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/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/swap.h>
  32 #include <linux/string.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/mount.h>
  46 #include <linux/security.h>
  47 #include <linux/syscalls.h>
  48 #include <linux/tsacct_kern.h>
  49 #include <linux/cn_proc.h>
  50 #include <linux/audit.h>
  51 #include <linux/tracehook.h>
  52 #include <linux/kmod.h>
  53 #include <linux/fsnotify.h>
  54 #include <linux/fs_struct.h>
  55 #include <linux/pipe_fs_i.h>
  56 #include <linux/oom.h>
  57 #include <linux/compat.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62
  63 #include <trace/events/task.h>
  64 #include "internal.h"
  65
  66 #include <trace/events/sched.h>
  67
  68 int suid_dumpable = 0;
  69
  70 static LIST_HEAD(formats);
  71 static DEFINE_RWLOCK(binfmt_lock);
  72
  73 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  74 {
  75         BUG_ON(!fmt);
  76         if (WARN_ON(!fmt->load_binary))
  77                 return;
  78         write_lock(&binfmt_lock);
  79         insert ? list_add(&fmt->lh, &formats) :
  80                  list_add_tail(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82 }
  83
  84 EXPORT_SYMBOL(__register_binfmt);
  85
  86 void unregister_binfmt(struct linux_binfmt * fmt)
  87 {
  88         write_lock(&binfmt_lock);
  89         list_del(&fmt->lh);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(unregister_binfmt);
  94
  95 static inline void put_binfmt(struct linux_binfmt * fmt)
  96 {
  97         module_put(fmt->module);
  98 }
  99
 100 /*
 101  * Note that a shared library must be both readable and executable due to
 102  * security reasons.
 103  *
 104  * Also note that we take the address to load from from the file itself.
 105  */
 106 SYSCALL_DEFINE1(uselib, const char __user *, library)
 107 {
 108         struct linux_binfmt *fmt;
 109         struct file *file;
 110         struct filename *tmp = getname(library);
 111         int error = PTR_ERR(tmp);
 112         static const struct open_flags uselib_flags = {
 113                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 114                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 115                 .intent = LOOKUP_OPEN,
 116                 .lookup_flags = LOOKUP_FOLLOW,
 117         };
 118
 119         if (IS_ERR(tmp))
 120                 goto out;
 121
 122         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 123         putname(tmp);
 124         error = PTR_ERR(file);
 125         if (IS_ERR(file))
 126                 goto out;
 127
 128         error = -EINVAL;
 129         if (!S_ISREG(file_inode(file)->i_mode))
 130                 goto exit;
 131
 132         error = -EACCES;
 133         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 134                 goto exit;
 135
 136         fsnotify_open(file);
 137
 138         error = -ENOEXEC;
 139
 140         read_lock(&binfmt_lock);
 141         list_for_each_entry(fmt, &formats, lh) {
 142                 if (!fmt->load_shlib)
 143                         continue;
 144                 if (!try_module_get(fmt->module))
 145                         continue;
 146                 read_unlock(&binfmt_lock);
 147                 error = fmt->load_shlib(file);
 148                 read_lock(&binfmt_lock);
 149                 put_binfmt(fmt);
 150                 if (error != -ENOEXEC)
 151                         break;
 152         }
 153         read_unlock(&binfmt_lock);
 154 exit:
 155         fput(file);
 156 out:
 157         return error;
 158 }
 159
 160 #ifdef CONFIG_MMU
 161 /*
 162  * The nascent bprm->mm is not visible until exec_mmap() but it can
 163  * use a lot of memory, account these pages in current->mm temporary
 164  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 165  * change the counter back via acct_arg_size(0).
 166  */
 167 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 168 {
 169         struct mm_struct *mm = current->mm;
 170         long diff = (long)(pages - bprm->vma_pages);
 171
 172         if (!mm || !diff)
 173                 return;
 174
 175         bprm->vma_pages = pages;
 176         add_mm_counter(mm, MM_ANONPAGES, diff);
 177 }
 178
 179 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 180                 int write)
 181 {
 182         struct page *page;
 183         int ret;
 184
 185 #ifdef CONFIG_STACK_GROWSUP
 186         if (write) {
 187                 ret = expand_downwards(bprm->vma, pos);
 188                 if (ret < 0)
 189                         return NULL;
 190         }
 191 #endif
 192         ret = get_user_pages(current, bprm->mm, pos,
 193                         1, write, 1, &page, NULL);
 194         if (ret <= 0)
 195                 return NULL;
 196
 197         if (write) {
 198                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 199                 struct rlimit *rlim;
 200
 201                 acct_arg_size(bprm, size / PAGE_SIZE);
 202
 203                 /*
 204                  * We've historically supported up to 32 pages (ARG_MAX)
 205                  * of argument strings even with small stacks
 206                  */
 207                 if (size <= ARG_MAX)
 208                         return page;
 209
 210                 /*
 211                  * Limit to 1/4-th the stack size for the argv+env strings.
 212                  * This ensures that:
 213                  *  - the remaining binfmt code will not run out of stack space,
 214                  *  - the program will have a reasonable amount of stack left
 215                  *    to work from.
 216                  */
 217                 rlim = current->signal->rlim;
 218                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 219                         put_page(page);
 220                         return NULL;
 221                 }
 222         }
 223
 224         return page;
 225 }
 226
 227 static void put_arg_page(struct page *page)
 228 {
 229         put_page(page);
 230 }
 231
 232 static void free_arg_page(struct linux_binprm *bprm, int i)
 233 {
 234 }
 235
 236 static void free_arg_pages(struct linux_binprm *bprm)
 237 {
 238 }
 239
 240 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 241                 struct page *page)
 242 {
 243         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 244 }
 245
 246 static int __bprm_mm_init(struct linux_binprm *bprm)
 247 {
 248         int err;
 249         struct vm_area_struct *vma = NULL;
 250         struct mm_struct *mm = bprm->mm;
 251
 252         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 253         if (!vma)
 254                 return -ENOMEM;
 255
 256         down_write(&mm->mmap_sem);
 257         vma->vm_mm = mm;
 258
 259         /*
 260          * Place the stack at the largest stack address the architecture
 261          * supports. Later, we'll move this to an appropriate place. We don't
 262          * use STACK_TOP because that can depend on attributes which aren't
 263          * configured yet.
 264          */
 265         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 266         vma->vm_end = STACK_TOP_MAX;
 267         vma->vm_start = vma->vm_end - PAGE_SIZE;
 268         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 269         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 270         INIT_LIST_HEAD(&vma->anon_vma_chain);
 271
 272         err = insert_vm_struct(mm, vma);
 273         if (err)
 274                 goto err;
 275
 276         mm->stack_vm = mm->total_vm = 1;
 277         up_write(&mm->mmap_sem);
 278         bprm->p = vma->vm_end - sizeof(void *);
 279         return 0;
 280 err:
 281         up_write(&mm->mmap_sem);
 282         bprm->vma = NULL;
 283         kmem_cache_free(vm_area_cachep, vma);
 284         return err;
 285 }
 286
 287 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 288 {
 289         return len <= MAX_ARG_STRLEN;
 290 }
 291
 292 #else
 293
 294 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 295 {
 296 }
 297
 298 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 299                 int write)
 300 {
 301         struct page *page;
 302
 303         page = bprm->page[pos / PAGE_SIZE];
 304         if (!page && write) {
 305                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 306                 if (!page)
 307                         return NULL;
 308                 bprm->page[pos / PAGE_SIZE] = page;
 309         }
 310
 311         return page;
 312 }
 313
 314 static void put_arg_page(struct page *page)
 315 {
 316 }
 317
 318 static void free_arg_page(struct linux_binprm *bprm, int i)
 319 {
 320         if (bprm->page[i]) {
 321                 __free_page(bprm->page[i]);
 322                 bprm->page[i] = NULL;
 323         }
 324 }
 325
 326 static void free_arg_pages(struct linux_binprm *bprm)
 327 {
 328         int i;
 329
 330         for (i = 0; i < MAX_ARG_PAGES; i++)
 331                 free_arg_page(bprm, i);
 332 }
 333
 334 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 335                 struct page *page)
 336 {
 337 }
 338
 339 static int __bprm_mm_init(struct linux_binprm *bprm)
 340 {
 341         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 342         return 0;
 343 }
 344
 345 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 346 {
 347         return len <= bprm->p;
 348 }
 349
 350 #endif /* CONFIG_MMU */
 351
 352 /*
 353  * Create a new mm_struct and populate it with a temporary stack
 354  * vm_area_struct.  We don't have enough context at this point to set the stack
 355  * flags, permissions, and offset, so we use temporary values.  We'll update
 356  * them later in setup_arg_pages().
 357  */
 358 static int bprm_mm_init(struct linux_binprm *bprm)
 359 {
 360         int err;
 361         struct mm_struct *mm = NULL;
 362
 363         bprm->mm = mm = mm_alloc();
 364         err = -ENOMEM;
 365         if (!mm)
 366                 goto err;
 367
 368         err = init_new_context(current, mm);
 369         if (err)
 370                 goto err;
 371
 372         err = __bprm_mm_init(bprm);
 373         if (err)
 374                 goto err;
 375
 376         return 0;
 377
 378 err:
 379         if (mm) {
 380                 bprm->mm = NULL;
 381                 mmdrop(mm);
 382         }
 383
 384         return err;
 385 }
 386
 387 struct user_arg_ptr {
 388 #ifdef CONFIG_COMPAT
 389         bool is_compat;
 390 #endif
 391         union {
 392                 const char __user *const __user *native;
 393 #ifdef CONFIG_COMPAT
 394                 const compat_uptr_t __user *compat;
 395 #endif
 396         } ptr;
 397 };
 398
 399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 400 {
 401         const char __user *native;
 402
 403 #ifdef CONFIG_COMPAT
 404         if (unlikely(argv.is_compat)) {
 405                 compat_uptr_t compat;
 406
 407                 if (get_user(compat, argv.ptr.compat + nr))
 408                         return ERR_PTR(-EFAULT);
 409
 410                 return compat_ptr(compat);
 411         }
 412 #endif
 413
 414         if (get_user(native, argv.ptr.native + nr))
 415                 return ERR_PTR(-EFAULT);
 416
 417         return native;
 418 }
 419
 420 /*
 421  * count() counts the number of strings in array ARGV.
 422  */
 423 static int count(struct user_arg_ptr argv, int max)
 424 {
 425         int i = 0;
 426
 427         if (argv.ptr.native != NULL) {
 428                 for (;;) {
 429                         const char __user *p = get_user_arg_ptr(argv, i);
 430
 431                         if (!p)
 432                                 break;
 433
 434                         if (IS_ERR(p))
 435                                 return -EFAULT;
 436
 437                         if (i >= max)
 438                                 return -E2BIG;
 439                         ++i;
 440
 441                         if (fatal_signal_pending(current))
 442                                 return -ERESTARTNOHAND;
 443                         cond_resched();
 444                 }
 445         }
 446         return i;
 447 }
 448
 449 /*
 450  * 'copy_strings()' copies argument/environment strings from the old
 451  * processes's memory to the new process's stack.  The call to get_user_pages()
 452  * ensures the destination page is created and not swapped out.
 453  */
 454 static int copy_strings(int argc, struct user_arg_ptr argv,
 455                         struct linux_binprm *bprm)
 456 {
 457         struct page *kmapped_page = NULL;
 458         char *kaddr = NULL;
 459         unsigned long kpos = 0;
 460         int ret;
 461
 462         while (argc-- > 0) {
 463                 const char __user *str;
 464                 int len;
 465                 unsigned long pos;
 466
 467                 ret = -EFAULT;
 468                 str = get_user_arg_ptr(argv, argc);
 469                 if (IS_ERR(str))
 470                         goto out;
 471
 472                 len = strnlen_user(str, MAX_ARG_STRLEN);
 473                 if (!len)
 474                         goto out;
 475
 476                 ret = -E2BIG;
 477                 if (!valid_arg_len(bprm, len))
 478                         goto out;
 479
 480                 /* We're going to work our way backwords. */
 481                 pos = bprm->p;
 482                 str += len;
 483                 bprm->p -= len;
 484
 485                 while (len > 0) {
 486                         int offset, bytes_to_copy;
 487
 488                         if (fatal_signal_pending(current)) {
 489                                 ret = -ERESTARTNOHAND;
 490                                 goto out;
 491                         }
 492                         cond_resched();
 493
 494                         offset = pos % PAGE_SIZE;
 495                         if (offset == 0)
 496                                 offset = PAGE_SIZE;
 497
 498                         bytes_to_copy = offset;
 499                         if (bytes_to_copy > len)
 500                                 bytes_to_copy = len;
 501
 502                         offset -= bytes_to_copy;
 503                         pos -= bytes_to_copy;
 504                         str -= bytes_to_copy;
 505                         len -= bytes_to_copy;
 506
 507                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 508                                 struct page *page;
 509
 510                                 page = get_arg_page(bprm, pos, 1);
 511                                 if (!page) {
 512                                         ret = -E2BIG;
 513                                         goto out;
 514                                 }
 515
 516                                 if (kmapped_page) {
 517                                         flush_kernel_dcache_page(kmapped_page);
 518                                         kunmap(kmapped_page);
 519                                         put_arg_page(kmapped_page);
 520                                 }
 521                                 kmapped_page = page;
 522                                 kaddr = kmap(kmapped_page);
 523                                 kpos = pos & PAGE_MASK;
 524                                 flush_arg_page(bprm, kpos, kmapped_page);
 525                         }
 526                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 527                                 ret = -EFAULT;
 528                                 goto out;
 529                         }
 530                 }
 531         }
 532         ret = 0;
 533 out:
 534         if (kmapped_page) {
 535                 flush_kernel_dcache_page(kmapped_page);
 536                 kunmap(kmapped_page);
 537                 put_arg_page(kmapped_page);
 538         }
 539         return ret;
 540 }
 541
 542 /*
 543  * Like copy_strings, but get argv and its values from kernel memory.
 544  */
 545 int copy_strings_kernel(int argc, const char *const *__argv,
 546                         struct linux_binprm *bprm)
 547 {
 548         int r;
 549         mm_segment_t oldfs = get_fs();
 550         struct user_arg_ptr argv = {
 551                 .ptr.native = (const char __user *const  __user *)__argv,
 552         };
 553
 554         set_fs(KERNEL_DS);
 555         r = copy_strings(argc, argv, bprm);
 556         set_fs(oldfs);
 557
 558         return r;
 559 }
 560 EXPORT_SYMBOL(copy_strings_kernel);
 561
 562 #ifdef CONFIG_MMU
 563
 564 /*
 565  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 566  * the binfmt code determines where the new stack should reside, we shift it to
 567  * its final location.  The process proceeds as follows:
 568  *
 569  * 1) Use shift to calculate the new vma endpoints.
 570  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 571  *    arguments passed to subsequent functions are consistent.
 572  * 3) Move vma's page tables to the new range.
 573  * 4) Free up any cleared pgd range.
 574  * 5) Shrink the vma to cover only the new range.
 575  */
 576 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 577 {
 578         struct mm_struct *mm = vma->vm_mm;
 579         unsigned long old_start = vma->vm_start;
 580         unsigned long old_end = vma->vm_end;
 581         unsigned long length = old_end - old_start;
 582         unsigned long new_start = old_start - shift;
 583         unsigned long new_end = old_end - shift;
 584         struct mmu_gather tlb;
 585
 586         BUG_ON(new_start > new_end);
 587
 588         /*
 589          * ensure there are no vmas between where we want to go
 590          * and where we are
 591          */
 592         if (vma != find_vma(mm, new_start))
 593                 return -EFAULT;
 594
 595         /*
 596          * cover the whole range: [new_start, old_end)
 597          */
 598         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 599                 return -ENOMEM;
 600
 601         /*
 602          * move the page tables downwards, on failure we rely on
 603          * process cleanup to remove whatever mess we made.
 604          */
 605         if (length != move_page_tables(vma, old_start,
 606                                        vma, new_start, length, false))
 607                 return -ENOMEM;
 608
 609         lru_add_drain();
 610         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 611         if (new_end > old_start) {
 612                 /*
 613                  * when the old and new regions overlap clear from new_end.
 614                  */
 615                 free_pgd_range(&tlb, new_end, old_end, new_end,
 616                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 617         } else {
 618                 /*
 619                  * otherwise, clean from old_start; this is done to not touch
 620                  * the address space in [new_end, old_start) some architectures
 621                  * have constraints on va-space that make this illegal (IA64) -
 622                  * for the others its just a little faster.
 623                  */
 624                 free_pgd_range(&tlb, old_start, old_end, new_end,
 625                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 626         }
 627         tlb_finish_mmu(&tlb, old_start, old_end);
 628
 629         /*
 630          * Shrink the vma to just the new range.  Always succeeds.
 631          */
 632         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 633
 634         return 0;
 635 }
 636
 637 /*
 638  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 639  * the stack is optionally relocated, and some extra space is added.
 640  */
 641 int setup_arg_pages(struct linux_binprm *bprm,
 642                     unsigned long stack_top,
 643                     int executable_stack)
 644 {
 645         unsigned long ret;
 646         unsigned long stack_shift;
 647         struct mm_struct *mm = current->mm;
 648         struct vm_area_struct *vma = bprm->vma;
 649         struct vm_area_struct *prev = NULL;
 650         unsigned long vm_flags;
 651         unsigned long stack_base;
 652         unsigned long stack_size;
 653         unsigned long stack_expand;
 654         unsigned long rlim_stack;
 655
 656 #ifdef CONFIG_STACK_GROWSUP
 657         /* Limit stack size to 1GB */
 658         stack_base = rlimit_max(RLIMIT_STACK);
 659         if (stack_base > (1 << 30))
 660                 stack_base = 1 << 30;
 661
 662         /* Make sure we didn't let the argument array grow too large. */
 663         if (vma->vm_end - vma->vm_start > stack_base)
 664                 return -ENOMEM;
 665
 666         stack_base = PAGE_ALIGN(stack_top - stack_base);
 667
 668         stack_shift = vma->vm_start - stack_base;
 669         mm->arg_start = bprm->p - stack_shift;
 670         bprm->p = vma->vm_end - stack_shift;
 671 #else
 672         stack_top = arch_align_stack(stack_top);
 673         stack_top = PAGE_ALIGN(stack_top);
 674
 675         if (unlikely(stack_top < mmap_min_addr) ||
 676             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 677                 return -ENOMEM;
 678
 679         stack_shift = vma->vm_end - stack_top;
 680
 681         bprm->p -= stack_shift;
 682         mm->arg_start = bprm->p;
 683 #endif
 684
 685         if (bprm->loader)
 686                 bprm->loader -= stack_shift;
 687         bprm->exec -= stack_shift;
 688
 689         down_write(&mm->mmap_sem);
 690         vm_flags = VM_STACK_FLAGS;
 691
 692         /*
 693          * Adjust stack execute permissions; explicitly enable for
 694          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 695          * (arch default) otherwise.
 696          */
 697         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 698                 vm_flags |= VM_EXEC;
 699         else if (executable_stack == EXSTACK_DISABLE_X)
 700                 vm_flags &= ~VM_EXEC;
 701         vm_flags |= mm->def_flags;
 702         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 703
 704         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 705                         vm_flags);
 706         if (ret)
 707                 goto out_unlock;
 708         BUG_ON(prev != vma);
 709
 710         /* Move stack pages down in memory. */
 711         if (stack_shift) {
 712                 ret = shift_arg_pages(vma, stack_shift);
 713                 if (ret)
 714                         goto out_unlock;
 715         }
 716
 717         /* mprotect_fixup is overkill to remove the temporary stack flags */
 718         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 719
 720         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 721         stack_size = vma->vm_end - vma->vm_start;
 722         /*
 723          * Align this down to a page boundary as expand_stack
 724          * will align it up.
 725          */
 726         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 727 #ifdef CONFIG_STACK_GROWSUP
 728         if (stack_size + stack_expand > rlim_stack)
 729                 stack_base = vma->vm_start + rlim_stack;
 730         else
 731                 stack_base = vma->vm_end + stack_expand;
 732 #else
 733         if (stack_size + stack_expand > rlim_stack)
 734                 stack_base = vma->vm_end - rlim_stack;
 735         else
 736                 stack_base = vma->vm_start - stack_expand;
 737 #endif
 738         current->mm->start_stack = bprm->p;
 739         ret = expand_stack(vma, stack_base);
 740         if (ret)
 741                 ret = -EFAULT;
 742
 743 out_unlock:
 744         up_write(&mm->mmap_sem);
 745         return ret;
 746 }
 747 EXPORT_SYMBOL(setup_arg_pages);
 748
 749 #endif /* CONFIG_MMU */
 750
 751 static struct file *do_open_exec(struct filename *name)
 752 {
 753         struct file *file;
 754         int err;
 755         static const struct open_flags open_exec_flags = {
 756                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 757                 .acc_mode = MAY_EXEC | MAY_OPEN,
 758                 .intent = LOOKUP_OPEN,
 759                 .lookup_flags = LOOKUP_FOLLOW,
 760         };
 761
 762         file = do_filp_open(AT_FDCWD, name, &open_exec_flags);
 763         if (IS_ERR(file))
 764                 goto out;
 765
 766         err = -EACCES;
 767         if (!S_ISREG(file_inode(file)->i_mode))
 768                 goto exit;
 769
 770         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 771                 goto exit;
 772
 773         fsnotify_open(file);
 774
 775         err = deny_write_access(file);
 776         if (err)
 777                 goto exit;
 778
 779 out:
 780         return file;
 781
 782 exit:
 783         fput(file);
 784         return ERR_PTR(err);
 785 }
 786
 787 struct file *open_exec(const char *name)
 788 {
 789         struct filename tmp = { .name = name };
 790         return do_open_exec(&tmp);
 791 }
 792 EXPORT_SYMBOL(open_exec);
 793
 794 int kernel_read(struct file *file, loff_t offset,
 795                 char *addr, unsigned long count)
 796 {
 797         mm_segment_t old_fs;
 798         loff_t pos = offset;
 799         int result;
 800
 801         old_fs = get_fs();
 802         set_fs(get_ds());
 803         /* The cast to a user pointer is valid due to the set_fs() */
 804         result = vfs_read(file, (void __user *)addr, count, &pos);
 805         set_fs(old_fs);
 806         return result;
 807 }
 808
 809 EXPORT_SYMBOL(kernel_read);
 810
 811 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 812 {
 813         ssize_t res = file->f_op->read(file, (void __user *)addr, len, &pos);
 814         if (res > 0)
 815                 flush_icache_range(addr, addr + len);
 816         return res;
 817 }
 818 EXPORT_SYMBOL(read_code);
 819
 820 static int exec_mmap(struct mm_struct *mm)
 821 {
 822         struct task_struct *tsk;
 823         struct mm_struct * old_mm, *active_mm;
 824
 825         /* Notify parent that we're no longer interested in the old VM */
 826         tsk = current;
 827         old_mm = current->mm;
 828         mm_release(tsk, old_mm);
 829
 830         if (old_mm) {
 831                 sync_mm_rss(old_mm);
 832                 /*
 833                  * Make sure that if there is a core dump in progress
 834                  * for the old mm, we get out and die instead of going
 835                  * through with the exec.  We must hold mmap_sem around
 836                  * checking core_state and changing tsk->mm.
 837                  */
 838                 down_read(&old_mm->mmap_sem);
 839                 if (unlikely(old_mm->core_state)) {
 840                         up_read(&old_mm->mmap_sem);
 841                         return -EINTR;
 842                 }
 843         }
 844         task_lock(tsk);
 845         active_mm = tsk->active_mm;
 846         tsk->mm = mm;
 847         tsk->active_mm = mm;
 848         activate_mm(active_mm, mm);
 849         task_unlock(tsk);
 850         if (old_mm) {
 851                 up_read(&old_mm->mmap_sem);
 852                 BUG_ON(active_mm != old_mm);
 853                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 854                 mm_update_next_owner(old_mm);
 855                 mmput(old_mm);
 856                 return 0;
 857         }
 858         mmdrop(active_mm);
 859         return 0;
 860 }
 861
 862 /*
 863  * This function makes sure the current process has its own signal table,
 864  * so that flush_signal_handlers can later reset the handlers without
 865  * disturbing other processes.  (Other processes might share the signal
 866  * table via the CLONE_SIGHAND option to clone().)
 867  */
 868 static int de_thread(struct task_struct *tsk)
 869 {
 870         struct signal_struct *sig = tsk->signal;
 871         struct sighand_struct *oldsighand = tsk->sighand;
 872         spinlock_t *lock = &oldsighand->siglock;
 873
 874         if (thread_group_empty(tsk))
 875                 goto no_thread_group;
 876
 877         /*
 878          * Kill all other threads in the thread group.
 879          */
 880         spin_lock_irq(lock);
 881         if (signal_group_exit(sig)) {
 882                 /*
 883                  * Another group action in progress, just
 884                  * return so that the signal is processed.
 885                  */
 886                 spin_unlock_irq(lock);
 887                 return -EAGAIN;
 888         }
 889
 890         sig->group_exit_task = tsk;
 891         sig->notify_count = zap_other_threads(tsk);
 892         if (!thread_group_leader(tsk))
 893                 sig->notify_count--;
 894
 895         while (sig->notify_count) {
 896                 __set_current_state(TASK_KILLABLE);
 897                 spin_unlock_irq(lock);
 898                 schedule();
 899                 if (unlikely(__fatal_signal_pending(tsk)))
 900                         goto killed;
 901                 spin_lock_irq(lock);
 902         }
 903         spin_unlock_irq(lock);
 904
 905         /*
 906          * At this point all other threads have exited, all we have to
 907          * do is to wait for the thread group leader to become inactive,
 908          * and to assume its PID:
 909          */
 910         if (!thread_group_leader(tsk)) {
 911                 struct task_struct *leader = tsk->group_leader;
 912
 913                 sig->notify_count = -1; /* for exit_notify() */
 914                 for (;;) {
 915                         threadgroup_change_begin(tsk);
 916                         write_lock_irq(&tasklist_lock);
 917                         if (likely(leader->exit_state))
 918                                 break;
 919                         __set_current_state(TASK_KILLABLE);
 920                         write_unlock_irq(&tasklist_lock);
 921                         threadgroup_change_end(tsk);
 922                         schedule();
 923                         if (unlikely(__fatal_signal_pending(tsk)))
 924                                 goto killed;
 925                 }
 926
 927                 /*
 928                  * The only record we have of the real-time age of a
 929                  * process, regardless of execs it's done, is start_time.
 930                  * All the past CPU time is accumulated in signal_struct
 931                  * from sister threads now dead.  But in this non-leader
 932                  * exec, nothing survives from the original leader thread,
 933                  * whose birth marks the true age of this process now.
 934                  * When we take on its identity by switching to its PID, we
 935                  * also take its birthdate (always earlier than our own).
 936                  */
 937                 tsk->start_time = leader->start_time;
 938                 tsk->real_start_time = leader->real_start_time;
 939
 940                 BUG_ON(!same_thread_group(leader, tsk));
 941                 BUG_ON(has_group_leader_pid(tsk));
 942                 /*
 943                  * An exec() starts a new thread group with the
 944                  * TGID of the previous thread group. Rehash the
 945                  * two threads with a switched PID, and release
 946                  * the former thread group leader:
 947                  */
 948
 949                 /* Become a process group leader with the old leader's pid.
 950                  * The old leader becomes a thread of the this thread group.
 951                  * Note: The old leader also uses this pid until release_task
 952                  *       is called.  Odd but simple and correct.
 953                  */
 954                 tsk->pid = leader->pid;
 955                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
 956                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 957                 transfer_pid(leader, tsk, PIDTYPE_SID);
 958
 959                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 960                 list_replace_init(&leader->sibling, &tsk->sibling);
 961
 962                 tsk->group_leader = tsk;
 963                 leader->group_leader = tsk;
 964
 965                 tsk->exit_signal = SIGCHLD;
 966                 leader->exit_signal = -1;
 967
 968                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 969                 leader->exit_state = EXIT_DEAD;
 970
 971                 /*
 972                  * We are going to release_task()->ptrace_unlink() silently,
 973                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 974                  * the tracer wont't block again waiting for this thread.
 975                  */
 976                 if (unlikely(leader->ptrace))
 977                         __wake_up_parent(leader, leader->parent);
 978                 write_unlock_irq(&tasklist_lock);
 979                 threadgroup_change_end(tsk);
 980
 981                 release_task(leader);
 982         }
 983
 984         sig->group_exit_task = NULL;
 985         sig->notify_count = 0;
 986
 987 no_thread_group:
 988         /* we have changed execution domain */
 989         tsk->exit_signal = SIGCHLD;
 990
 991         exit_itimers(sig);
 992         flush_itimer_signals();
 993
 994         if (atomic_read(&oldsighand->count) != 1) {
 995                 struct sighand_struct *newsighand;
 996                 /*
 997                  * This ->sighand is shared with the CLONE_SIGHAND
 998                  * but not CLONE_THREAD task, switch to the new one.
 999                  */
1000                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1001                 if (!newsighand)
1002                         return -ENOMEM;
1003
1004                 atomic_set(&newsighand->count, 1);
1005                 memcpy(newsighand->action, oldsighand->action,
1006                        sizeof(newsighand->action));
1007
1008                 write_lock_irq(&tasklist_lock);
1009                 spin_lock(&oldsighand->siglock);
1010                 rcu_assign_pointer(tsk->sighand, newsighand);
1011                 spin_unlock(&oldsighand->siglock);
1012                 write_unlock_irq(&tasklist_lock);
1013
1014                 __cleanup_sighand(oldsighand);
1015         }
1016
1017         BUG_ON(!thread_group_leader(tsk));
1018         return 0;
1019
1020 killed:
1021         /* protects against exit_notify() and __exit_signal() */
1022         read_lock(&tasklist_lock);
1023         sig->group_exit_task = NULL;
1024         sig->notify_count = 0;
1025         read_unlock(&tasklist_lock);
1026         return -EAGAIN;
1027 }
1028
1029 char *get_task_comm(char *buf, struct task_struct *tsk)
1030 {
1031         /* buf must be at least sizeof(tsk->comm) in size */
1032         task_lock(tsk);
1033         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1034         task_unlock(tsk);
1035         return buf;
1036 }
1037 EXPORT_SYMBOL_GPL(get_task_comm);
1038
1039 /*
1040  * These functions flushes out all traces of the currently running executable
1041  * so that a new one can be started
1042  */
1043
1044 void set_task_comm(struct task_struct *tsk, char *buf)
1045 {
1046         task_lock(tsk);
1047         trace_task_rename(tsk, buf);
1048         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1049         task_unlock(tsk);
1050         perf_event_comm(tsk);
1051 }
1052
1053 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1054 {
1055         int i, ch;
1056
1057         /* Copies the binary name from after last slash */
1058         for (i = 0; (ch = *(fn++)) != '\0';) {
1059                 if (ch == '/')
1060                         i = 0; /* overwrite what we wrote */
1061                 else
1062                         if (i < len - 1)
1063                                 tcomm[i++] = ch;
1064         }
1065         tcomm[i] = '\0';
1066 }
1067
1068 int flush_old_exec(struct linux_binprm * bprm)
1069 {
1070         int retval;
1071
1072         /*
1073          * Make sure we have a private signal table and that
1074          * we are unassociated from the previous thread group.
1075          */
1076         retval = de_thread(current);
1077         if (retval)
1078                 goto out;
1079
1080         set_mm_exe_file(bprm->mm, bprm->file);
1081
1082         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1083         /*
1084          * Release all of the old mmap stuff
1085          */
1086         acct_arg_size(bprm, 0);
1087         retval = exec_mmap(bprm->mm);
1088         if (retval)
1089                 goto out;
1090
1091         bprm->mm = NULL;                /* We're using it now */
1092
1093         set_fs(USER_DS);
1094         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1095                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1096         flush_thread();
1097         current->personality &= ~bprm->per_clear;
1098
1099         return 0;
1100
1101 out:
1102         return retval;
1103 }
1104 EXPORT_SYMBOL(flush_old_exec);
1105
1106 void would_dump(struct linux_binprm *bprm, struct file *file)
1107 {
1108         if (inode_permission(file_inode(file), MAY_READ) < 0)
1109                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1110 }
1111 EXPORT_SYMBOL(would_dump);
1112
1113 void setup_new_exec(struct linux_binprm * bprm)
1114 {
1115         arch_pick_mmap_layout(current->mm);
1116
1117         /* This is the point of no return */
1118         current->sas_ss_sp = current->sas_ss_size = 0;
1119
1120         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1121                 set_dumpable(current->mm, SUID_DUMP_USER);
1122         else
1123                 set_dumpable(current->mm, suid_dumpable);
1124
1125         set_task_comm(current, bprm->tcomm);
1126
1127         /* Set the new mm task size. We have to do that late because it may
1128          * depend on TIF_32BIT which is only updated in flush_thread() on
1129          * some architectures like powerpc
1130          */
1131         current->mm->task_size = TASK_SIZE;
1132
1133         /* install the new credentials */
1134         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1135             !gid_eq(bprm->cred->gid, current_egid())) {
1136                 current->pdeath_signal = 0;
1137         } else {
1138                 would_dump(bprm, bprm->file);
1139                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1140                         set_dumpable(current->mm, suid_dumpable);
1141         }
1142
1143         /* An exec changes our domain. We are no longer part of the thread
1144            group */
1145         current->self_exec_id++;
1146         flush_signal_handlers(current, 0);
1147         do_close_on_exec(current->files);
1148 }
1149 EXPORT_SYMBOL(setup_new_exec);
1150
1151 /*
1152  * Prepare credentials and lock ->cred_guard_mutex.
1153  * install_exec_creds() commits the new creds and drops the lock.
1154  * Or, if exec fails before, free_bprm() should release ->cred and
1155  * and unlock.
1156  */
1157 int prepare_bprm_creds(struct linux_binprm *bprm)
1158 {
1159         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1160                 return -ERESTARTNOINTR;
1161
1162         bprm->cred = prepare_exec_creds();
1163         if (likely(bprm->cred))
1164                 return 0;
1165
1166         mutex_unlock(&current->signal->cred_guard_mutex);
1167         return -ENOMEM;
1168 }
1169
1170 static void free_bprm(struct linux_binprm *bprm)
1171 {
1172         free_arg_pages(bprm);
1173         if (bprm->cred) {
1174                 mutex_unlock(&current->signal->cred_guard_mutex);
1175                 abort_creds(bprm->cred);
1176         }
1177         if (bprm->file) {
1178                 allow_write_access(bprm->file);
1179                 fput(bprm->file);
1180         }
1181         /* If a binfmt changed the interp, free it. */
1182         if (bprm->interp != bprm->filename)
1183                 kfree(bprm->interp);
1184         kfree(bprm);
1185 }
1186
1187 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1188 {
1189         /* If a binfmt changed the interp, free it first. */
1190         if (bprm->interp != bprm->filename)
1191                 kfree(bprm->interp);
1192         bprm->interp = kstrdup(interp, GFP_KERNEL);
1193         if (!bprm->interp)
1194                 return -ENOMEM;
1195         return 0;
1196 }
1197 EXPORT_SYMBOL(bprm_change_interp);
1198
1199 /*
1200  * install the new credentials for this executable
1201  */
1202 void install_exec_creds(struct linux_binprm *bprm)
1203 {
1204         security_bprm_committing_creds(bprm);
1205
1206         commit_creds(bprm->cred);
1207         bprm->cred = NULL;
1208
1209         /*
1210          * Disable monitoring for regular users
1211          * when executing setuid binaries. Must
1212          * wait until new credentials are committed
1213          * by commit_creds() above
1214          */
1215         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1216                 perf_event_exit_task(current);
1217         /*
1218          * cred_guard_mutex must be held at least to this point to prevent
1219          * ptrace_attach() from altering our determination of the task's
1220          * credentials; any time after this it may be unlocked.
1221          */
1222         security_bprm_committed_creds(bprm);
1223         mutex_unlock(&current->signal->cred_guard_mutex);
1224 }
1225 EXPORT_SYMBOL(install_exec_creds);
1226
1227 /*
1228  * determine how safe it is to execute the proposed program
1229  * - the caller must hold ->cred_guard_mutex to protect against
1230  *   PTRACE_ATTACH
1231  */
1232 static void check_unsafe_exec(struct linux_binprm *bprm)
1233 {
1234         struct task_struct *p = current, *t;
1235         unsigned n_fs;
1236
1237         if (p->ptrace) {
1238                 if (p->ptrace & PT_PTRACE_CAP)
1239                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1240                 else
1241                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1242         }
1243
1244         /*
1245          * This isn't strictly necessary, but it makes it harder for LSMs to
1246          * mess up.
1247          */
1248         if (current->no_new_privs)
1249                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1250
1251         t = p;
1252         n_fs = 1;
1253         spin_lock(&p->fs->lock);
1254         rcu_read_lock();
1255         while_each_thread(p, t) {
1256                 if (t->fs == p->fs)
1257                         n_fs++;
1258         }
1259         rcu_read_unlock();
1260
1261         if (p->fs->users > n_fs)
1262                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1263         else
1264                 p->fs->in_exec = 1;
1265         spin_unlock(&p->fs->lock);
1266 }
1267
1268 /*
1269  * Fill the binprm structure from the inode.
1270  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1271  *
1272  * This may be called multiple times for binary chains (scripts for example).
1273  */
1274 int prepare_binprm(struct linux_binprm *bprm)
1275 {
1276         struct inode *inode = file_inode(bprm->file);
1277         umode_t mode = inode->i_mode;
1278         int retval;
1279
1280
1281         /* clear any previous set[ug]id data from a previous binary */
1282         bprm->cred->euid = current_euid();
1283         bprm->cred->egid = current_egid();
1284
1285         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1286             !current->no_new_privs &&
1287             kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1288             kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1289                 /* Set-uid? */
1290                 if (mode & S_ISUID) {
1291                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1292                         bprm->cred->euid = inode->i_uid;
1293                 }
1294
1295                 /* Set-gid? */
1296                 /*
1297                  * If setgid is set but no group execute bit then this
1298                  * is a candidate for mandatory locking, not a setgid
1299                  * executable.
1300                  */
1301                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1302                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1303                         bprm->cred->egid = inode->i_gid;
1304                 }
1305         }
1306
1307         /* fill in binprm security blob */
1308         retval = security_bprm_set_creds(bprm);
1309         if (retval)
1310                 return retval;
1311         bprm->cred_prepared = 1;
1312
1313         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1314         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1315 }
1316
1317 EXPORT_SYMBOL(prepare_binprm);
1318
1319 /*
1320  * Arguments are '\0' separated strings found at the location bprm->p
1321  * points to; chop off the first by relocating brpm->p to right after
1322  * the first '\0' encountered.
1323  */
1324 int remove_arg_zero(struct linux_binprm *bprm)
1325 {
1326         int ret = 0;
1327         unsigned long offset;
1328         char *kaddr;
1329         struct page *page;
1330
1331         if (!bprm->argc)
1332                 return 0;
1333
1334         do {
1335                 offset = bprm->p & ~PAGE_MASK;
1336                 page = get_arg_page(bprm, bprm->p, 0);
1337                 if (!page) {
1338                         ret = -EFAULT;
1339                         goto out;
1340                 }
1341                 kaddr = kmap_atomic(page);
1342
1343                 for (; offset < PAGE_SIZE && kaddr[offset];
1344                                 offset++, bprm->p++)
1345                         ;
1346
1347                 kunmap_atomic(kaddr);
1348                 put_arg_page(page);
1349
1350                 if (offset == PAGE_SIZE)
1351                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1352         } while (offset == PAGE_SIZE);
1353
1354         bprm->p++;
1355         bprm->argc--;
1356         ret = 0;
1357
1358 out:
1359         return ret;
1360 }
1361 EXPORT_SYMBOL(remove_arg_zero);
1362
1363 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1364 /*
1365  * cycle the list of binary formats handler, until one recognizes the image
1366  */
1367 int search_binary_handler(struct linux_binprm *bprm)
1368 {
1369         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1370         struct linux_binfmt *fmt;
1371         int retval;
1372
1373         /* This allows 4 levels of binfmt rewrites before failing hard. */
1374         if (bprm->recursion_depth > 5)
1375                 return -ELOOP;
1376
1377         retval = security_bprm_check(bprm);
1378         if (retval)
1379                 return retval;
1380
1381         retval = -ENOENT;
1382  retry:
1383         read_lock(&binfmt_lock);
1384         list_for_each_entry(fmt, &formats, lh) {
1385                 if (!try_module_get(fmt->module))
1386                         continue;
1387                 read_unlock(&binfmt_lock);
1388                 bprm->recursion_depth++;
1389                 retval = fmt->load_binary(bprm);
1390                 bprm->recursion_depth--;
1391                 if (retval >= 0 || retval != -ENOEXEC ||
1392                     bprm->mm == NULL || bprm->file == NULL) {
1393                         put_binfmt(fmt);
1394                         return retval;
1395                 }
1396                 read_lock(&binfmt_lock);
1397                 put_binfmt(fmt);
1398         }
1399         read_unlock(&binfmt_lock);
1400
1401         if (need_retry && retval == -ENOEXEC) {
1402                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1403                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1404                         return retval;
1405                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1406                         return retval;
1407                 need_retry = false;
1408                 goto retry;
1409         }
1410
1411         return retval;
1412 }
1413 EXPORT_SYMBOL(search_binary_handler);
1414
1415 static int exec_binprm(struct linux_binprm *bprm)
1416 {
1417         pid_t old_pid, old_vpid;
1418         int ret;
1419
1420         /* Need to fetch pid before load_binary changes it */
1421         old_pid = current->pid;
1422         rcu_read_lock();
1423         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1424         rcu_read_unlock();
1425
1426         ret = search_binary_handler(bprm);
1427         if (ret >= 0) {
1428                 audit_bprm(bprm);
1429                 trace_sched_process_exec(current, old_pid, bprm);
1430                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1431                 proc_exec_connector(current);
1432         }
1433
1434         return ret;
1435 }
1436
1437 /*
1438  * sys_execve() executes a new program.
1439  */
1440 static int do_execve_common(struct filename *filename,
1441                                 struct user_arg_ptr argv,
1442                                 struct user_arg_ptr envp)
1443 {
1444         struct linux_binprm *bprm;
1445         struct file *file;
1446         struct files_struct *displaced;
1447         int retval;
1448
1449         if (IS_ERR(filename))
1450                 return PTR_ERR(filename);
1451
1452         /*
1453          * We move the actual failure in case of RLIMIT_NPROC excess from
1454          * set*uid() to execve() because too many poorly written programs
1455          * don't check setuid() return code.  Here we additionally recheck
1456          * whether NPROC limit is still exceeded.
1457          */
1458         if ((current->flags & PF_NPROC_EXCEEDED) &&
1459             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1460                 retval = -EAGAIN;
1461                 goto out_ret;
1462         }
1463
1464         /* We're below the limit (still or again), so we don't want to make
1465          * further execve() calls fail. */
1466         current->flags &= ~PF_NPROC_EXCEEDED;
1467
1468         retval = unshare_files(&displaced);
1469         if (retval)
1470                 goto out_ret;
1471
1472         retval = -ENOMEM;
1473         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1474         if (!bprm)
1475                 goto out_files;
1476
1477         retval = prepare_bprm_creds(bprm);
1478         if (retval)
1479                 goto out_free;
1480
1481         check_unsafe_exec(bprm);
1482         current->in_execve = 1;
1483
1484         file = do_open_exec(filename);
1485         retval = PTR_ERR(file);
1486         if (IS_ERR(file))
1487                 goto out_unmark;
1488
1489         sched_exec();
1490
1491         bprm->file = file;
1492         bprm->filename = bprm->interp = filename->name;
1493
1494         retval = bprm_mm_init(bprm);
1495         if (retval)
1496                 goto out_unmark;
1497
1498         bprm->argc = count(argv, MAX_ARG_STRINGS);
1499         if ((retval = bprm->argc) < 0)
1500                 goto out;
1501
1502         bprm->envc = count(envp, MAX_ARG_STRINGS);
1503         if ((retval = bprm->envc) < 0)
1504                 goto out;
1505
1506         retval = prepare_binprm(bprm);
1507         if (retval < 0)
1508                 goto out;
1509
1510         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1511         if (retval < 0)
1512                 goto out;
1513
1514         bprm->exec = bprm->p;
1515         retval = copy_strings(bprm->envc, envp, bprm);
1516         if (retval < 0)
1517                 goto out;
1518
1519         retval = copy_strings(bprm->argc, argv, bprm);
1520         if (retval < 0)
1521                 goto out;
1522
1523         retval = exec_binprm(bprm);
1524         if (retval < 0)
1525                 goto out;
1526
1527         /* execve succeeded */
1528         current->fs->in_exec = 0;
1529         current->in_execve = 0;
1530         acct_update_integrals(current);
1531         task_numa_free(current);
1532         free_bprm(bprm);
1533         putname(filename);
1534         if (displaced)
1535                 put_files_struct(displaced);
1536         return retval;
1537
1538 out:
1539         if (bprm->mm) {
1540                 acct_arg_size(bprm, 0);
1541                 mmput(bprm->mm);
1542         }
1543
1544 out_unmark:
1545         current->fs->in_exec = 0;
1546         current->in_execve = 0;
1547
1548 out_free:
1549         free_bprm(bprm);
1550
1551 out_files:
1552         if (displaced)
1553                 reset_files_struct(displaced);
1554 out_ret:
1555         putname(filename);
1556         return retval;
1557 }
1558
1559 int do_execve(struct filename *filename,
1560         const char __user *const __user *__argv,
1561         const char __user *const __user *__envp)
1562 {
1563         struct user_arg_ptr argv = { .ptr.native = __argv };
1564         struct user_arg_ptr envp = { .ptr.native = __envp };
1565         return do_execve_common(filename, argv, envp);
1566 }
1567
1568 #ifdef CONFIG_COMPAT
1569 static int compat_do_execve(struct filename *filename,
1570         const compat_uptr_t __user *__argv,
1571         const compat_uptr_t __user *__envp)
1572 {
1573         struct user_arg_ptr argv = {
1574                 .is_compat = true,
1575                 .ptr.compat = __argv,
1576         };
1577         struct user_arg_ptr envp = {
1578                 .is_compat = true,
1579                 .ptr.compat = __envp,
1580         };
1581         return do_execve_common(filename, argv, envp);
1582 }
1583 #endif
1584
1585 void set_binfmt(struct linux_binfmt *new)
1586 {
1587         struct mm_struct *mm = current->mm;
1588
1589         if (mm->binfmt)
1590                 module_put(mm->binfmt->module);
1591
1592         mm->binfmt = new;
1593         if (new)
1594                 __module_get(new->module);
1595 }
1596 EXPORT_SYMBOL(set_binfmt);
1597
1598 /*
1599  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1600  */
1601 void set_dumpable(struct mm_struct *mm, int value)
1602 {
1603         unsigned long old, new;
1604
1605         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1606                 return;
1607
1608         do {
1609                 old = ACCESS_ONCE(mm->flags);
1610                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1611         } while (cmpxchg(&mm->flags, old, new) != old);
1612 }
1613
1614 SYSCALL_DEFINE3(execve,
1615                 const char __user *, filename,
1616                 const char __user *const __user *, argv,
1617                 const char __user *const __user *, envp)
1618 {
1619         return do_execve(getname(filename), argv, envp);
1620 }
1621 #ifdef CONFIG_COMPAT
1622 asmlinkage long compat_sys_execve(const char __user * filename,
1623         const compat_uptr_t __user * argv,
1624         const compat_uptr_t __user * envp)
1625 {
1626         return compat_do_execve(getname(filename), argv, envp);
1627 }
1628 #endif