arch/x86/vdso/vdso32-setup.c

   1 /*
   2  * (C) Copyright 2002 Linus Torvalds
   3  * Portions based on the vdso-randomization code from exec-shield:
   4  * Copyright(C) 2005-2006, Red Hat, Inc., Ingo Molnar
   5  *
   6  * This file contains the needed initializations to support sysenter.
   7  */
   8
   9 #include <linux/init.h>
  10 #include <linux/smp.h>
  11 #include <linux/thread_info.h>
  12 #include <linux/sched.h>
  13 #include <linux/gfp.h>
  14 #include <linux/string.h>
  15 #include <linux/elf.h>
  16 #include <linux/mm.h>
  17 #include <linux/err.h>
  18 #include <linux/module.h>
  19
  20 #include <asm/cpufeature.h>
  21 #include <asm/msr.h>
  22 #include <asm/pgtable.h>
  23 #include <asm/unistd.h>
  24 #include <asm/elf.h>
  25 #include <asm/tlbflush.h>
  26 #include <asm/vdso.h>
  27 #include <asm/proto.h>
  28
  29 enum {
  30         VDSO_DISABLED = 0,
  31         VDSO_ENABLED = 1,
  32         VDSO_COMPAT = 2,
  33 };
  34
  35 #ifdef CONFIG_COMPAT_VDSO
  36 #define VDSO_DEFAULT    VDSO_COMPAT
  37 #else
  38 #define VDSO_DEFAULT    VDSO_ENABLED
  39 #endif
  40
  41 #ifdef CONFIG_X86_64
  42 #define vdso_enabled                    sysctl_vsyscall32
  43 #define arch_setup_additional_pages     syscall32_setup_pages
  44 extern int sysctl_ldt16;
  45 #endif
  46
  47 /*
  48  * This is the difference between the prelinked addresses in the vDSO images
  49  * and the VDSO_HIGH_BASE address where CONFIG_COMPAT_VDSO places the vDSO
  50  * in the user address space.
  51  */
  52 #define VDSO_ADDR_ADJUST        (VDSO_HIGH_BASE - (unsigned long)VDSO32_PRELINK)
  53
  54 /*
  55  * Should the kernel map a VDSO page into processes and pass its
  56  * address down to glibc upon exec()?
  57  */
  58 unsigned int __read_mostly vdso_enabled = VDSO_DEFAULT;
  59
  60 static int __init vdso_setup(char *s)
  61 {
  62         vdso_enabled = simple_strtoul(s, NULL, 0);
  63
  64         return 1;
  65 }
  66
  67 /*
  68  * For consistency, the argument vdso32=[012] affects the 32-bit vDSO
  69  * behavior on both 64-bit and 32-bit kernels.
  70  * On 32-bit kernels, vdso=[012] means the same thing.
  71  */
  72 __setup("vdso32=", vdso_setup);
  73
  74 #ifdef CONFIG_X86_32
  75 __setup_param("vdso=", vdso32_setup, vdso_setup, 0);
  76
  77 EXPORT_SYMBOL_GPL(vdso_enabled);
  78 #endif
  79
  80 static __init void reloc_symtab(Elf32_Ehdr *ehdr,
  81                                 unsigned offset, unsigned size)
  82 {
  83         Elf32_Sym *sym = (void *)ehdr + offset;
  84         unsigned nsym = size / sizeof(*sym);
  85         unsigned i;
  86
  87         for(i = 0; i < nsym; i++, sym++) {
  88                 if (sym->st_shndx == SHN_UNDEF ||
  89                     sym->st_shndx == SHN_ABS)
  90                         continue;  /* skip */
  91
  92                 if (sym->st_shndx > SHN_LORESERVE) {
  93                         printk(KERN_INFO "VDSO: unexpected st_shndx %x\n",
  94                                sym->st_shndx);
  95                         continue;
  96                 }
  97
  98                 switch(ELF_ST_TYPE(sym->st_info)) {
  99                 case STT_OBJECT:
 100                 case STT_FUNC:
 101                 case STT_SECTION:
 102                 case STT_FILE:
 103                         sym->st_value += VDSO_ADDR_ADJUST;
 104                 }
 105         }
 106 }
 107
 108 static __init void reloc_dyn(Elf32_Ehdr *ehdr, unsigned offset)
 109 {
 110         Elf32_Dyn *dyn = (void *)ehdr + offset;
 111
 112         for(; dyn->d_tag != DT_NULL; dyn++)
 113                 switch(dyn->d_tag) {
 114                 case DT_PLTGOT:
 115                 case DT_HASH:
 116                 case DT_STRTAB:
 117                 case DT_SYMTAB:
 118                 case DT_RELA:
 119                 case DT_INIT:
 120                 case DT_FINI:
 121                 case DT_REL:
 122                 case DT_DEBUG:
 123                 case DT_JMPREL:
 124                 case DT_VERSYM:
 125                 case DT_VERDEF:
 126                 case DT_VERNEED:
 127                 case DT_ADDRRNGLO ... DT_ADDRRNGHI:
 128                         /* definitely pointers needing relocation */
 129                         dyn->d_un.d_ptr += VDSO_ADDR_ADJUST;
 130                         break;
 131
 132                 case DT_ENCODING ... OLD_DT_LOOS-1:
 133                 case DT_LOOS ... DT_HIOS-1:
 134                         /* Tags above DT_ENCODING are pointers if
 135                            they're even */
 136                         if (dyn->d_tag >= DT_ENCODING &&
 137                             (dyn->d_tag & 1) == 0)
 138                                 dyn->d_un.d_ptr += VDSO_ADDR_ADJUST;
 139                         break;
 140
 141                 case DT_VERDEFNUM:
 142                 case DT_VERNEEDNUM:
 143                 case DT_FLAGS_1:
 144                 case DT_RELACOUNT:
 145                 case DT_RELCOUNT:
 146                 case DT_VALRNGLO ... DT_VALRNGHI:
 147                         /* definitely not pointers */
 148                         break;
 149
 150                 case OLD_DT_LOOS ... DT_LOOS-1:
 151                 case DT_HIOS ... DT_VALRNGLO-1:
 152                 default:
 153                         if (dyn->d_tag > DT_ENCODING)
 154                                 printk(KERN_INFO "VDSO: unexpected DT_tag %x\n",
 155                                        dyn->d_tag);
 156                         break;
 157                 }
 158 }
 159
 160 static __init void relocate_vdso(Elf32_Ehdr *ehdr)
 161 {
 162         Elf32_Phdr *phdr;
 163         Elf32_Shdr *shdr;
 164         int i;
 165
 166         BUG_ON(memcmp(ehdr->e_ident, ELFMAG, SELFMAG) != 0 ||
 167                !elf_check_arch_ia32(ehdr) ||
 168                ehdr->e_type != ET_DYN);
 169
 170         ehdr->e_entry += VDSO_ADDR_ADJUST;
 171
 172         /* rebase phdrs */
 173         phdr = (void *)ehdr + ehdr->e_phoff;
 174         for (i = 0; i < ehdr->e_phnum; i++) {
 175                 phdr[i].p_vaddr += VDSO_ADDR_ADJUST;
 176
 177                 /* relocate dynamic stuff */
 178                 if (phdr[i].p_type == PT_DYNAMIC)
 179                         reloc_dyn(ehdr, phdr[i].p_offset);
 180         }
 181
 182         /* rebase sections */
 183         shdr = (void *)ehdr + ehdr->e_shoff;
 184         for(i = 0; i < ehdr->e_shnum; i++) {
 185                 if (!(shdr[i].sh_flags & SHF_ALLOC))
 186                         continue;
 187
 188                 shdr[i].sh_addr += VDSO_ADDR_ADJUST;
 189
 190                 if (shdr[i].sh_type == SHT_SYMTAB ||
 191                     shdr[i].sh_type == SHT_DYNSYM)
 192                         reloc_symtab(ehdr, shdr[i].sh_offset,
 193                                      shdr[i].sh_size);
 194         }
 195 }
 196
 197 static struct page *vdso32_pages[1];
 198
 199 #ifdef CONFIG_X86_64
 200
 201 #define vdso32_sysenter()       (boot_cpu_has(X86_FEATURE_SYSENTER32))
 202 #define vdso32_syscall()        (boot_cpu_has(X86_FEATURE_SYSCALL32))
 203
 204 /* May not be __init: called during resume */
 205 void syscall32_cpu_init(void)
 206 {
 207         /* Load these always in case some future AMD CPU supports
 208            SYSENTER from compat mode too. */
 209         wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
 210         wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
 211         wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)ia32_sysenter_target);
 212
 213         wrmsrl(MSR_CSTAR, ia32_cstar_target);
 214 }
 215
 216 #define compat_uses_vma         1
 217
 218 static inline void map_compat_vdso(int map)
 219 {
 220 }
 221
 222 #else  /* CONFIG_X86_32 */
 223
 224 #define vdso32_sysenter()       (boot_cpu_has(X86_FEATURE_SEP))
 225 #define vdso32_syscall()        (0)
 226
 227 void enable_sep_cpu(void)
 228 {
 229         int cpu = get_cpu();
 230         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 231
 232         if (!boot_cpu_has(X86_FEATURE_SEP)) {
 233                 put_cpu();
 234                 return;
 235         }
 236
 237         tss->x86_tss.ss1 = __KERNEL_CS;
 238         tss->x86_tss.sp1 = sizeof(struct tss_struct) + (unsigned long) tss;
 239         wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
 240         wrmsr(MSR_IA32_SYSENTER_ESP, tss->x86_tss.sp1, 0);
 241         wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) ia32_sysenter_target, 0);
 242         put_cpu();
 243 }
 244
 245 static struct vm_area_struct gate_vma;
 246
 247 static int __init gate_vma_init(void)
 248 {
 249         gate_vma.vm_mm = NULL;
 250         gate_vma.vm_start = FIXADDR_USER_START;
 251         gate_vma.vm_end = FIXADDR_USER_END;
 252         gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
 253         gate_vma.vm_page_prot = __P101;
 254
 255         return 0;
 256 }
 257
 258 #define compat_uses_vma         0
 259
 260 static void map_compat_vdso(int map)
 261 {
 262         static int vdso_mapped;
 263
 264         if (map == vdso_mapped)
 265                 return;
 266
 267         vdso_mapped = map;
 268
 269         __set_fixmap(FIX_VDSO, page_to_pfn(vdso32_pages[0]) << PAGE_SHIFT,
 270                      map ? PAGE_READONLY_EXEC : PAGE_NONE);
 271
 272         /* flush stray tlbs */
 273         flush_tlb_all();
 274 }
 275
 276 #endif  /* CONFIG_X86_64 */
 277
 278 int __init sysenter_setup(void)
 279 {
 280         void *syscall_page = (void *)get_zeroed_page(GFP_ATOMIC);
 281         const void *vsyscall;
 282         size_t vsyscall_len;
 283
 284         vdso32_pages[0] = virt_to_page(syscall_page);
 285
 286 #ifdef CONFIG_X86_32
 287         gate_vma_init();
 288 #endif
 289
 290         if (vdso32_syscall()) {
 291                 vsyscall = &vdso32_syscall_start;
 292                 vsyscall_len = &vdso32_syscall_end - &vdso32_syscall_start;
 293         } else if (vdso32_sysenter()){
 294                 vsyscall = &vdso32_sysenter_start;
 295                 vsyscall_len = &vdso32_sysenter_end - &vdso32_sysenter_start;
 296         } else {
 297                 vsyscall = &vdso32_int80_start;
 298                 vsyscall_len = &vdso32_int80_end - &vdso32_int80_start;
 299         }
 300
 301         memcpy(syscall_page, vsyscall, vsyscall_len);
 302         relocate_vdso(syscall_page);
 303
 304         return 0;
 305 }
 306
 307 /* Setup a VMA at program startup for the vsyscall page */
 308 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 309 {
 310         struct mm_struct *mm = current->mm;
 311         unsigned long addr;
 312         int ret = 0;
 313         bool compat;
 314
 315 #ifdef CONFIG_X86_X32_ABI
 316         if (test_thread_flag(TIF_X32))
 317                 return x32_setup_additional_pages(bprm, uses_interp);
 318 #endif
 319
 320         if (vdso_enabled == VDSO_DISABLED)
 321                 return 0;
 322
 323         down_write(&mm->mmap_sem);
 324
 325         /* Test compat mode once here, in case someone
 326            changes it via sysctl */
 327         compat = (vdso_enabled == VDSO_COMPAT);
 328
 329         map_compat_vdso(compat);
 330
 331         if (compat)
 332                 addr = VDSO_HIGH_BASE;
 333         else {
 334                 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 335                 if (IS_ERR_VALUE(addr)) {
 336                         ret = addr;
 337                         goto up_fail;
 338                 }
 339         }
 340
 341         current->mm->context.vdso = (void *)addr;
 342
 343         if (compat_uses_vma || !compat) {
 344                 /*
 345                  * MAYWRITE to allow gdb to COW and set breakpoints
 346                  */
 347                 ret = install_special_mapping(mm, addr, PAGE_SIZE,
 348                                               VM_READ|VM_EXEC|
 349                                               VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 350                                               vdso32_pages);
 351
 352                 if (ret)
 353                         goto up_fail;
 354         }
 355
 356         current_thread_info()->sysenter_return =
 357                 VDSO32_SYMBOL(addr, SYSENTER_RETURN);
 358
 359   up_fail:
 360         if (ret)
 361                 current->mm->context.vdso = NULL;
 362
 363         up_write(&mm->mmap_sem);
 364
 365         return ret;
 366 }
 367
 368 #ifdef CONFIG_X86_64
 369
 370 subsys_initcall(sysenter_setup);
 371
 372 #ifdef CONFIG_SYSCTL
 373 /* Register vsyscall32 into the ABI table */
 374 #include <linux/sysctl.h>
 375
 376 static ctl_table abi_table2[] = {
 377         {
 378                 .procname       = "vsyscall32",
 379                 .data           = &sysctl_vsyscall32,
 380                 .maxlen         = sizeof(int),
 381                 .mode           = 0644,
 382                 .proc_handler   = proc_dointvec
 383         },
 384         {
 385                 .procname       = "ldt16",
 386                 .data           = &sysctl_ldt16,
 387                 .maxlen         = sizeof(int),
 388                 .mode           = 0644,
 389                 .proc_handler   = proc_dointvec
 390         },
 391         {}
 392 };
 393
 394 static ctl_table abi_root_table2[] = {
 395         {
 396                 .procname = "abi",
 397                 .mode = 0555,
 398                 .child = abi_table2
 399         },
 400         {}
 401 };
 402
 403 static __init int ia32_binfmt_init(void)
 404 {
 405         register_sysctl_table(abi_root_table2);
 406         return 0;
 407 }
 408 __initcall(ia32_binfmt_init);
 409 #endif
 410
 411 #else  /* CONFIG_X86_32 */
 412
 413 const char *arch_vma_name(struct vm_area_struct *vma)
 414 {
 415         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 416                 return "[vdso]";
 417         return NULL;
 418 }
 419
 420 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
 421 {
 422         /*
 423          * Check to see if the corresponding task was created in compat vdso
 424          * mode.
 425          */
 426         if (mm && mm->context.vdso == (void *)VDSO_HIGH_BASE)
 427                 return &gate_vma;
 428         return NULL;
 429 }
 430
 431 int in_gate_area(struct mm_struct *mm, unsigned long addr)
 432 {
 433         const struct vm_area_struct *vma = get_gate_vma(mm);
 434
 435         return vma && addr >= vma->vm_start && addr < vma->vm_end;
 436 }
 437
 438 int in_gate_area_no_mm(unsigned long addr)
 439 {
 440         return 0;
 441 }
 442
 443 #endif  /* CONFIG_X86_64 */