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/sched/mm.h>
  36 #include <linux/sched/coredump.h>
  37 #include <linux/sched/signal.h>
  38 #include <linux/sched/numa_balancing.h>
  39 #include <linux/sched/task.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/perf_event.h>
  42 #include <linux/highmem.h>
  43 #include <linux/spinlock.h>
  44 #include <linux/key.h>
  45 #include <linux/personality.h>
  46 #include <linux/binfmts.h>
  47 #include <linux/utsname.h>
  48 #include <linux/pid_namespace.h>
  49 #include <linux/module.h>
  50 #include <linux/namei.h>
  51 #include <linux/mount.h>
  52 #include <linux/security.h>
  53 #include <linux/syscalls.h>
  54 #include <linux/tsacct_kern.h>
  55 #include <linux/cn_proc.h>
  56 #include <linux/audit.h>
  57 #include <linux/tracehook.h>
  58 #include <linux/kmod.h>
  59 #include <linux/fsnotify.h>
  60 #include <linux/fs_struct.h>
  61 #include <linux/pipe_fs_i.h>
  62 #include <linux/oom.h>
  63 #include <linux/compat.h>
  64 #include <linux/vmalloc.h>
  65
  66 #include <linux/uaccess.h>
  67 #include <asm/mmu_context.h>
  68 #include <asm/tlb.h>
  69
  70 #include <trace/events/task.h>
  71 #include "internal.h"
  72
  73 #include <trace/events/sched.h>
  74
  75 int suid_dumpable = 0;
  76
  77 static LIST_HEAD(formats);
  78 static DEFINE_RWLOCK(binfmt_lock);
  79
  80 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  81 {
  82         BUG_ON(!fmt);
  83         if (WARN_ON(!fmt->load_binary))
  84                 return;
  85         write_lock(&binfmt_lock);
  86         insert ? list_add(&fmt->lh, &formats) :
  87                  list_add_tail(&fmt->lh, &formats);
  88         write_unlock(&binfmt_lock);
  89 }
  90
  91 EXPORT_SYMBOL(__register_binfmt);
  92
  93 void unregister_binfmt(struct linux_binfmt * fmt)
  94 {
  95         write_lock(&binfmt_lock);
  96         list_del(&fmt->lh);
  97         write_unlock(&binfmt_lock);
  98 }
  99
 100 EXPORT_SYMBOL(unregister_binfmt);
 101
 102 static inline void put_binfmt(struct linux_binfmt * fmt)
 103 {
 104         module_put(fmt->module);
 105 }
 106
 107 bool path_noexec(const struct path *path)
 108 {
 109         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 110                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 111 }
 112
 113 #ifdef CONFIG_USELIB
 114 /*
 115  * Note that a shared library must be both readable and executable due to
 116  * security reasons.
 117  *
 118  * Also note that we take the address to load from from the file itself.
 119  */
 120 SYSCALL_DEFINE1(uselib, const char __user *, library)
 121 {
 122         struct linux_binfmt *fmt;
 123         struct file *file;
 124         struct filename *tmp = getname(library);
 125         int error = PTR_ERR(tmp);
 126         static const struct open_flags uselib_flags = {
 127                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 128                 .acc_mode = MAY_READ | MAY_EXEC,
 129                 .intent = LOOKUP_OPEN,
 130                 .lookup_flags = LOOKUP_FOLLOW,
 131         };
 132
 133         if (IS_ERR(tmp))
 134                 goto out;
 135
 136         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 137         putname(tmp);
 138         error = PTR_ERR(file);
 139         if (IS_ERR(file))
 140                 goto out;
 141
 142         error = -EINVAL;
 143         if (!S_ISREG(file_inode(file)->i_mode))
 144                 goto exit;
 145
 146         error = -EACCES;
 147         if (path_noexec(&file->f_path))
 148                 goto exit;
 149
 150         fsnotify_open(file);
 151
 152         error = -ENOEXEC;
 153
 154         read_lock(&binfmt_lock);
 155         list_for_each_entry(fmt, &formats, lh) {
 156                 if (!fmt->load_shlib)
 157                         continue;
 158                 if (!try_module_get(fmt->module))
 159                         continue;
 160                 read_unlock(&binfmt_lock);
 161                 error = fmt->load_shlib(file);
 162                 read_lock(&binfmt_lock);
 163                 put_binfmt(fmt);
 164                 if (error != -ENOEXEC)
 165                         break;
 166         }
 167         read_unlock(&binfmt_lock);
 168 exit:
 169         fput(file);
 170 out:
 171         return error;
 172 }
 173 #endif /* #ifdef CONFIG_USELIB */
 174
 175 #ifdef CONFIG_MMU
 176 /*
 177  * The nascent bprm->mm is not visible until exec_mmap() but it can
 178  * use a lot of memory, account these pages in current->mm temporary
 179  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 180  * change the counter back via acct_arg_size(0).
 181  */
 182 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 183 {
 184         struct mm_struct *mm = current->mm;
 185         long diff = (long)(pages - bprm->vma_pages);
 186
 187         if (!mm || !diff)
 188                 return;
 189
 190         bprm->vma_pages = pages;
 191         add_mm_counter(mm, MM_ANONPAGES, diff);
 192 }
 193
 194 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 195                 int write)
 196 {
 197         struct page *page;
 198         int ret;
 199         unsigned int gup_flags = FOLL_FORCE;
 200
 201 #ifdef CONFIG_STACK_GROWSUP
 202         if (write) {
 203                 ret = expand_downwards(bprm->vma, pos);
 204                 if (ret < 0)
 205                         return NULL;
 206         }
 207 #endif
 208
 209         if (write)
 210                 gup_flags |= FOLL_WRITE;
 211
 212         /*
 213          * We are doing an exec().  'current' is the process
 214          * doing the exec and bprm->mm is the new process's mm.
 215          */
 216         ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 217                         &page, NULL, NULL);
 218         if (ret <= 0)
 219                 return NULL;
 220
 221         if (write) {
 222                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 223                 unsigned long ptr_size, limit;
 224
 225                 /*
 226                  * Since the stack will hold pointers to the strings, we
 227                  * must account for them as well.
 228                  *
 229                  * The size calculation is the entire vma while each arg page is
 230                  * built, so each time we get here it's calculating how far it
 231                  * is currently (rather than each call being just the newly
 232                  * added size from the arg page).  As a result, we need to
 233                  * always add the entire size of the pointers, so that on the
 234                  * last call to get_arg_page() we'll actually have the entire
 235                  * correct size.
 236                  */
 237                 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 238                 if (ptr_size > ULONG_MAX - size)
 239                         goto fail;
 240                 size += ptr_size;
 241
 242                 acct_arg_size(bprm, size / PAGE_SIZE);
 243
 244                 /*
 245                  * We've historically supported up to 32 pages (ARG_MAX)
 246                  * of argument strings even with small stacks
 247                  */
 248                 if (size <= ARG_MAX)
 249                         return page;
 250
 251                 /*
 252                  * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 253                  * (whichever is smaller) for the argv+env strings.
 254                  * This ensures that:
 255                  *  - the remaining binfmt code will not run out of stack space,
 256                  *  - the program will have a reasonable amount of stack left
 257                  *    to work from.
 258                  */
 259                 limit = _STK_LIM / 4 * 3;
 260                 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 261                 if (size > limit)
 262                         goto fail;
 263         }
 264
 265         return page;
 266
 267 fail:
 268         put_page(page);
 269         return NULL;
 270 }
 271
 272 static void put_arg_page(struct page *page)
 273 {
 274         put_page(page);
 275 }
 276
 277 static void free_arg_pages(struct linux_binprm *bprm)
 278 {
 279 }
 280
 281 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 282                 struct page *page)
 283 {
 284         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 285 }
 286
 287 static int __bprm_mm_init(struct linux_binprm *bprm)
 288 {
 289         int err;
 290         struct vm_area_struct *vma = NULL;
 291         struct mm_struct *mm = bprm->mm;
 292
 293         bprm->vma = vma = vm_area_alloc(mm);
 294         if (!vma)
 295                 return -ENOMEM;
 296         vma_set_anonymous(vma);
 297
 298         if (down_write_killable(&mm->mmap_sem)) {
 299                 err = -EINTR;
 300                 goto err_free;
 301         }
 302
 303         /*
 304          * Place the stack at the largest stack address the architecture
 305          * supports. Later, we'll move this to an appropriate place. We don't
 306          * use STACK_TOP because that can depend on attributes which aren't
 307          * configured yet.
 308          */
 309         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 310         vma->vm_end = STACK_TOP_MAX;
 311         vma->vm_start = vma->vm_end - PAGE_SIZE;
 312         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 313         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 314
 315         err = insert_vm_struct(mm, vma);
 316         if (err)
 317                 goto err;
 318
 319         mm->stack_vm = mm->total_vm = 1;
 320         arch_bprm_mm_init(mm, vma);
 321         up_write(&mm->mmap_sem);
 322         bprm->p = vma->vm_end - sizeof(void *);
 323         return 0;
 324 err:
 325         up_write(&mm->mmap_sem);
 326 err_free:
 327         bprm->vma = NULL;
 328         vm_area_free(vma);
 329         return err;
 330 }
 331
 332 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 333 {
 334         return len <= MAX_ARG_STRLEN;
 335 }
 336
 337 #else
 338
 339 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 340 {
 341 }
 342
 343 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 344                 int write)
 345 {
 346         struct page *page;
 347
 348         page = bprm->page[pos / PAGE_SIZE];
 349         if (!page && write) {
 350                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 351                 if (!page)
 352                         return NULL;
 353                 bprm->page[pos / PAGE_SIZE] = page;
 354         }
 355
 356         return page;
 357 }
 358
 359 static void put_arg_page(struct page *page)
 360 {
 361 }
 362
 363 static void free_arg_page(struct linux_binprm *bprm, int i)
 364 {
 365         if (bprm->page[i]) {
 366                 __free_page(bprm->page[i]);
 367                 bprm->page[i] = NULL;
 368         }
 369 }
 370
 371 static void free_arg_pages(struct linux_binprm *bprm)
 372 {
 373         int i;
 374
 375         for (i = 0; i < MAX_ARG_PAGES; i++)
 376                 free_arg_page(bprm, i);
 377 }
 378
 379 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 380                 struct page *page)
 381 {
 382 }
 383
 384 static int __bprm_mm_init(struct linux_binprm *bprm)
 385 {
 386         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 387         return 0;
 388 }
 389
 390 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 391 {
 392         return len <= bprm->p;
 393 }
 394
 395 #endif /* CONFIG_MMU */
 396
 397 /*
 398  * Create a new mm_struct and populate it with a temporary stack
 399  * vm_area_struct.  We don't have enough context at this point to set the stack
 400  * flags, permissions, and offset, so we use temporary values.  We'll update
 401  * them later in setup_arg_pages().
 402  */
 403 static int bprm_mm_init(struct linux_binprm *bprm)
 404 {
 405         int err;
 406         struct mm_struct *mm = NULL;
 407
 408         bprm->mm = mm = mm_alloc();
 409         err = -ENOMEM;
 410         if (!mm)
 411                 goto err;
 412
 413         /* Save current stack limit for all calculations made during exec. */
 414         task_lock(current->group_leader);
 415         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 416         task_unlock(current->group_leader);
 417
 418         err = __bprm_mm_init(bprm);
 419         if (err)
 420                 goto err;
 421
 422         return 0;
 423
 424 err:
 425         if (mm) {
 426                 bprm->mm = NULL;
 427                 mmdrop(mm);
 428         }
 429
 430         return err;
 431 }
 432
 433 struct user_arg_ptr {
 434 #ifdef CONFIG_COMPAT
 435         bool is_compat;
 436 #endif
 437         union {
 438                 const char __user *const __user *native;
 439 #ifdef CONFIG_COMPAT
 440                 const compat_uptr_t __user *compat;
 441 #endif
 442         } ptr;
 443 };
 444
 445 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 446 {
 447         const char __user *native;
 448
 449 #ifdef CONFIG_COMPAT
 450         if (unlikely(argv.is_compat)) {
 451                 compat_uptr_t compat;
 452
 453                 if (get_user(compat, argv.ptr.compat + nr))
 454                         return ERR_PTR(-EFAULT);
 455
 456                 return compat_ptr(compat);
 457         }
 458 #endif
 459
 460         if (get_user(native, argv.ptr.native + nr))
 461                 return ERR_PTR(-EFAULT);
 462
 463         return native;
 464 }
 465
 466 /*
 467  * count() counts the number of strings in array ARGV.
 468  */
 469 static int count(struct user_arg_ptr argv, int max)
 470 {
 471         int i = 0;
 472
 473         if (argv.ptr.native != NULL) {
 474                 for (;;) {
 475                         const char __user *p = get_user_arg_ptr(argv, i);
 476
 477                         if (!p)
 478                                 break;
 479
 480                         if (IS_ERR(p))
 481                                 return -EFAULT;
 482
 483                         if (i >= max)
 484                                 return -E2BIG;
 485                         ++i;
 486
 487                         if (fatal_signal_pending(current))
 488                                 return -ERESTARTNOHAND;
 489                         cond_resched();
 490                 }
 491         }
 492         return i;
 493 }
 494
 495 /*
 496  * 'copy_strings()' copies argument/environment strings from the old
 497  * processes's memory to the new process's stack.  The call to get_user_pages()
 498  * ensures the destination page is created and not swapped out.
 499  */
 500 static int copy_strings(int argc, struct user_arg_ptr argv,
 501                         struct linux_binprm *bprm)
 502 {
 503         struct page *kmapped_page = NULL;
 504         char *kaddr = NULL;
 505         unsigned long kpos = 0;
 506         int ret;
 507
 508         while (argc-- > 0) {
 509                 const char __user *str;
 510                 int len;
 511                 unsigned long pos;
 512
 513                 ret = -EFAULT;
 514                 str = get_user_arg_ptr(argv, argc);
 515                 if (IS_ERR(str))
 516                         goto out;
 517
 518                 len = strnlen_user(str, MAX_ARG_STRLEN);
 519                 if (!len)
 520                         goto out;
 521
 522                 ret = -E2BIG;
 523                 if (!valid_arg_len(bprm, len))
 524                         goto out;
 525
 526                 /* We're going to work our way backwords. */
 527                 pos = bprm->p;
 528                 str += len;
 529                 bprm->p -= len;
 530
 531                 while (len > 0) {
 532                         int offset, bytes_to_copy;
 533
 534                         if (fatal_signal_pending(current)) {
 535                                 ret = -ERESTARTNOHAND;
 536                                 goto out;
 537                         }
 538                         cond_resched();
 539
 540                         offset = pos % PAGE_SIZE;
 541                         if (offset == 0)
 542                                 offset = PAGE_SIZE;
 543
 544                         bytes_to_copy = offset;
 545                         if (bytes_to_copy > len)
 546                                 bytes_to_copy = len;
 547
 548                         offset -= bytes_to_copy;
 549                         pos -= bytes_to_copy;
 550                         str -= bytes_to_copy;
 551                         len -= bytes_to_copy;
 552
 553                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 554                                 struct page *page;
 555
 556                                 page = get_arg_page(bprm, pos, 1);
 557                                 if (!page) {
 558                                         ret = -E2BIG;
 559                                         goto out;
 560                                 }
 561
 562                                 if (kmapped_page) {
 563                                         flush_kernel_dcache_page(kmapped_page);
 564                                         kunmap(kmapped_page);
 565                                         put_arg_page(kmapped_page);
 566                                 }
 567                                 kmapped_page = page;
 568                                 kaddr = kmap(kmapped_page);
 569                                 kpos = pos & PAGE_MASK;
 570                                 flush_arg_page(bprm, kpos, kmapped_page);
 571                         }
 572                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 573                                 ret = -EFAULT;
 574                                 goto out;
 575                         }
 576                 }
 577         }
 578         ret = 0;
 579 out:
 580         if (kmapped_page) {
 581                 flush_kernel_dcache_page(kmapped_page);
 582                 kunmap(kmapped_page);
 583                 put_arg_page(kmapped_page);
 584         }
 585         return ret;
 586 }
 587
 588 /*
 589  * Like copy_strings, but get argv and its values from kernel memory.
 590  */
 591 int copy_strings_kernel(int argc, const char *const *__argv,
 592                         struct linux_binprm *bprm)
 593 {
 594         int r;
 595         mm_segment_t oldfs = get_fs();
 596         struct user_arg_ptr argv = {
 597                 .ptr.native = (const char __user *const  __user *)__argv,
 598         };
 599
 600         set_fs(KERNEL_DS);
 601         r = copy_strings(argc, argv, bprm);
 602         set_fs(oldfs);
 603
 604         return r;
 605 }
 606 EXPORT_SYMBOL(copy_strings_kernel);
 607
 608 #ifdef CONFIG_MMU
 609
 610 /*
 611  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 612  * the binfmt code determines where the new stack should reside, we shift it to
 613  * its final location.  The process proceeds as follows:
 614  *
 615  * 1) Use shift to calculate the new vma endpoints.
 616  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 617  *    arguments passed to subsequent functions are consistent.
 618  * 3) Move vma's page tables to the new range.
 619  * 4) Free up any cleared pgd range.
 620  * 5) Shrink the vma to cover only the new range.
 621  */
 622 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 623 {
 624         struct mm_struct *mm = vma->vm_mm;
 625         unsigned long old_start = vma->vm_start;
 626         unsigned long old_end = vma->vm_end;
 627         unsigned long length = old_end - old_start;
 628         unsigned long new_start = old_start - shift;
 629         unsigned long new_end = old_end - shift;
 630         struct mmu_gather tlb;
 631
 632         BUG_ON(new_start > new_end);
 633
 634         /*
 635          * ensure there are no vmas between where we want to go
 636          * and where we are
 637          */
 638         if (vma != find_vma(mm, new_start))
 639                 return -EFAULT;
 640
 641         /*
 642          * cover the whole range: [new_start, old_end)
 643          */
 644         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 645                 return -ENOMEM;
 646
 647         /*
 648          * move the page tables downwards, on failure we rely on
 649          * process cleanup to remove whatever mess we made.
 650          */
 651         if (length != move_page_tables(vma, old_start,
 652                                        vma, new_start, length, false))
 653                 return -ENOMEM;
 654
 655         lru_add_drain();
 656         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 657         if (new_end > old_start) {
 658                 /*
 659                  * when the old and new regions overlap clear from new_end.
 660                  */
 661                 free_pgd_range(&tlb, new_end, old_end, new_end,
 662                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 663         } else {
 664                 /*
 665                  * otherwise, clean from old_start; this is done to not touch
 666                  * the address space in [new_end, old_start) some architectures
 667                  * have constraints on va-space that make this illegal (IA64) -
 668                  * for the others its just a little faster.
 669                  */
 670                 free_pgd_range(&tlb, old_start, old_end, new_end,
 671                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 672         }
 673         tlb_finish_mmu(&tlb, old_start, old_end);
 674
 675         /*
 676          * Shrink the vma to just the new range.  Always succeeds.
 677          */
 678         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 679
 680         return 0;
 681 }
 682
 683 /*
 684  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 685  * the stack is optionally relocated, and some extra space is added.
 686  */
 687 int setup_arg_pages(struct linux_binprm *bprm,
 688                     unsigned long stack_top,
 689                     int executable_stack)
 690 {
 691         unsigned long ret;
 692         unsigned long stack_shift;
 693         struct mm_struct *mm = current->mm;
 694         struct vm_area_struct *vma = bprm->vma;
 695         struct vm_area_struct *prev = NULL;
 696         unsigned long vm_flags;
 697         unsigned long stack_base;
 698         unsigned long stack_size;
 699         unsigned long stack_expand;
 700         unsigned long rlim_stack;
 701
 702 #ifdef CONFIG_STACK_GROWSUP
 703         /* Limit stack size */
 704         stack_base = bprm->rlim_stack.rlim_max;
 705         if (stack_base > STACK_SIZE_MAX)
 706                 stack_base = STACK_SIZE_MAX;
 707
 708         /* Add space for stack randomization. */
 709         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 710
 711         /* Make sure we didn't let the argument array grow too large. */
 712         if (vma->vm_end - vma->vm_start > stack_base)
 713                 return -ENOMEM;
 714
 715         stack_base = PAGE_ALIGN(stack_top - stack_base);
 716
 717         stack_shift = vma->vm_start - stack_base;
 718         mm->arg_start = bprm->p - stack_shift;
 719         bprm->p = vma->vm_end - stack_shift;
 720 #else
 721         stack_top = arch_align_stack(stack_top);
 722         stack_top = PAGE_ALIGN(stack_top);
 723
 724         if (unlikely(stack_top < mmap_min_addr) ||
 725             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 726                 return -ENOMEM;
 727
 728         stack_shift = vma->vm_end - stack_top;
 729
 730         bprm->p -= stack_shift;
 731         mm->arg_start = bprm->p;
 732 #endif
 733
 734         if (bprm->loader)
 735                 bprm->loader -= stack_shift;
 736         bprm->exec -= stack_shift;
 737
 738         if (down_write_killable(&mm->mmap_sem))
 739                 return -EINTR;
 740
 741         vm_flags = VM_STACK_FLAGS;
 742
 743         /*
 744          * Adjust stack execute permissions; explicitly enable for
 745          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 746          * (arch default) otherwise.
 747          */
 748         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 749                 vm_flags |= VM_EXEC;
 750         else if (executable_stack == EXSTACK_DISABLE_X)
 751                 vm_flags &= ~VM_EXEC;
 752         vm_flags |= mm->def_flags;
 753         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 754
 755         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 756                         vm_flags);
 757         if (ret)
 758                 goto out_unlock;
 759         BUG_ON(prev != vma);
 760
 761         /* Move stack pages down in memory. */
 762         if (stack_shift) {
 763                 ret = shift_arg_pages(vma, stack_shift);
 764                 if (ret)
 765                         goto out_unlock;
 766         }
 767
 768         /* mprotect_fixup is overkill to remove the temporary stack flags */
 769         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 770
 771         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 772         stack_size = vma->vm_end - vma->vm_start;
 773         /*
 774          * Align this down to a page boundary as expand_stack
 775          * will align it up.
 776          */
 777         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 778 #ifdef CONFIG_STACK_GROWSUP
 779         if (stack_size + stack_expand > rlim_stack)
 780                 stack_base = vma->vm_start + rlim_stack;
 781         else
 782                 stack_base = vma->vm_end + stack_expand;
 783 #else
 784         if (stack_size + stack_expand > rlim_stack)
 785                 stack_base = vma->vm_end - rlim_stack;
 786         else
 787                 stack_base = vma->vm_start - stack_expand;
 788 #endif
 789         current->mm->start_stack = bprm->p;
 790         ret = expand_stack(vma, stack_base);
 791         if (ret)
 792                 ret = -EFAULT;
 793
 794 out_unlock:
 795         up_write(&mm->mmap_sem);
 796         return ret;
 797 }
 798 EXPORT_SYMBOL(setup_arg_pages);
 799
 800 #else
 801
 802 /*
 803  * Transfer the program arguments and environment from the holding pages
 804  * onto the stack. The provided stack pointer is adjusted accordingly.
 805  */
 806 int transfer_args_to_stack(struct linux_binprm *bprm,
 807                            unsigned long *sp_location)
 808 {
 809         unsigned long index, stop, sp;
 810         int ret = 0;
 811
 812         stop = bprm->p >> PAGE_SHIFT;
 813         sp = *sp_location;
 814
 815         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 816                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 817                 char *src = kmap(bprm->page[index]) + offset;
 818                 sp -= PAGE_SIZE - offset;
 819                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 820                         ret = -EFAULT;
 821                 kunmap(bprm->page[index]);
 822                 if (ret)
 823                         goto out;
 824         }
 825
 826         *sp_location = sp;
 827
 828 out:
 829         return ret;
 830 }
 831 EXPORT_SYMBOL(transfer_args_to_stack);
 832
 833 #endif /* CONFIG_MMU */
 834
 835 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 836 {
 837         struct file *file;
 838         int err;
 839         struct open_flags open_exec_flags = {
 840                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 841                 .acc_mode = MAY_EXEC,
 842                 .intent = LOOKUP_OPEN,
 843                 .lookup_flags = LOOKUP_FOLLOW,
 844         };
 845
 846         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 847                 return ERR_PTR(-EINVAL);
 848         if (flags & AT_SYMLINK_NOFOLLOW)
 849                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 850         if (flags & AT_EMPTY_PATH)
 851                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 852
 853         file = do_filp_open(fd, name, &open_exec_flags);
 854         if (IS_ERR(file))
 855                 goto out;
 856
 857         err = -EACCES;
 858         if (!S_ISREG(file_inode(file)->i_mode))
 859                 goto exit;
 860
 861         if (path_noexec(&file->f_path))
 862                 goto exit;
 863
 864         err = deny_write_access(file);
 865         if (err)
 866                 goto exit;
 867
 868         if (name->name[0] != '\0')
 869                 fsnotify_open(file);
 870
 871 out:
 872         return file;
 873
 874 exit:
 875         fput(file);
 876         return ERR_PTR(err);
 877 }
 878
 879 struct file *open_exec(const char *name)
 880 {
 881         struct filename *filename = getname_kernel(name);
 882         struct file *f = ERR_CAST(filename);
 883
 884         if (!IS_ERR(filename)) {
 885                 f = do_open_execat(AT_FDCWD, filename, 0);
 886                 putname(filename);
 887         }
 888         return f;
 889 }
 890 EXPORT_SYMBOL(open_exec);
 891
 892 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 893                      loff_t max_size, enum kernel_read_file_id id)
 894 {
 895         loff_t i_size, pos;
 896         ssize_t bytes = 0;
 897         int ret;
 898
 899         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 900                 return -EINVAL;
 901
 902         ret = deny_write_access(file);
 903         if (ret)
 904                 return ret;
 905
 906         ret = security_kernel_read_file(file, id);
 907         if (ret)
 908                 goto out;
 909
 910         i_size = i_size_read(file_inode(file));
 911         if (i_size <= 0) {
 912                 ret = -EINVAL;
 913                 goto out;
 914         }
 915         if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 916                 ret = -EFBIG;
 917                 goto out;
 918         }
 919
 920         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 921                 *buf = vmalloc(i_size);
 922         if (!*buf) {
 923                 ret = -ENOMEM;
 924                 goto out;
 925         }
 926
 927         pos = 0;
 928         while (pos < i_size) {
 929                 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 930                 if (bytes < 0) {
 931                         ret = bytes;
 932                         goto out;
 933                 }
 934
 935                 if (bytes == 0)
 936                         break;
 937         }
 938
 939         if (pos != i_size) {
 940                 ret = -EIO;
 941                 goto out_free;
 942         }
 943
 944         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 945         if (!ret)
 946                 *size = pos;
 947
 948 out_free:
 949         if (ret < 0) {
 950                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 951                         vfree(*buf);
 952                         *buf = NULL;
 953                 }
 954         }
 955
 956 out:
 957         allow_write_access(file);
 958         return ret;
 959 }
 960 EXPORT_SYMBOL_GPL(kernel_read_file);
 961
 962 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
 963                                loff_t max_size, enum kernel_read_file_id id)
 964 {
 965         struct file *file;
 966         int ret;
 967
 968         if (!path || !*path)
 969                 return -EINVAL;
 970
 971         file = filp_open(path, O_RDONLY, 0);
 972         if (IS_ERR(file))
 973                 return PTR_ERR(file);
 974
 975         ret = kernel_read_file(file, buf, size, max_size, id);
 976         fput(file);
 977         return ret;
 978 }
 979 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 980
 981 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 982                              enum kernel_read_file_id id)
 983 {
 984         struct fd f = fdget(fd);
 985         int ret = -EBADF;
 986
 987         if (!f.file)
 988                 goto out;
 989
 990         ret = kernel_read_file(f.file, buf, size, max_size, id);
 991 out:
 992         fdput(f);
 993         return ret;
 994 }
 995 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 996
 997 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 998 {
 999         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1000         if (res > 0)
1001                 flush_icache_range(addr, addr + len);
1002         return res;
1003 }
1004 EXPORT_SYMBOL(read_code);
1005
1006 static int exec_mmap(struct mm_struct *mm)
1007 {
1008         struct task_struct *tsk;
1009         struct mm_struct *old_mm, *active_mm;
1010
1011         /* Notify parent that we're no longer interested in the old VM */
1012         tsk = current;
1013         old_mm = current->mm;
1014         mm_release(tsk, old_mm);
1015
1016         if (old_mm) {
1017                 sync_mm_rss(old_mm);
1018                 /*
1019                  * Make sure that if there is a core dump in progress
1020                  * for the old mm, we get out and die instead of going
1021                  * through with the exec.  We must hold mmap_sem around
1022                  * checking core_state and changing tsk->mm.
1023                  */
1024                 down_read(&old_mm->mmap_sem);
1025                 if (unlikely(old_mm->core_state)) {
1026                         up_read(&old_mm->mmap_sem);
1027                         return -EINTR;
1028                 }
1029         }
1030         task_lock(tsk);
1031         active_mm = tsk->active_mm;
1032         tsk->mm = mm;
1033         tsk->active_mm = mm;
1034         activate_mm(active_mm, mm);
1035         tsk->mm->vmacache_seqnum = 0;
1036         vmacache_flush(tsk);
1037         task_unlock(tsk);
1038         if (old_mm) {
1039                 up_read(&old_mm->mmap_sem);
1040                 BUG_ON(active_mm != old_mm);
1041                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1042                 mm_update_next_owner(old_mm);
1043                 mmput(old_mm);
1044                 return 0;
1045         }
1046         mmdrop(active_mm);
1047         return 0;
1048 }
1049
1050 /*
1051  * This function makes sure the current process has its own signal table,
1052  * so that flush_signal_handlers can later reset the handlers without
1053  * disturbing other processes.  (Other processes might share the signal
1054  * table via the CLONE_SIGHAND option to clone().)
1055  */
1056 static int de_thread(struct task_struct *tsk)
1057 {
1058         struct signal_struct *sig = tsk->signal;
1059         struct sighand_struct *oldsighand = tsk->sighand;
1060         spinlock_t *lock = &oldsighand->siglock;
1061
1062         if (thread_group_empty(tsk))
1063                 goto no_thread_group;
1064
1065         /*
1066          * Kill all other threads in the thread group.
1067          */
1068         spin_lock_irq(lock);
1069         if (signal_group_exit(sig)) {
1070                 /*
1071                  * Another group action in progress, just
1072                  * return so that the signal is processed.
1073                  */
1074                 spin_unlock_irq(lock);
1075                 return -EAGAIN;
1076         }
1077
1078         sig->group_exit_task = tsk;
1079         sig->notify_count = zap_other_threads(tsk);
1080         if (!thread_group_leader(tsk))
1081                 sig->notify_count--;
1082
1083         while (sig->notify_count) {
1084                 __set_current_state(TASK_KILLABLE);
1085                 spin_unlock_irq(lock);
1086                 schedule();
1087                 if (unlikely(__fatal_signal_pending(tsk)))
1088                         goto killed;
1089                 spin_lock_irq(lock);
1090         }
1091         spin_unlock_irq(lock);
1092
1093         /*
1094          * At this point all other threads have exited, all we have to
1095          * do is to wait for the thread group leader to become inactive,
1096          * and to assume its PID:
1097          */
1098         if (!thread_group_leader(tsk)) {
1099                 struct task_struct *leader = tsk->group_leader;
1100
1101                 for (;;) {
1102                         cgroup_threadgroup_change_begin(tsk);
1103                         write_lock_irq(&tasklist_lock);
1104                         /*
1105                          * Do this under tasklist_lock to ensure that
1106                          * exit_notify() can't miss ->group_exit_task
1107                          */
1108                         sig->notify_count = -1;
1109                         if (likely(leader->exit_state))
1110                                 break;
1111                         __set_current_state(TASK_KILLABLE);
1112                         write_unlock_irq(&tasklist_lock);
1113                         cgroup_threadgroup_change_end(tsk);
1114                         schedule();
1115                         if (unlikely(__fatal_signal_pending(tsk)))
1116                                 goto killed;
1117                 }
1118
1119                 /*
1120                  * The only record we have of the real-time age of a
1121                  * process, regardless of execs it's done, is start_time.
1122                  * All the past CPU time is accumulated in signal_struct
1123                  * from sister threads now dead.  But in this non-leader
1124                  * exec, nothing survives from the original leader thread,
1125                  * whose birth marks the true age of this process now.
1126                  * When we take on its identity by switching to its PID, we
1127                  * also take its birthdate (always earlier than our own).
1128                  */
1129                 tsk->start_time = leader->start_time;
1130                 tsk->real_start_time = leader->real_start_time;
1131
1132                 BUG_ON(!same_thread_group(leader, tsk));
1133                 BUG_ON(has_group_leader_pid(tsk));
1134                 /*
1135                  * An exec() starts a new thread group with the
1136                  * TGID of the previous thread group. Rehash the
1137                  * two threads with a switched PID, and release
1138                  * the former thread group leader:
1139                  */
1140
1141                 /* Become a process group leader with the old leader's pid.
1142                  * The old leader becomes a thread of the this thread group.
1143                  * Note: The old leader also uses this pid until release_task
1144                  *       is called.  Odd but simple and correct.
1145                  */
1146                 tsk->pid = leader->pid;
1147                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1148                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1149                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1150                 transfer_pid(leader, tsk, PIDTYPE_SID);
1151
1152                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1153                 list_replace_init(&leader->sibling, &tsk->sibling);
1154
1155                 tsk->group_leader = tsk;
1156                 leader->group_leader = tsk;
1157
1158                 tsk->exit_signal = SIGCHLD;
1159                 leader->exit_signal = -1;
1160
1161                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1162                 leader->exit_state = EXIT_DEAD;
1163
1164                 /*
1165                  * We are going to release_task()->ptrace_unlink() silently,
1166                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1167                  * the tracer wont't block again waiting for this thread.
1168                  */
1169                 if (unlikely(leader->ptrace))
1170                         __wake_up_parent(leader, leader->parent);
1171                 write_unlock_irq(&tasklist_lock);
1172                 cgroup_threadgroup_change_end(tsk);
1173
1174                 release_task(leader);
1175         }
1176
1177         sig->group_exit_task = NULL;
1178         sig->notify_count = 0;
1179
1180 no_thread_group:
1181         /* we have changed execution domain */
1182         tsk->exit_signal = SIGCHLD;
1183
1184 #ifdef CONFIG_POSIX_TIMERS
1185         exit_itimers(sig);
1186         flush_itimer_signals();
1187 #endif
1188
1189         if (atomic_read(&oldsighand->count) != 1) {
1190                 struct sighand_struct *newsighand;
1191                 /*
1192                  * This ->sighand is shared with the CLONE_SIGHAND
1193                  * but not CLONE_THREAD task, switch to the new one.
1194                  */
1195                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1196                 if (!newsighand)
1197                         return -ENOMEM;
1198
1199                 atomic_set(&newsighand->count, 1);
1200                 memcpy(newsighand->action, oldsighand->action,
1201                        sizeof(newsighand->action));
1202
1203                 write_lock_irq(&tasklist_lock);
1204                 spin_lock(&oldsighand->siglock);
1205                 rcu_assign_pointer(tsk->sighand, newsighand);
1206                 spin_unlock(&oldsighand->siglock);
1207                 write_unlock_irq(&tasklist_lock);
1208
1209                 __cleanup_sighand(oldsighand);
1210         }
1211
1212         BUG_ON(!thread_group_leader(tsk));
1213         return 0;
1214
1215 killed:
1216         /* protects against exit_notify() and __exit_signal() */
1217         read_lock(&tasklist_lock);
1218         sig->group_exit_task = NULL;
1219         sig->notify_count = 0;
1220         read_unlock(&tasklist_lock);
1221         return -EAGAIN;
1222 }
1223
1224 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1225 {
1226         task_lock(tsk);
1227         strncpy(buf, tsk->comm, buf_size);
1228         task_unlock(tsk);
1229         return buf;
1230 }
1231 EXPORT_SYMBOL_GPL(__get_task_comm);
1232
1233 /*
1234  * These functions flushes out all traces of the currently running executable
1235  * so that a new one can be started
1236  */
1237
1238 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1239 {
1240         task_lock(tsk);
1241         trace_task_rename(tsk, buf);
1242         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1243         task_unlock(tsk);
1244         perf_event_comm(tsk, exec);
1245 }
1246
1247 /*
1248  * Calling this is the point of no return. None of the failures will be
1249  * seen by userspace since either the process is already taking a fatal
1250  * signal (via de_thread() or coredump), or will have SEGV raised
1251  * (after exec_mmap()) by search_binary_handlers (see below).
1252  */
1253 int flush_old_exec(struct linux_binprm * bprm)
1254 {
1255         int retval;
1256
1257         /*
1258          * Make sure we have a private signal table and that
1259          * we are unassociated from the previous thread group.
1260          */
1261         retval = de_thread(current);
1262         if (retval)
1263                 goto out;
1264
1265         /*
1266          * Must be called _before_ exec_mmap() as bprm->mm is
1267          * not visibile until then. This also enables the update
1268          * to be lockless.
1269          */
1270         set_mm_exe_file(bprm->mm, bprm->file);
1271
1272         /*
1273          * Release all of the old mmap stuff
1274          */
1275         acct_arg_size(bprm, 0);
1276         retval = exec_mmap(bprm->mm);
1277         if (retval)
1278                 goto out;
1279
1280         /*
1281          * After clearing bprm->mm (to mark that current is using the
1282          * prepared mm now), we have nothing left of the original
1283          * process. If anything from here on returns an error, the check
1284          * in search_binary_handler() will SEGV current.
1285          */
1286         bprm->mm = NULL;
1287
1288         set_fs(USER_DS);
1289         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1290                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1291         flush_thread();
1292         current->personality &= ~bprm->per_clear;
1293
1294         /*
1295          * We have to apply CLOEXEC before we change whether the process is
1296          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1297          * trying to access the should-be-closed file descriptors of a process
1298          * undergoing exec(2).
1299          */
1300         do_close_on_exec(current->files);
1301         return 0;
1302
1303 out:
1304         return retval;
1305 }
1306 EXPORT_SYMBOL(flush_old_exec);
1307
1308 void would_dump(struct linux_binprm *bprm, struct file *file)
1309 {
1310         struct inode *inode = file_inode(file);
1311         if (inode_permission(inode, MAY_READ) < 0) {
1312                 struct user_namespace *old, *user_ns;
1313                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1314
1315                 /* Ensure mm->user_ns contains the executable */
1316                 user_ns = old = bprm->mm->user_ns;
1317                 while ((user_ns != &init_user_ns) &&
1318                        !privileged_wrt_inode_uidgid(user_ns, inode))
1319                         user_ns = user_ns->parent;
1320
1321                 if (old != user_ns) {
1322                         bprm->mm->user_ns = get_user_ns(user_ns);
1323                         put_user_ns(old);
1324                 }
1325         }
1326 }
1327 EXPORT_SYMBOL(would_dump);
1328
1329 void setup_new_exec(struct linux_binprm * bprm)
1330 {
1331         /*
1332          * Once here, prepare_binrpm() will not be called any more, so
1333          * the final state of setuid/setgid/fscaps can be merged into the
1334          * secureexec flag.
1335          */
1336         bprm->secureexec |= bprm->cap_elevated;
1337
1338         if (bprm->secureexec) {
1339                 /* Make sure parent cannot signal privileged process. */
1340                 current->pdeath_signal = 0;
1341
1342                 /*
1343                  * For secureexec, reset the stack limit to sane default to
1344                  * avoid bad behavior from the prior rlimits. This has to
1345                  * happen before arch_pick_mmap_layout(), which examines
1346                  * RLIMIT_STACK, but after the point of no return to avoid
1347                  * needing to clean up the change on failure.
1348                  */
1349                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1350                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1351         }
1352
1353         arch_pick_mmap_layout(current->mm, &bprm->rlim_stack);
1354
1355         current->sas_ss_sp = current->sas_ss_size = 0;
1356
1357         /*
1358          * Figure out dumpability. Note that this checking only of current
1359          * is wrong, but userspace depends on it. This should be testing
1360          * bprm->secureexec instead.
1361          */
1362         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1363             !(uid_eq(current_euid(), current_uid()) &&
1364               gid_eq(current_egid(), current_gid())))
1365                 set_dumpable(current->mm, suid_dumpable);
1366         else
1367                 set_dumpable(current->mm, SUID_DUMP_USER);
1368
1369         arch_setup_new_exec();
1370         perf_event_exec();
1371         __set_task_comm(current, kbasename(bprm->filename), true);
1372
1373         /* Set the new mm task size. We have to do that late because it may
1374          * depend on TIF_32BIT which is only updated in flush_thread() on
1375          * some architectures like powerpc
1376          */
1377         current->mm->task_size = TASK_SIZE;
1378
1379         /* An exec changes our domain. We are no longer part of the thread
1380            group */
1381         current->self_exec_id++;
1382         flush_signal_handlers(current, 0);
1383 }
1384 EXPORT_SYMBOL(setup_new_exec);
1385
1386 /* Runs immediately before start_thread() takes over. */
1387 void finalize_exec(struct linux_binprm *bprm)
1388 {
1389         /* Store any stack rlimit changes before starting thread. */
1390         task_lock(current->group_leader);
1391         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1392         task_unlock(current->group_leader);
1393 }
1394 EXPORT_SYMBOL(finalize_exec);
1395
1396 /*
1397  * Prepare credentials and lock ->cred_guard_mutex.
1398  * install_exec_creds() commits the new creds and drops the lock.
1399  * Or, if exec fails before, free_bprm() should release ->cred and
1400  * and unlock.
1401  */
1402 int prepare_bprm_creds(struct linux_binprm *bprm)
1403 {
1404         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1405                 return -ERESTARTNOINTR;
1406
1407         bprm->cred = prepare_exec_creds();
1408         if (likely(bprm->cred))
1409                 return 0;
1410
1411         mutex_unlock(&current->signal->cred_guard_mutex);
1412         return -ENOMEM;
1413 }
1414
1415 static void free_bprm(struct linux_binprm *bprm)
1416 {
1417         free_arg_pages(bprm);
1418         if (bprm->cred) {
1419                 mutex_unlock(&current->signal->cred_guard_mutex);
1420                 abort_creds(bprm->cred);
1421         }
1422         if (bprm->file) {
1423                 allow_write_access(bprm->file);
1424                 fput(bprm->file);
1425         }
1426         /* If a binfmt changed the interp, free it. */
1427         if (bprm->interp != bprm->filename)
1428                 kfree(bprm->interp);
1429         kfree(bprm);
1430 }
1431
1432 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1433 {
1434         /* If a binfmt changed the interp, free it first. */
1435         if (bprm->interp != bprm->filename)
1436                 kfree(bprm->interp);
1437         bprm->interp = kstrdup(interp, GFP_KERNEL);
1438         if (!bprm->interp)
1439                 return -ENOMEM;
1440         return 0;
1441 }
1442 EXPORT_SYMBOL(bprm_change_interp);
1443
1444 /*
1445  * install the new credentials for this executable
1446  */
1447 void install_exec_creds(struct linux_binprm *bprm)
1448 {
1449         security_bprm_committing_creds(bprm);
1450
1451         commit_creds(bprm->cred);
1452         bprm->cred = NULL;
1453
1454         /*
1455          * Disable monitoring for regular users
1456          * when executing setuid binaries. Must
1457          * wait until new credentials are committed
1458          * by commit_creds() above
1459          */
1460         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1461                 perf_event_exit_task(current);
1462         /*
1463          * cred_guard_mutex must be held at least to this point to prevent
1464          * ptrace_attach() from altering our determination of the task's
1465          * credentials; any time after this it may be unlocked.
1466          */
1467         security_bprm_committed_creds(bprm);
1468         mutex_unlock(&current->signal->cred_guard_mutex);
1469 }
1470 EXPORT_SYMBOL(install_exec_creds);
1471
1472 /*
1473  * determine how safe it is to execute the proposed program
1474  * - the caller must hold ->cred_guard_mutex to protect against
1475  *   PTRACE_ATTACH or seccomp thread-sync
1476  */
1477 static void check_unsafe_exec(struct linux_binprm *bprm)
1478 {
1479         struct task_struct *p = current, *t;
1480         unsigned n_fs;
1481
1482         if (p->ptrace)
1483                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1484
1485         /*
1486          * This isn't strictly necessary, but it makes it harder for LSMs to
1487          * mess up.
1488          */
1489         if (task_no_new_privs(current))
1490                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1491
1492         t = p;
1493         n_fs = 1;
1494         spin_lock(&p->fs->lock);
1495         rcu_read_lock();
1496         while_each_thread(p, t) {
1497                 if (t->fs == p->fs)
1498                         n_fs++;
1499         }
1500         rcu_read_unlock();
1501
1502         if (p->fs->users > n_fs)
1503                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1504         else
1505                 p->fs->in_exec = 1;
1506         spin_unlock(&p->fs->lock);
1507 }
1508
1509 static void bprm_fill_uid(struct linux_binprm *bprm)
1510 {
1511         struct inode *inode;
1512         unsigned int mode;
1513         kuid_t uid;
1514         kgid_t gid;
1515
1516         /*
1517          * Since this can be called multiple times (via prepare_binprm),
1518          * we must clear any previous work done when setting set[ug]id
1519          * bits from any earlier bprm->file uses (for example when run
1520          * first for a setuid script then again for its interpreter).
1521          */
1522         bprm->cred->euid = current_euid();
1523         bprm->cred->egid = current_egid();
1524
1525         if (!mnt_may_suid(bprm->file->f_path.mnt))
1526                 return;
1527
1528         if (task_no_new_privs(current))
1529                 return;
1530
1531         inode = bprm->file->f_path.dentry->d_inode;
1532         mode = READ_ONCE(inode->i_mode);
1533         if (!(mode & (S_ISUID|S_ISGID)))
1534                 return;
1535
1536         /* Be careful if suid/sgid is set */
1537         inode_lock(inode);
1538
1539         /* reload atomically mode/uid/gid now that lock held */
1540         mode = inode->i_mode;
1541         uid = inode->i_uid;
1542         gid = inode->i_gid;
1543         inode_unlock(inode);
1544
1545         /* We ignore suid/sgid if there are no mappings for them in the ns */
1546         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1547                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1548                 return;
1549
1550         if (mode & S_ISUID) {
1551                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1552                 bprm->cred->euid = uid;
1553         }
1554
1555         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1556                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1557                 bprm->cred->egid = gid;
1558         }
1559 }
1560
1561 /*
1562  * Fill the binprm structure from the inode.
1563  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1564  *
1565  * This may be called multiple times for binary chains (scripts for example).
1566  */
1567 int prepare_binprm(struct linux_binprm *bprm)
1568 {
1569         int retval;
1570         loff_t pos = 0;
1571
1572         bprm_fill_uid(bprm);
1573
1574         /* fill in binprm security blob */
1575         retval = security_bprm_set_creds(bprm);
1576         if (retval)
1577                 return retval;
1578         bprm->called_set_creds = 1;
1579
1580         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1581         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1582 }
1583
1584 EXPORT_SYMBOL(prepare_binprm);
1585
1586 /*
1587  * Arguments are '\0' separated strings found at the location bprm->p
1588  * points to; chop off the first by relocating brpm->p to right after
1589  * the first '\0' encountered.
1590  */
1591 int remove_arg_zero(struct linux_binprm *bprm)
1592 {
1593         int ret = 0;
1594         unsigned long offset;
1595         char *kaddr;
1596         struct page *page;
1597
1598         if (!bprm->argc)
1599                 return 0;
1600
1601         do {
1602                 offset = bprm->p & ~PAGE_MASK;
1603                 page = get_arg_page(bprm, bprm->p, 0);
1604                 if (!page) {
1605                         ret = -EFAULT;
1606                         goto out;
1607                 }
1608                 kaddr = kmap_atomic(page);
1609
1610                 for (; offset < PAGE_SIZE && kaddr[offset];
1611                                 offset++, bprm->p++)
1612                         ;
1613
1614                 kunmap_atomic(kaddr);
1615                 put_arg_page(page);
1616         } while (offset == PAGE_SIZE);
1617
1618         bprm->p++;
1619         bprm->argc--;
1620         ret = 0;
1621
1622 out:
1623         return ret;
1624 }
1625 EXPORT_SYMBOL(remove_arg_zero);
1626
1627 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1628 /*
1629  * cycle the list of binary formats handler, until one recognizes the image
1630  */
1631 int search_binary_handler(struct linux_binprm *bprm)
1632 {
1633         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1634         struct linux_binfmt *fmt;
1635         int retval;
1636
1637         /* This allows 4 levels of binfmt rewrites before failing hard. */
1638         if (bprm->recursion_depth > 5)
1639                 return -ELOOP;
1640
1641         retval = security_bprm_check(bprm);
1642         if (retval)
1643                 return retval;
1644
1645         retval = -ENOENT;
1646  retry:
1647         read_lock(&binfmt_lock);
1648         list_for_each_entry(fmt, &formats, lh) {
1649                 if (!try_module_get(fmt->module))
1650                         continue;
1651                 read_unlock(&binfmt_lock);
1652                 bprm->recursion_depth++;
1653                 retval = fmt->load_binary(bprm);
1654                 read_lock(&binfmt_lock);
1655                 put_binfmt(fmt);
1656                 bprm->recursion_depth--;
1657                 if (retval < 0 && !bprm->mm) {
1658                         /* we got to flush_old_exec() and failed after it */
1659                         read_unlock(&binfmt_lock);
1660                         force_sigsegv(SIGSEGV, current);
1661                         return retval;
1662                 }
1663                 if (retval != -ENOEXEC || !bprm->file) {
1664                         read_unlock(&binfmt_lock);
1665                         return retval;
1666                 }
1667         }
1668         read_unlock(&binfmt_lock);
1669
1670         if (need_retry) {
1671                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1672                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1673                         return retval;
1674                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1675                         return retval;
1676                 need_retry = false;
1677                 goto retry;
1678         }
1679
1680         return retval;
1681 }
1682 EXPORT_SYMBOL(search_binary_handler);
1683
1684 static int exec_binprm(struct linux_binprm *bprm)
1685 {
1686         pid_t old_pid, old_vpid;
1687         int ret;
1688
1689         /* Need to fetch pid before load_binary changes it */
1690         old_pid = current->pid;
1691         rcu_read_lock();
1692         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1693         rcu_read_unlock();
1694
1695         ret = search_binary_handler(bprm);
1696         if (ret >= 0) {
1697                 audit_bprm(bprm);
1698                 trace_sched_process_exec(current, old_pid, bprm);
1699                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1700                 proc_exec_connector(current);
1701         }
1702
1703         return ret;
1704 }
1705
1706 /*
1707  * sys_execve() executes a new program.
1708  */
1709 static int __do_execve_file(int fd, struct filename *filename,
1710                             struct user_arg_ptr argv,
1711                             struct user_arg_ptr envp,
1712                             int flags, struct file *file)
1713 {
1714         char *pathbuf = NULL;
1715         struct linux_binprm *bprm;
1716         struct files_struct *displaced;
1717         int retval;
1718
1719         if (IS_ERR(filename))
1720                 return PTR_ERR(filename);
1721
1722         /*
1723          * We move the actual failure in case of RLIMIT_NPROC excess from
1724          * set*uid() to execve() because too many poorly written programs
1725          * don't check setuid() return code.  Here we additionally recheck
1726          * whether NPROC limit is still exceeded.
1727          */
1728         if ((current->flags & PF_NPROC_EXCEEDED) &&
1729             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1730                 retval = -EAGAIN;
1731                 goto out_ret;
1732         }
1733
1734         /* We're below the limit (still or again), so we don't want to make
1735          * further execve() calls fail. */
1736         current->flags &= ~PF_NPROC_EXCEEDED;
1737
1738         retval = unshare_files(&displaced);
1739         if (retval)
1740                 goto out_ret;
1741
1742         retval = -ENOMEM;
1743         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1744         if (!bprm)
1745                 goto out_files;
1746
1747         retval = prepare_bprm_creds(bprm);
1748         if (retval)
1749                 goto out_free;
1750
1751         check_unsafe_exec(bprm);
1752         current->in_execve = 1;
1753
1754         if (!file)
1755                 file = do_open_execat(fd, filename, flags);
1756         retval = PTR_ERR(file);
1757         if (IS_ERR(file))
1758                 goto out_unmark;
1759
1760         sched_exec();
1761
1762         bprm->file = file;
1763         if (!filename) {
1764                 bprm->filename = "none";
1765         } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1766                 bprm->filename = filename->name;
1767         } else {
1768                 if (filename->name[0] == '\0')
1769                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1770                 else
1771                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1772                                             fd, filename->name);
1773                 if (!pathbuf) {
1774                         retval = -ENOMEM;
1775                         goto out_unmark;
1776                 }
1777                 /*
1778                  * Record that a name derived from an O_CLOEXEC fd will be
1779                  * inaccessible after exec. Relies on having exclusive access to
1780                  * current->files (due to unshare_files above).
1781                  */
1782                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1783                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1784                 bprm->filename = pathbuf;
1785         }
1786         bprm->interp = bprm->filename;
1787
1788         retval = bprm_mm_init(bprm);
1789         if (retval)
1790                 goto out_unmark;
1791
1792         bprm->argc = count(argv, MAX_ARG_STRINGS);
1793         if ((retval = bprm->argc) < 0)
1794                 goto out;
1795
1796         bprm->envc = count(envp, MAX_ARG_STRINGS);
1797         if ((retval = bprm->envc) < 0)
1798                 goto out;
1799
1800         retval = prepare_binprm(bprm);
1801         if (retval < 0)
1802                 goto out;
1803
1804         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1805         if (retval < 0)
1806                 goto out;
1807
1808         bprm->exec = bprm->p;
1809         retval = copy_strings(bprm->envc, envp, bprm);
1810         if (retval < 0)
1811                 goto out;
1812
1813         retval = copy_strings(bprm->argc, argv, bprm);
1814         if (retval < 0)
1815                 goto out;
1816
1817         would_dump(bprm, bprm->file);
1818
1819         retval = exec_binprm(bprm);
1820         if (retval < 0)
1821                 goto out;
1822
1823         /* execve succeeded */
1824         current->fs->in_exec = 0;
1825         current->in_execve = 0;
1826         membarrier_execve(current);
1827         rseq_execve(current);
1828         acct_update_integrals(current);
1829         task_numa_free(current);
1830         free_bprm(bprm);
1831         kfree(pathbuf);
1832         if (filename)
1833                 putname(filename);
1834         if (displaced)
1835                 put_files_struct(displaced);
1836         return retval;
1837
1838 out:
1839         if (bprm->mm) {
1840                 acct_arg_size(bprm, 0);
1841                 mmput(bprm->mm);
1842         }
1843
1844 out_unmark:
1845         current->fs->in_exec = 0;
1846         current->in_execve = 0;
1847
1848 out_free:
1849         free_bprm(bprm);
1850         kfree(pathbuf);
1851
1852 out_files:
1853         if (displaced)
1854                 reset_files_struct(displaced);
1855 out_ret:
1856         if (filename)
1857                 putname(filename);
1858         return retval;
1859 }
1860
1861 static int do_execveat_common(int fd, struct filename *filename,
1862                               struct user_arg_ptr argv,
1863                               struct user_arg_ptr envp,
1864                               int flags)
1865 {
1866         return __do_execve_file(fd, filename, argv, envp, flags, NULL);
1867 }
1868
1869 int do_execve_file(struct file *file, void *__argv, void *__envp)
1870 {
1871         struct user_arg_ptr argv = { .ptr.native = __argv };
1872         struct user_arg_ptr envp = { .ptr.native = __envp };
1873
1874         return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file);
1875 }
1876
1877 int do_execve(struct filename *filename,
1878         const char __user *const __user *__argv,
1879         const char __user *const __user *__envp)
1880 {
1881         struct user_arg_ptr argv = { .ptr.native = __argv };
1882         struct user_arg_ptr envp = { .ptr.native = __envp };
1883         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1884 }
1885
1886 int do_execveat(int fd, struct filename *filename,
1887                 const char __user *const __user *__argv,
1888                 const char __user *const __user *__envp,
1889                 int flags)
1890 {
1891         struct user_arg_ptr argv = { .ptr.native = __argv };
1892         struct user_arg_ptr envp = { .ptr.native = __envp };
1893
1894         return do_execveat_common(fd, filename, argv, envp, flags);
1895 }
1896
1897 #ifdef CONFIG_COMPAT
1898 static int compat_do_execve(struct filename *filename,
1899         const compat_uptr_t __user *__argv,
1900         const compat_uptr_t __user *__envp)
1901 {
1902         struct user_arg_ptr argv = {
1903                 .is_compat = true,
1904                 .ptr.compat = __argv,
1905         };
1906         struct user_arg_ptr envp = {
1907                 .is_compat = true,
1908                 .ptr.compat = __envp,
1909         };
1910         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1911 }
1912
1913 static int compat_do_execveat(int fd, struct filename *filename,
1914                               const compat_uptr_t __user *__argv,
1915                               const compat_uptr_t __user *__envp,
1916                               int flags)
1917 {
1918         struct user_arg_ptr argv = {
1919                 .is_compat = true,
1920                 .ptr.compat = __argv,
1921         };
1922         struct user_arg_ptr envp = {
1923                 .is_compat = true,
1924                 .ptr.compat = __envp,
1925         };
1926         return do_execveat_common(fd, filename, argv, envp, flags);
1927 }
1928 #endif
1929
1930 void set_binfmt(struct linux_binfmt *new)
1931 {
1932         struct mm_struct *mm = current->mm;
1933
1934         if (mm->binfmt)
1935                 module_put(mm->binfmt->module);
1936
1937         mm->binfmt = new;
1938         if (new)
1939                 __module_get(new->module);
1940 }
1941 EXPORT_SYMBOL(set_binfmt);
1942
1943 /*
1944  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1945  */
1946 void set_dumpable(struct mm_struct *mm, int value)
1947 {
1948         unsigned long old, new;
1949
1950         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1951                 return;
1952
1953         do {
1954                 old = READ_ONCE(mm->flags);
1955                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1956         } while (cmpxchg(&mm->flags, old, new) != old);
1957 }
1958
1959 SYSCALL_DEFINE3(execve,
1960                 const char __user *, filename,
1961                 const char __user *const __user *, argv,
1962                 const char __user *const __user *, envp)
1963 {
1964         return do_execve(getname(filename), argv, envp);
1965 }
1966
1967 SYSCALL_DEFINE5(execveat,
1968                 int, fd, const char __user *, filename,
1969                 const char __user *const __user *, argv,
1970                 const char __user *const __user *, envp,
1971                 int, flags)
1972 {
1973         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1974
1975         return do_execveat(fd,
1976                            getname_flags(filename, lookup_flags, NULL),
1977                            argv, envp, flags);
1978 }
1979
1980 #ifdef CONFIG_COMPAT
1981 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1982         const compat_uptr_t __user *, argv,
1983         const compat_uptr_t __user *, envp)
1984 {
1985         return compat_do_execve(getname(filename), argv, envp);
1986 }
1987
1988 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1989                        const char __user *, filename,
1990                        const compat_uptr_t __user *, argv,
1991                        const compat_uptr_t __user *, envp,
1992                        int,  flags)
1993 {
1994         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1995
1996         return compat_do_execveat(fd,
1997                                   getname_flags(filename, lookup_flags, NULL),
1998                                   argv, envp, flags);
1999 }
2000 #endif