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 struct vdso_abi_info {
  38         const char *name;
  39         const char *vdso_code_start;
  40         const char *vdso_code_end;
  41         unsigned long vdso_pages;
  42         /* Data Mapping */
  43         struct vm_special_mapping *dm;
  44         /* Code Mapping */
  45         struct vm_special_mapping *cm;
  46 };
  47
  48 static struct vdso_abi_info vdso_info[] __ro_after_init = {
  49         [VDSO_ABI_AA64] = {
  50                 .name = "vdso",
  51                 .vdso_code_start = vdso_start,
  52                 .vdso_code_end = vdso_end,
  53         },
  54 #ifdef CONFIG_COMPAT_VDSO
  55         [VDSO_ABI_AA32] = {
  56                 .name = "vdso32",
  57                 .vdso_code_start = vdso32_start,
  58                 .vdso_code_end = vdso32_end,
  59         },
  60 #endif /* CONFIG_COMPAT_VDSO */
  61 };
  62
  63 /*
  64  * The vDSO data page.
  65  */
  66 static union vdso_data_store vdso_data_store __page_aligned_data;
  67 struct vdso_data *vdso_data = vdso_data_store.data;
  68
  69 static int vdso_mremap(const struct vm_special_mapping *sm,
  70                 struct vm_area_struct *new_vma)
  71 {
  72         current->mm->context.vdso = (void *)new_vma->vm_start;
  73
  74         return 0;
  75 }
  76
  77 static int __init __vdso_init(enum vdso_abi abi)
  78 {
  79         int i;
  80         struct page **vdso_pagelist;
  81         unsigned long pfn;
  82
  83         if (memcmp(vdso_info[abi].vdso_code_start, "\177ELF", 4)) {
  84                 pr_err("vDSO is not a valid ELF object!\n");
  85                 return -EINVAL;
  86         }
  87
  88         vdso_info[abi].vdso_pages = (
  89                         vdso_info[abi].vdso_code_end -
  90                         vdso_info[abi].vdso_code_start) >>
  91                         PAGE_SHIFT;
  92
  93         vdso_pagelist = kcalloc(vdso_info[abi].vdso_pages,
  94                                 sizeof(struct page *),
  95                                 GFP_KERNEL);
  96         if (vdso_pagelist == NULL)
  97                 return -ENOMEM;
  98
  99         /* Grab the vDSO code pages. */
 100         pfn = sym_to_pfn(vdso_info[abi].vdso_code_start);
 101
 102         for (i = 0; i < vdso_info[abi].vdso_pages; i++)
 103                 vdso_pagelist[i] = pfn_to_page(pfn + i);
 104
 105         vdso_info[abi].cm->pages = vdso_pagelist;
 106
 107         return 0;
 108 }
 109
 110 #ifdef CONFIG_TIME_NS
 111 struct vdso_data *arch_get_vdso_data(void *vvar_page)
 112 {
 113         return (struct vdso_data *)(vvar_page);
 114 }
 115
 116 /*
 117  * The vvar mapping contains data for a specific time namespace, so when a task
 118  * changes namespace we must unmap its vvar data for the old namespace.
 119  * Subsequent faults will map in data for the new namespace.
 120  *
 121  * For more details see timens_setup_vdso_data().
 122  */
 123 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
 124 {
 125         struct mm_struct *mm = task->mm;
 126         struct vm_area_struct *vma;
 127         VMA_ITERATOR(vmi, mm, 0);
 128
 129         mmap_read_lock(mm);
 130
 131         for_each_vma(vmi, vma) {
 132                 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA64].dm))
 133                         zap_vma_pages(vma);
 134 #ifdef CONFIG_COMPAT_VDSO
 135                 if (vma_is_special_mapping(vma, vdso_info[VDSO_ABI_AA32].dm))
 136                         zap_vma_pages(vma);
 137 #endif
 138         }
 139
 140         mmap_read_unlock(mm);
 141         return 0;
 142 }
 143 #endif
 144
 145 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
 146                              struct vm_area_struct *vma, struct vm_fault *vmf)
 147 {
 148         struct page *timens_page = find_timens_vvar_page(vma);
 149         unsigned long pfn;
 150
 151         switch (vmf->pgoff) {
 152         case VVAR_DATA_PAGE_OFFSET:
 153                 if (timens_page)
 154                         pfn = page_to_pfn(timens_page);
 155                 else
 156                         pfn = sym_to_pfn(vdso_data);
 157                 break;
 158 #ifdef CONFIG_TIME_NS
 159         case VVAR_TIMENS_PAGE_OFFSET:
 160                 /*
 161                  * If a task belongs to a time namespace then a namespace
 162                  * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
 163                  * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
 164                  * offset.
 165                  * See also the comment near timens_setup_vdso_data().
 166                  */
 167                 if (!timens_page)
 168                         return VM_FAULT_SIGBUS;
 169                 pfn = sym_to_pfn(vdso_data);
 170                 break;
 171 #endif /* CONFIG_TIME_NS */
 172         default:
 173                 return VM_FAULT_SIGBUS;
 174         }
 175
 176         return vmf_insert_pfn(vma, vmf->address, pfn);
 177 }
 178
 179 static int __setup_additional_pages(enum vdso_abi abi,
 180                                     struct mm_struct *mm,
 181                                     struct linux_binprm *bprm,
 182                                     int uses_interp)
 183 {
 184         unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
 185         unsigned long gp_flags = 0;
 186         void *ret;
 187
 188         BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
 189
 190         vdso_text_len = vdso_info[abi].vdso_pages << PAGE_SHIFT;
 191         /* Be sure to map the data page */
 192         vdso_mapping_len = vdso_text_len + VVAR_NR_PAGES * PAGE_SIZE;
 193
 194         vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
 195         if (IS_ERR_VALUE(vdso_base)) {
 196                 ret = ERR_PTR(vdso_base);
 197                 goto up_fail;
 198         }
 199
 200         ret = _install_special_mapping(mm, vdso_base, VVAR_NR_PAGES * PAGE_SIZE,
 201                                        VM_READ|VM_MAYREAD|VM_PFNMAP,
 202                                        vdso_info[abi].dm);
 203         if (IS_ERR(ret))
 204                 goto up_fail;
 205
 206         if (system_supports_bti_kernel())
 207                 gp_flags = VM_ARM64_BTI;
 208
 209         vdso_base += VVAR_NR_PAGES * PAGE_SIZE;
 210         mm->context.vdso = (void *)vdso_base;
 211         ret = _install_special_mapping(mm, vdso_base, vdso_text_len,
 212                                        VM_READ|VM_EXEC|gp_flags|
 213                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 214                                        vdso_info[abi].cm);
 215         if (IS_ERR(ret))
 216                 goto up_fail;
 217
 218         return 0;
 219
 220 up_fail:
 221         mm->context.vdso = NULL;
 222         return PTR_ERR(ret);
 223 }
 224
 225 #ifdef CONFIG_COMPAT
 226 /*
 227  * Create and map the vectors page for AArch32 tasks.
 228  */
 229 enum aarch32_map {
 230         AA32_MAP_VECTORS, /* kuser helpers */
 231         AA32_MAP_SIGPAGE,
 232         AA32_MAP_VVAR,
 233         AA32_MAP_VDSO,
 234 };
 235
 236 static struct page *aarch32_vectors_page __ro_after_init;
 237 static struct page *aarch32_sig_page __ro_after_init;
 238
 239 static int aarch32_sigpage_mremap(const struct vm_special_mapping *sm,
 240                                   struct vm_area_struct *new_vma)
 241 {
 242         current->mm->context.sigpage = (void *)new_vma->vm_start;
 243
 244         return 0;
 245 }
 246
 247 static struct vm_special_mapping aarch32_vdso_maps[] = {
 248         [AA32_MAP_VECTORS] = {
 249                 .name   = "[vectors]", /* ABI */
 250                 .pages  = &aarch32_vectors_page,
 251         },
 252         [AA32_MAP_SIGPAGE] = {
 253                 .name   = "[sigpage]", /* ABI */
 254                 .pages  = &aarch32_sig_page,
 255                 .mremap = aarch32_sigpage_mremap,
 256         },
 257         [AA32_MAP_VVAR] = {
 258                 .name = "[vvar]",
 259                 .fault = vvar_fault,
 260         },
 261         [AA32_MAP_VDSO] = {
 262                 .name = "[vdso]",
 263                 .mremap = vdso_mremap,
 264         },
 265 };
 266
 267 static int aarch32_alloc_kuser_vdso_page(void)
 268 {
 269         extern char __kuser_helper_start[], __kuser_helper_end[];
 270         int kuser_sz = __kuser_helper_end - __kuser_helper_start;
 271         unsigned long vdso_page;
 272
 273         if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
 274                 return 0;
 275
 276         vdso_page = get_zeroed_page(GFP_KERNEL);
 277         if (!vdso_page)
 278                 return -ENOMEM;
 279
 280         memcpy((void *)(vdso_page + 0x1000 - kuser_sz), __kuser_helper_start,
 281                kuser_sz);
 282         aarch32_vectors_page = virt_to_page((void *)vdso_page);
 283         return 0;
 284 }
 285
 286 #define COMPAT_SIGPAGE_POISON_WORD      0xe7fddef1
 287 static int aarch32_alloc_sigpage(void)
 288 {
 289         extern char __aarch32_sigret_code_start[], __aarch32_sigret_code_end[];
 290         int sigret_sz = __aarch32_sigret_code_end - __aarch32_sigret_code_start;
 291         __le32 poison = cpu_to_le32(COMPAT_SIGPAGE_POISON_WORD);
 292         void *sigpage;
 293
 294         sigpage = (void *)__get_free_page(GFP_KERNEL);
 295         if (!sigpage)
 296                 return -ENOMEM;
 297
 298         memset32(sigpage, (__force u32)poison, PAGE_SIZE / sizeof(poison));
 299         memcpy(sigpage, __aarch32_sigret_code_start, sigret_sz);
 300         aarch32_sig_page = virt_to_page(sigpage);
 301         return 0;
 302 }
 303
 304 static int __init __aarch32_alloc_vdso_pages(void)
 305 {
 306
 307         if (!IS_ENABLED(CONFIG_COMPAT_VDSO))
 308                 return 0;
 309
 310         vdso_info[VDSO_ABI_AA32].dm = &aarch32_vdso_maps[AA32_MAP_VVAR];
 311         vdso_info[VDSO_ABI_AA32].cm = &aarch32_vdso_maps[AA32_MAP_VDSO];
 312
 313         return __vdso_init(VDSO_ABI_AA32);
 314 }
 315
 316 static int __init aarch32_alloc_vdso_pages(void)
 317 {
 318         int ret;
 319
 320         ret = __aarch32_alloc_vdso_pages();
 321         if (ret)
 322                 return ret;
 323
 324         ret = aarch32_alloc_sigpage();
 325         if (ret)
 326                 return ret;
 327
 328         return aarch32_alloc_kuser_vdso_page();
 329 }
 330 arch_initcall(aarch32_alloc_vdso_pages);
 331
 332 static int aarch32_kuser_helpers_setup(struct mm_struct *mm)
 333 {
 334         void *ret;
 335
 336         if (!IS_ENABLED(CONFIG_KUSER_HELPERS))
 337                 return 0;
 338
 339         /*
 340          * Avoid VM_MAYWRITE for compatibility with arch/arm/, where it's
 341          * not safe to CoW the page containing the CPU exception vectors.
 342          */
 343         ret = _install_special_mapping(mm, AARCH32_VECTORS_BASE, PAGE_SIZE,
 344                                        VM_READ | VM_EXEC |
 345                                        VM_MAYREAD | VM_MAYEXEC,
 346                                        &aarch32_vdso_maps[AA32_MAP_VECTORS]);
 347
 348         return PTR_ERR_OR_ZERO(ret);
 349 }
 350
 351 static int aarch32_sigreturn_setup(struct mm_struct *mm)
 352 {
 353         unsigned long addr;
 354         void *ret;
 355
 356         addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 357         if (IS_ERR_VALUE(addr)) {
 358                 ret = ERR_PTR(addr);
 359                 goto out;
 360         }
 361
 362         /*
 363          * VM_MAYWRITE is required to allow gdb to Copy-on-Write and
 364          * set breakpoints.
 365          */
 366         ret = _install_special_mapping(mm, addr, PAGE_SIZE,
 367                                        VM_READ | VM_EXEC | VM_MAYREAD |
 368                                        VM_MAYWRITE | VM_MAYEXEC,
 369                                        &aarch32_vdso_maps[AA32_MAP_SIGPAGE]);
 370         if (IS_ERR(ret))
 371                 goto out;
 372
 373         mm->context.sigpage = (void *)addr;
 374
 375 out:
 376         return PTR_ERR_OR_ZERO(ret);
 377 }
 378
 379 int aarch32_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 380 {
 381         struct mm_struct *mm = current->mm;
 382         int ret;
 383
 384         if (mmap_write_lock_killable(mm))
 385                 return -EINTR;
 386
 387         ret = aarch32_kuser_helpers_setup(mm);
 388         if (ret)
 389                 goto out;
 390
 391         if (IS_ENABLED(CONFIG_COMPAT_VDSO)) {
 392                 ret = __setup_additional_pages(VDSO_ABI_AA32, mm, bprm,
 393                                                uses_interp);
 394                 if (ret)
 395                         goto out;
 396         }
 397
 398         ret = aarch32_sigreturn_setup(mm);
 399 out:
 400         mmap_write_unlock(mm);
 401         return ret;
 402 }
 403 #endif /* CONFIG_COMPAT */
 404
 405 enum aarch64_map {
 406         AA64_MAP_VVAR,
 407         AA64_MAP_VDSO,
 408 };
 409
 410 static struct vm_special_mapping aarch64_vdso_maps[] __ro_after_init = {
 411         [AA64_MAP_VVAR] = {
 412                 .name   = "[vvar]",
 413                 .fault = vvar_fault,
 414         },
 415         [AA64_MAP_VDSO] = {
 416                 .name   = "[vdso]",
 417                 .mremap = vdso_mremap,
 418         },
 419 };
 420
 421 static int __init vdso_init(void)
 422 {
 423         vdso_info[VDSO_ABI_AA64].dm = &aarch64_vdso_maps[AA64_MAP_VVAR];
 424         vdso_info[VDSO_ABI_AA64].cm = &aarch64_vdso_maps[AA64_MAP_VDSO];
 425
 426         return __vdso_init(VDSO_ABI_AA64);
 427 }
 428 arch_initcall(vdso_init);
 429
 430 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 431 {
 432         struct mm_struct *mm = current->mm;
 433         int ret;
 434
 435         if (mmap_write_lock_killable(mm))
 436                 return -EINTR;
 437
 438         ret = __setup_additional_pages(VDSO_ABI_AA64, mm, bprm, uses_interp);
 439         mmap_write_unlock(mm);
 440
 441         return ret;
 442 }