fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/vmacache.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57 #include <linux/oom.h>
  58 #include <linux/compat.h>
  59
  60 #include <asm/uaccess.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/tlb.h>
  63
  64 #include <trace/events/task.h>
  65 #include "internal.h"
  66
  67 #include <trace/events/sched.h>
  68
  69 int suid_dumpable = 0;
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         BUG_ON(!fmt);
  77         if (WARN_ON(!fmt->load_binary))
  78                 return;
  79         write_lock(&binfmt_lock);
  80         insert ? list_add(&fmt->lh, &formats) :
  81                  list_add_tail(&fmt->lh, &formats);
  82         write_unlock(&binfmt_lock);
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 #ifdef CONFIG_USELIB
 102 /*
 103  * Note that a shared library must be both readable and executable due to
 104  * security reasons.
 105  *
 106  * Also note that we take the address to load from from the file itself.
 107  */
 108 SYSCALL_DEFINE1(uselib, const char __user *, library)
 109 {
 110         struct linux_binfmt *fmt;
 111         struct file *file;
 112         struct filename *tmp = getname(library);
 113         int error = PTR_ERR(tmp);
 114         static const struct open_flags uselib_flags = {
 115                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 116                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 117                 .intent = LOOKUP_OPEN,
 118                 .lookup_flags = LOOKUP_FOLLOW,
 119         };
 120
 121         if (IS_ERR(tmp))
 122                 goto out;
 123
 124         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 125         putname(tmp);
 126         error = PTR_ERR(file);
 127         if (IS_ERR(file))
 128                 goto out;
 129
 130         error = -EINVAL;
 131         if (!S_ISREG(file_inode(file)->i_mode))
 132                 goto exit;
 133
 134         error = -EACCES;
 135         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 136                 goto exit;
 137
 138         fsnotify_open(file);
 139
 140         error = -ENOEXEC;
 141
 142         read_lock(&binfmt_lock);
 143         list_for_each_entry(fmt, &formats, lh) {
 144                 if (!fmt->load_shlib)
 145                         continue;
 146                 if (!try_module_get(fmt->module))
 147                         continue;
 148                 read_unlock(&binfmt_lock);
 149                 error = fmt->load_shlib(file);
 150                 read_lock(&binfmt_lock);
 151                 put_binfmt(fmt);
 152                 if (error != -ENOEXEC)
 153                         break;
 154         }
 155         read_unlock(&binfmt_lock);
 156 exit:
 157         fput(file);
 158 out:
 159         return error;
 160 }
 161 #endif /* #ifdef CONFIG_USELIB */
 162
 163 #ifdef CONFIG_MMU
 164 /*
 165  * The nascent bprm->mm is not visible until exec_mmap() but it can
 166  * use a lot of memory, account these pages in current->mm temporary
 167  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 168  * change the counter back via acct_arg_size(0).
 169  */
 170 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 171 {
 172         struct mm_struct *mm = current->mm;
 173         long diff = (long)(pages - bprm->vma_pages);
 174
 175         if (!mm || !diff)
 176                 return;
 177
 178         bprm->vma_pages = pages;
 179         add_mm_counter(mm, MM_ANONPAGES, diff);
 180 }
 181
 182 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 183                 int write)
 184 {
 185         struct page *page;
 186         int ret;
 187
 188 #ifdef CONFIG_STACK_GROWSUP
 189         if (write) {
 190                 ret = expand_downwards(bprm->vma, pos);
 191                 if (ret < 0)
 192                         return NULL;
 193         }
 194 #endif
 195         ret = get_user_pages(current, bprm->mm, pos,
 196                         1, write, 1, &page, NULL);
 197         if (ret <= 0)
 198                 return NULL;
 199
 200         if (write) {
 201                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 202                 struct rlimit *rlim;
 203
 204                 acct_arg_size(bprm, size / PAGE_SIZE);
 205
 206                 /*
 207                  * We've historically supported up to 32 pages (ARG_MAX)
 208                  * of argument strings even with small stacks
 209                  */
 210                 if (size <= ARG_MAX)
 211                         return page;
 212
 213                 /*
 214                  * Limit to 1/4-th the stack size for the argv+env strings.
 215                  * This ensures that:
 216                  *  - the remaining binfmt code will not run out of stack space,
 217                  *  - the program will have a reasonable amount of stack left
 218                  *    to work from.
 219                  */
 220                 rlim = current->signal->rlim;
 221                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 222                         put_page(page);
 223                         return NULL;
 224                 }
 225         }
 226
 227         return page;
 228 }
 229
 230 static void put_arg_page(struct page *page)
 231 {
 232         put_page(page);
 233 }
 234
 235 static void free_arg_page(struct linux_binprm *bprm, int i)
 236 {
 237 }
 238
 239 static void free_arg_pages(struct linux_binprm *bprm)
 240 {
 241 }
 242
 243 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 244                 struct page *page)
 245 {
 246         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 247 }
 248
 249 static int __bprm_mm_init(struct linux_binprm *bprm)
 250 {
 251         int err;
 252         struct vm_area_struct *vma = NULL;
 253         struct mm_struct *mm = bprm->mm;
 254
 255         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 256         if (!vma)
 257                 return -ENOMEM;
 258
 259         down_write(&mm->mmap_sem);
 260         vma->vm_mm = mm;
 261
 262         /*
 263          * Place the stack at the largest stack address the architecture
 264          * supports. Later, we'll move this to an appropriate place. We don't
 265          * use STACK_TOP because that can depend on attributes which aren't
 266          * configured yet.
 267          */
 268         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 269         vma->vm_end = STACK_TOP_MAX;
 270         vma->vm_start = vma->vm_end - PAGE_SIZE;
 271         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 272         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 273         INIT_LIST_HEAD(&vma->anon_vma_chain);
 274
 275         err = insert_vm_struct(mm, vma);
 276         if (err)
 277                 goto err;
 278
 279         mm->stack_vm = mm->total_vm = 1;
 280         arch_bprm_mm_init(mm, vma);
 281         up_write(&mm->mmap_sem);
 282         bprm->p = vma->vm_end - sizeof(void *);
 283         return 0;
 284 err:
 285         up_write(&mm->mmap_sem);
 286         bprm->vma = NULL;
 287         kmem_cache_free(vm_area_cachep, vma);
 288         return err;
 289 }
 290
 291 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 292 {
 293         return len <= MAX_ARG_STRLEN;
 294 }
 295
 296 #else
 297
 298 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 299 {
 300 }
 301
 302 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 303                 int write)
 304 {
 305         struct page *page;
 306
 307         page = bprm->page[pos / PAGE_SIZE];
 308         if (!page && write) {
 309                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 310                 if (!page)
 311                         return NULL;
 312                 bprm->page[pos / PAGE_SIZE] = page;
 313         }
 314
 315         return page;
 316 }
 317
 318 static void put_arg_page(struct page *page)
 319 {
 320 }
 321
 322 static void free_arg_page(struct linux_binprm *bprm, int i)
 323 {
 324         if (bprm->page[i]) {
 325                 __free_page(bprm->page[i]);
 326                 bprm->page[i] = NULL;
 327         }
 328 }
 329
 330 static void free_arg_pages(struct linux_binprm *bprm)
 331 {
 332         int i;
 333
 334         for (i = 0; i < MAX_ARG_PAGES; i++)
 335                 free_arg_page(bprm, i);
 336 }
 337
 338 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 339                 struct page *page)
 340 {
 341 }
 342
 343 static int __bprm_mm_init(struct linux_binprm *bprm)
 344 {
 345         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 346         return 0;
 347 }
 348
 349 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 350 {
 351         return len <= bprm->p;
 352 }
 353
 354 #endif /* CONFIG_MMU */
 355
 356 /*
 357  * Create a new mm_struct and populate it with a temporary stack
 358  * vm_area_struct.  We don't have enough context at this point to set the stack
 359  * flags, permissions, and offset, so we use temporary values.  We'll update
 360  * them later in setup_arg_pages().
 361  */
 362 static int bprm_mm_init(struct linux_binprm *bprm)
 363 {
 364         int err;
 365         struct mm_struct *mm = NULL;
 366
 367         bprm->mm = mm = mm_alloc();
 368         err = -ENOMEM;
 369         if (!mm)
 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 */
 658         stack_base = rlimit_max(RLIMIT_STACK);
 659         if (stack_base > STACK_SIZE_MAX)
 660                 stack_base = STACK_SIZE_MAX;
 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_execat(int fd, struct filename *name, int flags)
 752 {
 753         struct file *file;
 754         int err;
 755         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         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 763                 return ERR_PTR(-EINVAL);
 764         if (flags & AT_SYMLINK_NOFOLLOW)
 765                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 766         if (flags & AT_EMPTY_PATH)
 767                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 768
 769         file = do_filp_open(fd, name, &open_exec_flags);
 770         if (IS_ERR(file))
 771                 goto out;
 772
 773         err = -EACCES;
 774         if (!S_ISREG(file_inode(file)->i_mode))
 775                 goto exit;
 776
 777         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 778                 goto exit;
 779
 780         err = deny_write_access(file);
 781         if (err)
 782                 goto exit;
 783
 784         if (name->name[0] != '\0')
 785                 fsnotify_open(file);
 786
 787 out:
 788         return file;
 789
 790 exit:
 791         fput(file);
 792         return ERR_PTR(err);
 793 }
 794
 795 struct file *open_exec(const char *name)
 796 {
 797         struct filename *filename = getname_kernel(name);
 798         struct file *f = ERR_CAST(filename);
 799
 800         if (!IS_ERR(filename)) {
 801                 f = do_open_execat(AT_FDCWD, filename, 0);
 802                 putname(filename);
 803         }
 804         return f;
 805 }
 806 EXPORT_SYMBOL(open_exec);
 807
 808 int kernel_read(struct file *file, loff_t offset,
 809                 char *addr, unsigned long count)
 810 {
 811         mm_segment_t old_fs;
 812         loff_t pos = offset;
 813         int result;
 814
 815         old_fs = get_fs();
 816         set_fs(get_ds());
 817         /* The cast to a user pointer is valid due to the set_fs() */
 818         result = vfs_read(file, (void __user *)addr, count, &pos);
 819         set_fs(old_fs);
 820         return result;
 821 }
 822
 823 EXPORT_SYMBOL(kernel_read);
 824
 825 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 826 {
 827         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 828         if (res > 0)
 829                 flush_icache_range(addr, addr + len);
 830         return res;
 831 }
 832 EXPORT_SYMBOL(read_code);
 833
 834 static int exec_mmap(struct mm_struct *mm)
 835 {
 836         struct task_struct *tsk;
 837         struct mm_struct *old_mm, *active_mm;
 838
 839         /* Notify parent that we're no longer interested in the old VM */
 840         tsk = current;
 841         old_mm = current->mm;
 842         mm_release(tsk, old_mm);
 843
 844         if (old_mm) {
 845                 sync_mm_rss(old_mm);
 846                 /*
 847                  * Make sure that if there is a core dump in progress
 848                  * for the old mm, we get out and die instead of going
 849                  * through with the exec.  We must hold mmap_sem around
 850                  * checking core_state and changing tsk->mm.
 851                  */
 852                 down_read(&old_mm->mmap_sem);
 853                 if (unlikely(old_mm->core_state)) {
 854                         up_read(&old_mm->mmap_sem);
 855                         return -EINTR;
 856                 }
 857         }
 858         task_lock(tsk);
 859         active_mm = tsk->active_mm;
 860         tsk->mm = mm;
 861         tsk->active_mm = mm;
 862         activate_mm(active_mm, mm);
 863         tsk->mm->vmacache_seqnum = 0;
 864         vmacache_flush(tsk);
 865         task_unlock(tsk);
 866         if (old_mm) {
 867                 up_read(&old_mm->mmap_sem);
 868                 BUG_ON(active_mm != old_mm);
 869                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 870                 mm_update_next_owner(old_mm);
 871                 mmput(old_mm);
 872                 return 0;
 873         }
 874         mmdrop(active_mm);
 875         return 0;
 876 }
 877
 878 /*
 879  * This function makes sure the current process has its own signal table,
 880  * so that flush_signal_handlers can later reset the handlers without
 881  * disturbing other processes.  (Other processes might share the signal
 882  * table via the CLONE_SIGHAND option to clone().)
 883  */
 884 static int de_thread(struct task_struct *tsk)
 885 {
 886         struct signal_struct *sig = tsk->signal;
 887         struct sighand_struct *oldsighand = tsk->sighand;
 888         spinlock_t *lock = &oldsighand->siglock;
 889
 890         if (thread_group_empty(tsk))
 891                 goto no_thread_group;
 892
 893         /*
 894          * Kill all other threads in the thread group.
 895          */
 896         spin_lock_irq(lock);
 897         if (signal_group_exit(sig)) {
 898                 /*
 899                  * Another group action in progress, just
 900                  * return so that the signal is processed.
 901                  */
 902                 spin_unlock_irq(lock);
 903                 return -EAGAIN;
 904         }
 905
 906         sig->group_exit_task = tsk;
 907         sig->notify_count = zap_other_threads(tsk);
 908         if (!thread_group_leader(tsk))
 909                 sig->notify_count--;
 910
 911         while (sig->notify_count) {
 912                 __set_current_state(TASK_KILLABLE);
 913                 spin_unlock_irq(lock);
 914                 schedule();
 915                 if (unlikely(__fatal_signal_pending(tsk)))
 916                         goto killed;
 917                 spin_lock_irq(lock);
 918         }
 919         spin_unlock_irq(lock);
 920
 921         /*
 922          * At this point all other threads have exited, all we have to
 923          * do is to wait for the thread group leader to become inactive,
 924          * and to assume its PID:
 925          */
 926         if (!thread_group_leader(tsk)) {
 927                 struct task_struct *leader = tsk->group_leader;
 928
 929                 for (;;) {
 930                         threadgroup_change_begin(tsk);
 931                         write_lock_irq(&tasklist_lock);
 932                         /*
 933                          * Do this under tasklist_lock to ensure that
 934                          * exit_notify() can't miss ->group_exit_task
 935                          */
 936                         sig->notify_count = -1;
 937                         if (likely(leader->exit_state))
 938                                 break;
 939                         __set_current_state(TASK_KILLABLE);
 940                         write_unlock_irq(&tasklist_lock);
 941                         threadgroup_change_end(tsk);
 942                         schedule();
 943                         if (unlikely(__fatal_signal_pending(tsk)))
 944                                 goto killed;
 945                 }
 946
 947                 /*
 948                  * The only record we have of the real-time age of a
 949                  * process, regardless of execs it's done, is start_time.
 950                  * All the past CPU time is accumulated in signal_struct
 951                  * from sister threads now dead.  But in this non-leader
 952                  * exec, nothing survives from the original leader thread,
 953                  * whose birth marks the true age of this process now.
 954                  * When we take on its identity by switching to its PID, we
 955                  * also take its birthdate (always earlier than our own).
 956                  */
 957                 tsk->start_time = leader->start_time;
 958                 tsk->real_start_time = leader->real_start_time;
 959
 960                 BUG_ON(!same_thread_group(leader, tsk));
 961                 BUG_ON(has_group_leader_pid(tsk));
 962                 /*
 963                  * An exec() starts a new thread group with the
 964                  * TGID of the previous thread group. Rehash the
 965                  * two threads with a switched PID, and release
 966                  * the former thread group leader:
 967                  */
 968
 969                 /* Become a process group leader with the old leader's pid.
 970                  * The old leader becomes a thread of the this thread group.
 971                  * Note: The old leader also uses this pid until release_task
 972                  *       is called.  Odd but simple and correct.
 973                  */
 974                 tsk->pid = leader->pid;
 975                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
 976                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 977                 transfer_pid(leader, tsk, PIDTYPE_SID);
 978
 979                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 980                 list_replace_init(&leader->sibling, &tsk->sibling);
 981
 982                 tsk->group_leader = tsk;
 983                 leader->group_leader = tsk;
 984
 985                 tsk->exit_signal = SIGCHLD;
 986                 leader->exit_signal = -1;
 987
 988                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 989                 leader->exit_state = EXIT_DEAD;
 990
 991                 /*
 992                  * We are going to release_task()->ptrace_unlink() silently,
 993                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 994                  * the tracer wont't block again waiting for this thread.
 995                  */
 996                 if (unlikely(leader->ptrace))
 997                         __wake_up_parent(leader, leader->parent);
 998                 write_unlock_irq(&tasklist_lock);
 999                 threadgroup_change_end(tsk);
1000
1001                 release_task(leader);
1002         }
1003
1004         sig->group_exit_task = NULL;
1005         sig->notify_count = 0;
1006
1007 no_thread_group:
1008         /* we have changed execution domain */
1009         tsk->exit_signal = SIGCHLD;
1010
1011         exit_itimers(sig);
1012         flush_itimer_signals();
1013
1014         if (atomic_read(&oldsighand->count) != 1) {
1015                 struct sighand_struct *newsighand;
1016                 /*
1017                  * This ->sighand is shared with the CLONE_SIGHAND
1018                  * but not CLONE_THREAD task, switch to the new one.
1019                  */
1020                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1021                 if (!newsighand)
1022                         return -ENOMEM;
1023
1024                 atomic_set(&newsighand->count, 1);
1025                 memcpy(newsighand->action, oldsighand->action,
1026                        sizeof(newsighand->action));
1027
1028                 write_lock_irq(&tasklist_lock);
1029                 spin_lock(&oldsighand->siglock);
1030                 rcu_assign_pointer(tsk->sighand, newsighand);
1031                 spin_unlock(&oldsighand->siglock);
1032                 write_unlock_irq(&tasklist_lock);
1033
1034                 __cleanup_sighand(oldsighand);
1035         }
1036
1037         BUG_ON(!thread_group_leader(tsk));
1038         return 0;
1039
1040 killed:
1041         /* protects against exit_notify() and __exit_signal() */
1042         read_lock(&tasklist_lock);
1043         sig->group_exit_task = NULL;
1044         sig->notify_count = 0;
1045         read_unlock(&tasklist_lock);
1046         return -EAGAIN;
1047 }
1048
1049 char *get_task_comm(char *buf, struct task_struct *tsk)
1050 {
1051         /* buf must be at least sizeof(tsk->comm) in size */
1052         task_lock(tsk);
1053         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1054         task_unlock(tsk);
1055         return buf;
1056 }
1057 EXPORT_SYMBOL_GPL(get_task_comm);
1058
1059 /*
1060  * These functions flushes out all traces of the currently running executable
1061  * so that a new one can be started
1062  */
1063
1064 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1065 {
1066         task_lock(tsk);
1067         trace_task_rename(tsk, buf);
1068         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1069         task_unlock(tsk);
1070         perf_event_comm(tsk, exec);
1071 }
1072
1073 int flush_old_exec(struct linux_binprm * bprm)
1074 {
1075         int retval;
1076
1077         /*
1078          * Make sure we have a private signal table and that
1079          * we are unassociated from the previous thread group.
1080          */
1081         retval = de_thread(current);
1082         if (retval)
1083                 goto out;
1084
1085         /*
1086          * Must be called _before_ exec_mmap() as bprm->mm is
1087          * not visibile until then. This also enables the update
1088          * to be lockless.
1089          */
1090         set_mm_exe_file(bprm->mm, bprm->file);
1091
1092         /*
1093          * Release all of the old mmap stuff
1094          */
1095         acct_arg_size(bprm, 0);
1096         retval = exec_mmap(bprm->mm);
1097         if (retval)
1098                 goto out;
1099
1100         bprm->mm = NULL;                /* We're using it now */
1101
1102         set_fs(USER_DS);
1103         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1104                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1105         flush_thread();
1106         current->personality &= ~bprm->per_clear;
1107
1108         return 0;
1109
1110 out:
1111         return retval;
1112 }
1113 EXPORT_SYMBOL(flush_old_exec);
1114
1115 void would_dump(struct linux_binprm *bprm, struct file *file)
1116 {
1117         if (inode_permission(file_inode(file), MAY_READ) < 0)
1118                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1119 }
1120 EXPORT_SYMBOL(would_dump);
1121
1122 void setup_new_exec(struct linux_binprm * bprm)
1123 {
1124         arch_pick_mmap_layout(current->mm);
1125
1126         /* This is the point of no return */
1127         current->sas_ss_sp = current->sas_ss_size = 0;
1128
1129         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1130                 set_dumpable(current->mm, SUID_DUMP_USER);
1131         else
1132                 set_dumpable(current->mm, suid_dumpable);
1133
1134         perf_event_exec();
1135         __set_task_comm(current, kbasename(bprm->filename), true);
1136
1137         /* Set the new mm task size. We have to do that late because it may
1138          * depend on TIF_32BIT which is only updated in flush_thread() on
1139          * some architectures like powerpc
1140          */
1141         current->mm->task_size = TASK_SIZE;
1142
1143         /* install the new credentials */
1144         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1145             !gid_eq(bprm->cred->gid, current_egid())) {
1146                 current->pdeath_signal = 0;
1147         } else {
1148                 would_dump(bprm, bprm->file);
1149                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1150                         set_dumpable(current->mm, suid_dumpable);
1151         }
1152
1153         /* An exec changes our domain. We are no longer part of the thread
1154            group */
1155         current->self_exec_id++;
1156         flush_signal_handlers(current, 0);
1157         do_close_on_exec(current->files);
1158 }
1159 EXPORT_SYMBOL(setup_new_exec);
1160
1161 /*
1162  * Prepare credentials and lock ->cred_guard_mutex.
1163  * install_exec_creds() commits the new creds and drops the lock.
1164  * Or, if exec fails before, free_bprm() should release ->cred and
1165  * and unlock.
1166  */
1167 int prepare_bprm_creds(struct linux_binprm *bprm)
1168 {
1169         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1170                 return -ERESTARTNOINTR;
1171
1172         bprm->cred = prepare_exec_creds();
1173         if (likely(bprm->cred))
1174                 return 0;
1175
1176         mutex_unlock(&current->signal->cred_guard_mutex);
1177         return -ENOMEM;
1178 }
1179
1180 static void free_bprm(struct linux_binprm *bprm)
1181 {
1182         free_arg_pages(bprm);
1183         if (bprm->cred) {
1184                 mutex_unlock(&current->signal->cred_guard_mutex);
1185                 abort_creds(bprm->cred);
1186         }
1187         if (bprm->file) {
1188                 allow_write_access(bprm->file);
1189                 fput(bprm->file);
1190         }
1191         /* If a binfmt changed the interp, free it. */
1192         if (bprm->interp != bprm->filename)
1193                 kfree(bprm->interp);
1194         kfree(bprm);
1195 }
1196
1197 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1198 {
1199         /* If a binfmt changed the interp, free it first. */
1200         if (bprm->interp != bprm->filename)
1201                 kfree(bprm->interp);
1202         bprm->interp = kstrdup(interp, GFP_KERNEL);
1203         if (!bprm->interp)
1204                 return -ENOMEM;
1205         return 0;
1206 }
1207 EXPORT_SYMBOL(bprm_change_interp);
1208
1209 /*
1210  * install the new credentials for this executable
1211  */
1212 void install_exec_creds(struct linux_binprm *bprm)
1213 {
1214         security_bprm_committing_creds(bprm);
1215
1216         commit_creds(bprm->cred);
1217         bprm->cred = NULL;
1218
1219         /*
1220          * Disable monitoring for regular users
1221          * when executing setuid binaries. Must
1222          * wait until new credentials are committed
1223          * by commit_creds() above
1224          */
1225         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1226                 perf_event_exit_task(current);
1227         /*
1228          * cred_guard_mutex must be held at least to this point to prevent
1229          * ptrace_attach() from altering our determination of the task's
1230          * credentials; any time after this it may be unlocked.
1231          */
1232         security_bprm_committed_creds(bprm);
1233         mutex_unlock(&current->signal->cred_guard_mutex);
1234 }
1235 EXPORT_SYMBOL(install_exec_creds);
1236
1237 /*
1238  * determine how safe it is to execute the proposed program
1239  * - the caller must hold ->cred_guard_mutex to protect against
1240  *   PTRACE_ATTACH or seccomp thread-sync
1241  */
1242 static void check_unsafe_exec(struct linux_binprm *bprm)
1243 {
1244         struct task_struct *p = current, *t;
1245         unsigned n_fs;
1246
1247         if (p->ptrace) {
1248                 if (p->ptrace & PT_PTRACE_CAP)
1249                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1250                 else
1251                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1252         }
1253
1254         /*
1255          * This isn't strictly necessary, but it makes it harder for LSMs to
1256          * mess up.
1257          */
1258         if (task_no_new_privs(current))
1259                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1260
1261         t = p;
1262         n_fs = 1;
1263         spin_lock(&p->fs->lock);
1264         rcu_read_lock();
1265         while_each_thread(p, t) {
1266                 if (t->fs == p->fs)
1267                         n_fs++;
1268         }
1269         rcu_read_unlock();
1270
1271         if (p->fs->users > n_fs)
1272                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1273         else
1274                 p->fs->in_exec = 1;
1275         spin_unlock(&p->fs->lock);
1276 }
1277
1278 static void bprm_fill_uid(struct linux_binprm *bprm)
1279 {
1280         struct inode *inode;
1281         unsigned int mode;
1282         kuid_t uid;
1283         kgid_t gid;
1284
1285         /* clear any previous set[ug]id data from a previous binary */
1286         bprm->cred->euid = current_euid();
1287         bprm->cred->egid = current_egid();
1288
1289         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1290                 return;
1291
1292         if (task_no_new_privs(current))
1293                 return;
1294
1295         inode = file_inode(bprm->file);
1296         mode = READ_ONCE(inode->i_mode);
1297         if (!(mode & (S_ISUID|S_ISGID)))
1298                 return;
1299
1300         /* Be careful if suid/sgid is set */
1301         mutex_lock(&inode->i_mutex);
1302
1303         /* reload atomically mode/uid/gid now that lock held */
1304         mode = inode->i_mode;
1305         uid = inode->i_uid;
1306         gid = inode->i_gid;
1307         mutex_unlock(&inode->i_mutex);
1308
1309         /* We ignore suid/sgid if there are no mappings for them in the ns */
1310         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1311                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1312                 return;
1313
1314         if (mode & S_ISUID) {
1315                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1316                 bprm->cred->euid = uid;
1317         }
1318
1319         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1320                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1321                 bprm->cred->egid = gid;
1322         }
1323 }
1324
1325 /*
1326  * Fill the binprm structure from the inode.
1327  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1328  *
1329  * This may be called multiple times for binary chains (scripts for example).
1330  */
1331 int prepare_binprm(struct linux_binprm *bprm)
1332 {
1333         int retval;
1334
1335         bprm_fill_uid(bprm);
1336
1337         /* fill in binprm security blob */
1338         retval = security_bprm_set_creds(bprm);
1339         if (retval)
1340                 return retval;
1341         bprm->cred_prepared = 1;
1342
1343         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1344         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1345 }
1346
1347 EXPORT_SYMBOL(prepare_binprm);
1348
1349 /*
1350  * Arguments are '\0' separated strings found at the location bprm->p
1351  * points to; chop off the first by relocating brpm->p to right after
1352  * the first '\0' encountered.
1353  */
1354 int remove_arg_zero(struct linux_binprm *bprm)
1355 {
1356         int ret = 0;
1357         unsigned long offset;
1358         char *kaddr;
1359         struct page *page;
1360
1361         if (!bprm->argc)
1362                 return 0;
1363
1364         do {
1365                 offset = bprm->p & ~PAGE_MASK;
1366                 page = get_arg_page(bprm, bprm->p, 0);
1367                 if (!page) {
1368                         ret = -EFAULT;
1369                         goto out;
1370                 }
1371                 kaddr = kmap_atomic(page);
1372
1373                 for (; offset < PAGE_SIZE && kaddr[offset];
1374                                 offset++, bprm->p++)
1375                         ;
1376
1377                 kunmap_atomic(kaddr);
1378                 put_arg_page(page);
1379
1380                 if (offset == PAGE_SIZE)
1381                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1382         } while (offset == PAGE_SIZE);
1383
1384         bprm->p++;
1385         bprm->argc--;
1386         ret = 0;
1387
1388 out:
1389         return ret;
1390 }
1391 EXPORT_SYMBOL(remove_arg_zero);
1392
1393 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1394 /*
1395  * cycle the list of binary formats handler, until one recognizes the image
1396  */
1397 int search_binary_handler(struct linux_binprm *bprm)
1398 {
1399         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1400         struct linux_binfmt *fmt;
1401         int retval;
1402
1403         /* This allows 4 levels of binfmt rewrites before failing hard. */
1404         if (bprm->recursion_depth > 5)
1405                 return -ELOOP;
1406
1407         retval = security_bprm_check(bprm);
1408         if (retval)
1409                 return retval;
1410
1411         retval = -ENOENT;
1412  retry:
1413         read_lock(&binfmt_lock);
1414         list_for_each_entry(fmt, &formats, lh) {
1415                 if (!try_module_get(fmt->module))
1416                         continue;
1417                 read_unlock(&binfmt_lock);
1418                 bprm->recursion_depth++;
1419                 retval = fmt->load_binary(bprm);
1420                 read_lock(&binfmt_lock);
1421                 put_binfmt(fmt);
1422                 bprm->recursion_depth--;
1423                 if (retval < 0 && !bprm->mm) {
1424                         /* we got to flush_old_exec() and failed after it */
1425                         read_unlock(&binfmt_lock);
1426                         force_sigsegv(SIGSEGV, current);
1427                         return retval;
1428                 }
1429                 if (retval != -ENOEXEC || !bprm->file) {
1430                         read_unlock(&binfmt_lock);
1431                         return retval;
1432                 }
1433         }
1434         read_unlock(&binfmt_lock);
1435
1436         if (need_retry) {
1437                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1438                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1439                         return retval;
1440                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1441                         return retval;
1442                 need_retry = false;
1443                 goto retry;
1444         }
1445
1446         return retval;
1447 }
1448 EXPORT_SYMBOL(search_binary_handler);
1449
1450 static int exec_binprm(struct linux_binprm *bprm)
1451 {
1452         pid_t old_pid, old_vpid;
1453         int ret;
1454
1455         /* Need to fetch pid before load_binary changes it */
1456         old_pid = current->pid;
1457         rcu_read_lock();
1458         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1459         rcu_read_unlock();
1460
1461         ret = search_binary_handler(bprm);
1462         if (ret >= 0) {
1463                 audit_bprm(bprm);
1464                 trace_sched_process_exec(current, old_pid, bprm);
1465                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1466                 proc_exec_connector(current);
1467         }
1468
1469         return ret;
1470 }
1471
1472 /*
1473  * sys_execve() executes a new program.
1474  */
1475 static int do_execveat_common(int fd, struct filename *filename,
1476                               struct user_arg_ptr argv,
1477                               struct user_arg_ptr envp,
1478                               int flags)
1479 {
1480         char *pathbuf = NULL;
1481         struct linux_binprm *bprm;
1482         struct file *file;
1483         struct files_struct *displaced;
1484         int retval;
1485
1486         if (IS_ERR(filename))
1487                 return PTR_ERR(filename);
1488
1489         /*
1490          * We move the actual failure in case of RLIMIT_NPROC excess from
1491          * set*uid() to execve() because too many poorly written programs
1492          * don't check setuid() return code.  Here we additionally recheck
1493          * whether NPROC limit is still exceeded.
1494          */
1495         if ((current->flags & PF_NPROC_EXCEEDED) &&
1496             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1497                 retval = -EAGAIN;
1498                 goto out_ret;
1499         }
1500
1501         /* We're below the limit (still or again), so we don't want to make
1502          * further execve() calls fail. */
1503         current->flags &= ~PF_NPROC_EXCEEDED;
1504
1505         retval = unshare_files(&displaced);
1506         if (retval)
1507                 goto out_ret;
1508
1509         retval = -ENOMEM;
1510         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1511         if (!bprm)
1512                 goto out_files;
1513
1514         retval = prepare_bprm_creds(bprm);
1515         if (retval)
1516                 goto out_free;
1517
1518         check_unsafe_exec(bprm);
1519         current->in_execve = 1;
1520
1521         file = do_open_execat(fd, filename, flags);
1522         retval = PTR_ERR(file);
1523         if (IS_ERR(file))
1524                 goto out_unmark;
1525
1526         sched_exec();
1527
1528         bprm->file = file;
1529         if (fd == AT_FDCWD || filename->name[0] == '/') {
1530                 bprm->filename = filename->name;
1531         } else {
1532                 if (filename->name[0] == '\0')
1533                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1534                 else
1535                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1536                                             fd, filename->name);
1537                 if (!pathbuf) {
1538                         retval = -ENOMEM;
1539                         goto out_unmark;
1540                 }
1541                 /*
1542                  * Record that a name derived from an O_CLOEXEC fd will be
1543                  * inaccessible after exec. Relies on having exclusive access to
1544                  * current->files (due to unshare_files above).
1545                  */
1546                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1547                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1548                 bprm->filename = pathbuf;
1549         }
1550         bprm->interp = bprm->filename;
1551
1552         retval = bprm_mm_init(bprm);
1553         if (retval)
1554                 goto out_unmark;
1555
1556         bprm->argc = count(argv, MAX_ARG_STRINGS);
1557         if ((retval = bprm->argc) < 0)
1558                 goto out;
1559
1560         bprm->envc = count(envp, MAX_ARG_STRINGS);
1561         if ((retval = bprm->envc) < 0)
1562                 goto out;
1563
1564         retval = prepare_binprm(bprm);
1565         if (retval < 0)
1566                 goto out;
1567
1568         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1569         if (retval < 0)
1570                 goto out;
1571
1572         bprm->exec = bprm->p;
1573         retval = copy_strings(bprm->envc, envp, bprm);
1574         if (retval < 0)
1575                 goto out;
1576
1577         retval = copy_strings(bprm->argc, argv, bprm);
1578         if (retval < 0)
1579                 goto out;
1580
1581         retval = exec_binprm(bprm);
1582         if (retval < 0)
1583                 goto out;
1584
1585         /* execve succeeded */
1586         current->fs->in_exec = 0;
1587         current->in_execve = 0;
1588         acct_update_integrals(current);
1589         task_numa_free(current);
1590         free_bprm(bprm);
1591         kfree(pathbuf);
1592         putname(filename);
1593         if (displaced)
1594                 put_files_struct(displaced);
1595         return retval;
1596
1597 out:
1598         if (bprm->mm) {
1599                 acct_arg_size(bprm, 0);
1600                 mmput(bprm->mm);
1601         }
1602
1603 out_unmark:
1604         current->fs->in_exec = 0;
1605         current->in_execve = 0;
1606
1607 out_free:
1608         free_bprm(bprm);
1609         kfree(pathbuf);
1610
1611 out_files:
1612         if (displaced)
1613                 reset_files_struct(displaced);
1614 out_ret:
1615         putname(filename);
1616         return retval;
1617 }
1618
1619 int do_execve(struct filename *filename,
1620         const char __user *const __user *__argv,
1621         const char __user *const __user *__envp)
1622 {
1623         struct user_arg_ptr argv = { .ptr.native = __argv };
1624         struct user_arg_ptr envp = { .ptr.native = __envp };
1625         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1626 }
1627
1628 int do_execveat(int fd, struct filename *filename,
1629                 const char __user *const __user *__argv,
1630                 const char __user *const __user *__envp,
1631                 int flags)
1632 {
1633         struct user_arg_ptr argv = { .ptr.native = __argv };
1634         struct user_arg_ptr envp = { .ptr.native = __envp };
1635
1636         return do_execveat_common(fd, filename, argv, envp, flags);
1637 }
1638
1639 #ifdef CONFIG_COMPAT
1640 static int compat_do_execve(struct filename *filename,
1641         const compat_uptr_t __user *__argv,
1642         const compat_uptr_t __user *__envp)
1643 {
1644         struct user_arg_ptr argv = {
1645                 .is_compat = true,
1646                 .ptr.compat = __argv,
1647         };
1648         struct user_arg_ptr envp = {
1649                 .is_compat = true,
1650                 .ptr.compat = __envp,
1651         };
1652         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1653 }
1654
1655 static int compat_do_execveat(int fd, struct filename *filename,
1656                               const compat_uptr_t __user *__argv,
1657                               const compat_uptr_t __user *__envp,
1658                               int flags)
1659 {
1660         struct user_arg_ptr argv = {
1661                 .is_compat = true,
1662                 .ptr.compat = __argv,
1663         };
1664         struct user_arg_ptr envp = {
1665                 .is_compat = true,
1666                 .ptr.compat = __envp,
1667         };
1668         return do_execveat_common(fd, filename, argv, envp, flags);
1669 }
1670 #endif
1671
1672 void set_binfmt(struct linux_binfmt *new)
1673 {
1674         struct mm_struct *mm = current->mm;
1675
1676         if (mm->binfmt)
1677                 module_put(mm->binfmt->module);
1678
1679         mm->binfmt = new;
1680         if (new)
1681                 __module_get(new->module);
1682 }
1683 EXPORT_SYMBOL(set_binfmt);
1684
1685 /*
1686  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1687  */
1688 void set_dumpable(struct mm_struct *mm, int value)
1689 {
1690         unsigned long old, new;
1691
1692         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1693                 return;
1694
1695         do {
1696                 old = ACCESS_ONCE(mm->flags);
1697                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1698         } while (cmpxchg(&mm->flags, old, new) != old);
1699 }
1700
1701 SYSCALL_DEFINE3(execve,
1702                 const char __user *, filename,
1703                 const char __user *const __user *, argv,
1704                 const char __user *const __user *, envp)
1705 {
1706         return do_execve(getname(filename), argv, envp);
1707 }
1708
1709 SYSCALL_DEFINE5(execveat,
1710                 int, fd, const char __user *, filename,
1711                 const char __user *const __user *, argv,
1712                 const char __user *const __user *, envp,
1713                 int, flags)
1714 {
1715         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1716
1717         return do_execveat(fd,
1718                            getname_flags(filename, lookup_flags, NULL),
1719                            argv, envp, flags);
1720 }
1721
1722 #ifdef CONFIG_COMPAT
1723 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1724         const compat_uptr_t __user *, argv,
1725         const compat_uptr_t __user *, envp)
1726 {
1727         return compat_do_execve(getname(filename), argv, envp);
1728 }
1729
1730 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1731                        const char __user *, filename,
1732                        const compat_uptr_t __user *, argv,
1733                        const compat_uptr_t __user *, envp,
1734                        int,  flags)
1735 {
1736         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1737
1738         return compat_do_execveat(fd,
1739                                   getname_flags(filename, lookup_flags, NULL),
1740                                   argv, envp, flags);
1741 }
1742 #endif