arch/x86/entry/vdso/vma.c

   1 /*
   2  * Copyright 2007 Andi Kleen, SUSE Labs.
   3  * Subject to the GPL, v.2
   4  *
   5  * This contains most of the x86 vDSO kernel-side code.
   6  */
   7 #include <linux/mm.h>
   8 #include <linux/err.h>
   9 #include <linux/sched.h>
  10 #include <linux/slab.h>
  11 #include <linux/init.h>
  12 #include <linux/random.h>
  13 #include <linux/elf.h>
  14 #include <linux/cpu.h>
  15 #include <asm/pvclock.h>
  16 #include <asm/vgtod.h>
  17 #include <asm/proto.h>
  18 #include <asm/vdso.h>
  19 #include <asm/vvar.h>
  20 #include <asm/page.h>
  21 #include <asm/hpet.h>
  22 #include <asm/desc.h>
  23 #include <asm/cpufeature.h>
  24
  25 #if defined(CONFIG_X86_64)
  26 unsigned int __read_mostly vdso64_enabled = 1;
  27 #endif
  28
  29 void __init init_vdso_image(const struct vdso_image *image)
  30 {
  31         BUG_ON(image->size % PAGE_SIZE != 0);
  32
  33         apply_alternatives((struct alt_instr *)(image->data + image->alt),
  34                            (struct alt_instr *)(image->data + image->alt +
  35                                                 image->alt_len));
  36 }
  37
  38 struct linux_binprm;
  39
  40 /*
  41  * Put the vdso above the (randomized) stack with another randomized
  42  * offset.  This way there is no hole in the middle of address space.
  43  * To save memory make sure it is still in the same PTE as the stack
  44  * top.  This doesn't give that many random bits.
  45  *
  46  * Note that this algorithm is imperfect: the distribution of the vdso
  47  * start address within a PMD is biased toward the end.
  48  *
  49  * Only used for the 64-bit and x32 vdsos.
  50  */
  51 static unsigned long vdso_addr(unsigned long start, unsigned len)
  52 {
  53 #ifdef CONFIG_X86_32
  54         return 0;
  55 #else
  56         unsigned long addr, end;
  57         unsigned offset;
  58
  59         /*
  60          * Round up the start address.  It can start out unaligned as a result
  61          * of stack start randomization.
  62          */
  63         start = PAGE_ALIGN(start);
  64
  65         /* Round the lowest possible end address up to a PMD boundary. */
  66         end = (start + len + PMD_SIZE - 1) & PMD_MASK;
  67         if (end >= TASK_SIZE_MAX)
  68                 end = TASK_SIZE_MAX;
  69         end -= len;
  70
  71         if (end > start) {
  72                 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
  73                 addr = start + (offset << PAGE_SHIFT);
  74         } else {
  75                 addr = start;
  76         }
  77
  78         /*
  79          * Forcibly align the final address in case we have a hardware
  80          * issue that requires alignment for performance reasons.
  81          */
  82         addr = align_vdso_addr(addr);
  83
  84         return addr;
  85 #endif
  86 }
  87
  88 static int vdso_fault(const struct vm_special_mapping *sm,
  89                       struct vm_area_struct *vma, struct vm_fault *vmf)
  90 {
  91         const struct vdso_image *image = vma->vm_mm->context.vdso_image;
  92
  93         if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
  94                 return VM_FAULT_SIGBUS;
  95
  96         vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
  97         get_page(vmf->page);
  98         return 0;
  99 }
 100
 101 static const struct vm_special_mapping text_mapping = {
 102         .name = "[vdso]",
 103         .fault = vdso_fault,
 104 };
 105
 106 static int vvar_fault(const struct vm_special_mapping *sm,
 107                       struct vm_area_struct *vma, struct vm_fault *vmf)
 108 {
 109         const struct vdso_image *image = vma->vm_mm->context.vdso_image;
 110         long sym_offset;
 111         int ret = -EFAULT;
 112
 113         if (!image)
 114                 return VM_FAULT_SIGBUS;
 115
 116         sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
 117                 image->sym_vvar_start;
 118
 119         /*
 120          * Sanity check: a symbol offset of zero means that the page
 121          * does not exist for this vdso image, not that the page is at
 122          * offset zero relative to the text mapping.  This should be
 123          * impossible here, because sym_offset should only be zero for
 124          * the page past the end of the vvar mapping.
 125          */
 126         if (sym_offset == 0)
 127                 return VM_FAULT_SIGBUS;
 128
 129         if (sym_offset == image->sym_vvar_page) {
 130                 ret = vm_insert_pfn(vma, (unsigned long)vmf->virtual_address,
 131                                     __pa_symbol(&__vvar_page) >> PAGE_SHIFT);
 132         } else if (sym_offset == image->sym_hpet_page) {
 133 #ifdef CONFIG_HPET_TIMER
 134                 if (hpet_address && vclock_was_used(VCLOCK_HPET)) {
 135                         ret = vm_insert_pfn_prot(
 136                                 vma,
 137                                 (unsigned long)vmf->virtual_address,
 138                                 hpet_address >> PAGE_SHIFT,
 139                                 pgprot_noncached(PAGE_READONLY));
 140                 }
 141 #endif
 142         } else if (sym_offset == image->sym_pvclock_page) {
 143                 struct pvclock_vsyscall_time_info *pvti =
 144                         pvclock_pvti_cpu0_va();
 145                 if (pvti && vclock_was_used(VCLOCK_PVCLOCK)) {
 146                         ret = vm_insert_pfn(
 147                                 vma,
 148                                 (unsigned long)vmf->virtual_address,
 149                                 __pa(pvti) >> PAGE_SHIFT);
 150                 }
 151         }
 152
 153         if (ret == 0 || ret == -EBUSY)
 154                 return VM_FAULT_NOPAGE;
 155
 156         return VM_FAULT_SIGBUS;
 157 }
 158
 159 static int map_vdso(const struct vdso_image *image, bool calculate_addr)
 160 {
 161         struct mm_struct *mm = current->mm;
 162         struct vm_area_struct *vma;
 163         unsigned long addr, text_start;
 164         int ret = 0;
 165         static const struct vm_special_mapping vvar_mapping = {
 166                 .name = "[vvar]",
 167                 .fault = vvar_fault,
 168         };
 169
 170         if (calculate_addr) {
 171                 addr = vdso_addr(current->mm->start_stack,
 172                                  image->size - image->sym_vvar_start);
 173         } else {
 174                 addr = 0;
 175         }
 176
 177         down_write(&mm->mmap_sem);
 178
 179         addr = get_unmapped_area(NULL, addr,
 180                                  image->size - image->sym_vvar_start, 0, 0);
 181         if (IS_ERR_VALUE(addr)) {
 182                 ret = addr;
 183                 goto up_fail;
 184         }
 185
 186         text_start = addr - image->sym_vvar_start;
 187         current->mm->context.vdso = (void __user *)text_start;
 188         current->mm->context.vdso_image = image;
 189
 190         /*
 191          * MAYWRITE to allow gdb to COW and set breakpoints
 192          */
 193         vma = _install_special_mapping(mm,
 194                                        text_start,
 195                                        image->size,
 196                                        VM_READ|VM_EXEC|
 197                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 198                                        &text_mapping);
 199
 200         if (IS_ERR(vma)) {
 201                 ret = PTR_ERR(vma);
 202                 goto up_fail;
 203         }
 204
 205         vma = _install_special_mapping(mm,
 206                                        addr,
 207                                        -image->sym_vvar_start,
 208                                        VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
 209                                        VM_PFNMAP,
 210                                        &vvar_mapping);
 211
 212         if (IS_ERR(vma)) {
 213                 ret = PTR_ERR(vma);
 214                 goto up_fail;
 215         }
 216
 217 up_fail:
 218         if (ret)
 219                 current->mm->context.vdso = NULL;
 220
 221         up_write(&mm->mmap_sem);
 222         return ret;
 223 }
 224
 225 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 226 static int load_vdso32(void)
 227 {
 228         if (vdso32_enabled != 1)  /* Other values all mean "disabled" */
 229                 return 0;
 230
 231         return map_vdso(&vdso_image_32, false);
 232 }
 233 #endif
 234
 235 #ifdef CONFIG_X86_64
 236 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 237 {
 238         if (!vdso64_enabled)
 239                 return 0;
 240
 241         return map_vdso(&vdso_image_64, true);
 242 }
 243
 244 #ifdef CONFIG_COMPAT
 245 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
 246                                        int uses_interp)
 247 {
 248 #ifdef CONFIG_X86_X32_ABI
 249         if (test_thread_flag(TIF_X32)) {
 250                 if (!vdso64_enabled)
 251                         return 0;
 252
 253                 return map_vdso(&vdso_image_x32, true);
 254         }
 255 #endif
 256 #ifdef CONFIG_IA32_EMULATION
 257         return load_vdso32();
 258 #else
 259         return 0;
 260 #endif
 261 }
 262 #endif
 263 #else
 264 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 265 {
 266         return load_vdso32();
 267 }
 268 #endif
 269
 270 #ifdef CONFIG_X86_64
 271 static __init int vdso_setup(char *s)
 272 {
 273         vdso64_enabled = simple_strtoul(s, NULL, 0);
 274         return 0;
 275 }
 276 __setup("vdso=", vdso_setup);
 277 #endif
 278
 279 #ifdef CONFIG_X86_64
 280 static void vgetcpu_cpu_init(void *arg)
 281 {
 282         int cpu = smp_processor_id();
 283         struct desc_struct d = { };
 284         unsigned long node = 0;
 285 #ifdef CONFIG_NUMA
 286         node = cpu_to_node(cpu);
 287 #endif
 288         if (static_cpu_has(X86_FEATURE_RDTSCP))
 289                 write_rdtscp_aux((node << 12) | cpu);
 290
 291         /*
 292          * Store cpu number in limit so that it can be loaded
 293          * quickly in user space in vgetcpu. (12 bits for the CPU
 294          * and 8 bits for the node)
 295          */
 296         d.limit0 = cpu | ((node & 0xf) << 12);
 297         d.limit = node >> 4;
 298         d.type = 5;             /* RO data, expand down, accessed */
 299         d.dpl = 3;              /* Visible to user code */
 300         d.s = 1;                /* Not a system segment */
 301         d.p = 1;                /* Present */
 302         d.d = 1;                /* 32-bit */
 303
 304         write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 305 }
 306
 307 static int
 308 vgetcpu_cpu_notifier(struct notifier_block *n, unsigned long action, void *arg)
 309 {
 310         long cpu = (long)arg;
 311
 312         if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
 313                 smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
 314
 315         return NOTIFY_DONE;
 316 }
 317
 318 static int __init init_vdso(void)
 319 {
 320         init_vdso_image(&vdso_image_64);
 321
 322 #ifdef CONFIG_X86_X32_ABI
 323         init_vdso_image(&vdso_image_x32);
 324 #endif
 325
 326         cpu_notifier_register_begin();
 327
 328         on_each_cpu(vgetcpu_cpu_init, NULL, 1);
 329         /* notifier priority > KVM */
 330         __hotcpu_notifier(vgetcpu_cpu_notifier, 30);
 331
 332         cpu_notifier_register_done();
 333
 334         return 0;
 335 }
 336 subsys_initcall(init_vdso);
 337 #endif /* CONFIG_X86_64 */