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 #include <linux/vmalloc.h>
  60
  61 #include <asm/uaccess.h>
  62 #include <asm/mmu_context.h>
  63 #include <asm/tlb.h>
  64
  65 #include <trace/events/task.h>
  66 #include "internal.h"
  67
  68 #include <trace/events/sched.h>
  69
  70 int suid_dumpable = 0;
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  76 {
  77         BUG_ON(!fmt);
  78         if (WARN_ON(!fmt->load_binary))
  79                 return;
  80         write_lock(&binfmt_lock);
  81         insert ? list_add(&fmt->lh, &formats) :
  82                  list_add_tail(&fmt->lh, &formats);
  83         write_unlock(&binfmt_lock);
  84 }
  85
  86 EXPORT_SYMBOL(__register_binfmt);
  87
  88 void unregister_binfmt(struct linux_binfmt * fmt)
  89 {
  90         write_lock(&binfmt_lock);
  91         list_del(&fmt->lh);
  92         write_unlock(&binfmt_lock);
  93 }
  94
  95 EXPORT_SYMBOL(unregister_binfmt);
  96
  97 static inline void put_binfmt(struct linux_binfmt * fmt)
  98 {
  99         module_put(fmt->module);
 100 }
 101
 102 bool path_noexec(const struct path *path)
 103 {
 104         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 105                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 106 }
 107
 108 #ifdef CONFIG_USELIB
 109 /*
 110  * Note that a shared library must be both readable and executable due to
 111  * security reasons.
 112  *
 113  * Also note that we take the address to load from from the file itself.
 114  */
 115 SYSCALL_DEFINE1(uselib, const char __user *, library)
 116 {
 117         struct linux_binfmt *fmt;
 118         struct file *file;
 119         struct filename *tmp = getname(library);
 120         int error = PTR_ERR(tmp);
 121         static const struct open_flags uselib_flags = {
 122                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 123                 .acc_mode = MAY_READ | MAY_EXEC,
 124                 .intent = LOOKUP_OPEN,
 125                 .lookup_flags = LOOKUP_FOLLOW,
 126         };
 127
 128         if (IS_ERR(tmp))
 129                 goto out;
 130
 131         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 132         putname(tmp);
 133         error = PTR_ERR(file);
 134         if (IS_ERR(file))
 135                 goto out;
 136
 137         error = -EINVAL;
 138         if (!S_ISREG(file_inode(file)->i_mode))
 139                 goto exit;
 140
 141         error = -EACCES;
 142         if (path_noexec(&file->f_path))
 143                 goto exit;
 144
 145         fsnotify_open(file);
 146
 147         error = -ENOEXEC;
 148
 149         read_lock(&binfmt_lock);
 150         list_for_each_entry(fmt, &formats, lh) {
 151                 if (!fmt->load_shlib)
 152                         continue;
 153                 if (!try_module_get(fmt->module))
 154                         continue;
 155                 read_unlock(&binfmt_lock);
 156                 error = fmt->load_shlib(file);
 157                 read_lock(&binfmt_lock);
 158                 put_binfmt(fmt);
 159                 if (error != -ENOEXEC)
 160                         break;
 161         }
 162         read_unlock(&binfmt_lock);
 163 exit:
 164         fput(file);
 165 out:
 166         return error;
 167 }
 168 #endif /* #ifdef CONFIG_USELIB */
 169
 170 #ifdef CONFIG_MMU
 171 /*
 172  * The nascent bprm->mm is not visible until exec_mmap() but it can
 173  * use a lot of memory, account these pages in current->mm temporary
 174  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 175  * change the counter back via acct_arg_size(0).
 176  */
 177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 178 {
 179         struct mm_struct *mm = current->mm;
 180         long diff = (long)(pages - bprm->vma_pages);
 181
 182         if (!mm || !diff)
 183                 return;
 184
 185         bprm->vma_pages = pages;
 186         add_mm_counter(mm, MM_ANONPAGES, diff);
 187 }
 188
 189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 190                 int write)
 191 {
 192         struct page *page;
 193         int ret;
 194         unsigned int gup_flags = FOLL_FORCE;
 195
 196 #ifdef CONFIG_STACK_GROWSUP
 197         if (write) {
 198                 ret = expand_downwards(bprm->vma, pos);
 199                 if (ret < 0)
 200                         return NULL;
 201         }
 202 #endif
 203
 204         if (write)
 205                 gup_flags |= FOLL_WRITE;
 206
 207         /*
 208          * We are doing an exec().  'current' is the process
 209          * doing the exec and bprm->mm is the new process's mm.
 210          */
 211         ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 212                         &page, NULL);
 213         if (ret <= 0)
 214                 return NULL;
 215
 216         if (write) {
 217                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 218                 unsigned long ptr_size, limit;
 219
 220                 /*
 221                  * Since the stack will hold pointers to the strings, we
 222                  * must account for them as well.
 223                  *
 224                  * The size calculation is the entire vma while each arg page is
 225                  * built, so each time we get here it's calculating how far it
 226                  * is currently (rather than each call being just the newly
 227                  * added size from the arg page).  As a result, we need to
 228                  * always add the entire size of the pointers, so that on the
 229                  * last call to get_arg_page() we'll actually have the entire
 230                  * correct size.
 231                  */
 232                 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 233                 if (ptr_size > ULONG_MAX - size)
 234                         goto fail;
 235                 size += ptr_size;
 236
 237                 acct_arg_size(bprm, size / PAGE_SIZE);
 238
 239                 /*
 240                  * We've historically supported up to 32 pages (ARG_MAX)
 241                  * of argument strings even with small stacks
 242                  */
 243                 if (size <= ARG_MAX)
 244                         return page;
 245
 246                 /*
 247                  * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 248                  * (whichever is smaller) for the argv+env strings.
 249                  * This ensures that:
 250                  *  - the remaining binfmt code will not run out of stack space,
 251                  *  - the program will have a reasonable amount of stack left
 252                  *    to work from.
 253                  */
 254                 limit = _STK_LIM / 4 * 3;
 255                 limit = min(limit, rlimit(RLIMIT_STACK) / 4);
 256                 if (size > limit)
 257                         goto fail;
 258         }
 259
 260         return page;
 261
 262 fail:
 263         put_page(page);
 264         return NULL;
 265 }
 266
 267 static void put_arg_page(struct page *page)
 268 {
 269         put_page(page);
 270 }
 271
 272 static void free_arg_pages(struct linux_binprm *bprm)
 273 {
 274 }
 275
 276 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 277                 struct page *page)
 278 {
 279         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 280 }
 281
 282 static int __bprm_mm_init(struct linux_binprm *bprm)
 283 {
 284         int err;
 285         struct vm_area_struct *vma = NULL;
 286         struct mm_struct *mm = bprm->mm;
 287
 288         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 289         if (!vma)
 290                 return -ENOMEM;
 291
 292         if (down_write_killable(&mm->mmap_sem)) {
 293                 err = -EINTR;
 294                 goto err_free;
 295         }
 296         vma->vm_mm = mm;
 297
 298         /*
 299          * Place the stack at the largest stack address the architecture
 300          * supports. Later, we'll move this to an appropriate place. We don't
 301          * use STACK_TOP because that can depend on attributes which aren't
 302          * configured yet.
 303          */
 304         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 305         vma->vm_end = STACK_TOP_MAX;
 306         vma->vm_start = vma->vm_end - PAGE_SIZE;
 307         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 308         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 309         INIT_LIST_HEAD(&vma->anon_vma_chain);
 310
 311         err = insert_vm_struct(mm, vma);
 312         if (err)
 313                 goto err;
 314
 315         mm->stack_vm = mm->total_vm = 1;
 316         arch_bprm_mm_init(mm, vma);
 317         up_write(&mm->mmap_sem);
 318         bprm->p = vma->vm_end - sizeof(void *);
 319         return 0;
 320 err:
 321         up_write(&mm->mmap_sem);
 322 err_free:
 323         bprm->vma = NULL;
 324         kmem_cache_free(vm_area_cachep, vma);
 325         return err;
 326 }
 327
 328 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 329 {
 330         return len <= MAX_ARG_STRLEN;
 331 }
 332
 333 #else
 334
 335 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 336 {
 337 }
 338
 339 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 340                 int write)
 341 {
 342         struct page *page;
 343
 344         page = bprm->page[pos / PAGE_SIZE];
 345         if (!page && write) {
 346                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 347                 if (!page)
 348                         return NULL;
 349                 bprm->page[pos / PAGE_SIZE] = page;
 350         }
 351
 352         return page;
 353 }
 354
 355 static void put_arg_page(struct page *page)
 356 {
 357 }
 358
 359 static void free_arg_page(struct linux_binprm *bprm, int i)
 360 {
 361         if (bprm->page[i]) {
 362                 __free_page(bprm->page[i]);
 363                 bprm->page[i] = NULL;
 364         }
 365 }
 366
 367 static void free_arg_pages(struct linux_binprm *bprm)
 368 {
 369         int i;
 370
 371         for (i = 0; i < MAX_ARG_PAGES; i++)
 372                 free_arg_page(bprm, i);
 373 }
 374
 375 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 376                 struct page *page)
 377 {
 378 }
 379
 380 static int __bprm_mm_init(struct linux_binprm *bprm)
 381 {
 382         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 383         return 0;
 384 }
 385
 386 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 387 {
 388         return len <= bprm->p;
 389 }
 390
 391 #endif /* CONFIG_MMU */
 392
 393 /*
 394  * Create a new mm_struct and populate it with a temporary stack
 395  * vm_area_struct.  We don't have enough context at this point to set the stack
 396  * flags, permissions, and offset, so we use temporary values.  We'll update
 397  * them later in setup_arg_pages().
 398  */
 399 static int bprm_mm_init(struct linux_binprm *bprm)
 400 {
 401         int err;
 402         struct mm_struct *mm = NULL;
 403
 404         bprm->mm = mm = mm_alloc();
 405         err = -ENOMEM;
 406         if (!mm)
 407                 goto err;
 408
 409         err = __bprm_mm_init(bprm);
 410         if (err)
 411                 goto err;
 412
 413         return 0;
 414
 415 err:
 416         if (mm) {
 417                 bprm->mm = NULL;
 418                 mmdrop(mm);
 419         }
 420
 421         return err;
 422 }
 423
 424 struct user_arg_ptr {
 425 #ifdef CONFIG_COMPAT
 426         bool is_compat;
 427 #endif
 428         union {
 429                 const char __user *const __user *native;
 430 #ifdef CONFIG_COMPAT
 431                 const compat_uptr_t __user *compat;
 432 #endif
 433         } ptr;
 434 };
 435
 436 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 437 {
 438         const char __user *native;
 439
 440 #ifdef CONFIG_COMPAT
 441         if (unlikely(argv.is_compat)) {
 442                 compat_uptr_t compat;
 443
 444                 if (get_user(compat, argv.ptr.compat + nr))
 445                         return ERR_PTR(-EFAULT);
 446
 447                 return compat_ptr(compat);
 448         }
 449 #endif
 450
 451         if (get_user(native, argv.ptr.native + nr))
 452                 return ERR_PTR(-EFAULT);
 453
 454         return native;
 455 }
 456
 457 /*
 458  * count() counts the number of strings in array ARGV.
 459  */
 460 static int count(struct user_arg_ptr argv, int max)
 461 {
 462         int i = 0;
 463
 464         if (argv.ptr.native != NULL) {
 465                 for (;;) {
 466                         const char __user *p = get_user_arg_ptr(argv, i);
 467
 468                         if (!p)
 469                                 break;
 470
 471                         if (IS_ERR(p))
 472                                 return -EFAULT;
 473
 474                         if (i >= max)
 475                                 return -E2BIG;
 476                         ++i;
 477
 478                         if (fatal_signal_pending(current))
 479                                 return -ERESTARTNOHAND;
 480                         cond_resched();
 481                 }
 482         }
 483         return i;
 484 }
 485
 486 /*
 487  * 'copy_strings()' copies argument/environment strings from the old
 488  * processes's memory to the new process's stack.  The call to get_user_pages()
 489  * ensures the destination page is created and not swapped out.
 490  */
 491 static int copy_strings(int argc, struct user_arg_ptr argv,
 492                         struct linux_binprm *bprm)
 493 {
 494         struct page *kmapped_page = NULL;
 495         char *kaddr = NULL;
 496         unsigned long kpos = 0;
 497         int ret;
 498
 499         while (argc-- > 0) {
 500                 const char __user *str;
 501                 int len;
 502                 unsigned long pos;
 503
 504                 ret = -EFAULT;
 505                 str = get_user_arg_ptr(argv, argc);
 506                 if (IS_ERR(str))
 507                         goto out;
 508
 509                 len = strnlen_user(str, MAX_ARG_STRLEN);
 510                 if (!len)
 511                         goto out;
 512
 513                 ret = -E2BIG;
 514                 if (!valid_arg_len(bprm, len))
 515                         goto out;
 516
 517                 /* We're going to work our way backwords. */
 518                 pos = bprm->p;
 519                 str += len;
 520                 bprm->p -= len;
 521
 522                 while (len > 0) {
 523                         int offset, bytes_to_copy;
 524
 525                         if (fatal_signal_pending(current)) {
 526                                 ret = -ERESTARTNOHAND;
 527                                 goto out;
 528                         }
 529                         cond_resched();
 530
 531                         offset = pos % PAGE_SIZE;
 532                         if (offset == 0)
 533                                 offset = PAGE_SIZE;
 534
 535                         bytes_to_copy = offset;
 536                         if (bytes_to_copy > len)
 537                                 bytes_to_copy = len;
 538
 539                         offset -= bytes_to_copy;
 540                         pos -= bytes_to_copy;
 541                         str -= bytes_to_copy;
 542                         len -= bytes_to_copy;
 543
 544                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 545                                 struct page *page;
 546
 547                                 page = get_arg_page(bprm, pos, 1);
 548                                 if (!page) {
 549                                         ret = -E2BIG;
 550                                         goto out;
 551                                 }
 552
 553                                 if (kmapped_page) {
 554                                         flush_kernel_dcache_page(kmapped_page);
 555                                         kunmap(kmapped_page);
 556                                         put_arg_page(kmapped_page);
 557                                 }
 558                                 kmapped_page = page;
 559                                 kaddr = kmap(kmapped_page);
 560                                 kpos = pos & PAGE_MASK;
 561                                 flush_arg_page(bprm, kpos, kmapped_page);
 562                         }
 563                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 564                                 ret = -EFAULT;
 565                                 goto out;
 566                         }
 567                 }
 568         }
 569         ret = 0;
 570 out:
 571         if (kmapped_page) {
 572                 flush_kernel_dcache_page(kmapped_page);
 573                 kunmap(kmapped_page);
 574                 put_arg_page(kmapped_page);
 575         }
 576         return ret;
 577 }
 578
 579 /*
 580  * Like copy_strings, but get argv and its values from kernel memory.
 581  */
 582 int copy_strings_kernel(int argc, const char *const *__argv,
 583                         struct linux_binprm *bprm)
 584 {
 585         int r;
 586         mm_segment_t oldfs = get_fs();
 587         struct user_arg_ptr argv = {
 588                 .ptr.native = (const char __user *const  __user *)__argv,
 589         };
 590
 591         set_fs(KERNEL_DS);
 592         r = copy_strings(argc, argv, bprm);
 593         set_fs(oldfs);
 594
 595         return r;
 596 }
 597 EXPORT_SYMBOL(copy_strings_kernel);
 598
 599 #ifdef CONFIG_MMU
 600
 601 /*
 602  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 603  * the binfmt code determines where the new stack should reside, we shift it to
 604  * its final location.  The process proceeds as follows:
 605  *
 606  * 1) Use shift to calculate the new vma endpoints.
 607  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 608  *    arguments passed to subsequent functions are consistent.
 609  * 3) Move vma's page tables to the new range.
 610  * 4) Free up any cleared pgd range.
 611  * 5) Shrink the vma to cover only the new range.
 612  */
 613 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 614 {
 615         struct mm_struct *mm = vma->vm_mm;
 616         unsigned long old_start = vma->vm_start;
 617         unsigned long old_end = vma->vm_end;
 618         unsigned long length = old_end - old_start;
 619         unsigned long new_start = old_start - shift;
 620         unsigned long new_end = old_end - shift;
 621         struct mmu_gather tlb;
 622
 623         BUG_ON(new_start > new_end);
 624
 625         /*
 626          * ensure there are no vmas between where we want to go
 627          * and where we are
 628          */
 629         if (vma != find_vma(mm, new_start))
 630                 return -EFAULT;
 631
 632         /*
 633          * cover the whole range: [new_start, old_end)
 634          */
 635         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 636                 return -ENOMEM;
 637
 638         /*
 639          * move the page tables downwards, on failure we rely on
 640          * process cleanup to remove whatever mess we made.
 641          */
 642         if (length != move_page_tables(vma, old_start,
 643                                        vma, new_start, length, false))
 644                 return -ENOMEM;
 645
 646         lru_add_drain();
 647         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 648         if (new_end > old_start) {
 649                 /*
 650                  * when the old and new regions overlap clear from new_end.
 651                  */
 652                 free_pgd_range(&tlb, new_end, old_end, new_end,
 653                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 654         } else {
 655                 /*
 656                  * otherwise, clean from old_start; this is done to not touch
 657                  * the address space in [new_end, old_start) some architectures
 658                  * have constraints on va-space that make this illegal (IA64) -
 659                  * for the others its just a little faster.
 660                  */
 661                 free_pgd_range(&tlb, old_start, old_end, new_end,
 662                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 663         }
 664         tlb_finish_mmu(&tlb, old_start, old_end);
 665
 666         /*
 667          * Shrink the vma to just the new range.  Always succeeds.
 668          */
 669         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 670
 671         return 0;
 672 }
 673
 674 /*
 675  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 676  * the stack is optionally relocated, and some extra space is added.
 677  */
 678 int setup_arg_pages(struct linux_binprm *bprm,
 679                     unsigned long stack_top,
 680                     int executable_stack)
 681 {
 682         unsigned long ret;
 683         unsigned long stack_shift;
 684         struct mm_struct *mm = current->mm;
 685         struct vm_area_struct *vma = bprm->vma;
 686         struct vm_area_struct *prev = NULL;
 687         unsigned long vm_flags;
 688         unsigned long stack_base;
 689         unsigned long stack_size;
 690         unsigned long stack_expand;
 691         unsigned long rlim_stack;
 692
 693 #ifdef CONFIG_STACK_GROWSUP
 694         /* Limit stack size */
 695         stack_base = rlimit_max(RLIMIT_STACK);
 696         if (stack_base > STACK_SIZE_MAX)
 697                 stack_base = STACK_SIZE_MAX;
 698
 699         /* Add space for stack randomization. */
 700         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 701
 702         /* Make sure we didn't let the argument array grow too large. */
 703         if (vma->vm_end - vma->vm_start > stack_base)
 704                 return -ENOMEM;
 705
 706         stack_base = PAGE_ALIGN(stack_top - stack_base);
 707
 708         stack_shift = vma->vm_start - stack_base;
 709         mm->arg_start = bprm->p - stack_shift;
 710         bprm->p = vma->vm_end - stack_shift;
 711 #else
 712         stack_top = arch_align_stack(stack_top);
 713         stack_top = PAGE_ALIGN(stack_top);
 714
 715         if (unlikely(stack_top < mmap_min_addr) ||
 716             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 717                 return -ENOMEM;
 718
 719         stack_shift = vma->vm_end - stack_top;
 720
 721         bprm->p -= stack_shift;
 722         mm->arg_start = bprm->p;
 723 #endif
 724
 725         if (bprm->loader)
 726                 bprm->loader -= stack_shift;
 727         bprm->exec -= stack_shift;
 728
 729         if (down_write_killable(&mm->mmap_sem))
 730                 return -EINTR;
 731
 732         vm_flags = VM_STACK_FLAGS;
 733
 734         /*
 735          * Adjust stack execute permissions; explicitly enable for
 736          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 737          * (arch default) otherwise.
 738          */
 739         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 740                 vm_flags |= VM_EXEC;
 741         else if (executable_stack == EXSTACK_DISABLE_X)
 742                 vm_flags &= ~VM_EXEC;
 743         vm_flags |= mm->def_flags;
 744         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 745
 746         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 747                         vm_flags);
 748         if (ret)
 749                 goto out_unlock;
 750         BUG_ON(prev != vma);
 751
 752         /* Move stack pages down in memory. */
 753         if (stack_shift) {
 754                 ret = shift_arg_pages(vma, stack_shift);
 755                 if (ret)
 756                         goto out_unlock;
 757         }
 758
 759         /* mprotect_fixup is overkill to remove the temporary stack flags */
 760         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 761
 762         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 763         stack_size = vma->vm_end - vma->vm_start;
 764         /*
 765          * Align this down to a page boundary as expand_stack
 766          * will align it up.
 767          */
 768         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 769 #ifdef CONFIG_STACK_GROWSUP
 770         if (stack_size + stack_expand > rlim_stack)
 771                 stack_base = vma->vm_start + rlim_stack;
 772         else
 773                 stack_base = vma->vm_end + stack_expand;
 774 #else
 775         if (stack_size + stack_expand > rlim_stack)
 776                 stack_base = vma->vm_end - rlim_stack;
 777         else
 778                 stack_base = vma->vm_start - stack_expand;
 779 #endif
 780         current->mm->start_stack = bprm->p;
 781         ret = expand_stack(vma, stack_base);
 782         if (ret)
 783                 ret = -EFAULT;
 784
 785 out_unlock:
 786         up_write(&mm->mmap_sem);
 787         return ret;
 788 }
 789 EXPORT_SYMBOL(setup_arg_pages);
 790
 791 #else
 792
 793 /*
 794  * Transfer the program arguments and environment from the holding pages
 795  * onto the stack. The provided stack pointer is adjusted accordingly.
 796  */
 797 int transfer_args_to_stack(struct linux_binprm *bprm,
 798                            unsigned long *sp_location)
 799 {
 800         unsigned long index, stop, sp;
 801         int ret = 0;
 802
 803         stop = bprm->p >> PAGE_SHIFT;
 804         sp = *sp_location;
 805
 806         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 807                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 808                 char *src = kmap(bprm->page[index]) + offset;
 809                 sp -= PAGE_SIZE - offset;
 810                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 811                         ret = -EFAULT;
 812                 kunmap(bprm->page[index]);
 813                 if (ret)
 814                         goto out;
 815         }
 816
 817         *sp_location = sp;
 818
 819 out:
 820         return ret;
 821 }
 822 EXPORT_SYMBOL(transfer_args_to_stack);
 823
 824 #endif /* CONFIG_MMU */
 825
 826 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 827 {
 828         struct file *file;
 829         int err;
 830         struct open_flags open_exec_flags = {
 831                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 832                 .acc_mode = MAY_EXEC,
 833                 .intent = LOOKUP_OPEN,
 834                 .lookup_flags = LOOKUP_FOLLOW,
 835         };
 836
 837         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 838                 return ERR_PTR(-EINVAL);
 839         if (flags & AT_SYMLINK_NOFOLLOW)
 840                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 841         if (flags & AT_EMPTY_PATH)
 842                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 843
 844         file = do_filp_open(fd, name, &open_exec_flags);
 845         if (IS_ERR(file))
 846                 goto out;
 847
 848         err = -EACCES;
 849         if (!S_ISREG(file_inode(file)->i_mode))
 850                 goto exit;
 851
 852         if (path_noexec(&file->f_path))
 853                 goto exit;
 854
 855         err = deny_write_access(file);
 856         if (err)
 857                 goto exit;
 858
 859         if (name->name[0] != '\0')
 860                 fsnotify_open(file);
 861
 862 out:
 863         return file;
 864
 865 exit:
 866         fput(file);
 867         return ERR_PTR(err);
 868 }
 869
 870 struct file *open_exec(const char *name)
 871 {
 872         struct filename *filename = getname_kernel(name);
 873         struct file *f = ERR_CAST(filename);
 874
 875         if (!IS_ERR(filename)) {
 876                 f = do_open_execat(AT_FDCWD, filename, 0);
 877                 putname(filename);
 878         }
 879         return f;
 880 }
 881 EXPORT_SYMBOL(open_exec);
 882
 883 int kernel_read(struct file *file, loff_t offset,
 884                 char *addr, unsigned long count)
 885 {
 886         mm_segment_t old_fs;
 887         loff_t pos = offset;
 888         int result;
 889
 890         old_fs = get_fs();
 891         set_fs(get_ds());
 892         /* The cast to a user pointer is valid due to the set_fs() */
 893         result = vfs_read(file, (void __user *)addr, count, &pos);
 894         set_fs(old_fs);
 895         return result;
 896 }
 897
 898 EXPORT_SYMBOL(kernel_read);
 899
 900 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 901                      loff_t max_size, enum kernel_read_file_id id)
 902 {
 903         loff_t i_size, pos;
 904         ssize_t bytes = 0;
 905         int ret;
 906
 907         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 908                 return -EINVAL;
 909
 910         ret = security_kernel_read_file(file, id);
 911         if (ret)
 912                 return ret;
 913
 914         ret = deny_write_access(file);
 915         if (ret)
 916                 return ret;
 917
 918         i_size = i_size_read(file_inode(file));
 919         if (max_size > 0 && i_size > max_size) {
 920                 ret = -EFBIG;
 921                 goto out;
 922         }
 923         if (i_size <= 0) {
 924                 ret = -EINVAL;
 925                 goto out;
 926         }
 927
 928         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 929                 *buf = vmalloc(i_size);
 930         if (!*buf) {
 931                 ret = -ENOMEM;
 932                 goto out;
 933         }
 934
 935         pos = 0;
 936         while (pos < i_size) {
 937                 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 938                                     i_size - pos);
 939                 if (bytes < 0) {
 940                         ret = bytes;
 941                         goto out_free;
 942                 }
 943
 944                 if (bytes == 0)
 945                         break;
 946                 pos += bytes;
 947         }
 948
 949         if (pos != i_size) {
 950                 ret = -EIO;
 951                 goto out_free;
 952         }
 953
 954         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 955         if (!ret)
 956                 *size = pos;
 957
 958 out_free:
 959         if (ret < 0) {
 960                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 961                         vfree(*buf);
 962                         *buf = NULL;
 963                 }
 964         }
 965
 966 out:
 967         allow_write_access(file);
 968         return ret;
 969 }
 970 EXPORT_SYMBOL_GPL(kernel_read_file);
 971
 972 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 973                                loff_t max_size, enum kernel_read_file_id id)
 974 {
 975         struct file *file;
 976         int ret;
 977
 978         if (!path || !*path)
 979                 return -EINVAL;
 980
 981         file = filp_open(path, O_RDONLY, 0);
 982         if (IS_ERR(file))
 983                 return PTR_ERR(file);
 984
 985         ret = kernel_read_file(file, buf, size, max_size, id);
 986         fput(file);
 987         return ret;
 988 }
 989 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 990
 991 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 992                              enum kernel_read_file_id id)
 993 {
 994         struct fd f = fdget(fd);
 995         int ret = -EBADF;
 996
 997         if (!f.file)
 998                 goto out;
 999
1000         ret = kernel_read_file(f.file, buf, size, max_size, id);
1001 out:
1002         fdput(f);
1003         return ret;
1004 }
1005 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1006
1007 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1008 {
1009         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1010         if (res > 0)
1011                 flush_icache_range(addr, addr + len);
1012         return res;
1013 }
1014 EXPORT_SYMBOL(read_code);
1015
1016 static int exec_mmap(struct mm_struct *mm)
1017 {
1018         struct task_struct *tsk;
1019         struct mm_struct *old_mm, *active_mm;
1020
1021         /* Notify parent that we're no longer interested in the old VM */
1022         tsk = current;
1023         old_mm = current->mm;
1024         mm_release(tsk, old_mm);
1025
1026         if (old_mm) {
1027                 sync_mm_rss(old_mm);
1028                 /*
1029                  * Make sure that if there is a core dump in progress
1030                  * for the old mm, we get out and die instead of going
1031                  * through with the exec.  We must hold mmap_sem around
1032                  * checking core_state and changing tsk->mm.
1033                  */
1034                 down_read(&old_mm->mmap_sem);
1035                 if (unlikely(old_mm->core_state)) {
1036                         up_read(&old_mm->mmap_sem);
1037                         return -EINTR;
1038                 }
1039         }
1040         task_lock(tsk);
1041         active_mm = tsk->active_mm;
1042         tsk->mm = mm;
1043         tsk->active_mm = mm;
1044         activate_mm(active_mm, mm);
1045         tsk->mm->vmacache_seqnum = 0;
1046         vmacache_flush(tsk);
1047         task_unlock(tsk);
1048         if (old_mm) {
1049                 up_read(&old_mm->mmap_sem);
1050                 BUG_ON(active_mm != old_mm);
1051                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1052                 mm_update_next_owner(old_mm);
1053                 mmput(old_mm);
1054                 return 0;
1055         }
1056         mmdrop(active_mm);
1057         return 0;
1058 }
1059
1060 /*
1061  * This function makes sure the current process has its own signal table,
1062  * so that flush_signal_handlers can later reset the handlers without
1063  * disturbing other processes.  (Other processes might share the signal
1064  * table via the CLONE_SIGHAND option to clone().)
1065  */
1066 static int de_thread(struct task_struct *tsk)
1067 {
1068         struct signal_struct *sig = tsk->signal;
1069         struct sighand_struct *oldsighand = tsk->sighand;
1070         spinlock_t *lock = &oldsighand->siglock;
1071
1072         if (thread_group_empty(tsk))
1073                 goto no_thread_group;
1074
1075         /*
1076          * Kill all other threads in the thread group.
1077          */
1078         spin_lock_irq(lock);
1079         if (signal_group_exit(sig)) {
1080                 /*
1081                  * Another group action in progress, just
1082                  * return so that the signal is processed.
1083                  */
1084                 spin_unlock_irq(lock);
1085                 return -EAGAIN;
1086         }
1087
1088         sig->group_exit_task = tsk;
1089         sig->notify_count = zap_other_threads(tsk);
1090         if (!thread_group_leader(tsk))
1091                 sig->notify_count--;
1092
1093         while (sig->notify_count) {
1094                 __set_current_state(TASK_KILLABLE);
1095                 spin_unlock_irq(lock);
1096                 schedule();
1097                 if (unlikely(__fatal_signal_pending(tsk)))
1098                         goto killed;
1099                 spin_lock_irq(lock);
1100         }
1101         spin_unlock_irq(lock);
1102
1103         /*
1104          * At this point all other threads have exited, all we have to
1105          * do is to wait for the thread group leader to become inactive,
1106          * and to assume its PID:
1107          */
1108         if (!thread_group_leader(tsk)) {
1109                 struct task_struct *leader = tsk->group_leader;
1110
1111                 for (;;) {
1112                         threadgroup_change_begin(tsk);
1113                         write_lock_irq(&tasklist_lock);
1114                         /*
1115                          * Do this under tasklist_lock to ensure that
1116                          * exit_notify() can't miss ->group_exit_task
1117                          */
1118                         sig->notify_count = -1;
1119                         if (likely(leader->exit_state))
1120                                 break;
1121                         __set_current_state(TASK_KILLABLE);
1122                         write_unlock_irq(&tasklist_lock);
1123                         threadgroup_change_end(tsk);
1124                         schedule();
1125                         if (unlikely(__fatal_signal_pending(tsk)))
1126                                 goto killed;
1127                 }
1128
1129                 /*
1130                  * The only record we have of the real-time age of a
1131                  * process, regardless of execs it's done, is start_time.
1132                  * All the past CPU time is accumulated in signal_struct
1133                  * from sister threads now dead.  But in this non-leader
1134                  * exec, nothing survives from the original leader thread,
1135                  * whose birth marks the true age of this process now.
1136                  * When we take on its identity by switching to its PID, we
1137                  * also take its birthdate (always earlier than our own).
1138                  */
1139                 tsk->start_time = leader->start_time;
1140                 tsk->real_start_time = leader->real_start_time;
1141
1142                 BUG_ON(!same_thread_group(leader, tsk));
1143                 BUG_ON(has_group_leader_pid(tsk));
1144                 /*
1145                  * An exec() starts a new thread group with the
1146                  * TGID of the previous thread group. Rehash the
1147                  * two threads with a switched PID, and release
1148                  * the former thread group leader:
1149                  */
1150
1151                 /* Become a process group leader with the old leader's pid.
1152                  * The old leader becomes a thread of the this thread group.
1153                  * Note: The old leader also uses this pid until release_task
1154                  *       is called.  Odd but simple and correct.
1155                  */
1156                 tsk->pid = leader->pid;
1157                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1158                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1159                 transfer_pid(leader, tsk, PIDTYPE_SID);
1160
1161                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1162                 list_replace_init(&leader->sibling, &tsk->sibling);
1163
1164                 tsk->group_leader = tsk;
1165                 leader->group_leader = tsk;
1166
1167                 tsk->exit_signal = SIGCHLD;
1168                 leader->exit_signal = -1;
1169
1170                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1171                 leader->exit_state = EXIT_DEAD;
1172
1173                 /*
1174                  * We are going to release_task()->ptrace_unlink() silently,
1175                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1176                  * the tracer wont't block again waiting for this thread.
1177                  */
1178                 if (unlikely(leader->ptrace))
1179                         __wake_up_parent(leader, leader->parent);
1180                 write_unlock_irq(&tasklist_lock);
1181                 threadgroup_change_end(tsk);
1182
1183                 release_task(leader);
1184         }
1185
1186         sig->group_exit_task = NULL;
1187         sig->notify_count = 0;
1188
1189 no_thread_group:
1190         /* we have changed execution domain */
1191         tsk->exit_signal = SIGCHLD;
1192
1193         exit_itimers(sig);
1194         flush_itimer_signals();
1195
1196         if (atomic_read(&oldsighand->count) != 1) {
1197                 struct sighand_struct *newsighand;
1198                 /*
1199                  * This ->sighand is shared with the CLONE_SIGHAND
1200                  * but not CLONE_THREAD task, switch to the new one.
1201                  */
1202                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1203                 if (!newsighand)
1204                         return -ENOMEM;
1205
1206                 atomic_set(&newsighand->count, 1);
1207                 memcpy(newsighand->action, oldsighand->action,
1208                        sizeof(newsighand->action));
1209
1210                 write_lock_irq(&tasklist_lock);
1211                 spin_lock(&oldsighand->siglock);
1212                 rcu_assign_pointer(tsk->sighand, newsighand);
1213                 spin_unlock(&oldsighand->siglock);
1214                 write_unlock_irq(&tasklist_lock);
1215
1216                 __cleanup_sighand(oldsighand);
1217         }
1218
1219         BUG_ON(!thread_group_leader(tsk));
1220         return 0;
1221
1222 killed:
1223         /* protects against exit_notify() and __exit_signal() */
1224         read_lock(&tasklist_lock);
1225         sig->group_exit_task = NULL;
1226         sig->notify_count = 0;
1227         read_unlock(&tasklist_lock);
1228         return -EAGAIN;
1229 }
1230
1231 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1232 {
1233         task_lock(tsk);
1234         strncpy(buf, tsk->comm, buf_size);
1235         task_unlock(tsk);
1236         return buf;
1237 }
1238 EXPORT_SYMBOL_GPL(__get_task_comm);
1239
1240 /*
1241  * These functions flushes out all traces of the currently running executable
1242  * so that a new one can be started
1243  */
1244
1245 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1246 {
1247         task_lock(tsk);
1248         trace_task_rename(tsk, buf);
1249         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1250         task_unlock(tsk);
1251         perf_event_comm(tsk, exec);
1252 }
1253
1254 int flush_old_exec(struct linux_binprm * bprm)
1255 {
1256         int retval;
1257
1258         /*
1259          * Make sure we have a private signal table and that
1260          * we are unassociated from the previous thread group.
1261          */
1262         retval = de_thread(current);
1263         if (retval)
1264                 goto out;
1265
1266         /*
1267          * Must be called _before_ exec_mmap() as bprm->mm is
1268          * not visibile until then. This also enables the update
1269          * to be lockless.
1270          */
1271         set_mm_exe_file(bprm->mm, bprm->file);
1272
1273         /*
1274          * Release all of the old mmap stuff
1275          */
1276         acct_arg_size(bprm, 0);
1277         retval = exec_mmap(bprm->mm);
1278         if (retval)
1279                 goto out;
1280
1281         bprm->mm = NULL;                /* We're using it now */
1282
1283         set_fs(USER_DS);
1284         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1285                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1286         flush_thread();
1287         current->personality &= ~bprm->per_clear;
1288
1289         /*
1290          * We have to apply CLOEXEC before we change whether the process is
1291          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1292          * trying to access the should-be-closed file descriptors of a process
1293          * undergoing exec(2).
1294          */
1295         do_close_on_exec(current->files);
1296         return 0;
1297
1298 out:
1299         return retval;
1300 }
1301 EXPORT_SYMBOL(flush_old_exec);
1302
1303 void would_dump(struct linux_binprm *bprm, struct file *file)
1304 {
1305         struct inode *inode = file_inode(file);
1306         if (inode_permission(inode, MAY_READ) < 0) {
1307                 struct user_namespace *old, *user_ns;
1308                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1309
1310                 /* Ensure mm->user_ns contains the executable */
1311                 user_ns = old = bprm->mm->user_ns;
1312                 while ((user_ns != &init_user_ns) &&
1313                        !privileged_wrt_inode_uidgid(user_ns, inode))
1314                         user_ns = user_ns->parent;
1315
1316                 if (old != user_ns) {
1317                         bprm->mm->user_ns = get_user_ns(user_ns);
1318                         put_user_ns(old);
1319                 }
1320         }
1321 }
1322 EXPORT_SYMBOL(would_dump);
1323
1324 void setup_new_exec(struct linux_binprm * bprm)
1325 {
1326         arch_pick_mmap_layout(current->mm);
1327
1328         /* This is the point of no return */
1329         current->sas_ss_sp = current->sas_ss_size = 0;
1330
1331         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1332                 set_dumpable(current->mm, SUID_DUMP_USER);
1333         else
1334                 set_dumpable(current->mm, suid_dumpable);
1335
1336         perf_event_exec();
1337         __set_task_comm(current, kbasename(bprm->filename), true);
1338
1339         /* Set the new mm task size. We have to do that late because it may
1340          * depend on TIF_32BIT which is only updated in flush_thread() on
1341          * some architectures like powerpc
1342          */
1343         current->mm->task_size = TASK_SIZE;
1344
1345         /* install the new credentials */
1346         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1347             !gid_eq(bprm->cred->gid, current_egid())) {
1348                 current->pdeath_signal = 0;
1349         } else {
1350                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1351                         set_dumpable(current->mm, suid_dumpable);
1352         }
1353
1354         /* An exec changes our domain. We are no longer part of the thread
1355            group */
1356         current->self_exec_id++;
1357         flush_signal_handlers(current, 0);
1358 }
1359 EXPORT_SYMBOL(setup_new_exec);
1360
1361 /*
1362  * Prepare credentials and lock ->cred_guard_mutex.
1363  * install_exec_creds() commits the new creds and drops the lock.
1364  * Or, if exec fails before, free_bprm() should release ->cred and
1365  * and unlock.
1366  */
1367 int prepare_bprm_creds(struct linux_binprm *bprm)
1368 {
1369         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1370                 return -ERESTARTNOINTR;
1371
1372         bprm->cred = prepare_exec_creds();
1373         if (likely(bprm->cred))
1374                 return 0;
1375
1376         mutex_unlock(&current->signal->cred_guard_mutex);
1377         return -ENOMEM;
1378 }
1379
1380 static void free_bprm(struct linux_binprm *bprm)
1381 {
1382         free_arg_pages(bprm);
1383         if (bprm->cred) {
1384                 mutex_unlock(&current->signal->cred_guard_mutex);
1385                 abort_creds(bprm->cred);
1386         }
1387         if (bprm->file) {
1388                 allow_write_access(bprm->file);
1389                 fput(bprm->file);
1390         }
1391         /* If a binfmt changed the interp, free it. */
1392         if (bprm->interp != bprm->filename)
1393                 kfree(bprm->interp);
1394         kfree(bprm);
1395 }
1396
1397 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1398 {
1399         /* If a binfmt changed the interp, free it first. */
1400         if (bprm->interp != bprm->filename)
1401                 kfree(bprm->interp);
1402         bprm->interp = kstrdup(interp, GFP_KERNEL);
1403         if (!bprm->interp)
1404                 return -ENOMEM;
1405         return 0;
1406 }
1407 EXPORT_SYMBOL(bprm_change_interp);
1408
1409 /*
1410  * install the new credentials for this executable
1411  */
1412 void install_exec_creds(struct linux_binprm *bprm)
1413 {
1414         security_bprm_committing_creds(bprm);
1415
1416         commit_creds(bprm->cred);
1417         bprm->cred = NULL;
1418
1419         /*
1420          * Disable monitoring for regular users
1421          * when executing setuid binaries. Must
1422          * wait until new credentials are committed
1423          * by commit_creds() above
1424          */
1425         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1426                 perf_event_exit_task(current);
1427         /*
1428          * cred_guard_mutex must be held at least to this point to prevent
1429          * ptrace_attach() from altering our determination of the task's
1430          * credentials; any time after this it may be unlocked.
1431          */
1432         security_bprm_committed_creds(bprm);
1433         mutex_unlock(&current->signal->cred_guard_mutex);
1434 }
1435 EXPORT_SYMBOL(install_exec_creds);
1436
1437 /*
1438  * determine how safe it is to execute the proposed program
1439  * - the caller must hold ->cred_guard_mutex to protect against
1440  *   PTRACE_ATTACH or seccomp thread-sync
1441  */
1442 static void check_unsafe_exec(struct linux_binprm *bprm)
1443 {
1444         struct task_struct *p = current, *t;
1445         unsigned n_fs;
1446
1447         if (p->ptrace) {
1448                 if (ptracer_capable(p, current_user_ns()))
1449                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1450                 else
1451                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1452         }
1453
1454         /*
1455          * This isn't strictly necessary, but it makes it harder for LSMs to
1456          * mess up.
1457          */
1458         if (task_no_new_privs(current))
1459                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1460
1461         t = p;
1462         n_fs = 1;
1463         spin_lock(&p->fs->lock);
1464         rcu_read_lock();
1465         while_each_thread(p, t) {
1466                 if (t->fs == p->fs)
1467                         n_fs++;
1468         }
1469         rcu_read_unlock();
1470
1471         if (p->fs->users > n_fs)
1472                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1473         else
1474                 p->fs->in_exec = 1;
1475         spin_unlock(&p->fs->lock);
1476 }
1477
1478 static void bprm_fill_uid(struct linux_binprm *bprm)
1479 {
1480         struct inode *inode;
1481         unsigned int mode;
1482         kuid_t uid;
1483         kgid_t gid;
1484
1485         /*
1486          * Since this can be called multiple times (via prepare_binprm),
1487          * we must clear any previous work done when setting set[ug]id
1488          * bits from any earlier bprm->file uses (for example when run
1489          * first for a setuid script then again for its interpreter).
1490          */
1491         bprm->cred->euid = current_euid();
1492         bprm->cred->egid = current_egid();
1493
1494         if (!mnt_may_suid(bprm->file->f_path.mnt))
1495                 return;
1496
1497         if (task_no_new_privs(current))
1498                 return;
1499
1500         inode = file_inode(bprm->file);
1501         mode = READ_ONCE(inode->i_mode);
1502         if (!(mode & (S_ISUID|S_ISGID)))
1503                 return;
1504
1505         /* Be careful if suid/sgid is set */
1506         inode_lock(inode);
1507
1508         /* reload atomically mode/uid/gid now that lock held */
1509         mode = inode->i_mode;
1510         uid = inode->i_uid;
1511         gid = inode->i_gid;
1512         inode_unlock(inode);
1513
1514         /* We ignore suid/sgid if there are no mappings for them in the ns */
1515         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1516                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1517                 return;
1518
1519         if (mode & S_ISUID) {
1520                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1521                 bprm->cred->euid = uid;
1522         }
1523
1524         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1525                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1526                 bprm->cred->egid = gid;
1527         }
1528 }
1529
1530 /*
1531  * Fill the binprm structure from the inode.
1532  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1533  *
1534  * This may be called multiple times for binary chains (scripts for example).
1535  */
1536 int prepare_binprm(struct linux_binprm *bprm)
1537 {
1538         int retval;
1539
1540         bprm_fill_uid(bprm);
1541
1542         /* fill in binprm security blob */
1543         retval = security_bprm_set_creds(bprm);
1544         if (retval)
1545                 return retval;
1546         bprm->cred_prepared = 1;
1547
1548         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1549         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1550 }
1551
1552 EXPORT_SYMBOL(prepare_binprm);
1553
1554 /*
1555  * Arguments are '\0' separated strings found at the location bprm->p
1556  * points to; chop off the first by relocating brpm->p to right after
1557  * the first '\0' encountered.
1558  */
1559 int remove_arg_zero(struct linux_binprm *bprm)
1560 {
1561         int ret = 0;
1562         unsigned long offset;
1563         char *kaddr;
1564         struct page *page;
1565
1566         if (!bprm->argc)
1567                 return 0;
1568
1569         do {
1570                 offset = bprm->p & ~PAGE_MASK;
1571                 page = get_arg_page(bprm, bprm->p, 0);
1572                 if (!page) {
1573                         ret = -EFAULT;
1574                         goto out;
1575                 }
1576                 kaddr = kmap_atomic(page);
1577
1578                 for (; offset < PAGE_SIZE && kaddr[offset];
1579                                 offset++, bprm->p++)
1580                         ;
1581
1582                 kunmap_atomic(kaddr);
1583                 put_arg_page(page);
1584         } while (offset == PAGE_SIZE);
1585
1586         bprm->p++;
1587         bprm->argc--;
1588         ret = 0;
1589
1590 out:
1591         return ret;
1592 }
1593 EXPORT_SYMBOL(remove_arg_zero);
1594
1595 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1596 /*
1597  * cycle the list of binary formats handler, until one recognizes the image
1598  */
1599 int search_binary_handler(struct linux_binprm *bprm)
1600 {
1601         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1602         struct linux_binfmt *fmt;
1603         int retval;
1604
1605         /* This allows 4 levels of binfmt rewrites before failing hard. */
1606         if (bprm->recursion_depth > 5)
1607                 return -ELOOP;
1608
1609         retval = security_bprm_check(bprm);
1610         if (retval)
1611                 return retval;
1612
1613         retval = -ENOENT;
1614  retry:
1615         read_lock(&binfmt_lock);
1616         list_for_each_entry(fmt, &formats, lh) {
1617                 if (!try_module_get(fmt->module))
1618                         continue;
1619                 read_unlock(&binfmt_lock);
1620                 bprm->recursion_depth++;
1621                 retval = fmt->load_binary(bprm);
1622                 read_lock(&binfmt_lock);
1623                 put_binfmt(fmt);
1624                 bprm->recursion_depth--;
1625                 if (retval < 0 && !bprm->mm) {
1626                         /* we got to flush_old_exec() and failed after it */
1627                         read_unlock(&binfmt_lock);
1628                         force_sigsegv(SIGSEGV, current);
1629                         return retval;
1630                 }
1631                 if (retval != -ENOEXEC || !bprm->file) {
1632                         read_unlock(&binfmt_lock);
1633                         return retval;
1634                 }
1635         }
1636         read_unlock(&binfmt_lock);
1637
1638         if (need_retry) {
1639                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1640                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1641                         return retval;
1642                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1643                         return retval;
1644                 need_retry = false;
1645                 goto retry;
1646         }
1647
1648         return retval;
1649 }
1650 EXPORT_SYMBOL(search_binary_handler);
1651
1652 static int exec_binprm(struct linux_binprm *bprm)
1653 {
1654         pid_t old_pid, old_vpid;
1655         int ret;
1656
1657         /* Need to fetch pid before load_binary changes it */
1658         old_pid = current->pid;
1659         rcu_read_lock();
1660         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1661         rcu_read_unlock();
1662
1663         ret = search_binary_handler(bprm);
1664         if (ret >= 0) {
1665                 audit_bprm(bprm);
1666                 trace_sched_process_exec(current, old_pid, bprm);
1667                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1668                 proc_exec_connector(current);
1669         }
1670
1671         return ret;
1672 }
1673
1674 /*
1675  * sys_execve() executes a new program.
1676  */
1677 static int do_execveat_common(int fd, struct filename *filename,
1678                               struct user_arg_ptr argv,
1679                               struct user_arg_ptr envp,
1680                               int flags)
1681 {
1682         char *pathbuf = NULL;
1683         struct linux_binprm *bprm;
1684         struct file *file;
1685         struct files_struct *displaced;
1686         int retval;
1687
1688         if (IS_ERR(filename))
1689                 return PTR_ERR(filename);
1690
1691         /*
1692          * We move the actual failure in case of RLIMIT_NPROC excess from
1693          * set*uid() to execve() because too many poorly written programs
1694          * don't check setuid() return code.  Here we additionally recheck
1695          * whether NPROC limit is still exceeded.
1696          */
1697         if ((current->flags & PF_NPROC_EXCEEDED) &&
1698             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1699                 retval = -EAGAIN;
1700                 goto out_ret;
1701         }
1702
1703         /* We're below the limit (still or again), so we don't want to make
1704          * further execve() calls fail. */
1705         current->flags &= ~PF_NPROC_EXCEEDED;
1706
1707         retval = unshare_files(&displaced);
1708         if (retval)
1709                 goto out_ret;
1710
1711         retval = -ENOMEM;
1712         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1713         if (!bprm)
1714                 goto out_files;
1715
1716         retval = prepare_bprm_creds(bprm);
1717         if (retval)
1718                 goto out_free;
1719
1720         check_unsafe_exec(bprm);
1721         current->in_execve = 1;
1722
1723         file = do_open_execat(fd, filename, flags);
1724         retval = PTR_ERR(file);
1725         if (IS_ERR(file))
1726                 goto out_unmark;
1727
1728         sched_exec();
1729
1730         bprm->file = file;
1731         if (fd == AT_FDCWD || filename->name[0] == '/') {
1732                 bprm->filename = filename->name;
1733         } else {
1734                 if (filename->name[0] == '\0')
1735                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1736                 else
1737                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1738                                             fd, filename->name);
1739                 if (!pathbuf) {
1740                         retval = -ENOMEM;
1741                         goto out_unmark;
1742                 }
1743                 /*
1744                  * Record that a name derived from an O_CLOEXEC fd will be
1745                  * inaccessible after exec. Relies on having exclusive access to
1746                  * current->files (due to unshare_files above).
1747                  */
1748                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1749                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1750                 bprm->filename = pathbuf;
1751         }
1752         bprm->interp = bprm->filename;
1753
1754         retval = bprm_mm_init(bprm);
1755         if (retval)
1756                 goto out_unmark;
1757
1758         bprm->argc = count(argv, MAX_ARG_STRINGS);
1759         if ((retval = bprm->argc) < 0)
1760                 goto out;
1761
1762         bprm->envc = count(envp, MAX_ARG_STRINGS);
1763         if ((retval = bprm->envc) < 0)
1764                 goto out;
1765
1766         retval = prepare_binprm(bprm);
1767         if (retval < 0)
1768                 goto out;
1769
1770         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1771         if (retval < 0)
1772                 goto out;
1773
1774         bprm->exec = bprm->p;
1775         retval = copy_strings(bprm->envc, envp, bprm);
1776         if (retval < 0)
1777                 goto out;
1778
1779         retval = copy_strings(bprm->argc, argv, bprm);
1780         if (retval < 0)
1781                 goto out;
1782
1783         would_dump(bprm, bprm->file);
1784
1785         retval = exec_binprm(bprm);
1786         if (retval < 0)
1787                 goto out;
1788
1789         /* execve succeeded */
1790         current->fs->in_exec = 0;
1791         current->in_execve = 0;
1792         acct_update_integrals(current);
1793         task_numa_free(current);
1794         free_bprm(bprm);
1795         kfree(pathbuf);
1796         putname(filename);
1797         if (displaced)
1798                 put_files_struct(displaced);
1799         return retval;
1800
1801 out:
1802         if (bprm->mm) {
1803                 acct_arg_size(bprm, 0);
1804                 mmput(bprm->mm);
1805         }
1806
1807 out_unmark:
1808         current->fs->in_exec = 0;
1809         current->in_execve = 0;
1810
1811 out_free:
1812         free_bprm(bprm);
1813         kfree(pathbuf);
1814
1815 out_files:
1816         if (displaced)
1817                 reset_files_struct(displaced);
1818 out_ret:
1819         putname(filename);
1820         return retval;
1821 }
1822
1823 int do_execve(struct filename *filename,
1824         const char __user *const __user *__argv,
1825         const char __user *const __user *__envp)
1826 {
1827         struct user_arg_ptr argv = { .ptr.native = __argv };
1828         struct user_arg_ptr envp = { .ptr.native = __envp };
1829         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1830 }
1831
1832 int do_execveat(int fd, struct filename *filename,
1833                 const char __user *const __user *__argv,
1834                 const char __user *const __user *__envp,
1835                 int flags)
1836 {
1837         struct user_arg_ptr argv = { .ptr.native = __argv };
1838         struct user_arg_ptr envp = { .ptr.native = __envp };
1839
1840         return do_execveat_common(fd, filename, argv, envp, flags);
1841 }
1842
1843 #ifdef CONFIG_COMPAT
1844 static int compat_do_execve(struct filename *filename,
1845         const compat_uptr_t __user *__argv,
1846         const compat_uptr_t __user *__envp)
1847 {
1848         struct user_arg_ptr argv = {
1849                 .is_compat = true,
1850                 .ptr.compat = __argv,
1851         };
1852         struct user_arg_ptr envp = {
1853                 .is_compat = true,
1854                 .ptr.compat = __envp,
1855         };
1856         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1857 }
1858
1859 static int compat_do_execveat(int fd, struct filename *filename,
1860                               const compat_uptr_t __user *__argv,
1861                               const compat_uptr_t __user *__envp,
1862                               int flags)
1863 {
1864         struct user_arg_ptr argv = {
1865                 .is_compat = true,
1866                 .ptr.compat = __argv,
1867         };
1868         struct user_arg_ptr envp = {
1869                 .is_compat = true,
1870                 .ptr.compat = __envp,
1871         };
1872         return do_execveat_common(fd, filename, argv, envp, flags);
1873 }
1874 #endif
1875
1876 void set_binfmt(struct linux_binfmt *new)
1877 {
1878         struct mm_struct *mm = current->mm;
1879
1880         if (mm->binfmt)
1881                 module_put(mm->binfmt->module);
1882
1883         mm->binfmt = new;
1884         if (new)
1885                 __module_get(new->module);
1886 }
1887 EXPORT_SYMBOL(set_binfmt);
1888
1889 /*
1890  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1891  */
1892 void set_dumpable(struct mm_struct *mm, int value)
1893 {
1894         unsigned long old, new;
1895
1896         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1897                 return;
1898
1899         do {
1900                 old = ACCESS_ONCE(mm->flags);
1901                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1902         } while (cmpxchg(&mm->flags, old, new) != old);
1903 }
1904
1905 SYSCALL_DEFINE3(execve,
1906                 const char __user *, filename,
1907                 const char __user *const __user *, argv,
1908                 const char __user *const __user *, envp)
1909 {
1910         return do_execve(getname(filename), argv, envp);
1911 }
1912
1913 SYSCALL_DEFINE5(execveat,
1914                 int, fd, const char __user *, filename,
1915                 const char __user *const __user *, argv,
1916                 const char __user *const __user *, envp,
1917                 int, flags)
1918 {
1919         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1920
1921         return do_execveat(fd,
1922                            getname_flags(filename, lookup_flags, NULL),
1923                            argv, envp, flags);
1924 }
1925
1926 #ifdef CONFIG_COMPAT
1927 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1928         const compat_uptr_t __user *, argv,
1929         const compat_uptr_t __user *, envp)
1930 {
1931         return compat_do_execve(getname(filename), argv, envp);
1932 }
1933
1934 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1935                        const char __user *, filename,
1936                        const compat_uptr_t __user *, argv,
1937                        const compat_uptr_t __user *, envp,
1938                        int,  flags)
1939 {
1940         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1941
1942         return compat_do_execveat(fd,
1943                                   getname_flags(filename, lookup_flags, NULL),
1944                                   argv, envp, flags);
1945 }
1946 #endif