fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/vmacache.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57 #include <linux/oom.h>
  58 #include <linux/compat.h>
  59
  60 #include <asm/uaccess.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/tlb.h>
  63
  64 #include <trace/events/task.h>
  65 #include "internal.h"
  66
  67 #include <trace/events/sched.h>
  68
  69 int suid_dumpable = 0;
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         BUG_ON(!fmt);
  77         if (WARN_ON(!fmt->load_binary))
  78                 return;
  79         write_lock(&binfmt_lock);
  80         insert ? list_add(&fmt->lh, &formats) :
  81                  list_add_tail(&fmt->lh, &formats);
  82         write_unlock(&binfmt_lock);
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 #ifdef CONFIG_USELIB
 102 /*
 103  * Note that a shared library must be both readable and executable due to
 104  * security reasons.
 105  *
 106  * Also note that we take the address to load from from the file itself.
 107  */
 108 SYSCALL_DEFINE1(uselib, const char __user *, library)
 109 {
 110         struct linux_binfmt *fmt;
 111         struct file *file;
 112         struct filename *tmp = getname(library);
 113         int error = PTR_ERR(tmp);
 114         static const struct open_flags uselib_flags = {
 115                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 116                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 117                 .intent = LOOKUP_OPEN,
 118                 .lookup_flags = LOOKUP_FOLLOW,
 119         };
 120
 121         if (IS_ERR(tmp))
 122                 goto out;
 123
 124         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 125         putname(tmp);
 126         error = PTR_ERR(file);
 127         if (IS_ERR(file))
 128                 goto out;
 129
 130         error = -EINVAL;
 131         if (!S_ISREG(file_inode(file)->i_mode))
 132                 goto exit;
 133
 134         error = -EACCES;
 135         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 136                 goto exit;
 137
 138         fsnotify_open(file);
 139
 140         error = -ENOEXEC;
 141
 142         read_lock(&binfmt_lock);
 143         list_for_each_entry(fmt, &formats, lh) {
 144                 if (!fmt->load_shlib)
 145                         continue;
 146                 if (!try_module_get(fmt->module))
 147                         continue;
 148                 read_unlock(&binfmt_lock);
 149                 error = fmt->load_shlib(file);
 150                 read_lock(&binfmt_lock);
 151                 put_binfmt(fmt);
 152                 if (error != -ENOEXEC)
 153                         break;
 154         }
 155         read_unlock(&binfmt_lock);
 156 exit:
 157         fput(file);
 158 out:
 159         return error;
 160 }
 161 #endif /* #ifdef CONFIG_USELIB */
 162
 163 #ifdef CONFIG_MMU
 164 /*
 165  * The nascent bprm->mm is not visible until exec_mmap() but it can
 166  * use a lot of memory, account these pages in current->mm temporary
 167  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 168  * change the counter back via acct_arg_size(0).
 169  */
 170 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 171 {
 172         struct mm_struct *mm = current->mm;
 173         long diff = (long)(pages - bprm->vma_pages);
 174
 175         if (!mm || !diff)
 176                 return;
 177
 178         bprm->vma_pages = pages;
 179         add_mm_counter(mm, MM_ANONPAGES, diff);
 180 }
 181
 182 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 183                 int write)
 184 {
 185         struct page *page;
 186         int ret;
 187
 188 #ifdef CONFIG_STACK_GROWSUP
 189         if (write) {
 190                 ret = expand_downwards(bprm->vma, pos);
 191                 if (ret < 0)
 192                         return NULL;
 193         }
 194 #endif
 195         ret = get_user_pages(current, bprm->mm, pos,
 196                         1, write, 1, &page, NULL);
 197         if (ret <= 0)
 198                 return NULL;
 199
 200         if (write) {
 201                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 202                 struct rlimit *rlim;
 203
 204                 acct_arg_size(bprm, size / PAGE_SIZE);
 205
 206                 /*
 207                  * We've historically supported up to 32 pages (ARG_MAX)
 208                  * of argument strings even with small stacks
 209                  */
 210                 if (size <= ARG_MAX)
 211                         return page;
 212
 213                 /*
 214                  * Limit to 1/4-th the stack size for the argv+env strings.
 215                  * This ensures that:
 216                  *  - the remaining binfmt code will not run out of stack space,
 217                  *  - the program will have a reasonable amount of stack left
 218                  *    to work from.
 219                  */
 220                 rlim = current->signal->rlim;
 221                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 222                         put_page(page);
 223                         return NULL;
 224                 }
 225         }
 226
 227         return page;
 228 }
 229
 230 static void put_arg_page(struct page *page)
 231 {
 232         put_page(page);
 233 }
 234
 235 static void free_arg_page(struct linux_binprm *bprm, int i)
 236 {
 237 }
 238
 239 static void free_arg_pages(struct linux_binprm *bprm)
 240 {
 241 }
 242
 243 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 244                 struct page *page)
 245 {
 246         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 247 }
 248
 249 static int __bprm_mm_init(struct linux_binprm *bprm)
 250 {
 251         int err;
 252         struct vm_area_struct *vma = NULL;
 253         struct mm_struct *mm = bprm->mm;
 254
 255         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 256         if (!vma)
 257                 return -ENOMEM;
 258
 259         down_write(&mm->mmap_sem);
 260         vma->vm_mm = mm;
 261
 262         /*
 263          * Place the stack at the largest stack address the architecture
 264          * supports. Later, we'll move this to an appropriate place. We don't
 265          * use STACK_TOP because that can depend on attributes which aren't
 266          * configured yet.
 267          */
 268         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 269         vma->vm_end = STACK_TOP_MAX;
 270         vma->vm_start = vma->vm_end - PAGE_SIZE;
 271         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 272         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 273         INIT_LIST_HEAD(&vma->anon_vma_chain);
 274
 275         err = insert_vm_struct(mm, vma);
 276         if (err)
 277                 goto err;
 278
 279         mm->stack_vm = mm->total_vm = 1;
 280         up_write(&mm->mmap_sem);
 281         bprm->p = vma->vm_end - sizeof(void *);
 282         return 0;
 283 err:
 284         up_write(&mm->mmap_sem);
 285         bprm->vma = NULL;
 286         kmem_cache_free(vm_area_cachep, vma);
 287         return err;
 288 }
 289
 290 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 291 {
 292         return len <= MAX_ARG_STRLEN;
 293 }
 294
 295 #else
 296
 297 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 298 {
 299 }
 300
 301 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 302                 int write)
 303 {
 304         struct page *page;
 305
 306         page = bprm->page[pos / PAGE_SIZE];
 307         if (!page && write) {
 308                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 309                 if (!page)
 310                         return NULL;
 311                 bprm->page[pos / PAGE_SIZE] = page;
 312         }
 313
 314         return page;
 315 }
 316
 317 static void put_arg_page(struct page *page)
 318 {
 319 }
 320
 321 static void free_arg_page(struct linux_binprm *bprm, int i)
 322 {
 323         if (bprm->page[i]) {
 324                 __free_page(bprm->page[i]);
 325                 bprm->page[i] = NULL;
 326         }
 327 }
 328
 329 static void free_arg_pages(struct linux_binprm *bprm)
 330 {
 331         int i;
 332
 333         for (i = 0; i < MAX_ARG_PAGES; i++)
 334                 free_arg_page(bprm, i);
 335 }
 336
 337 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 338                 struct page *page)
 339 {
 340 }
 341
 342 static int __bprm_mm_init(struct linux_binprm *bprm)
 343 {
 344         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 345         return 0;
 346 }
 347
 348 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 349 {
 350         return len <= bprm->p;
 351 }
 352
 353 #endif /* CONFIG_MMU */
 354
 355 /*
 356  * Create a new mm_struct and populate it with a temporary stack
 357  * vm_area_struct.  We don't have enough context at this point to set the stack
 358  * flags, permissions, and offset, so we use temporary values.  We'll update
 359  * them later in setup_arg_pages().
 360  */
 361 static int bprm_mm_init(struct linux_binprm *bprm)
 362 {
 363         int err;
 364         struct mm_struct *mm = NULL;
 365
 366         bprm->mm = mm = mm_alloc();
 367         err = -ENOMEM;
 368         if (!mm)
 369                 goto err;
 370
 371         err = __bprm_mm_init(bprm);
 372         if (err)
 373                 goto err;
 374
 375         return 0;
 376
 377 err:
 378         if (mm) {
 379                 bprm->mm = NULL;
 380                 mmdrop(mm);
 381         }
 382
 383         return err;
 384 }
 385
 386 struct user_arg_ptr {
 387 #ifdef CONFIG_COMPAT
 388         bool is_compat;
 389 #endif
 390         union {
 391                 const char __user *const __user *native;
 392 #ifdef CONFIG_COMPAT
 393                 const compat_uptr_t __user *compat;
 394 #endif
 395         } ptr;
 396 };
 397
 398 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 399 {
 400         const char __user *native;
 401
 402 #ifdef CONFIG_COMPAT
 403         if (unlikely(argv.is_compat)) {
 404                 compat_uptr_t compat;
 405
 406                 if (get_user(compat, argv.ptr.compat + nr))
 407                         return ERR_PTR(-EFAULT);
 408
 409                 return compat_ptr(compat);
 410         }
 411 #endif
 412
 413         if (get_user(native, argv.ptr.native + nr))
 414                 return ERR_PTR(-EFAULT);
 415
 416         return native;
 417 }
 418
 419 /*
 420  * count() counts the number of strings in array ARGV.
 421  */
 422 static int count(struct user_arg_ptr argv, int max)
 423 {
 424         int i = 0;
 425
 426         if (argv.ptr.native != NULL) {
 427                 for (;;) {
 428                         const char __user *p = get_user_arg_ptr(argv, i);
 429
 430                         if (!p)
 431                                 break;
 432
 433                         if (IS_ERR(p))
 434                                 return -EFAULT;
 435
 436                         if (i >= max)
 437                                 return -E2BIG;
 438                         ++i;
 439
 440                         if (fatal_signal_pending(current))
 441                                 return -ERESTARTNOHAND;
 442                         cond_resched();
 443                 }
 444         }
 445         return i;
 446 }
 447
 448 /*
 449  * 'copy_strings()' copies argument/environment strings from the old
 450  * processes's memory to the new process's stack.  The call to get_user_pages()
 451  * ensures the destination page is created and not swapped out.
 452  */
 453 static int copy_strings(int argc, struct user_arg_ptr argv,
 454                         struct linux_binprm *bprm)
 455 {
 456         struct page *kmapped_page = NULL;
 457         char *kaddr = NULL;
 458         unsigned long kpos = 0;
 459         int ret;
 460
 461         while (argc-- > 0) {
 462                 const char __user *str;
 463                 int len;
 464                 unsigned long pos;
 465
 466                 ret = -EFAULT;
 467                 str = get_user_arg_ptr(argv, argc);
 468                 if (IS_ERR(str))
 469                         goto out;
 470
 471                 len = strnlen_user(str, MAX_ARG_STRLEN);
 472                 if (!len)
 473                         goto out;
 474
 475                 ret = -E2BIG;
 476                 if (!valid_arg_len(bprm, len))
 477                         goto out;
 478
 479                 /* We're going to work our way backwords. */
 480                 pos = bprm->p;
 481                 str += len;
 482                 bprm->p -= len;
 483
 484                 while (len > 0) {
 485                         int offset, bytes_to_copy;
 486
 487                         if (fatal_signal_pending(current)) {
 488                                 ret = -ERESTARTNOHAND;
 489                                 goto out;
 490                         }
 491                         cond_resched();
 492
 493                         offset = pos % PAGE_SIZE;
 494                         if (offset == 0)
 495                                 offset = PAGE_SIZE;
 496
 497                         bytes_to_copy = offset;
 498                         if (bytes_to_copy > len)
 499                                 bytes_to_copy = len;
 500
 501                         offset -= bytes_to_copy;
 502                         pos -= bytes_to_copy;
 503                         str -= bytes_to_copy;
 504                         len -= bytes_to_copy;
 505
 506                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 507                                 struct page *page;
 508
 509                                 page = get_arg_page(bprm, pos, 1);
 510                                 if (!page) {
 511                                         ret = -E2BIG;
 512                                         goto out;
 513                                 }
 514
 515                                 if (kmapped_page) {
 516                                         flush_kernel_dcache_page(kmapped_page);
 517                                         kunmap(kmapped_page);
 518                                         put_arg_page(kmapped_page);
 519                                 }
 520                                 kmapped_page = page;
 521                                 kaddr = kmap(kmapped_page);
 522                                 kpos = pos & PAGE_MASK;
 523                                 flush_arg_page(bprm, kpos, kmapped_page);
 524                         }
 525                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 526                                 ret = -EFAULT;
 527                                 goto out;
 528                         }
 529                 }
 530         }
 531         ret = 0;
 532 out:
 533         if (kmapped_page) {
 534                 flush_kernel_dcache_page(kmapped_page);
 535                 kunmap(kmapped_page);
 536                 put_arg_page(kmapped_page);
 537         }
 538         return ret;
 539 }
 540
 541 /*
 542  * Like copy_strings, but get argv and its values from kernel memory.
 543  */
 544 int copy_strings_kernel(int argc, const char *const *__argv,
 545                         struct linux_binprm *bprm)
 546 {
 547         int r;
 548         mm_segment_t oldfs = get_fs();
 549         struct user_arg_ptr argv = {
 550                 .ptr.native = (const char __user *const  __user *)__argv,
 551         };
 552
 553         set_fs(KERNEL_DS);
 554         r = copy_strings(argc, argv, bprm);
 555         set_fs(oldfs);
 556
 557         return r;
 558 }
 559 EXPORT_SYMBOL(copy_strings_kernel);
 560
 561 #ifdef CONFIG_MMU
 562
 563 /*
 564  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 565  * the binfmt code determines where the new stack should reside, we shift it to
 566  * its final location.  The process proceeds as follows:
 567  *
 568  * 1) Use shift to calculate the new vma endpoints.
 569  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 570  *    arguments passed to subsequent functions are consistent.
 571  * 3) Move vma's page tables to the new range.
 572  * 4) Free up any cleared pgd range.
 573  * 5) Shrink the vma to cover only the new range.
 574  */
 575 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 576 {
 577         struct mm_struct *mm = vma->vm_mm;
 578         unsigned long old_start = vma->vm_start;
 579         unsigned long old_end = vma->vm_end;
 580         unsigned long length = old_end - old_start;
 581         unsigned long new_start = old_start - shift;
 582         unsigned long new_end = old_end - shift;
 583         struct mmu_gather tlb;
 584
 585         BUG_ON(new_start > new_end);
 586
 587         /*
 588          * ensure there are no vmas between where we want to go
 589          * and where we are
 590          */
 591         if (vma != find_vma(mm, new_start))
 592                 return -EFAULT;
 593
 594         /*
 595          * cover the whole range: [new_start, old_end)
 596          */
 597         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 598                 return -ENOMEM;
 599
 600         /*
 601          * move the page tables downwards, on failure we rely on
 602          * process cleanup to remove whatever mess we made.
 603          */
 604         if (length != move_page_tables(vma, old_start,
 605                                        vma, new_start, length, false))
 606                 return -ENOMEM;
 607
 608         lru_add_drain();
 609         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 610         if (new_end > old_start) {
 611                 /*
 612                  * when the old and new regions overlap clear from new_end.
 613                  */
 614                 free_pgd_range(&tlb, new_end, old_end, new_end,
 615                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 616         } else {
 617                 /*
 618                  * otherwise, clean from old_start; this is done to not touch
 619                  * the address space in [new_end, old_start) some architectures
 620                  * have constraints on va-space that make this illegal (IA64) -
 621                  * for the others its just a little faster.
 622                  */
 623                 free_pgd_range(&tlb, old_start, old_end, new_end,
 624                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 625         }
 626         tlb_finish_mmu(&tlb, old_start, old_end);
 627
 628         /*
 629          * Shrink the vma to just the new range.  Always succeeds.
 630          */
 631         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 632
 633         return 0;
 634 }
 635
 636 /*
 637  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 638  * the stack is optionally relocated, and some extra space is added.
 639  */
 640 int setup_arg_pages(struct linux_binprm *bprm,
 641                     unsigned long stack_top,
 642                     int executable_stack)
 643 {
 644         unsigned long ret;
 645         unsigned long stack_shift;
 646         struct mm_struct *mm = current->mm;
 647         struct vm_area_struct *vma = bprm->vma;
 648         struct vm_area_struct *prev = NULL;
 649         unsigned long vm_flags;
 650         unsigned long stack_base;
 651         unsigned long stack_size;
 652         unsigned long stack_expand;
 653         unsigned long rlim_stack;
 654
 655 #ifdef CONFIG_STACK_GROWSUP
 656         /* Limit stack size */
 657         stack_base = rlimit_max(RLIMIT_STACK);
 658         if (stack_base > STACK_SIZE_MAX)
 659                 stack_base = STACK_SIZE_MAX;
 660
 661         /* Make sure we didn't let the argument array grow too large. */
 662         if (vma->vm_end - vma->vm_start > stack_base)
 663                 return -ENOMEM;
 664
 665         stack_base = PAGE_ALIGN(stack_top - stack_base);
 666
 667         stack_shift = vma->vm_start - stack_base;
 668         mm->arg_start = bprm->p - stack_shift;
 669         bprm->p = vma->vm_end - stack_shift;
 670 #else
 671         stack_top = arch_align_stack(stack_top);
 672         stack_top = PAGE_ALIGN(stack_top);
 673
 674         if (unlikely(stack_top < mmap_min_addr) ||
 675             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 676                 return -ENOMEM;
 677
 678         stack_shift = vma->vm_end - stack_top;
 679
 680         bprm->p -= stack_shift;
 681         mm->arg_start = bprm->p;
 682 #endif
 683
 684         if (bprm->loader)
 685                 bprm->loader -= stack_shift;
 686         bprm->exec -= stack_shift;
 687
 688         down_write(&mm->mmap_sem);
 689         vm_flags = VM_STACK_FLAGS;
 690
 691         /*
 692          * Adjust stack execute permissions; explicitly enable for
 693          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 694          * (arch default) otherwise.
 695          */
 696         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 697                 vm_flags |= VM_EXEC;
 698         else if (executable_stack == EXSTACK_DISABLE_X)
 699                 vm_flags &= ~VM_EXEC;
 700         vm_flags |= mm->def_flags;
 701         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 702
 703         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 704                         vm_flags);
 705         if (ret)
 706                 goto out_unlock;
 707         BUG_ON(prev != vma);
 708
 709         /* Move stack pages down in memory. */
 710         if (stack_shift) {
 711                 ret = shift_arg_pages(vma, stack_shift);
 712                 if (ret)
 713                         goto out_unlock;
 714         }
 715
 716         /* mprotect_fixup is overkill to remove the temporary stack flags */
 717         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 718
 719         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 720         stack_size = vma->vm_end - vma->vm_start;
 721         /*
 722          * Align this down to a page boundary as expand_stack
 723          * will align it up.
 724          */
 725         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 726 #ifdef CONFIG_STACK_GROWSUP
 727         if (stack_size + stack_expand > rlim_stack)
 728                 stack_base = vma->vm_start + rlim_stack;
 729         else
 730                 stack_base = vma->vm_end + stack_expand;
 731 #else
 732         if (stack_size + stack_expand > rlim_stack)
 733                 stack_base = vma->vm_end - rlim_stack;
 734         else
 735                 stack_base = vma->vm_start - stack_expand;
 736 #endif
 737         current->mm->start_stack = bprm->p;
 738         ret = expand_stack(vma, stack_base);
 739         if (ret)
 740                 ret = -EFAULT;
 741
 742 out_unlock:
 743         up_write(&mm->mmap_sem);
 744         return ret;
 745 }
 746 EXPORT_SYMBOL(setup_arg_pages);
 747
 748 #endif /* CONFIG_MMU */
 749
 750 static struct file *do_open_exec(struct filename *name)
 751 {
 752         struct file *file;
 753         int err;
 754         static const struct open_flags open_exec_flags = {
 755                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 756                 .acc_mode = MAY_EXEC | MAY_OPEN,
 757                 .intent = LOOKUP_OPEN,
 758                 .lookup_flags = LOOKUP_FOLLOW,
 759         };
 760
 761         file = do_filp_open(AT_FDCWD, name, &open_exec_flags);
 762         if (IS_ERR(file))
 763                 goto out;
 764
 765         err = -EACCES;
 766         if (!S_ISREG(file_inode(file)->i_mode))
 767                 goto exit;
 768
 769         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 770                 goto exit;
 771
 772         fsnotify_open(file);
 773
 774         err = deny_write_access(file);
 775         if (err)
 776                 goto exit;
 777
 778 out:
 779         return file;
 780
 781 exit:
 782         fput(file);
 783         return ERR_PTR(err);
 784 }
 785
 786 struct file *open_exec(const char *name)
 787 {
 788         struct filename tmp = { .name = name };
 789         return do_open_exec(&tmp);
 790 }
 791 EXPORT_SYMBOL(open_exec);
 792
 793 int kernel_read(struct file *file, loff_t offset,
 794                 char *addr, unsigned long count)
 795 {
 796         mm_segment_t old_fs;
 797         loff_t pos = offset;
 798         int result;
 799
 800         old_fs = get_fs();
 801         set_fs(get_ds());
 802         /* The cast to a user pointer is valid due to the set_fs() */
 803         result = vfs_read(file, (void __user *)addr, count, &pos);
 804         set_fs(old_fs);
 805         return result;
 806 }
 807
 808 EXPORT_SYMBOL(kernel_read);
 809
 810 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 811 {
 812         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 813         if (res > 0)
 814                 flush_icache_range(addr, addr + len);
 815         return res;
 816 }
 817 EXPORT_SYMBOL(read_code);
 818
 819 static int exec_mmap(struct mm_struct *mm)
 820 {
 821         struct task_struct *tsk;
 822         struct mm_struct *old_mm, *active_mm;
 823
 824         /* Notify parent that we're no longer interested in the old VM */
 825         tsk = current;
 826         old_mm = current->mm;
 827         mm_release(tsk, old_mm);
 828
 829         if (old_mm) {
 830                 sync_mm_rss(old_mm);
 831                 /*
 832                  * Make sure that if there is a core dump in progress
 833                  * for the old mm, we get out and die instead of going
 834                  * through with the exec.  We must hold mmap_sem around
 835                  * checking core_state and changing tsk->mm.
 836                  */
 837                 down_read(&old_mm->mmap_sem);
 838                 if (unlikely(old_mm->core_state)) {
 839                         up_read(&old_mm->mmap_sem);
 840                         return -EINTR;
 841                 }
 842         }
 843         task_lock(tsk);
 844         active_mm = tsk->active_mm;
 845         tsk->mm = mm;
 846         tsk->active_mm = mm;
 847         activate_mm(active_mm, mm);
 848         tsk->mm->vmacache_seqnum = 0;
 849         vmacache_flush(tsk);
 850         task_unlock(tsk);
 851         if (old_mm) {
 852                 up_read(&old_mm->mmap_sem);
 853                 BUG_ON(active_mm != old_mm);
 854                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 855                 mm_update_next_owner(old_mm);
 856                 mmput(old_mm);
 857                 return 0;
 858         }
 859         mmdrop(active_mm);
 860         return 0;
 861 }
 862
 863 /*
 864  * This function makes sure the current process has its own signal table,
 865  * so that flush_signal_handlers can later reset the handlers without
 866  * disturbing other processes.  (Other processes might share the signal
 867  * table via the CLONE_SIGHAND option to clone().)
 868  */
 869 static int de_thread(struct task_struct *tsk)
 870 {
 871         struct signal_struct *sig = tsk->signal;
 872         struct sighand_struct *oldsighand = tsk->sighand;
 873         spinlock_t *lock = &oldsighand->siglock;
 874
 875         if (thread_group_empty(tsk))
 876                 goto no_thread_group;
 877
 878         /*
 879          * Kill all other threads in the thread group.
 880          */
 881         spin_lock_irq(lock);
 882         if (signal_group_exit(sig)) {
 883                 /*
 884                  * Another group action in progress, just
 885                  * return so that the signal is processed.
 886                  */
 887                 spin_unlock_irq(lock);
 888                 return -EAGAIN;
 889         }
 890
 891         sig->group_exit_task = tsk;
 892         sig->notify_count = zap_other_threads(tsk);
 893         if (!thread_group_leader(tsk))
 894                 sig->notify_count--;
 895
 896         while (sig->notify_count) {
 897                 __set_current_state(TASK_KILLABLE);
 898                 spin_unlock_irq(lock);
 899                 schedule();
 900                 if (unlikely(__fatal_signal_pending(tsk)))
 901                         goto killed;
 902                 spin_lock_irq(lock);
 903         }
 904         spin_unlock_irq(lock);
 905
 906         /*
 907          * At this point all other threads have exited, all we have to
 908          * do is to wait for the thread group leader to become inactive,
 909          * and to assume its PID:
 910          */
 911         if (!thread_group_leader(tsk)) {
 912                 struct task_struct *leader = tsk->group_leader;
 913
 914                 sig->notify_count = -1; /* for exit_notify() */
 915                 for (;;) {
 916                         threadgroup_change_begin(tsk);
 917                         write_lock_irq(&tasklist_lock);
 918                         if (likely(leader->exit_state))
 919                                 break;
 920                         __set_current_state(TASK_KILLABLE);
 921                         write_unlock_irq(&tasklist_lock);
 922                         threadgroup_change_end(tsk);
 923                         schedule();
 924                         if (unlikely(__fatal_signal_pending(tsk)))
 925                                 goto killed;
 926                 }
 927
 928                 /*
 929                  * The only record we have of the real-time age of a
 930                  * process, regardless of execs it's done, is start_time.
 931                  * All the past CPU time is accumulated in signal_struct
 932                  * from sister threads now dead.  But in this non-leader
 933                  * exec, nothing survives from the original leader thread,
 934                  * whose birth marks the true age of this process now.
 935                  * When we take on its identity by switching to its PID, we
 936                  * also take its birthdate (always earlier than our own).
 937                  */
 938                 tsk->start_time = leader->start_time;
 939                 tsk->real_start_time = leader->real_start_time;
 940
 941                 BUG_ON(!same_thread_group(leader, tsk));
 942                 BUG_ON(has_group_leader_pid(tsk));
 943                 /*
 944                  * An exec() starts a new thread group with the
 945                  * TGID of the previous thread group. Rehash the
 946                  * two threads with a switched PID, and release
 947                  * the former thread group leader:
 948                  */
 949
 950                 /* Become a process group leader with the old leader's pid.
 951                  * The old leader becomes a thread of the this thread group.
 952                  * Note: The old leader also uses this pid until release_task
 953                  *       is called.  Odd but simple and correct.
 954                  */
 955                 tsk->pid = leader->pid;
 956                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
 957                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 958                 transfer_pid(leader, tsk, PIDTYPE_SID);
 959
 960                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 961                 list_replace_init(&leader->sibling, &tsk->sibling);
 962
 963                 tsk->group_leader = tsk;
 964                 leader->group_leader = tsk;
 965
 966                 tsk->exit_signal = SIGCHLD;
 967                 leader->exit_signal = -1;
 968
 969                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 970                 leader->exit_state = EXIT_DEAD;
 971
 972                 /*
 973                  * We are going to release_task()->ptrace_unlink() silently,
 974                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 975                  * the tracer wont't block again waiting for this thread.
 976                  */
 977                 if (unlikely(leader->ptrace))
 978                         __wake_up_parent(leader, leader->parent);
 979                 write_unlock_irq(&tasklist_lock);
 980                 threadgroup_change_end(tsk);
 981
 982                 release_task(leader);
 983         }
 984
 985         sig->group_exit_task = NULL;
 986         sig->notify_count = 0;
 987
 988 no_thread_group:
 989         /* we have changed execution domain */
 990         tsk->exit_signal = SIGCHLD;
 991
 992         exit_itimers(sig);
 993         flush_itimer_signals();
 994
 995         if (atomic_read(&oldsighand->count) != 1) {
 996                 struct sighand_struct *newsighand;
 997                 /*
 998                  * This ->sighand is shared with the CLONE_SIGHAND
 999                  * but not CLONE_THREAD task, switch to the new one.
1000                  */
1001                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1002                 if (!newsighand)
1003                         return -ENOMEM;
1004
1005                 atomic_set(&newsighand->count, 1);
1006                 memcpy(newsighand->action, oldsighand->action,
1007                        sizeof(newsighand->action));
1008
1009                 write_lock_irq(&tasklist_lock);
1010                 spin_lock(&oldsighand->siglock);
1011                 rcu_assign_pointer(tsk->sighand, newsighand);
1012                 spin_unlock(&oldsighand->siglock);
1013                 write_unlock_irq(&tasklist_lock);
1014
1015                 __cleanup_sighand(oldsighand);
1016         }
1017
1018         BUG_ON(!thread_group_leader(tsk));
1019         return 0;
1020
1021 killed:
1022         /* protects against exit_notify() and __exit_signal() */
1023         read_lock(&tasklist_lock);
1024         sig->group_exit_task = NULL;
1025         sig->notify_count = 0;
1026         read_unlock(&tasklist_lock);
1027         return -EAGAIN;
1028 }
1029
1030 char *get_task_comm(char *buf, struct task_struct *tsk)
1031 {
1032         /* buf must be at least sizeof(tsk->comm) in size */
1033         task_lock(tsk);
1034         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1035         task_unlock(tsk);
1036         return buf;
1037 }
1038 EXPORT_SYMBOL_GPL(get_task_comm);
1039
1040 /*
1041  * These functions flushes out all traces of the currently running executable
1042  * so that a new one can be started
1043  */
1044
1045 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1046 {
1047         task_lock(tsk);
1048         trace_task_rename(tsk, buf);
1049         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1050         task_unlock(tsk);
1051         perf_event_comm(tsk, exec);
1052 }
1053
1054 int flush_old_exec(struct linux_binprm * bprm)
1055 {
1056         int retval;
1057
1058         /*
1059          * Make sure we have a private signal table and that
1060          * we are unassociated from the previous thread group.
1061          */
1062         retval = de_thread(current);
1063         if (retval)
1064                 goto out;
1065
1066         set_mm_exe_file(bprm->mm, bprm->file);
1067         /*
1068          * Release all of the old mmap stuff
1069          */
1070         acct_arg_size(bprm, 0);
1071         retval = exec_mmap(bprm->mm);
1072         if (retval)
1073                 goto out;
1074
1075         bprm->mm = NULL;                /* We're using it now */
1076
1077         set_fs(USER_DS);
1078         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1079                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1080         flush_thread();
1081         current->personality &= ~bprm->per_clear;
1082
1083         return 0;
1084
1085 out:
1086         return retval;
1087 }
1088 EXPORT_SYMBOL(flush_old_exec);
1089
1090 void would_dump(struct linux_binprm *bprm, struct file *file)
1091 {
1092         if (inode_permission(file_inode(file), MAY_READ) < 0)
1093                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1094 }
1095 EXPORT_SYMBOL(would_dump);
1096
1097 void setup_new_exec(struct linux_binprm * bprm)
1098 {
1099         arch_pick_mmap_layout(current->mm);
1100
1101         /* This is the point of no return */
1102         current->sas_ss_sp = current->sas_ss_size = 0;
1103
1104         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1105                 set_dumpable(current->mm, SUID_DUMP_USER);
1106         else
1107                 set_dumpable(current->mm, suid_dumpable);
1108
1109         perf_event_exec();
1110         __set_task_comm(current, kbasename(bprm->filename), true);
1111
1112         /* Set the new mm task size. We have to do that late because it may
1113          * depend on TIF_32BIT which is only updated in flush_thread() on
1114          * some architectures like powerpc
1115          */
1116         current->mm->task_size = TASK_SIZE;
1117
1118         /* install the new credentials */
1119         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1120             !gid_eq(bprm->cred->gid, current_egid())) {
1121                 current->pdeath_signal = 0;
1122         } else {
1123                 would_dump(bprm, bprm->file);
1124                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1125                         set_dumpable(current->mm, suid_dumpable);
1126         }
1127
1128         /* An exec changes our domain. We are no longer part of the thread
1129            group */
1130         current->self_exec_id++;
1131         flush_signal_handlers(current, 0);
1132         do_close_on_exec(current->files);
1133 }
1134 EXPORT_SYMBOL(setup_new_exec);
1135
1136 /*
1137  * Prepare credentials and lock ->cred_guard_mutex.
1138  * install_exec_creds() commits the new creds and drops the lock.
1139  * Or, if exec fails before, free_bprm() should release ->cred and
1140  * and unlock.
1141  */
1142 int prepare_bprm_creds(struct linux_binprm *bprm)
1143 {
1144         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1145                 return -ERESTARTNOINTR;
1146
1147         bprm->cred = prepare_exec_creds();
1148         if (likely(bprm->cred))
1149                 return 0;
1150
1151         mutex_unlock(&current->signal->cred_guard_mutex);
1152         return -ENOMEM;
1153 }
1154
1155 static void free_bprm(struct linux_binprm *bprm)
1156 {
1157         free_arg_pages(bprm);
1158         if (bprm->cred) {
1159                 mutex_unlock(&current->signal->cred_guard_mutex);
1160                 abort_creds(bprm->cred);
1161         }
1162         if (bprm->file) {
1163                 allow_write_access(bprm->file);
1164                 fput(bprm->file);
1165         }
1166         /* If a binfmt changed the interp, free it. */
1167         if (bprm->interp != bprm->filename)
1168                 kfree(bprm->interp);
1169         kfree(bprm);
1170 }
1171
1172 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1173 {
1174         /* If a binfmt changed the interp, free it first. */
1175         if (bprm->interp != bprm->filename)
1176                 kfree(bprm->interp);
1177         bprm->interp = kstrdup(interp, GFP_KERNEL);
1178         if (!bprm->interp)
1179                 return -ENOMEM;
1180         return 0;
1181 }
1182 EXPORT_SYMBOL(bprm_change_interp);
1183
1184 /*
1185  * install the new credentials for this executable
1186  */
1187 void install_exec_creds(struct linux_binprm *bprm)
1188 {
1189         security_bprm_committing_creds(bprm);
1190
1191         commit_creds(bprm->cred);
1192         bprm->cred = NULL;
1193
1194         /*
1195          * Disable monitoring for regular users
1196          * when executing setuid binaries. Must
1197          * wait until new credentials are committed
1198          * by commit_creds() above
1199          */
1200         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1201                 perf_event_exit_task(current);
1202         /*
1203          * cred_guard_mutex must be held at least to this point to prevent
1204          * ptrace_attach() from altering our determination of the task's
1205          * credentials; any time after this it may be unlocked.
1206          */
1207         security_bprm_committed_creds(bprm);
1208         mutex_unlock(&current->signal->cred_guard_mutex);
1209 }
1210 EXPORT_SYMBOL(install_exec_creds);
1211
1212 /*
1213  * determine how safe it is to execute the proposed program
1214  * - the caller must hold ->cred_guard_mutex to protect against
1215  *   PTRACE_ATTACH or seccomp thread-sync
1216  */
1217 static void check_unsafe_exec(struct linux_binprm *bprm)
1218 {
1219         struct task_struct *p = current, *t;
1220         unsigned n_fs;
1221
1222         if (p->ptrace) {
1223                 if (p->ptrace & PT_PTRACE_CAP)
1224                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1225                 else
1226                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1227         }
1228
1229         /*
1230          * This isn't strictly necessary, but it makes it harder for LSMs to
1231          * mess up.
1232          */
1233         if (task_no_new_privs(current))
1234                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1235
1236         t = p;
1237         n_fs = 1;
1238         spin_lock(&p->fs->lock);
1239         rcu_read_lock();
1240         while_each_thread(p, t) {
1241                 if (t->fs == p->fs)
1242                         n_fs++;
1243         }
1244         rcu_read_unlock();
1245
1246         if (p->fs->users > n_fs)
1247                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1248         else
1249                 p->fs->in_exec = 1;
1250         spin_unlock(&p->fs->lock);
1251 }
1252
1253 /*
1254  * Fill the binprm structure from the inode.
1255  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1256  *
1257  * This may be called multiple times for binary chains (scripts for example).
1258  */
1259 int prepare_binprm(struct linux_binprm *bprm)
1260 {
1261         struct inode *inode = file_inode(bprm->file);
1262         umode_t mode = inode->i_mode;
1263         int retval;
1264
1265
1266         /* clear any previous set[ug]id data from a previous binary */
1267         bprm->cred->euid = current_euid();
1268         bprm->cred->egid = current_egid();
1269
1270         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1271             !task_no_new_privs(current) &&
1272             kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1273             kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1274                 /* Set-uid? */
1275                 if (mode & S_ISUID) {
1276                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1277                         bprm->cred->euid = inode->i_uid;
1278                 }
1279
1280                 /* Set-gid? */
1281                 /*
1282                  * If setgid is set but no group execute bit then this
1283                  * is a candidate for mandatory locking, not a setgid
1284                  * executable.
1285                  */
1286                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1287                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1288                         bprm->cred->egid = inode->i_gid;
1289                 }
1290         }
1291
1292         /* fill in binprm security blob */
1293         retval = security_bprm_set_creds(bprm);
1294         if (retval)
1295                 return retval;
1296         bprm->cred_prepared = 1;
1297
1298         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1299         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1300 }
1301
1302 EXPORT_SYMBOL(prepare_binprm);
1303
1304 /*
1305  * Arguments are '\0' separated strings found at the location bprm->p
1306  * points to; chop off the first by relocating brpm->p to right after
1307  * the first '\0' encountered.
1308  */
1309 int remove_arg_zero(struct linux_binprm *bprm)
1310 {
1311         int ret = 0;
1312         unsigned long offset;
1313         char *kaddr;
1314         struct page *page;
1315
1316         if (!bprm->argc)
1317                 return 0;
1318
1319         do {
1320                 offset = bprm->p & ~PAGE_MASK;
1321                 page = get_arg_page(bprm, bprm->p, 0);
1322                 if (!page) {
1323                         ret = -EFAULT;
1324                         goto out;
1325                 }
1326                 kaddr = kmap_atomic(page);
1327
1328                 for (; offset < PAGE_SIZE && kaddr[offset];
1329                                 offset++, bprm->p++)
1330                         ;
1331
1332                 kunmap_atomic(kaddr);
1333                 put_arg_page(page);
1334
1335                 if (offset == PAGE_SIZE)
1336                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1337         } while (offset == PAGE_SIZE);
1338
1339         bprm->p++;
1340         bprm->argc--;
1341         ret = 0;
1342
1343 out:
1344         return ret;
1345 }
1346 EXPORT_SYMBOL(remove_arg_zero);
1347
1348 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1349 /*
1350  * cycle the list of binary formats handler, until one recognizes the image
1351  */
1352 int search_binary_handler(struct linux_binprm *bprm)
1353 {
1354         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1355         struct linux_binfmt *fmt;
1356         int retval;
1357
1358         /* This allows 4 levels of binfmt rewrites before failing hard. */
1359         if (bprm->recursion_depth > 5)
1360                 return -ELOOP;
1361
1362         retval = security_bprm_check(bprm);
1363         if (retval)
1364                 return retval;
1365
1366         retval = -ENOENT;
1367  retry:
1368         read_lock(&binfmt_lock);
1369         list_for_each_entry(fmt, &formats, lh) {
1370                 if (!try_module_get(fmt->module))
1371                         continue;
1372                 read_unlock(&binfmt_lock);
1373                 bprm->recursion_depth++;
1374                 retval = fmt->load_binary(bprm);
1375                 bprm->recursion_depth--;
1376                 if (retval >= 0 || retval != -ENOEXEC ||
1377                     bprm->mm == NULL || bprm->file == NULL) {
1378                         put_binfmt(fmt);
1379                         return retval;
1380                 }
1381                 read_lock(&binfmt_lock);
1382                 put_binfmt(fmt);
1383         }
1384         read_unlock(&binfmt_lock);
1385
1386         if (need_retry && retval == -ENOEXEC) {
1387                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1388                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1389                         return retval;
1390                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1391                         return retval;
1392                 need_retry = false;
1393                 goto retry;
1394         }
1395
1396         return retval;
1397 }
1398 EXPORT_SYMBOL(search_binary_handler);
1399
1400 static int exec_binprm(struct linux_binprm *bprm)
1401 {
1402         pid_t old_pid, old_vpid;
1403         int ret;
1404
1405         /* Need to fetch pid before load_binary changes it */
1406         old_pid = current->pid;
1407         rcu_read_lock();
1408         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1409         rcu_read_unlock();
1410
1411         ret = search_binary_handler(bprm);
1412         if (ret >= 0) {
1413                 audit_bprm(bprm);
1414                 trace_sched_process_exec(current, old_pid, bprm);
1415                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1416                 proc_exec_connector(current);
1417         }
1418
1419         return ret;
1420 }
1421
1422 /*
1423  * sys_execve() executes a new program.
1424  */
1425 static int do_execve_common(struct filename *filename,
1426                                 struct user_arg_ptr argv,
1427                                 struct user_arg_ptr envp)
1428 {
1429         struct linux_binprm *bprm;
1430         struct file *file;
1431         struct files_struct *displaced;
1432         int retval;
1433
1434         if (IS_ERR(filename))
1435                 return PTR_ERR(filename);
1436
1437         /*
1438          * We move the actual failure in case of RLIMIT_NPROC excess from
1439          * set*uid() to execve() because too many poorly written programs
1440          * don't check setuid() return code.  Here we additionally recheck
1441          * whether NPROC limit is still exceeded.
1442          */
1443         if ((current->flags & PF_NPROC_EXCEEDED) &&
1444             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1445                 retval = -EAGAIN;
1446                 goto out_ret;
1447         }
1448
1449         /* We're below the limit (still or again), so we don't want to make
1450          * further execve() calls fail. */
1451         current->flags &= ~PF_NPROC_EXCEEDED;
1452
1453         retval = unshare_files(&displaced);
1454         if (retval)
1455                 goto out_ret;
1456
1457         retval = -ENOMEM;
1458         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1459         if (!bprm)
1460                 goto out_files;
1461
1462         retval = prepare_bprm_creds(bprm);
1463         if (retval)
1464                 goto out_free;
1465
1466         check_unsafe_exec(bprm);
1467         current->in_execve = 1;
1468
1469         file = do_open_exec(filename);
1470         retval = PTR_ERR(file);
1471         if (IS_ERR(file))
1472                 goto out_unmark;
1473
1474         sched_exec();
1475
1476         bprm->file = file;
1477         bprm->filename = bprm->interp = filename->name;
1478
1479         retval = bprm_mm_init(bprm);
1480         if (retval)
1481                 goto out_unmark;
1482
1483         bprm->argc = count(argv, MAX_ARG_STRINGS);
1484         if ((retval = bprm->argc) < 0)
1485                 goto out;
1486
1487         bprm->envc = count(envp, MAX_ARG_STRINGS);
1488         if ((retval = bprm->envc) < 0)
1489                 goto out;
1490
1491         retval = prepare_binprm(bprm);
1492         if (retval < 0)
1493                 goto out;
1494
1495         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1496         if (retval < 0)
1497                 goto out;
1498
1499         bprm->exec = bprm->p;
1500         retval = copy_strings(bprm->envc, envp, bprm);
1501         if (retval < 0)
1502                 goto out;
1503
1504         retval = copy_strings(bprm->argc, argv, bprm);
1505         if (retval < 0)
1506                 goto out;
1507
1508         retval = exec_binprm(bprm);
1509         if (retval < 0)
1510                 goto out;
1511
1512         /* execve succeeded */
1513         current->fs->in_exec = 0;
1514         current->in_execve = 0;
1515         acct_update_integrals(current);
1516         task_numa_free(current);
1517         free_bprm(bprm);
1518         putname(filename);
1519         if (displaced)
1520                 put_files_struct(displaced);
1521         return retval;
1522
1523 out:
1524         if (bprm->mm) {
1525                 acct_arg_size(bprm, 0);
1526                 mmput(bprm->mm);
1527         }
1528
1529 out_unmark:
1530         current->fs->in_exec = 0;
1531         current->in_execve = 0;
1532
1533 out_free:
1534         free_bprm(bprm);
1535
1536 out_files:
1537         if (displaced)
1538                 reset_files_struct(displaced);
1539 out_ret:
1540         putname(filename);
1541         return retval;
1542 }
1543
1544 int do_execve(struct filename *filename,
1545         const char __user *const __user *__argv,
1546         const char __user *const __user *__envp)
1547 {
1548         struct user_arg_ptr argv = { .ptr.native = __argv };
1549         struct user_arg_ptr envp = { .ptr.native = __envp };
1550         return do_execve_common(filename, argv, envp);
1551 }
1552
1553 #ifdef CONFIG_COMPAT
1554 static int compat_do_execve(struct filename *filename,
1555         const compat_uptr_t __user *__argv,
1556         const compat_uptr_t __user *__envp)
1557 {
1558         struct user_arg_ptr argv = {
1559                 .is_compat = true,
1560                 .ptr.compat = __argv,
1561         };
1562         struct user_arg_ptr envp = {
1563                 .is_compat = true,
1564                 .ptr.compat = __envp,
1565         };
1566         return do_execve_common(filename, argv, envp);
1567 }
1568 #endif
1569
1570 void set_binfmt(struct linux_binfmt *new)
1571 {
1572         struct mm_struct *mm = current->mm;
1573
1574         if (mm->binfmt)
1575                 module_put(mm->binfmt->module);
1576
1577         mm->binfmt = new;
1578         if (new)
1579                 __module_get(new->module);
1580 }
1581 EXPORT_SYMBOL(set_binfmt);
1582
1583 /*
1584  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1585  */
1586 void set_dumpable(struct mm_struct *mm, int value)
1587 {
1588         unsigned long old, new;
1589
1590         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1591                 return;
1592
1593         do {
1594                 old = ACCESS_ONCE(mm->flags);
1595                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1596         } while (cmpxchg(&mm->flags, old, new) != old);
1597 }
1598
1599 SYSCALL_DEFINE3(execve,
1600                 const char __user *, filename,
1601                 const char __user *const __user *, argv,
1602                 const char __user *const __user *, envp)
1603 {
1604         return do_execve(getname(filename), argv, envp);
1605 }
1606 #ifdef CONFIG_COMPAT
1607 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1608         const compat_uptr_t __user *, argv,
1609         const compat_uptr_t __user *, envp)
1610 {
1611         return compat_do_execve(getname(filename), argv, envp);
1612 }
1613 #endif