fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/vmacache.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57 #include <linux/oom.h>
  58 #include <linux/compat.h>
  59 #include <linux/vmalloc.h>
  60
  61 #include <asm/uaccess.h>
  62 #include <asm/mmu_context.h>
  63 #include <asm/tlb.h>
  64
  65 #include <trace/events/task.h>
  66 #include "internal.h"
  67
  68 #include <trace/events/sched.h>
  69
  70 int suid_dumpable = 0;
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  76 {
  77         BUG_ON(!fmt);
  78         if (WARN_ON(!fmt->load_binary))
  79                 return;
  80         write_lock(&binfmt_lock);
  81         insert ? list_add(&fmt->lh, &formats) :
  82                  list_add_tail(&fmt->lh, &formats);
  83         write_unlock(&binfmt_lock);
  84 }
  85
  86 EXPORT_SYMBOL(__register_binfmt);
  87
  88 void unregister_binfmt(struct linux_binfmt * fmt)
  89 {
  90         write_lock(&binfmt_lock);
  91         list_del(&fmt->lh);
  92         write_unlock(&binfmt_lock);
  93 }
  94
  95 EXPORT_SYMBOL(unregister_binfmt);
  96
  97 static inline void put_binfmt(struct linux_binfmt * fmt)
  98 {
  99         module_put(fmt->module);
 100 }
 101
 102 bool path_noexec(const struct path *path)
 103 {
 104         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 105                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 106 }
 107
 108 #ifdef CONFIG_USELIB
 109 /*
 110  * Note that a shared library must be both readable and executable due to
 111  * security reasons.
 112  *
 113  * Also note that we take the address to load from from the file itself.
 114  */
 115 SYSCALL_DEFINE1(uselib, const char __user *, library)
 116 {
 117         struct linux_binfmt *fmt;
 118         struct file *file;
 119         struct filename *tmp = getname(library);
 120         int error = PTR_ERR(tmp);
 121         static const struct open_flags uselib_flags = {
 122                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 123                 .acc_mode = MAY_READ | MAY_EXEC,
 124                 .intent = LOOKUP_OPEN,
 125                 .lookup_flags = LOOKUP_FOLLOW,
 126         };
 127
 128         if (IS_ERR(tmp))
 129                 goto out;
 130
 131         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 132         putname(tmp);
 133         error = PTR_ERR(file);
 134         if (IS_ERR(file))
 135                 goto out;
 136
 137         error = -EINVAL;
 138         if (!S_ISREG(file_inode(file)->i_mode))
 139                 goto exit;
 140
 141         error = -EACCES;
 142         if (path_noexec(&file->f_path))
 143                 goto exit;
 144
 145         fsnotify_open(file);
 146
 147         error = -ENOEXEC;
 148
 149         read_lock(&binfmt_lock);
 150         list_for_each_entry(fmt, &formats, lh) {
 151                 if (!fmt->load_shlib)
 152                         continue;
 153                 if (!try_module_get(fmt->module))
 154                         continue;
 155                 read_unlock(&binfmt_lock);
 156                 error = fmt->load_shlib(file);
 157                 read_lock(&binfmt_lock);
 158                 put_binfmt(fmt);
 159                 if (error != -ENOEXEC)
 160                         break;
 161         }
 162         read_unlock(&binfmt_lock);
 163 exit:
 164         fput(file);
 165 out:
 166         return error;
 167 }
 168 #endif /* #ifdef CONFIG_USELIB */
 169
 170 #ifdef CONFIG_MMU
 171 /*
 172  * The nascent bprm->mm is not visible until exec_mmap() but it can
 173  * use a lot of memory, account these pages in current->mm temporary
 174  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 175  * change the counter back via acct_arg_size(0).
 176  */
 177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 178 {
 179         struct mm_struct *mm = current->mm;
 180         long diff = (long)(pages - bprm->vma_pages);
 181
 182         if (!mm || !diff)
 183                 return;
 184
 185         bprm->vma_pages = pages;
 186         add_mm_counter(mm, MM_ANONPAGES, diff);
 187 }
 188
 189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 190                 int write)
 191 {
 192         struct page *page;
 193         int ret;
 194
 195 #ifdef CONFIG_STACK_GROWSUP
 196         if (write) {
 197                 ret = expand_downwards(bprm->vma, pos);
 198                 if (ret < 0)
 199                         return NULL;
 200         }
 201 #endif
 202         /*
 203          * We are doing an exec().  'current' is the process
 204          * doing the exec and bprm->mm is the new process's mm.
 205          */
 206         ret = get_user_pages_remote(current, bprm->mm, pos, 1, write,
 207                         1, &page, NULL);
 208         if (ret <= 0)
 209                 return NULL;
 210
 211         if (write) {
 212                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 213                 struct rlimit *rlim;
 214
 215                 acct_arg_size(bprm, size / PAGE_SIZE);
 216
 217                 /*
 218                  * We've historically supported up to 32 pages (ARG_MAX)
 219                  * of argument strings even with small stacks
 220                  */
 221                 if (size <= ARG_MAX)
 222                         return page;
 223
 224                 /*
 225                  * Limit to 1/4-th the stack size for the argv+env strings.
 226                  * This ensures that:
 227                  *  - the remaining binfmt code will not run out of stack space,
 228                  *  - the program will have a reasonable amount of stack left
 229                  *    to work from.
 230                  */
 231                 rlim = current->signal->rlim;
 232                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 233                         put_page(page);
 234                         return NULL;
 235                 }
 236         }
 237
 238         return page;
 239 }
 240
 241 static void put_arg_page(struct page *page)
 242 {
 243         put_page(page);
 244 }
 245
 246 static void free_arg_page(struct linux_binprm *bprm, int i)
 247 {
 248 }
 249
 250 static void free_arg_pages(struct linux_binprm *bprm)
 251 {
 252 }
 253
 254 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 255                 struct page *page)
 256 {
 257         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 258 }
 259
 260 static int __bprm_mm_init(struct linux_binprm *bprm)
 261 {
 262         int err;
 263         struct vm_area_struct *vma = NULL;
 264         struct mm_struct *mm = bprm->mm;
 265
 266         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 267         if (!vma)
 268                 return -ENOMEM;
 269
 270         down_write(&mm->mmap_sem);
 271         vma->vm_mm = mm;
 272
 273         /*
 274          * Place the stack at the largest stack address the architecture
 275          * supports. Later, we'll move this to an appropriate place. We don't
 276          * use STACK_TOP because that can depend on attributes which aren't
 277          * configured yet.
 278          */
 279         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 280         vma->vm_end = STACK_TOP_MAX;
 281         vma->vm_start = vma->vm_end - PAGE_SIZE;
 282         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 283         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 284         INIT_LIST_HEAD(&vma->anon_vma_chain);
 285
 286         err = insert_vm_struct(mm, vma);
 287         if (err)
 288                 goto err;
 289
 290         mm->stack_vm = mm->total_vm = 1;
 291         arch_bprm_mm_init(mm, vma);
 292         up_write(&mm->mmap_sem);
 293         bprm->p = vma->vm_end - sizeof(void *);
 294         return 0;
 295 err:
 296         up_write(&mm->mmap_sem);
 297         bprm->vma = NULL;
 298         kmem_cache_free(vm_area_cachep, vma);
 299         return err;
 300 }
 301
 302 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 303 {
 304         return len <= MAX_ARG_STRLEN;
 305 }
 306
 307 #else
 308
 309 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 310 {
 311 }
 312
 313 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 314                 int write)
 315 {
 316         struct page *page;
 317
 318         page = bprm->page[pos / PAGE_SIZE];
 319         if (!page && write) {
 320                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 321                 if (!page)
 322                         return NULL;
 323                 bprm->page[pos / PAGE_SIZE] = page;
 324         }
 325
 326         return page;
 327 }
 328
 329 static void put_arg_page(struct page *page)
 330 {
 331 }
 332
 333 static void free_arg_page(struct linux_binprm *bprm, int i)
 334 {
 335         if (bprm->page[i]) {
 336                 __free_page(bprm->page[i]);
 337                 bprm->page[i] = NULL;
 338         }
 339 }
 340
 341 static void free_arg_pages(struct linux_binprm *bprm)
 342 {
 343         int i;
 344
 345         for (i = 0; i < MAX_ARG_PAGES; i++)
 346                 free_arg_page(bprm, i);
 347 }
 348
 349 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 350                 struct page *page)
 351 {
 352 }
 353
 354 static int __bprm_mm_init(struct linux_binprm *bprm)
 355 {
 356         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 357         return 0;
 358 }
 359
 360 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 361 {
 362         return len <= bprm->p;
 363 }
 364
 365 #endif /* CONFIG_MMU */
 366
 367 /*
 368  * Create a new mm_struct and populate it with a temporary stack
 369  * vm_area_struct.  We don't have enough context at this point to set the stack
 370  * flags, permissions, and offset, so we use temporary values.  We'll update
 371  * them later in setup_arg_pages().
 372  */
 373 static int bprm_mm_init(struct linux_binprm *bprm)
 374 {
 375         int err;
 376         struct mm_struct *mm = NULL;
 377
 378         bprm->mm = mm = mm_alloc();
 379         err = -ENOMEM;
 380         if (!mm)
 381                 goto err;
 382
 383         err = __bprm_mm_init(bprm);
 384         if (err)
 385                 goto err;
 386
 387         return 0;
 388
 389 err:
 390         if (mm) {
 391                 bprm->mm = NULL;
 392                 mmdrop(mm);
 393         }
 394
 395         return err;
 396 }
 397
 398 struct user_arg_ptr {
 399 #ifdef CONFIG_COMPAT
 400         bool is_compat;
 401 #endif
 402         union {
 403                 const char __user *const __user *native;
 404 #ifdef CONFIG_COMPAT
 405                 const compat_uptr_t __user *compat;
 406 #endif
 407         } ptr;
 408 };
 409
 410 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 411 {
 412         const char __user *native;
 413
 414 #ifdef CONFIG_COMPAT
 415         if (unlikely(argv.is_compat)) {
 416                 compat_uptr_t compat;
 417
 418                 if (get_user(compat, argv.ptr.compat + nr))
 419                         return ERR_PTR(-EFAULT);
 420
 421                 return compat_ptr(compat);
 422         }
 423 #endif
 424
 425         if (get_user(native, argv.ptr.native + nr))
 426                 return ERR_PTR(-EFAULT);
 427
 428         return native;
 429 }
 430
 431 /*
 432  * count() counts the number of strings in array ARGV.
 433  */
 434 static int count(struct user_arg_ptr argv, int max)
 435 {
 436         int i = 0;
 437
 438         if (argv.ptr.native != NULL) {
 439                 for (;;) {
 440                         const char __user *p = get_user_arg_ptr(argv, i);
 441
 442                         if (!p)
 443                                 break;
 444
 445                         if (IS_ERR(p))
 446                                 return -EFAULT;
 447
 448                         if (i >= max)
 449                                 return -E2BIG;
 450                         ++i;
 451
 452                         if (fatal_signal_pending(current))
 453                                 return -ERESTARTNOHAND;
 454                         cond_resched();
 455                 }
 456         }
 457         return i;
 458 }
 459
 460 /*
 461  * 'copy_strings()' copies argument/environment strings from the old
 462  * processes's memory to the new process's stack.  The call to get_user_pages()
 463  * ensures the destination page is created and not swapped out.
 464  */
 465 static int copy_strings(int argc, struct user_arg_ptr argv,
 466                         struct linux_binprm *bprm)
 467 {
 468         struct page *kmapped_page = NULL;
 469         char *kaddr = NULL;
 470         unsigned long kpos = 0;
 471         int ret;
 472
 473         while (argc-- > 0) {
 474                 const char __user *str;
 475                 int len;
 476                 unsigned long pos;
 477
 478                 ret = -EFAULT;
 479                 str = get_user_arg_ptr(argv, argc);
 480                 if (IS_ERR(str))
 481                         goto out;
 482
 483                 len = strnlen_user(str, MAX_ARG_STRLEN);
 484                 if (!len)
 485                         goto out;
 486
 487                 ret = -E2BIG;
 488                 if (!valid_arg_len(bprm, len))
 489                         goto out;
 490
 491                 /* We're going to work our way backwords. */
 492                 pos = bprm->p;
 493                 str += len;
 494                 bprm->p -= len;
 495
 496                 while (len > 0) {
 497                         int offset, bytes_to_copy;
 498
 499                         if (fatal_signal_pending(current)) {
 500                                 ret = -ERESTARTNOHAND;
 501                                 goto out;
 502                         }
 503                         cond_resched();
 504
 505                         offset = pos % PAGE_SIZE;
 506                         if (offset == 0)
 507                                 offset = PAGE_SIZE;
 508
 509                         bytes_to_copy = offset;
 510                         if (bytes_to_copy > len)
 511                                 bytes_to_copy = len;
 512
 513                         offset -= bytes_to_copy;
 514                         pos -= bytes_to_copy;
 515                         str -= bytes_to_copy;
 516                         len -= bytes_to_copy;
 517
 518                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 519                                 struct page *page;
 520
 521                                 page = get_arg_page(bprm, pos, 1);
 522                                 if (!page) {
 523                                         ret = -E2BIG;
 524                                         goto out;
 525                                 }
 526
 527                                 if (kmapped_page) {
 528                                         flush_kernel_dcache_page(kmapped_page);
 529                                         kunmap(kmapped_page);
 530                                         put_arg_page(kmapped_page);
 531                                 }
 532                                 kmapped_page = page;
 533                                 kaddr = kmap(kmapped_page);
 534                                 kpos = pos & PAGE_MASK;
 535                                 flush_arg_page(bprm, kpos, kmapped_page);
 536                         }
 537                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 538                                 ret = -EFAULT;
 539                                 goto out;
 540                         }
 541                 }
 542         }
 543         ret = 0;
 544 out:
 545         if (kmapped_page) {
 546                 flush_kernel_dcache_page(kmapped_page);
 547                 kunmap(kmapped_page);
 548                 put_arg_page(kmapped_page);
 549         }
 550         return ret;
 551 }
 552
 553 /*
 554  * Like copy_strings, but get argv and its values from kernel memory.
 555  */
 556 int copy_strings_kernel(int argc, const char *const *__argv,
 557                         struct linux_binprm *bprm)
 558 {
 559         int r;
 560         mm_segment_t oldfs = get_fs();
 561         struct user_arg_ptr argv = {
 562                 .ptr.native = (const char __user *const  __user *)__argv,
 563         };
 564
 565         set_fs(KERNEL_DS);
 566         r = copy_strings(argc, argv, bprm);
 567         set_fs(oldfs);
 568
 569         return r;
 570 }
 571 EXPORT_SYMBOL(copy_strings_kernel);
 572
 573 #ifdef CONFIG_MMU
 574
 575 /*
 576  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 577  * the binfmt code determines where the new stack should reside, we shift it to
 578  * its final location.  The process proceeds as follows:
 579  *
 580  * 1) Use shift to calculate the new vma endpoints.
 581  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 582  *    arguments passed to subsequent functions are consistent.
 583  * 3) Move vma's page tables to the new range.
 584  * 4) Free up any cleared pgd range.
 585  * 5) Shrink the vma to cover only the new range.
 586  */
 587 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 588 {
 589         struct mm_struct *mm = vma->vm_mm;
 590         unsigned long old_start = vma->vm_start;
 591         unsigned long old_end = vma->vm_end;
 592         unsigned long length = old_end - old_start;
 593         unsigned long new_start = old_start - shift;
 594         unsigned long new_end = old_end - shift;
 595         struct mmu_gather tlb;
 596
 597         BUG_ON(new_start > new_end);
 598
 599         /*
 600          * ensure there are no vmas between where we want to go
 601          * and where we are
 602          */
 603         if (vma != find_vma(mm, new_start))
 604                 return -EFAULT;
 605
 606         /*
 607          * cover the whole range: [new_start, old_end)
 608          */
 609         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 610                 return -ENOMEM;
 611
 612         /*
 613          * move the page tables downwards, on failure we rely on
 614          * process cleanup to remove whatever mess we made.
 615          */
 616         if (length != move_page_tables(vma, old_start,
 617                                        vma, new_start, length, false))
 618                 return -ENOMEM;
 619
 620         lru_add_drain();
 621         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 622         if (new_end > old_start) {
 623                 /*
 624                  * when the old and new regions overlap clear from new_end.
 625                  */
 626                 free_pgd_range(&tlb, new_end, old_end, new_end,
 627                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 628         } else {
 629                 /*
 630                  * otherwise, clean from old_start; this is done to not touch
 631                  * the address space in [new_end, old_start) some architectures
 632                  * have constraints on va-space that make this illegal (IA64) -
 633                  * for the others its just a little faster.
 634                  */
 635                 free_pgd_range(&tlb, old_start, old_end, new_end,
 636                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 637         }
 638         tlb_finish_mmu(&tlb, old_start, old_end);
 639
 640         /*
 641          * Shrink the vma to just the new range.  Always succeeds.
 642          */
 643         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 644
 645         return 0;
 646 }
 647
 648 /*
 649  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 650  * the stack is optionally relocated, and some extra space is added.
 651  */
 652 int setup_arg_pages(struct linux_binprm *bprm,
 653                     unsigned long stack_top,
 654                     int executable_stack)
 655 {
 656         unsigned long ret;
 657         unsigned long stack_shift;
 658         struct mm_struct *mm = current->mm;
 659         struct vm_area_struct *vma = bprm->vma;
 660         struct vm_area_struct *prev = NULL;
 661         unsigned long vm_flags;
 662         unsigned long stack_base;
 663         unsigned long stack_size;
 664         unsigned long stack_expand;
 665         unsigned long rlim_stack;
 666
 667 #ifdef CONFIG_STACK_GROWSUP
 668         /* Limit stack size */
 669         stack_base = rlimit_max(RLIMIT_STACK);
 670         if (stack_base > STACK_SIZE_MAX)
 671                 stack_base = STACK_SIZE_MAX;
 672
 673         /* Add space for stack randomization. */
 674         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 675
 676         /* Make sure we didn't let the argument array grow too large. */
 677         if (vma->vm_end - vma->vm_start > stack_base)
 678                 return -ENOMEM;
 679
 680         stack_base = PAGE_ALIGN(stack_top - stack_base);
 681
 682         stack_shift = vma->vm_start - stack_base;
 683         mm->arg_start = bprm->p - stack_shift;
 684         bprm->p = vma->vm_end - stack_shift;
 685 #else
 686         stack_top = arch_align_stack(stack_top);
 687         stack_top = PAGE_ALIGN(stack_top);
 688
 689         if (unlikely(stack_top < mmap_min_addr) ||
 690             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 691                 return -ENOMEM;
 692
 693         stack_shift = vma->vm_end - stack_top;
 694
 695         bprm->p -= stack_shift;
 696         mm->arg_start = bprm->p;
 697 #endif
 698
 699         if (bprm->loader)
 700                 bprm->loader -= stack_shift;
 701         bprm->exec -= stack_shift;
 702
 703         down_write(&mm->mmap_sem);
 704         vm_flags = VM_STACK_FLAGS;
 705
 706         /*
 707          * Adjust stack execute permissions; explicitly enable for
 708          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 709          * (arch default) otherwise.
 710          */
 711         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 712                 vm_flags |= VM_EXEC;
 713         else if (executable_stack == EXSTACK_DISABLE_X)
 714                 vm_flags &= ~VM_EXEC;
 715         vm_flags |= mm->def_flags;
 716         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 717
 718         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 719                         vm_flags);
 720         if (ret)
 721                 goto out_unlock;
 722         BUG_ON(prev != vma);
 723
 724         /* Move stack pages down in memory. */
 725         if (stack_shift) {
 726                 ret = shift_arg_pages(vma, stack_shift);
 727                 if (ret)
 728                         goto out_unlock;
 729         }
 730
 731         /* mprotect_fixup is overkill to remove the temporary stack flags */
 732         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 733
 734         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 735         stack_size = vma->vm_end - vma->vm_start;
 736         /*
 737          * Align this down to a page boundary as expand_stack
 738          * will align it up.
 739          */
 740         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 741 #ifdef CONFIG_STACK_GROWSUP
 742         if (stack_size + stack_expand > rlim_stack)
 743                 stack_base = vma->vm_start + rlim_stack;
 744         else
 745                 stack_base = vma->vm_end + stack_expand;
 746 #else
 747         if (stack_size + stack_expand > rlim_stack)
 748                 stack_base = vma->vm_end - rlim_stack;
 749         else
 750                 stack_base = vma->vm_start - stack_expand;
 751 #endif
 752         current->mm->start_stack = bprm->p;
 753         ret = expand_stack(vma, stack_base);
 754         if (ret)
 755                 ret = -EFAULT;
 756
 757 out_unlock:
 758         up_write(&mm->mmap_sem);
 759         return ret;
 760 }
 761 EXPORT_SYMBOL(setup_arg_pages);
 762
 763 #endif /* CONFIG_MMU */
 764
 765 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 766 {
 767         struct file *file;
 768         int err;
 769         struct open_flags open_exec_flags = {
 770                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 771                 .acc_mode = MAY_EXEC,
 772                 .intent = LOOKUP_OPEN,
 773                 .lookup_flags = LOOKUP_FOLLOW,
 774         };
 775
 776         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 777                 return ERR_PTR(-EINVAL);
 778         if (flags & AT_SYMLINK_NOFOLLOW)
 779                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 780         if (flags & AT_EMPTY_PATH)
 781                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 782
 783         file = do_filp_open(fd, name, &open_exec_flags);
 784         if (IS_ERR(file))
 785                 goto out;
 786
 787         err = -EACCES;
 788         if (!S_ISREG(file_inode(file)->i_mode))
 789                 goto exit;
 790
 791         if (path_noexec(&file->f_path))
 792                 goto exit;
 793
 794         err = deny_write_access(file);
 795         if (err)
 796                 goto exit;
 797
 798         if (name->name[0] != '\0')
 799                 fsnotify_open(file);
 800
 801 out:
 802         return file;
 803
 804 exit:
 805         fput(file);
 806         return ERR_PTR(err);
 807 }
 808
 809 struct file *open_exec(const char *name)
 810 {
 811         struct filename *filename = getname_kernel(name);
 812         struct file *f = ERR_CAST(filename);
 813
 814         if (!IS_ERR(filename)) {
 815                 f = do_open_execat(AT_FDCWD, filename, 0);
 816                 putname(filename);
 817         }
 818         return f;
 819 }
 820 EXPORT_SYMBOL(open_exec);
 821
 822 int kernel_read(struct file *file, loff_t offset,
 823                 char *addr, unsigned long count)
 824 {
 825         mm_segment_t old_fs;
 826         loff_t pos = offset;
 827         int result;
 828
 829         old_fs = get_fs();
 830         set_fs(get_ds());
 831         /* The cast to a user pointer is valid due to the set_fs() */
 832         result = vfs_read(file, (void __user *)addr, count, &pos);
 833         set_fs(old_fs);
 834         return result;
 835 }
 836
 837 EXPORT_SYMBOL(kernel_read);
 838
 839 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 840                      loff_t max_size, enum kernel_read_file_id id)
 841 {
 842         loff_t i_size, pos;
 843         ssize_t bytes = 0;
 844         int ret;
 845
 846         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 847                 return -EINVAL;
 848
 849         ret = security_kernel_read_file(file, id);
 850         if (ret)
 851                 return ret;
 852
 853         ret = deny_write_access(file);
 854         if (ret)
 855                 return ret;
 856
 857         i_size = i_size_read(file_inode(file));
 858         if (max_size > 0 && i_size > max_size) {
 859                 ret = -EFBIG;
 860                 goto out;
 861         }
 862         if (i_size <= 0) {
 863                 ret = -EINVAL;
 864                 goto out;
 865         }
 866
 867         *buf = vmalloc(i_size);
 868         if (!*buf) {
 869                 ret = -ENOMEM;
 870                 goto out;
 871         }
 872
 873         pos = 0;
 874         while (pos < i_size) {
 875                 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 876                                     i_size - pos);
 877                 if (bytes < 0) {
 878                         ret = bytes;
 879                         goto out;
 880                 }
 881
 882                 if (bytes == 0)
 883                         break;
 884                 pos += bytes;
 885         }
 886
 887         if (pos != i_size) {
 888                 ret = -EIO;
 889                 goto out_free;
 890         }
 891
 892         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 893         if (!ret)
 894                 *size = pos;
 895
 896 out_free:
 897         if (ret < 0) {
 898                 vfree(*buf);
 899                 *buf = NULL;
 900         }
 901
 902 out:
 903         allow_write_access(file);
 904         return ret;
 905 }
 906 EXPORT_SYMBOL_GPL(kernel_read_file);
 907
 908 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 909                                loff_t max_size, enum kernel_read_file_id id)
 910 {
 911         struct file *file;
 912         int ret;
 913
 914         if (!path || !*path)
 915                 return -EINVAL;
 916
 917         file = filp_open(path, O_RDONLY, 0);
 918         if (IS_ERR(file))
 919                 return PTR_ERR(file);
 920
 921         ret = kernel_read_file(file, buf, size, max_size, id);
 922         fput(file);
 923         return ret;
 924 }
 925 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 926
 927 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 928                              enum kernel_read_file_id id)
 929 {
 930         struct fd f = fdget(fd);
 931         int ret = -EBADF;
 932
 933         if (!f.file)
 934                 goto out;
 935
 936         ret = kernel_read_file(f.file, buf, size, max_size, id);
 937 out:
 938         fdput(f);
 939         return ret;
 940 }
 941 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 942
 943 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 944 {
 945         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 946         if (res > 0)
 947                 flush_icache_range(addr, addr + len);
 948         return res;
 949 }
 950 EXPORT_SYMBOL(read_code);
 951
 952 static int exec_mmap(struct mm_struct *mm)
 953 {
 954         struct task_struct *tsk;
 955         struct mm_struct *old_mm, *active_mm;
 956
 957         /* Notify parent that we're no longer interested in the old VM */
 958         tsk = current;
 959         old_mm = current->mm;
 960         mm_release(tsk, old_mm);
 961
 962         if (old_mm) {
 963                 sync_mm_rss(old_mm);
 964                 /*
 965                  * Make sure that if there is a core dump in progress
 966                  * for the old mm, we get out and die instead of going
 967                  * through with the exec.  We must hold mmap_sem around
 968                  * checking core_state and changing tsk->mm.
 969                  */
 970                 down_read(&old_mm->mmap_sem);
 971                 if (unlikely(old_mm->core_state)) {
 972                         up_read(&old_mm->mmap_sem);
 973                         return -EINTR;
 974                 }
 975         }
 976         task_lock(tsk);
 977         active_mm = tsk->active_mm;
 978         tsk->mm = mm;
 979         tsk->active_mm = mm;
 980         activate_mm(active_mm, mm);
 981         tsk->mm->vmacache_seqnum = 0;
 982         vmacache_flush(tsk);
 983         task_unlock(tsk);
 984         if (old_mm) {
 985                 up_read(&old_mm->mmap_sem);
 986                 BUG_ON(active_mm != old_mm);
 987                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 988                 mm_update_next_owner(old_mm);
 989                 mmput(old_mm);
 990                 return 0;
 991         }
 992         mmdrop(active_mm);
 993         return 0;
 994 }
 995
 996 /*
 997  * This function makes sure the current process has its own signal table,
 998  * so that flush_signal_handlers can later reset the handlers without
 999  * disturbing other processes.  (Other processes might share the signal
1000  * table via the CLONE_SIGHAND option to clone().)
1001  */
1002 static int de_thread(struct task_struct *tsk)
1003 {
1004         struct signal_struct *sig = tsk->signal;
1005         struct sighand_struct *oldsighand = tsk->sighand;
1006         spinlock_t *lock = &oldsighand->siglock;
1007
1008         if (thread_group_empty(tsk))
1009                 goto no_thread_group;
1010
1011         /*
1012          * Kill all other threads in the thread group.
1013          */
1014         spin_lock_irq(lock);
1015         if (signal_group_exit(sig)) {
1016                 /*
1017                  * Another group action in progress, just
1018                  * return so that the signal is processed.
1019                  */
1020                 spin_unlock_irq(lock);
1021                 return -EAGAIN;
1022         }
1023
1024         sig->group_exit_task = tsk;
1025         sig->notify_count = zap_other_threads(tsk);
1026         if (!thread_group_leader(tsk))
1027                 sig->notify_count--;
1028
1029         while (sig->notify_count) {
1030                 __set_current_state(TASK_KILLABLE);
1031                 spin_unlock_irq(lock);
1032                 schedule();
1033                 if (unlikely(__fatal_signal_pending(tsk)))
1034                         goto killed;
1035                 spin_lock_irq(lock);
1036         }
1037         spin_unlock_irq(lock);
1038
1039         /*
1040          * At this point all other threads have exited, all we have to
1041          * do is to wait for the thread group leader to become inactive,
1042          * and to assume its PID:
1043          */
1044         if (!thread_group_leader(tsk)) {
1045                 struct task_struct *leader = tsk->group_leader;
1046
1047                 for (;;) {
1048                         threadgroup_change_begin(tsk);
1049                         write_lock_irq(&tasklist_lock);
1050                         /*
1051                          * Do this under tasklist_lock to ensure that
1052                          * exit_notify() can't miss ->group_exit_task
1053                          */
1054                         sig->notify_count = -1;
1055                         if (likely(leader->exit_state))
1056                                 break;
1057                         __set_current_state(TASK_KILLABLE);
1058                         write_unlock_irq(&tasklist_lock);
1059                         threadgroup_change_end(tsk);
1060                         schedule();
1061                         if (unlikely(__fatal_signal_pending(tsk)))
1062                                 goto killed;
1063                 }
1064
1065                 /*
1066                  * The only record we have of the real-time age of a
1067                  * process, regardless of execs it's done, is start_time.
1068                  * All the past CPU time is accumulated in signal_struct
1069                  * from sister threads now dead.  But in this non-leader
1070                  * exec, nothing survives from the original leader thread,
1071                  * whose birth marks the true age of this process now.
1072                  * When we take on its identity by switching to its PID, we
1073                  * also take its birthdate (always earlier than our own).
1074                  */
1075                 tsk->start_time = leader->start_time;
1076                 tsk->real_start_time = leader->real_start_time;
1077
1078                 BUG_ON(!same_thread_group(leader, tsk));
1079                 BUG_ON(has_group_leader_pid(tsk));
1080                 /*
1081                  * An exec() starts a new thread group with the
1082                  * TGID of the previous thread group. Rehash the
1083                  * two threads with a switched PID, and release
1084                  * the former thread group leader:
1085                  */
1086
1087                 /* Become a process group leader with the old leader's pid.
1088                  * The old leader becomes a thread of the this thread group.
1089                  * Note: The old leader also uses this pid until release_task
1090                  *       is called.  Odd but simple and correct.
1091                  */
1092                 tsk->pid = leader->pid;
1093                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1094                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1095                 transfer_pid(leader, tsk, PIDTYPE_SID);
1096
1097                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1098                 list_replace_init(&leader->sibling, &tsk->sibling);
1099
1100                 tsk->group_leader = tsk;
1101                 leader->group_leader = tsk;
1102
1103                 tsk->exit_signal = SIGCHLD;
1104                 leader->exit_signal = -1;
1105
1106                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1107                 leader->exit_state = EXIT_DEAD;
1108
1109                 /*
1110                  * We are going to release_task()->ptrace_unlink() silently,
1111                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1112                  * the tracer wont't block again waiting for this thread.
1113                  */
1114                 if (unlikely(leader->ptrace))
1115                         __wake_up_parent(leader, leader->parent);
1116                 write_unlock_irq(&tasklist_lock);
1117                 threadgroup_change_end(tsk);
1118
1119                 release_task(leader);
1120         }
1121
1122         sig->group_exit_task = NULL;
1123         sig->notify_count = 0;
1124
1125 no_thread_group:
1126         /* we have changed execution domain */
1127         tsk->exit_signal = SIGCHLD;
1128
1129         exit_itimers(sig);
1130         flush_itimer_signals();
1131
1132         if (atomic_read(&oldsighand->count) != 1) {
1133                 struct sighand_struct *newsighand;
1134                 /*
1135                  * This ->sighand is shared with the CLONE_SIGHAND
1136                  * but not CLONE_THREAD task, switch to the new one.
1137                  */
1138                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1139                 if (!newsighand)
1140                         return -ENOMEM;
1141
1142                 atomic_set(&newsighand->count, 1);
1143                 memcpy(newsighand->action, oldsighand->action,
1144                        sizeof(newsighand->action));
1145
1146                 write_lock_irq(&tasklist_lock);
1147                 spin_lock(&oldsighand->siglock);
1148                 rcu_assign_pointer(tsk->sighand, newsighand);
1149                 spin_unlock(&oldsighand->siglock);
1150                 write_unlock_irq(&tasklist_lock);
1151
1152                 __cleanup_sighand(oldsighand);
1153         }
1154
1155         BUG_ON(!thread_group_leader(tsk));
1156         return 0;
1157
1158 killed:
1159         /* protects against exit_notify() and __exit_signal() */
1160         read_lock(&tasklist_lock);
1161         sig->group_exit_task = NULL;
1162         sig->notify_count = 0;
1163         read_unlock(&tasklist_lock);
1164         return -EAGAIN;
1165 }
1166
1167 char *get_task_comm(char *buf, struct task_struct *tsk)
1168 {
1169         /* buf must be at least sizeof(tsk->comm) in size */
1170         task_lock(tsk);
1171         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1172         task_unlock(tsk);
1173         return buf;
1174 }
1175 EXPORT_SYMBOL_GPL(get_task_comm);
1176
1177 /*
1178  * These functions flushes out all traces of the currently running executable
1179  * so that a new one can be started
1180  */
1181
1182 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1183 {
1184         task_lock(tsk);
1185         trace_task_rename(tsk, buf);
1186         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1187         task_unlock(tsk);
1188         perf_event_comm(tsk, exec);
1189 }
1190
1191 int flush_old_exec(struct linux_binprm * bprm)
1192 {
1193         int retval;
1194
1195         /*
1196          * Make sure we have a private signal table and that
1197          * we are unassociated from the previous thread group.
1198          */
1199         retval = de_thread(current);
1200         if (retval)
1201                 goto out;
1202
1203         /*
1204          * Must be called _before_ exec_mmap() as bprm->mm is
1205          * not visibile until then. This also enables the update
1206          * to be lockless.
1207          */
1208         set_mm_exe_file(bprm->mm, bprm->file);
1209
1210         /*
1211          * Release all of the old mmap stuff
1212          */
1213         acct_arg_size(bprm, 0);
1214         retval = exec_mmap(bprm->mm);
1215         if (retval)
1216                 goto out;
1217
1218         bprm->mm = NULL;                /* We're using it now */
1219
1220         set_fs(USER_DS);
1221         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1222                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1223         flush_thread();
1224         current->personality &= ~bprm->per_clear;
1225
1226         return 0;
1227
1228 out:
1229         return retval;
1230 }
1231 EXPORT_SYMBOL(flush_old_exec);
1232
1233 void would_dump(struct linux_binprm *bprm, struct file *file)
1234 {
1235         if (inode_permission(file_inode(file), MAY_READ) < 0)
1236                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1237 }
1238 EXPORT_SYMBOL(would_dump);
1239
1240 void setup_new_exec(struct linux_binprm * bprm)
1241 {
1242         arch_pick_mmap_layout(current->mm);
1243
1244         /* This is the point of no return */
1245         current->sas_ss_sp = current->sas_ss_size = 0;
1246
1247         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1248                 set_dumpable(current->mm, SUID_DUMP_USER);
1249         else
1250                 set_dumpable(current->mm, suid_dumpable);
1251
1252         perf_event_exec();
1253         __set_task_comm(current, kbasename(bprm->filename), true);
1254
1255         /* Set the new mm task size. We have to do that late because it may
1256          * depend on TIF_32BIT which is only updated in flush_thread() on
1257          * some architectures like powerpc
1258          */
1259         current->mm->task_size = TASK_SIZE;
1260
1261         /* install the new credentials */
1262         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1263             !gid_eq(bprm->cred->gid, current_egid())) {
1264                 current->pdeath_signal = 0;
1265         } else {
1266                 would_dump(bprm, bprm->file);
1267                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1268                         set_dumpable(current->mm, suid_dumpable);
1269         }
1270
1271         /* An exec changes our domain. We are no longer part of the thread
1272            group */
1273         current->self_exec_id++;
1274         flush_signal_handlers(current, 0);
1275         do_close_on_exec(current->files);
1276 }
1277 EXPORT_SYMBOL(setup_new_exec);
1278
1279 /*
1280  * Prepare credentials and lock ->cred_guard_mutex.
1281  * install_exec_creds() commits the new creds and drops the lock.
1282  * Or, if exec fails before, free_bprm() should release ->cred and
1283  * and unlock.
1284  */
1285 int prepare_bprm_creds(struct linux_binprm *bprm)
1286 {
1287         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1288                 return -ERESTARTNOINTR;
1289
1290         bprm->cred = prepare_exec_creds();
1291         if (likely(bprm->cred))
1292                 return 0;
1293
1294         mutex_unlock(&current->signal->cred_guard_mutex);
1295         return -ENOMEM;
1296 }
1297
1298 static void free_bprm(struct linux_binprm *bprm)
1299 {
1300         free_arg_pages(bprm);
1301         if (bprm->cred) {
1302                 mutex_unlock(&current->signal->cred_guard_mutex);
1303                 abort_creds(bprm->cred);
1304         }
1305         if (bprm->file) {
1306                 allow_write_access(bprm->file);
1307                 fput(bprm->file);
1308         }
1309         /* If a binfmt changed the interp, free it. */
1310         if (bprm->interp != bprm->filename)
1311                 kfree(bprm->interp);
1312         kfree(bprm);
1313 }
1314
1315 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1316 {
1317         /* If a binfmt changed the interp, free it first. */
1318         if (bprm->interp != bprm->filename)
1319                 kfree(bprm->interp);
1320         bprm->interp = kstrdup(interp, GFP_KERNEL);
1321         if (!bprm->interp)
1322                 return -ENOMEM;
1323         return 0;
1324 }
1325 EXPORT_SYMBOL(bprm_change_interp);
1326
1327 /*
1328  * install the new credentials for this executable
1329  */
1330 void install_exec_creds(struct linux_binprm *bprm)
1331 {
1332         security_bprm_committing_creds(bprm);
1333
1334         commit_creds(bprm->cred);
1335         bprm->cred = NULL;
1336
1337         /*
1338          * Disable monitoring for regular users
1339          * when executing setuid binaries. Must
1340          * wait until new credentials are committed
1341          * by commit_creds() above
1342          */
1343         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1344                 perf_event_exit_task(current);
1345         /*
1346          * cred_guard_mutex must be held at least to this point to prevent
1347          * ptrace_attach() from altering our determination of the task's
1348          * credentials; any time after this it may be unlocked.
1349          */
1350         security_bprm_committed_creds(bprm);
1351         mutex_unlock(&current->signal->cred_guard_mutex);
1352 }
1353 EXPORT_SYMBOL(install_exec_creds);
1354
1355 /*
1356  * determine how safe it is to execute the proposed program
1357  * - the caller must hold ->cred_guard_mutex to protect against
1358  *   PTRACE_ATTACH or seccomp thread-sync
1359  */
1360 static void check_unsafe_exec(struct linux_binprm *bprm)
1361 {
1362         struct task_struct *p = current, *t;
1363         unsigned n_fs;
1364
1365         if (p->ptrace) {
1366                 if (p->ptrace & PT_PTRACE_CAP)
1367                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1368                 else
1369                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1370         }
1371
1372         /*
1373          * This isn't strictly necessary, but it makes it harder for LSMs to
1374          * mess up.
1375          */
1376         if (task_no_new_privs(current))
1377                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1378
1379         t = p;
1380         n_fs = 1;
1381         spin_lock(&p->fs->lock);
1382         rcu_read_lock();
1383         while_each_thread(p, t) {
1384                 if (t->fs == p->fs)
1385                         n_fs++;
1386         }
1387         rcu_read_unlock();
1388
1389         if (p->fs->users > n_fs)
1390                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1391         else
1392                 p->fs->in_exec = 1;
1393         spin_unlock(&p->fs->lock);
1394 }
1395
1396 static void bprm_fill_uid(struct linux_binprm *bprm)
1397 {
1398         struct inode *inode;
1399         unsigned int mode;
1400         kuid_t uid;
1401         kgid_t gid;
1402
1403         /*
1404          * Since this can be called multiple times (via prepare_binprm),
1405          * we must clear any previous work done when setting set[ug]id
1406          * bits from any earlier bprm->file uses (for example when run
1407          * first for a setuid script then again for its interpreter).
1408          */
1409         bprm->cred->euid = current_euid();
1410         bprm->cred->egid = current_egid();
1411
1412         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1413                 return;
1414
1415         if (task_no_new_privs(current))
1416                 return;
1417
1418         inode = file_inode(bprm->file);
1419         mode = READ_ONCE(inode->i_mode);
1420         if (!(mode & (S_ISUID|S_ISGID)))
1421                 return;
1422
1423         /* Be careful if suid/sgid is set */
1424         inode_lock(inode);
1425
1426         /* reload atomically mode/uid/gid now that lock held */
1427         mode = inode->i_mode;
1428         uid = inode->i_uid;
1429         gid = inode->i_gid;
1430         inode_unlock(inode);
1431
1432         /* We ignore suid/sgid if there are no mappings for them in the ns */
1433         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1434                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1435                 return;
1436
1437         if (mode & S_ISUID) {
1438                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1439                 bprm->cred->euid = uid;
1440         }
1441
1442         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1443                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1444                 bprm->cred->egid = gid;
1445         }
1446 }
1447
1448 /*
1449  * Fill the binprm structure from the inode.
1450  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1451  *
1452  * This may be called multiple times for binary chains (scripts for example).
1453  */
1454 int prepare_binprm(struct linux_binprm *bprm)
1455 {
1456         int retval;
1457
1458         bprm_fill_uid(bprm);
1459
1460         /* fill in binprm security blob */
1461         retval = security_bprm_set_creds(bprm);
1462         if (retval)
1463                 return retval;
1464         bprm->cred_prepared = 1;
1465
1466         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1467         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1468 }
1469
1470 EXPORT_SYMBOL(prepare_binprm);
1471
1472 /*
1473  * Arguments are '\0' separated strings found at the location bprm->p
1474  * points to; chop off the first by relocating brpm->p to right after
1475  * the first '\0' encountered.
1476  */
1477 int remove_arg_zero(struct linux_binprm *bprm)
1478 {
1479         int ret = 0;
1480         unsigned long offset;
1481         char *kaddr;
1482         struct page *page;
1483
1484         if (!bprm->argc)
1485                 return 0;
1486
1487         do {
1488                 offset = bprm->p & ~PAGE_MASK;
1489                 page = get_arg_page(bprm, bprm->p, 0);
1490                 if (!page) {
1491                         ret = -EFAULT;
1492                         goto out;
1493                 }
1494                 kaddr = kmap_atomic(page);
1495
1496                 for (; offset < PAGE_SIZE && kaddr[offset];
1497                                 offset++, bprm->p++)
1498                         ;
1499
1500                 kunmap_atomic(kaddr);
1501                 put_arg_page(page);
1502
1503                 if (offset == PAGE_SIZE)
1504                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1505         } while (offset == PAGE_SIZE);
1506
1507         bprm->p++;
1508         bprm->argc--;
1509         ret = 0;
1510
1511 out:
1512         return ret;
1513 }
1514 EXPORT_SYMBOL(remove_arg_zero);
1515
1516 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1517 /*
1518  * cycle the list of binary formats handler, until one recognizes the image
1519  */
1520 int search_binary_handler(struct linux_binprm *bprm)
1521 {
1522         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1523         struct linux_binfmt *fmt;
1524         int retval;
1525
1526         /* This allows 4 levels of binfmt rewrites before failing hard. */
1527         if (bprm->recursion_depth > 5)
1528                 return -ELOOP;
1529
1530         retval = security_bprm_check(bprm);
1531         if (retval)
1532                 return retval;
1533
1534         retval = -ENOENT;
1535  retry:
1536         read_lock(&binfmt_lock);
1537         list_for_each_entry(fmt, &formats, lh) {
1538                 if (!try_module_get(fmt->module))
1539                         continue;
1540                 read_unlock(&binfmt_lock);
1541                 bprm->recursion_depth++;
1542                 retval = fmt->load_binary(bprm);
1543                 read_lock(&binfmt_lock);
1544                 put_binfmt(fmt);
1545                 bprm->recursion_depth--;
1546                 if (retval < 0 && !bprm->mm) {
1547                         /* we got to flush_old_exec() and failed after it */
1548                         read_unlock(&binfmt_lock);
1549                         force_sigsegv(SIGSEGV, current);
1550                         return retval;
1551                 }
1552                 if (retval != -ENOEXEC || !bprm->file) {
1553                         read_unlock(&binfmt_lock);
1554                         return retval;
1555                 }
1556         }
1557         read_unlock(&binfmt_lock);
1558
1559         if (need_retry) {
1560                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1561                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1562                         return retval;
1563                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1564                         return retval;
1565                 need_retry = false;
1566                 goto retry;
1567         }
1568
1569         return retval;
1570 }
1571 EXPORT_SYMBOL(search_binary_handler);
1572
1573 static int exec_binprm(struct linux_binprm *bprm)
1574 {
1575         pid_t old_pid, old_vpid;
1576         int ret;
1577
1578         /* Need to fetch pid before load_binary changes it */
1579         old_pid = current->pid;
1580         rcu_read_lock();
1581         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1582         rcu_read_unlock();
1583
1584         ret = search_binary_handler(bprm);
1585         if (ret >= 0) {
1586                 audit_bprm(bprm);
1587                 trace_sched_process_exec(current, old_pid, bprm);
1588                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1589                 proc_exec_connector(current);
1590         }
1591
1592         return ret;
1593 }
1594
1595 /*
1596  * sys_execve() executes a new program.
1597  */
1598 static int do_execveat_common(int fd, struct filename *filename,
1599                               struct user_arg_ptr argv,
1600                               struct user_arg_ptr envp,
1601                               int flags)
1602 {
1603         char *pathbuf = NULL;
1604         struct linux_binprm *bprm;
1605         struct file *file;
1606         struct files_struct *displaced;
1607         int retval;
1608
1609         if (IS_ERR(filename))
1610                 return PTR_ERR(filename);
1611
1612         /*
1613          * We move the actual failure in case of RLIMIT_NPROC excess from
1614          * set*uid() to execve() because too many poorly written programs
1615          * don't check setuid() return code.  Here we additionally recheck
1616          * whether NPROC limit is still exceeded.
1617          */
1618         if ((current->flags & PF_NPROC_EXCEEDED) &&
1619             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1620                 retval = -EAGAIN;
1621                 goto out_ret;
1622         }
1623
1624         /* We're below the limit (still or again), so we don't want to make
1625          * further execve() calls fail. */
1626         current->flags &= ~PF_NPROC_EXCEEDED;
1627
1628         retval = unshare_files(&displaced);
1629         if (retval)
1630                 goto out_ret;
1631
1632         retval = -ENOMEM;
1633         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1634         if (!bprm)
1635                 goto out_files;
1636
1637         retval = prepare_bprm_creds(bprm);
1638         if (retval)
1639                 goto out_free;
1640
1641         check_unsafe_exec(bprm);
1642         current->in_execve = 1;
1643
1644         file = do_open_execat(fd, filename, flags);
1645         retval = PTR_ERR(file);
1646         if (IS_ERR(file))
1647                 goto out_unmark;
1648
1649         sched_exec();
1650
1651         bprm->file = file;
1652         if (fd == AT_FDCWD || filename->name[0] == '/') {
1653                 bprm->filename = filename->name;
1654         } else {
1655                 if (filename->name[0] == '\0')
1656                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1657                 else
1658                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1659                                             fd, filename->name);
1660                 if (!pathbuf) {
1661                         retval = -ENOMEM;
1662                         goto out_unmark;
1663                 }
1664                 /*
1665                  * Record that a name derived from an O_CLOEXEC fd will be
1666                  * inaccessible after exec. Relies on having exclusive access to
1667                  * current->files (due to unshare_files above).
1668                  */
1669                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1670                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1671                 bprm->filename = pathbuf;
1672         }
1673         bprm->interp = bprm->filename;
1674
1675         retval = bprm_mm_init(bprm);
1676         if (retval)
1677                 goto out_unmark;
1678
1679         bprm->argc = count(argv, MAX_ARG_STRINGS);
1680         if ((retval = bprm->argc) < 0)
1681                 goto out;
1682
1683         bprm->envc = count(envp, MAX_ARG_STRINGS);
1684         if ((retval = bprm->envc) < 0)
1685                 goto out;
1686
1687         retval = prepare_binprm(bprm);
1688         if (retval < 0)
1689                 goto out;
1690
1691         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1692         if (retval < 0)
1693                 goto out;
1694
1695         bprm->exec = bprm->p;
1696         retval = copy_strings(bprm->envc, envp, bprm);
1697         if (retval < 0)
1698                 goto out;
1699
1700         retval = copy_strings(bprm->argc, argv, bprm);
1701         if (retval < 0)
1702                 goto out;
1703
1704         retval = exec_binprm(bprm);
1705         if (retval < 0)
1706                 goto out;
1707
1708         /* execve succeeded */
1709         current->fs->in_exec = 0;
1710         current->in_execve = 0;
1711         acct_update_integrals(current);
1712         task_numa_free(current);
1713         free_bprm(bprm);
1714         kfree(pathbuf);
1715         putname(filename);
1716         if (displaced)
1717                 put_files_struct(displaced);
1718         return retval;
1719
1720 out:
1721         if (bprm->mm) {
1722                 acct_arg_size(bprm, 0);
1723                 mmput(bprm->mm);
1724         }
1725
1726 out_unmark:
1727         current->fs->in_exec = 0;
1728         current->in_execve = 0;
1729
1730 out_free:
1731         free_bprm(bprm);
1732         kfree(pathbuf);
1733
1734 out_files:
1735         if (displaced)
1736                 reset_files_struct(displaced);
1737 out_ret:
1738         putname(filename);
1739         return retval;
1740 }
1741
1742 int do_execve(struct filename *filename,
1743         const char __user *const __user *__argv,
1744         const char __user *const __user *__envp)
1745 {
1746         struct user_arg_ptr argv = { .ptr.native = __argv };
1747         struct user_arg_ptr envp = { .ptr.native = __envp };
1748         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1749 }
1750
1751 int do_execveat(int fd, struct filename *filename,
1752                 const char __user *const __user *__argv,
1753                 const char __user *const __user *__envp,
1754                 int flags)
1755 {
1756         struct user_arg_ptr argv = { .ptr.native = __argv };
1757         struct user_arg_ptr envp = { .ptr.native = __envp };
1758
1759         return do_execveat_common(fd, filename, argv, envp, flags);
1760 }
1761
1762 #ifdef CONFIG_COMPAT
1763 static int compat_do_execve(struct filename *filename,
1764         const compat_uptr_t __user *__argv,
1765         const compat_uptr_t __user *__envp)
1766 {
1767         struct user_arg_ptr argv = {
1768                 .is_compat = true,
1769                 .ptr.compat = __argv,
1770         };
1771         struct user_arg_ptr envp = {
1772                 .is_compat = true,
1773                 .ptr.compat = __envp,
1774         };
1775         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1776 }
1777
1778 static int compat_do_execveat(int fd, struct filename *filename,
1779                               const compat_uptr_t __user *__argv,
1780                               const compat_uptr_t __user *__envp,
1781                               int flags)
1782 {
1783         struct user_arg_ptr argv = {
1784                 .is_compat = true,
1785                 .ptr.compat = __argv,
1786         };
1787         struct user_arg_ptr envp = {
1788                 .is_compat = true,
1789                 .ptr.compat = __envp,
1790         };
1791         return do_execveat_common(fd, filename, argv, envp, flags);
1792 }
1793 #endif
1794
1795 void set_binfmt(struct linux_binfmt *new)
1796 {
1797         struct mm_struct *mm = current->mm;
1798
1799         if (mm->binfmt)
1800                 module_put(mm->binfmt->module);
1801
1802         mm->binfmt = new;
1803         if (new)
1804                 __module_get(new->module);
1805 }
1806 EXPORT_SYMBOL(set_binfmt);
1807
1808 /*
1809  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1810  */
1811 void set_dumpable(struct mm_struct *mm, int value)
1812 {
1813         unsigned long old, new;
1814
1815         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1816                 return;
1817
1818         do {
1819                 old = ACCESS_ONCE(mm->flags);
1820                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1821         } while (cmpxchg(&mm->flags, old, new) != old);
1822 }
1823
1824 SYSCALL_DEFINE3(execve,
1825                 const char __user *, filename,
1826                 const char __user *const __user *, argv,
1827                 const char __user *const __user *, envp)
1828 {
1829         return do_execve(getname(filename), argv, envp);
1830 }
1831
1832 SYSCALL_DEFINE5(execveat,
1833                 int, fd, const char __user *, filename,
1834                 const char __user *const __user *, argv,
1835                 const char __user *const __user *, envp,
1836                 int, flags)
1837 {
1838         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1839
1840         return do_execveat(fd,
1841                            getname_flags(filename, lookup_flags, NULL),
1842                            argv, envp, flags);
1843 }
1844
1845 #ifdef CONFIG_COMPAT
1846 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1847         const compat_uptr_t __user *, argv,
1848         const compat_uptr_t __user *, envp)
1849 {
1850         return compat_do_execve(getname(filename), argv, envp);
1851 }
1852
1853 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1854                        const char __user *, filename,
1855                        const compat_uptr_t __user *, argv,
1856                        const compat_uptr_t __user *, envp,
1857                        int,  flags)
1858 {
1859         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1860
1861         return compat_do_execveat(fd,
1862                                   getname_flags(filename, lookup_flags, NULL),
1863                                   argv, envp, flags);
1864 }
1865 #endif