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/sched/task_stack.h>
  11 #include <linux/slab.h>
  12 #include <linux/init.h>
  13 #include <linux/random.h>
  14 #include <linux/elf.h>
  15 #include <linux/cpu.h>
  16 #include <linux/ptrace.h>
  17 #include <asm/pvclock.h>
  18 #include <asm/vgtod.h>
  19 #include <asm/proto.h>
  20 #include <asm/vdso.h>
  21 #include <asm/vvar.h>
  22 #include <asm/page.h>
  23 #include <asm/desc.h>
  24 #include <asm/cpufeature.h>
  25 #include <asm/mshyperv.h>
  26
  27 #if defined(CONFIG_X86_64)
  28 unsigned int __read_mostly vdso64_enabled = 1;
  29 #endif
  30
  31 void __init init_vdso_image(const struct vdso_image *image)
  32 {
  33         BUG_ON(image->size % PAGE_SIZE != 0);
  34
  35         apply_alternatives((struct alt_instr *)(image->data + image->alt),
  36                            (struct alt_instr *)(image->data + image->alt +
  37                                                 image->alt_len));
  38 }
  39
  40 struct linux_binprm;
  41
  42 static int vdso_fault(const struct vm_special_mapping *sm,
  43                       struct vm_area_struct *vma, struct vm_fault *vmf)
  44 {
  45         const struct vdso_image *image = vma->vm_mm->context.vdso_image;
  46
  47         if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
  48                 return VM_FAULT_SIGBUS;
  49
  50         vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
  51         get_page(vmf->page);
  52         return 0;
  53 }
  54
  55 static void vdso_fix_landing(const struct vdso_image *image,
  56                 struct vm_area_struct *new_vma)
  57 {
  58 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
  59         if (in_ia32_syscall() && image == &vdso_image_32) {
  60                 struct pt_regs *regs = current_pt_regs();
  61                 unsigned long vdso_land = image->sym_int80_landing_pad;
  62                 unsigned long old_land_addr = vdso_land +
  63                         (unsigned long)current->mm->context.vdso;
  64
  65                 /* Fixing userspace landing - look at do_fast_syscall_32 */
  66                 if (regs->ip == old_land_addr)
  67                         regs->ip = new_vma->vm_start + vdso_land;
  68         }
  69 #endif
  70 }
  71
  72 static int vdso_mremap(const struct vm_special_mapping *sm,
  73                 struct vm_area_struct *new_vma)
  74 {
  75         unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
  76         const struct vdso_image *image = current->mm->context.vdso_image;
  77
  78         if (image->size != new_size)
  79                 return -EINVAL;
  80
  81         vdso_fix_landing(image, new_vma);
  82         current->mm->context.vdso = (void __user *)new_vma->vm_start;
  83
  84         return 0;
  85 }
  86
  87 static int vvar_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         long sym_offset;
  92         int ret = -EFAULT;
  93
  94         if (!image)
  95                 return VM_FAULT_SIGBUS;
  96
  97         sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
  98                 image->sym_vvar_start;
  99
 100         /*
 101          * Sanity check: a symbol offset of zero means that the page
 102          * does not exist for this vdso image, not that the page is at
 103          * offset zero relative to the text mapping.  This should be
 104          * impossible here, because sym_offset should only be zero for
 105          * the page past the end of the vvar mapping.
 106          */
 107         if (sym_offset == 0)
 108                 return VM_FAULT_SIGBUS;
 109
 110         if (sym_offset == image->sym_vvar_page) {
 111                 ret = vm_insert_pfn(vma, vmf->address,
 112                                     __pa_symbol(&__vvar_page) >> PAGE_SHIFT);
 113         } else if (sym_offset == image->sym_pvclock_page) {
 114                 struct pvclock_vsyscall_time_info *pvti =
 115                         pvclock_get_pvti_cpu0_va();
 116                 if (pvti && vclock_was_used(VCLOCK_PVCLOCK)) {
 117                         ret = vm_insert_pfn_prot(
 118                                 vma,
 119                                 vmf->address,
 120                                 __pa(pvti) >> PAGE_SHIFT,
 121                                 pgprot_decrypted(vma->vm_page_prot));
 122                 }
 123         } else if (sym_offset == image->sym_hvclock_page) {
 124                 struct ms_hyperv_tsc_page *tsc_pg = hv_get_tsc_page();
 125
 126                 if (tsc_pg && vclock_was_used(VCLOCK_HVCLOCK))
 127                         ret = vm_insert_pfn(vma, vmf->address,
 128                                             vmalloc_to_pfn(tsc_pg));
 129         }
 130
 131         if (ret == 0 || ret == -EBUSY)
 132                 return VM_FAULT_NOPAGE;
 133
 134         return VM_FAULT_SIGBUS;
 135 }
 136
 137 static const struct vm_special_mapping vdso_mapping = {
 138         .name = "[vdso]",
 139         .fault = vdso_fault,
 140         .mremap = vdso_mremap,
 141 };
 142 static const struct vm_special_mapping vvar_mapping = {
 143         .name = "[vvar]",
 144         .fault = vvar_fault,
 145 };
 146
 147 /*
 148  * Add vdso and vvar mappings to current process.
 149  * @image          - blob to map
 150  * @addr           - request a specific address (zero to map at free addr)
 151  */
 152 static int map_vdso(const struct vdso_image *image, unsigned long addr)
 153 {
 154         struct mm_struct *mm = current->mm;
 155         struct vm_area_struct *vma;
 156         unsigned long text_start;
 157         int ret = 0;
 158
 159         if (down_write_killable(&mm->mmap_sem))
 160                 return -EINTR;
 161
 162         addr = get_unmapped_area(NULL, addr,
 163                                  image->size - image->sym_vvar_start, 0, 0);
 164         if (IS_ERR_VALUE(addr)) {
 165                 ret = addr;
 166                 goto up_fail;
 167         }
 168
 169         text_start = addr - image->sym_vvar_start;
 170
 171         /*
 172          * MAYWRITE to allow gdb to COW and set breakpoints
 173          */
 174         vma = _install_special_mapping(mm,
 175                                        text_start,
 176                                        image->size,
 177                                        VM_READ|VM_EXEC|
 178                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 179                                        &vdso_mapping);
 180
 181         if (IS_ERR(vma)) {
 182                 ret = PTR_ERR(vma);
 183                 goto up_fail;
 184         }
 185
 186         vma = _install_special_mapping(mm,
 187                                        addr,
 188                                        -image->sym_vvar_start,
 189                                        VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
 190                                        VM_PFNMAP,
 191                                        &vvar_mapping);
 192
 193         if (IS_ERR(vma)) {
 194                 ret = PTR_ERR(vma);
 195                 do_munmap(mm, text_start, image->size, NULL);
 196         } else {
 197                 current->mm->context.vdso = (void __user *)text_start;
 198                 current->mm->context.vdso_image = image;
 199         }
 200
 201 up_fail:
 202         up_write(&mm->mmap_sem);
 203         return ret;
 204 }
 205
 206 #ifdef CONFIG_X86_64
 207 /*
 208  * Put the vdso above the (randomized) stack with another randomized
 209  * offset.  This way there is no hole in the middle of address space.
 210  * To save memory make sure it is still in the same PTE as the stack
 211  * top.  This doesn't give that many random bits.
 212  *
 213  * Note that this algorithm is imperfect: the distribution of the vdso
 214  * start address within a PMD is biased toward the end.
 215  *
 216  * Only used for the 64-bit and x32 vdsos.
 217  */
 218 static unsigned long vdso_addr(unsigned long start, unsigned len)
 219 {
 220         unsigned long addr, end;
 221         unsigned offset;
 222
 223         /*
 224          * Round up the start address.  It can start out unaligned as a result
 225          * of stack start randomization.
 226          */
 227         start = PAGE_ALIGN(start);
 228
 229         /* Round the lowest possible end address up to a PMD boundary. */
 230         end = (start + len + PMD_SIZE - 1) & PMD_MASK;
 231         if (end >= TASK_SIZE_MAX)
 232                 end = TASK_SIZE_MAX;
 233         end -= len;
 234
 235         if (end > start) {
 236                 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
 237                 addr = start + (offset << PAGE_SHIFT);
 238         } else {
 239                 addr = start;
 240         }
 241
 242         /*
 243          * Forcibly align the final address in case we have a hardware
 244          * issue that requires alignment for performance reasons.
 245          */
 246         addr = align_vdso_addr(addr);
 247
 248         return addr;
 249 }
 250
 251 static int map_vdso_randomized(const struct vdso_image *image)
 252 {
 253         unsigned long addr = vdso_addr(current->mm->start_stack, image->size-image->sym_vvar_start);
 254
 255         return map_vdso(image, addr);
 256 }
 257 #endif
 258
 259 int map_vdso_once(const struct vdso_image *image, unsigned long addr)
 260 {
 261         struct mm_struct *mm = current->mm;
 262         struct vm_area_struct *vma;
 263
 264         down_write(&mm->mmap_sem);
 265         /*
 266          * Check if we have already mapped vdso blob - fail to prevent
 267          * abusing from userspace install_speciall_mapping, which may
 268          * not do accounting and rlimit right.
 269          * We could search vma near context.vdso, but it's a slowpath,
 270          * so let's explicitely check all VMAs to be completely sure.
 271          */
 272         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 273                 if (vma_is_special_mapping(vma, &vdso_mapping) ||
 274                                 vma_is_special_mapping(vma, &vvar_mapping)) {
 275                         up_write(&mm->mmap_sem);
 276                         return -EEXIST;
 277                 }
 278         }
 279         up_write(&mm->mmap_sem);
 280
 281         return map_vdso(image, addr);
 282 }
 283
 284 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 285 static int load_vdso32(void)
 286 {
 287         if (vdso32_enabled != 1)  /* Other values all mean "disabled" */
 288                 return 0;
 289
 290         return map_vdso(&vdso_image_32, 0);
 291 }
 292 #endif
 293
 294 #ifdef CONFIG_X86_64
 295 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 296 {
 297         if (!vdso64_enabled)
 298                 return 0;
 299
 300         return map_vdso_randomized(&vdso_image_64);
 301 }
 302
 303 #ifdef CONFIG_COMPAT
 304 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
 305                                        int uses_interp)
 306 {
 307 #ifdef CONFIG_X86_X32_ABI
 308         if (test_thread_flag(TIF_X32)) {
 309                 if (!vdso64_enabled)
 310                         return 0;
 311                 return map_vdso_randomized(&vdso_image_x32);
 312         }
 313 #endif
 314 #ifdef CONFIG_IA32_EMULATION
 315         return load_vdso32();
 316 #else
 317         return 0;
 318 #endif
 319 }
 320 #endif
 321 #else
 322 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 323 {
 324         return load_vdso32();
 325 }
 326 #endif
 327
 328 #ifdef CONFIG_X86_64
 329 static __init int vdso_setup(char *s)
 330 {
 331         vdso64_enabled = simple_strtoul(s, NULL, 0);
 332         return 0;
 333 }
 334 __setup("vdso=", vdso_setup);
 335 #endif
 336
 337 #ifdef CONFIG_X86_64
 338 static void vgetcpu_cpu_init(void *arg)
 339 {
 340         int cpu = smp_processor_id();
 341         struct desc_struct d = { };
 342         unsigned long node = 0;
 343 #ifdef CONFIG_NUMA
 344         node = cpu_to_node(cpu);
 345 #endif
 346         if (static_cpu_has(X86_FEATURE_RDTSCP))
 347                 write_rdtscp_aux((node << 12) | cpu);
 348
 349         /*
 350          * Store cpu number in limit so that it can be loaded
 351          * quickly in user space in vgetcpu. (12 bits for the CPU
 352          * and 8 bits for the node)
 353          */
 354         d.limit0 = cpu | ((node & 0xf) << 12);
 355         d.limit1 = node >> 4;
 356         d.type = 5;             /* RO data, expand down, accessed */
 357         d.dpl = 3;              /* Visible to user code */
 358         d.s = 1;                /* Not a system segment */
 359         d.p = 1;                /* Present */
 360         d.d = 1;                /* 32-bit */
 361
 362         write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_PER_CPU, &d, DESCTYPE_S);
 363 }
 364
 365 static int vgetcpu_online(unsigned int cpu)
 366 {
 367         return smp_call_function_single(cpu, vgetcpu_cpu_init, NULL, 1);
 368 }
 369
 370 static int __init init_vdso(void)
 371 {
 372         init_vdso_image(&vdso_image_64);
 373
 374 #ifdef CONFIG_X86_X32_ABI
 375         init_vdso_image(&vdso_image_x32);
 376 #endif
 377
 378         /* notifier priority > KVM */
 379         return cpuhp_setup_state(CPUHP_AP_X86_VDSO_VMA_ONLINE,
 380                                  "x86/vdso/vma:online", vgetcpu_online, NULL);
 381 }
 382 subsys_initcall(init_vdso);
 383 #endif /* CONFIG_X86_64 */