
201 	}
202 
203 	mm->context.vdso = (void *)vdso_addr;
204 	ret = 0;
205 
206 out:
207 	mmap_write_unlock(mm);
208 	return ret;
209 }
210