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 static int bprm_creds_from_file(struct linux_binprm *bprm);
  76
  77 int suid_dumpable = 0;
  78
  79 static LIST_HEAD(formats);
  80 static DEFINE_RWLOCK(binfmt_lock);
  81
  82 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  83 {
  84         BUG_ON(!fmt);
  85         if (WARN_ON(!fmt->load_binary))
  86                 return;
  87         write_lock(&binfmt_lock);
  88         insert ? list_add(&fmt->lh, &formats) :
  89                  list_add_tail(&fmt->lh, &formats);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(__register_binfmt);
  94
  95 void unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         write_lock(&binfmt_lock);
  98         list_del(&fmt->lh);
  99         write_unlock(&binfmt_lock);
 100 }
 101
 102 EXPORT_SYMBOL(unregister_binfmt);
 103
 104 static inline void put_binfmt(struct linux_binfmt * fmt)
 105 {
 106         module_put(fmt->module);
 107 }
 108
 109 bool path_noexec(const struct path *path)
 110 {
 111         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 112                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 113 }
 114
 115 #ifdef CONFIG_USELIB
 116 /*
 117  * Note that a shared library must be both readable and executable due to
 118  * security reasons.
 119  *
 120  * Also note that we take the address to load from from the file itself.
 121  */
 122 SYSCALL_DEFINE1(uselib, const char __user *, library)
 123 {
 124         struct linux_binfmt *fmt;
 125         struct file *file;
 126         struct filename *tmp = getname(library);
 127         int error = PTR_ERR(tmp);
 128         static const struct open_flags uselib_flags = {
 129                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 130                 .acc_mode = MAY_READ | MAY_EXEC,
 131                 .intent = LOOKUP_OPEN,
 132                 .lookup_flags = LOOKUP_FOLLOW,
 133         };
 134
 135         if (IS_ERR(tmp))
 136                 goto out;
 137
 138         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 139         putname(tmp);
 140         error = PTR_ERR(file);
 141         if (IS_ERR(file))
 142                 goto out;
 143
 144         error = -EINVAL;
 145         if (!S_ISREG(file_inode(file)->i_mode))
 146                 goto exit;
 147
 148         error = -EACCES;
 149         if (path_noexec(&file->f_path))
 150                 goto exit;
 151
 152         fsnotify_open(file);
 153
 154         error = -ENOEXEC;
 155
 156         read_lock(&binfmt_lock);
 157         list_for_each_entry(fmt, &formats, lh) {
 158                 if (!fmt->load_shlib)
 159                         continue;
 160                 if (!try_module_get(fmt->module))
 161                         continue;
 162                 read_unlock(&binfmt_lock);
 163                 error = fmt->load_shlib(file);
 164                 read_lock(&binfmt_lock);
 165                 put_binfmt(fmt);
 166                 if (error != -ENOEXEC)
 167                         break;
 168         }
 169         read_unlock(&binfmt_lock);
 170 exit:
 171         fput(file);
 172 out:
 173         return error;
 174 }
 175 #endif /* #ifdef CONFIG_USELIB */
 176
 177 #ifdef CONFIG_MMU
 178 /*
 179  * The nascent bprm->mm is not visible until exec_mmap() but it can
 180  * use a lot of memory, account these pages in current->mm temporary
 181  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 182  * change the counter back via acct_arg_size(0).
 183  */
 184 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 185 {
 186         struct mm_struct *mm = current->mm;
 187         long diff = (long)(pages - bprm->vma_pages);
 188
 189         if (!mm || !diff)
 190                 return;
 191
 192         bprm->vma_pages = pages;
 193         add_mm_counter(mm, MM_ANONPAGES, diff);
 194 }
 195
 196 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 197                 int write)
 198 {
 199         struct page *page;
 200         int ret;
 201         unsigned int gup_flags = FOLL_FORCE;
 202
 203 #ifdef CONFIG_STACK_GROWSUP
 204         if (write) {
 205                 ret = expand_downwards(bprm->vma, pos);
 206                 if (ret < 0)
 207                         return NULL;
 208         }
 209 #endif
 210
 211         if (write)
 212                 gup_flags |= FOLL_WRITE;
 213
 214         /*
 215          * We are doing an exec().  'current' is the process
 216          * doing the exec and bprm->mm is the new process's mm.
 217          */
 218         ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 219                         &page, NULL, NULL);
 220         if (ret <= 0)
 221                 return NULL;
 222
 223         if (write)
 224                 acct_arg_size(bprm, vma_pages(bprm->vma));
 225
 226         return page;
 227 }
 228
 229 static void put_arg_page(struct page *page)
 230 {
 231         put_page(page);
 232 }
 233
 234 static void free_arg_pages(struct linux_binprm *bprm)
 235 {
 236 }
 237
 238 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 239                 struct page *page)
 240 {
 241         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 242 }
 243
 244 static int __bprm_mm_init(struct linux_binprm *bprm)
 245 {
 246         int err;
 247         struct vm_area_struct *vma = NULL;
 248         struct mm_struct *mm = bprm->mm;
 249
 250         bprm->vma = vma = vm_area_alloc(mm);
 251         if (!vma)
 252                 return -ENOMEM;
 253         vma_set_anonymous(vma);
 254
 255         if (down_write_killable(&mm->mmap_sem)) {
 256                 err = -EINTR;
 257                 goto err_free;
 258         }
 259
 260         /*
 261          * Place the stack at the largest stack address the architecture
 262          * supports. Later, we'll move this to an appropriate place. We don't
 263          * use STACK_TOP because that can depend on attributes which aren't
 264          * configured yet.
 265          */
 266         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 267         vma->vm_end = STACK_TOP_MAX;
 268         vma->vm_start = vma->vm_end - PAGE_SIZE;
 269         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 270         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 271
 272         err = insert_vm_struct(mm, vma);
 273         if (err)
 274                 goto err;
 275
 276         mm->stack_vm = mm->total_vm = 1;
 277         up_write(&mm->mmap_sem);
 278         bprm->p = vma->vm_end - sizeof(void *);
 279         return 0;
 280 err:
 281         up_write(&mm->mmap_sem);
 282 err_free:
 283         bprm->vma = NULL;
 284         vm_area_free(vma);
 285         return err;
 286 }
 287
 288 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 289 {
 290         return len <= MAX_ARG_STRLEN;
 291 }
 292
 293 #else
 294
 295 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 296 {
 297 }
 298
 299 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 300                 int write)
 301 {
 302         struct page *page;
 303
 304         page = bprm->page[pos / PAGE_SIZE];
 305         if (!page && write) {
 306                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 307                 if (!page)
 308                         return NULL;
 309                 bprm->page[pos / PAGE_SIZE] = page;
 310         }
 311
 312         return page;
 313 }
 314
 315 static void put_arg_page(struct page *page)
 316 {
 317 }
 318
 319 static void free_arg_page(struct linux_binprm *bprm, int i)
 320 {
 321         if (bprm->page[i]) {
 322                 __free_page(bprm->page[i]);
 323                 bprm->page[i] = NULL;
 324         }
 325 }
 326
 327 static void free_arg_pages(struct linux_binprm *bprm)
 328 {
 329         int i;
 330
 331         for (i = 0; i < MAX_ARG_PAGES; i++)
 332                 free_arg_page(bprm, i);
 333 }
 334
 335 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 336                 struct page *page)
 337 {
 338 }
 339
 340 static int __bprm_mm_init(struct linux_binprm *bprm)
 341 {
 342         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 343         return 0;
 344 }
 345
 346 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 347 {
 348         return len <= bprm->p;
 349 }
 350
 351 #endif /* CONFIG_MMU */
 352
 353 /*
 354  * Create a new mm_struct and populate it with a temporary stack
 355  * vm_area_struct.  We don't have enough context at this point to set the stack
 356  * flags, permissions, and offset, so we use temporary values.  We'll update
 357  * them later in setup_arg_pages().
 358  */
 359 static int bprm_mm_init(struct linux_binprm *bprm)
 360 {
 361         int err;
 362         struct mm_struct *mm = NULL;
 363
 364         bprm->mm = mm = mm_alloc();
 365         err = -ENOMEM;
 366         if (!mm)
 367                 goto err;
 368
 369         /* Save current stack limit for all calculations made during exec. */
 370         task_lock(current->group_leader);
 371         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 372         task_unlock(current->group_leader);
 373
 374         err = __bprm_mm_init(bprm);
 375         if (err)
 376                 goto err;
 377
 378         return 0;
 379
 380 err:
 381         if (mm) {
 382                 bprm->mm = NULL;
 383                 mmdrop(mm);
 384         }
 385
 386         return err;
 387 }
 388
 389 struct user_arg_ptr {
 390 #ifdef CONFIG_COMPAT
 391         bool is_compat;
 392 #endif
 393         union {
 394                 const char __user *const __user *native;
 395 #ifdef CONFIG_COMPAT
 396                 const compat_uptr_t __user *compat;
 397 #endif
 398         } ptr;
 399 };
 400
 401 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 402 {
 403         const char __user *native;
 404
 405 #ifdef CONFIG_COMPAT
 406         if (unlikely(argv.is_compat)) {
 407                 compat_uptr_t compat;
 408
 409                 if (get_user(compat, argv.ptr.compat + nr))
 410                         return ERR_PTR(-EFAULT);
 411
 412                 return compat_ptr(compat);
 413         }
 414 #endif
 415
 416         if (get_user(native, argv.ptr.native + nr))
 417                 return ERR_PTR(-EFAULT);
 418
 419         return native;
 420 }
 421
 422 /*
 423  * count() counts the number of strings in array ARGV.
 424  */
 425 static int count(struct user_arg_ptr argv, int max)
 426 {
 427         int i = 0;
 428
 429         if (argv.ptr.native != NULL) {
 430                 for (;;) {
 431                         const char __user *p = get_user_arg_ptr(argv, i);
 432
 433                         if (!p)
 434                                 break;
 435
 436                         if (IS_ERR(p))
 437                                 return -EFAULT;
 438
 439                         if (i >= max)
 440                                 return -E2BIG;
 441                         ++i;
 442
 443                         if (fatal_signal_pending(current))
 444                                 return -ERESTARTNOHAND;
 445                         cond_resched();
 446                 }
 447         }
 448         return i;
 449 }
 450
 451 static int prepare_arg_pages(struct linux_binprm *bprm,
 452                         struct user_arg_ptr argv, struct user_arg_ptr envp)
 453 {
 454         unsigned long limit, ptr_size;
 455
 456         bprm->argc = count(argv, MAX_ARG_STRINGS);
 457         if (bprm->argc < 0)
 458                 return bprm->argc;
 459
 460         bprm->envc = count(envp, MAX_ARG_STRINGS);
 461         if (bprm->envc < 0)
 462                 return bprm->envc;
 463
 464         /*
 465          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 466          * (whichever is smaller) for the argv+env strings.
 467          * This ensures that:
 468          *  - the remaining binfmt code will not run out of stack space,
 469          *  - the program will have a reasonable amount of stack left
 470          *    to work from.
 471          */
 472         limit = _STK_LIM / 4 * 3;
 473         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 474         /*
 475          * We've historically supported up to 32 pages (ARG_MAX)
 476          * of argument strings even with small stacks
 477          */
 478         limit = max_t(unsigned long, limit, ARG_MAX);
 479         /*
 480          * We must account for the size of all the argv and envp pointers to
 481          * the argv and envp strings, since they will also take up space in
 482          * the stack. They aren't stored until much later when we can't
 483          * signal to the parent that the child has run out of stack space.
 484          * Instead, calculate it here so it's possible to fail gracefully.
 485          */
 486         ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 487         if (limit <= ptr_size)
 488                 return -E2BIG;
 489         limit -= ptr_size;
 490
 491         bprm->argmin = bprm->p - limit;
 492         return 0;
 493 }
 494
 495 /*
 496  * 'copy_strings()' copies argument/environment strings from the old
 497  * processes's memory to the new process's stack.  The call to get_user_pages()
 498  * ensures the destination page is created and not swapped out.
 499  */
 500 static int copy_strings(int argc, struct user_arg_ptr argv,
 501                         struct linux_binprm *bprm)
 502 {
 503         struct page *kmapped_page = NULL;
 504         char *kaddr = NULL;
 505         unsigned long kpos = 0;
 506         int ret;
 507
 508         while (argc-- > 0) {
 509                 const char __user *str;
 510                 int len;
 511                 unsigned long pos;
 512
 513                 ret = -EFAULT;
 514                 str = get_user_arg_ptr(argv, argc);
 515                 if (IS_ERR(str))
 516                         goto out;
 517
 518                 len = strnlen_user(str, MAX_ARG_STRLEN);
 519                 if (!len)
 520                         goto out;
 521
 522                 ret = -E2BIG;
 523                 if (!valid_arg_len(bprm, len))
 524                         goto out;
 525
 526                 /* We're going to work our way backwords. */
 527                 pos = bprm->p;
 528                 str += len;
 529                 bprm->p -= len;
 530 #ifdef CONFIG_MMU
 531                 if (bprm->p < bprm->argmin)
 532                         goto out;
 533 #endif
 534
 535                 while (len > 0) {
 536                         int offset, bytes_to_copy;
 537
 538                         if (fatal_signal_pending(current)) {
 539                                 ret = -ERESTARTNOHAND;
 540                                 goto out;
 541                         }
 542                         cond_resched();
 543
 544                         offset = pos % PAGE_SIZE;
 545                         if (offset == 0)
 546                                 offset = PAGE_SIZE;
 547
 548                         bytes_to_copy = offset;
 549                         if (bytes_to_copy > len)
 550                                 bytes_to_copy = len;
 551
 552                         offset -= bytes_to_copy;
 553                         pos -= bytes_to_copy;
 554                         str -= bytes_to_copy;
 555                         len -= bytes_to_copy;
 556
 557                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 558                                 struct page *page;
 559
 560                                 page = get_arg_page(bprm, pos, 1);
 561                                 if (!page) {
 562                                         ret = -E2BIG;
 563                                         goto out;
 564                                 }
 565
 566                                 if (kmapped_page) {
 567                                         flush_kernel_dcache_page(kmapped_page);
 568                                         kunmap(kmapped_page);
 569                                         put_arg_page(kmapped_page);
 570                                 }
 571                                 kmapped_page = page;
 572                                 kaddr = kmap(kmapped_page);
 573                                 kpos = pos & PAGE_MASK;
 574                                 flush_arg_page(bprm, kpos, kmapped_page);
 575                         }
 576                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 577                                 ret = -EFAULT;
 578                                 goto out;
 579                         }
 580                 }
 581         }
 582         ret = 0;
 583 out:
 584         if (kmapped_page) {
 585                 flush_kernel_dcache_page(kmapped_page);
 586                 kunmap(kmapped_page);
 587                 put_arg_page(kmapped_page);
 588         }
 589         return ret;
 590 }
 591
 592 /*
 593  * Copy and argument/environment string from the kernel to the processes stack.
 594  */
 595 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 596 {
 597         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 598         unsigned long pos = bprm->p;
 599
 600         if (len == 0)
 601                 return -EFAULT;
 602         if (!valid_arg_len(bprm, len))
 603                 return -E2BIG;
 604
 605         /* We're going to work our way backwards. */
 606         arg += len;
 607         bprm->p -= len;
 608         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 609                 return -E2BIG;
 610
 611         while (len > 0) {
 612                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 613                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 614                 struct page *page;
 615                 char *kaddr;
 616
 617                 pos -= bytes_to_copy;
 618                 arg -= bytes_to_copy;
 619                 len -= bytes_to_copy;
 620
 621                 page = get_arg_page(bprm, pos, 1);
 622                 if (!page)
 623                         return -E2BIG;
 624                 kaddr = kmap_atomic(page);
 625                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 626                 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
 627                 flush_kernel_dcache_page(page);
 628                 kunmap_atomic(kaddr);
 629                 put_arg_page(page);
 630         }
 631
 632         return 0;
 633 }
 634 EXPORT_SYMBOL(copy_string_kernel);
 635
 636 #ifdef CONFIG_MMU
 637
 638 /*
 639  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 640  * the binfmt code determines where the new stack should reside, we shift it to
 641  * its final location.  The process proceeds as follows:
 642  *
 643  * 1) Use shift to calculate the new vma endpoints.
 644  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 645  *    arguments passed to subsequent functions are consistent.
 646  * 3) Move vma's page tables to the new range.
 647  * 4) Free up any cleared pgd range.
 648  * 5) Shrink the vma to cover only the new range.
 649  */
 650 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 651 {
 652         struct mm_struct *mm = vma->vm_mm;
 653         unsigned long old_start = vma->vm_start;
 654         unsigned long old_end = vma->vm_end;
 655         unsigned long length = old_end - old_start;
 656         unsigned long new_start = old_start - shift;
 657         unsigned long new_end = old_end - shift;
 658         struct mmu_gather tlb;
 659
 660         BUG_ON(new_start > new_end);
 661
 662         /*
 663          * ensure there are no vmas between where we want to go
 664          * and where we are
 665          */
 666         if (vma != find_vma(mm, new_start))
 667                 return -EFAULT;
 668
 669         /*
 670          * cover the whole range: [new_start, old_end)
 671          */
 672         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 673                 return -ENOMEM;
 674
 675         /*
 676          * move the page tables downwards, on failure we rely on
 677          * process cleanup to remove whatever mess we made.
 678          */
 679         if (length != move_page_tables(vma, old_start,
 680                                        vma, new_start, length, false))
 681                 return -ENOMEM;
 682
 683         lru_add_drain();
 684         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 685         if (new_end > old_start) {
 686                 /*
 687                  * when the old and new regions overlap clear from new_end.
 688                  */
 689                 free_pgd_range(&tlb, new_end, old_end, new_end,
 690                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 691         } else {
 692                 /*
 693                  * otherwise, clean from old_start; this is done to not touch
 694                  * the address space in [new_end, old_start) some architectures
 695                  * have constraints on va-space that make this illegal (IA64) -
 696                  * for the others its just a little faster.
 697                  */
 698                 free_pgd_range(&tlb, old_start, old_end, new_end,
 699                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 700         }
 701         tlb_finish_mmu(&tlb, old_start, old_end);
 702
 703         /*
 704          * Shrink the vma to just the new range.  Always succeeds.
 705          */
 706         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 707
 708         return 0;
 709 }
 710
 711 /*
 712  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 713  * the stack is optionally relocated, and some extra space is added.
 714  */
 715 int setup_arg_pages(struct linux_binprm *bprm,
 716                     unsigned long stack_top,
 717                     int executable_stack)
 718 {
 719         unsigned long ret;
 720         unsigned long stack_shift;
 721         struct mm_struct *mm = current->mm;
 722         struct vm_area_struct *vma = bprm->vma;
 723         struct vm_area_struct *prev = NULL;
 724         unsigned long vm_flags;
 725         unsigned long stack_base;
 726         unsigned long stack_size;
 727         unsigned long stack_expand;
 728         unsigned long rlim_stack;
 729
 730 #ifdef CONFIG_STACK_GROWSUP
 731         /* Limit stack size */
 732         stack_base = bprm->rlim_stack.rlim_max;
 733         if (stack_base > STACK_SIZE_MAX)
 734                 stack_base = STACK_SIZE_MAX;
 735
 736         /* Add space for stack randomization. */
 737         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 738
 739         /* Make sure we didn't let the argument array grow too large. */
 740         if (vma->vm_end - vma->vm_start > stack_base)
 741                 return -ENOMEM;
 742
 743         stack_base = PAGE_ALIGN(stack_top - stack_base);
 744
 745         stack_shift = vma->vm_start - stack_base;
 746         mm->arg_start = bprm->p - stack_shift;
 747         bprm->p = vma->vm_end - stack_shift;
 748 #else
 749         stack_top = arch_align_stack(stack_top);
 750         stack_top = PAGE_ALIGN(stack_top);
 751
 752         if (unlikely(stack_top < mmap_min_addr) ||
 753             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 754                 return -ENOMEM;
 755
 756         stack_shift = vma->vm_end - stack_top;
 757
 758         bprm->p -= stack_shift;
 759         mm->arg_start = bprm->p;
 760 #endif
 761
 762         if (bprm->loader)
 763                 bprm->loader -= stack_shift;
 764         bprm->exec -= stack_shift;
 765
 766         if (down_write_killable(&mm->mmap_sem))
 767                 return -EINTR;
 768
 769         vm_flags = VM_STACK_FLAGS;
 770
 771         /*
 772          * Adjust stack execute permissions; explicitly enable for
 773          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 774          * (arch default) otherwise.
 775          */
 776         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 777                 vm_flags |= VM_EXEC;
 778         else if (executable_stack == EXSTACK_DISABLE_X)
 779                 vm_flags &= ~VM_EXEC;
 780         vm_flags |= mm->def_flags;
 781         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 782
 783         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 784                         vm_flags);
 785         if (ret)
 786                 goto out_unlock;
 787         BUG_ON(prev != vma);
 788
 789         if (unlikely(vm_flags & VM_EXEC)) {
 790                 pr_warn_once("process '%pD4' started with executable stack\n",
 791                              bprm->file);
 792         }
 793
 794         /* Move stack pages down in memory. */
 795         if (stack_shift) {
 796                 ret = shift_arg_pages(vma, stack_shift);
 797                 if (ret)
 798                         goto out_unlock;
 799         }
 800
 801         /* mprotect_fixup is overkill to remove the temporary stack flags */
 802         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 803
 804         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 805         stack_size = vma->vm_end - vma->vm_start;
 806         /*
 807          * Align this down to a page boundary as expand_stack
 808          * will align it up.
 809          */
 810         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 811 #ifdef CONFIG_STACK_GROWSUP
 812         if (stack_size + stack_expand > rlim_stack)
 813                 stack_base = vma->vm_start + rlim_stack;
 814         else
 815                 stack_base = vma->vm_end + stack_expand;
 816 #else
 817         if (stack_size + stack_expand > rlim_stack)
 818                 stack_base = vma->vm_end - rlim_stack;
 819         else
 820                 stack_base = vma->vm_start - stack_expand;
 821 #endif
 822         current->mm->start_stack = bprm->p;
 823         ret = expand_stack(vma, stack_base);
 824         if (ret)
 825                 ret = -EFAULT;
 826
 827 out_unlock:
 828         up_write(&mm->mmap_sem);
 829         return ret;
 830 }
 831 EXPORT_SYMBOL(setup_arg_pages);
 832
 833 #else
 834
 835 /*
 836  * Transfer the program arguments and environment from the holding pages
 837  * onto the stack. The provided stack pointer is adjusted accordingly.
 838  */
 839 int transfer_args_to_stack(struct linux_binprm *bprm,
 840                            unsigned long *sp_location)
 841 {
 842         unsigned long index, stop, sp;
 843         int ret = 0;
 844
 845         stop = bprm->p >> PAGE_SHIFT;
 846         sp = *sp_location;
 847
 848         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 849                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 850                 char *src = kmap(bprm->page[index]) + offset;
 851                 sp -= PAGE_SIZE - offset;
 852                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 853                         ret = -EFAULT;
 854                 kunmap(bprm->page[index]);
 855                 if (ret)
 856                         goto out;
 857         }
 858
 859         *sp_location = sp;
 860
 861 out:
 862         return ret;
 863 }
 864 EXPORT_SYMBOL(transfer_args_to_stack);
 865
 866 #endif /* CONFIG_MMU */
 867
 868 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 869 {
 870         struct file *file;
 871         int err;
 872         struct open_flags open_exec_flags = {
 873                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 874                 .acc_mode = MAY_EXEC,
 875                 .intent = LOOKUP_OPEN,
 876                 .lookup_flags = LOOKUP_FOLLOW,
 877         };
 878
 879         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 880                 return ERR_PTR(-EINVAL);
 881         if (flags & AT_SYMLINK_NOFOLLOW)
 882                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 883         if (flags & AT_EMPTY_PATH)
 884                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 885
 886         file = do_filp_open(fd, name, &open_exec_flags);
 887         if (IS_ERR(file))
 888                 goto out;
 889
 890         err = -EACCES;
 891         if (!S_ISREG(file_inode(file)->i_mode))
 892                 goto exit;
 893
 894         if (path_noexec(&file->f_path))
 895                 goto exit;
 896
 897         err = deny_write_access(file);
 898         if (err)
 899                 goto exit;
 900
 901         if (name->name[0] != '\0')
 902                 fsnotify_open(file);
 903
 904 out:
 905         return file;
 906
 907 exit:
 908         fput(file);
 909         return ERR_PTR(err);
 910 }
 911
 912 struct file *open_exec(const char *name)
 913 {
 914         struct filename *filename = getname_kernel(name);
 915         struct file *f = ERR_CAST(filename);
 916
 917         if (!IS_ERR(filename)) {
 918                 f = do_open_execat(AT_FDCWD, filename, 0);
 919                 putname(filename);
 920         }
 921         return f;
 922 }
 923 EXPORT_SYMBOL(open_exec);
 924
 925 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 926                      loff_t max_size, enum kernel_read_file_id id)
 927 {
 928         loff_t i_size, pos;
 929         ssize_t bytes = 0;
 930         int ret;
 931
 932         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 933                 return -EINVAL;
 934
 935         ret = deny_write_access(file);
 936         if (ret)
 937                 return ret;
 938
 939         ret = security_kernel_read_file(file, id);
 940         if (ret)
 941                 goto out;
 942
 943         i_size = i_size_read(file_inode(file));
 944         if (i_size <= 0) {
 945                 ret = -EINVAL;
 946                 goto out;
 947         }
 948         if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 949                 ret = -EFBIG;
 950                 goto out;
 951         }
 952
 953         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 954                 *buf = vmalloc(i_size);
 955         if (!*buf) {
 956                 ret = -ENOMEM;
 957                 goto out;
 958         }
 959
 960         pos = 0;
 961         while (pos < i_size) {
 962                 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 963                 if (bytes < 0) {
 964                         ret = bytes;
 965                         goto out_free;
 966                 }
 967
 968                 if (bytes == 0)
 969                         break;
 970         }
 971
 972         if (pos != i_size) {
 973                 ret = -EIO;
 974                 goto out_free;
 975         }
 976
 977         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 978         if (!ret)
 979                 *size = pos;
 980
 981 out_free:
 982         if (ret < 0) {
 983                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 984                         vfree(*buf);
 985                         *buf = NULL;
 986                 }
 987         }
 988
 989 out:
 990         allow_write_access(file);
 991         return ret;
 992 }
 993 EXPORT_SYMBOL_GPL(kernel_read_file);
 994
 995 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
 996                                loff_t max_size, enum kernel_read_file_id id)
 997 {
 998         struct file *file;
 999         int ret;
1000
1001         if (!path || !*path)
1002                 return -EINVAL;
1003
1004         file = filp_open(path, O_RDONLY, 0);
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);
1013
1014 int kernel_read_file_from_path_initns(const char *path, void **buf,
1015                                       loff_t *size, loff_t max_size,
1016                                       enum kernel_read_file_id id)
1017 {
1018         struct file *file;
1019         struct path root;
1020         int ret;
1021
1022         if (!path || !*path)
1023                 return -EINVAL;
1024
1025         task_lock(&init_task);
1026         get_fs_root(init_task.fs, &root);
1027         task_unlock(&init_task);
1028
1029         file = file_open_root(root.dentry, root.mnt, path, O_RDONLY, 0);
1030         path_put(&root);
1031         if (IS_ERR(file))
1032                 return PTR_ERR(file);
1033
1034         ret = kernel_read_file(file, buf, size, max_size, id);
1035         fput(file);
1036         return ret;
1037 }
1038 EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns);
1039
1040 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
1041                              enum kernel_read_file_id id)
1042 {
1043         struct fd f = fdget(fd);
1044         int ret = -EBADF;
1045
1046         if (!f.file)
1047                 goto out;
1048
1049         ret = kernel_read_file(f.file, buf, size, max_size, id);
1050 out:
1051         fdput(f);
1052         return ret;
1053 }
1054 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1055
1056 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1057 {
1058         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1059         if (res > 0)
1060                 flush_icache_range(addr, addr + len);
1061         return res;
1062 }
1063 EXPORT_SYMBOL(read_code);
1064
1065 /*
1066  * Maps the mm_struct mm into the current task struct.
1067  * On success, this function returns with the mutex
1068  * exec_update_mutex locked.
1069  */
1070 static int exec_mmap(struct mm_struct *mm)
1071 {
1072         struct task_struct *tsk;
1073         struct mm_struct *old_mm, *active_mm;
1074         int ret;
1075
1076         /* Notify parent that we're no longer interested in the old VM */
1077         tsk = current;
1078         old_mm = current->mm;
1079         exec_mm_release(tsk, old_mm);
1080         if (old_mm)
1081                 sync_mm_rss(old_mm);
1082
1083         ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
1084         if (ret)
1085                 return ret;
1086
1087         if (old_mm) {
1088                 /*
1089                  * Make sure that if there is a core dump in progress
1090                  * for the old mm, we get out and die instead of going
1091                  * through with the exec.  We must hold mmap_sem around
1092                  * checking core_state and changing tsk->mm.
1093                  */
1094                 down_read(&old_mm->mmap_sem);
1095                 if (unlikely(old_mm->core_state)) {
1096                         up_read(&old_mm->mmap_sem);
1097                         mutex_unlock(&tsk->signal->exec_update_mutex);
1098                         return -EINTR;
1099                 }
1100         }
1101
1102         task_lock(tsk);
1103         active_mm = tsk->active_mm;
1104         membarrier_exec_mmap(mm);
1105         tsk->mm = mm;
1106         tsk->active_mm = mm;
1107         activate_mm(active_mm, mm);
1108         tsk->mm->vmacache_seqnum = 0;
1109         vmacache_flush(tsk);
1110         task_unlock(tsk);
1111         if (old_mm) {
1112                 up_read(&old_mm->mmap_sem);
1113                 BUG_ON(active_mm != old_mm);
1114                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1115                 mm_update_next_owner(old_mm);
1116                 mmput(old_mm);
1117                 return 0;
1118         }
1119         mmdrop(active_mm);
1120         return 0;
1121 }
1122
1123 static int de_thread(struct task_struct *tsk)
1124 {
1125         struct signal_struct *sig = tsk->signal;
1126         struct sighand_struct *oldsighand = tsk->sighand;
1127         spinlock_t *lock = &oldsighand->siglock;
1128
1129         if (thread_group_empty(tsk))
1130                 goto no_thread_group;
1131
1132         /*
1133          * Kill all other threads in the thread group.
1134          */
1135         spin_lock_irq(lock);
1136         if (signal_group_exit(sig)) {
1137                 /*
1138                  * Another group action in progress, just
1139                  * return so that the signal is processed.
1140                  */
1141                 spin_unlock_irq(lock);
1142                 return -EAGAIN;
1143         }
1144
1145         sig->group_exit_task = tsk;
1146         sig->notify_count = zap_other_threads(tsk);
1147         if (!thread_group_leader(tsk))
1148                 sig->notify_count--;
1149
1150         while (sig->notify_count) {
1151                 __set_current_state(TASK_KILLABLE);
1152                 spin_unlock_irq(lock);
1153                 schedule();
1154                 if (__fatal_signal_pending(tsk))
1155                         goto killed;
1156                 spin_lock_irq(lock);
1157         }
1158         spin_unlock_irq(lock);
1159
1160         /*
1161          * At this point all other threads have exited, all we have to
1162          * do is to wait for the thread group leader to become inactive,
1163          * and to assume its PID:
1164          */
1165         if (!thread_group_leader(tsk)) {
1166                 struct task_struct *leader = tsk->group_leader;
1167
1168                 for (;;) {
1169                         cgroup_threadgroup_change_begin(tsk);
1170                         write_lock_irq(&tasklist_lock);
1171                         /*
1172                          * Do this under tasklist_lock to ensure that
1173                          * exit_notify() can't miss ->group_exit_task
1174                          */
1175                         sig->notify_count = -1;
1176                         if (likely(leader->exit_state))
1177                                 break;
1178                         __set_current_state(TASK_KILLABLE);
1179                         write_unlock_irq(&tasklist_lock);
1180                         cgroup_threadgroup_change_end(tsk);
1181                         schedule();
1182                         if (__fatal_signal_pending(tsk))
1183                                 goto killed;
1184                 }
1185
1186                 /*
1187                  * The only record we have of the real-time age of a
1188                  * process, regardless of execs it's done, is start_time.
1189                  * All the past CPU time is accumulated in signal_struct
1190                  * from sister threads now dead.  But in this non-leader
1191                  * exec, nothing survives from the original leader thread,
1192                  * whose birth marks the true age of this process now.
1193                  * When we take on its identity by switching to its PID, we
1194                  * also take its birthdate (always earlier than our own).
1195                  */
1196                 tsk->start_time = leader->start_time;
1197                 tsk->start_boottime = leader->start_boottime;
1198
1199                 BUG_ON(!same_thread_group(leader, tsk));
1200                 /*
1201                  * An exec() starts a new thread group with the
1202                  * TGID of the previous thread group. Rehash the
1203                  * two threads with a switched PID, and release
1204                  * the former thread group leader:
1205                  */
1206
1207                 /* Become a process group leader with the old leader's pid.
1208                  * The old leader becomes a thread of the this thread group.
1209                  */
1210                 exchange_tids(tsk, leader);
1211                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1212                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1213                 transfer_pid(leader, tsk, PIDTYPE_SID);
1214
1215                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1216                 list_replace_init(&leader->sibling, &tsk->sibling);
1217
1218                 tsk->group_leader = tsk;
1219                 leader->group_leader = tsk;
1220
1221                 tsk->exit_signal = SIGCHLD;
1222                 leader->exit_signal = -1;
1223
1224                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1225                 leader->exit_state = EXIT_DEAD;
1226
1227                 /*
1228                  * We are going to release_task()->ptrace_unlink() silently,
1229                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1230                  * the tracer wont't block again waiting for this thread.
1231                  */
1232                 if (unlikely(leader->ptrace))
1233                         __wake_up_parent(leader, leader->parent);
1234                 write_unlock_irq(&tasklist_lock);
1235                 cgroup_threadgroup_change_end(tsk);
1236
1237                 release_task(leader);
1238         }
1239
1240         sig->group_exit_task = NULL;
1241         sig->notify_count = 0;
1242
1243 no_thread_group:
1244         /* we have changed execution domain */
1245         tsk->exit_signal = SIGCHLD;
1246
1247         BUG_ON(!thread_group_leader(tsk));
1248         return 0;
1249
1250 killed:
1251         /* protects against exit_notify() and __exit_signal() */
1252         read_lock(&tasklist_lock);
1253         sig->group_exit_task = NULL;
1254         sig->notify_count = 0;
1255         read_unlock(&tasklist_lock);
1256         return -EAGAIN;
1257 }
1258
1259
1260 /*
1261  * This function makes sure the current process has its own signal table,
1262  * so that flush_signal_handlers can later reset the handlers without
1263  * disturbing other processes.  (Other processes might share the signal
1264  * table via the CLONE_SIGHAND option to clone().)
1265  */
1266 static int unshare_sighand(struct task_struct *me)
1267 {
1268         struct sighand_struct *oldsighand = me->sighand;
1269
1270         if (refcount_read(&oldsighand->count) != 1) {
1271                 struct sighand_struct *newsighand;
1272                 /*
1273                  * This ->sighand is shared with the CLONE_SIGHAND
1274                  * but not CLONE_THREAD task, switch to the new one.
1275                  */
1276                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1277                 if (!newsighand)
1278                         return -ENOMEM;
1279
1280                 refcount_set(&newsighand->count, 1);
1281                 memcpy(newsighand->action, oldsighand->action,
1282                        sizeof(newsighand->action));
1283
1284                 write_lock_irq(&tasklist_lock);
1285                 spin_lock(&oldsighand->siglock);
1286                 rcu_assign_pointer(me->sighand, newsighand);
1287                 spin_unlock(&oldsighand->siglock);
1288                 write_unlock_irq(&tasklist_lock);
1289
1290                 __cleanup_sighand(oldsighand);
1291         }
1292         return 0;
1293 }
1294
1295 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1296 {
1297         task_lock(tsk);
1298         strncpy(buf, tsk->comm, buf_size);
1299         task_unlock(tsk);
1300         return buf;
1301 }
1302 EXPORT_SYMBOL_GPL(__get_task_comm);
1303
1304 /*
1305  * These functions flushes out all traces of the currently running executable
1306  * so that a new one can be started
1307  */
1308
1309 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1310 {
1311         task_lock(tsk);
1312         trace_task_rename(tsk, buf);
1313         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1314         task_unlock(tsk);
1315         perf_event_comm(tsk, exec);
1316 }
1317
1318 /*
1319  * Calling this is the point of no return. None of the failures will be
1320  * seen by userspace since either the process is already taking a fatal
1321  * signal (via de_thread() or coredump), or will have SEGV raised
1322  * (after exec_mmap()) by search_binary_handler (see below).
1323  */
1324 int begin_new_exec(struct linux_binprm * bprm)
1325 {
1326         struct task_struct *me = current;
1327         int retval;
1328
1329         /* Once we are committed compute the creds */
1330         retval = bprm_creds_from_file(bprm);
1331         if (retval)
1332                 return retval;
1333
1334         /*
1335          * Ensure all future errors are fatal.
1336          */
1337         bprm->point_of_no_return = true;
1338
1339         /*
1340          * Make this the only thread in the thread group.
1341          */
1342         retval = de_thread(me);
1343         if (retval)
1344                 goto out;
1345
1346         /*
1347          * Must be called _before_ exec_mmap() as bprm->mm is
1348          * not visibile until then. This also enables the update
1349          * to be lockless.
1350          */
1351         set_mm_exe_file(bprm->mm, bprm->file);
1352
1353         /* If the binary is not readable then enforce mm->dumpable=0 */
1354         would_dump(bprm, bprm->file);
1355         if (bprm->have_execfd)
1356                 would_dump(bprm, bprm->executable);
1357
1358         /*
1359          * Release all of the old mmap stuff
1360          */
1361         acct_arg_size(bprm, 0);
1362         retval = exec_mmap(bprm->mm);
1363         if (retval)
1364                 goto out;
1365
1366         bprm->mm = NULL;
1367
1368 #ifdef CONFIG_POSIX_TIMERS
1369         exit_itimers(me->signal);
1370         flush_itimer_signals();
1371 #endif
1372
1373         /*
1374          * Make the signal table private.
1375          */
1376         retval = unshare_sighand(me);
1377         if (retval)
1378                 goto out_unlock;
1379
1380         set_fs(USER_DS);
1381         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1382                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1383         flush_thread();
1384         me->personality &= ~bprm->per_clear;
1385
1386         /*
1387          * We have to apply CLOEXEC before we change whether the process is
1388          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1389          * trying to access the should-be-closed file descriptors of a process
1390          * undergoing exec(2).
1391          */
1392         do_close_on_exec(me->files);
1393
1394         if (bprm->secureexec) {
1395                 /* Make sure parent cannot signal privileged process. */
1396                 me->pdeath_signal = 0;
1397
1398                 /*
1399                  * For secureexec, reset the stack limit to sane default to
1400                  * avoid bad behavior from the prior rlimits. This has to
1401                  * happen before arch_pick_mmap_layout(), which examines
1402                  * RLIMIT_STACK, but after the point of no return to avoid
1403                  * needing to clean up the change on failure.
1404                  */
1405                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1406                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1407         }
1408
1409         me->sas_ss_sp = me->sas_ss_size = 0;
1410
1411         /*
1412          * Figure out dumpability. Note that this checking only of current
1413          * is wrong, but userspace depends on it. This should be testing
1414          * bprm->secureexec instead.
1415          */
1416         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1417             !(uid_eq(current_euid(), current_uid()) &&
1418               gid_eq(current_egid(), current_gid())))
1419                 set_dumpable(current->mm, suid_dumpable);
1420         else
1421                 set_dumpable(current->mm, SUID_DUMP_USER);
1422
1423         perf_event_exec();
1424         __set_task_comm(me, kbasename(bprm->filename), true);
1425
1426         /* An exec changes our domain. We are no longer part of the thread
1427            group */
1428         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1429         flush_signal_handlers(me, 0);
1430
1431         /*
1432          * install the new credentials for this executable
1433          */
1434         security_bprm_committing_creds(bprm);
1435
1436         commit_creds(bprm->cred);
1437         bprm->cred = NULL;
1438
1439         /*
1440          * Disable monitoring for regular users
1441          * when executing setuid binaries. Must
1442          * wait until new credentials are committed
1443          * by commit_creds() above
1444          */
1445         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1446                 perf_event_exit_task(me);
1447         /*
1448          * cred_guard_mutex must be held at least to this point to prevent
1449          * ptrace_attach() from altering our determination of the task's
1450          * credentials; any time after this it may be unlocked.
1451          */
1452         security_bprm_committed_creds(bprm);
1453
1454         /* Pass the opened binary to the interpreter. */
1455         if (bprm->have_execfd) {
1456                 retval = get_unused_fd_flags(0);
1457                 if (retval < 0)
1458                         goto out_unlock;
1459                 fd_install(retval, bprm->executable);
1460                 bprm->executable = NULL;
1461                 bprm->execfd = retval;
1462         }
1463         return 0;
1464
1465 out_unlock:
1466         mutex_unlock(&me->signal->exec_update_mutex);
1467 out:
1468         return retval;
1469 }
1470 EXPORT_SYMBOL(begin_new_exec);
1471
1472 void would_dump(struct linux_binprm *bprm, struct file *file)
1473 {
1474         struct inode *inode = file_inode(file);
1475         if (inode_permission(inode, MAY_READ) < 0) {
1476                 struct user_namespace *old, *user_ns;
1477                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1478
1479                 /* Ensure mm->user_ns contains the executable */
1480                 user_ns = old = bprm->mm->user_ns;
1481                 while ((user_ns != &init_user_ns) &&
1482                        !privileged_wrt_inode_uidgid(user_ns, inode))
1483                         user_ns = user_ns->parent;
1484
1485                 if (old != user_ns) {
1486                         bprm->mm->user_ns = get_user_ns(user_ns);
1487                         put_user_ns(old);
1488                 }
1489         }
1490 }
1491 EXPORT_SYMBOL(would_dump);
1492
1493 void setup_new_exec(struct linux_binprm * bprm)
1494 {
1495         /* Setup things that can depend upon the personality */
1496         struct task_struct *me = current;
1497
1498         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1499
1500         arch_setup_new_exec();
1501
1502         /* Set the new mm task size. We have to do that late because it may
1503          * depend on TIF_32BIT which is only updated in flush_thread() on
1504          * some architectures like powerpc
1505          */
1506         me->mm->task_size = TASK_SIZE;
1507         mutex_unlock(&me->signal->exec_update_mutex);
1508         mutex_unlock(&me->signal->cred_guard_mutex);
1509 }
1510 EXPORT_SYMBOL(setup_new_exec);
1511
1512 /* Runs immediately before start_thread() takes over. */
1513 void finalize_exec(struct linux_binprm *bprm)
1514 {
1515         /* Store any stack rlimit changes before starting thread. */
1516         task_lock(current->group_leader);
1517         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1518         task_unlock(current->group_leader);
1519 }
1520 EXPORT_SYMBOL(finalize_exec);
1521
1522 /*
1523  * Prepare credentials and lock ->cred_guard_mutex.
1524  * setup_new_exec() commits the new creds and drops the lock.
1525  * Or, if exec fails before, free_bprm() should release ->cred and
1526  * and unlock.
1527  */
1528 static int prepare_bprm_creds(struct linux_binprm *bprm)
1529 {
1530         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1531                 return -ERESTARTNOINTR;
1532
1533         bprm->cred = prepare_exec_creds();
1534         if (likely(bprm->cred))
1535                 return 0;
1536
1537         mutex_unlock(&current->signal->cred_guard_mutex);
1538         return -ENOMEM;
1539 }
1540
1541 static void free_bprm(struct linux_binprm *bprm)
1542 {
1543         free_arg_pages(bprm);
1544         if (bprm->cred) {
1545                 mutex_unlock(&current->signal->cred_guard_mutex);
1546                 abort_creds(bprm->cred);
1547         }
1548         if (bprm->file) {
1549                 allow_write_access(bprm->file);
1550                 fput(bprm->file);
1551         }
1552         if (bprm->executable)
1553                 fput(bprm->executable);
1554         /* If a binfmt changed the interp, free it. */
1555         if (bprm->interp != bprm->filename)
1556                 kfree(bprm->interp);
1557         kfree(bprm);
1558 }
1559
1560 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1561 {
1562         /* If a binfmt changed the interp, free it first. */
1563         if (bprm->interp != bprm->filename)
1564                 kfree(bprm->interp);
1565         bprm->interp = kstrdup(interp, GFP_KERNEL);
1566         if (!bprm->interp)
1567                 return -ENOMEM;
1568         return 0;
1569 }
1570 EXPORT_SYMBOL(bprm_change_interp);
1571
1572 /*
1573  * determine how safe it is to execute the proposed program
1574  * - the caller must hold ->cred_guard_mutex to protect against
1575  *   PTRACE_ATTACH or seccomp thread-sync
1576  */
1577 static void check_unsafe_exec(struct linux_binprm *bprm)
1578 {
1579         struct task_struct *p = current, *t;
1580         unsigned n_fs;
1581
1582         if (p->ptrace)
1583                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1584
1585         /*
1586          * This isn't strictly necessary, but it makes it harder for LSMs to
1587          * mess up.
1588          */
1589         if (task_no_new_privs(current))
1590                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1591
1592         t = p;
1593         n_fs = 1;
1594         spin_lock(&p->fs->lock);
1595         rcu_read_lock();
1596         while_each_thread(p, t) {
1597                 if (t->fs == p->fs)
1598                         n_fs++;
1599         }
1600         rcu_read_unlock();
1601
1602         if (p->fs->users > n_fs)
1603                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1604         else
1605                 p->fs->in_exec = 1;
1606         spin_unlock(&p->fs->lock);
1607 }
1608
1609 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1610 {
1611         /* Handle suid and sgid on files */
1612         struct inode *inode;
1613         unsigned int mode;
1614         kuid_t uid;
1615         kgid_t gid;
1616
1617         if (!mnt_may_suid(file->f_path.mnt))
1618                 return;
1619
1620         if (task_no_new_privs(current))
1621                 return;
1622
1623         inode = file->f_path.dentry->d_inode;
1624         mode = READ_ONCE(inode->i_mode);
1625         if (!(mode & (S_ISUID|S_ISGID)))
1626                 return;
1627
1628         /* Be careful if suid/sgid is set */
1629         inode_lock(inode);
1630
1631         /* reload atomically mode/uid/gid now that lock held */
1632         mode = inode->i_mode;
1633         uid = inode->i_uid;
1634         gid = inode->i_gid;
1635         inode_unlock(inode);
1636
1637         /* We ignore suid/sgid if there are no mappings for them in the ns */
1638         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1639                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1640                 return;
1641
1642         if (mode & S_ISUID) {
1643                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1644                 bprm->cred->euid = uid;
1645         }
1646
1647         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1648                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1649                 bprm->cred->egid = gid;
1650         }
1651 }
1652
1653 /*
1654  * Compute brpm->cred based upon the final binary.
1655  */
1656 static int bprm_creds_from_file(struct linux_binprm *bprm)
1657 {
1658         /* Compute creds based on which file? */
1659         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1660
1661         bprm_fill_uid(bprm, file);
1662         return security_bprm_creds_from_file(bprm, file);
1663 }
1664
1665 /*
1666  * Fill the binprm structure from the inode.
1667  * Read the first BINPRM_BUF_SIZE bytes
1668  *
1669  * This may be called multiple times for binary chains (scripts for example).
1670  */
1671 static int prepare_binprm(struct linux_binprm *bprm)
1672 {
1673         loff_t pos = 0;
1674
1675         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1676         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1677 }
1678
1679 /*
1680  * Arguments are '\0' separated strings found at the location bprm->p
1681  * points to; chop off the first by relocating brpm->p to right after
1682  * the first '\0' encountered.
1683  */
1684 int remove_arg_zero(struct linux_binprm *bprm)
1685 {
1686         int ret = 0;
1687         unsigned long offset;
1688         char *kaddr;
1689         struct page *page;
1690
1691         if (!bprm->argc)
1692                 return 0;
1693
1694         do {
1695                 offset = bprm->p & ~PAGE_MASK;
1696                 page = get_arg_page(bprm, bprm->p, 0);
1697                 if (!page) {
1698                         ret = -EFAULT;
1699                         goto out;
1700                 }
1701                 kaddr = kmap_atomic(page);
1702
1703                 for (; offset < PAGE_SIZE && kaddr[offset];
1704                                 offset++, bprm->p++)
1705                         ;
1706
1707                 kunmap_atomic(kaddr);
1708                 put_arg_page(page);
1709         } while (offset == PAGE_SIZE);
1710
1711         bprm->p++;
1712         bprm->argc--;
1713         ret = 0;
1714
1715 out:
1716         return ret;
1717 }
1718 EXPORT_SYMBOL(remove_arg_zero);
1719
1720 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1721 /*
1722  * cycle the list of binary formats handler, until one recognizes the image
1723  */
1724 static int search_binary_handler(struct linux_binprm *bprm)
1725 {
1726         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1727         struct linux_binfmt *fmt;
1728         int retval;
1729
1730         retval = prepare_binprm(bprm);
1731         if (retval < 0)
1732                 return retval;
1733
1734         retval = security_bprm_check(bprm);
1735         if (retval)
1736                 return retval;
1737
1738         retval = -ENOENT;
1739  retry:
1740         read_lock(&binfmt_lock);
1741         list_for_each_entry(fmt, &formats, lh) {
1742                 if (!try_module_get(fmt->module))
1743                         continue;
1744                 read_unlock(&binfmt_lock);
1745
1746                 retval = fmt->load_binary(bprm);
1747
1748                 read_lock(&binfmt_lock);
1749                 put_binfmt(fmt);
1750                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1751                         read_unlock(&binfmt_lock);
1752                         return retval;
1753                 }
1754         }
1755         read_unlock(&binfmt_lock);
1756
1757         if (need_retry) {
1758                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1759                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1760                         return retval;
1761                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1762                         return retval;
1763                 need_retry = false;
1764                 goto retry;
1765         }
1766
1767         return retval;
1768 }
1769
1770 static int exec_binprm(struct linux_binprm *bprm)
1771 {
1772         pid_t old_pid, old_vpid;
1773         int ret, depth;
1774
1775         /* Need to fetch pid before load_binary changes it */
1776         old_pid = current->pid;
1777         rcu_read_lock();
1778         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1779         rcu_read_unlock();
1780
1781         /* This allows 4 levels of binfmt rewrites before failing hard. */
1782         for (depth = 0;; depth++) {
1783                 struct file *exec;
1784                 if (depth > 5)
1785                         return -ELOOP;
1786
1787                 ret = search_binary_handler(bprm);
1788                 if (ret < 0)
1789                         return ret;
1790                 if (!bprm->interpreter)
1791                         break;
1792
1793                 exec = bprm->file;
1794                 bprm->file = bprm->interpreter;
1795                 bprm->interpreter = NULL;
1796
1797                 allow_write_access(exec);
1798                 if (unlikely(bprm->have_execfd)) {
1799                         if (bprm->executable) {
1800                                 fput(exec);
1801                                 return -ENOEXEC;
1802                         }
1803                         bprm->executable = exec;
1804                 } else
1805                         fput(exec);
1806         }
1807
1808         audit_bprm(bprm);
1809         trace_sched_process_exec(current, old_pid, bprm);
1810         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1811         proc_exec_connector(current);
1812         return 0;
1813 }
1814
1815 /*
1816  * sys_execve() executes a new program.
1817  */
1818 static int __do_execve_file(int fd, struct filename *filename,
1819                             struct user_arg_ptr argv,
1820                             struct user_arg_ptr envp,
1821                             int flags, struct file *file)
1822 {
1823         char *pathbuf = NULL;
1824         struct linux_binprm *bprm;
1825         struct files_struct *displaced;
1826         int retval;
1827
1828         if (IS_ERR(filename))
1829                 return PTR_ERR(filename);
1830
1831         /*
1832          * We move the actual failure in case of RLIMIT_NPROC excess from
1833          * set*uid() to execve() because too many poorly written programs
1834          * don't check setuid() return code.  Here we additionally recheck
1835          * whether NPROC limit is still exceeded.
1836          */
1837         if ((current->flags & PF_NPROC_EXCEEDED) &&
1838             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1839                 retval = -EAGAIN;
1840                 goto out_ret;
1841         }
1842
1843         /* We're below the limit (still or again), so we don't want to make
1844          * further execve() calls fail. */
1845         current->flags &= ~PF_NPROC_EXCEEDED;
1846
1847         retval = unshare_files(&displaced);
1848         if (retval)
1849                 goto out_ret;
1850
1851         retval = -ENOMEM;
1852         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1853         if (!bprm)
1854                 goto out_files;
1855
1856         retval = prepare_bprm_creds(bprm);
1857         if (retval)
1858                 goto out_free;
1859
1860         check_unsafe_exec(bprm);
1861         current->in_execve = 1;
1862
1863         if (!file)
1864                 file = do_open_execat(fd, filename, flags);
1865         retval = PTR_ERR(file);
1866         if (IS_ERR(file))
1867                 goto out_unmark;
1868
1869         sched_exec();
1870
1871         bprm->file = file;
1872         if (!filename) {
1873                 bprm->filename = "none";
1874         } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1875                 bprm->filename = filename->name;
1876         } else {
1877                 if (filename->name[0] == '\0')
1878                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1879                 else
1880                         pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1881                                             fd, filename->name);
1882                 if (!pathbuf) {
1883                         retval = -ENOMEM;
1884                         goto out_unmark;
1885                 }
1886                 /*
1887                  * Record that a name derived from an O_CLOEXEC fd will be
1888                  * inaccessible after exec. Relies on having exclusive access to
1889                  * current->files (due to unshare_files above).
1890                  */
1891                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1892                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1893                 bprm->filename = pathbuf;
1894         }
1895         bprm->interp = bprm->filename;
1896
1897         retval = bprm_mm_init(bprm);
1898         if (retval)
1899                 goto out_unmark;
1900
1901         retval = prepare_arg_pages(bprm, argv, envp);
1902         if (retval < 0)
1903                 goto out;
1904
1905         /* Set the unchanging part of bprm->cred */
1906         retval = security_bprm_creds_for_exec(bprm);
1907         if (retval)
1908                 goto out;
1909
1910         retval = copy_string_kernel(bprm->filename, bprm);
1911         if (retval < 0)
1912                 goto out;
1913
1914         bprm->exec = bprm->p;
1915         retval = copy_strings(bprm->envc, envp, bprm);
1916         if (retval < 0)
1917                 goto out;
1918
1919         retval = copy_strings(bprm->argc, argv, bprm);
1920         if (retval < 0)
1921                 goto out;
1922
1923         retval = exec_binprm(bprm);
1924         if (retval < 0)
1925                 goto out;
1926
1927         /* execve succeeded */
1928         current->fs->in_exec = 0;
1929         current->in_execve = 0;
1930         rseq_execve(current);
1931         acct_update_integrals(current);
1932         task_numa_free(current, false);
1933         free_bprm(bprm);
1934         kfree(pathbuf);
1935         if (filename)
1936                 putname(filename);
1937         if (displaced)
1938                 put_files_struct(displaced);
1939         return retval;
1940
1941 out:
1942         /*
1943          * If past the point of no return ensure the the code never
1944          * returns to the userspace process.  Use an existing fatal
1945          * signal if present otherwise terminate the process with
1946          * SIGSEGV.
1947          */
1948         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1949                 force_sigsegv(SIGSEGV);
1950         if (bprm->mm) {
1951                 acct_arg_size(bprm, 0);
1952                 mmput(bprm->mm);
1953         }
1954
1955 out_unmark:
1956         current->fs->in_exec = 0;
1957         current->in_execve = 0;
1958
1959 out_free:
1960         free_bprm(bprm);
1961         kfree(pathbuf);
1962
1963 out_files:
1964         if (displaced)
1965                 reset_files_struct(displaced);
1966 out_ret:
1967         if (filename)
1968                 putname(filename);
1969         return retval;
1970 }
1971
1972 static int do_execveat_common(int fd, struct filename *filename,
1973                               struct user_arg_ptr argv,
1974                               struct user_arg_ptr envp,
1975                               int flags)
1976 {
1977         return __do_execve_file(fd, filename, argv, envp, flags, NULL);
1978 }
1979
1980 int do_execve_file(struct file *file, void *__argv, void *__envp)
1981 {
1982         struct user_arg_ptr argv = { .ptr.native = __argv };
1983         struct user_arg_ptr envp = { .ptr.native = __envp };
1984
1985         return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file);
1986 }
1987
1988 int do_execve(struct filename *filename,
1989         const char __user *const __user *__argv,
1990         const char __user *const __user *__envp)
1991 {
1992         struct user_arg_ptr argv = { .ptr.native = __argv };
1993         struct user_arg_ptr envp = { .ptr.native = __envp };
1994         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1995 }
1996
1997 int do_execveat(int fd, struct filename *filename,
1998                 const char __user *const __user *__argv,
1999                 const char __user *const __user *__envp,
2000                 int flags)
2001 {
2002         struct user_arg_ptr argv = { .ptr.native = __argv };
2003         struct user_arg_ptr envp = { .ptr.native = __envp };
2004
2005         return do_execveat_common(fd, filename, argv, envp, flags);
2006 }
2007
2008 #ifdef CONFIG_COMPAT
2009 static int compat_do_execve(struct filename *filename,
2010         const compat_uptr_t __user *__argv,
2011         const compat_uptr_t __user *__envp)
2012 {
2013         struct user_arg_ptr argv = {
2014                 .is_compat = true,
2015                 .ptr.compat = __argv,
2016         };
2017         struct user_arg_ptr envp = {
2018                 .is_compat = true,
2019                 .ptr.compat = __envp,
2020         };
2021         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2022 }
2023
2024 static int compat_do_execveat(int fd, struct filename *filename,
2025                               const compat_uptr_t __user *__argv,
2026                               const compat_uptr_t __user *__envp,
2027                               int flags)
2028 {
2029         struct user_arg_ptr argv = {
2030                 .is_compat = true,
2031                 .ptr.compat = __argv,
2032         };
2033         struct user_arg_ptr envp = {
2034                 .is_compat = true,
2035                 .ptr.compat = __envp,
2036         };
2037         return do_execveat_common(fd, filename, argv, envp, flags);
2038 }
2039 #endif
2040
2041 void set_binfmt(struct linux_binfmt *new)
2042 {
2043         struct mm_struct *mm = current->mm;
2044
2045         if (mm->binfmt)
2046                 module_put(mm->binfmt->module);
2047
2048         mm->binfmt = new;
2049         if (new)
2050                 __module_get(new->module);
2051 }
2052 EXPORT_SYMBOL(set_binfmt);
2053
2054 /*
2055  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2056  */
2057 void set_dumpable(struct mm_struct *mm, int value)
2058 {
2059         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2060                 return;
2061
2062         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2063 }
2064
2065 SYSCALL_DEFINE3(execve,
2066                 const char __user *, filename,
2067                 const char __user *const __user *, argv,
2068                 const char __user *const __user *, envp)
2069 {
2070         return do_execve(getname(filename), argv, envp);
2071 }
2072
2073 SYSCALL_DEFINE5(execveat,
2074                 int, fd, const char __user *, filename,
2075                 const char __user *const __user *, argv,
2076                 const char __user *const __user *, envp,
2077                 int, flags)
2078 {
2079         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2080
2081         return do_execveat(fd,
2082                            getname_flags(filename, lookup_flags, NULL),
2083                            argv, envp, flags);
2084 }
2085
2086 #ifdef CONFIG_COMPAT
2087 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2088         const compat_uptr_t __user *, argv,
2089         const compat_uptr_t __user *, envp)
2090 {
2091         return compat_do_execve(getname(filename), argv, envp);
2092 }
2093
2094 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2095                        const char __user *, filename,
2096                        const compat_uptr_t __user *, argv,
2097                        const compat_uptr_t __user *, envp,
2098                        int,  flags)
2099 {
2100         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2101
2102         return compat_do_execveat(fd,
2103                                   getname_flags(filename, lookup_flags, NULL),
2104                                   argv, envp, flags);
2105 }
2106 #endif