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