arch/riscv/kernel/vdso.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
   4  *                    <benh@kernel.crashing.org>
   5  * Copyright (C) 2012 ARM Limited
   6  * Copyright (C) 2015 Regents of the University of California
   7  */
   8
   9 #include <linux/elf.h>
  10 #include <linux/mm.h>
  11 #include <linux/slab.h>
  12 #include <linux/binfmts.h>
  13 #include <linux/err.h>
  14 #include <asm/page.h>
  15 #include <asm/vdso.h>
  16 #include <linux/time_namespace.h>
  17
  18 #ifdef CONFIG_GENERIC_TIME_VSYSCALL
  19 #include <vdso/datapage.h>
  20 #else
  21 struct vdso_data {
  22 };
  23 #endif
  24
  25 extern char vdso_start[], vdso_end[];
  26 #ifdef CONFIG_COMPAT
  27 extern char compat_vdso_start[], compat_vdso_end[];
  28 #endif
  29
  30 enum vvar_pages {
  31         VVAR_DATA_PAGE_OFFSET,
  32         VVAR_TIMENS_PAGE_OFFSET,
  33         VVAR_NR_PAGES,
  34 };
  35
  36 enum rv_vdso_map {
  37         RV_VDSO_MAP_VVAR,
  38         RV_VDSO_MAP_VDSO,
  39 };
  40
  41 #define VVAR_SIZE  (VVAR_NR_PAGES << PAGE_SHIFT)
  42
  43 /*
  44  * The vDSO data page.
  45  */
  46 static union {
  47         struct vdso_data        data;
  48         u8                      page[PAGE_SIZE];
  49 } vdso_data_store __page_aligned_data;
  50 struct vdso_data *vdso_data = &vdso_data_store.data;
  51
  52 struct __vdso_info {
  53         const char *name;
  54         const char *vdso_code_start;
  55         const char *vdso_code_end;
  56         unsigned long vdso_pages;
  57         /* Data Mapping */
  58         struct vm_special_mapping *dm;
  59         /* Code Mapping */
  60         struct vm_special_mapping *cm;
  61 };
  62
  63 static int vdso_mremap(const struct vm_special_mapping *sm,
  64                        struct vm_area_struct *new_vma)
  65 {
  66         current->mm->context.vdso = (void *)new_vma->vm_start;
  67
  68         return 0;
  69 }
  70
  71 static void __init __vdso_init(struct __vdso_info *vdso_info)
  72 {
  73         unsigned int i;
  74         struct page **vdso_pagelist;
  75         unsigned long pfn;
  76
  77         if (memcmp(vdso_info->vdso_code_start, "\177ELF", 4))
  78                 panic("vDSO is not a valid ELF object!\n");
  79
  80         vdso_info->vdso_pages = (
  81                 vdso_info->vdso_code_end -
  82                 vdso_info->vdso_code_start) >>
  83                 PAGE_SHIFT;
  84
  85         vdso_pagelist = kcalloc(vdso_info->vdso_pages,
  86                                 sizeof(struct page *),
  87                                 GFP_KERNEL);
  88         if (vdso_pagelist == NULL)
  89                 panic("vDSO kcalloc failed!\n");
  90
  91         /* Grab the vDSO code pages. */
  92         pfn = sym_to_pfn(vdso_info->vdso_code_start);
  93
  94         for (i = 0; i < vdso_info->vdso_pages; i++)
  95                 vdso_pagelist[i] = pfn_to_page(pfn + i);
  96
  97         vdso_info->cm->pages = vdso_pagelist;
  98 }
  99
 100 #ifdef CONFIG_TIME_NS
 101 struct vdso_data *arch_get_vdso_data(void *vvar_page)
 102 {
 103         return (struct vdso_data *)(vvar_page);
 104 }
 105
 106 /*
 107  * The vvar mapping contains data for a specific time namespace, so when a task
 108  * changes namespace we must unmap its vvar data for the old namespace.
 109  * Subsequent faults will map in data for the new namespace.
 110  *
 111  * For more details see timens_setup_vdso_data().
 112  */
 113 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
 114 {
 115         struct mm_struct *mm = task->mm;
 116         struct vm_area_struct *vma;
 117         struct __vdso_info *vdso_info = mm->context.vdso_info;
 118
 119         mmap_read_lock(mm);
 120
 121         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 122                 unsigned long size = vma->vm_end - vma->vm_start;
 123
 124                 if (vma_is_special_mapping(vma, vdso_info->dm))
 125                         zap_page_range(vma, vma->vm_start, size);
 126         }
 127
 128         mmap_read_unlock(mm);
 129         return 0;
 130 }
 131
 132 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
 133 {
 134         if (likely(vma->vm_mm == current->mm))
 135                 return current->nsproxy->time_ns->vvar_page;
 136
 137         /*
 138          * VM_PFNMAP | VM_IO protect .fault() handler from being called
 139          * through interfaces like /proc/$pid/mem or
 140          * process_vm_{readv,writev}() as long as there's no .access()
 141          * in special_mapping_vmops.
 142          * For more details check_vma_flags() and __access_remote_vm()
 143          */
 144         WARN(1, "vvar_page accessed remotely");
 145
 146         return NULL;
 147 }
 148 #else
 149 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
 150 {
 151         return NULL;
 152 }
 153 #endif
 154
 155 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
 156                              struct vm_area_struct *vma, struct vm_fault *vmf)
 157 {
 158         struct page *timens_page = find_timens_vvar_page(vma);
 159         unsigned long pfn;
 160
 161         switch (vmf->pgoff) {
 162         case VVAR_DATA_PAGE_OFFSET:
 163                 if (timens_page)
 164                         pfn = page_to_pfn(timens_page);
 165                 else
 166                         pfn = sym_to_pfn(vdso_data);
 167                 break;
 168 #ifdef CONFIG_TIME_NS
 169         case VVAR_TIMENS_PAGE_OFFSET:
 170                 /*
 171                  * If a task belongs to a time namespace then a namespace
 172                  * specific VVAR is mapped with the VVAR_DATA_PAGE_OFFSET and
 173                  * the real VVAR page is mapped with the VVAR_TIMENS_PAGE_OFFSET
 174                  * offset.
 175                  * See also the comment near timens_setup_vdso_data().
 176                  */
 177                 if (!timens_page)
 178                         return VM_FAULT_SIGBUS;
 179                 pfn = sym_to_pfn(vdso_data);
 180                 break;
 181 #endif /* CONFIG_TIME_NS */
 182         default:
 183                 return VM_FAULT_SIGBUS;
 184         }
 185
 186         return vmf_insert_pfn(vma, vmf->address, pfn);
 187 }
 188
 189 static struct vm_special_mapping rv_vdso_maps[] __ro_after_init = {
 190         [RV_VDSO_MAP_VVAR] = {
 191                 .name   = "[vvar]",
 192                 .fault = vvar_fault,
 193         },
 194         [RV_VDSO_MAP_VDSO] = {
 195                 .name   = "[vdso]",
 196                 .mremap = vdso_mremap,
 197         },
 198 };
 199
 200 static struct __vdso_info vdso_info __ro_after_init = {
 201         .name = "vdso",
 202         .vdso_code_start = vdso_start,
 203         .vdso_code_end = vdso_end,
 204         .dm = &rv_vdso_maps[RV_VDSO_MAP_VVAR],
 205         .cm = &rv_vdso_maps[RV_VDSO_MAP_VDSO],
 206 };
 207
 208 #ifdef CONFIG_COMPAT
 209 static struct vm_special_mapping rv_compat_vdso_maps[] __ro_after_init = {
 210         [RV_VDSO_MAP_VVAR] = {
 211                 .name   = "[vvar]",
 212                 .fault = vvar_fault,
 213         },
 214         [RV_VDSO_MAP_VDSO] = {
 215                 .name   = "[vdso]",
 216                 .mremap = vdso_mremap,
 217         },
 218 };
 219
 220 static struct __vdso_info compat_vdso_info __ro_after_init = {
 221         .name = "compat_vdso",
 222         .vdso_code_start = compat_vdso_start,
 223         .vdso_code_end = compat_vdso_end,
 224         .dm = &rv_compat_vdso_maps[RV_VDSO_MAP_VVAR],
 225         .cm = &rv_compat_vdso_maps[RV_VDSO_MAP_VDSO],
 226 };
 227 #endif
 228
 229 static int __init vdso_init(void)
 230 {
 231         __vdso_init(&vdso_info);
 232 #ifdef CONFIG_COMPAT
 233         __vdso_init(&compat_vdso_info);
 234 #endif
 235
 236         return 0;
 237 }
 238 arch_initcall(vdso_init);
 239
 240 static int __setup_additional_pages(struct mm_struct *mm,
 241                                     struct linux_binprm *bprm,
 242                                     int uses_interp,
 243                                     struct __vdso_info *vdso_info)
 244 {
 245         unsigned long vdso_base, vdso_text_len, vdso_mapping_len;
 246         void *ret;
 247
 248         BUILD_BUG_ON(VVAR_NR_PAGES != __VVAR_PAGES);
 249
 250         vdso_text_len = vdso_info->vdso_pages << PAGE_SHIFT;
 251         /* Be sure to map the data page */
 252         vdso_mapping_len = vdso_text_len + VVAR_SIZE;
 253
 254         vdso_base = get_unmapped_area(NULL, 0, vdso_mapping_len, 0, 0);
 255         if (IS_ERR_VALUE(vdso_base)) {
 256                 ret = ERR_PTR(vdso_base);
 257                 goto up_fail;
 258         }
 259
 260         ret = _install_special_mapping(mm, vdso_base, VVAR_SIZE,
 261                 (VM_READ | VM_MAYREAD | VM_PFNMAP), vdso_info->dm);
 262         if (IS_ERR(ret))
 263                 goto up_fail;
 264
 265         vdso_base += VVAR_SIZE;
 266         mm->context.vdso = (void *)vdso_base;
 267         mm->context.vdso_info = (void *)vdso_info;
 268
 269         ret =
 270            _install_special_mapping(mm, vdso_base, vdso_text_len,
 271                 (VM_READ | VM_EXEC | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC),
 272                 vdso_info->cm);
 273
 274         if (IS_ERR(ret))
 275                 goto up_fail;
 276
 277         return 0;
 278
 279 up_fail:
 280         mm->context.vdso = NULL;
 281         return PTR_ERR(ret);
 282 }
 283
 284 #ifdef CONFIG_COMPAT
 285 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
 286                                        int uses_interp)
 287 {
 288         struct mm_struct *mm = current->mm;
 289         int ret;
 290
 291         if (mmap_write_lock_killable(mm))
 292                 return -EINTR;
 293
 294         ret = __setup_additional_pages(mm, bprm, uses_interp,
 295                                                         &compat_vdso_info);
 296         mmap_write_unlock(mm);
 297
 298         return ret;
 299 }
 300 #endif
 301
 302 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 303 {
 304         struct mm_struct *mm = current->mm;
 305         int ret;
 306
 307         if (mmap_write_lock_killable(mm))
 308                 return -EINTR;
 309
 310         ret = __setup_additional_pages(mm, bprm, uses_interp, &vdso_info);
 311         mmap_write_unlock(mm);
 312
 313         return ret;
 314 }