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