1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp. 4 * <benh@kernel.crashing.org> 5 * Copyright (C) 2012 ARM Limited 6 * Copyright (C) 2015 Regents of the University of California 7 */ 8 9 #include <linux/elf.h> 10 #include <linux/mm.h> 11 #include <linux/slab.h> 12 #include <linux/binfmts.h> 13 #include <linux/err.h> 14 #include <asm/page.h> 15 #include <asm/vdso.h> 16 #include <linux/time_namespace.h> 17 18 #ifdef CONFIG_GENERIC_TIME_VSYSCALL 19 #include <vdso/datapage.h> 20 #else 21 struct vdso_data { 22 }; 23 #endif 24 25 extern char vdso_start[], vdso_end[]; 26 27 enum vvar_pages { 28 VVAR_DATA_PAGE_OFFSET, 29 VVAR_TIMENS_PAGE_OFFSET, 30 VVAR_NR_PAGES, 31 }; 32 33 #define VVAR_SIZE (VVAR_NR_PAGES << PAGE_SHIFT) 34 35 /* 36 * The vDSO data page. 37 */ 38 static union { 39 struct vdso_data data; 40 u8 page[PAGE_SIZE]; 41 } vdso_data_store __page_aligned_data; 42 struct vdso_data *vdso_data = &vdso_data_store.data; 43 44 struct __vdso_info { 45 const char *name; 46 const char *vdso_code_start; 47 const char *vdso_code_end; 48 unsigned long vdso_pages; 49 /* Data Mapping */ 50 struct vm_special_mapping *dm; 51 /* Code Mapping */ 52 struct vm_special_mapping *cm; 53 }; 54 55 static struct __vdso_info vdso_info __ro_after_init = { 56 .name = "vdso", 57 .vdso_code_start = vdso_start, 58 .vdso_code_end = vdso_end, 59 }; 60 61 static int vdso_mremap(const struct vm_special_mapping *sm, 62 struct vm_area_struct *new_vma) 63 { 64 current->mm->context.vdso = (void *)new_vma->vm_start; 65 66 return 0; 67 } 68 69 static int __init __vdso_init(void) 70 { 71 unsigned int i; 72 struct page **vdso_pagelist; 73 unsigned long pfn; 74 75 if (memcmp(vdso_info.vdso_code_start, "\177ELF", 4)) { 76 pr_err("vDSO is not a valid ELF object!\n"); 77 return -EINVAL; 78 } 79 80 vdso_info.vdso_pages = ( 81 vdso_info.vdso_code_end - 82 vdso_info.vdso_code_start) >> 83 PAGE_SHIFT; 84 85 vdso_pagelist = kcalloc(vdso_info.vdso_pages, 86 sizeof(struct page *), 87 GFP_KERNEL); 88 if (vdso_pagelist == NULL) 89 return -ENOMEM; 90 91 /* Grab the vDSO code pages. */ 92 pfn = sym_to_pfn(vdso_info.vdso_code_start); 93 94 for (i = 0; i < vdso_info.vdso_pages; i++) 95 vdso_pagelist[i] = pfn_to_page(pfn + i); 96 97 vdso_info.cm->pages = vdso_pagelist; 98 99 return 0; 100 } 101 102 #ifdef CONFIG_TIME_NS 103 struct vdso_data *arch_get_vdso_data(void *vvar_page) 104 { 105 return (struct vdso_data *)(vvar_page); 106 } 107 108 /* 109 * The vvar mapping contains data for a specific time namespace, so when a task 110 * changes namespace we must unmap its vvar data for the old namespace. 111 * Subsequent faults will map in data for the new namespace. 112 * 113 * For more details see timens_setup_vdso_data(). 114 */ 115 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns) 116 { 117 struct mm_struct *mm = task->mm; 118 struct vm_area_struct *vma; 119 120 mmap_read_lock(mm); 121 122 for (vma = mm->mmap; vma; vma = vma->vm_next) { 123 unsigned long size = vma->vm_end - vma->vm_start; 124 125 if (vma_is_special_mapping(vma, vdso_info.dm)) 126 zap_page_range(vma, vma->vm_start, size); 127 } 128 129 mmap_read_unlock(mm); 130 return 0; 131 } 132 133 static struct page *find_timens_vvar_page(struct vm_area_struct *vma) 134 { 135 if (likely(vma->vm_mm == current->mm)) 136 return current->nsproxy->time_ns->vvar_page; 137 138 /* 139 * VM_PFNMAP | VM_IO protect .fault() handler from being called 140 * through interfaces like /proc/$pid/mem or 141 * process_vm_{readv,writev}() as long as there's no .access() 142 * in special_mapping_vmops. 143 * For more details check_vma_flags() and __access_remote_vm() 144 */ 145 WARN(1, "vvar_page accessed remotely"); 146 147 return NULL; 148 } 149 #else 150 static struct page *find_timens_vvar_page(struct vm_area_struct *vma) 151 { 152 return NULL; 153 } 154 #endif 155 156 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm, 157 struct vm_area_struct *vma, struct vm_fault *vmf) 158 { 159 struct page *timens_page = find_timens_vvar_page(vma); 160 unsigned long pfn; 161 162 switch (vmf->pgoff) { 163 case VVAR_DATA_PAGE_OFFSET: 164 if (timens_page) 165 pfn = page_to_pfn(timens_page); 166 else 167 pfn = sym_to_pfn(vdso_data); 168 break; 169 #ifdef CONFIG_TIME_NS 170 case VVAR_TIMENS_PAGE_OFFSET: 171 /* 172 * If a task belongs to a time namespace then a namespace 173 * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and 174 * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET 175 * offset. 176 * See also the comment near timens_setup_vdso_data(). 177 */ 178 if (!timens_page) 179 return VM_FAULT_SIGBUS; 180 pfn = sym_to_pfn(vdso_data); 181 break; 182 #endif /* CONFIG_TIME_NS */ 183 default: 184 return VM_FAULT_SIGBUS; 185 } 186 187 return vmf_insert_pfn(vma, vmf->address, pfn); 188 } 189 190 enum rv_vdso_map { 191 RV_VDSO_MAP_VVAR, 192 RV_VDSO_MAP_VDSO, 193 }; 194 195 static struct vm_special_mapping rv_vdso_maps[] __ro_after_init = { 196 [RV_VDSO_MAP_VVAR] = { 197 .name = "[vvar]", 198 .fault = vvar_fault, 199 }, 200 [RV_VDSO_MAP_VDSO] = { 201 .name = "[vdso]", 202 .mremap = vdso_mremap, 203 }, 204 }; 205 206 static int __init vdso_init(void) 207 { 208 vdso_info.dm = &rv_vdso_maps[RV_VDSO_MAP_VVAR]; 209 vdso_info.cm = &rv_vdso_maps[RV_VDSO_MAP_VDSO]; 210 211 return __vdso_init(); 212 } 213 arch_initcall(vdso_init); 214 215 static int __setup_additional_pages(struct mm_struct *mm, 216 struct linux_binprm *bprm, 217 int uses_interp) 218 { 219 unsigned long vdso_base, vdso_text_len, vdso_mapping_len; 220 void *ret; 221 222 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES); 223 224 vdso_text_len = vdso_info.vdso_pages << PAGE_SHIFT; 225 /* Be sure to map the data page */ 226 vdso_mapping_len = vdso_text_len + VVAR_SIZE; 227 228 vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0); 229 if (IS_ERR_VALUE(vdso_base)) { 230 ret = ERR_PTR(vdso_base); 231 goto up_fail; 232 } 233 234 ret = _install_special_mapping(mm, vdso_base, VVAR_SIZE, 235 (VM_READ | VM_MAYREAD | VM_PFNMAP), vdso_info.dm); 236 if (IS_ERR(ret)) 237 goto up_fail; 238 239 vdso_base += VVAR_SIZE; 240 mm->context.vdso = (void *)vdso_base; 241 ret = 242 _install_special_mapping(mm, vdso_base, vdso_text_len, 243 (VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC), 244 vdso_info.cm); 245 246 if (IS_ERR(ret)) 247 goto up_fail; 248 249 return 0; 250 251 up_fail: 252 mm->context.vdso = NULL; 253 return PTR_ERR(ret); 254 } 255 256 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp) 257 { 258 struct mm_struct *mm = current->mm; 259 int ret; 260 261 if (mmap_write_lock_killable(mm)) 262 return -EINTR; 263 264 ret = __setup_additional_pages(mm, bprm, uses_interp); 265 mmap_write_unlock(mm); 266 267 return ret; 268 } 269