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 enum vdso_abi { 33 VDSO_ABI_AA64, 34 VDSO_ABI_AA32, 35 }; 36 37 enum vvar_pages { 38 VVAR_DATA_PAGE_OFFSET, 39 VVAR_TIMENS_PAGE_OFFSET, 40 VVAR_NR_PAGES, 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_mremap(const struct vm_special_mapping *sm, 79 struct vm_area_struct *new_vma) 80 { 81 current->mm->context.vdso = (void *)new_vma->vm_start; 82 83 return 0; 84 } 85 86 static int __init __vdso_init(enum vdso_abi abi) 87 { 88 int i; 89 struct page **vdso_pagelist; 90 unsigned long pfn; 91 92 if (memcmp(vdso_info[abi].vdso_code_start, "\177ELF", 4)) { 93 pr_err("vDSO is not a valid ELF object!\n"); 94 return -EINVAL; 95 } 96 97 vdso_info[abi].vdso_pages = ( 98 vdso_info[abi].vdso_code_end - 99 vdso_info[abi].vdso_code_start) >> 100 PAGE_SHIFT; 101 102 vdso_pagelist = kcalloc(vdso_info[abi].vdso_pages, 103 sizeof(struct page *), 104 GFP_KERNEL); 105 if (vdso_pagelist == NULL) 106 return -ENOMEM; 107 108 /* Grab the vDSO code pages. */ 109 pfn = sym_to_pfn(vdso_info[abi].vdso_code_start); 110 111 for (i = 0; i < vdso_info[abi].vdso_pages; i++) 112 vdso_pagelist[i] = pfn_to_page(pfn + i); 113 114 vdso_info[abi].cm->pages = vdso_pagelist; 115 116 return 0; 117 } 118 119 #ifdef CONFIG_TIME_NS 120 struct vdso_data *arch_get_vdso_data(void *vvar_page) 121 { 122 return (struct vdso_data *)(vvar_page); 123 } 124 125 /* 126 * The vvar mapping contains data for a specific time namespace, so when a task 127 * changes namespace we must unmap its vvar data for the old namespace. 128 * Subsequent faults will map in data for the new namespace. 129 * 130 * For more details see timens_setup_vdso_data(). 131 */ 132 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns) 133 { 134 struct mm_struct *mm = task->mm; 135 struct vm_area_struct *vma; 136 VMA_ITERATOR(vmi, mm, 0); 137 138 mmap_read_lock(mm); 139 140 for_each_vma(vmi, vma) { 141 unsigned long size = vma->vm_end - vma->vm_start; 142 143 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA64].dm)) 144 zap_page_range(vma, vma->vm_start, size); 145 #ifdef CONFIG_COMPAT_VDSO 146 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA32].dm)) 147 zap_page_range(vma, vma->vm_start, size); 148 #endif 149 } 150 151 mmap_read_unlock(mm); 152 return 0; 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 static int __setup_additional_pages(enum vdso_abi abi, 191 struct mm_struct *mm, 192 struct linux_binprm *bprm, 193 int uses_interp) 194 { 195 unsigned long vdso_base, vdso_text_len, vdso_mapping_len; 196 unsigned long gp_flags = 0; 197 void *ret; 198 199 BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES); 200 201 vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT; 202 /* Be sure to map the data page */ 203 vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE; 204 205 vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0); 206 if (IS_ERR_VALUE(vdso_base)) { 207 ret = ERR_PTR(vdso_base); 208 goto up_fail; 209 } 210 211 ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE, 212 VM_READ|VM_MAYREAD|VM_PFNMAP, 213 vdso_info[abi].dm); 214 if (IS_ERR(ret)) 215 goto up_fail; 216 217 if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) && system_supports_bti()) 218 gp_flags = VM_ARM64_BTI; 219 220 vdso_base += VVAR_NR_PAGES * PAGE_SIZE; 221 mm->context.vdso = (void *)vdso_base; 222 ret = _install_special_mapping(mm, vdso_base, vdso_text_len, 223 VM_READ|VM_EXEC|gp_flags| 224 VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC, 225 vdso_info[abi].cm); 226 if (IS_ERR(ret)) 227 goto up_fail; 228 229 return 0; 230 231 up_fail: 232 mm->context.vdso = NULL; 233 return PTR_ERR(ret); 234 } 235 236 #ifdef CONFIG_COMPAT 237 /* 238 * Create and map the vectors page for AArch32 tasks. 239 */ 240 enum aarch32_map { 241 AA32_MAP_VECTORS, /* kuser helpers */ 242 AA32_MAP_SIGPAGE, 243 AA32_MAP_VVAR, 244 AA32_MAP_VDSO, 245 }; 246 247 static struct page *aarch32_vectors_page __ro_after_init; 248 static struct page *aarch32_sig_page __ro_after_init; 249 250 static int aarch32_sigpage_mremap(const struct vm_special_mapping *sm, 251 struct vm_area_struct *new_vma) 252 { 253 current->mm->context.sigpage = (void *)new_vma->vm_start; 254 255 return 0; 256 } 257 258 static struct vm_special_mapping aarch32_vdso_maps[] = { 259 [AA32_MAP_VECTORS] = { 260 .name = "[vectors]", /* ABI */ 261 .pages = &aarch32_vectors_page, 262 }, 263 [AA32_MAP_SIGPAGE] = { 264 .name = "[sigpage]", /* ABI */ 265 .pages = &aarch32_sig_page, 266 .mremap = aarch32_sigpage_mremap, 267 }, 268 [AA32_MAP_VVAR] = { 269 .name = "[vvar]", 270 .fault = vvar_fault, 271 }, 272 [AA32_MAP_VDSO] = { 273 .name = "[vdso]", 274 .mremap = vdso_mremap, 275 }, 276 }; 277 278 static int aarch32_alloc_kuser_vdso_page(void) 279 { 280 extern char __kuser_helper_start[], __kuser_helper_end[]; 281 int kuser_sz = __kuser_helper_end - __kuser_helper_start; 282 unsigned long vdso_page; 283 284 if (!IS_ENABLED(CONFIG_KUSER_HELPERS)) 285 return 0; 286 287 vdso_page = get_zeroed_page(GFP_KERNEL); 288 if (!vdso_page) 289 return -ENOMEM; 290 291 memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start, 292 kuser_sz); 293 aarch32_vectors_page = virt_to_page(vdso_page); 294 return 0; 295 } 296 297 #define COMPAT_SIGPAGE_POISON_WORD 0xe7fddef1 298 static int aarch32_alloc_sigpage(void) 299 { 300 extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[]; 301 int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start; 302 __le32 poison = cpu_to_le32(COMPAT_SIGPAGE_POISON_WORD); 303 void *sigpage; 304 305 sigpage = (void *)__get_free_page(GFP_KERNEL); 306 if (!sigpage) 307 return -ENOMEM; 308 309 memset32(sigpage, (__force u32)poison, PAGE_SIZE / sizeof(poison)); 310 memcpy(sigpage, __aarch32_sigret_code_start, sigret_sz); 311 aarch32_sig_page = virt_to_page(sigpage); 312 return 0; 313 } 314 315 static int __init __aarch32_alloc_vdso_pages(void) 316 { 317 318 if (!IS_ENABLED(CONFIG_COMPAT_VDSO)) 319 return 0; 320 321 vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR]; 322 vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO]; 323 324 return __vdso_init(VDSO_ABI_AA32); 325 } 326 327 static int __init aarch32_alloc_vdso_pages(void) 328 { 329 int ret; 330 331 ret = __aarch32_alloc_vdso_pages(); 332 if (ret) 333 return ret; 334 335 ret = aarch32_alloc_sigpage(); 336 if (ret) 337 return ret; 338 339 return aarch32_alloc_kuser_vdso_page(); 340 } 341 arch_initcall(aarch32_alloc_vdso_pages); 342 343 static int aarch32_kuser_helpers_setup(struct mm_struct *mm) 344 { 345 void *ret; 346 347 if (!IS_ENABLED(CONFIG_KUSER_HELPERS)) 348 return 0; 349 350 /* 351 * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's 352 * not safe to CoW the page containing the CPU exception vectors. 353 */ 354 ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE, 355 VM_READ | VM_EXEC | 356 VM_MAYREAD | VM_MAYEXEC, 357 &aarch32_vdso_maps[AA32_MAP_VECTORS]); 358 359 return PTR_ERR_OR_ZERO(ret); 360 } 361 362 static int aarch32_sigreturn_setup(struct mm_struct *mm) 363 { 364 unsigned long addr; 365 void *ret; 366 367 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0); 368 if (IS_ERR_VALUE(addr)) { 369 ret = ERR_PTR(addr); 370 goto out; 371 } 372 373 /* 374 * VM_MAYWRITE is required to allow gdb to Copy-on-Write and 375 * set breakpoints. 376 */ 377 ret = _install_special_mapping(mm, addr, PAGE_SIZE, 378 VM_READ | VM_EXEC | VM_MAYREAD | 379 VM_MAYWRITE | VM_MAYEXEC, 380 &aarch32_vdso_maps[AA32_MAP_SIGPAGE]); 381 if (IS_ERR(ret)) 382 goto out; 383 384 mm->context.sigpage = (void *)addr; 385 386 out: 387 return PTR_ERR_OR_ZERO(ret); 388 } 389 390 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp) 391 { 392 struct mm_struct *mm = current->mm; 393 int ret; 394 395 if (mmap_write_lock_killable(mm)) 396 return -EINTR; 397 398 ret = aarch32_kuser_helpers_setup(mm); 399 if (ret) 400 goto out; 401 402 if (IS_ENABLED(CONFIG_COMPAT_VDSO)) { 403 ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm, 404 uses_interp); 405 if (ret) 406 goto out; 407 } 408 409 ret = aarch32_sigreturn_setup(mm); 410 out: 411 mmap_write_unlock(mm); 412 return ret; 413 } 414 #endif /* CONFIG_COMPAT */ 415 416 enum aarch64_map { 417 AA64_MAP_VVAR, 418 AA64_MAP_VDSO, 419 }; 420 421 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = { 422 [AA64_MAP_VVAR] = { 423 .name = "[vvar]", 424 .fault = vvar_fault, 425 }, 426 [AA64_MAP_VDSO] = { 427 .name = "[vdso]", 428 .mremap = vdso_mremap, 429 }, 430 }; 431 432 static int __init vdso_init(void) 433 { 434 vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR]; 435 vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO]; 436 437 return __vdso_init(VDSO_ABI_AA64); 438 } 439 arch_initcall(vdso_init); 440 441 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp) 442 { 443 struct mm_struct *mm = current->mm; 444 int ret; 445 446 if (mmap_write_lock_killable(mm)) 447 return -EINTR; 448 449 ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp); 450 mmap_write_unlock(mm); 451 452 return ret; 453 } 454