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_pages(struct linux_binprm *bprm)
 247 {
 248 }
 249
 250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 251                 struct page *page)
 252 {
 253         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 254 }
 255
 256 static int __bprm_mm_init(struct linux_binprm *bprm)
 257 {
 258         int err;
 259         struct vm_area_struct *vma = NULL;
 260         struct mm_struct *mm = bprm->mm;
 261
 262         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 263         if (!vma)
 264                 return -ENOMEM;
 265
 266         if (down_write_killable(&mm->mmap_sem)) {
 267                 err = -EINTR;
 268                 goto err_free;
 269         }
 270         vma->vm_mm = mm;
 271
 272         /*
 273          * Place the stack at the largest stack address the architecture
 274          * supports. Later, we'll move this to an appropriate place. We don't
 275          * use STACK_TOP because that can depend on attributes which aren't
 276          * configured yet.
 277          */
 278         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 279         vma->vm_end = STACK_TOP_MAX;
 280         vma->vm_start = vma->vm_end - PAGE_SIZE;
 281         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 282         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 283         INIT_LIST_HEAD(&vma->anon_vma_chain);
 284
 285         err = insert_vm_struct(mm, vma);
 286         if (err)
 287                 goto err;
 288
 289         mm->stack_vm = mm->total_vm = 1;
 290         arch_bprm_mm_init(mm, vma);
 291         up_write(&mm->mmap_sem);
 292         bprm->p = vma->vm_end - sizeof(void *);
 293         return 0;
 294 err:
 295         up_write(&mm->mmap_sem);
 296 err_free:
 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         if (down_write_killable(&mm->mmap_sem))
 704                 return -EINTR;
 705
 706         vm_flags = VM_STACK_FLAGS;
 707
 708         /*
 709          * Adjust stack execute permissions; explicitly enable for
 710          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 711          * (arch default) otherwise.
 712          */
 713         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 714                 vm_flags |= VM_EXEC;
 715         else if (executable_stack == EXSTACK_DISABLE_X)
 716                 vm_flags &= ~VM_EXEC;
 717         vm_flags |= mm->def_flags;
 718         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 719
 720         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 721                         vm_flags);
 722         if (ret)
 723                 goto out_unlock;
 724         BUG_ON(prev != vma);
 725
 726         /* Move stack pages down in memory. */
 727         if (stack_shift) {
 728                 ret = shift_arg_pages(vma, stack_shift);
 729                 if (ret)
 730                         goto out_unlock;
 731         }
 732
 733         /* mprotect_fixup is overkill to remove the temporary stack flags */
 734         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 735
 736         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 737         stack_size = vma->vm_end - vma->vm_start;
 738         /*
 739          * Align this down to a page boundary as expand_stack
 740          * will align it up.
 741          */
 742         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 743 #ifdef CONFIG_STACK_GROWSUP
 744         if (stack_size + stack_expand > rlim_stack)
 745                 stack_base = vma->vm_start + rlim_stack;
 746         else
 747                 stack_base = vma->vm_end + stack_expand;
 748 #else
 749         if (stack_size + stack_expand > rlim_stack)
 750                 stack_base = vma->vm_end - rlim_stack;
 751         else
 752                 stack_base = vma->vm_start - stack_expand;
 753 #endif
 754         current->mm->start_stack = bprm->p;
 755         ret = expand_stack(vma, stack_base);
 756         if (ret)
 757                 ret = -EFAULT;
 758
 759 out_unlock:
 760         up_write(&mm->mmap_sem);
 761         return ret;
 762 }
 763 EXPORT_SYMBOL(setup_arg_pages);
 764
 765 #endif /* CONFIG_MMU */
 766
 767 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 768 {
 769         struct file *file;
 770         int err;
 771         struct open_flags open_exec_flags = {
 772                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 773                 .acc_mode = MAY_EXEC,
 774                 .intent = LOOKUP_OPEN,
 775                 .lookup_flags = LOOKUP_FOLLOW,
 776         };
 777
 778         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 779                 return ERR_PTR(-EINVAL);
 780         if (flags & AT_SYMLINK_NOFOLLOW)
 781                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 782         if (flags & AT_EMPTY_PATH)
 783                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 784
 785         file = do_filp_open(fd, name, &open_exec_flags);
 786         if (IS_ERR(file))
 787                 goto out;
 788
 789         err = -EACCES;
 790         if (!S_ISREG(file_inode(file)->i_mode))
 791                 goto exit;
 792
 793         if (path_noexec(&file->f_path))
 794                 goto exit;
 795
 796         err = deny_write_access(file);
 797         if (err)
 798                 goto exit;
 799
 800         if (name->name[0] != '\0')
 801                 fsnotify_open(file);
 802
 803 out:
 804         return file;
 805
 806 exit:
 807         fput(file);
 808         return ERR_PTR(err);
 809 }
 810
 811 struct file *open_exec(const char *name)
 812 {
 813         struct filename *filename = getname_kernel(name);
 814         struct file *f = ERR_CAST(filename);
 815
 816         if (!IS_ERR(filename)) {
 817                 f = do_open_execat(AT_FDCWD, filename, 0);
 818                 putname(filename);
 819         }
 820         return f;
 821 }
 822 EXPORT_SYMBOL(open_exec);
 823
 824 int kernel_read(struct file *file, loff_t offset,
 825                 char *addr, unsigned long count)
 826 {
 827         mm_segment_t old_fs;
 828         loff_t pos = offset;
 829         int result;
 830
 831         old_fs = get_fs();
 832         set_fs(get_ds());
 833         /* The cast to a user pointer is valid due to the set_fs() */
 834         result = vfs_read(file, (void __user *)addr, count, &pos);
 835         set_fs(old_fs);
 836         return result;
 837 }
 838
 839 EXPORT_SYMBOL(kernel_read);
 840
 841 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 842                      loff_t max_size, enum kernel_read_file_id id)
 843 {
 844         loff_t i_size, pos;
 845         ssize_t bytes = 0;
 846         int ret;
 847
 848         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 849                 return -EINVAL;
 850
 851         ret = security_kernel_read_file(file, id);
 852         if (ret)
 853                 return ret;
 854
 855         ret = deny_write_access(file);
 856         if (ret)
 857                 return ret;
 858
 859         i_size = i_size_read(file_inode(file));
 860         if (max_size > 0 && i_size > max_size) {
 861                 ret = -EFBIG;
 862                 goto out;
 863         }
 864         if (i_size <= 0) {
 865                 ret = -EINVAL;
 866                 goto out;
 867         }
 868
 869         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 870                 *buf = vmalloc(i_size);
 871         if (!*buf) {
 872                 ret = -ENOMEM;
 873                 goto out;
 874         }
 875
 876         pos = 0;
 877         while (pos < i_size) {
 878                 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 879                                     i_size - pos);
 880                 if (bytes < 0) {
 881                         ret = bytes;
 882                         goto out;
 883                 }
 884
 885                 if (bytes == 0)
 886                         break;
 887                 pos += bytes;
 888         }
 889
 890         if (pos != i_size) {
 891                 ret = -EIO;
 892                 goto out_free;
 893         }
 894
 895         ret = security_kernel_post_read_file(file, *buf, i_size, id);
 896         if (!ret)
 897                 *size = pos;
 898
 899 out_free:
 900         if (ret < 0) {
 901                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 902                         vfree(*buf);
 903                         *buf = NULL;
 904                 }
 905         }
 906
 907 out:
 908         allow_write_access(file);
 909         return ret;
 910 }
 911 EXPORT_SYMBOL_GPL(kernel_read_file);
 912
 913 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 914                                loff_t max_size, enum kernel_read_file_id id)
 915 {
 916         struct file *file;
 917         int ret;
 918
 919         if (!path || !*path)
 920                 return -EINVAL;
 921
 922         file = filp_open(path, O_RDONLY, 0);
 923         if (IS_ERR(file))
 924                 return PTR_ERR(file);
 925
 926         ret = kernel_read_file(file, buf, size, max_size, id);
 927         fput(file);
 928         return ret;
 929 }
 930 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 931
 932 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 933                              enum kernel_read_file_id id)
 934 {
 935         struct fd f = fdget(fd);
 936         int ret = -EBADF;
 937
 938         if (!f.file)
 939                 goto out;
 940
 941         ret = kernel_read_file(f.file, buf, size, max_size, id);
 942 out:
 943         fdput(f);
 944         return ret;
 945 }
 946 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 947
 948 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 949 {
 950         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 951         if (res > 0)
 952                 flush_icache_range(addr, addr + len);
 953         return res;
 954 }
 955 EXPORT_SYMBOL(read_code);
 956
 957 static int exec_mmap(struct mm_struct *mm)
 958 {
 959         struct task_struct *tsk;
 960         struct mm_struct *old_mm, *active_mm;
 961
 962         /* Notify parent that we're no longer interested in the old VM */
 963         tsk = current;
 964         old_mm = current->mm;
 965         mm_release(tsk, old_mm);
 966
 967         if (old_mm) {
 968                 sync_mm_rss(old_mm);
 969                 /*
 970                  * Make sure that if there is a core dump in progress
 971                  * for the old mm, we get out and die instead of going
 972                  * through with the exec.  We must hold mmap_sem around
 973                  * checking core_state and changing tsk->mm.
 974                  */
 975                 down_read(&old_mm->mmap_sem);
 976                 if (unlikely(old_mm->core_state)) {
 977                         up_read(&old_mm->mmap_sem);
 978                         return -EINTR;
 979                 }
 980         }
 981         task_lock(tsk);
 982         active_mm = tsk->active_mm;
 983         tsk->mm = mm;
 984         tsk->active_mm = mm;
 985         activate_mm(active_mm, mm);
 986         tsk->mm->vmacache_seqnum = 0;
 987         vmacache_flush(tsk);
 988         task_unlock(tsk);
 989         if (old_mm) {
 990                 up_read(&old_mm->mmap_sem);
 991                 BUG_ON(active_mm != old_mm);
 992                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 993                 mm_update_next_owner(old_mm);
 994                 mmput(old_mm);
 995                 return 0;
 996         }
 997         mmdrop(active_mm);
 998         return 0;
 999 }
1000
1001 /*
1002  * This function makes sure the current process has its own signal table,
1003  * so that flush_signal_handlers can later reset the handlers without
1004  * disturbing other processes.  (Other processes might share the signal
1005  * table via the CLONE_SIGHAND option to clone().)
1006  */
1007 static int de_thread(struct task_struct *tsk)
1008 {
1009         struct signal_struct *sig = tsk->signal;
1010         struct sighand_struct *oldsighand = tsk->sighand;
1011         spinlock_t *lock = &oldsighand->siglock;
1012
1013         if (thread_group_empty(tsk))
1014                 goto no_thread_group;
1015
1016         /*
1017          * Kill all other threads in the thread group.
1018          */
1019         spin_lock_irq(lock);
1020         if (signal_group_exit(sig)) {
1021                 /*
1022                  * Another group action in progress, just
1023                  * return so that the signal is processed.
1024                  */
1025                 spin_unlock_irq(lock);
1026                 return -EAGAIN;
1027         }
1028
1029         sig->group_exit_task = tsk;
1030         sig->notify_count = zap_other_threads(tsk);
1031         if (!thread_group_leader(tsk))
1032                 sig->notify_count--;
1033
1034         while (sig->notify_count) {
1035                 __set_current_state(TASK_KILLABLE);
1036                 spin_unlock_irq(lock);
1037                 schedule();
1038                 if (unlikely(__fatal_signal_pending(tsk)))
1039                         goto killed;
1040                 spin_lock_irq(lock);
1041         }
1042         spin_unlock_irq(lock);
1043
1044         /*
1045          * At this point all other threads have exited, all we have to
1046          * do is to wait for the thread group leader to become inactive,
1047          * and to assume its PID:
1048          */
1049         if (!thread_group_leader(tsk)) {
1050                 struct task_struct *leader = tsk->group_leader;
1051
1052                 for (;;) {
1053                         threadgroup_change_begin(tsk);
1054                         write_lock_irq(&tasklist_lock);
1055                         /*
1056                          * Do this under tasklist_lock to ensure that
1057                          * exit_notify() can't miss ->group_exit_task
1058                          */
1059                         sig->notify_count = -1;
1060                         if (likely(leader->exit_state))
1061                                 break;
1062                         __set_current_state(TASK_KILLABLE);
1063                         write_unlock_irq(&tasklist_lock);
1064                         threadgroup_change_end(tsk);
1065                         schedule();
1066                         if (unlikely(__fatal_signal_pending(tsk)))
1067                                 goto killed;
1068                 }
1069
1070                 /*
1071                  * The only record we have of the real-time age of a
1072                  * process, regardless of execs it's done, is start_time.
1073                  * All the past CPU time is accumulated in signal_struct
1074                  * from sister threads now dead.  But in this non-leader
1075                  * exec, nothing survives from the original leader thread,
1076                  * whose birth marks the true age of this process now.
1077                  * When we take on its identity by switching to its PID, we
1078                  * also take its birthdate (always earlier than our own).
1079                  */
1080                 tsk->start_time = leader->start_time;
1081                 tsk->real_start_time = leader->real_start_time;
1082
1083                 BUG_ON(!same_thread_group(leader, tsk));
1084                 BUG_ON(has_group_leader_pid(tsk));
1085                 /*
1086                  * An exec() starts a new thread group with the
1087                  * TGID of the previous thread group. Rehash the
1088                  * two threads with a switched PID, and release
1089                  * the former thread group leader:
1090                  */
1091
1092                 /* Become a process group leader with the old leader's pid.
1093                  * The old leader becomes a thread of the this thread group.
1094                  * Note: The old leader also uses this pid until release_task
1095                  *       is called.  Odd but simple and correct.
1096                  */
1097                 tsk->pid = leader->pid;
1098                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1099                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1100                 transfer_pid(leader, tsk, PIDTYPE_SID);
1101
1102                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1103                 list_replace_init(&leader->sibling, &tsk->sibling);
1104
1105                 tsk->group_leader = tsk;
1106                 leader->group_leader = tsk;
1107
1108                 tsk->exit_signal = SIGCHLD;
1109                 leader->exit_signal = -1;
1110
1111                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1112                 leader->exit_state = EXIT_DEAD;
1113
1114                 /*
1115                  * We are going to release_task()->ptrace_unlink() silently,
1116                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1117                  * the tracer wont't block again waiting for this thread.
1118                  */
1119                 if (unlikely(leader->ptrace))
1120                         __wake_up_parent(leader, leader->parent);
1121                 write_unlock_irq(&tasklist_lock);
1122                 threadgroup_change_end(tsk);
1123
1124                 release_task(leader);
1125         }
1126
1127         sig->group_exit_task = NULL;
1128         sig->notify_count = 0;
1129
1130 no_thread_group:
1131         /* we have changed execution domain */
1132         tsk->exit_signal = SIGCHLD;
1133
1134         exit_itimers(sig);
1135         flush_itimer_signals();
1136
1137         if (atomic_read(&oldsighand->count) != 1) {
1138                 struct sighand_struct *newsighand;
1139                 /*
1140                  * This ->sighand is shared with the CLONE_SIGHAND
1141                  * but not CLONE_THREAD task, switch to the new one.
1142                  */
1143                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1144                 if (!newsighand)
1145                         return -ENOMEM;
1146
1147                 atomic_set(&newsighand->count, 1);
1148                 memcpy(newsighand->action, oldsighand->action,
1149                        sizeof(newsighand->action));
1150
1151                 write_lock_irq(&tasklist_lock);
1152                 spin_lock(&oldsighand->siglock);
1153                 rcu_assign_pointer(tsk->sighand, newsighand);
1154                 spin_unlock(&oldsighand->siglock);
1155                 write_unlock_irq(&tasklist_lock);
1156
1157                 __cleanup_sighand(oldsighand);
1158         }
1159
1160         BUG_ON(!thread_group_leader(tsk));
1161         return 0;
1162
1163 killed:
1164         /* protects against exit_notify() and __exit_signal() */
1165         read_lock(&tasklist_lock);
1166         sig->group_exit_task = NULL;
1167         sig->notify_count = 0;
1168         read_unlock(&tasklist_lock);
1169         return -EAGAIN;
1170 }
1171
1172 char *get_task_comm(char *buf, struct task_struct *tsk)
1173 {
1174         /* buf must be at least sizeof(tsk->comm) in size */
1175         task_lock(tsk);
1176         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1177         task_unlock(tsk);
1178         return buf;
1179 }
1180 EXPORT_SYMBOL_GPL(get_task_comm);
1181
1182 /*
1183  * These functions flushes out all traces of the currently running executable
1184  * so that a new one can be started
1185  */
1186
1187 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1188 {
1189         task_lock(tsk);
1190         trace_task_rename(tsk, buf);
1191         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1192         task_unlock(tsk);
1193         perf_event_comm(tsk, exec);
1194 }
1195
1196 int flush_old_exec(struct linux_binprm * bprm)
1197 {
1198         int retval;
1199
1200         /*
1201          * Make sure we have a private signal table and that
1202          * we are unassociated from the previous thread group.
1203          */
1204         retval = de_thread(current);
1205         if (retval)
1206                 goto out;
1207
1208         /*
1209          * Must be called _before_ exec_mmap() as bprm->mm is
1210          * not visibile until then. This also enables the update
1211          * to be lockless.
1212          */
1213         set_mm_exe_file(bprm->mm, bprm->file);
1214
1215         /*
1216          * Release all of the old mmap stuff
1217          */
1218         acct_arg_size(bprm, 0);
1219         retval = exec_mmap(bprm->mm);
1220         if (retval)
1221                 goto out;
1222
1223         bprm->mm = NULL;                /* We're using it now */
1224
1225         set_fs(USER_DS);
1226         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1227                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1228         flush_thread();
1229         current->personality &= ~bprm->per_clear;
1230
1231         return 0;
1232
1233 out:
1234         return retval;
1235 }
1236 EXPORT_SYMBOL(flush_old_exec);
1237
1238 void would_dump(struct linux_binprm *bprm, struct file *file)
1239 {
1240         if (inode_permission(file_inode(file), MAY_READ) < 0)
1241                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1242 }
1243 EXPORT_SYMBOL(would_dump);
1244
1245 void setup_new_exec(struct linux_binprm * bprm)
1246 {
1247         arch_pick_mmap_layout(current->mm);
1248
1249         /* This is the point of no return */
1250         current->sas_ss_sp = current->sas_ss_size = 0;
1251
1252         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1253                 set_dumpable(current->mm, SUID_DUMP_USER);
1254         else
1255                 set_dumpable(current->mm, suid_dumpable);
1256
1257         perf_event_exec();
1258         __set_task_comm(current, kbasename(bprm->filename), true);
1259
1260         /* Set the new mm task size. We have to do that late because it may
1261          * depend on TIF_32BIT which is only updated in flush_thread() on
1262          * some architectures like powerpc
1263          */
1264         current->mm->task_size = TASK_SIZE;
1265
1266         /* install the new credentials */
1267         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1268             !gid_eq(bprm->cred->gid, current_egid())) {
1269                 current->pdeath_signal = 0;
1270         } else {
1271                 would_dump(bprm, bprm->file);
1272                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1273                         set_dumpable(current->mm, suid_dumpable);
1274         }
1275
1276         /* An exec changes our domain. We are no longer part of the thread
1277            group */
1278         current->self_exec_id++;
1279         flush_signal_handlers(current, 0);
1280         do_close_on_exec(current->files);
1281 }
1282 EXPORT_SYMBOL(setup_new_exec);
1283
1284 /*
1285  * Prepare credentials and lock ->cred_guard_mutex.
1286  * install_exec_creds() commits the new creds and drops the lock.
1287  * Or, if exec fails before, free_bprm() should release ->cred and
1288  * and unlock.
1289  */
1290 int prepare_bprm_creds(struct linux_binprm *bprm)
1291 {
1292         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1293                 return -ERESTARTNOINTR;
1294
1295         bprm->cred = prepare_exec_creds();
1296         if (likely(bprm->cred))
1297                 return 0;
1298
1299         mutex_unlock(&current->signal->cred_guard_mutex);
1300         return -ENOMEM;
1301 }
1302
1303 static void free_bprm(struct linux_binprm *bprm)
1304 {
1305         free_arg_pages(bprm);
1306         if (bprm->cred) {
1307                 mutex_unlock(&current->signal->cred_guard_mutex);
1308                 abort_creds(bprm->cred);
1309         }
1310         if (bprm->file) {
1311                 allow_write_access(bprm->file);
1312                 fput(bprm->file);
1313         }
1314         /* If a binfmt changed the interp, free it. */
1315         if (bprm->interp != bprm->filename)
1316                 kfree(bprm->interp);
1317         kfree(bprm);
1318 }
1319
1320 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1321 {
1322         /* If a binfmt changed the interp, free it first. */
1323         if (bprm->interp != bprm->filename)
1324                 kfree(bprm->interp);
1325         bprm->interp = kstrdup(interp, GFP_KERNEL);
1326         if (!bprm->interp)
1327                 return -ENOMEM;
1328         return 0;
1329 }
1330 EXPORT_SYMBOL(bprm_change_interp);
1331
1332 /*
1333  * install the new credentials for this executable
1334  */
1335 void install_exec_creds(struct linux_binprm *bprm)
1336 {
1337         security_bprm_committing_creds(bprm);
1338
1339         commit_creds(bprm->cred);
1340         bprm->cred = NULL;
1341
1342         /*
1343          * Disable monitoring for regular users
1344          * when executing setuid binaries. Must
1345          * wait until new credentials are committed
1346          * by commit_creds() above
1347          */
1348         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1349                 perf_event_exit_task(current);
1350         /*
1351          * cred_guard_mutex must be held at least to this point to prevent
1352          * ptrace_attach() from altering our determination of the task's
1353          * credentials; any time after this it may be unlocked.
1354          */
1355         security_bprm_committed_creds(bprm);
1356         mutex_unlock(&current->signal->cred_guard_mutex);
1357 }
1358 EXPORT_SYMBOL(install_exec_creds);
1359
1360 /*
1361  * determine how safe it is to execute the proposed program
1362  * - the caller must hold ->cred_guard_mutex to protect against
1363  *   PTRACE_ATTACH or seccomp thread-sync
1364  */
1365 static void check_unsafe_exec(struct linux_binprm *bprm)
1366 {
1367         struct task_struct *p = current, *t;
1368         unsigned n_fs;
1369
1370         if (p->ptrace) {
1371                 if (p->ptrace & PT_PTRACE_CAP)
1372                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1373                 else
1374                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1375         }
1376
1377         /*
1378          * This isn't strictly necessary, but it makes it harder for LSMs to
1379          * mess up.
1380          */
1381         if (task_no_new_privs(current))
1382                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1383
1384         t = p;
1385         n_fs = 1;
1386         spin_lock(&p->fs->lock);
1387         rcu_read_lock();
1388         while_each_thread(p, t) {
1389                 if (t->fs == p->fs)
1390                         n_fs++;
1391         }
1392         rcu_read_unlock();
1393
1394         if (p->fs->users > n_fs)
1395                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1396         else
1397                 p->fs->in_exec = 1;
1398         spin_unlock(&p->fs->lock);
1399 }
1400
1401 static void bprm_fill_uid(struct linux_binprm *bprm)
1402 {
1403         struct inode *inode;
1404         unsigned int mode;
1405         kuid_t uid;
1406         kgid_t gid;
1407
1408         /*
1409          * Since this can be called multiple times (via prepare_binprm),
1410          * we must clear any previous work done when setting set[ug]id
1411          * bits from any earlier bprm->file uses (for example when run
1412          * first for a setuid script then again for its interpreter).
1413          */
1414         bprm->cred->euid = current_euid();
1415         bprm->cred->egid = current_egid();
1416
1417         if (!mnt_may_suid(bprm->file->f_path.mnt))
1418                 return;
1419
1420         if (task_no_new_privs(current))
1421                 return;
1422
1423         inode = file_inode(bprm->file);
1424         mode = READ_ONCE(inode->i_mode);
1425         if (!(mode & (S_ISUID|S_ISGID)))
1426                 return;
1427
1428         /* Be careful if suid/sgid is set */
1429         inode_lock(inode);
1430
1431         /* reload atomically mode/uid/gid now that lock held */
1432         mode = inode->i_mode;
1433         uid = inode->i_uid;
1434         gid = inode->i_gid;
1435         inode_unlock(inode);
1436
1437         /* We ignore suid/sgid if there are no mappings for them in the ns */
1438         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1439                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1440                 return;
1441
1442         if (mode & S_ISUID) {
1443                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1444                 bprm->cred->euid = uid;
1445         }
1446
1447         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1448                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1449                 bprm->cred->egid = gid;
1450         }
1451 }
1452
1453 /*
1454  * Fill the binprm structure from the inode.
1455  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1456  *
1457  * This may be called multiple times for binary chains (scripts for example).
1458  */
1459 int prepare_binprm(struct linux_binprm *bprm)
1460 {
1461         int retval;
1462
1463         bprm_fill_uid(bprm);
1464
1465         /* fill in binprm security blob */
1466         retval = security_bprm_set_creds(bprm);
1467         if (retval)
1468                 return retval;
1469         bprm->cred_prepared = 1;
1470
1471         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1472         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1473 }
1474
1475 EXPORT_SYMBOL(prepare_binprm);
1476
1477 /*
1478  * Arguments are '\0' separated strings found at the location bprm->p
1479  * points to; chop off the first by relocating brpm->p to right after
1480  * the first '\0' encountered.
1481  */
1482 int remove_arg_zero(struct linux_binprm *bprm)
1483 {
1484         int ret = 0;
1485         unsigned long offset;
1486         char *kaddr;
1487         struct page *page;
1488
1489         if (!bprm->argc)
1490                 return 0;
1491
1492         do {
1493                 offset = bprm->p & ~PAGE_MASK;
1494                 page = get_arg_page(bprm, bprm->p, 0);
1495                 if (!page) {
1496                         ret = -EFAULT;
1497                         goto out;
1498                 }
1499                 kaddr = kmap_atomic(page);
1500
1501                 for (; offset < PAGE_SIZE && kaddr[offset];
1502                                 offset++, bprm->p++)
1503                         ;
1504
1505                 kunmap_atomic(kaddr);
1506                 put_arg_page(page);
1507         } while (offset == PAGE_SIZE);
1508
1509         bprm->p++;
1510         bprm->argc--;
1511         ret = 0;
1512
1513 out:
1514         return ret;
1515 }
1516 EXPORT_SYMBOL(remove_arg_zero);
1517
1518 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1519 /*
1520  * cycle the list of binary formats handler, until one recognizes the image
1521  */
1522 int search_binary_handler(struct linux_binprm *bprm)
1523 {
1524         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1525         struct linux_binfmt *fmt;
1526         int retval;
1527
1528         /* This allows 4 levels of binfmt rewrites before failing hard. */
1529         if (bprm->recursion_depth > 5)
1530                 return -ELOOP;
1531
1532         retval = security_bprm_check(bprm);
1533         if (retval)
1534                 return retval;
1535
1536         retval = -ENOENT;
1537  retry:
1538         read_lock(&binfmt_lock);
1539         list_for_each_entry(fmt, &formats, lh) {
1540                 if (!try_module_get(fmt->module))
1541                         continue;
1542                 read_unlock(&binfmt_lock);
1543                 bprm->recursion_depth++;
1544                 retval = fmt->load_binary(bprm);
1545                 read_lock(&binfmt_lock);
1546                 put_binfmt(fmt);
1547                 bprm->recursion_depth--;
1548                 if (retval < 0 && !bprm->mm) {
1549                         /* we got to flush_old_exec() and failed after it */
1550                         read_unlock(&binfmt_lock);
1551                         force_sigsegv(SIGSEGV, current);
1552                         return retval;
1553                 }
1554                 if (retval != -ENOEXEC || !bprm->file) {
1555                         read_unlock(&binfmt_lock);
1556                         return retval;
1557                 }
1558         }
1559         read_unlock(&binfmt_lock);
1560
1561         if (need_retry) {
1562                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1563                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1564                         return retval;
1565                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1566                         return retval;
1567                 need_retry = false;
1568                 goto retry;
1569         }
1570
1571         return retval;
1572 }
1573 EXPORT_SYMBOL(search_binary_handler);
1574
1575 static int exec_binprm(struct linux_binprm *bprm)
1576 {
1577         pid_t old_pid, old_vpid;
1578         int ret;
1579
1580         /* Need to fetch pid before load_binary changes it */
1581         old_pid = current->pid;
1582         rcu_read_lock();
1583         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1584         rcu_read_unlock();
1585
1586         ret = search_binary_handler(bprm);
1587         if (ret >= 0) {
1588                 audit_bprm(bprm);
1589                 trace_sched_process_exec(current, old_pid, bprm);
1590                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1591                 proc_exec_connector(current);
1592         }
1593
1594         return ret;
1595 }
1596
1597 /*
1598  * sys_execve() executes a new program.
1599  */
1600 static int do_execveat_common(int fd, struct filename *filename,
1601                               struct user_arg_ptr argv,
1602                               struct user_arg_ptr envp,
1603                               int flags)
1604 {
1605         char *pathbuf = NULL;
1606         struct linux_binprm *bprm;
1607         struct file *file;
1608         struct files_struct *displaced;
1609         int retval;
1610
1611         if (IS_ERR(filename))
1612                 return PTR_ERR(filename);
1613
1614         /*
1615          * We move the actual failure in case of RLIMIT_NPROC excess from
1616          * set*uid() to execve() because too many poorly written programs
1617          * don't check setuid() return code.  Here we additionally recheck
1618          * whether NPROC limit is still exceeded.
1619          */
1620         if ((current->flags & PF_NPROC_EXCEEDED) &&
1621             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1622                 retval = -EAGAIN;
1623                 goto out_ret;
1624         }
1625
1626         /* We're below the limit (still or again), so we don't want to make
1627          * further execve() calls fail. */
1628         current->flags &= ~PF_NPROC_EXCEEDED;
1629
1630         retval = unshare_files(&displaced);
1631         if (retval)
1632                 goto out_ret;
1633
1634         retval = -ENOMEM;
1635         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1636         if (!bprm)
1637                 goto out_files;
1638
1639         retval = prepare_bprm_creds(bprm);
1640         if (retval)
1641                 goto out_free;
1642
1643         check_unsafe_exec(bprm);
1644         current->in_execve = 1;
1645
1646         file = do_open_execat(fd, filename, flags);
1647         retval = PTR_ERR(file);
1648         if (IS_ERR(file))
1649                 goto out_unmark;
1650
1651         sched_exec();
1652
1653         bprm->file = file;
1654         if (fd == AT_FDCWD || filename->name[0] == '/') {
1655                 bprm->filename = filename->name;
1656         } else {
1657                 if (filename->name[0] == '\0')
1658                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1659                 else
1660                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1661                                             fd, filename->name);
1662                 if (!pathbuf) {
1663                         retval = -ENOMEM;
1664                         goto out_unmark;
1665                 }
1666                 /*
1667                  * Record that a name derived from an O_CLOEXEC fd will be
1668                  * inaccessible after exec. Relies on having exclusive access to
1669                  * current->files (due to unshare_files above).
1670                  */
1671                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1672                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1673                 bprm->filename = pathbuf;
1674         }
1675         bprm->interp = bprm->filename;
1676
1677         retval = bprm_mm_init(bprm);
1678         if (retval)
1679                 goto out_unmark;
1680
1681         bprm->argc = count(argv, MAX_ARG_STRINGS);
1682         if ((retval = bprm->argc) < 0)
1683                 goto out;
1684
1685         bprm->envc = count(envp, MAX_ARG_STRINGS);
1686         if ((retval = bprm->envc) < 0)
1687                 goto out;
1688
1689         retval = prepare_binprm(bprm);
1690         if (retval < 0)
1691                 goto out;
1692
1693         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1694         if (retval < 0)
1695                 goto out;
1696
1697         bprm->exec = bprm->p;
1698         retval = copy_strings(bprm->envc, envp, bprm);
1699         if (retval < 0)
1700                 goto out;
1701
1702         retval = copy_strings(bprm->argc, argv, bprm);
1703         if (retval < 0)
1704                 goto out;
1705
1706         retval = exec_binprm(bprm);
1707         if (retval < 0)
1708                 goto out;
1709
1710         /* execve succeeded */
1711         current->fs->in_exec = 0;
1712         current->in_execve = 0;
1713         acct_update_integrals(current);
1714         task_numa_free(current);
1715         free_bprm(bprm);
1716         kfree(pathbuf);
1717         putname(filename);
1718         if (displaced)
1719                 put_files_struct(displaced);
1720         return retval;
1721
1722 out:
1723         if (bprm->mm) {
1724                 acct_arg_size(bprm, 0);
1725                 mmput(bprm->mm);
1726         }
1727
1728 out_unmark:
1729         current->fs->in_exec = 0;
1730         current->in_execve = 0;
1731
1732 out_free:
1733         free_bprm(bprm);
1734         kfree(pathbuf);
1735
1736 out_files:
1737         if (displaced)
1738                 reset_files_struct(displaced);
1739 out_ret:
1740         putname(filename);
1741         return retval;
1742 }
1743
1744 int do_execve(struct filename *filename,
1745         const char __user *const __user *__argv,
1746         const char __user *const __user *__envp)
1747 {
1748         struct user_arg_ptr argv = { .ptr.native = __argv };
1749         struct user_arg_ptr envp = { .ptr.native = __envp };
1750         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1751 }
1752
1753 int do_execveat(int fd, struct filename *filename,
1754                 const char __user *const __user *__argv,
1755                 const char __user *const __user *__envp,
1756                 int flags)
1757 {
1758         struct user_arg_ptr argv = { .ptr.native = __argv };
1759         struct user_arg_ptr envp = { .ptr.native = __envp };
1760
1761         return do_execveat_common(fd, filename, argv, envp, flags);
1762 }
1763
1764 #ifdef CONFIG_COMPAT
1765 static int compat_do_execve(struct filename *filename,
1766         const compat_uptr_t __user *__argv,
1767         const compat_uptr_t __user *__envp)
1768 {
1769         struct user_arg_ptr argv = {
1770                 .is_compat = true,
1771                 .ptr.compat = __argv,
1772         };
1773         struct user_arg_ptr envp = {
1774                 .is_compat = true,
1775                 .ptr.compat = __envp,
1776         };
1777         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1778 }
1779
1780 static int compat_do_execveat(int fd, struct filename *filename,
1781                               const compat_uptr_t __user *__argv,
1782                               const compat_uptr_t __user *__envp,
1783                               int flags)
1784 {
1785         struct user_arg_ptr argv = {
1786                 .is_compat = true,
1787                 .ptr.compat = __argv,
1788         };
1789         struct user_arg_ptr envp = {
1790                 .is_compat = true,
1791                 .ptr.compat = __envp,
1792         };
1793         return do_execveat_common(fd, filename, argv, envp, flags);
1794 }
1795 #endif
1796
1797 void set_binfmt(struct linux_binfmt *new)
1798 {
1799         struct mm_struct *mm = current->mm;
1800
1801         if (mm->binfmt)
1802                 module_put(mm->binfmt->module);
1803
1804         mm->binfmt = new;
1805         if (new)
1806                 __module_get(new->module);
1807 }
1808 EXPORT_SYMBOL(set_binfmt);
1809
1810 /*
1811  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1812  */
1813 void set_dumpable(struct mm_struct *mm, int value)
1814 {
1815         unsigned long old, new;
1816
1817         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1818                 return;
1819
1820         do {
1821                 old = ACCESS_ONCE(mm->flags);
1822                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1823         } while (cmpxchg(&mm->flags, old, new) != old);
1824 }
1825
1826 SYSCALL_DEFINE3(execve,
1827                 const char __user *, filename,
1828                 const char __user *const __user *, argv,
1829                 const char __user *const __user *, envp)
1830 {
1831         return do_execve(getname(filename), argv, envp);
1832 }
1833
1834 SYSCALL_DEFINE5(execveat,
1835                 int, fd, const char __user *, filename,
1836                 const char __user *const __user *, argv,
1837                 const char __user *const __user *, envp,
1838                 int, flags)
1839 {
1840         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1841
1842         return do_execveat(fd,
1843                            getname_flags(filename, lookup_flags, NULL),
1844                            argv, envp, flags);
1845 }
1846
1847 #ifdef CONFIG_COMPAT
1848 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1849         const compat_uptr_t __user *, argv,
1850         const compat_uptr_t __user *, envp)
1851 {
1852         return compat_do_execve(getname(filename), argv, envp);
1853 }
1854
1855 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1856                        const char __user *, filename,
1857                        const compat_uptr_t __user *, argv,
1858                        const compat_uptr_t __user *, envp,
1859                        int,  flags)
1860 {
1861         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1862
1863         return compat_do_execveat(fd,
1864                                   getname_flags(filename, lookup_flags, NULL),
1865                                   argv, envp, flags);
1866 }
1867 #endif