arch/x86/vdso/vdso32-setup-xen.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 #include <xen/interface/callback.h>
  30
  31 enum {
  32         VDSO_DISABLED = 0,
  33         VDSO_ENABLED = 1,
  34         VDSO_COMPAT = 2,
  35 };
  36
  37 #ifdef CONFIG_COMPAT_VDSO
  38 #define VDSO_DEFAULT    VDSO_COMPAT
  39 #else
  40 #define VDSO_DEFAULT    VDSO_ENABLED
  41 #endif
  42
  43 #ifdef CONFIG_X86_64
  44 #define vdso_enabled                    sysctl_vsyscall32
  45 #define arch_setup_additional_pages     syscall32_setup_pages
  46 #endif
  47
  48 /*
  49  * This is the difference between the prelinked addresses in the vDSO images
  50  * and the VDSO_HIGH_BASE address where CONFIG_COMPAT_VDSO places the vDSO
  51  * in the user address space.
  52  */
  53 #define VDSO_ADDR_ADJUST        (VDSO_HIGH_BASE - (unsigned long)VDSO32_PRELINK)
  54
  55 /*
  56  * Should the kernel map a VDSO page into processes and pass its
  57  * address down to glibc upon exec()?
  58  */
  59 unsigned int __read_mostly vdso_enabled = VDSO_DEFAULT;
  60
  61 static int __init vdso_setup(char *s)
  62 {
  63         vdso_enabled = simple_strtoul(s, NULL, 0);
  64
  65         return 1;
  66 }
  67
  68 /*
  69  * For consistency, the argument vdso32=[012] affects the 32-bit vDSO
  70  * behavior on both 64-bit and 32-bit kernels.
  71  * On 32-bit kernels, vdso=[012] means the same thing.
  72  */
  73 __setup("vdso32=", vdso_setup);
  74
  75 #ifdef CONFIG_X86_32
  76 __setup_param("vdso=", vdso32_setup, vdso_setup, 0);
  77
  78 EXPORT_SYMBOL_GPL(vdso_enabled);
  79 #endif
  80
  81 static __init void reloc_symtab(Elf32_Ehdr *ehdr,
  82                                 unsigned offset, unsigned size)
  83 {
  84         Elf32_Sym *sym = (void *)ehdr + offset;
  85         unsigned nsym = size / sizeof(*sym);
  86         unsigned i;
  87
  88         for(i = 0; i < nsym; i++, sym++) {
  89                 if (sym->st_shndx == SHN_UNDEF ||
  90                     sym->st_shndx == SHN_ABS)
  91                         continue;  /* skip */
  92
  93                 if (sym->st_shndx > SHN_LORESERVE) {
  94                         printk(KERN_INFO "VDSO: unexpected st_shndx %x\n",
  95                                sym->st_shndx);
  96                         continue;
  97                 }
  98
  99                 switch(ELF_ST_TYPE(sym->st_info)) {
 100                 case STT_OBJECT:
 101                 case STT_FUNC:
 102                 case STT_SECTION:
 103                 case STT_FILE:
 104                         sym->st_value += VDSO_ADDR_ADJUST;
 105                 }
 106         }
 107 }
 108
 109 static __init void reloc_dyn(Elf32_Ehdr *ehdr, unsigned offset)
 110 {
 111         Elf32_Dyn *dyn = (void *)ehdr + offset;
 112
 113         for(; dyn->d_tag != DT_NULL; dyn++)
 114                 switch(dyn->d_tag) {
 115                 case DT_PLTGOT:
 116                 case DT_HASH:
 117                 case DT_STRTAB:
 118                 case DT_SYMTAB:
 119                 case DT_RELA:
 120                 case DT_INIT:
 121                 case DT_FINI:
 122                 case DT_REL:
 123                 case DT_DEBUG:
 124                 case DT_JMPREL:
 125                 case DT_VERSYM:
 126                 case DT_VERDEF:
 127                 case DT_VERNEED:
 128                 case DT_ADDRRNGLO ... DT_ADDRRNGHI:
 129                         /* definitely pointers needing relocation */
 130                         dyn->d_un.d_ptr += VDSO_ADDR_ADJUST;
 131                         break;
 132
 133                 case DT_ENCODING ... OLD_DT_LOOS-1:
 134                 case DT_LOOS ... DT_HIOS-1:
 135                         /* Tags above DT_ENCODING are pointers if
 136                            they're even */
 137                         if (dyn->d_tag >= DT_ENCODING &&
 138                             (dyn->d_tag & 1) == 0)
 139                                 dyn->d_un.d_ptr += VDSO_ADDR_ADJUST;
 140                         break;
 141
 142                 case DT_VERDEFNUM:
 143                 case DT_VERNEEDNUM:
 144                 case DT_FLAGS_1:
 145                 case DT_RELACOUNT:
 146                 case DT_RELCOUNT:
 147                 case DT_VALRNGLO ... DT_VALRNGHI:
 148                         /* definitely not pointers */
 149                         break;
 150
 151                 case OLD_DT_LOOS ... DT_LOOS-1:
 152                 case DT_HIOS ... DT_VALRNGLO-1:
 153                 default:
 154                         if (dyn->d_tag > DT_ENCODING)
 155                                 printk(KERN_INFO "VDSO: unexpected DT_tag %x\n",
 156                                        dyn->d_tag);
 157                         break;
 158                 }
 159 }
 160
 161 static __init void relocate_vdso(Elf32_Ehdr *ehdr)
 162 {
 163         Elf32_Phdr *phdr;
 164         Elf32_Shdr *shdr;
 165         int i;
 166
 167         BUG_ON(memcmp(ehdr->e_ident, ELFMAG, SELFMAG) != 0 ||
 168                !elf_check_arch_ia32(ehdr) ||
 169                ehdr->e_type != ET_DYN);
 170
 171         ehdr->e_entry += VDSO_ADDR_ADJUST;
 172
 173         /* rebase phdrs */
 174         phdr = (void *)ehdr + ehdr->e_phoff;
 175         for (i = 0; i < ehdr->e_phnum; i++) {
 176                 phdr[i].p_vaddr += VDSO_ADDR_ADJUST;
 177
 178                 /* relocate dynamic stuff */
 179                 if (phdr[i].p_type == PT_DYNAMIC)
 180                         reloc_dyn(ehdr, phdr[i].p_offset);
 181         }
 182
 183         /* rebase sections */
 184         shdr = (void *)ehdr + ehdr->e_shoff;
 185         for(i = 0; i < ehdr->e_shnum; i++) {
 186                 if (!(shdr[i].sh_flags & SHF_ALLOC))
 187                         continue;
 188
 189                 shdr[i].sh_addr += VDSO_ADDR_ADJUST;
 190
 191                 if (shdr[i].sh_type == SHT_SYMTAB ||
 192                     shdr[i].sh_type == SHT_DYNSYM)
 193                         reloc_symtab(ehdr, shdr[i].sh_offset,
 194                                      shdr[i].sh_size);
 195         }
 196 }
 197
 198 static struct page *vdso32_pages[1];
 199
 200 #ifdef CONFIG_X86_64
 201
 202 #define vdso32_sysenter()       (boot_cpu_has(X86_FEATURE_SYSENTER32))
 203 #define vdso32_syscall()        (boot_cpu_has(X86_FEATURE_SYSCALL32))
 204
 205 void __cpuinit syscall32_cpu_init(void)
 206 {
 207         static const struct callback_register __cpuinitconst cstar = {
 208                 .type = CALLBACKTYPE_syscall32,
 209                 .address = (unsigned long)ia32_cstar_target
 210         };
 211         static const struct callback_register __cpuinitconst sysenter = {
 212                 .type = CALLBACKTYPE_sysenter,
 213                 .address = (unsigned long)ia32_sysenter_target
 214         };
 215
 216         if (HYPERVISOR_callback_op(CALLBACKOP_register, &sysenter) < 0)
 217                 setup_clear_cpu_cap(X86_FEATURE_SYSENTER32);
 218         if (HYPERVISOR_callback_op(CALLBACKOP_register, &cstar) < 0)
 219                 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 220 }
 221
 222 #define compat_uses_vma         1
 223
 224 static inline void map_compat_vdso(int map)
 225 {
 226 }
 227
 228 #else  /* CONFIG_X86_32 */
 229
 230 #define vdso32_sysenter()       (boot_cpu_has(X86_FEATURE_SEP))
 231 #define vdso32_syscall()        (boot_cpu_has(X86_FEATURE_SYSCALL32))
 232
 233 extern asmlinkage void ia32pv_cstar_target(void);
 234 static const struct callback_register __cpuinitconst cstar = {
 235         .type = CALLBACKTYPE_syscall32,
 236         .address = { __KERNEL_CS, (unsigned long)ia32pv_cstar_target },
 237 };
 238
 239 void __cpuinit enable_sep_cpu(void)
 240 {
 241         extern asmlinkage void ia32pv_sysenter_target(void);
 242         static struct callback_register __cpuinitdata sysenter = {
 243                 .type = CALLBACKTYPE_sysenter,
 244                 .address = { __KERNEL_CS, (unsigned long)ia32pv_sysenter_target },
 245         };
 246
 247         if (vdso32_syscall()) {
 248                 if (HYPERVISOR_callback_op(CALLBACKOP_register, &cstar) != 0)
 249                         BUG();
 250                 return;
 251         }
 252
 253         if (!vdso32_sysenter())
 254                 return;
 255
 256         if (xen_feature(XENFEAT_supervisor_mode_kernel))
 257                 sysenter.address.eip = (unsigned long)ia32_sysenter_target;
 258
 259         switch (HYPERVISOR_callback_op(CALLBACKOP_register, &sysenter)) {
 260         case 0:
 261                 break;
 262 #if CONFIG_XEN_COMPAT < 0x030200
 263         case -ENOSYS:
 264                 sysenter.type = CALLBACKTYPE_sysenter_deprecated;
 265                 if (HYPERVISOR_callback_op(CALLBACKOP_register, &sysenter) == 0)
 266                         break;
 267 #endif
 268         default:
 269                 setup_clear_cpu_cap(X86_FEATURE_SEP);
 270                 break;
 271         }
 272 }
 273
 274 static struct vm_area_struct gate_vma;
 275
 276 static int __init gate_vma_init(void)
 277 {
 278         gate_vma.vm_mm = NULL;
 279         gate_vma.vm_start = FIXADDR_USER_START;
 280         gate_vma.vm_end = FIXADDR_USER_END;
 281         gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
 282         gate_vma.vm_page_prot = __P101;
 283         /*
 284          * Make sure the vDSO gets into every core dump.
 285          * Dumping its contents makes post-mortem fully interpretable later
 286          * without matching up the same kernel and hardware config to see
 287          * what PC values meant.
 288          */
 289         gate_vma.vm_flags |= VM_ALWAYSDUMP;
 290         return 0;
 291 }
 292
 293 #define compat_uses_vma         0
 294
 295 static void map_compat_vdso(int map)
 296 {
 297         static int vdso_mapped;
 298
 299         if (map == vdso_mapped)
 300                 return;
 301
 302         vdso_mapped = map;
 303
 304         __set_fixmap(FIX_VDSO, page_to_pfn(vdso32_pages[0]) << PAGE_SHIFT,
 305                      map ? PAGE_READONLY_EXEC : PAGE_NONE);
 306
 307         /* flush stray tlbs */
 308         flush_tlb_all();
 309 }
 310
 311 #endif  /* CONFIG_X86_64 */
 312
 313 int __init sysenter_setup(void)
 314 {
 315         void *syscall_page = (void *)get_zeroed_page(GFP_ATOMIC);
 316         const void *vsyscall;
 317         size_t vsyscall_len;
 318
 319         vdso32_pages[0] = virt_to_page(syscall_page);
 320
 321 #ifdef CONFIG_X86_32
 322         gate_vma_init();
 323
 324         if (boot_cpu_has(X86_FEATURE_SYSCALL)) {
 325                 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD
 326                     && HYPERVISOR_callback_op(CALLBACKOP_register, &cstar) == 0)
 327                         setup_force_cpu_cap(X86_FEATURE_SYSCALL32);
 328                 else {
 329                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL);
 330                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
 331                 }
 332         }
 333 #endif
 334         if (vdso32_syscall()) {
 335                 vsyscall = &vdso32_syscall_start;
 336                 vsyscall_len = &vdso32_syscall_end - &vdso32_syscall_start;
 337         } else if (vdso32_sysenter()){
 338                 vsyscall = &vdso32_sysenter_start;
 339                 vsyscall_len = &vdso32_sysenter_end - &vdso32_sysenter_start;
 340         } else {
 341                 vsyscall = &vdso32_int80_start;
 342                 vsyscall_len = &vdso32_int80_end - &vdso32_int80_start;
 343         }
 344
 345         memcpy(syscall_page, vsyscall, vsyscall_len);
 346         relocate_vdso(syscall_page);
 347
 348         return 0;
 349 }
 350
 351 /* Setup a VMA at program startup for the vsyscall page */
 352 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 353 {
 354         struct mm_struct *mm = current->mm;
 355         unsigned long addr;
 356         int ret = 0;
 357         bool compat;
 358
 359         if (vdso_enabled == VDSO_DISABLED)
 360                 return 0;
 361
 362         down_write(&mm->mmap_sem);
 363
 364         /* Test compat mode once here, in case someone
 365            changes it via sysctl */
 366         compat = (vdso_enabled == VDSO_COMPAT);
 367
 368         map_compat_vdso(compat);
 369
 370         if (compat)
 371                 addr = VDSO_HIGH_BASE;
 372         else {
 373                 addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 374                 if (IS_ERR_VALUE(addr)) {
 375                         ret = addr;
 376                         goto up_fail;
 377                 }
 378         }
 379
 380         current->mm->context.vdso = (void *)addr;
 381
 382         if (compat_uses_vma || !compat) {
 383                 /*
 384                  * MAYWRITE to allow gdb to COW and set breakpoints
 385                  *
 386                  * Make sure the vDSO gets into every core dump.
 387                  * Dumping its contents makes post-mortem fully
 388                  * interpretable later without matching up the same
 389                  * kernel and hardware config to see what PC values
 390                  * meant.
 391                  */
 392                 ret = install_special_mapping(mm, addr, PAGE_SIZE,
 393                                               VM_READ|VM_EXEC|
 394                                               VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
 395                                               VM_ALWAYSDUMP,
 396                                               vdso32_pages);
 397
 398                 if (ret)
 399                         goto up_fail;
 400         }
 401
 402         current_thread_info()->sysenter_return =
 403                 VDSO32_SYMBOL(addr, SYSENTER_RETURN);
 404
 405   up_fail:
 406         if (ret)
 407                 current->mm->context.vdso = NULL;
 408
 409         up_write(&mm->mmap_sem);
 410
 411         return ret;
 412 }
 413
 414 #ifdef CONFIG_X86_64
 415
 416 /*
 417  * This must be done early in case we have an initrd containing 32-bit
 418  * binaries (e.g., hotplug). This could be pushed upstream.
 419  */
 420 core_initcall(sysenter_setup);
 421
 422 #ifdef CONFIG_SYSCTL
 423 /* Register vsyscall32 into the ABI table */
 424 #include <linux/sysctl.h>
 425
 426 static ctl_table abi_table2[] = {
 427         {
 428                 .procname       = "vsyscall32",
 429                 .data           = &sysctl_vsyscall32,
 430                 .maxlen         = sizeof(int),
 431                 .mode           = 0644,
 432                 .proc_handler   = proc_dointvec
 433         },
 434         {}
 435 };
 436
 437 static ctl_table abi_root_table2[] = {
 438         {
 439                 .procname = "abi",
 440                 .mode = 0555,
 441                 .child = abi_table2
 442         },
 443         {}
 444 };
 445
 446 static __init int ia32_binfmt_init(void)
 447 {
 448         register_sysctl_table(abi_root_table2);
 449         return 0;
 450 }
 451 __initcall(ia32_binfmt_init);
 452 #endif
 453
 454 #else  /* CONFIG_X86_32 */
 455
 456 const char *arch_vma_name(struct vm_area_struct *vma)
 457 {
 458         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 459                 return "[vdso]";
 460         return NULL;
 461 }
 462
 463 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 464 {
 465         struct mm_struct *mm = tsk->mm;
 466
 467         /* Check to see if this task was created in compat vdso mode */
 468         if (mm && mm->context.vdso == (void *)VDSO_HIGH_BASE)
 469                 return &gate_vma;
 470         return NULL;
 471 }
 472
 473 int in_gate_area(struct task_struct *task, unsigned long addr)
 474 {
 475         const struct vm_area_struct *vma = get_gate_vma(task);
 476
 477         return vma && addr >= vma->vm_start && addr < vma->vm_end;
 478 }
 479
 480 int in_gate_area_no_task(unsigned long addr)
 481 {
 482         return 0;
 483 }
 484
 485 #endif  /* CONFIG_X86_64 */