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         /* Move stack pages down in memory. */
 764         if (stack_shift) {
 765                 ret = shift_arg_pages(vma, stack_shift);
 766                 if (ret)
 767                         goto out_unlock;
 768         }
 769
 770         /* mprotect_fixup is overkill to remove the temporary stack flags */
 771         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 772
 773         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 774         stack_size = vma->vm_end - vma->vm_start;
 775         /*
 776          * Align this down to a page boundary as expand_stack
 777          * will align it up.
 778          */
 779         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 780 #ifdef CONFIG_STACK_GROWSUP
 781         if (stack_size + stack_expand > rlim_stack)
 782                 stack_base = vma->vm_start + rlim_stack;
 783         else
 784                 stack_base = vma->vm_end + stack_expand;
 785 #else
 786         if (stack_size + stack_expand > rlim_stack)
 787                 stack_base = vma->vm_end - rlim_stack;
 788         else
 789                 stack_base = vma->vm_start - stack_expand;
 790 #endif
 791         current->mm->start_stack = bprm->p;
 792         ret = expand_stack(vma, stack_base);
 793         if (ret)
 794                 ret = -EFAULT;
 795
 796 out_unlock:
 797         up_write(&mm->mmap_sem);
 798         return ret;
 799 }
 800 EXPORT_SYMBOL(setup_arg_pages);
 801
 802 #else
 803
 804 /*
 805  * Transfer the program arguments and environment from the holding pages
 806  * onto the stack. The provided stack pointer is adjusted accordingly.
 807  */
 808 int transfer_args_to_stack(struct linux_binprm *bprm,
 809                            unsigned long *sp_location)
 810 {
 811         unsigned long index, stop, sp;
 812         int ret = 0;
 813
 814         stop = bprm->p >> PAGE_SHIFT;
 815         sp = *sp_location;
 816
 817         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 818                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 819                 char *src = kmap(bprm->page[index]) + offset;
 820                 sp -= PAGE_SIZE - offset;
 821                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 822                         ret = -EFAULT;
 823                 kunmap(bprm->page[index]);
 824                 if (ret)
 825                         goto out;
 826         }
 827
 828         *sp_location = sp;
 829
 830 out:
 831         return ret;
 832 }
 833 EXPORT_SYMBOL(transfer_args_to_stack);
 834
 835 #endif /* CONFIG_MMU */
 836
 837 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 838 {
 839         struct file *file;
 840         int err;
 841         struct open_flags open_exec_flags = {
 842                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 843                 .acc_mode = MAY_EXEC,
 844                 .intent = LOOKUP_OPEN,
 845                 .lookup_flags = LOOKUP_FOLLOW,
 846         };
 847
 848         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 849                 return ERR_PTR(-EINVAL);
 850         if (flags & AT_SYMLINK_NOFOLLOW)
 851                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 852         if (flags & AT_EMPTY_PATH)
 853                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 854
 855         file = do_filp_open(fd, name, &open_exec_flags);
 856         if (IS_ERR(file))
 857                 goto out;
 858
 859         err = -EACCES;
 860         if (!S_ISREG(file_inode(file)->i_mode))
 861                 goto exit;
 862
 863         if (path_noexec(&file->f_path))
 864                 goto exit;
 865
 866         err = deny_write_access(file);
 867         if (err)
 868                 goto exit;
 869
 870         if (name->name[0] != '\0')
 871                 fsnotify_open(file);
 872
 873 out:
 874         return file;
 875
 876 exit:
 877         fput(file);
 878         return ERR_PTR(err);
 879 }
 880
 881 struct file *open_exec(const char *name)
 882 {
 883         struct filename *filename = getname_kernel(name);
 884         struct file *f = ERR_CAST(filename);
 885
 886         if (!IS_ERR(filename)) {
 887                 f = do_open_execat(AT_FDCWD, filename, 0);
 888                 putname(filename);
 889         }
 890         return f;
 891 }
 892 EXPORT_SYMBOL(open_exec);
 893
 894 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 895                      loff_t max_size, enum kernel_read_file_id id)
 896 {
 897         loff_t i_size, pos;
 898         ssize_t bytes = 0;
 899         int ret;
 900
 901         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 902                 return -EINVAL;
 903
 904         ret = deny_write_access(file);
 905         if (ret)
 906                 return ret;
 907
 908         ret = security_kernel_read_file(file, id);
 909         if (ret)
 910                 goto out;
 911
 912         i_size = i_size_read(file_inode(file));
 913         if (i_size <= 0) {
 914                 ret = -EINVAL;
 915                 goto out;
 916         }
 917         if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 918                 ret = -EFBIG;
 919                 goto out;
 920         }
 921
 922         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 923                 *buf = vmalloc(i_size);
 924         if (!*buf) {
 925                 ret = -ENOMEM;
 926                 goto out;
 927         }
 928
 929         pos = 0;
 930         while (pos < i_size) {
 931                 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 932                 if (bytes < 0) {
 933                         ret = bytes;
 934                         goto out_free;
 935                 }
 936
 937                 if (bytes == 0)
 938                         break;
 939         }
 940
 941         if (pos != i_size) {
 942                 ret = -EIO;
 943                 goto out_free;
 944         }
 945
 946         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 947         if (!ret)
 948                 *size = pos;
 949
 950 out_free:
 951         if (ret < 0) {
 952                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 953                         vfree(*buf);
 954                         *buf = NULL;
 955                 }
 956         }
 957
 958 out:
 959         allow_write_access(file);
 960         return ret;
 961 }
 962 EXPORT_SYMBOL_GPL(kernel_read_file);
 963
 964 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
 965                                loff_t max_size, enum kernel_read_file_id id)
 966 {
 967         struct file *file;
 968         int ret;
 969
 970         if (!path || !*path)
 971                 return -EINVAL;
 972
 973         file = filp_open(path, O_RDONLY, 0);
 974         if (IS_ERR(file))
 975                 return PTR_ERR(file);
 976
 977         ret = kernel_read_file(file, buf, size, max_size, id);
 978         fput(file);
 979         return ret;
 980 }
 981 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 982
 983 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 984                              enum kernel_read_file_id id)
 985 {
 986         struct fd f = fdget(fd);
 987         int ret = -EBADF;
 988
 989         if (!f.file)
 990                 goto out;
 991
 992         ret = kernel_read_file(f.file, buf, size, max_size, id);
 993 out:
 994         fdput(f);
 995         return ret;
 996 }
 997 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 998
 999 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1000 {
1001         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1002         if (res > 0)
1003                 flush_icache_range(addr, addr + len);
1004         return res;
1005 }
1006 EXPORT_SYMBOL(read_code);
1007
1008 static int exec_mmap(struct mm_struct *mm)
1009 {
1010         struct task_struct *tsk;
1011         struct mm_struct *old_mm, *active_mm;
1012
1013         /* Notify parent that we're no longer interested in the old VM */
1014         tsk = current;
1015         old_mm = current->mm;
1016         exec_mm_release(tsk, old_mm);
1017
1018         if (old_mm) {
1019                 sync_mm_rss(old_mm);
1020                 /*
1021                  * Make sure that if there is a core dump in progress
1022                  * for the old mm, we get out and die instead of going
1023                  * through with the exec.  We must hold mmap_sem around
1024                  * checking core_state and changing tsk->mm.
1025                  */
1026                 down_read(&old_mm->mmap_sem);
1027                 if (unlikely(old_mm->core_state)) {
1028                         up_read(&old_mm->mmap_sem);
1029                         return -EINTR;
1030                 }
1031         }
1032         task_lock(tsk);
1033         active_mm = tsk->active_mm;
1034         membarrier_exec_mmap(mm);
1035         tsk->mm = mm;
1036         tsk->active_mm = mm;
1037         activate_mm(active_mm, mm);
1038         tsk->mm->vmacache_seqnum = 0;
1039         vmacache_flush(tsk);
1040         task_unlock(tsk);
1041         if (old_mm) {
1042                 up_read(&old_mm->mmap_sem);
1043                 BUG_ON(active_mm != old_mm);
1044                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1045                 mm_update_next_owner(old_mm);
1046                 mmput(old_mm);
1047                 return 0;
1048         }
1049         mmdrop(active_mm);
1050         return 0;
1051 }
1052
1053 /*
1054  * This function makes sure the current process has its own signal table,
1055  * so that flush_signal_handlers can later reset the handlers without
1056  * disturbing other processes.  (Other processes might share the signal
1057  * table via the CLONE_SIGHAND option to clone().)
1058  */
1059 static int de_thread(struct task_struct *tsk)
1060 {
1061         struct signal_struct *sig = tsk->signal;
1062         struct sighand_struct *oldsighand = tsk->sighand;
1063         spinlock_t *lock = &oldsighand->siglock;
1064
1065         if (thread_group_empty(tsk))
1066                 goto no_thread_group;
1067
1068         /*
1069          * Kill all other threads in the thread group.
1070          */
1071         spin_lock_irq(lock);
1072         if (signal_group_exit(sig)) {
1073                 /*
1074                  * Another group action in progress, just
1075                  * return so that the signal is processed.
1076                  */
1077                 spin_unlock_irq(lock);
1078                 return -EAGAIN;
1079         }
1080
1081         sig->group_exit_task = tsk;
1082         sig->notify_count = zap_other_threads(tsk);
1083         if (!thread_group_leader(tsk))
1084                 sig->notify_count--;
1085
1086         while (sig->notify_count) {
1087                 __set_current_state(TASK_KILLABLE);
1088                 spin_unlock_irq(lock);
1089                 schedule();
1090                 if (__fatal_signal_pending(tsk))
1091                         goto killed;
1092                 spin_lock_irq(lock);
1093         }
1094         spin_unlock_irq(lock);
1095
1096         /*
1097          * At this point all other threads have exited, all we have to
1098          * do is to wait for the thread group leader to become inactive,
1099          * and to assume its PID:
1100          */
1101         if (!thread_group_leader(tsk)) {
1102                 struct task_struct *leader = tsk->group_leader;
1103
1104                 for (;;) {
1105                         cgroup_threadgroup_change_begin(tsk);
1106                         write_lock_irq(&tasklist_lock);
1107                         /*
1108                          * Do this under tasklist_lock to ensure that
1109                          * exit_notify() can't miss ->group_exit_task
1110                          */
1111                         sig->notify_count = -1;
1112                         if (likely(leader->exit_state))
1113                                 break;
1114                         __set_current_state(TASK_KILLABLE);
1115                         write_unlock_irq(&tasklist_lock);
1116                         cgroup_threadgroup_change_end(tsk);
1117                         schedule();
1118                         if (__fatal_signal_pending(tsk))
1119                                 goto killed;
1120                 }
1121
1122                 /*
1123                  * The only record we have of the real-time age of a
1124                  * process, regardless of execs it's done, is start_time.
1125                  * All the past CPU time is accumulated in signal_struct
1126                  * from sister threads now dead.  But in this non-leader
1127                  * exec, nothing survives from the original leader thread,
1128                  * whose birth marks the true age of this process now.
1129                  * When we take on its identity by switching to its PID, we
1130                  * also take its birthdate (always earlier than our own).
1131                  */
1132                 tsk->start_time = leader->start_time;
1133                 tsk->start_boottime = leader->start_boottime;
1134
1135                 BUG_ON(!same_thread_group(leader, tsk));
1136                 BUG_ON(has_group_leader_pid(tsk));
1137                 /*
1138                  * An exec() starts a new thread group with the
1139                  * TGID of the previous thread group. Rehash the
1140                  * two threads with a switched PID, and release
1141                  * the former thread group leader:
1142                  */
1143
1144                 /* Become a process group leader with the old leader's pid.
1145                  * The old leader becomes a thread of the this thread group.
1146                  * Note: The old leader also uses this pid until release_task
1147                  *       is called.  Odd but simple and correct.
1148                  */
1149                 tsk->pid = leader->pid;
1150                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1151                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1152                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1153                 transfer_pid(leader, tsk, PIDTYPE_SID);
1154
1155                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1156                 list_replace_init(&leader->sibling, &tsk->sibling);
1157
1158                 tsk->group_leader = tsk;
1159                 leader->group_leader = tsk;
1160
1161                 tsk->exit_signal = SIGCHLD;
1162                 leader->exit_signal = -1;
1163
1164                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1165                 leader->exit_state = EXIT_DEAD;
1166
1167                 /*
1168                  * We are going to release_task()->ptrace_unlink() silently,
1169                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1170                  * the tracer wont't block again waiting for this thread.
1171                  */
1172                 if (unlikely(leader->ptrace))
1173                         __wake_up_parent(leader, leader->parent);
1174                 write_unlock_irq(&tasklist_lock);
1175                 cgroup_threadgroup_change_end(tsk);
1176
1177                 release_task(leader);
1178         }
1179
1180         sig->group_exit_task = NULL;
1181         sig->notify_count = 0;
1182
1183 no_thread_group:
1184         /* we have changed execution domain */
1185         tsk->exit_signal = SIGCHLD;
1186
1187 #ifdef CONFIG_POSIX_TIMERS
1188         exit_itimers(sig);
1189         flush_itimer_signals();
1190 #endif
1191
1192         if (refcount_read(&oldsighand->count) != 1) {
1193                 struct sighand_struct *newsighand;
1194                 /*
1195                  * This ->sighand is shared with the CLONE_SIGHAND
1196                  * but not CLONE_THREAD task, switch to the new one.
1197                  */
1198                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1199                 if (!newsighand)
1200                         return -ENOMEM;
1201
1202                 refcount_set(&newsighand->count, 1);
1203                 memcpy(newsighand->action, oldsighand->action,
1204                        sizeof(newsighand->action));
1205
1206                 write_lock_irq(&tasklist_lock);
1207                 spin_lock(&oldsighand->siglock);
1208                 rcu_assign_pointer(tsk->sighand, newsighand);
1209                 spin_unlock(&oldsighand->siglock);
1210                 write_unlock_irq(&tasklist_lock);
1211
1212                 __cleanup_sighand(oldsighand);
1213         }
1214
1215         BUG_ON(!thread_group_leader(tsk));
1216         return 0;
1217
1218 killed:
1219         /* protects against exit_notify() and __exit_signal() */
1220         read_lock(&tasklist_lock);
1221         sig->group_exit_task = NULL;
1222         sig->notify_count = 0;
1223         read_unlock(&tasklist_lock);
1224         return -EAGAIN;
1225 }
1226
1227 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1228 {
1229         task_lock(tsk);
1230         strncpy(buf, tsk->comm, buf_size);
1231         task_unlock(tsk);
1232         return buf;
1233 }
1234 EXPORT_SYMBOL_GPL(__get_task_comm);
1235
1236 /*
1237  * These functions flushes out all traces of the currently running executable
1238  * so that a new one can be started
1239  */
1240
1241 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1242 {
1243         task_lock(tsk);
1244         trace_task_rename(tsk, buf);
1245         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1246         task_unlock(tsk);
1247         perf_event_comm(tsk, exec);
1248 }
1249
1250 /*
1251  * Calling this is the point of no return. None of the failures will be
1252  * seen by userspace since either the process is already taking a fatal
1253  * signal (via de_thread() or coredump), or will have SEGV raised
1254  * (after exec_mmap()) by search_binary_handlers (see below).
1255  */
1256 int flush_old_exec(struct linux_binprm * bprm)
1257 {
1258         int retval;
1259
1260         /*
1261          * Make sure we have a private signal table and that
1262          * we are unassociated from the previous thread group.
1263          */
1264         retval = de_thread(current);
1265         if (retval)
1266                 goto out;
1267
1268         /*
1269          * Must be called _before_ exec_mmap() as bprm->mm is
1270          * not visibile until then. This also enables the update
1271          * to be lockless.
1272          */
1273         set_mm_exe_file(bprm->mm, bprm->file);
1274
1275         /*
1276          * Release all of the old mmap stuff
1277          */
1278         acct_arg_size(bprm, 0);
1279         retval = exec_mmap(bprm->mm);
1280         if (retval)
1281                 goto out;
1282
1283         /*
1284          * After clearing bprm->mm (to mark that current is using the
1285          * prepared mm now), we have nothing left of the original
1286          * process. If anything from here on returns an error, the check
1287          * in search_binary_handler() will SEGV current.
1288          */
1289         bprm->mm = NULL;
1290
1291         set_fs(USER_DS);
1292         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1293                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1294         flush_thread();
1295         current->personality &= ~bprm->per_clear;
1296
1297         /*
1298          * We have to apply CLOEXEC before we change whether the process is
1299          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1300          * trying to access the should-be-closed file descriptors of a process
1301          * undergoing exec(2).
1302          */
1303         do_close_on_exec(current->files);
1304         return 0;
1305
1306 out:
1307         return retval;
1308 }
1309 EXPORT_SYMBOL(flush_old_exec);
1310
1311 void would_dump(struct linux_binprm *bprm, struct file *file)
1312 {
1313         struct inode *inode = file_inode(file);
1314         if (inode_permission(inode, MAY_READ) < 0) {
1315                 struct user_namespace *old, *user_ns;
1316                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1317
1318                 /* Ensure mm->user_ns contains the executable */
1319                 user_ns = old = bprm->mm->user_ns;
1320                 while ((user_ns != &init_user_ns) &&
1321                        !privileged_wrt_inode_uidgid(user_ns, inode))
1322                         user_ns = user_ns->parent;
1323
1324                 if (old != user_ns) {
1325                         bprm->mm->user_ns = get_user_ns(user_ns);
1326                         put_user_ns(old);
1327                 }
1328         }
1329 }
1330 EXPORT_SYMBOL(would_dump);
1331
1332 void setup_new_exec(struct linux_binprm * bprm)
1333 {
1334         /*
1335          * Once here, prepare_binrpm() will not be called any more, so
1336          * the final state of setuid/setgid/fscaps can be merged into the
1337          * secureexec flag.
1338          */
1339         bprm->secureexec |= bprm->cap_elevated;
1340
1341         if (bprm->secureexec) {
1342                 /* Make sure parent cannot signal privileged process. */
1343                 current->pdeath_signal = 0;
1344
1345                 /*
1346                  * For secureexec, reset the stack limit to sane default to
1347                  * avoid bad behavior from the prior rlimits. This has to
1348                  * happen before arch_pick_mmap_layout(), which examines
1349                  * RLIMIT_STACK, but after the point of no return to avoid
1350                  * needing to clean up the change on failure.
1351                  */
1352                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1353                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1354         }
1355
1356         arch_pick_mmap_layout(current->mm, &bprm->rlim_stack);
1357
1358         current->sas_ss_sp = current->sas_ss_size = 0;
1359
1360         /*
1361          * Figure out dumpability. Note that this checking only of current
1362          * is wrong, but userspace depends on it. This should be testing
1363          * bprm->secureexec instead.
1364          */
1365         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1366             !(uid_eq(current_euid(), current_uid()) &&
1367               gid_eq(current_egid(), current_gid())))
1368                 set_dumpable(current->mm, suid_dumpable);
1369         else
1370                 set_dumpable(current->mm, SUID_DUMP_USER);
1371
1372         arch_setup_new_exec();
1373         perf_event_exec();
1374         __set_task_comm(current, kbasename(bprm->filename), true);
1375
1376         /* Set the new mm task size. We have to do that late because it may
1377          * depend on TIF_32BIT which is only updated in flush_thread() on
1378          * some architectures like powerpc
1379          */
1380         current->mm->task_size = TASK_SIZE;
1381
1382         /* An exec changes our domain. We are no longer part of the thread
1383            group */
1384         current->self_exec_id++;
1385         flush_signal_handlers(current, 0);
1386 }
1387 EXPORT_SYMBOL(setup_new_exec);
1388
1389 /* Runs immediately before start_thread() takes over. */
1390 void finalize_exec(struct linux_binprm *bprm)
1391 {
1392         /* Store any stack rlimit changes before starting thread. */
1393         task_lock(current->group_leader);
1394         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1395         task_unlock(current->group_leader);
1396 }
1397 EXPORT_SYMBOL(finalize_exec);
1398
1399 /*
1400  * Prepare credentials and lock ->cred_guard_mutex.
1401  * install_exec_creds() commits the new creds and drops the lock.
1402  * Or, if exec fails before, free_bprm() should release ->cred and
1403  * and unlock.
1404  */
1405 static int prepare_bprm_creds(struct linux_binprm *bprm)
1406 {
1407         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1408                 return -ERESTARTNOINTR;
1409
1410         bprm->cred = prepare_exec_creds();
1411         if (likely(bprm->cred))
1412                 return 0;
1413
1414         mutex_unlock(&current->signal->cred_guard_mutex);
1415         return -ENOMEM;
1416 }
1417
1418 static void free_bprm(struct linux_binprm *bprm)
1419 {
1420         free_arg_pages(bprm);
1421         if (bprm->cred) {
1422                 mutex_unlock(&current->signal->cred_guard_mutex);
1423                 abort_creds(bprm->cred);
1424         }
1425         if (bprm->file) {
1426                 allow_write_access(bprm->file);
1427                 fput(bprm->file);
1428         }
1429         /* If a binfmt changed the interp, free it. */
1430         if (bprm->interp != bprm->filename)
1431                 kfree(bprm->interp);
1432         kfree(bprm);
1433 }
1434
1435 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1436 {
1437         /* If a binfmt changed the interp, free it first. */
1438         if (bprm->interp != bprm->filename)
1439                 kfree(bprm->interp);
1440         bprm->interp = kstrdup(interp, GFP_KERNEL);
1441         if (!bprm->interp)
1442                 return -ENOMEM;
1443         return 0;
1444 }
1445 EXPORT_SYMBOL(bprm_change_interp);
1446
1447 /*
1448  * install the new credentials for this executable
1449  */
1450 void install_exec_creds(struct linux_binprm *bprm)
1451 {
1452         security_bprm_committing_creds(bprm);
1453
1454         commit_creds(bprm->cred);
1455         bprm->cred = NULL;
1456
1457         /*
1458          * Disable monitoring for regular users
1459          * when executing setuid binaries. Must
1460          * wait until new credentials are committed
1461          * by commit_creds() above
1462          */
1463         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1464                 perf_event_exit_task(current);
1465         /*
1466          * cred_guard_mutex must be held at least to this point to prevent
1467          * ptrace_attach() from altering our determination of the task's
1468          * credentials; any time after this it may be unlocked.
1469          */
1470         security_bprm_committed_creds(bprm);
1471         mutex_unlock(&current->signal->cred_guard_mutex);
1472 }
1473 EXPORT_SYMBOL(install_exec_creds);
1474
1475 /*
1476  * determine how safe it is to execute the proposed program
1477  * - the caller must hold ->cred_guard_mutex to protect against
1478  *   PTRACE_ATTACH or seccomp thread-sync
1479  */
1480 static void check_unsafe_exec(struct linux_binprm *bprm)
1481 {
1482         struct task_struct *p = current, *t;
1483         unsigned n_fs;
1484
1485         if (p->ptrace)
1486                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1487
1488         /*
1489          * This isn't strictly necessary, but it makes it harder for LSMs to
1490          * mess up.
1491          */
1492         if (task_no_new_privs(current))
1493                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1494
1495         t = p;
1496         n_fs = 1;
1497         spin_lock(&p->fs->lock);
1498         rcu_read_lock();
1499         while_each_thread(p, t) {
1500                 if (t->fs == p->fs)
1501                         n_fs++;
1502         }
1503         rcu_read_unlock();
1504
1505         if (p->fs->users > n_fs)
1506                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1507         else
1508                 p->fs->in_exec = 1;
1509         spin_unlock(&p->fs->lock);
1510 }
1511
1512 static void bprm_fill_uid(struct linux_binprm *bprm)
1513 {
1514         struct inode *inode;
1515         unsigned int mode;
1516         kuid_t uid;
1517         kgid_t gid;
1518
1519         /*
1520          * Since this can be called multiple times (via prepare_binprm),
1521          * we must clear any previous work done when setting set[ug]id
1522          * bits from any earlier bprm->file uses (for example when run
1523          * first for a setuid script then again for its interpreter).
1524          */
1525         bprm->cred->euid = current_euid();
1526         bprm->cred->egid = current_egid();
1527
1528         if (!mnt_may_suid(bprm->file->f_path.mnt))
1529                 return;
1530
1531         if (task_no_new_privs(current))
1532                 return;
1533
1534         inode = bprm->file->f_path.dentry->d_inode;
1535         mode = READ_ONCE(inode->i_mode);
1536         if (!(mode & (S_ISUID|S_ISGID)))
1537                 return;
1538
1539         /* Be careful if suid/sgid is set */
1540         inode_lock(inode);
1541
1542         /* reload atomically mode/uid/gid now that lock held */
1543         mode = inode->i_mode;
1544         uid = inode->i_uid;
1545         gid = inode->i_gid;
1546         inode_unlock(inode);
1547
1548         /* We ignore suid/sgid if there are no mappings for them in the ns */
1549         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1550                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1551                 return;
1552
1553         if (mode & S_ISUID) {
1554                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1555                 bprm->cred->euid = uid;
1556         }
1557
1558         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1559                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1560                 bprm->cred->egid = gid;
1561         }
1562 }
1563
1564 /*
1565  * Fill the binprm structure from the inode.
1566  * Check permissions, then read the first BINPRM_BUF_SIZE bytes
1567  *
1568  * This may be called multiple times for binary chains (scripts for example).
1569  */
1570 int prepare_binprm(struct linux_binprm *bprm)
1571 {
1572         int retval;
1573         loff_t pos = 0;
1574
1575         bprm_fill_uid(bprm);
1576
1577         /* fill in binprm security blob */
1578         retval = security_bprm_set_creds(bprm);
1579         if (retval)
1580                 return retval;
1581         bprm->called_set_creds = 1;
1582
1583         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1584         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1585 }
1586
1587 EXPORT_SYMBOL(prepare_binprm);
1588
1589 /*
1590  * Arguments are '\0' separated strings found at the location bprm->p
1591  * points to; chop off the first by relocating brpm->p to right after
1592  * the first '\0' encountered.
1593  */
1594 int remove_arg_zero(struct linux_binprm *bprm)
1595 {
1596         int ret = 0;
1597         unsigned long offset;
1598         char *kaddr;
1599         struct page *page;
1600
1601         if (!bprm->argc)
1602                 return 0;
1603
1604         do {
1605                 offset = bprm->p & ~PAGE_MASK;
1606                 page = get_arg_page(bprm, bprm->p, 0);
1607                 if (!page) {
1608                         ret = -EFAULT;
1609                         goto out;
1610                 }
1611                 kaddr = kmap_atomic(page);
1612
1613                 for (; offset < PAGE_SIZE && kaddr[offset];
1614                                 offset++, bprm->p++)
1615                         ;
1616
1617                 kunmap_atomic(kaddr);
1618                 put_arg_page(page);
1619         } while (offset == PAGE_SIZE);
1620
1621         bprm->p++;
1622         bprm->argc--;
1623         ret = 0;
1624
1625 out:
1626         return ret;
1627 }
1628 EXPORT_SYMBOL(remove_arg_zero);
1629
1630 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1631 /*
1632  * cycle the list of binary formats handler, until one recognizes the image
1633  */
1634 int search_binary_handler(struct linux_binprm *bprm)
1635 {
1636         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1637         struct linux_binfmt *fmt;
1638         int retval;
1639
1640         /* This allows 4 levels of binfmt rewrites before failing hard. */
1641         if (bprm->recursion_depth > 5)
1642                 return -ELOOP;
1643
1644         retval = security_bprm_check(bprm);
1645         if (retval)
1646                 return retval;
1647
1648         retval = -ENOENT;
1649  retry:
1650         read_lock(&binfmt_lock);
1651         list_for_each_entry(fmt, &formats, lh) {
1652                 if (!try_module_get(fmt->module))
1653                         continue;
1654                 read_unlock(&binfmt_lock);
1655
1656                 bprm->recursion_depth++;
1657                 retval = fmt->load_binary(bprm);
1658                 bprm->recursion_depth--;
1659
1660                 read_lock(&binfmt_lock);
1661                 put_binfmt(fmt);
1662                 if (retval < 0 && !bprm->mm) {
1663                         /* we got to flush_old_exec() and failed after it */
1664                         read_unlock(&binfmt_lock);
1665                         force_sigsegv(SIGSEGV);
1666                         return retval;
1667                 }
1668                 if (retval != -ENOEXEC || !bprm->file) {
1669                         read_unlock(&binfmt_lock);
1670                         return retval;
1671                 }
1672         }
1673         read_unlock(&binfmt_lock);
1674
1675         if (need_retry) {
1676                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1677                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1678                         return retval;
1679                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1680                         return retval;
1681                 need_retry = false;
1682                 goto retry;
1683         }
1684
1685         return retval;
1686 }
1687 EXPORT_SYMBOL(search_binary_handler);
1688
1689 static int exec_binprm(struct linux_binprm *bprm)
1690 {
1691         pid_t old_pid, old_vpid;
1692         int ret;
1693
1694         /* Need to fetch pid before load_binary changes it */
1695         old_pid = current->pid;
1696         rcu_read_lock();
1697         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1698         rcu_read_unlock();
1699
1700         ret = search_binary_handler(bprm);
1701         if (ret >= 0) {
1702                 audit_bprm(bprm);
1703                 trace_sched_process_exec(current, old_pid, bprm);
1704                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1705                 proc_exec_connector(current);
1706         }
1707
1708         return ret;
1709 }
1710
1711 /*
1712  * sys_execve() executes a new program.
1713  */
1714 static int __do_execve_file(int fd, struct filename *filename,
1715                             struct user_arg_ptr argv,
1716                             struct user_arg_ptr envp,
1717                             int flags, struct file *file)
1718 {
1719         char *pathbuf = NULL;
1720         struct linux_binprm *bprm;
1721         struct files_struct *displaced;
1722         int retval;
1723
1724         if (IS_ERR(filename))
1725                 return PTR_ERR(filename);
1726
1727         /*
1728          * We move the actual failure in case of RLIMIT_NPROC excess from
1729          * set*uid() to execve() because too many poorly written programs
1730          * don't check setuid() return code.  Here we additionally recheck
1731          * whether NPROC limit is still exceeded.
1732          */
1733         if ((current->flags & PF_NPROC_EXCEEDED) &&
1734             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1735                 retval = -EAGAIN;
1736                 goto out_ret;
1737         }
1738
1739         /* We're below the limit (still or again), so we don't want to make
1740          * further execve() calls fail. */
1741         current->flags &= ~PF_NPROC_EXCEEDED;
1742
1743         retval = unshare_files(&displaced);
1744         if (retval)
1745                 goto out_ret;
1746
1747         retval = -ENOMEM;
1748         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1749         if (!bprm)
1750                 goto out_files;
1751
1752         retval = prepare_bprm_creds(bprm);
1753         if (retval)
1754                 goto out_free;
1755
1756         check_unsafe_exec(bprm);
1757         current->in_execve = 1;
1758
1759         if (!file)
1760                 file = do_open_execat(fd, filename, flags);
1761         retval = PTR_ERR(file);
1762         if (IS_ERR(file))
1763                 goto out_unmark;
1764
1765         sched_exec();
1766
1767         bprm->file = file;
1768         if (!filename) {
1769                 bprm->filename = "none";
1770         } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1771                 bprm->filename = filename->name;
1772         } else {
1773                 if (filename->name[0] == '\0')
1774                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1775                 else
1776                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1777                                             fd, filename->name);
1778                 if (!pathbuf) {
1779                         retval = -ENOMEM;
1780                         goto out_unmark;
1781                 }
1782                 /*
1783                  * Record that a name derived from an O_CLOEXEC fd will be
1784                  * inaccessible after exec. Relies on having exclusive access to
1785                  * current->files (due to unshare_files above).
1786                  */
1787                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1788                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1789                 bprm->filename = pathbuf;
1790         }
1791         bprm->interp = bprm->filename;
1792
1793         retval = bprm_mm_init(bprm);
1794         if (retval)
1795                 goto out_unmark;
1796
1797         retval = prepare_arg_pages(bprm, argv, envp);
1798         if (retval < 0)
1799                 goto out;
1800
1801         retval = prepare_binprm(bprm);
1802         if (retval < 0)
1803                 goto out;
1804
1805         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1806         if (retval < 0)
1807                 goto out;
1808
1809         bprm->exec = bprm->p;
1810         retval = copy_strings(bprm->envc, envp, bprm);
1811         if (retval < 0)
1812                 goto out;
1813
1814         retval = copy_strings(bprm->argc, argv, bprm);
1815         if (retval < 0)
1816                 goto out;
1817
1818         would_dump(bprm, bprm->file);
1819
1820         retval = exec_binprm(bprm);
1821         if (retval < 0)
1822                 goto out;
1823
1824         /* execve succeeded */
1825         current->fs->in_exec = 0;
1826         current->in_execve = 0;
1827         rseq_execve(current);
1828         acct_update_integrals(current);
1829         task_numa_free(current, false);
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         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1949                 return;
1950
1951         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
1952 }
1953
1954 SYSCALL_DEFINE3(execve,
1955                 const char __user *, filename,
1956                 const char __user *const __user *, argv,
1957                 const char __user *const __user *, envp)
1958 {
1959         return do_execve(getname(filename), argv, envp);
1960 }
1961
1962 SYSCALL_DEFINE5(execveat,
1963                 int, fd, const char __user *, filename,
1964                 const char __user *const __user *, argv,
1965                 const char __user *const __user *, envp,
1966                 int, flags)
1967 {
1968         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1969
1970         return do_execveat(fd,
1971                            getname_flags(filename, lookup_flags, NULL),
1972                            argv, envp, flags);
1973 }
1974
1975 #ifdef CONFIG_COMPAT
1976 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1977         const compat_uptr_t __user *, argv,
1978         const compat_uptr_t __user *, envp)
1979 {
1980         return compat_do_execve(getname(filename), argv, envp);
1981 }
1982
1983 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1984                        const char __user *, filename,
1985                        const compat_uptr_t __user *, argv,
1986                        const compat_uptr_t __user *, envp,
1987                        int,  flags)
1988 {
1989         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1990
1991         return compat_do_execveat(fd,
1992                                   getname_flags(filename, lookup_flags, NULL),
1993                                   argv, envp, flags);
1994 }
1995 #endif