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/timekeeper_internal.h> 22 #include <linux/vmalloc.h> 23 #include <vdso/datapage.h> 24 #include <vdso/helpers.h> 25 #include <vdso/vsyscall.h> 26 27 #include <asm/cacheflush.h> 28 #include <asm/signal32.h> 29 #include <asm/vdso.h> 30 31 extern char vdso_start[], vdso_end[]; 32 #ifdef CONFIG_COMPAT_VDSO 33 extern char vdso32_start[], vdso32_end[]; 34 #endif /* CONFIG_COMPAT_VDSO */ 35 36 enum vdso_abi { 37 VDSO_ABI_AA64, 38 #ifdef CONFIG_COMPAT_VDSO 39 VDSO_ABI_AA32, 40 #endif /* CONFIG_COMPAT_VDSO */ 41 }; 42 43 struct 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 54 static 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 */ 72 static union { 73 struct vdso_data data[CS_BASES]; 74 u8 page[PAGE_SIZE]; 75 } vdso_data_store __page_aligned_data; 76 struct vdso_data *vdso_data = vdso_data_store.data; 77 78 static int __vdso_remap(enum vdso_abi abi, 79 const struct vm_special_mapping *sm, 80 struct vm_area_struct *new_vma) 81 { 82 unsigned long new_size = new_vma->vm_end - new_vma->vm_start; 83 unsigned long vdso_size = vdso_info[abi].vdso_code_end - 84 vdso_info[abi].vdso_code_start; 85 86 if (vdso_size != new_size) 87 return -EINVAL; 88 89 current->mm->context.vdso = (void *)new_vma->vm_start; 90 91 return 0; 92 } 93 94 static int __vdso_init(enum vdso_abi abi) 95 { 96 int i; 97 struct page **vdso_pagelist; 98 unsigned long pfn; 99 100 if (memcmp(vdso_info[abi].vdso_code_start, "\177ELF", 4)) { 101 pr_err("vDSO is not a valid ELF object!\n"); 102 return -EINVAL; 103 } 104 105 vdso_info[abi].vdso_pages = ( 106 vdso_info[abi].vdso_code_end - 107 vdso_info[abi].vdso_code_start) >> 108 PAGE_SHIFT; 109 110 vdso_pagelist = kcalloc(vdso_info[abi].vdso_pages, 111 sizeof(struct page *), 112 GFP_KERNEL); 113 if (vdso_pagelist == NULL) 114 return -ENOMEM; 115 116 /* Grab the vDSO code pages. */ 117 pfn = sym_to_pfn(vdso_info[abi].vdso_code_start); 118 119 for (i = 0; i < vdso_info[abi].vdso_pages; i++) 120 vdso_pagelist[i] = pfn_to_page(pfn + i); 121 122 vdso_info[abi].cm->pages = vdso_pagelist; 123 124 return 0; 125 } 126 127 #ifdef CONFIG_TIME_NS 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; 139 140 mmap_read_lock(mm); 141 142 for (vma = mm->mmap; vma; vma = vma->vm_next) { 143 unsigned long size = vma->vm_end - vma->vm_start; 144 145 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA64].dm)) 146 zap_page_range(vma, vma->vm_start, size); 147 #ifdef CONFIG_COMPAT_VDSO 148 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA32].dm)) 149 zap_page_range(vma, vma->vm_start, size); 150 #endif 151 } 152 153 mmap_read_unlock(mm); 154 return 0; 155 } 156 #endif 157 158 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm, 159 struct vm_area_struct *vma, struct vm_fault *vmf) 160 { 161 if (vmf->pgoff == 0) 162 return vmf_insert_pfn(vma, vmf->address, 163 sym_to_pfn(vdso_data)); 164 return VM_FAULT_SIGBUS; 165 } 166 167 static int __setup_additional_pages(enum vdso_abi abi, 168 struct mm_struct *mm, 169 struct linux_binprm *bprm, 170 int uses_interp) 171 { 172 unsigned long vdso_base, vdso_text_len, vdso_mapping_len; 173 unsigned long gp_flags = 0; 174 void *ret; 175 176 vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT; 177 /* Be sure to map the data page */ 178 vdso_mapping_len = vdso_text_len + PAGE_SIZE; 179 180 vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0); 181 if (IS_ERR_VALUE(vdso_base)) { 182 ret = ERR_PTR(vdso_base); 183 goto up_fail; 184 } 185 186 ret = _install_special_mapping(mm, vdso_base, PAGE_SIZE, 187 VM_READ|VM_MAYREAD|VM_PFNMAP, 188 vdso_info[abi].dm); 189 if (IS_ERR(ret)) 190 goto up_fail; 191 192 if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) && system_supports_bti()) 193 gp_flags = VM_ARM64_BTI; 194 195 vdso_base += PAGE_SIZE; 196 mm->context.vdso = (void *)vdso_base; 197 ret = _install_special_mapping(mm, vdso_base, vdso_text_len, 198 VM_READ|VM_EXEC|gp_flags| 199 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC, 200 vdso_info[abi].cm); 201 if (IS_ERR(ret)) 202 goto up_fail; 203 204 return 0; 205 206 up_fail: 207 mm->context.vdso = NULL; 208 return PTR_ERR(ret); 209 } 210 211 #ifdef CONFIG_COMPAT 212 /* 213 * Create and map the vectors page for AArch32 tasks. 214 */ 215 #ifdef CONFIG_COMPAT_VDSO 216 static int aarch32_vdso_mremap(const struct vm_special_mapping *sm, 217 struct vm_area_struct *new_vma) 218 { 219 return __vdso_remap(VDSO_ABI_AA32, sm, new_vma); 220 } 221 #endif /* CONFIG_COMPAT_VDSO */ 222 223 enum aarch32_map { 224 AA32_MAP_VECTORS, /* kuser helpers */ 225 #ifdef CONFIG_COMPAT_VDSO 226 AA32_MAP_VVAR, 227 AA32_MAP_VDSO, 228 #endif 229 AA32_MAP_SIGPAGE 230 }; 231 232 static struct page *aarch32_vectors_page __ro_after_init; 233 static struct page *aarch32_sig_page __ro_after_init; 234 235 static struct vm_special_mapping aarch32_vdso_maps[] = { 236 [AA32_MAP_VECTORS] = { 237 .name = "[vectors]", /* ABI */ 238 .pages = &aarch32_vectors_page, 239 }, 240 #ifdef CONFIG_COMPAT_VDSO 241 [AA32_MAP_VVAR] = { 242 .name = "[vvar]", 243 .fault = vvar_fault, 244 }, 245 [AA32_MAP_VDSO] = { 246 .name = "[vdso]", 247 .mremap = aarch32_vdso_mremap, 248 }, 249 #endif /* CONFIG_COMPAT_VDSO */ 250 [AA32_MAP_SIGPAGE] = { 251 .name = "[sigpage]", /* ABI */ 252 .pages = &aarch32_sig_page, 253 }, 254 }; 255 256 static int aarch32_alloc_kuser_vdso_page(void) 257 { 258 extern char __kuser_helper_start[], __kuser_helper_end[]; 259 int kuser_sz = __kuser_helper_end - __kuser_helper_start; 260 unsigned long vdso_page; 261 262 if (!IS_ENABLED(CONFIG_KUSER_HELPERS)) 263 return 0; 264 265 vdso_page = get_zeroed_page(GFP_ATOMIC); 266 if (!vdso_page) 267 return -ENOMEM; 268 269 memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start, 270 kuser_sz); 271 aarch32_vectors_page = virt_to_page(vdso_page); 272 flush_dcache_page(aarch32_vectors_page); 273 return 0; 274 } 275 276 static int aarch32_alloc_sigpage(void) 277 { 278 extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[]; 279 int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start; 280 unsigned long sigpage; 281 282 sigpage = get_zeroed_page(GFP_ATOMIC); 283 if (!sigpage) 284 return -ENOMEM; 285 286 memcpy((void *)sigpage, __aarch32_sigret_code_start, sigret_sz); 287 aarch32_sig_page = virt_to_page(sigpage); 288 flush_dcache_page(aarch32_sig_page); 289 return 0; 290 } 291 292 #ifdef CONFIG_COMPAT_VDSO 293 static int __aarch32_alloc_vdso_pages(void) 294 { 295 vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR]; 296 vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO]; 297 298 return __vdso_init(VDSO_ABI_AA32); 299 } 300 #endif /* CONFIG_COMPAT_VDSO */ 301 302 static int __init aarch32_alloc_vdso_pages(void) 303 { 304 int ret; 305 306 #ifdef CONFIG_COMPAT_VDSO 307 ret = __aarch32_alloc_vdso_pages(); 308 if (ret) 309 return ret; 310 #endif 311 312 ret = aarch32_alloc_sigpage(); 313 if (ret) 314 return ret; 315 316 return aarch32_alloc_kuser_vdso_page(); 317 } 318 arch_initcall(aarch32_alloc_vdso_pages); 319 320 static int aarch32_kuser_helpers_setup(struct mm_struct *mm) 321 { 322 void *ret; 323 324 if (!IS_ENABLED(CONFIG_KUSER_HELPERS)) 325 return 0; 326 327 /* 328 * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's 329 * not safe to CoW the page containing the CPU exception vectors. 330 */ 331 ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE, 332 VM_READ | VM_EXEC | 333 VM_MAYREAD | VM_MAYEXEC, 334 &aarch32_vdso_maps[AA32_MAP_VECTORS]); 335 336 return PTR_ERR_OR_ZERO(ret); 337 } 338 339 static int aarch32_sigreturn_setup(struct mm_struct *mm) 340 { 341 unsigned long addr; 342 void *ret; 343 344 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0); 345 if (IS_ERR_VALUE(addr)) { 346 ret = ERR_PTR(addr); 347 goto out; 348 } 349 350 /* 351 * VM_MAYWRITE is required to allow gdb to Copy-on-Write and 352 * set breakpoints. 353 */ 354 ret = _install_special_mapping(mm, addr, PAGE_SIZE, 355 VM_READ | VM_EXEC | VM_MAYREAD | 356 VM_MAYWRITE | VM_MAYEXEC, 357 &aarch32_vdso_maps[AA32_MAP_SIGPAGE]); 358 if (IS_ERR(ret)) 359 goto out; 360 361 mm->context.sigpage = (void *)addr; 362 363 out: 364 return PTR_ERR_OR_ZERO(ret); 365 } 366 367 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp) 368 { 369 struct mm_struct *mm = current->mm; 370 int ret; 371 372 if (mmap_write_lock_killable(mm)) 373 return -EINTR; 374 375 ret = aarch32_kuser_helpers_setup(mm); 376 if (ret) 377 goto out; 378 379 #ifdef CONFIG_COMPAT_VDSO 380 ret = __setup_additional_pages(VDSO_ABI_AA32, 381 mm, 382 bprm, 383 uses_interp); 384 if (ret) 385 goto out; 386 #endif /* CONFIG_COMPAT_VDSO */ 387 388 ret = aarch32_sigreturn_setup(mm); 389 out: 390 mmap_write_unlock(mm); 391 return ret; 392 } 393 #endif /* CONFIG_COMPAT */ 394 395 static int vdso_mremap(const struct vm_special_mapping *sm, 396 struct vm_area_struct *new_vma) 397 { 398 return __vdso_remap(VDSO_ABI_AA64, sm, new_vma); 399 } 400 401 enum aarch64_map { 402 AA64_MAP_VVAR, 403 AA64_MAP_VDSO, 404 }; 405 406 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = { 407 [AA64_MAP_VVAR] = { 408 .name = "[vvar]", 409 .fault = vvar_fault, 410 }, 411 [AA64_MAP_VDSO] = { 412 .name = "[vdso]", 413 .mremap = vdso_mremap, 414 }, 415 }; 416 417 static int __init vdso_init(void) 418 { 419 vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR]; 420 vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO]; 421 422 return __vdso_init(VDSO_ABI_AA64); 423 } 424 arch_initcall(vdso_init); 425 426 int arch_setup_additional_pages(struct linux_binprm *bprm, 427 int uses_interp) 428 { 429 struct mm_struct *mm = current->mm; 430 int ret; 431 432 if (mmap_write_lock_killable(mm)) 433 return -EINTR; 434 435 ret = __setup_additional_pages(VDSO_ABI_AA64, 436 mm, 437 bprm, 438 uses_interp); 439 440 mmap_write_unlock(mm); 441 442 return ret; 443 } 444