arch/x86/entry/vdso/vma.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright 2007 Andi Kleen, SUSE Labs.
   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 <linux/time_namespace.h>
  18
  19 #include <asm/pvclock.h>
  20 #include <asm/vgtod.h>
  21 #include <asm/proto.h>
  22 #include <asm/vdso.h>
  23 #include <asm/vvar.h>
  24 #include <asm/tlb.h>
  25 #include <asm/page.h>
  26 #include <asm/desc.h>
  27 #include <asm/cpufeature.h>
  28 #include <clocksource/hyperv_timer.h>
  29
  30 #undef _ASM_X86_VVAR_H
  31 #define EMIT_VVAR(name, offset) \
  32         const size_t name ## _offset = offset;
  33 #include <asm/vvar.h>
  34
  35 struct vdso_data *arch_get_vdso_data(void *vvar_page)
  36 {
  37         return (struct vdso_data *)(vvar_page + _vdso_data_offset);
  38 }
  39 #undef EMIT_VVAR
  40
  41 #if defined(CONFIG_X86_64)
  42 unsigned int __read_mostly vdso64_enabled = 1;
  43 #endif
  44
  45 void __init init_vdso_image(const struct vdso_image *image)
  46 {
  47         BUG_ON(image->size % PAGE_SIZE != 0);
  48
  49         apply_alternatives((struct alt_instr *)(image->data + image->alt),
  50                            (struct alt_instr *)(image->data + image->alt +
  51                                                 image->alt_len));
  52 }
  53
  54 static const struct vm_special_mapping vvar_mapping;
  55 struct linux_binprm;
  56
  57 static vm_fault_t vdso_fault(const struct vm_special_mapping *sm,
  58                       struct vm_area_struct *vma, struct vm_fault *vmf)
  59 {
  60         const struct vdso_image *image = vma->vm_mm->context.vdso_image;
  61
  62         if (!image || (vmf->pgoff << PAGE_SHIFT) >= image->size)
  63                 return VM_FAULT_SIGBUS;
  64
  65         vmf->page = virt_to_page(image->data + (vmf->pgoff << PAGE_SHIFT));
  66         get_page(vmf->page);
  67         return 0;
  68 }
  69
  70 static void vdso_fix_landing(const struct vdso_image *image,
  71                 struct vm_area_struct *new_vma)
  72 {
  73 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
  74         if (in_ia32_syscall() && image == &vdso_image_32) {
  75                 struct pt_regs *regs = current_pt_regs();
  76                 unsigned long vdso_land = image->sym_int80_landing_pad;
  77                 unsigned long old_land_addr = vdso_land +
  78                         (unsigned long)current->mm->context.vdso;
  79
  80                 /* Fixing userspace landing - look at do_fast_syscall_32 */
  81                 if (regs->ip == old_land_addr)
  82                         regs->ip = new_vma->vm_start + vdso_land;
  83         }
  84 #endif
  85 }
  86
  87 static int vdso_mremap(const struct vm_special_mapping *sm,
  88                 struct vm_area_struct *new_vma)
  89 {
  90         unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
  91         const struct vdso_image *image = current->mm->context.vdso_image;
  92
  93         if (image->size != new_size)
  94                 return -EINVAL;
  95
  96         vdso_fix_landing(image, new_vma);
  97         current->mm->context.vdso = (void __user *)new_vma->vm_start;
  98
  99         return 0;
 100 }
 101
 102 static int vvar_mremap(const struct vm_special_mapping *sm,
 103                 struct vm_area_struct *new_vma)
 104 {
 105         const struct vdso_image *image = new_vma->vm_mm->context.vdso_image;
 106         unsigned long new_size = new_vma->vm_end - new_vma->vm_start;
 107
 108         if (new_size != -image->sym_vvar_start)
 109                 return -EINVAL;
 110
 111         return 0;
 112 }
 113
 114 #ifdef CONFIG_TIME_NS
 115 static struct page *find_timens_vvar_page(struct vm_area_struct *vma)
 116 {
 117         if (likely(vma->vm_mm == current->mm))
 118                 return current->nsproxy->time_ns->vvar_page;
 119
 120         /*
 121          * VM_PFNMAP | VM_IO protect .fault() handler from being called
 122          * through interfaces like /proc/$pid/mem or
 123          * process_vm_{readv,writev}() as long as there's no .access()
 124          * in special_mapping_vmops().
 125          * For more details check_vma_flags() and __access_remote_vm()
 126          */
 127
 128         WARN(1, "vvar_page accessed remotely");
 129
 130         return NULL;
 131 }
 132
 133 /*
 134  * The vvar page layout depends on whether a task belongs to the root or
 135  * non-root time namespace. Whenever a task changes its namespace, the VVAR
 136  * page tables are cleared and then they will re-faulted with a
 137  * corresponding layout.
 138  * See also the comment near timens_setup_vdso_data() for details.
 139  */
 140 int vdso_join_timens(struct task_struct *task, struct time_namespace *ns)
 141 {
 142         struct mm_struct *mm = task->mm;
 143         struct vm_area_struct *vma;
 144
 145         if (down_write_killable(&mm->mmap_sem))
 146                 return -EINTR;
 147
 148         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 149                 unsigned long size = vma->vm_end - vma->vm_start;
 150
 151                 if (vma_is_special_mapping(vma, &vvar_mapping))
 152                         zap_page_range(vma, vma->vm_start, size);
 153         }
 154
 155         up_write(&mm->mmap_sem);
 156         return 0;
 157 }
 158 #else
 159 static inline struct page *find_timens_vvar_page(struct vm_area_struct *vma)
 160 {
 161         return NULL;
 162 }
 163 #endif
 164
 165 static vm_fault_t vvar_fault(const struct vm_special_mapping *sm,
 166                       struct vm_area_struct *vma, struct vm_fault *vmf)
 167 {
 168         const struct vdso_image *image = vma->vm_mm->context.vdso_image;
 169         unsigned long pfn;
 170         long sym_offset;
 171
 172         if (!image)
 173                 return VM_FAULT_SIGBUS;
 174
 175         sym_offset = (long)(vmf->pgoff << PAGE_SHIFT) +
 176                 image->sym_vvar_start;
 177
 178         /*
 179          * Sanity check: a symbol offset of zero means that the page
 180          * does not exist for this vdso image, not that the page is at
 181          * offset zero relative to the text mapping.  This should be
 182          * impossible here, because sym_offset should only be zero for
 183          * the page past the end of the vvar mapping.
 184          */
 185         if (sym_offset == 0)
 186                 return VM_FAULT_SIGBUS;
 187
 188         if (sym_offset == image->sym_vvar_page) {
 189                 struct page *timens_page = find_timens_vvar_page(vma);
 190
 191                 pfn = __pa_symbol(&__vvar_page) >> PAGE_SHIFT;
 192
 193                 /*
 194                  * If a task belongs to a time namespace then a namespace
 195                  * specific VVAR is mapped with the sym_vvar_page offset and
 196                  * the real VVAR page is mapped with the sym_timens_page
 197                  * offset.
 198                  * See also the comment near timens_setup_vdso_data().
 199                  */
 200                 if (timens_page) {
 201                         unsigned long addr;
 202                         vm_fault_t err;
 203
 204                         /*
 205                          * Optimization: inside time namespace pre-fault
 206                          * VVAR page too. As on timens page there are only
 207                          * offsets for clocks on VVAR, it'll be faulted
 208                          * shortly by VDSO code.
 209                          */
 210                         addr = vmf->address + (image->sym_timens_page - sym_offset);
 211                         err = vmf_insert_pfn(vma, addr, pfn);
 212                         if (unlikely(err & VM_FAULT_ERROR))
 213                                 return err;
 214
 215                         pfn = page_to_pfn(timens_page);
 216                 }
 217
 218                 return vmf_insert_pfn(vma, vmf->address, pfn);
 219         } else if (sym_offset == image->sym_pvclock_page) {
 220                 struct pvclock_vsyscall_time_info *pvti =
 221                         pvclock_get_pvti_cpu0_va();
 222                 if (pvti && vclock_was_used(VCLOCK_PVCLOCK)) {
 223                         return vmf_insert_pfn_prot(vma, vmf->address,
 224                                         __pa(pvti) >> PAGE_SHIFT,
 225                                         pgprot_decrypted(vma->vm_page_prot));
 226                 }
 227         } else if (sym_offset == image->sym_hvclock_page) {
 228                 struct ms_hyperv_tsc_page *tsc_pg = hv_get_tsc_page();
 229
 230                 if (tsc_pg && vclock_was_used(VCLOCK_HVCLOCK))
 231                         return vmf_insert_pfn(vma, vmf->address,
 232                                         virt_to_phys(tsc_pg) >> PAGE_SHIFT);
 233         } else if (sym_offset == image->sym_timens_page) {
 234                 struct page *timens_page = find_timens_vvar_page(vma);
 235
 236                 if (!timens_page)
 237                         return VM_FAULT_SIGBUS;
 238
 239                 pfn = __pa_symbol(&__vvar_page) >> PAGE_SHIFT;
 240                 return vmf_insert_pfn(vma, vmf->address, pfn);
 241         }
 242
 243         return VM_FAULT_SIGBUS;
 244 }
 245
 246 static const struct vm_special_mapping vdso_mapping = {
 247         .name = "[vdso]",
 248         .fault = vdso_fault,
 249         .mremap = vdso_mremap,
 250 };
 251 static const struct vm_special_mapping vvar_mapping = {
 252         .name = "[vvar]",
 253         .fault = vvar_fault,
 254         .mremap = vvar_mremap,
 255 };
 256
 257 /*
 258  * Add vdso and vvar mappings to current process.
 259  * @image          - blob to map
 260  * @addr           - request a specific address (zero to map at free addr)
 261  */
 262 static int map_vdso(const struct vdso_image *image, unsigned long addr)
 263 {
 264         struct mm_struct *mm = current->mm;
 265         struct vm_area_struct *vma;
 266         unsigned long text_start;
 267         int ret = 0;
 268
 269         if (down_write_killable(&mm->mmap_sem))
 270                 return -EINTR;
 271
 272         addr = get_unmapped_area(NULL, addr,
 273                                  image->size - image->sym_vvar_start, 0, 0);
 274         if (IS_ERR_VALUE(addr)) {
 275                 ret = addr;
 276                 goto up_fail;
 277         }
 278
 279         text_start = addr - image->sym_vvar_start;
 280
 281         /*
 282          * MAYWRITE to allow gdb to COW and set breakpoints
 283          */
 284         vma = _install_special_mapping(mm,
 285                                        text_start,
 286                                        image->size,
 287                                        VM_READ|VM_EXEC|
 288                                        VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
 289                                        &vdso_mapping);
 290
 291         if (IS_ERR(vma)) {
 292                 ret = PTR_ERR(vma);
 293                 goto up_fail;
 294         }
 295
 296         vma = _install_special_mapping(mm,
 297                                        addr,
 298                                        -image->sym_vvar_start,
 299                                        VM_READ|VM_MAYREAD|VM_IO|VM_DONTDUMP|
 300                                        VM_PFNMAP,
 301                                        &vvar_mapping);
 302
 303         if (IS_ERR(vma)) {
 304                 ret = PTR_ERR(vma);
 305                 do_munmap(mm, text_start, image->size, NULL);
 306         } else {
 307                 current->mm->context.vdso = (void __user *)text_start;
 308                 current->mm->context.vdso_image = image;
 309         }
 310
 311 up_fail:
 312         up_write(&mm->mmap_sem);
 313         return ret;
 314 }
 315
 316 #ifdef CONFIG_X86_64
 317 /*
 318  * Put the vdso above the (randomized) stack with another randomized
 319  * offset.  This way there is no hole in the middle of address space.
 320  * To save memory make sure it is still in the same PTE as the stack
 321  * top.  This doesn't give that many random bits.
 322  *
 323  * Note that this algorithm is imperfect: the distribution of the vdso
 324  * start address within a PMD is biased toward the end.
 325  *
 326  * Only used for the 64-bit and x32 vdsos.
 327  */
 328 static unsigned long vdso_addr(unsigned long start, unsigned len)
 329 {
 330         unsigned long addr, end;
 331         unsigned offset;
 332
 333         /*
 334          * Round up the start address.  It can start out unaligned as a result
 335          * of stack start randomization.
 336          */
 337         start = PAGE_ALIGN(start);
 338
 339         /* Round the lowest possible end address up to a PMD boundary. */
 340         end = (start + len + PMD_SIZE - 1) & PMD_MASK;
 341         if (end >= TASK_SIZE_MAX)
 342                 end = TASK_SIZE_MAX;
 343         end -= len;
 344
 345         if (end > start) {
 346                 offset = get_random_int() % (((end - start) >> PAGE_SHIFT) + 1);
 347                 addr = start + (offset << PAGE_SHIFT);
 348         } else {
 349                 addr = start;
 350         }
 351
 352         /*
 353          * Forcibly align the final address in case we have a hardware
 354          * issue that requires alignment for performance reasons.
 355          */
 356         addr = align_vdso_addr(addr);
 357
 358         return addr;
 359 }
 360
 361 static int map_vdso_randomized(const struct vdso_image *image)
 362 {
 363         unsigned long addr = vdso_addr(current->mm->start_stack, image->size-image->sym_vvar_start);
 364
 365         return map_vdso(image, addr);
 366 }
 367 #endif
 368
 369 int map_vdso_once(const struct vdso_image *image, unsigned long addr)
 370 {
 371         struct mm_struct *mm = current->mm;
 372         struct vm_area_struct *vma;
 373
 374         down_write(&mm->mmap_sem);
 375         /*
 376          * Check if we have already mapped vdso blob - fail to prevent
 377          * abusing from userspace install_speciall_mapping, which may
 378          * not do accounting and rlimit right.
 379          * We could search vma near context.vdso, but it's a slowpath,
 380          * so let's explicitly check all VMAs to be completely sure.
 381          */
 382         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 383                 if (vma_is_special_mapping(vma, &vdso_mapping) ||
 384                                 vma_is_special_mapping(vma, &vvar_mapping)) {
 385                         up_write(&mm->mmap_sem);
 386                         return -EEXIST;
 387                 }
 388         }
 389         up_write(&mm->mmap_sem);
 390
 391         return map_vdso(image, addr);
 392 }
 393
 394 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 395 static int load_vdso32(void)
 396 {
 397         if (vdso32_enabled != 1)  /* Other values all mean "disabled" */
 398                 return 0;
 399
 400         return map_vdso(&vdso_image_32, 0);
 401 }
 402 #endif
 403
 404 #ifdef CONFIG_X86_64
 405 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 406 {
 407         if (!vdso64_enabled)
 408                 return 0;
 409
 410         return map_vdso_randomized(&vdso_image_64);
 411 }
 412
 413 #ifdef CONFIG_COMPAT
 414 int compat_arch_setup_additional_pages(struct linux_binprm *bprm,
 415                                        int uses_interp)
 416 {
 417 #ifdef CONFIG_X86_X32_ABI
 418         if (test_thread_flag(TIF_X32)) {
 419                 if (!vdso64_enabled)
 420                         return 0;
 421                 return map_vdso_randomized(&vdso_image_x32);
 422         }
 423 #endif
 424 #ifdef CONFIG_IA32_EMULATION
 425         return load_vdso32();
 426 #else
 427         return 0;
 428 #endif
 429 }
 430 #endif
 431 #else
 432 int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 433 {
 434         return load_vdso32();
 435 }
 436 #endif
 437
 438 #ifdef CONFIG_X86_64
 439 static __init int vdso_setup(char *s)
 440 {
 441         vdso64_enabled = simple_strtoul(s, NULL, 0);
 442         return 0;
 443 }
 444 __setup("vdso=", vdso_setup);
 445
 446 static int __init init_vdso(void)
 447 {
 448         init_vdso_image(&vdso_image_64);
 449
 450 #ifdef CONFIG_X86_X32_ABI
 451         init_vdso_image(&vdso_image_x32);
 452 #endif
 453
 454         return 0;
 455 }
 456 subsys_initcall(init_vdso);
 457 #endif /* CONFIG_X86_64 */