fs/exec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/exec.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  * #!-checking implemented by tytso.
  10  */
  11 /*
  12  * Demand-loading implemented 01.12.91 - no need to read anything but
  13  * the header into memory. The inode of the executable is put into
  14  * "current->executable", and page faults do the actual loading. Clean.
  15  *
  16  * Once more I can proudly say that linux stood up to being changed: it
  17  * was less than 2 hours work to get demand-loading completely implemented.
  18  *
  19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20  * current->executable is only used by the procfs.  This allows a dispatch
  21  * table to check for several different types  of binary formats.  We keep
  22  * trying until we recognize the file or we run out of supported binary
  23  * formats.
  24  */
  25
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/fdtable.h>
  29 #include <linux/mm.h>
  30 #include <linux/vmacache.h>
  31 #include <linux/stat.h>
  32 #include <linux/fcntl.h>
  33 #include <linux/swap.h>
  34 #include <linux/string.h>
  35 #include <linux/init.h>
  36 #include <linux/sched/mm.h>
  37 #include <linux/sched/coredump.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/sched/numa_balancing.h>
  40 #include <linux/sched/task.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/perf_event.h>
  43 #include <linux/highmem.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/key.h>
  46 #include <linux/personality.h>
  47 #include <linux/binfmts.h>
  48 #include <linux/utsname.h>
  49 #include <linux/pid_namespace.h>
  50 #include <linux/module.h>
  51 #include <linux/namei.h>
  52 #include <linux/mount.h>
  53 #include <linux/security.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/audit.h>
  58 #include <linux/tracehook.h>
  59 #include <linux/kmod.h>
  60 #include <linux/fsnotify.h>
  61 #include <linux/fs_struct.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                 acct_arg_size(bprm, vma_pages(bprm->vma));
 223
 224         return page;
 225 }
 226
 227 static void put_arg_page(struct page *page)
 228 {
 229         put_page(page);
 230 }
 231
 232 static void free_arg_pages(struct linux_binprm *bprm)
 233 {
 234 }
 235
 236 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 237                 struct page *page)
 238 {
 239         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 240 }
 241
 242 static int __bprm_mm_init(struct linux_binprm *bprm)
 243 {
 244         int err;
 245         struct vm_area_struct *vma = NULL;
 246         struct mm_struct *mm = bprm->mm;
 247
 248         bprm->vma = vma = vm_area_alloc(mm);
 249         if (!vma)
 250                 return -ENOMEM;
 251         vma_set_anonymous(vma);
 252
 253         if (down_write_killable(&mm->mmap_sem)) {
 254                 err = -EINTR;
 255                 goto err_free;
 256         }
 257
 258         /*
 259          * Place the stack at the largest stack address the architecture
 260          * supports. Later, we'll move this to an appropriate place. We don't
 261          * use STACK_TOP because that can depend on attributes which aren't
 262          * configured yet.
 263          */
 264         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 265         vma->vm_end = STACK_TOP_MAX;
 266         vma->vm_start = vma->vm_end - PAGE_SIZE;
 267         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 268         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 269
 270         err = insert_vm_struct(mm, vma);
 271         if (err)
 272                 goto err;
 273
 274         mm->stack_vm = mm->total_vm = 1;
 275         up_write(&mm->mmap_sem);
 276         bprm->p = vma->vm_end - sizeof(void *);
 277         return 0;
 278 err:
 279         up_write(&mm->mmap_sem);
 280 err_free:
 281         bprm->vma = NULL;
 282         vm_area_free(vma);
 283         return err;
 284 }
 285
 286 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 287 {
 288         return len <= MAX_ARG_STRLEN;
 289 }
 290
 291 #else
 292
 293 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 294 {
 295 }
 296
 297 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 298                 int write)
 299 {
 300         struct page *page;
 301
 302         page = bprm->page[pos / PAGE_SIZE];
 303         if (!page && write) {
 304                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 305                 if (!page)
 306                         return NULL;
 307                 bprm->page[pos / PAGE_SIZE] = page;
 308         }
 309
 310         return page;
 311 }
 312
 313 static void put_arg_page(struct page *page)
 314 {
 315 }
 316
 317 static void free_arg_page(struct linux_binprm *bprm, int i)
 318 {
 319         if (bprm->page[i]) {
 320                 __free_page(bprm->page[i]);
 321                 bprm->page[i] = NULL;
 322         }
 323 }
 324
 325 static void free_arg_pages(struct linux_binprm *bprm)
 326 {
 327         int i;
 328
 329         for (i = 0; i < MAX_ARG_PAGES; i++)
 330                 free_arg_page(bprm, i);
 331 }
 332
 333 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 334                 struct page *page)
 335 {
 336 }
 337
 338 static int __bprm_mm_init(struct linux_binprm *bprm)
 339 {
 340         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 341         return 0;
 342 }
 343
 344 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 345 {
 346         return len <= bprm->p;
 347 }
 348
 349 #endif /* CONFIG_MMU */
 350
 351 /*
 352  * Create a new mm_struct and populate it with a temporary stack
 353  * vm_area_struct.  We don't have enough context at this point to set the stack
 354  * flags, permissions, and offset, so we use temporary values.  We'll update
 355  * them later in setup_arg_pages().
 356  */
 357 static int bprm_mm_init(struct linux_binprm *bprm)
 358 {
 359         int err;
 360         struct mm_struct *mm = NULL;
 361
 362         bprm->mm = mm = mm_alloc();
 363         err = -ENOMEM;
 364         if (!mm)
 365                 goto err;
 366
 367         /* Save current stack limit for all calculations made during exec. */
 368         task_lock(current->group_leader);
 369         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 370         task_unlock(current->group_leader);
 371
 372         err = __bprm_mm_init(bprm);
 373         if (err)
 374                 goto err;
 375
 376         return 0;
 377
 378 err:
 379         if (mm) {
 380                 bprm->mm = NULL;
 381                 mmdrop(mm);
 382         }
 383
 384         return err;
 385 }
 386
 387 struct user_arg_ptr {
 388 #ifdef CONFIG_COMPAT
 389         bool is_compat;
 390 #endif
 391         union {
 392                 const char __user *const __user *native;
 393 #ifdef CONFIG_COMPAT
 394                 const compat_uptr_t __user *compat;
 395 #endif
 396         } ptr;
 397 };
 398
 399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 400 {
 401         const char __user *native;
 402
 403 #ifdef CONFIG_COMPAT
 404         if (unlikely(argv.is_compat)) {
 405                 compat_uptr_t compat;
 406
 407                 if (get_user(compat, argv.ptr.compat + nr))
 408                         return ERR_PTR(-EFAULT);
 409
 410                 return compat_ptr(compat);
 411         }
 412 #endif
 413
 414         if (get_user(native, argv.ptr.native + nr))
 415                 return ERR_PTR(-EFAULT);
 416
 417         return native;
 418 }
 419
 420 /*
 421  * count() counts the number of strings in array ARGV.
 422  */
 423 static int count(struct user_arg_ptr argv, int max)
 424 {
 425         int i = 0;
 426
 427         if (argv.ptr.native != NULL) {
 428                 for (;;) {
 429                         const char __user *p = get_user_arg_ptr(argv, i);
 430
 431                         if (!p)
 432                                 break;
 433
 434                         if (IS_ERR(p))
 435                                 return -EFAULT;
 436
 437                         if (i >= max)
 438                                 return -E2BIG;
 439                         ++i;
 440
 441                         if (fatal_signal_pending(current))
 442                                 return -ERESTARTNOHAND;
 443                         cond_resched();
 444                 }
 445         }
 446         return i;
 447 }
 448
 449 static int prepare_arg_pages(struct linux_binprm *bprm,
 450                         struct user_arg_ptr argv, struct user_arg_ptr envp)
 451 {
 452         unsigned long limit, ptr_size;
 453
 454         bprm->argc = count(argv, MAX_ARG_STRINGS);
 455         if (bprm->argc < 0)
 456                 return bprm->argc;
 457
 458         bprm->envc = count(envp, MAX_ARG_STRINGS);
 459         if (bprm->envc < 0)
 460                 return bprm->envc;
 461
 462         /*
 463          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 464          * (whichever is smaller) for the argv+env strings.
 465          * This ensures that:
 466          *  - the remaining binfmt code will not run out of stack space,
 467          *  - the program will have a reasonable amount of stack left
 468          *    to work from.
 469          */
 470         limit = _STK_LIM / 4 * 3;
 471         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 472         /*
 473          * We've historically supported up to 32 pages (ARG_MAX)
 474          * of argument strings even with small stacks
 475          */
 476         limit = max_t(unsigned long, limit, ARG_MAX);
 477         /*
 478          * We must account for the size of all the argv and envp pointers to
 479          * the argv and envp strings, since they will also take up space in
 480          * the stack. They aren't stored until much later when we can't
 481          * signal to the parent that the child has run out of stack space.
 482          * Instead, calculate it here so it's possible to fail gracefully.
 483          */
 484         ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 485         if (limit <= ptr_size)
 486                 return -E2BIG;
 487         limit -= ptr_size;
 488
 489         bprm->argmin = bprm->p - limit;
 490         return 0;
 491 }
 492
 493 /*
 494  * 'copy_strings()' copies argument/environment strings from the old
 495  * processes's memory to the new process's stack.  The call to get_user_pages()
 496  * ensures the destination page is created and not swapped out.
 497  */
 498 static int copy_strings(int argc, struct user_arg_ptr argv,
 499                         struct linux_binprm *bprm)
 500 {
 501         struct page *kmapped_page = NULL;
 502         char *kaddr = NULL;
 503         unsigned long kpos = 0;
 504         int ret;
 505
 506         while (argc-- > 0) {
 507                 const char __user *str;
 508                 int len;
 509                 unsigned long pos;
 510
 511                 ret = -EFAULT;
 512                 str = get_user_arg_ptr(argv, argc);
 513                 if (IS_ERR(str))
 514                         goto out;
 515
 516                 len = strnlen_user(str, MAX_ARG_STRLEN);
 517                 if (!len)
 518                         goto out;
 519
 520                 ret = -E2BIG;
 521                 if (!valid_arg_len(bprm, len))
 522                         goto out;
 523
 524                 /* We're going to work our way backwords. */
 525                 pos = bprm->p;
 526                 str += len;
 527                 bprm->p -= len;
 528 #ifdef CONFIG_MMU
 529                 if (bprm->p < bprm->argmin)
 530                         goto out;
 531 #endif
 532
 533                 while (len > 0) {
 534                         int offset, bytes_to_copy;
 535
 536                         if (fatal_signal_pending(current)) {
 537                                 ret = -ERESTARTNOHAND;
 538                                 goto out;
 539                         }
 540                         cond_resched();
 541
 542                         offset = pos % PAGE_SIZE;
 543                         if (offset == 0)
 544                                 offset = PAGE_SIZE;
 545
 546                         bytes_to_copy = offset;
 547                         if (bytes_to_copy > len)
 548                                 bytes_to_copy = len;
 549
 550                         offset -= bytes_to_copy;
 551                         pos -= bytes_to_copy;
 552                         str -= bytes_to_copy;
 553                         len -= bytes_to_copy;
 554
 555                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 556                                 struct page *page;
 557
 558                                 page = get_arg_page(bprm, pos, 1);
 559                                 if (!page) {
 560                                         ret = -E2BIG;
 561                                         goto out;
 562                                 }
 563
 564                                 if (kmapped_page) {
 565                                         flush_kernel_dcache_page(kmapped_page);
 566                                         kunmap(kmapped_page);
 567                                         put_arg_page(kmapped_page);
 568                                 }
 569                                 kmapped_page = page;
 570                                 kaddr = kmap(kmapped_page);
 571                                 kpos = pos & PAGE_MASK;
 572                                 flush_arg_page(bprm, kpos, kmapped_page);
 573                         }
 574                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 575                                 ret = -EFAULT;
 576                                 goto out;
 577                         }
 578                 }
 579         }
 580         ret = 0;
 581 out:
 582         if (kmapped_page) {
 583                 flush_kernel_dcache_page(kmapped_page);
 584                 kunmap(kmapped_page);
 585                 put_arg_page(kmapped_page);
 586         }
 587         return ret;
 588 }
 589
 590 /*
 591  * Like copy_strings, but get argv and its values from kernel memory.
 592  */
 593 int copy_strings_kernel(int argc, const char *const *__argv,
 594                         struct linux_binprm *bprm)
 595 {
 596         int r;
 597         mm_segment_t oldfs = get_fs();
 598         struct user_arg_ptr argv = {
 599                 .ptr.native = (const char __user *const  __user *)__argv,
 600         };
 601
 602         set_fs(KERNEL_DS);
 603         r = copy_strings(argc, argv, bprm);
 604         set_fs(oldfs);
 605
 606         return r;
 607 }
 608 EXPORT_SYMBOL(copy_strings_kernel);
 609
 610 #ifdef CONFIG_MMU
 611
 612 /*
 613  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 614  * the binfmt code determines where the new stack should reside, we shift it to
 615  * its final location.  The process proceeds as follows:
 616  *
 617  * 1) Use shift to calculate the new vma endpoints.
 618  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 619  *    arguments passed to subsequent functions are consistent.
 620  * 3) Move vma's page tables to the new range.
 621  * 4) Free up any cleared pgd range.
 622  * 5) Shrink the vma to cover only the new range.
 623  */
 624 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 625 {
 626         struct mm_struct *mm = vma->vm_mm;
 627         unsigned long old_start = vma->vm_start;
 628         unsigned long old_end = vma->vm_end;
 629         unsigned long length = old_end - old_start;
 630         unsigned long new_start = old_start - shift;
 631         unsigned long new_end = old_end - shift;
 632         struct mmu_gather tlb;
 633
 634         BUG_ON(new_start > new_end);
 635
 636         /*
 637          * ensure there are no vmas between where we want to go
 638          * and where we are
 639          */
 640         if (vma != find_vma(mm, new_start))
 641                 return -EFAULT;
 642
 643         /*
 644          * cover the whole range: [new_start, old_end)
 645          */
 646         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 647                 return -ENOMEM;
 648
 649         /*
 650          * move the page tables downwards, on failure we rely on
 651          * process cleanup to remove whatever mess we made.
 652          */
 653         if (length != move_page_tables(vma, old_start,
 654                                        vma, new_start, length, false))
 655                 return -ENOMEM;
 656
 657         lru_add_drain();
 658         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 659         if (new_end > old_start) {
 660                 /*
 661                  * when the old and new regions overlap clear from new_end.
 662                  */
 663                 free_pgd_range(&tlb, new_end, old_end, new_end,
 664                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 665         } else {
 666                 /*
 667                  * otherwise, clean from old_start; this is done to not touch
 668                  * the address space in [new_end, old_start) some architectures
 669                  * have constraints on va-space that make this illegal (IA64) -
 670                  * for the others its just a little faster.
 671                  */
 672                 free_pgd_range(&tlb, old_start, old_end, new_end,
 673                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 674         }
 675         tlb_finish_mmu(&tlb, old_start, old_end);
 676
 677         /*
 678          * Shrink the vma to just the new range.  Always succeeds.
 679          */
 680         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 681
 682         return 0;
 683 }
 684
 685 /*
 686  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 687  * the stack is optionally relocated, and some extra space is added.
 688  */
 689 int setup_arg_pages(struct linux_binprm *bprm,
 690                     unsigned long stack_top,
 691                     int executable_stack)
 692 {
 693         unsigned long ret;
 694         unsigned long stack_shift;
 695         struct mm_struct *mm = current->mm;
 696         struct vm_area_struct *vma = bprm->vma;
 697         struct vm_area_struct *prev = NULL;
 698         unsigned long vm_flags;
 699         unsigned long stack_base;
 700         unsigned long stack_size;
 701         unsigned long stack_expand;
 702         unsigned long rlim_stack;
 703
 704 #ifdef CONFIG_STACK_GROWSUP
 705         /* Limit stack size */
 706         stack_base = bprm->rlim_stack.rlim_max;
 707         if (stack_base > STACK_SIZE_MAX)
 708                 stack_base = STACK_SIZE_MAX;
 709
 710         /* Add space for stack randomization. */
 711         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 712
 713         /* Make sure we didn't let the argument array grow too large. */
 714         if (vma->vm_end - vma->vm_start > stack_base)
 715                 return -ENOMEM;
 716
 717         stack_base = PAGE_ALIGN(stack_top - stack_base);
 718
 719         stack_shift = vma->vm_start - stack_base;
 720         mm->arg_start = bprm->p - stack_shift;
 721         bprm->p = vma->vm_end - stack_shift;
 722 #else
 723         stack_top = arch_align_stack(stack_top);
 724         stack_top = PAGE_ALIGN(stack_top);
 725
 726         if (unlikely(stack_top < mmap_min_addr) ||
 727             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 728                 return -ENOMEM;
 729
 730         stack_shift = vma->vm_end - stack_top;
 731
 732         bprm->p -= stack_shift;
 733         mm->arg_start = bprm->p;
 734 #endif
 735
 736         if (bprm->loader)
 737                 bprm->loader -= stack_shift;
 738         bprm->exec -= stack_shift;
 739
 740         if (down_write_killable(&mm->mmap_sem))
 741                 return -EINTR;
 742
 743         vm_flags = VM_STACK_FLAGS;
 744
 745         /*
 746          * Adjust stack execute permissions; explicitly enable for
 747          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 748          * (arch default) otherwise.
 749          */
 750         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 751                 vm_flags |= VM_EXEC;
 752         else if (executable_stack == EXSTACK_DISABLE_X)
 753                 vm_flags &= ~VM_EXEC;
 754         vm_flags |= mm->def_flags;
 755         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 756
 757         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 758                         vm_flags);
 759         if (ret)
 760                 goto out_unlock;
 761         BUG_ON(prev != vma);
 762
 763         if (unlikely(vm_flags & VM_EXEC)) {
 764                 pr_warn_once("process '%pD4' started with executable stack\n",
 765                              bprm->file);
 766         }
 767
 768         /* Move stack pages down in memory. */
 769         if (stack_shift) {
 770                 ret = shift_arg_pages(vma, stack_shift);
 771                 if (ret)
 772                         goto out_unlock;
 773         }
 774
 775         /* mprotect_fixup is overkill to remove the temporary stack flags */
 776         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 777
 778         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 779         stack_size = vma->vm_end - vma->vm_start;
 780         /*
 781          * Align this down to a page boundary as expand_stack
 782          * will align it up.
 783          */
 784         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 785 #ifdef CONFIG_STACK_GROWSUP
 786         if (stack_size + stack_expand > rlim_stack)
 787                 stack_base = vma->vm_start + rlim_stack;
 788         else
 789                 stack_base = vma->vm_end + stack_expand;
 790 #else
 791         if (stack_size + stack_expand > rlim_stack)
 792                 stack_base = vma->vm_end - rlim_stack;
 793         else
 794                 stack_base = vma->vm_start - stack_expand;
 795 #endif
 796         current->mm->start_stack = bprm->p;
 797         ret = expand_stack(vma, stack_base);
 798         if (ret)
 799                 ret = -EFAULT;
 800
 801 out_unlock:
 802         up_write(&mm->mmap_sem);
 803         return ret;
 804 }
 805 EXPORT_SYMBOL(setup_arg_pages);
 806
 807 #else
 808
 809 /*
 810  * Transfer the program arguments and environment from the holding pages
 811  * onto the stack. The provided stack pointer is adjusted accordingly.
 812  */
 813 int transfer_args_to_stack(struct linux_binprm *bprm,
 814                            unsigned long *sp_location)
 815 {
 816         unsigned long index, stop, sp;
 817         int ret = 0;
 818
 819         stop = bprm->p >> PAGE_SHIFT;
 820         sp = *sp_location;
 821
 822         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 823                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 824                 char *src = kmap(bprm->page[index]) + offset;
 825                 sp -= PAGE_SIZE - offset;
 826                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 827                         ret = -EFAULT;
 828                 kunmap(bprm->page[index]);
 829                 if (ret)
 830                         goto out;
 831         }
 832
 833         *sp_location = sp;
 834
 835 out:
 836         return ret;
 837 }
 838 EXPORT_SYMBOL(transfer_args_to_stack);
 839
 840 #endif /* CONFIG_MMU */
 841
 842 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 843 {
 844         struct file *file;
 845         int err;
 846         struct open_flags open_exec_flags = {
 847                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 848                 .acc_mode = MAY_EXEC,
 849                 .intent = LOOKUP_OPEN,
 850                 .lookup_flags = LOOKUP_FOLLOW,
 851         };
 852
 853         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 854                 return ERR_PTR(-EINVAL);
 855         if (flags & AT_SYMLINK_NOFOLLOW)
 856                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 857         if (flags & AT_EMPTY_PATH)
 858                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 859
 860         file = do_filp_open(fd, name, &open_exec_flags);
 861         if (IS_ERR(file))
 862                 goto out;
 863
 864         err = -EACCES;
 865         if (!S_ISREG(file_inode(file)->i_mode))
 866                 goto exit;
 867
 868         if (path_noexec(&file->f_path))
 869                 goto exit;
 870
 871         err = deny_write_access(file);
 872         if (err)
 873                 goto exit;
 874
 875         if (name->name[0] != '\0')
 876                 fsnotify_open(file);
 877
 878 out:
 879         return file;
 880
 881 exit:
 882         fput(file);
 883         return ERR_PTR(err);
 884 }
 885
 886 struct file *open_exec(const char *name)
 887 {
 888         struct filename *filename = getname_kernel(name);
 889         struct file *f = ERR_CAST(filename);
 890
 891         if (!IS_ERR(filename)) {
 892                 f = do_open_execat(AT_FDCWD, filename, 0);
 893                 putname(filename);
 894         }
 895         return f;
 896 }
 897 EXPORT_SYMBOL(open_exec);
 898
 899 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 900                      loff_t max_size, enum kernel_read_file_id id)
 901 {
 902         loff_t i_size, pos;
 903         ssize_t bytes = 0;
 904         int ret;
 905
 906         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 907                 return -EINVAL;
 908
 909         ret = deny_write_access(file);
 910         if (ret)
 911                 return ret;
 912
 913         ret = security_kernel_read_file(file, id);
 914         if (ret)
 915                 goto out;
 916
 917         i_size = i_size_read(file_inode(file));
 918         if (i_size <= 0) {
 919                 ret = -EINVAL;
 920                 goto out;
 921         }
 922         if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 923                 ret = -EFBIG;
 924                 goto out;
 925         }
 926
 927         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 928                 *buf = vmalloc(i_size);
 929         if (!*buf) {
 930                 ret = -ENOMEM;
 931                 goto out;
 932         }
 933
 934         pos = 0;
 935         while (pos < i_size) {
 936                 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 937                 if (bytes < 0) {
 938                         ret = bytes;
 939                         goto out_free;
 940                 }
 941
 942                 if (bytes == 0)
 943                         break;
 944         }
 945
 946         if (pos != i_size) {
 947                 ret = -EIO;
 948                 goto out_free;
 949         }
 950
 951         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 952         if (!ret)
 953                 *size = pos;
 954
 955 out_free:
 956         if (ret < 0) {
 957                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 958                         vfree(*buf);
 959                         *buf = NULL;
 960                 }
 961         }
 962
 963 out:
 964         allow_write_access(file);
 965         return ret;
 966 }
 967 EXPORT_SYMBOL_GPL(kernel_read_file);
 968
 969 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
 970                                loff_t max_size, enum kernel_read_file_id id)
 971 {
 972         struct file *file;
 973         int ret;
 974
 975         if (!path || !*path)
 976                 return -EINVAL;
 977
 978         file = filp_open(path, O_RDONLY, 0);
 979         if (IS_ERR(file))
 980                 return PTR_ERR(file);
 981
 982         ret = kernel_read_file(file, buf, size, max_size, id);
 983         fput(file);
 984         return ret;
 985 }
 986 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 987
 988 int kernel_read_file_from_path_initns(const char *path, void **buf,
 989                                       loff_t *size, loff_t max_size,
 990                                       enum kernel_read_file_id id)
 991 {
 992         struct file *file;
 993         struct path root;
 994         int ret;
 995
 996         if (!path || !*path)
 997                 return -EINVAL;
 998
 999         task_lock(&init_task);
1000         get_fs_root(init_task.fs, &root);
1001         task_unlock(&init_task);
1002
1003         file = file_open_root(root.dentry, root.mnt, path, O_RDONLY, 0);
1004         path_put(&root);
1005         if (IS_ERR(file))
1006                 return PTR_ERR(file);
1007
1008         ret = kernel_read_file(file, buf, size, max_size, id);
1009         fput(file);
1010         return ret;
1011 }
1012 EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns);
1013
1014 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
1015                              enum kernel_read_file_id id)
1016 {
1017         struct fd f = fdget(fd);
1018         int ret = -EBADF;
1019
1020         if (!f.file)
1021                 goto out;
1022
1023         ret = kernel_read_file(f.file, buf, size, max_size, id);
1024 out:
1025         fdput(f);
1026         return ret;
1027 }
1028 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1029
1030 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1031 {
1032         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1033         if (res > 0)
1034                 flush_icache_range(addr, addr + len);
1035         return res;
1036 }
1037 EXPORT_SYMBOL(read_code);
1038
1039 /*
1040  * Maps the mm_struct mm into the current task struct.
1041  * On success, this function returns with the mutex
1042  * exec_update_mutex locked.
1043  */
1044 static int exec_mmap(struct mm_struct *mm)
1045 {
1046         struct task_struct *tsk;
1047         struct mm_struct *old_mm, *active_mm;
1048         int ret;
1049
1050         /* Notify parent that we're no longer interested in the old VM */
1051         tsk = current;
1052         old_mm = current->mm;
1053         exec_mm_release(tsk, old_mm);
1054
1055         ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
1056         if (ret)
1057                 return ret;
1058
1059         if (old_mm) {
1060                 sync_mm_rss(old_mm);
1061                 /*
1062                  * Make sure that if there is a core dump in progress
1063                  * for the old mm, we get out and die instead of going
1064                  * through with the exec.  We must hold mmap_sem around
1065                  * checking core_state and changing tsk->mm.
1066                  */
1067                 down_read(&old_mm->mmap_sem);
1068                 if (unlikely(old_mm->core_state)) {
1069                         up_read(&old_mm->mmap_sem);
1070                         mutex_unlock(&tsk->signal->exec_update_mutex);
1071                         return -EINTR;
1072                 }
1073         }
1074
1075         task_lock(tsk);
1076         active_mm = tsk->active_mm;
1077         membarrier_exec_mmap(mm);
1078         tsk->mm = mm;
1079         tsk->active_mm = mm;
1080         activate_mm(active_mm, mm);
1081         tsk->mm->vmacache_seqnum = 0;
1082         vmacache_flush(tsk);
1083         task_unlock(tsk);
1084         if (old_mm) {
1085                 up_read(&old_mm->mmap_sem);
1086                 BUG_ON(active_mm != old_mm);
1087                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1088                 mm_update_next_owner(old_mm);
1089                 mmput(old_mm);
1090                 return 0;
1091         }
1092         mmdrop(active_mm);
1093         return 0;
1094 }
1095
1096 /*
1097  * This function makes sure the current process has its own signal table,
1098  * so that flush_signal_handlers can later reset the handlers without
1099  * disturbing other processes.  (Other processes might share the signal
1100  * table via the CLONE_SIGHAND option to clone().)
1101  */
1102 static int de_thread(struct task_struct *tsk)
1103 {
1104         struct signal_struct *sig = tsk->signal;
1105         struct sighand_struct *oldsighand = tsk->sighand;
1106         spinlock_t *lock = &oldsighand->siglock;
1107
1108         if (thread_group_empty(tsk))
1109                 goto no_thread_group;
1110
1111         /*
1112          * Kill all other threads in the thread group.
1113          */
1114         spin_lock_irq(lock);
1115         if (signal_group_exit(sig)) {
1116                 /*
1117                  * Another group action in progress, just
1118                  * return so that the signal is processed.
1119                  */
1120                 spin_unlock_irq(lock);
1121                 return -EAGAIN;
1122         }
1123
1124         sig->group_exit_task = tsk;
1125         sig->notify_count = zap_other_threads(tsk);
1126         if (!thread_group_leader(tsk))
1127                 sig->notify_count--;
1128
1129         while (sig->notify_count) {
1130                 __set_current_state(TASK_KILLABLE);
1131                 spin_unlock_irq(lock);
1132                 schedule();
1133                 if (__fatal_signal_pending(tsk))
1134                         goto killed;
1135                 spin_lock_irq(lock);
1136         }
1137         spin_unlock_irq(lock);
1138
1139         /*
1140          * At this point all other threads have exited, all we have to
1141          * do is to wait for the thread group leader to become inactive,
1142          * and to assume its PID:
1143          */
1144         if (!thread_group_leader(tsk)) {
1145                 struct task_struct *leader = tsk->group_leader;
1146
1147                 for (;;) {
1148                         cgroup_threadgroup_change_begin(tsk);
1149                         write_lock_irq(&tasklist_lock);
1150                         /*
1151                          * Do this under tasklist_lock to ensure that
1152                          * exit_notify() can't miss ->group_exit_task
1153                          */
1154                         sig->notify_count = -1;
1155                         if (likely(leader->exit_state))
1156                                 break;
1157                         __set_current_state(TASK_KILLABLE);
1158                         write_unlock_irq(&tasklist_lock);
1159                         cgroup_threadgroup_change_end(tsk);
1160                         schedule();
1161                         if (__fatal_signal_pending(tsk))
1162                                 goto killed;
1163                 }
1164
1165                 /*
1166                  * The only record we have of the real-time age of a
1167                  * process, regardless of execs it's done, is start_time.
1168                  * All the past CPU time is accumulated in signal_struct
1169                  * from sister threads now dead.  But in this non-leader
1170                  * exec, nothing survives from the original leader thread,
1171                  * whose birth marks the true age of this process now.
1172                  * When we take on its identity by switching to its PID, we
1173                  * also take its birthdate (always earlier than our own).
1174                  */
1175                 tsk->start_time = leader->start_time;
1176                 tsk->start_boottime = leader->start_boottime;
1177
1178                 BUG_ON(!same_thread_group(leader, tsk));
1179                 BUG_ON(has_group_leader_pid(tsk));
1180                 /*
1181                  * An exec() starts a new thread group with the
1182                  * TGID of the previous thread group. Rehash the
1183                  * two threads with a switched PID, and release
1184                  * the former thread group leader:
1185                  */
1186
1187                 /* Become a process group leader with the old leader's pid.
1188                  * The old leader becomes a thread of the this thread group.
1189                  * Note: The old leader also uses this pid until release_task
1190                  *       is called.  Odd but simple and correct.
1191                  */
1192                 tsk->pid = leader->pid;
1193                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1194                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1195                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1196                 transfer_pid(leader, tsk, PIDTYPE_SID);
1197
1198                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1199                 list_replace_init(&leader->sibling, &tsk->sibling);
1200
1201                 tsk->group_leader = tsk;
1202                 leader->group_leader = tsk;
1203
1204                 tsk->exit_signal = SIGCHLD;
1205                 leader->exit_signal = -1;
1206
1207                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1208                 leader->exit_state = EXIT_DEAD;
1209
1210                 /*
1211                  * We are going to release_task()->ptrace_unlink() silently,
1212                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1213                  * the tracer wont't block again waiting for this thread.
1214                  */
1215                 if (unlikely(leader->ptrace))
1216                         __wake_up_parent(leader, leader->parent);
1217                 write_unlock_irq(&tasklist_lock);
1218                 cgroup_threadgroup_change_end(tsk);
1219
1220                 release_task(leader);
1221         }
1222
1223         sig->group_exit_task = NULL;
1224         sig->notify_count = 0;
1225
1226 no_thread_group:
1227         /* we have changed execution domain */
1228         tsk->exit_signal = SIGCHLD;
1229
1230         BUG_ON(!thread_group_leader(tsk));
1231         return 0;
1232
1233 killed:
1234         /* protects against exit_notify() and __exit_signal() */
1235         read_lock(&tasklist_lock);
1236         sig->group_exit_task = NULL;
1237         sig->notify_count = 0;
1238         read_unlock(&tasklist_lock);
1239         return -EAGAIN;
1240 }
1241
1242
1243 static int unshare_sighand(struct task_struct *me)
1244 {
1245         struct sighand_struct *oldsighand = me->sighand;
1246
1247         if (refcount_read(&oldsighand->count) != 1) {
1248                 struct sighand_struct *newsighand;
1249                 /*
1250                  * This ->sighand is shared with the CLONE_SIGHAND
1251                  * but not CLONE_THREAD task, switch to the new one.
1252                  */
1253                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1254                 if (!newsighand)
1255                         return -ENOMEM;
1256
1257                 refcount_set(&newsighand->count, 1);
1258                 memcpy(newsighand->action, oldsighand->action,
1259                        sizeof(newsighand->action));
1260
1261                 write_lock_irq(&tasklist_lock);
1262                 spin_lock(&oldsighand->siglock);
1263                 rcu_assign_pointer(me->sighand, newsighand);
1264                 spin_unlock(&oldsighand->siglock);
1265                 write_unlock_irq(&tasklist_lock);
1266
1267                 __cleanup_sighand(oldsighand);
1268         }
1269         return 0;
1270 }
1271
1272 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1273 {
1274         task_lock(tsk);
1275         strncpy(buf, tsk->comm, buf_size);
1276         task_unlock(tsk);
1277         return buf;
1278 }
1279 EXPORT_SYMBOL_GPL(__get_task_comm);
1280
1281 /*
1282  * These functions flushes out all traces of the currently running executable
1283  * so that a new one can be started
1284  */
1285
1286 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1287 {
1288         task_lock(tsk);
1289         trace_task_rename(tsk, buf);
1290         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1291         task_unlock(tsk);
1292         perf_event_comm(tsk, exec);
1293 }
1294
1295 /*
1296  * Calling this is the point of no return. None of the failures will be
1297  * seen by userspace since either the process is already taking a fatal
1298  * signal (via de_thread() or coredump), or will have SEGV raised
1299  * (after exec_mmap()) by search_binary_handlers (see below).
1300  */
1301 int flush_old_exec(struct linux_binprm * bprm)
1302 {
1303         struct task_struct *me = current;
1304         int retval;
1305
1306         /*
1307          * Make this the only thread in the thread group.
1308          */
1309         retval = de_thread(me);
1310         if (retval)
1311                 goto out;
1312
1313         /*
1314          * Must be called _before_ exec_mmap() as bprm->mm is
1315          * not visibile until then. This also enables the update
1316          * to be lockless.
1317          */
1318         set_mm_exe_file(bprm->mm, bprm->file);
1319
1320         /*
1321          * Release all of the old mmap stuff
1322          */
1323         acct_arg_size(bprm, 0);
1324         retval = exec_mmap(bprm->mm);
1325         if (retval)
1326                 goto out;
1327
1328         /*
1329          * After setting bprm->called_exec_mmap (to mark that current is
1330          * using the prepared mm now), we have nothing left of the original
1331          * process. If anything from here on returns an error, the check
1332          * in search_binary_handler() will SEGV current.
1333          */
1334         bprm->called_exec_mmap = 1;
1335         bprm->mm = NULL;
1336
1337 #ifdef CONFIG_POSIX_TIMERS
1338         exit_itimers(me->signal);
1339         flush_itimer_signals();
1340 #endif
1341
1342         /*
1343          * Make the signal table private.
1344          */
1345         retval = unshare_sighand(me);
1346         if (retval)
1347                 goto out;
1348
1349         set_fs(USER_DS);
1350         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1351                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1352         flush_thread();
1353         me->personality &= ~bprm->per_clear;
1354
1355         /*
1356          * We have to apply CLOEXEC before we change whether the process is
1357          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1358          * trying to access the should-be-closed file descriptors of a process
1359          * undergoing exec(2).
1360          */
1361         do_close_on_exec(me->files);
1362         return 0;
1363
1364 out:
1365         return retval;
1366 }
1367 EXPORT_SYMBOL(flush_old_exec);
1368
1369 void would_dump(struct linux_binprm *bprm, struct file *file)
1370 {
1371         struct inode *inode = file_inode(file);
1372         if (inode_permission(inode, MAY_READ) < 0) {
1373                 struct user_namespace *old, *user_ns;
1374                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1375
1376                 /* Ensure mm->user_ns contains the executable */
1377                 user_ns = old = bprm->mm->user_ns;
1378                 while ((user_ns != &init_user_ns) &&
1379                        !privileged_wrt_inode_uidgid(user_ns, inode))
1380                         user_ns = user_ns->parent;
1381
1382                 if (old != user_ns) {
1383                         bprm->mm->user_ns = get_user_ns(user_ns);
1384                         put_user_ns(old);
1385                 }
1386         }
1387 }
1388 EXPORT_SYMBOL(would_dump);
1389
1390 void setup_new_exec(struct linux_binprm * bprm)
1391 {
1392         /*
1393          * Once here, prepare_binrpm() will not be called any more, so
1394          * the final state of setuid/setgid/fscaps can be merged into the
1395          * secureexec flag.
1396          */
1397         bprm->secureexec |= bprm->cap_elevated;
1398
1399         if (bprm->secureexec) {
1400                 /* Make sure parent cannot signal privileged process. */
1401                 current->pdeath_signal = 0;
1402
1403                 /*
1404                  * For secureexec, reset the stack limit to sane default to
1405                  * avoid bad behavior from the prior rlimits. This has to
1406                  * happen before arch_pick_mmap_layout(), which examines
1407                  * RLIMIT_STACK, but after the point of no return to avoid
1408                  * needing to clean up the change on failure.
1409                  */
1410                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1411                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1412         }
1413
1414         arch_pick_mmap_layout(current->mm, &bprm->rlim_stack);
1415
1416         current->sas_ss_sp = current->sas_ss_size = 0;
1417
1418         /*
1419          * Figure out dumpability. Note that this checking only of current
1420          * is wrong, but userspace depends on it. This should be testing
1421          * bprm->secureexec instead.
1422          */
1423         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1424             !(uid_eq(current_euid(), current_uid()) &&
1425               gid_eq(current_egid(), current_gid())))
1426                 set_dumpable(current->mm, suid_dumpable);
1427         else
1428                 set_dumpable(current->mm, SUID_DUMP_USER);
1429
1430         arch_setup_new_exec();
1431         perf_event_exec();
1432         __set_task_comm(current, kbasename(bprm->filename), true);
1433
1434         /* Set the new mm task size. We have to do that late because it may
1435          * depend on TIF_32BIT which is only updated in flush_thread() on
1436          * some architectures like powerpc
1437          */
1438         current->mm->task_size = TASK_SIZE;
1439
1440         /* An exec changes our domain. We are no longer part of the thread
1441            group */
1442         WRITE_ONCE(current->self_exec_id, current->self_exec_id + 1);
1443         flush_signal_handlers(current, 0);
1444 }
1445 EXPORT_SYMBOL(setup_new_exec);
1446
1447 /* Runs immediately before start_thread() takes over. */
1448 void finalize_exec(struct linux_binprm *bprm)
1449 {
1450         /* Store any stack rlimit changes before starting thread. */
1451         task_lock(current->group_leader);
1452         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1453         task_unlock(current->group_leader);
1454 }
1455 EXPORT_SYMBOL(finalize_exec);
1456
1457 /*
1458  * Prepare credentials and lock ->cred_guard_mutex.
1459  * install_exec_creds() commits the new creds and drops the lock.
1460  * Or, if exec fails before, free_bprm() should release ->cred and
1461  * and unlock.
1462  */
1463 static int prepare_bprm_creds(struct linux_binprm *bprm)
1464 {
1465         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1466                 return -ERESTARTNOINTR;
1467
1468         bprm->cred = prepare_exec_creds();
1469         if (likely(bprm->cred))
1470                 return 0;
1471
1472         mutex_unlock(&current->signal->cred_guard_mutex);
1473         return -ENOMEM;
1474 }
1475
1476 static void free_bprm(struct linux_binprm *bprm)
1477 {
1478         free_arg_pages(bprm);
1479         if (bprm->cred) {
1480                 if (bprm->called_exec_mmap)
1481                         mutex_unlock(&current->signal->exec_update_mutex);
1482                 mutex_unlock(&current->signal->cred_guard_mutex);
1483                 abort_creds(bprm->cred);
1484         }
1485         if (bprm->file) {
1486                 allow_write_access(bprm->file);
1487                 fput(bprm->file);
1488         }
1489         /* If a binfmt changed the interp, free it. */
1490         if (bprm->interp != bprm->filename)
1491                 kfree(bprm->interp);
1492         kfree(bprm);
1493 }
1494
1495 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1496 {
1497         /* If a binfmt changed the interp, free it first. */
1498         if (bprm->interp != bprm->filename)
1499                 kfree(bprm->interp);
1500         bprm->interp = kstrdup(interp, GFP_KERNEL);
1501         if (!bprm->interp)
1502                 return -ENOMEM;
1503         return 0;
1504 }
1505 EXPORT_SYMBOL(bprm_change_interp);
1506
1507 /*
1508  * install the new credentials for this executable
1509  */
1510 void install_exec_creds(struct linux_binprm *bprm)
1511 {
1512         security_bprm_committing_creds(bprm);
1513
1514         commit_creds(bprm->cred);
1515         bprm->cred = NULL;
1516
1517         /*
1518          * Disable monitoring for regular users
1519          * when executing setuid binaries. Must
1520          * wait until new credentials are committed
1521          * by commit_creds() above
1522          */
1523         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1524                 perf_event_exit_task(current);
1525         /*
1526          * cred_guard_mutex must be held at least to this point to prevent
1527          * ptrace_attach() from altering our determination of the task's
1528          * credentials; any time after this it may be unlocked.
1529          */
1530         security_bprm_committed_creds(bprm);
1531         mutex_unlock(&current->signal->exec_update_mutex);
1532         mutex_unlock(&current->signal->cred_guard_mutex);
1533 }
1534 EXPORT_SYMBOL(install_exec_creds);
1535
1536 /*
1537  * determine how safe it is to execute the proposed program
1538  * - the caller must hold ->cred_guard_mutex to protect against
1539  *   PTRACE_ATTACH or seccomp thread-sync
1540  */
1541 static void check_unsafe_exec(struct linux_binprm *bprm)
1542 {
1543         struct task_struct *p = current, *t;
1544         unsigned n_fs;
1545
1546         if (p->ptrace)
1547                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1548
1549         /*
1550          * This isn't strictly necessary, but it makes it harder for LSMs to
1551          * mess up.
1552          */
1553         if (task_no_new_privs(current))
1554                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1555
1556         t = p;
1557         n_fs = 1;
1558         spin_lock(&p->fs->lock);
1559         rcu_read_lock();
1560         while_each_thread(p, t) {
1561                 if (t->fs == p->fs)
1562                         n_fs++;
1563         }
1564         rcu_read_unlock();
1565
1566         if (p->fs->users > n_fs)
1567                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1568         else
1569                 p->fs->in_exec = 1;
1570         spin_unlock(&p->fs->lock);
1571 }
1572
1573 static void bprm_fill_uid(struct linux_binprm *bprm)
1574 {
1575         struct inode *inode;
1576         unsigned int mode;
1577         kuid_t uid;
1578         kgid_t gid;
1579
1580         /*
1581          * Since this can be called multiple times (via prepare_binprm),
1582          * we must clear any previous work done when setting set[ug]id
1583          * bits from any earlier bprm->file uses (for example when run
1584          * first for a setuid script then again for its interpreter).
1585          */
1586         bprm->cred->euid = current_euid();
1587         bprm->cred->egid = current_egid();
1588
1589         if (!mnt_may_suid(bprm->file->f_path.mnt))
1590                 return;
1591
1592         if (task_no_new_privs(current))
1593                 return;
1594
1595         inode = bprm->file->f_path.dentry->d_inode;
1596         mode = READ_ONCE(inode->i_mode);
1597         if (!(mode & (S_ISUID|S_ISGID)))
1598                 return;
1599
1600         /* Be careful if suid/sgid is set */
1601         inode_lock(inode);
1602
1603         /* reload atomically mode/uid/gid now that lock held */
1604         mode = inode->i_mode;
1605         uid = inode->i_uid;
1606         gid = inode->i_gid;
1607         inode_unlock(inode);
1608
1609         /* We ignore suid/sgid if there are no mappings for them in the ns */
1610         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1611                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1612                 return;
1613
1614         if (mode & S_ISUID) {
1615                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1616                 bprm->cred->euid = uid;
1617         }
1618
1619         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1620                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1621                 bprm->cred->egid = gid;
1622         }
1623 }
1624
1625 /*
1626  * Fill the binprm structure from the inode.
1627  * Check permissions, then read the first BINPRM_BUF_SIZE bytes
1628  *
1629  * This may be called multiple times for binary chains (scripts for example).
1630  */
1631 int prepare_binprm(struct linux_binprm *bprm)
1632 {
1633         int retval;
1634         loff_t pos = 0;
1635
1636         bprm_fill_uid(bprm);
1637
1638         /* fill in binprm security blob */
1639         retval = security_bprm_set_creds(bprm);
1640         if (retval)
1641                 return retval;
1642         bprm->called_set_creds = 1;
1643
1644         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1645         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1646 }
1647
1648 EXPORT_SYMBOL(prepare_binprm);
1649
1650 /*
1651  * Arguments are '\0' separated strings found at the location bprm->p
1652  * points to; chop off the first by relocating brpm->p to right after
1653  * the first '\0' encountered.
1654  */
1655 int remove_arg_zero(struct linux_binprm *bprm)
1656 {
1657         int ret = 0;
1658         unsigned long offset;
1659         char *kaddr;
1660         struct page *page;
1661
1662         if (!bprm->argc)
1663                 return 0;
1664
1665         do {
1666                 offset = bprm->p & ~PAGE_MASK;
1667                 page = get_arg_page(bprm, bprm->p, 0);
1668                 if (!page) {
1669                         ret = -EFAULT;
1670                         goto out;
1671                 }
1672                 kaddr = kmap_atomic(page);
1673
1674                 for (; offset < PAGE_SIZE && kaddr[offset];
1675                                 offset++, bprm->p++)
1676                         ;
1677
1678                 kunmap_atomic(kaddr);
1679                 put_arg_page(page);
1680         } while (offset == PAGE_SIZE);
1681
1682         bprm->p++;
1683         bprm->argc--;
1684         ret = 0;
1685
1686 out:
1687         return ret;
1688 }
1689 EXPORT_SYMBOL(remove_arg_zero);
1690
1691 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1692 /*
1693  * cycle the list of binary formats handler, until one recognizes the image
1694  */
1695 int search_binary_handler(struct linux_binprm *bprm)
1696 {
1697         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1698         struct linux_binfmt *fmt;
1699         int retval;
1700
1701         /* This allows 4 levels of binfmt rewrites before failing hard. */
1702         if (bprm->recursion_depth > 5)
1703                 return -ELOOP;
1704
1705         retval = security_bprm_check(bprm);
1706         if (retval)
1707                 return retval;
1708
1709         retval = -ENOENT;
1710  retry:
1711         read_lock(&binfmt_lock);
1712         list_for_each_entry(fmt, &formats, lh) {
1713                 if (!try_module_get(fmt->module))
1714                         continue;
1715                 read_unlock(&binfmt_lock);
1716
1717                 bprm->recursion_depth++;
1718                 retval = fmt->load_binary(bprm);
1719                 bprm->recursion_depth--;
1720
1721                 read_lock(&binfmt_lock);
1722                 put_binfmt(fmt);
1723                 if (retval < 0 && bprm->called_exec_mmap) {
1724                         /* we got to flush_old_exec() and failed after it */
1725                         read_unlock(&binfmt_lock);
1726                         force_sigsegv(SIGSEGV);
1727                         return retval;
1728                 }
1729                 if (retval != -ENOEXEC || !bprm->file) {
1730                         read_unlock(&binfmt_lock);
1731                         return retval;
1732                 }
1733         }
1734         read_unlock(&binfmt_lock);
1735
1736         if (need_retry) {
1737                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1738                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1739                         return retval;
1740                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1741                         return retval;
1742                 need_retry = false;
1743                 goto retry;
1744         }
1745
1746         return retval;
1747 }
1748 EXPORT_SYMBOL(search_binary_handler);
1749
1750 static int exec_binprm(struct linux_binprm *bprm)
1751 {
1752         pid_t old_pid, old_vpid;
1753         int ret;
1754
1755         /* Need to fetch pid before load_binary changes it */
1756         old_pid = current->pid;
1757         rcu_read_lock();
1758         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1759         rcu_read_unlock();
1760
1761         ret = search_binary_handler(bprm);
1762         if (ret >= 0) {
1763                 audit_bprm(bprm);
1764                 trace_sched_process_exec(current, old_pid, bprm);
1765                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1766                 proc_exec_connector(current);
1767         }
1768
1769         return ret;
1770 }
1771
1772 /*
1773  * sys_execve() executes a new program.
1774  */
1775 static int __do_execve_file(int fd, struct filename *filename,
1776                             struct user_arg_ptr argv,
1777                             struct user_arg_ptr envp,
1778                             int flags, struct file *file)
1779 {
1780         char *pathbuf = NULL;
1781         struct linux_binprm *bprm;
1782         struct files_struct *displaced;
1783         int retval;
1784
1785         if (IS_ERR(filename))
1786                 return PTR_ERR(filename);
1787
1788         /*
1789          * We move the actual failure in case of RLIMIT_NPROC excess from
1790          * set*uid() to execve() because too many poorly written programs
1791          * don't check setuid() return code.  Here we additionally recheck
1792          * whether NPROC limit is still exceeded.
1793          */
1794         if ((current->flags & PF_NPROC_EXCEEDED) &&
1795             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1796                 retval = -EAGAIN;
1797                 goto out_ret;
1798         }
1799
1800         /* We're below the limit (still or again), so we don't want to make
1801          * further execve() calls fail. */
1802         current->flags &= ~PF_NPROC_EXCEEDED;
1803
1804         retval = unshare_files(&displaced);
1805         if (retval)
1806                 goto out_ret;
1807
1808         retval = -ENOMEM;
1809         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1810         if (!bprm)
1811                 goto out_files;
1812
1813         retval = prepare_bprm_creds(bprm);
1814         if (retval)
1815                 goto out_free;
1816
1817         check_unsafe_exec(bprm);
1818         current->in_execve = 1;
1819
1820         if (!file)
1821                 file = do_open_execat(fd, filename, flags);
1822         retval = PTR_ERR(file);
1823         if (IS_ERR(file))
1824                 goto out_unmark;
1825
1826         sched_exec();
1827
1828         bprm->file = file;
1829         if (!filename) {
1830                 bprm->filename = "none";
1831         } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1832                 bprm->filename = filename->name;
1833         } else {
1834                 if (filename->name[0] == '\0')
1835                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1836                 else
1837                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1838                                             fd, filename->name);
1839                 if (!pathbuf) {
1840                         retval = -ENOMEM;
1841                         goto out_unmark;
1842                 }
1843                 /*
1844                  * Record that a name derived from an O_CLOEXEC fd will be
1845                  * inaccessible after exec. Relies on having exclusive access to
1846                  * current->files (due to unshare_files above).
1847                  */
1848                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1849                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1850                 bprm->filename = pathbuf;
1851         }
1852         bprm->interp = bprm->filename;
1853
1854         retval = bprm_mm_init(bprm);
1855         if (retval)
1856                 goto out_unmark;
1857
1858         retval = prepare_arg_pages(bprm, argv, envp);
1859         if (retval < 0)
1860                 goto out;
1861
1862         retval = prepare_binprm(bprm);
1863         if (retval < 0)
1864                 goto out;
1865
1866         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1867         if (retval < 0)
1868                 goto out;
1869
1870         bprm->exec = bprm->p;
1871         retval = copy_strings(bprm->envc, envp, bprm);
1872         if (retval < 0)
1873                 goto out;
1874
1875         retval = copy_strings(bprm->argc, argv, bprm);
1876         if (retval < 0)
1877                 goto out;
1878
1879         would_dump(bprm, bprm->file);
1880
1881         retval = exec_binprm(bprm);
1882         if (retval < 0)
1883                 goto out;
1884
1885         /* execve succeeded */
1886         current->fs->in_exec = 0;
1887         current->in_execve = 0;
1888         rseq_execve(current);
1889         acct_update_integrals(current);
1890         task_numa_free(current, false);
1891         free_bprm(bprm);
1892         kfree(pathbuf);
1893         if (filename)
1894                 putname(filename);
1895         if (displaced)
1896                 put_files_struct(displaced);
1897         return retval;
1898
1899 out:
1900         if (bprm->mm) {
1901                 acct_arg_size(bprm, 0);
1902                 mmput(bprm->mm);
1903         }
1904
1905 out_unmark:
1906         current->fs->in_exec = 0;
1907         current->in_execve = 0;
1908
1909 out_free:
1910         free_bprm(bprm);
1911         kfree(pathbuf);
1912
1913 out_files:
1914         if (displaced)
1915                 reset_files_struct(displaced);
1916 out_ret:
1917         if (filename)
1918                 putname(filename);
1919         return retval;
1920 }
1921
1922 static int do_execveat_common(int fd, struct filename *filename,
1923                               struct user_arg_ptr argv,
1924                               struct user_arg_ptr envp,
1925                               int flags)
1926 {
1927         return __do_execve_file(fd, filename, argv, envp, flags, NULL);
1928 }
1929
1930 int do_execve_file(struct file *file, void *__argv, void *__envp)
1931 {
1932         struct user_arg_ptr argv = { .ptr.native = __argv };
1933         struct user_arg_ptr envp = { .ptr.native = __envp };
1934
1935         return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file);
1936 }
1937
1938 int do_execve(struct filename *filename,
1939         const char __user *const __user *__argv,
1940         const char __user *const __user *__envp)
1941 {
1942         struct user_arg_ptr argv = { .ptr.native = __argv };
1943         struct user_arg_ptr envp = { .ptr.native = __envp };
1944         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1945 }
1946
1947 int do_execveat(int fd, struct filename *filename,
1948                 const char __user *const __user *__argv,
1949                 const char __user *const __user *__envp,
1950                 int flags)
1951 {
1952         struct user_arg_ptr argv = { .ptr.native = __argv };
1953         struct user_arg_ptr envp = { .ptr.native = __envp };
1954
1955         return do_execveat_common(fd, filename, argv, envp, flags);
1956 }
1957
1958 #ifdef CONFIG_COMPAT
1959 static int compat_do_execve(struct filename *filename,
1960         const compat_uptr_t __user *__argv,
1961         const compat_uptr_t __user *__envp)
1962 {
1963         struct user_arg_ptr argv = {
1964                 .is_compat = true,
1965                 .ptr.compat = __argv,
1966         };
1967         struct user_arg_ptr envp = {
1968                 .is_compat = true,
1969                 .ptr.compat = __envp,
1970         };
1971         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1972 }
1973
1974 static int compat_do_execveat(int fd, struct filename *filename,
1975                               const compat_uptr_t __user *__argv,
1976                               const compat_uptr_t __user *__envp,
1977                               int flags)
1978 {
1979         struct user_arg_ptr argv = {
1980                 .is_compat = true,
1981                 .ptr.compat = __argv,
1982         };
1983         struct user_arg_ptr envp = {
1984                 .is_compat = true,
1985                 .ptr.compat = __envp,
1986         };
1987         return do_execveat_common(fd, filename, argv, envp, flags);
1988 }
1989 #endif
1990
1991 void set_binfmt(struct linux_binfmt *new)
1992 {
1993         struct mm_struct *mm = current->mm;
1994
1995         if (mm->binfmt)
1996                 module_put(mm->binfmt->module);
1997
1998         mm->binfmt = new;
1999         if (new)
2000                 __module_get(new->module);
2001 }
2002 EXPORT_SYMBOL(set_binfmt);
2003
2004 /*
2005  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2006  */
2007 void set_dumpable(struct mm_struct *mm, int value)
2008 {
2009         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2010                 return;
2011
2012         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2013 }
2014
2015 SYSCALL_DEFINE3(execve,
2016                 const char __user *, filename,
2017                 const char __user *const __user *, argv,
2018                 const char __user *const __user *, envp)
2019 {
2020         return do_execve(getname(filename), argv, envp);
2021 }
2022
2023 SYSCALL_DEFINE5(execveat,
2024                 int, fd, const char __user *, filename,
2025                 const char __user *const __user *, argv,
2026                 const char __user *const __user *, envp,
2027                 int, flags)
2028 {
2029         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2030
2031         return do_execveat(fd,
2032                            getname_flags(filename, lookup_flags, NULL),
2033                            argv, envp, flags);
2034 }
2035
2036 #ifdef CONFIG_COMPAT
2037 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2038         const compat_uptr_t __user *, argv,
2039         const compat_uptr_t __user *, envp)
2040 {
2041         return compat_do_execve(getname(filename), argv, envp);
2042 }
2043
2044 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2045                        const char __user *, filename,
2046                        const compat_uptr_t __user *, argv,
2047                        const compat_uptr_t __user *, envp,
2048                        int,  flags)
2049 {
2050         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2051
2052         return compat_do_execveat(fd,
2053                                   getname_flags(filename, lookup_flags, NULL),
2054                                   argv, envp, flags);
2055 }
2056 #endif