arch/arm64/kernel/vdso.c

   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
 155 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
 156 {
 157         if (likely(vma->vm_mm == current->mm))
 158                 return current->nsproxy->time_ns->vvar_page;
 159
 160         /*
 161          * VM_PFNMAP | VM_IO protect .fault() handler from being called
 162          * through interfaces like /proc/$pid/mem or
 163          * process_vm_{readv,writev}() as long as there's no .access()
 164          * in special_mapping_vmops.
 165          * For more details check_vma_flags() and __access_remote_vm()
 166          */
 167         WARN(1, "vvar_page accessed remotely");
 168
 169         return NULL;
 170 }
 171 #else
 172 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
 173 {
 174         return NULL;
 175 }
 176 #endif
 177
 178 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
 179                              struct vm_area_struct *vma, struct vm_fault *vmf)
 180 {
 181         struct page *timens_page = find_timens_vvar_page(vma);
 182         unsigned long pfn;
 183
 184         switch (vmf->pgoff) {
 185         case VVAR_DATA_PAGE_OFFSET:
 186                 if (timens_page)
 187                         pfn = page_to_pfn(timens_page);
 188                 else
 189                         pfn = sym_to_pfn(vdso_data);
 190                 break;
 191 #ifdef CONFIG_TIME_NS
 192         case VVAR_TIMENS_PAGE_OFFSET:
 193                 /*
 194                  * If a task belongs to a time namespace then a namespace
 195                  * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
 196                  * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
 197                  * offset.
 198                  * See also the comment near timens_setup_vdso_data().
 199                  */
 200                 if (!timens_page)
 201                         return VM_FAULT_SIGBUS;
 202                 pfn = sym_to_pfn(vdso_data);
 203                 break;
 204 #endif /* CONFIG_TIME_NS */
 205         default:
 206                 return VM_FAULT_SIGBUS;
 207         }
 208
 209         return vmf_insert_pfn(vma, vmf->address, pfn);
 210 }
 211
 212 static int __setup_additional_pages(enum vdso_abi abi,
 213                                     struct mm_struct *mm,
 214                                     struct linux_binprm *bprm,
 215                                     int uses_interp)
 216 {
 217         unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
 218         unsigned long gp_flags = 0;
 219         void *ret;
 220
 221         BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
 222
 223         vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT;
 224         /* Be sure to map the data page */
 225         vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
 226
 227         vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
 228         if (IS_ERR_VALUE(vdso_base)) {
 229                 ret = ERR_PTR(vdso_base);
 230                 goto up_fail;
 231         }
 232
 233         ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE,
 234                                        VM_READ|VM_MAYREAD|VM_PFNMAP,
 235                                        vdso_info[abi].dm);
 236         if (IS_ERR(ret))
 237                 goto up_fail;
 238
 239         if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) && system_supports_bti())
 240                 gp_flags = VM_ARM64_BTI;
 241
 242         vdso_base += VVAR_NR_PAGES * PAGE_SIZE;
 243         mm->context.vdso = (void *)vdso_base;
 244         ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
 245                                        VM_READ|VM_EXEC|gp_flags|
 246                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 247                                        vdso_info[abi].cm);
 248         if (IS_ERR(ret))
 249                 goto up_fail;
 250
 251         return 0;
 252
 253 up_fail:
 254         mm->context.vdso = NULL;
 255         return PTR_ERR(ret);
 256 }
 257
 258 #ifdef CONFIG_COMPAT
 259 /*
 260  * Create and map the vectors page for AArch32 tasks.
 261  */
 262 enum aarch32_map {
 263         AA32_MAP_VECTORS, /* kuser helpers */
 264         AA32_MAP_SIGPAGE,
 265         AA32_MAP_VVAR,
 266         AA32_MAP_VDSO,
 267 };
 268
 269 static struct page *aarch32_vectors_page __ro_after_init;
 270 static struct page *aarch32_sig_page __ro_after_init;
 271
 272 static int aarch32_sigpage_mremap(const struct vm_special_mapping *sm,
 273                                   struct vm_area_struct *new_vma)
 274 {
 275         current->mm->context.sigpage = (void *)new_vma->vm_start;
 276
 277         return 0;
 278 }
 279
 280 static struct vm_special_mapping aarch32_vdso_maps[] = {
 281         [AA32_MAP_VECTORS] = {
 282                 .name   = "[vectors]", /* ABI */
 283                 .pages  = &aarch32_vectors_page,
 284         },
 285         [AA32_MAP_SIGPAGE] = {
 286                 .name   = "[sigpage]", /* ABI */
 287                 .pages  = &aarch32_sig_page,
 288                 .mremap = aarch32_sigpage_mremap,
 289         },
 290         [AA32_MAP_VVAR] = {
 291                 .name = "[vvar]",
 292                 .fault = vvar_fault,
 293         },
 294         [AA32_MAP_VDSO] = {
 295                 .name = "[vdso]",
 296                 .mremap = vdso_mremap,
 297         },
 298 };
 299
 300 static int aarch32_alloc_kuser_vdso_page(void)
 301 {
 302         extern char __kuser_helper_start[], __kuser_helper_end[];
 303         int kuser_sz = __kuser_helper_end - __kuser_helper_start;
 304         unsigned long vdso_page;
 305
 306         if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
 307                 return 0;
 308
 309         vdso_page = get_zeroed_page(GFP_KERNEL);
 310         if (!vdso_page)
 311                 return -ENOMEM;
 312
 313         memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
 314                kuser_sz);
 315         aarch32_vectors_page = virt_to_page((void *)vdso_page);
 316         return 0;
 317 }
 318
 319 #define COMPAT_SIGPAGE_POISON_WORD      0xe7fddef1
 320 static int aarch32_alloc_sigpage(void)
 321 {
 322         extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
 323         int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
 324         __le32 poison = cpu_to_le32(COMPAT_SIGPAGE_POISON_WORD);
 325         void *sigpage;
 326
 327         sigpage = (void *)__get_free_page(GFP_KERNEL);
 328         if (!sigpage)
 329                 return -ENOMEM;
 330
 331         memset32(sigpage, (__force u32)poison, PAGE_SIZE / sizeof(poison));
 332         memcpy(sigpage, __aarch32_sigret_code_start, sigret_sz);
 333         aarch32_sig_page = virt_to_page(sigpage);
 334         return 0;
 335 }
 336
 337 static int __init __aarch32_alloc_vdso_pages(void)
 338 {
 339
 340         if (!IS_ENABLED(CONFIG_COMPAT_VDSO))
 341                 return 0;
 342
 343         vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR];
 344         vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO];
 345
 346         return __vdso_init(VDSO_ABI_AA32);
 347 }
 348
 349 static int __init aarch32_alloc_vdso_pages(void)
 350 {
 351         int ret;
 352
 353         ret = __aarch32_alloc_vdso_pages();
 354         if (ret)
 355                 return ret;
 356
 357         ret = aarch32_alloc_sigpage();
 358         if (ret)
 359                 return ret;
 360
 361         return aarch32_alloc_kuser_vdso_page();
 362 }
 363 arch_initcall(aarch32_alloc_vdso_pages);
 364
 365 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
 366 {
 367         void *ret;
 368
 369         if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
 370                 return 0;
 371
 372         /*
 373          * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
 374          * not safe to CoW the page containing the CPU exception vectors.
 375          */
 376         ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
 377                                        VM_READ | VM_EXEC |
 378                                        VM_MAYREAD | VM_MAYEXEC,
 379                                        &aarch32_vdso_maps[AA32_MAP_VECTORS]);
 380
 381         return PTR_ERR_OR_ZERO(ret);
 382 }
 383
 384 static int aarch32_sigreturn_setup(struct mm_struct *mm)
 385 {
 386         unsigned long addr;
 387         void *ret;
 388
 389         addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 390         if (IS_ERR_VALUE(addr)) {
 391                 ret = ERR_PTR(addr);
 392                 goto out;
 393         }
 394
 395         /*
 396          * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
 397          * set breakpoints.
 398          */
 399         ret = _install_special_mapping(mm, addr, PAGE_SIZE,
 400                                        VM_READ | VM_EXEC | VM_MAYREAD |
 401                                        VM_MAYWRITE | VM_MAYEXEC,
 402                                        &aarch32_vdso_maps[AA32_MAP_SIGPAGE]);
 403         if (IS_ERR(ret))
 404                 goto out;
 405
 406         mm->context.sigpage = (void *)addr;
 407
 408 out:
 409         return PTR_ERR_OR_ZERO(ret);
 410 }
 411
 412 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 413 {
 414         struct mm_struct *mm = current->mm;
 415         int ret;
 416
 417         if (mmap_write_lock_killable(mm))
 418                 return -EINTR;
 419
 420         ret = aarch32_kuser_helpers_setup(mm);
 421         if (ret)
 422                 goto out;
 423
 424         if (IS_ENABLED(CONFIG_COMPAT_VDSO)) {
 425                 ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm,
 426                                                uses_interp);
 427                 if (ret)
 428                         goto out;
 429         }
 430
 431         ret = aarch32_sigreturn_setup(mm);
 432 out:
 433         mmap_write_unlock(mm);
 434         return ret;
 435 }
 436 #endif /* CONFIG_COMPAT */
 437
 438 enum aarch64_map {
 439         AA64_MAP_VVAR,
 440         AA64_MAP_VDSO,
 441 };
 442
 443 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = {
 444         [AA64_MAP_VVAR] = {
 445                 .name   = "[vvar]",
 446                 .fault = vvar_fault,
 447         },
 448         [AA64_MAP_VDSO] = {
 449                 .name   = "[vdso]",
 450                 .mremap = vdso_mremap,
 451         },
 452 };
 453
 454 static int __init vdso_init(void)
 455 {
 456         vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR];
 457         vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO];
 458
 459         return __vdso_init(VDSO_ABI_AA64);
 460 }
 461 arch_initcall(vdso_init);
 462
 463 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 464 {
 465         struct mm_struct *mm = current->mm;
 466         int ret;
 467
 468         if (mmap_write_lock_killable(mm))
 469                 return -EINTR;
 470
 471         ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp);
 472         mmap_write_unlock(mm);
 473
 474         return ret;
 475 }