arch/x86/vdso/vma.c

   1 /*
   2  * Set up the VMAs to tell the VM about the vDSO.
   3  * Copyright 2007 Andi Kleen, SUSE Labs.
   4  * Subject to the GPL, v.2
   5  */
   6 #include <linux/mm.h>
   7 #include <linux/err.h>
   8 #include <linux/sched.h>
   9 #include <linux/slab.h>
  10 #include <linux/init.h>
  11 #include <linux/random.h>
  12 #include <linux/elf.h>
  13 #include <asm/vsyscall.h>
  14 #include <asm/vgtod.h>
  15 #include <asm/proto.h>
  16 #include <asm/vdso.h>
  17 #include <asm/page.h>
  18 #include <asm/hpet.h>
  19
  20 #if defined(CONFIG_X86_64)
  21 unsigned int __read_mostly vdso64_enabled = 1;
  22
  23 extern unsigned short vdso_sync_cpuid;
  24 #endif
  25
  26 void __init init_vdso_image(const struct vdso_image *image)
  27 {
  28         int i;
  29         int npages = (image->size) / PAGE_SIZE;
  30
  31         BUG_ON(image->size % PAGE_SIZE != 0);
  32         for (i = 0; i < npages; i++)
  33                 image->text_mapping.pages[i] =
  34                         virt_to_page(image->data + i*PAGE_SIZE);
  35
  36         apply_alternatives((struct alt_instr *)(image->data + image->alt),
  37                            (struct alt_instr *)(image->data + image->alt +
  38                                                 image->alt_len));
  39 }
  40
  41 #if defined(CONFIG_X86_64)
  42 static int __init init_vdso(void)
  43 {
  44         init_vdso_image(&vdso_image_64);
  45
  46 #ifdef CONFIG_X86_X32_ABI
  47         init_vdso_image(&vdso_image_x32);
  48 #endif
  49
  50         return 0;
  51 }
  52 subsys_initcall(init_vdso);
  53 #endif
  54
  55 struct linux_binprm;
  56
  57 /* Put the vdso above the (randomized) stack with another randomized offset.
  58    This way there is no hole in the middle of address space.
  59    To save memory make sure it is still in the same PTE as the stack top.
  60    This doesn't give that many random bits.
  61
  62    Only used for the 64-bit and x32 vdsos. */
  63 static unsigned long vdso_addr(unsigned long start, unsigned len)
  64 {
  65 #ifdef CONFIG_X86_32
  66         return 0;
  67 #else
  68         unsigned long addr, end;
  69         unsigned offset;
  70         end = (start + PMD_SIZE - 1) & PMD_MASK;
  71         if (end >= TASK_SIZE_MAX)
  72                 end = TASK_SIZE_MAX;
  73         end -= len;
  74         /* This loses some more bits than a modulo, but is cheaper */
  75         offset = get_random_int() & (PTRS_PER_PTE - 1);
  76         addr = start + (offset << PAGE_SHIFT);
  77         if (addr >= end)
  78                 addr = end;
  79
  80         /*
  81          * page-align it here so that get_unmapped_area doesn't
  82          * align it wrongfully again to the next page. addr can come in 4K
  83          * unaligned here as a result of stack start randomization.
  84          */
  85         addr = PAGE_ALIGN(addr);
  86         addr = align_vdso_addr(addr);
  87
  88         return addr;
  89 #endif
  90 }
  91
  92 static int map_vdso(const struct vdso_image *image, bool calculate_addr)
  93 {
  94         struct mm_struct *mm = current->mm;
  95         struct vm_area_struct *vma;
  96         unsigned long addr, text_start;
  97         int ret = 0;
  98         static struct page *no_pages[] = {NULL};
  99         static struct vm_special_mapping vvar_mapping = {
 100                 .name = "[vvar]",
 101                 .pages = no_pages,
 102         };
 103
 104         if (calculate_addr) {
 105                 addr = vdso_addr(current->mm->start_stack,
 106                                  image->size - image->sym_vvar_start);
 107         } else {
 108                 addr = 0;
 109         }
 110
 111         down_write(&mm->mmap_sem);
 112
 113         addr = get_unmapped_area(NULL, addr,
 114                                  image->size - image->sym_vvar_start, 0, 0);
 115         if (IS_ERR_VALUE(addr)) {
 116                 ret = addr;
 117                 goto up_fail;
 118         }
 119
 120         text_start = addr - image->sym_vvar_start;
 121         current->mm->context.vdso = (void __user *)text_start;
 122
 123         /*
 124          * MAYWRITE to allow gdb to COW and set breakpoints
 125          */
 126         vma = _install_special_mapping(mm,
 127                                        text_start,
 128                                        image->size,
 129                                        VM_READ|VM_EXEC|
 130                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 131                                        &image->text_mapping);
 132
 133         if (IS_ERR(vma)) {
 134                 ret = PTR_ERR(vma);
 135                 goto up_fail;
 136         }
 137
 138         vma = _install_special_mapping(mm,
 139                                        addr,
 140                                        -image->sym_vvar_start,
 141                                        VM_READ,
 142                                        &vvar_mapping);
 143
 144         if (IS_ERR(vma)) {
 145                 ret = PTR_ERR(vma);
 146                 goto up_fail;
 147         }
 148
 149         if (image->sym_vvar_page)
 150                 ret = remap_pfn_range(vma,
 151                                       text_start + image->sym_vvar_page,
 152                                       __pa_symbol(&__vvar_page) >> PAGE_SHIFT,
 153                                       PAGE_SIZE,
 154                                       PAGE_READONLY);
 155
 156         if (ret)
 157                 goto up_fail;
 158
 159 #ifdef CONFIG_HPET_TIMER
 160         if (hpet_address && image->sym_hpet_page) {
 161                 ret = io_remap_pfn_range(vma,
 162                         text_start + image->sym_hpet_page,
 163                         hpet_address >> PAGE_SHIFT,
 164                         PAGE_SIZE,
 165                         pgprot_noncached(PAGE_READONLY));
 166
 167                 if (ret)
 168                         goto up_fail;
 169         }
 170 #endif
 171
 172 up_fail:
 173         if (ret)
 174                 current->mm->context.vdso = NULL;
 175
 176         up_write(&mm->mmap_sem);
 177         return ret;
 178 }
 179
 180 #if defined(CONFIG_X86_32) || defined(CONFIG_COMPAT)
 181 static int load_vdso32(void)
 182 {
 183         int ret;
 184
 185         if (vdso32_enabled != 1)  /* Other values all mean "disabled" */
 186                 return 0;
 187
 188         ret = map_vdso(selected_vdso32, false);
 189         if (ret)
 190                 return ret;
 191
 192         if (selected_vdso32->sym_VDSO32_SYSENTER_RETURN)
 193                 current_thread_info()->sysenter_return =
 194                         current->mm->context.vdso +
 195                         selected_vdso32->sym_VDSO32_SYSENTER_RETURN;
 196
 197         return 0;
 198 }
 199 #endif
 200
 201 #ifdef CONFIG_X86_64
 202 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 203 {
 204         if (!vdso64_enabled)
 205                 return 0;
 206
 207         return map_vdso(&vdso_image_64, true);
 208 }
 209
 210 #ifdef CONFIG_COMPAT
 211 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
 212                                        int uses_interp)
 213 {
 214 #ifdef CONFIG_X86_X32_ABI
 215         if (test_thread_flag(TIF_X32)) {
 216                 if (!vdso64_enabled)
 217                         return 0;
 218
 219                 return map_vdso(&vdso_image_x32, true);
 220         }
 221 #endif
 222
 223         return load_vdso32();
 224 }
 225 #endif
 226 #else
 227 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 228 {
 229         return load_vdso32();
 230 }
 231 #endif
 232
 233 #ifdef CONFIG_X86_64
 234 static __init int vdso_setup(char *s)
 235 {
 236         vdso64_enabled = simple_strtoul(s, NULL, 0);
 237         return 0;
 238 }
 239 __setup("vdso=", vdso_setup);
 240 #endif