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
 137         mmap_read_lock(mm);
 138
 139         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 140                 unsigned long size = vma->vm_end - vma->vm_start;
 141
 142                 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA64].dm))
 143                         zap_page_range(vma, vma->vm_start, size);
 144 #ifdef CONFIG_COMPAT_VDSO
 145                 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA32].dm))
 146                         zap_page_range(vma, vma->vm_start, size);
 147 #endif
 148         }
 149
 150         mmap_read_unlock(mm);
 151         return 0;
 152 }
 153
 154 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
 155 {
 156         if (likely(vma->vm_mm == current->mm))
 157                 return current->nsproxy->time_ns->vvar_page;
 158
 159         /*
 160          * VM_PFNMAP | VM_IO protect .fault() handler from being called
 161          * through interfaces like /proc/$pid/mem or
 162          * process_vm_{readv,writev}() as long as there's no .access()
 163          * in special_mapping_vmops.
 164          * For more details check_vma_flags() and __access_remote_vm()
 165          */
 166         WARN(1, "vvar_page accessed remotely");
 167
 168         return NULL;
 169 }
 170 #else
 171 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
 172 {
 173         return NULL;
 174 }
 175 #endif
 176
 177 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
 178                              struct vm_area_struct *vma, struct vm_fault *vmf)
 179 {
 180         struct page *timens_page = find_timens_vvar_page(vma);
 181         unsigned long pfn;
 182
 183         switch (vmf->pgoff) {
 184         case VVAR_DATA_PAGE_OFFSET:
 185                 if (timens_page)
 186                         pfn = page_to_pfn(timens_page);
 187                 else
 188                         pfn = sym_to_pfn(vdso_data);
 189                 break;
 190 #ifdef CONFIG_TIME_NS
 191         case VVAR_TIMENS_PAGE_OFFSET:
 192                 /*
 193                  * If a task belongs to a time namespace then a namespace
 194                  * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
 195                  * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
 196                  * offset.
 197                  * See also the comment near timens_setup_vdso_data().
 198                  */
 199                 if (!timens_page)
 200                         return VM_FAULT_SIGBUS;
 201                 pfn = sym_to_pfn(vdso_data);
 202                 break;
 203 #endif /* CONFIG_TIME_NS */
 204         default:
 205                 return VM_FAULT_SIGBUS;
 206         }
 207
 208         return vmf_insert_pfn(vma, vmf->address, pfn);
 209 }
 210
 211 static int __setup_additional_pages(enum vdso_abi abi,
 212                                     struct mm_struct *mm,
 213                                     struct linux_binprm *bprm,
 214                                     int uses_interp)
 215 {
 216         unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
 217         unsigned long gp_flags = 0;
 218         void *ret;
 219
 220         BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
 221
 222         vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT;
 223         /* Be sure to map the data page */
 224         vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
 225
 226         vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
 227         if (IS_ERR_VALUE(vdso_base)) {
 228                 ret = ERR_PTR(vdso_base);
 229                 goto up_fail;
 230         }
 231
 232         ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE,
 233                                        VM_READ|VM_MAYREAD|VM_PFNMAP,
 234                                        vdso_info[abi].dm);
 235         if (IS_ERR(ret))
 236                 goto up_fail;
 237
 238         if (IS_ENABLED(CONFIG_ARM64_BTI_KERNEL) && system_supports_bti())
 239                 gp_flags = VM_ARM64_BTI;
 240
 241         vdso_base += VVAR_NR_PAGES * PAGE_SIZE;
 242         mm->context.vdso = (void *)vdso_base;
 243         ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
 244                                        VM_READ|VM_EXEC|gp_flags|
 245                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 246                                        vdso_info[abi].cm);
 247         if (IS_ERR(ret))
 248                 goto up_fail;
 249
 250         return 0;
 251
 252 up_fail:
 253         mm->context.vdso = NULL;
 254         return PTR_ERR(ret);
 255 }
 256
 257 #ifdef CONFIG_COMPAT
 258 /*
 259  * Create and map the vectors page for AArch32 tasks.
 260  */
 261 enum aarch32_map {
 262         AA32_MAP_VECTORS, /* kuser helpers */
 263         AA32_MAP_SIGPAGE,
 264         AA32_MAP_VVAR,
 265         AA32_MAP_VDSO,
 266 };
 267
 268 static struct page *aarch32_vectors_page __ro_after_init;
 269 static struct page *aarch32_sig_page __ro_after_init;
 270
 271 static int aarch32_sigpage_mremap(const struct vm_special_mapping *sm,
 272                                   struct vm_area_struct *new_vma)
 273 {
 274         current->mm->context.sigpage = (void *)new_vma->vm_start;
 275
 276         return 0;
 277 }
 278
 279 static struct vm_special_mapping aarch32_vdso_maps[] = {
 280         [AA32_MAP_VECTORS] = {
 281                 .name   = "[vectors]", /* ABI */
 282                 .pages  = &aarch32_vectors_page,
 283         },
 284         [AA32_MAP_SIGPAGE] = {
 285                 .name   = "[sigpage]", /* ABI */
 286                 .pages  = &aarch32_sig_page,
 287                 .mremap = aarch32_sigpage_mremap,
 288         },
 289         [AA32_MAP_VVAR] = {
 290                 .name = "[vvar]",
 291                 .fault = vvar_fault,
 292         },
 293         [AA32_MAP_VDSO] = {
 294                 .name = "[vdso]",
 295                 .mremap = vdso_mremap,
 296         },
 297 };
 298
 299 static int aarch32_alloc_kuser_vdso_page(void)
 300 {
 301         extern char __kuser_helper_start[], __kuser_helper_end[];
 302         int kuser_sz = __kuser_helper_end - __kuser_helper_start;
 303         unsigned long vdso_page;
 304
 305         if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
 306                 return 0;
 307
 308         vdso_page = get_zeroed_page(GFP_KERNEL);
 309         if (!vdso_page)
 310                 return -ENOMEM;
 311
 312         memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
 313                kuser_sz);
 314         aarch32_vectors_page = virt_to_page(vdso_page);
 315         return 0;
 316 }
 317
 318 #define COMPAT_SIGPAGE_POISON_WORD      0xe7fddef1
 319 static int aarch32_alloc_sigpage(void)
 320 {
 321         extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
 322         int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
 323         __le32 poison = cpu_to_le32(COMPAT_SIGPAGE_POISON_WORD);
 324         void *sigpage;
 325
 326         sigpage = (void *)__get_free_page(GFP_KERNEL);
 327         if (!sigpage)
 328                 return -ENOMEM;
 329
 330         memset32(sigpage, (__force u32)poison, PAGE_SIZE / sizeof(poison));
 331         memcpy(sigpage, __aarch32_sigret_code_start, sigret_sz);
 332         aarch32_sig_page = virt_to_page(sigpage);
 333         return 0;
 334 }
 335
 336 static int __init __aarch32_alloc_vdso_pages(void)
 337 {
 338
 339         if (!IS_ENABLED(CONFIG_COMPAT_VDSO))
 340                 return 0;
 341
 342         vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR];
 343         vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO];
 344
 345         return __vdso_init(VDSO_ABI_AA32);
 346 }
 347
 348 static int __init aarch32_alloc_vdso_pages(void)
 349 {
 350         int ret;
 351
 352         ret = __aarch32_alloc_vdso_pages();
 353         if (ret)
 354                 return ret;
 355
 356         ret = aarch32_alloc_sigpage();
 357         if (ret)
 358                 return ret;
 359
 360         return aarch32_alloc_kuser_vdso_page();
 361 }
 362 arch_initcall(aarch32_alloc_vdso_pages);
 363
 364 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
 365 {
 366         void *ret;
 367
 368         if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
 369                 return 0;
 370
 371         /*
 372          * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
 373          * not safe to CoW the page containing the CPU exception vectors.
 374          */
 375         ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
 376                                        VM_READ | VM_EXEC |
 377                                        VM_MAYREAD | VM_MAYEXEC,
 378                                        &aarch32_vdso_maps[AA32_MAP_VECTORS]);
 379
 380         return PTR_ERR_OR_ZERO(ret);
 381 }
 382
 383 static int aarch32_sigreturn_setup(struct mm_struct *mm)
 384 {
 385         unsigned long addr;
 386         void *ret;
 387
 388         addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 389         if (IS_ERR_VALUE(addr)) {
 390                 ret = ERR_PTR(addr);
 391                 goto out;
 392         }
 393
 394         /*
 395          * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
 396          * set breakpoints.
 397          */
 398         ret = _install_special_mapping(mm, addr, PAGE_SIZE,
 399                                        VM_READ | VM_EXEC | VM_MAYREAD |
 400                                        VM_MAYWRITE | VM_MAYEXEC,
 401                                        &aarch32_vdso_maps[AA32_MAP_SIGPAGE]);
 402         if (IS_ERR(ret))
 403                 goto out;
 404
 405         mm->context.sigpage = (void *)addr;
 406
 407 out:
 408         return PTR_ERR_OR_ZERO(ret);
 409 }
 410
 411 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 412 {
 413         struct mm_struct *mm = current->mm;
 414         int ret;
 415
 416         if (mmap_write_lock_killable(mm))
 417                 return -EINTR;
 418
 419         ret = aarch32_kuser_helpers_setup(mm);
 420         if (ret)
 421                 goto out;
 422
 423         if (IS_ENABLED(CONFIG_COMPAT_VDSO)) {
 424                 ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm,
 425                                                uses_interp);
 426                 if (ret)
 427                         goto out;
 428         }
 429
 430         ret = aarch32_sigreturn_setup(mm);
 431 out:
 432         mmap_write_unlock(mm);
 433         return ret;
 434 }
 435 #endif /* CONFIG_COMPAT */
 436
 437 enum aarch64_map {
 438         AA64_MAP_VVAR,
 439         AA64_MAP_VDSO,
 440 };
 441
 442 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = {
 443         [AA64_MAP_VVAR] = {
 444                 .name   = "[vvar]",
 445                 .fault = vvar_fault,
 446         },
 447         [AA64_MAP_VDSO] = {
 448                 .name   = "[vdso]",
 449                 .mremap = vdso_mremap,
 450         },
 451 };
 452
 453 static int __init vdso_init(void)
 454 {
 455         vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR];
 456         vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO];
 457
 458         return __vdso_init(VDSO_ABI_AA64);
 459 }
 460 arch_initcall(vdso_init);
 461
 462 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 463 {
 464         struct mm_struct *mm = current->mm;
 465         int ret;
 466
 467         if (mmap_write_lock_killable(mm))
 468                 return -EINTR;
 469
 470         ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp);
 471         mmap_write_unlock(mm);
 472
 473         return ret;
 474 }