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 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 SYSCALL_DEFINE1(uselib, const char __user *, library)
 108 {
 109         struct linux_binfmt *fmt;
 110         struct file *file;
 111         struct filename *tmp = getname(library);
 112         int error = PTR_ERR(tmp);
 113         static const struct open_flags uselib_flags = {
 114                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 115                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 116                 .intent = LOOKUP_OPEN,
 117                 .lookup_flags = LOOKUP_FOLLOW,
 118         };
 119
 120         if (IS_ERR(tmp))
 121                 goto out;
 122
 123         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 124         putname(tmp);
 125         error = PTR_ERR(file);
 126         if (IS_ERR(file))
 127                 goto out;
 128
 129         error = -EINVAL;
 130         if (!S_ISREG(file_inode(file)->i_mode))
 131                 goto exit;
 132
 133         error = -EACCES;
 134         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 135                 goto exit;
 136
 137         fsnotify_open(file);
 138
 139         error = -ENOEXEC;
 140
 141         read_lock(&binfmt_lock);
 142         list_for_each_entry(fmt, &formats, lh) {
 143                 if (!fmt->load_shlib)
 144                         continue;
 145                 if (!try_module_get(fmt->module))
 146                         continue;
 147                 read_unlock(&binfmt_lock);
 148                 error = fmt->load_shlib(file);
 149                 read_lock(&binfmt_lock);
 150                 put_binfmt(fmt);
 151                 if (error != -ENOEXEC)
 152                         break;
 153         }
 154         read_unlock(&binfmt_lock);
 155 exit:
 156         fput(file);
 157 out:
 158         return error;
 159 }
 160
 161 #ifdef CONFIG_MMU
 162 /*
 163  * The nascent bprm->mm is not visible until exec_mmap() but it can
 164  * use a lot of memory, account these pages in current->mm temporary
 165  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 166  * change the counter back via acct_arg_size(0).
 167  */
 168 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 169 {
 170         struct mm_struct *mm = current->mm;
 171         long diff = (long)(pages - bprm->vma_pages);
 172
 173         if (!mm || !diff)
 174                 return;
 175
 176         bprm->vma_pages = pages;
 177         add_mm_counter(mm, MM_ANONPAGES, diff);
 178 }
 179
 180 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 181                 int write)
 182 {
 183         struct page *page;
 184         int ret;
 185
 186 #ifdef CONFIG_STACK_GROWSUP
 187         if (write) {
 188                 ret = expand_downwards(bprm->vma, pos);
 189                 if (ret < 0)
 190                         return NULL;
 191         }
 192 #endif
 193         ret = get_user_pages(current, bprm->mm, pos,
 194                         1, write, 1, &page, NULL);
 195         if (ret <= 0)
 196                 return NULL;
 197
 198         if (write) {
 199                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 200                 struct rlimit *rlim;
 201
 202                 acct_arg_size(bprm, size / PAGE_SIZE);
 203
 204                 /*
 205                  * We've historically supported up to 32 pages (ARG_MAX)
 206                  * of argument strings even with small stacks
 207                  */
 208                 if (size <= ARG_MAX)
 209                         return page;
 210
 211                 /*
 212                  * Limit to 1/4-th the stack size for the argv+env strings.
 213                  * This ensures that:
 214                  *  - the remaining binfmt code will not run out of stack space,
 215                  *  - the program will have a reasonable amount of stack left
 216                  *    to work from.
 217                  */
 218                 rlim = current->signal->rlim;
 219                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 220                         put_page(page);
 221                         return NULL;
 222                 }
 223         }
 224
 225         return page;
 226 }
 227
 228 static void put_arg_page(struct page *page)
 229 {
 230         put_page(page);
 231 }
 232
 233 static void free_arg_page(struct linux_binprm *bprm, int i)
 234 {
 235 }
 236
 237 static void free_arg_pages(struct linux_binprm *bprm)
 238 {
 239 }
 240
 241 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 242                 struct page *page)
 243 {
 244         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 245 }
 246
 247 static int __bprm_mm_init(struct linux_binprm *bprm)
 248 {
 249         int err;
 250         struct vm_area_struct *vma = NULL;
 251         struct mm_struct *mm = bprm->mm;
 252
 253         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 254         if (!vma)
 255                 return -ENOMEM;
 256
 257         down_write(&mm->mmap_sem);
 258         vma->vm_mm = mm;
 259
 260         /*
 261          * Place the stack at the largest stack address the architecture
 262          * supports. Later, we'll move this to an appropriate place. We don't
 263          * use STACK_TOP because that can depend on attributes which aren't
 264          * configured yet.
 265          */
 266         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 267         vma->vm_end = STACK_TOP_MAX;
 268         vma->vm_start = vma->vm_end - PAGE_SIZE;
 269         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 270         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 271         INIT_LIST_HEAD(&vma->anon_vma_chain);
 272
 273         err = insert_vm_struct(mm, vma);
 274         if (err)
 275                 goto err;
 276
 277         mm->stack_vm = mm->total_vm = 1;
 278         up_write(&mm->mmap_sem);
 279         bprm->p = vma->vm_end - sizeof(void *);
 280         return 0;
 281 err:
 282         up_write(&mm->mmap_sem);
 283         bprm->vma = NULL;
 284         kmem_cache_free(vm_area_cachep, vma);
 285         return err;
 286 }
 287
 288 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 289 {
 290         return len <= MAX_ARG_STRLEN;
 291 }
 292
 293 #else
 294
 295 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 296 {
 297 }
 298
 299 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 300                 int write)
 301 {
 302         struct page *page;
 303
 304         page = bprm->page[pos / PAGE_SIZE];
 305         if (!page && write) {
 306                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 307                 if (!page)
 308                         return NULL;
 309                 bprm->page[pos / PAGE_SIZE] = page;
 310         }
 311
 312         return page;
 313 }
 314
 315 static void put_arg_page(struct page *page)
 316 {
 317 }
 318
 319 static void free_arg_page(struct linux_binprm *bprm, int i)
 320 {
 321         if (bprm->page[i]) {
 322                 __free_page(bprm->page[i]);
 323                 bprm->page[i] = NULL;
 324         }
 325 }
 326
 327 static void free_arg_pages(struct linux_binprm *bprm)
 328 {
 329         int i;
 330
 331         for (i = 0; i < MAX_ARG_PAGES; i++)
 332                 free_arg_page(bprm, i);
 333 }
 334
 335 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 336                 struct page *page)
 337 {
 338 }
 339
 340 static int __bprm_mm_init(struct linux_binprm *bprm)
 341 {
 342         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 343         return 0;
 344 }
 345
 346 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 347 {
 348         return len <= bprm->p;
 349 }
 350
 351 #endif /* CONFIG_MMU */
 352
 353 /*
 354  * Create a new mm_struct and populate it with a temporary stack
 355  * vm_area_struct.  We don't have enough context at this point to set the stack
 356  * flags, permissions, and offset, so we use temporary values.  We'll update
 357  * them later in setup_arg_pages().
 358  */
 359 static int bprm_mm_init(struct linux_binprm *bprm)
 360 {
 361         int err;
 362         struct mm_struct *mm = NULL;
 363
 364         bprm->mm = mm = mm_alloc();
 365         err = -ENOMEM;
 366         if (!mm)
 367                 goto err;
 368
 369         err = init_new_context(current, mm);
 370         if (err)
 371                 goto err;
 372
 373         err = __bprm_mm_init(bprm);
 374         if (err)
 375                 goto err;
 376
 377         return 0;
 378
 379 err:
 380         if (mm) {
 381                 bprm->mm = NULL;
 382                 mmdrop(mm);
 383         }
 384
 385         return err;
 386 }
 387
 388 struct user_arg_ptr {
 389 #ifdef CONFIG_COMPAT
 390         bool is_compat;
 391 #endif
 392         union {
 393                 const char __user *const __user *native;
 394 #ifdef CONFIG_COMPAT
 395                 const compat_uptr_t __user *compat;
 396 #endif
 397         } ptr;
 398 };
 399
 400 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 401 {
 402         const char __user *native;
 403
 404 #ifdef CONFIG_COMPAT
 405         if (unlikely(argv.is_compat)) {
 406                 compat_uptr_t compat;
 407
 408                 if (get_user(compat, argv.ptr.compat + nr))
 409                         return ERR_PTR(-EFAULT);
 410
 411                 return compat_ptr(compat);
 412         }
 413 #endif
 414
 415         if (get_user(native, argv.ptr.native + nr))
 416                 return ERR_PTR(-EFAULT);
 417
 418         return native;
 419 }
 420
 421 /*
 422  * count() counts the number of strings in array ARGV.
 423  */
 424 static int count(struct user_arg_ptr argv, int max)
 425 {
 426         int i = 0;
 427
 428         if (argv.ptr.native != NULL) {
 429                 for (;;) {
 430                         const char __user *p = get_user_arg_ptr(argv, i);
 431
 432                         if (!p)
 433                                 break;
 434
 435                         if (IS_ERR(p))
 436                                 return -EFAULT;
 437
 438                         if (i >= max)
 439                                 return -E2BIG;
 440                         ++i;
 441
 442                         if (fatal_signal_pending(current))
 443                                 return -ERESTARTNOHAND;
 444                         cond_resched();
 445                 }
 446         }
 447         return i;
 448 }
 449
 450 /*
 451  * 'copy_strings()' copies argument/environment strings from the old
 452  * processes's memory to the new process's stack.  The call to get_user_pages()
 453  * ensures the destination page is created and not swapped out.
 454  */
 455 static int copy_strings(int argc, struct user_arg_ptr argv,
 456                         struct linux_binprm *bprm)
 457 {
 458         struct page *kmapped_page = NULL;
 459         char *kaddr = NULL;
 460         unsigned long kpos = 0;
 461         int ret;
 462
 463         while (argc-- > 0) {
 464                 const char __user *str;
 465                 int len;
 466                 unsigned long pos;
 467
 468                 ret = -EFAULT;
 469                 str = get_user_arg_ptr(argv, argc);
 470                 if (IS_ERR(str))
 471                         goto out;
 472
 473                 len = strnlen_user(str, MAX_ARG_STRLEN);
 474                 if (!len)
 475                         goto out;
 476
 477                 ret = -E2BIG;
 478                 if (!valid_arg_len(bprm, len))
 479                         goto out;
 480
 481                 /* We're going to work our way backwords. */
 482                 pos = bprm->p;
 483                 str += len;
 484                 bprm->p -= len;
 485
 486                 while (len > 0) {
 487                         int offset, bytes_to_copy;
 488
 489                         if (fatal_signal_pending(current)) {
 490                                 ret = -ERESTARTNOHAND;
 491                                 goto out;
 492                         }
 493                         cond_resched();
 494
 495                         offset = pos % PAGE_SIZE;
 496                         if (offset == 0)
 497                                 offset = PAGE_SIZE;
 498
 499                         bytes_to_copy = offset;
 500                         if (bytes_to_copy > len)
 501                                 bytes_to_copy = len;
 502
 503                         offset -= bytes_to_copy;
 504                         pos -= bytes_to_copy;
 505                         str -= bytes_to_copy;
 506                         len -= bytes_to_copy;
 507
 508                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 509                                 struct page *page;
 510
 511                                 page = get_arg_page(bprm, pos, 1);
 512                                 if (!page) {
 513                                         ret = -E2BIG;
 514                                         goto out;
 515                                 }
 516
 517                                 if (kmapped_page) {
 518                                         flush_kernel_dcache_page(kmapped_page);
 519                                         kunmap(kmapped_page);
 520                                         put_arg_page(kmapped_page);
 521                                 }
 522                                 kmapped_page = page;
 523                                 kaddr = kmap(kmapped_page);
 524                                 kpos = pos & PAGE_MASK;
 525                                 flush_arg_page(bprm, kpos, kmapped_page);
 526                         }
 527                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 528                                 ret = -EFAULT;
 529                                 goto out;
 530                         }
 531                 }
 532         }
 533         ret = 0;
 534 out:
 535         if (kmapped_page) {
 536                 flush_kernel_dcache_page(kmapped_page);
 537                 kunmap(kmapped_page);
 538                 put_arg_page(kmapped_page);
 539         }
 540         return ret;
 541 }
 542
 543 /*
 544  * Like copy_strings, but get argv and its values from kernel memory.
 545  */
 546 int copy_strings_kernel(int argc, const char *const *__argv,
 547                         struct linux_binprm *bprm)
 548 {
 549         int r;
 550         mm_segment_t oldfs = get_fs();
 551         struct user_arg_ptr argv = {
 552                 .ptr.native = (const char __user *const  __user *)__argv,
 553         };
 554
 555         set_fs(KERNEL_DS);
 556         r = copy_strings(argc, argv, bprm);
 557         set_fs(oldfs);
 558
 559         return r;
 560 }
 561 EXPORT_SYMBOL(copy_strings_kernel);
 562
 563 #ifdef CONFIG_MMU
 564
 565 /*
 566  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 567  * the binfmt code determines where the new stack should reside, we shift it to
 568  * its final location.  The process proceeds as follows:
 569  *
 570  * 1) Use shift to calculate the new vma endpoints.
 571  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 572  *    arguments passed to subsequent functions are consistent.
 573  * 3) Move vma's page tables to the new range.
 574  * 4) Free up any cleared pgd range.
 575  * 5) Shrink the vma to cover only the new range.
 576  */
 577 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 578 {
 579         struct mm_struct *mm = vma->vm_mm;
 580         unsigned long old_start = vma->vm_start;
 581         unsigned long old_end = vma->vm_end;
 582         unsigned long length = old_end - old_start;
 583         unsigned long new_start = old_start - shift;
 584         unsigned long new_end = old_end - shift;
 585         struct mmu_gather tlb;
 586
 587         BUG_ON(new_start > new_end);
 588
 589         /*
 590          * ensure there are no vmas between where we want to go
 591          * and where we are
 592          */
 593         if (vma != find_vma(mm, new_start))
 594                 return -EFAULT;
 595
 596         /*
 597          * cover the whole range: [new_start, old_end)
 598          */
 599         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 600                 return -ENOMEM;
 601
 602         /*
 603          * move the page tables downwards, on failure we rely on
 604          * process cleanup to remove whatever mess we made.
 605          */
 606         if (length != move_page_tables(vma, old_start,
 607                                        vma, new_start, length, false))
 608                 return -ENOMEM;
 609
 610         lru_add_drain();
 611         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 612         if (new_end > old_start) {
 613                 /*
 614                  * when the old and new regions overlap clear from new_end.
 615                  */
 616                 free_pgd_range(&tlb, new_end, old_end, new_end,
 617                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 618         } else {
 619                 /*
 620                  * otherwise, clean from old_start; this is done to not touch
 621                  * the address space in [new_end, old_start) some architectures
 622                  * have constraints on va-space that make this illegal (IA64) -
 623                  * for the others its just a little faster.
 624                  */
 625                 free_pgd_range(&tlb, old_start, old_end, new_end,
 626                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 627         }
 628         tlb_finish_mmu(&tlb, old_start, old_end);
 629
 630         /*
 631          * Shrink the vma to just the new range.  Always succeeds.
 632          */
 633         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 634
 635         return 0;
 636 }
 637
 638 /*
 639  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 640  * the stack is optionally relocated, and some extra space is added.
 641  */
 642 int setup_arg_pages(struct linux_binprm *bprm,
 643                     unsigned long stack_top,
 644                     int executable_stack)
 645 {
 646         unsigned long ret;
 647         unsigned long stack_shift;
 648         struct mm_struct *mm = current->mm;
 649         struct vm_area_struct *vma = bprm->vma;
 650         struct vm_area_struct *prev = NULL;
 651         unsigned long vm_flags;
 652         unsigned long stack_base;
 653         unsigned long stack_size;
 654         unsigned long stack_expand;
 655         unsigned long rlim_stack;
 656
 657 #ifdef CONFIG_STACK_GROWSUP
 658         /* Limit stack size */
 659         stack_base = rlimit_max(RLIMIT_STACK);
 660         if (stack_base > STACK_SIZE_MAX)
 661                 stack_base = STACK_SIZE_MAX;
 662
 663         /* Make sure we didn't let the argument array grow too large. */
 664         if (vma->vm_end - vma->vm_start > stack_base)
 665                 return -ENOMEM;
 666
 667         stack_base = PAGE_ALIGN(stack_top - stack_base);
 668
 669         stack_shift = vma->vm_start - stack_base;
 670         mm->arg_start = bprm->p - stack_shift;
 671         bprm->p = vma->vm_end - stack_shift;
 672 #else
 673         stack_top = arch_align_stack(stack_top);
 674         stack_top = PAGE_ALIGN(stack_top);
 675
 676         if (unlikely(stack_top < mmap_min_addr) ||
 677             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 678                 return -ENOMEM;
 679
 680         stack_shift = vma->vm_end - stack_top;
 681
 682         bprm->p -= stack_shift;
 683         mm->arg_start = bprm->p;
 684 #endif
 685
 686         if (bprm->loader)
 687                 bprm->loader -= stack_shift;
 688         bprm->exec -= stack_shift;
 689
 690         down_write(&mm->mmap_sem);
 691         vm_flags = VM_STACK_FLAGS;
 692
 693         /*
 694          * Adjust stack execute permissions; explicitly enable for
 695          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 696          * (arch default) otherwise.
 697          */
 698         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 699                 vm_flags |= VM_EXEC;
 700         else if (executable_stack == EXSTACK_DISABLE_X)
 701                 vm_flags &= ~VM_EXEC;
 702         vm_flags |= mm->def_flags;
 703         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 704
 705         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 706                         vm_flags);
 707         if (ret)
 708                 goto out_unlock;
 709         BUG_ON(prev != vma);
 710
 711         /* Move stack pages down in memory. */
 712         if (stack_shift) {
 713                 ret = shift_arg_pages(vma, stack_shift);
 714                 if (ret)
 715                         goto out_unlock;
 716         }
 717
 718         /* mprotect_fixup is overkill to remove the temporary stack flags */
 719         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 720
 721         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 722         stack_size = vma->vm_end - vma->vm_start;
 723         /*
 724          * Align this down to a page boundary as expand_stack
 725          * will align it up.
 726          */
 727         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 728 #ifdef CONFIG_STACK_GROWSUP
 729         if (stack_size + stack_expand > rlim_stack)
 730                 stack_base = vma->vm_start + rlim_stack;
 731         else
 732                 stack_base = vma->vm_end + stack_expand;
 733 #else
 734         if (stack_size + stack_expand > rlim_stack)
 735                 stack_base = vma->vm_end - rlim_stack;
 736         else
 737                 stack_base = vma->vm_start - stack_expand;
 738 #endif
 739         current->mm->start_stack = bprm->p;
 740         ret = expand_stack(vma, stack_base);
 741         if (ret)
 742                 ret = -EFAULT;
 743
 744 out_unlock:
 745         up_write(&mm->mmap_sem);
 746         return ret;
 747 }
 748 EXPORT_SYMBOL(setup_arg_pages);
 749
 750 #endif /* CONFIG_MMU */
 751
 752 static struct file *do_open_exec(struct filename *name)
 753 {
 754         struct file *file;
 755         int err;
 756         static const struct open_flags open_exec_flags = {
 757                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 758                 .acc_mode = MAY_EXEC | MAY_OPEN,
 759                 .intent = LOOKUP_OPEN,
 760                 .lookup_flags = LOOKUP_FOLLOW,
 761         };
 762
 763         file = do_filp_open(AT_FDCWD, name, &open_exec_flags);
 764         if (IS_ERR(file))
 765                 goto out;
 766
 767         err = -EACCES;
 768         if (!S_ISREG(file_inode(file)->i_mode))
 769                 goto exit;
 770
 771         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 772                 goto exit;
 773
 774         fsnotify_open(file);
 775
 776         err = deny_write_access(file);
 777         if (err)
 778                 goto exit;
 779
 780 out:
 781         return file;
 782
 783 exit:
 784         fput(file);
 785         return ERR_PTR(err);
 786 }
 787
 788 struct file *open_exec(const char *name)
 789 {
 790         struct filename tmp = { .name = name };
 791         return do_open_exec(&tmp);
 792 }
 793 EXPORT_SYMBOL(open_exec);
 794
 795 int kernel_read(struct file *file, loff_t offset,
 796                 char *addr, unsigned long count)
 797 {
 798         mm_segment_t old_fs;
 799         loff_t pos = offset;
 800         int result;
 801
 802         old_fs = get_fs();
 803         set_fs(get_ds());
 804         /* The cast to a user pointer is valid due to the set_fs() */
 805         result = vfs_read(file, (void __user *)addr, count, &pos);
 806         set_fs(old_fs);
 807         return result;
 808 }
 809
 810 EXPORT_SYMBOL(kernel_read);
 811
 812 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 813 {
 814         ssize_t res = file->f_op->read(file, (void __user *)addr, len, &pos);
 815         if (res > 0)
 816                 flush_icache_range(addr, addr + len);
 817         return res;
 818 }
 819 EXPORT_SYMBOL(read_code);
 820
 821 static int exec_mmap(struct mm_struct *mm)
 822 {
 823         struct task_struct *tsk;
 824         struct mm_struct *old_mm, *active_mm;
 825
 826         /* Notify parent that we're no longer interested in the old VM */
 827         tsk = current;
 828         old_mm = current->mm;
 829         mm_release(tsk, old_mm);
 830
 831         if (old_mm) {
 832                 sync_mm_rss(old_mm);
 833                 /*
 834                  * Make sure that if there is a core dump in progress
 835                  * for the old mm, we get out and die instead of going
 836                  * through with the exec.  We must hold mmap_sem around
 837                  * checking core_state and changing tsk->mm.
 838                  */
 839                 down_read(&old_mm->mmap_sem);
 840                 if (unlikely(old_mm->core_state)) {
 841                         up_read(&old_mm->mmap_sem);
 842                         return -EINTR;
 843                 }
 844         }
 845         task_lock(tsk);
 846         active_mm = tsk->active_mm;
 847         tsk->mm = mm;
 848         tsk->active_mm = mm;
 849         activate_mm(active_mm, mm);
 850         tsk->mm->vmacache_seqnum = 0;
 851         vmacache_flush(tsk);
 852         task_unlock(tsk);
 853         if (old_mm) {
 854                 up_read(&old_mm->mmap_sem);
 855                 BUG_ON(active_mm != old_mm);
 856                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 857                 mm_update_next_owner(old_mm);
 858                 mmput(old_mm);
 859                 return 0;
 860         }
 861         mmdrop(active_mm);
 862         return 0;
 863 }
 864
 865 /*
 866  * This function makes sure the current process has its own signal table,
 867  * so that flush_signal_handlers can later reset the handlers without
 868  * disturbing other processes.  (Other processes might share the signal
 869  * table via the CLONE_SIGHAND option to clone().)
 870  */
 871 static int de_thread(struct task_struct *tsk)
 872 {
 873         struct signal_struct *sig = tsk->signal;
 874         struct sighand_struct *oldsighand = tsk->sighand;
 875         spinlock_t *lock = &oldsighand->siglock;
 876
 877         if (thread_group_empty(tsk))
 878                 goto no_thread_group;
 879
 880         /*
 881          * Kill all other threads in the thread group.
 882          */
 883         spin_lock_irq(lock);
 884         if (signal_group_exit(sig)) {
 885                 /*
 886                  * Another group action in progress, just
 887                  * return so that the signal is processed.
 888                  */
 889                 spin_unlock_irq(lock);
 890                 return -EAGAIN;
 891         }
 892
 893         sig->group_exit_task = tsk;
 894         sig->notify_count = zap_other_threads(tsk);
 895         if (!thread_group_leader(tsk))
 896                 sig->notify_count--;
 897
 898         while (sig->notify_count) {
 899                 __set_current_state(TASK_KILLABLE);
 900                 spin_unlock_irq(lock);
 901                 schedule();
 902                 if (unlikely(__fatal_signal_pending(tsk)))
 903                         goto killed;
 904                 spin_lock_irq(lock);
 905         }
 906         spin_unlock_irq(lock);
 907
 908         /*
 909          * At this point all other threads have exited, all we have to
 910          * do is to wait for the thread group leader to become inactive,
 911          * and to assume its PID:
 912          */
 913         if (!thread_group_leader(tsk)) {
 914                 struct task_struct *leader = tsk->group_leader;
 915
 916                 sig->notify_count = -1; /* for exit_notify() */
 917                 for (;;) {
 918                         threadgroup_change_begin(tsk);
 919                         write_lock_irq(&tasklist_lock);
 920                         if (likely(leader->exit_state))
 921                                 break;
 922                         __set_current_state(TASK_KILLABLE);
 923                         write_unlock_irq(&tasklist_lock);
 924                         threadgroup_change_end(tsk);
 925                         schedule();
 926                         if (unlikely(__fatal_signal_pending(tsk)))
 927                                 goto killed;
 928                 }
 929
 930                 /*
 931                  * The only record we have of the real-time age of a
 932                  * process, regardless of execs it's done, is start_time.
 933                  * All the past CPU time is accumulated in signal_struct
 934                  * from sister threads now dead.  But in this non-leader
 935                  * exec, nothing survives from the original leader thread,
 936                  * whose birth marks the true age of this process now.
 937                  * When we take on its identity by switching to its PID, we
 938                  * also take its birthdate (always earlier than our own).
 939                  */
 940                 tsk->start_time = leader->start_time;
 941                 tsk->real_start_time = leader->real_start_time;
 942
 943                 BUG_ON(!same_thread_group(leader, tsk));
 944                 BUG_ON(has_group_leader_pid(tsk));
 945                 /*
 946                  * An exec() starts a new thread group with the
 947                  * TGID of the previous thread group. Rehash the
 948                  * two threads with a switched PID, and release
 949                  * the former thread group leader:
 950                  */
 951
 952                 /* Become a process group leader with the old leader's pid.
 953                  * The old leader becomes a thread of the this thread group.
 954                  * Note: The old leader also uses this pid until release_task
 955                  *       is called.  Odd but simple and correct.
 956                  */
 957                 tsk->pid = leader->pid;
 958                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
 959                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 960                 transfer_pid(leader, tsk, PIDTYPE_SID);
 961
 962                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 963                 list_replace_init(&leader->sibling, &tsk->sibling);
 964
 965                 tsk->group_leader = tsk;
 966                 leader->group_leader = tsk;
 967
 968                 tsk->exit_signal = SIGCHLD;
 969                 leader->exit_signal = -1;
 970
 971                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 972                 leader->exit_state = EXIT_DEAD;
 973
 974                 /*
 975                  * We are going to release_task()->ptrace_unlink() silently,
 976                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 977                  * the tracer wont't block again waiting for this thread.
 978                  */
 979                 if (unlikely(leader->ptrace))
 980                         __wake_up_parent(leader, leader->parent);
 981                 write_unlock_irq(&tasklist_lock);
 982                 threadgroup_change_end(tsk);
 983
 984                 release_task(leader);
 985         }
 986
 987         sig->group_exit_task = NULL;
 988         sig->notify_count = 0;
 989
 990 no_thread_group:
 991         /* we have changed execution domain */
 992         tsk->exit_signal = SIGCHLD;
 993
 994         exit_itimers(sig);
 995         flush_itimer_signals();
 996
 997         if (atomic_read(&oldsighand->count) != 1) {
 998                 struct sighand_struct *newsighand;
 999                 /*
1000                  * This ->sighand is shared with the CLONE_SIGHAND
1001                  * but not CLONE_THREAD task, switch to the new one.
1002                  */
1003                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1004                 if (!newsighand)
1005                         return -ENOMEM;
1006
1007                 atomic_set(&newsighand->count, 1);
1008                 memcpy(newsighand->action, oldsighand->action,
1009                        sizeof(newsighand->action));
1010
1011                 write_lock_irq(&tasklist_lock);
1012                 spin_lock(&oldsighand->siglock);
1013                 rcu_assign_pointer(tsk->sighand, newsighand);
1014                 spin_unlock(&oldsighand->siglock);
1015                 write_unlock_irq(&tasklist_lock);
1016
1017                 __cleanup_sighand(oldsighand);
1018         }
1019
1020         BUG_ON(!thread_group_leader(tsk));
1021         return 0;
1022
1023 killed:
1024         /* protects against exit_notify() and __exit_signal() */
1025         read_lock(&tasklist_lock);
1026         sig->group_exit_task = NULL;
1027         sig->notify_count = 0;
1028         read_unlock(&tasklist_lock);
1029         return -EAGAIN;
1030 }
1031
1032 char *get_task_comm(char *buf, struct task_struct *tsk)
1033 {
1034         /* buf must be at least sizeof(tsk->comm) in size */
1035         task_lock(tsk);
1036         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1037         task_unlock(tsk);
1038         return buf;
1039 }
1040 EXPORT_SYMBOL_GPL(get_task_comm);
1041
1042 /*
1043  * These functions flushes out all traces of the currently running executable
1044  * so that a new one can be started
1045  */
1046
1047 void set_task_comm(struct task_struct *tsk, char *buf)
1048 {
1049         task_lock(tsk);
1050         trace_task_rename(tsk, buf);
1051         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1052         task_unlock(tsk);
1053         perf_event_comm(tsk);
1054 }
1055
1056 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1057 {
1058         int i, ch;
1059
1060         /* Copies the binary name from after last slash */
1061         for (i = 0; (ch = *(fn++)) != '\0';) {
1062                 if (ch == '/')
1063                         i = 0; /* overwrite what we wrote */
1064                 else
1065                         if (i < len - 1)
1066                                 tcomm[i++] = ch;
1067         }
1068         tcomm[i] = '\0';
1069 }
1070
1071 int flush_old_exec(struct linux_binprm * bprm)
1072 {
1073         int retval;
1074
1075         /*
1076          * Make sure we have a private signal table and that
1077          * we are unassociated from the previous thread group.
1078          */
1079         retval = de_thread(current);
1080         if (retval)
1081                 goto out;
1082
1083         set_mm_exe_file(bprm->mm, bprm->file);
1084
1085         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1086         /*
1087          * Release all of the old mmap stuff
1088          */
1089         acct_arg_size(bprm, 0);
1090         retval = exec_mmap(bprm->mm);
1091         if (retval)
1092                 goto out;
1093
1094         bprm->mm = NULL;                /* We're using it now */
1095
1096         set_fs(USER_DS);
1097         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1098                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1099         flush_thread();
1100         current->personality &= ~bprm->per_clear;
1101
1102         return 0;
1103
1104 out:
1105         return retval;
1106 }
1107 EXPORT_SYMBOL(flush_old_exec);
1108
1109 void would_dump(struct linux_binprm *bprm, struct file *file)
1110 {
1111         if (inode_permission(file_inode(file), MAY_READ) < 0)
1112                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1113 }
1114 EXPORT_SYMBOL(would_dump);
1115
1116 void setup_new_exec(struct linux_binprm * bprm)
1117 {
1118         arch_pick_mmap_layout(current->mm);
1119
1120         /* This is the point of no return */
1121         current->sas_ss_sp = current->sas_ss_size = 0;
1122
1123         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1124                 set_dumpable(current->mm, SUID_DUMP_USER);
1125         else
1126                 set_dumpable(current->mm, suid_dumpable);
1127
1128         set_task_comm(current, bprm->tcomm);
1129
1130         /* Set the new mm task size. We have to do that late because it may
1131          * depend on TIF_32BIT which is only updated in flush_thread() on
1132          * some architectures like powerpc
1133          */
1134         current->mm->task_size = TASK_SIZE;
1135
1136         /* install the new credentials */
1137         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1138             !gid_eq(bprm->cred->gid, current_egid())) {
1139                 current->pdeath_signal = 0;
1140         } else {
1141                 would_dump(bprm, bprm->file);
1142                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1143                         set_dumpable(current->mm, suid_dumpable);
1144         }
1145
1146         /* An exec changes our domain. We are no longer part of the thread
1147            group */
1148         current->self_exec_id++;
1149         flush_signal_handlers(current, 0);
1150         do_close_on_exec(current->files);
1151 }
1152 EXPORT_SYMBOL(setup_new_exec);
1153
1154 /*
1155  * Prepare credentials and lock ->cred_guard_mutex.
1156  * install_exec_creds() commits the new creds and drops the lock.
1157  * Or, if exec fails before, free_bprm() should release ->cred and
1158  * and unlock.
1159  */
1160 int prepare_bprm_creds(struct linux_binprm *bprm)
1161 {
1162         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1163                 return -ERESTARTNOINTR;
1164
1165         bprm->cred = prepare_exec_creds();
1166         if (likely(bprm->cred))
1167                 return 0;
1168
1169         mutex_unlock(&current->signal->cred_guard_mutex);
1170         return -ENOMEM;
1171 }
1172
1173 static void free_bprm(struct linux_binprm *bprm)
1174 {
1175         free_arg_pages(bprm);
1176         if (bprm->cred) {
1177                 mutex_unlock(&current->signal->cred_guard_mutex);
1178                 abort_creds(bprm->cred);
1179         }
1180         if (bprm->file) {
1181                 allow_write_access(bprm->file);
1182                 fput(bprm->file);
1183         }
1184         /* If a binfmt changed the interp, free it. */
1185         if (bprm->interp != bprm->filename)
1186                 kfree(bprm->interp);
1187         kfree(bprm);
1188 }
1189
1190 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1191 {
1192         /* If a binfmt changed the interp, free it first. */
1193         if (bprm->interp != bprm->filename)
1194                 kfree(bprm->interp);
1195         bprm->interp = kstrdup(interp, GFP_KERNEL);
1196         if (!bprm->interp)
1197                 return -ENOMEM;
1198         return 0;
1199 }
1200 EXPORT_SYMBOL(bprm_change_interp);
1201
1202 /*
1203  * install the new credentials for this executable
1204  */
1205 void install_exec_creds(struct linux_binprm *bprm)
1206 {
1207         security_bprm_committing_creds(bprm);
1208
1209         commit_creds(bprm->cred);
1210         bprm->cred = NULL;
1211
1212         /*
1213          * Disable monitoring for regular users
1214          * when executing setuid binaries. Must
1215          * wait until new credentials are committed
1216          * by commit_creds() above
1217          */
1218         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1219                 perf_event_exit_task(current);
1220         /*
1221          * cred_guard_mutex must be held at least to this point to prevent
1222          * ptrace_attach() from altering our determination of the task's
1223          * credentials; any time after this it may be unlocked.
1224          */
1225         security_bprm_committed_creds(bprm);
1226         mutex_unlock(&current->signal->cred_guard_mutex);
1227 }
1228 EXPORT_SYMBOL(install_exec_creds);
1229
1230 /*
1231  * determine how safe it is to execute the proposed program
1232  * - the caller must hold ->cred_guard_mutex to protect against
1233  *   PTRACE_ATTACH
1234  */
1235 static void check_unsafe_exec(struct linux_binprm *bprm)
1236 {
1237         struct task_struct *p = current, *t;
1238         unsigned n_fs;
1239
1240         if (p->ptrace) {
1241                 if (p->ptrace & PT_PTRACE_CAP)
1242                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1243                 else
1244                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1245         }
1246
1247         /*
1248          * This isn't strictly necessary, but it makes it harder for LSMs to
1249          * mess up.
1250          */
1251         if (current->no_new_privs)
1252                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1253
1254         t = p;
1255         n_fs = 1;
1256         spin_lock(&p->fs->lock);
1257         rcu_read_lock();
1258         while_each_thread(p, t) {
1259                 if (t->fs == p->fs)
1260                         n_fs++;
1261         }
1262         rcu_read_unlock();
1263
1264         if (p->fs->users > n_fs)
1265                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1266         else
1267                 p->fs->in_exec = 1;
1268         spin_unlock(&p->fs->lock);
1269 }
1270
1271 /*
1272  * Fill the binprm structure from the inode.
1273  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1274  *
1275  * This may be called multiple times for binary chains (scripts for example).
1276  */
1277 int prepare_binprm(struct linux_binprm *bprm)
1278 {
1279         struct inode *inode = file_inode(bprm->file);
1280         umode_t mode = inode->i_mode;
1281         int retval;
1282
1283
1284         /* clear any previous set[ug]id data from a previous binary */
1285         bprm->cred->euid = current_euid();
1286         bprm->cred->egid = current_egid();
1287
1288         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1289             !current->no_new_privs &&
1290             kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1291             kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1292                 /* Set-uid? */
1293                 if (mode & S_ISUID) {
1294                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1295                         bprm->cred->euid = inode->i_uid;
1296                 }
1297
1298                 /* Set-gid? */
1299                 /*
1300                  * If setgid is set but no group execute bit then this
1301                  * is a candidate for mandatory locking, not a setgid
1302                  * executable.
1303                  */
1304                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1305                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1306                         bprm->cred->egid = inode->i_gid;
1307                 }
1308         }
1309
1310         /* fill in binprm security blob */
1311         retval = security_bprm_set_creds(bprm);
1312         if (retval)
1313                 return retval;
1314         bprm->cred_prepared = 1;
1315
1316         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1317         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1318 }
1319
1320 EXPORT_SYMBOL(prepare_binprm);
1321
1322 /*
1323  * Arguments are '\0' separated strings found at the location bprm->p
1324  * points to; chop off the first by relocating brpm->p to right after
1325  * the first '\0' encountered.
1326  */
1327 int remove_arg_zero(struct linux_binprm *bprm)
1328 {
1329         int ret = 0;
1330         unsigned long offset;
1331         char *kaddr;
1332         struct page *page;
1333
1334         if (!bprm->argc)
1335                 return 0;
1336
1337         do {
1338                 offset = bprm->p & ~PAGE_MASK;
1339                 page = get_arg_page(bprm, bprm->p, 0);
1340                 if (!page) {
1341                         ret = -EFAULT;
1342                         goto out;
1343                 }
1344                 kaddr = kmap_atomic(page);
1345
1346                 for (; offset < PAGE_SIZE && kaddr[offset];
1347                                 offset++, bprm->p++)
1348                         ;
1349
1350                 kunmap_atomic(kaddr);
1351                 put_arg_page(page);
1352
1353                 if (offset == PAGE_SIZE)
1354                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1355         } while (offset == PAGE_SIZE);
1356
1357         bprm->p++;
1358         bprm->argc--;
1359         ret = 0;
1360
1361 out:
1362         return ret;
1363 }
1364 EXPORT_SYMBOL(remove_arg_zero);
1365
1366 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1367 /*
1368  * cycle the list of binary formats handler, until one recognizes the image
1369  */
1370 int search_binary_handler(struct linux_binprm *bprm)
1371 {
1372         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1373         struct linux_binfmt *fmt;
1374         int retval;
1375
1376         /* This allows 4 levels of binfmt rewrites before failing hard. */
1377         if (bprm->recursion_depth > 5)
1378                 return -ELOOP;
1379
1380         retval = security_bprm_check(bprm);
1381         if (retval)
1382                 return retval;
1383
1384         retval = -ENOENT;
1385  retry:
1386         read_lock(&binfmt_lock);
1387         list_for_each_entry(fmt, &formats, lh) {
1388                 if (!try_module_get(fmt->module))
1389                         continue;
1390                 read_unlock(&binfmt_lock);
1391                 bprm->recursion_depth++;
1392                 retval = fmt->load_binary(bprm);
1393                 bprm->recursion_depth--;
1394                 if (retval >= 0 || retval != -ENOEXEC ||
1395                     bprm->mm == NULL || bprm->file == NULL) {
1396                         put_binfmt(fmt);
1397                         return retval;
1398                 }
1399                 read_lock(&binfmt_lock);
1400                 put_binfmt(fmt);
1401         }
1402         read_unlock(&binfmt_lock);
1403
1404         if (need_retry && retval == -ENOEXEC) {
1405                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1406                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1407                         return retval;
1408                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1409                         return retval;
1410                 need_retry = false;
1411                 goto retry;
1412         }
1413
1414         return retval;
1415 }
1416 EXPORT_SYMBOL(search_binary_handler);
1417
1418 static int exec_binprm(struct linux_binprm *bprm)
1419 {
1420         pid_t old_pid, old_vpid;
1421         int ret;
1422
1423         /* Need to fetch pid before load_binary changes it */
1424         old_pid = current->pid;
1425         rcu_read_lock();
1426         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1427         rcu_read_unlock();
1428
1429         ret = search_binary_handler(bprm);
1430         if (ret >= 0) {
1431                 audit_bprm(bprm);
1432                 trace_sched_process_exec(current, old_pid, bprm);
1433                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1434                 proc_exec_connector(current);
1435         }
1436
1437         return ret;
1438 }
1439
1440 /*
1441  * sys_execve() executes a new program.
1442  */
1443 static int do_execve_common(struct filename *filename,
1444                                 struct user_arg_ptr argv,
1445                                 struct user_arg_ptr envp)
1446 {
1447         struct linux_binprm *bprm;
1448         struct file *file;
1449         struct files_struct *displaced;
1450         int retval;
1451
1452         if (IS_ERR(filename))
1453                 return PTR_ERR(filename);
1454
1455         /*
1456          * We move the actual failure in case of RLIMIT_NPROC excess from
1457          * set*uid() to execve() because too many poorly written programs
1458          * don't check setuid() return code.  Here we additionally recheck
1459          * whether NPROC limit is still exceeded.
1460          */
1461         if ((current->flags & PF_NPROC_EXCEEDED) &&
1462             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1463                 retval = -EAGAIN;
1464                 goto out_ret;
1465         }
1466
1467         /* We're below the limit (still or again), so we don't want to make
1468          * further execve() calls fail. */
1469         current->flags &= ~PF_NPROC_EXCEEDED;
1470
1471         retval = unshare_files(&displaced);
1472         if (retval)
1473                 goto out_ret;
1474
1475         retval = -ENOMEM;
1476         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1477         if (!bprm)
1478                 goto out_files;
1479
1480         retval = prepare_bprm_creds(bprm);
1481         if (retval)
1482                 goto out_free;
1483
1484         check_unsafe_exec(bprm);
1485         current->in_execve = 1;
1486
1487         file = do_open_exec(filename);
1488         retval = PTR_ERR(file);
1489         if (IS_ERR(file))
1490                 goto out_unmark;
1491
1492         sched_exec();
1493
1494         bprm->file = file;
1495         bprm->filename = bprm->interp = filename->name;
1496
1497         retval = bprm_mm_init(bprm);
1498         if (retval)
1499                 goto out_unmark;
1500
1501         bprm->argc = count(argv, MAX_ARG_STRINGS);
1502         if ((retval = bprm->argc) < 0)
1503                 goto out;
1504
1505         bprm->envc = count(envp, MAX_ARG_STRINGS);
1506         if ((retval = bprm->envc) < 0)
1507                 goto out;
1508
1509         retval = prepare_binprm(bprm);
1510         if (retval < 0)
1511                 goto out;
1512
1513         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1514         if (retval < 0)
1515                 goto out;
1516
1517         bprm->exec = bprm->p;
1518         retval = copy_strings(bprm->envc, envp, bprm);
1519         if (retval < 0)
1520                 goto out;
1521
1522         retval = copy_strings(bprm->argc, argv, bprm);
1523         if (retval < 0)
1524                 goto out;
1525
1526         retval = exec_binprm(bprm);
1527         if (retval < 0)
1528                 goto out;
1529
1530         /* execve succeeded */
1531         current->fs->in_exec = 0;
1532         current->in_execve = 0;
1533         acct_update_integrals(current);
1534         task_numa_free(current);
1535         free_bprm(bprm);
1536         putname(filename);
1537         if (displaced)
1538                 put_files_struct(displaced);
1539         return retval;
1540
1541 out:
1542         if (bprm->mm) {
1543                 acct_arg_size(bprm, 0);
1544                 mmput(bprm->mm);
1545         }
1546
1547 out_unmark:
1548         current->fs->in_exec = 0;
1549         current->in_execve = 0;
1550
1551 out_free:
1552         free_bprm(bprm);
1553
1554 out_files:
1555         if (displaced)
1556                 reset_files_struct(displaced);
1557 out_ret:
1558         putname(filename);
1559         return retval;
1560 }
1561
1562 int do_execve(struct filename *filename,
1563         const char __user *const __user *__argv,
1564         const char __user *const __user *__envp)
1565 {
1566         struct user_arg_ptr argv = { .ptr.native = __argv };
1567         struct user_arg_ptr envp = { .ptr.native = __envp };
1568         return do_execve_common(filename, argv, envp);
1569 }
1570
1571 #ifdef CONFIG_COMPAT
1572 static int compat_do_execve(struct filename *filename,
1573         const compat_uptr_t __user *__argv,
1574         const compat_uptr_t __user *__envp)
1575 {
1576         struct user_arg_ptr argv = {
1577                 .is_compat = true,
1578                 .ptr.compat = __argv,
1579         };
1580         struct user_arg_ptr envp = {
1581                 .is_compat = true,
1582                 .ptr.compat = __envp,
1583         };
1584         return do_execve_common(filename, argv, envp);
1585 }
1586 #endif
1587
1588 void set_binfmt(struct linux_binfmt *new)
1589 {
1590         struct mm_struct *mm = current->mm;
1591
1592         if (mm->binfmt)
1593                 module_put(mm->binfmt->module);
1594
1595         mm->binfmt = new;
1596         if (new)
1597                 __module_get(new->module);
1598 }
1599 EXPORT_SYMBOL(set_binfmt);
1600
1601 /*
1602  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1603  */
1604 void set_dumpable(struct mm_struct *mm, int value)
1605 {
1606         unsigned long old, new;
1607
1608         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1609                 return;
1610
1611         do {
1612                 old = ACCESS_ONCE(mm->flags);
1613                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1614         } while (cmpxchg(&mm->flags, old, new) != old);
1615 }
1616
1617 SYSCALL_DEFINE3(execve,
1618                 const char __user *, filename,
1619                 const char __user *const __user *, argv,
1620                 const char __user *const __user *, envp)
1621 {
1622         return do_execve(getname(filename), argv, envp);
1623 }
1624 #ifdef CONFIG_COMPAT
1625 asmlinkage long compat_sys_execve(const char __user * filename,
1626         const compat_uptr_t __user * argv,
1627         const compat_uptr_t __user * envp)
1628 {
1629         return compat_do_execve(getname(filename), argv, envp);
1630 }
1631 #endif