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/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/swap.h>
  32 #include <linux/string.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/mount.h>
  46 #include <linux/security.h>
  47 #include <linux/syscalls.h>
  48 #include <linux/tsacct_kern.h>
  49 #include <linux/cn_proc.h>
  50 #include <linux/audit.h>
  51 #include <linux/tracehook.h>
  52 #include <linux/kmod.h>
  53 #include <linux/fsnotify.h>
  54 #include <linux/fs_struct.h>
  55 #include <linux/pipe_fs_i.h>
  56 #include <linux/oom.h>
  57 #include <linux/compat.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62 #include <asm/exec.h>
  63
  64 #include <trace/events/task.h>
  65 #include "internal.h"
  66
  67 #include <trace/events/sched.h>
  68
  69 int suid_dumpable = 0;
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         BUG_ON(!fmt);
  77         write_lock(&binfmt_lock);
  78         insert ? list_add(&fmt->lh, &formats) :
  79                  list_add_tail(&fmt->lh, &formats);
  80         write_unlock(&binfmt_lock);
  81 }
  82
  83 EXPORT_SYMBOL(__register_binfmt);
  84
  85 void unregister_binfmt(struct linux_binfmt * fmt)
  86 {
  87         write_lock(&binfmt_lock);
  88         list_del(&fmt->lh);
  89         write_unlock(&binfmt_lock);
  90 }
  91
  92 EXPORT_SYMBOL(unregister_binfmt);
  93
  94 static inline void put_binfmt(struct linux_binfmt * fmt)
  95 {
  96         module_put(fmt->module);
  97 }
  98
  99 /*
 100  * Note that a shared library must be both readable and executable due to
 101  * security reasons.
 102  *
 103  * Also note that we take the address to load from from the file itself.
 104  */
 105 SYSCALL_DEFINE1(uselib, const char __user *, library)
 106 {
 107         struct file *file;
 108         char *tmp = getname(library);
 109         int error = PTR_ERR(tmp);
 110         static const struct open_flags uselib_flags = {
 111                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 112                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 113                 .intent = LOOKUP_OPEN
 114         };
 115
 116         if (IS_ERR(tmp))
 117                 goto out;
 118
 119         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 120         putname(tmp);
 121         error = PTR_ERR(file);
 122         if (IS_ERR(file))
 123                 goto out;
 124
 125         error = -EINVAL;
 126         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 127                 goto exit;
 128
 129         error = -EACCES;
 130         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 131                 goto exit;
 132
 133         fsnotify_open(file);
 134
 135         error = -ENOEXEC;
 136         if(file->f_op) {
 137                 struct linux_binfmt * fmt;
 138
 139                 read_lock(&binfmt_lock);
 140                 list_for_each_entry(fmt, &formats, lh) {
 141                         if (!fmt->load_shlib)
 142                                 continue;
 143                         if (!try_module_get(fmt->module))
 144                                 continue;
 145                         read_unlock(&binfmt_lock);
 146                         error = fmt->load_shlib(file);
 147                         read_lock(&binfmt_lock);
 148                         put_binfmt(fmt);
 149                         if (error != -ENOEXEC)
 150                                 break;
 151                 }
 152                 read_unlock(&binfmt_lock);
 153         }
 154 exit:
 155         fput(file);
 156 out:
 157         return error;
 158 }
 159
 160 #ifdef CONFIG_MMU
 161 /*
 162  * The nascent bprm->mm is not visible until exec_mmap() but it can
 163  * use a lot of memory, account these pages in current->mm temporary
 164  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 165  * change the counter back via acct_arg_size(0).
 166  */
 167 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 168 {
 169         struct mm_struct *mm = current->mm;
 170         long diff = (long)(pages - bprm->vma_pages);
 171
 172         if (!mm || !diff)
 173                 return;
 174
 175         bprm->vma_pages = pages;
 176         add_mm_counter(mm, MM_ANONPAGES, diff);
 177 }
 178
 179 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 180                 int write)
 181 {
 182         struct page *page;
 183         int ret;
 184
 185 #ifdef CONFIG_STACK_GROWSUP
 186         if (write) {
 187                 ret = expand_downwards(bprm->vma, pos);
 188                 if (ret < 0)
 189                         return NULL;
 190         }
 191 #endif
 192         ret = get_user_pages(current, bprm->mm, pos,
 193                         1, write, 1, &page, NULL);
 194         if (ret <= 0)
 195                 return NULL;
 196
 197         if (write) {
 198                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 199                 struct rlimit *rlim;
 200
 201                 acct_arg_size(bprm, size / PAGE_SIZE);
 202
 203                 /*
 204                  * We've historically supported up to 32 pages (ARG_MAX)
 205                  * of argument strings even with small stacks
 206                  */
 207                 if (size <= ARG_MAX)
 208                         return page;
 209
 210                 /*
 211                  * Limit to 1/4-th the stack size for the argv+env strings.
 212                  * This ensures that:
 213                  *  - the remaining binfmt code will not run out of stack space,
 214                  *  - the program will have a reasonable amount of stack left
 215                  *    to work from.
 216                  */
 217                 rlim = current->signal->rlim;
 218                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 219                         put_page(page);
 220                         return NULL;
 221                 }
 222         }
 223
 224         return page;
 225 }
 226
 227 static void put_arg_page(struct page *page)
 228 {
 229         put_page(page);
 230 }
 231
 232 static void free_arg_page(struct linux_binprm *bprm, int i)
 233 {
 234 }
 235
 236 static void free_arg_pages(struct linux_binprm *bprm)
 237 {
 238 }
 239
 240 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 241                 struct page *page)
 242 {
 243         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 244 }
 245
 246 static int __bprm_mm_init(struct linux_binprm *bprm)
 247 {
 248         int err;
 249         struct vm_area_struct *vma = NULL;
 250         struct mm_struct *mm = bprm->mm;
 251
 252         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 253         if (!vma)
 254                 return -ENOMEM;
 255
 256         down_write(&mm->mmap_sem);
 257         vma->vm_mm = mm;
 258
 259         /*
 260          * Place the stack at the largest stack address the architecture
 261          * supports. Later, we'll move this to an appropriate place. We don't
 262          * use STACK_TOP because that can depend on attributes which aren't
 263          * configured yet.
 264          */
 265         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 266         vma->vm_end = STACK_TOP_MAX;
 267         vma->vm_start = vma->vm_end - PAGE_SIZE;
 268         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 269         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 270         INIT_LIST_HEAD(&vma->anon_vma_chain);
 271
 272         err = insert_vm_struct(mm, vma);
 273         if (err)
 274                 goto err;
 275
 276         mm->stack_vm = mm->total_vm = 1;
 277         up_write(&mm->mmap_sem);
 278         bprm->p = vma->vm_end - sizeof(void *);
 279         return 0;
 280 err:
 281         up_write(&mm->mmap_sem);
 282         bprm->vma = NULL;
 283         kmem_cache_free(vm_area_cachep, vma);
 284         return err;
 285 }
 286
 287 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 288 {
 289         return len <= MAX_ARG_STRLEN;
 290 }
 291
 292 #else
 293
 294 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 295 {
 296 }
 297
 298 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 299                 int write)
 300 {
 301         struct page *page;
 302
 303         page = bprm->page[pos / PAGE_SIZE];
 304         if (!page && write) {
 305                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 306                 if (!page)
 307                         return NULL;
 308                 bprm->page[pos / PAGE_SIZE] = page;
 309         }
 310
 311         return page;
 312 }
 313
 314 static void put_arg_page(struct page *page)
 315 {
 316 }
 317
 318 static void free_arg_page(struct linux_binprm *bprm, int i)
 319 {
 320         if (bprm->page[i]) {
 321                 __free_page(bprm->page[i]);
 322                 bprm->page[i] = NULL;
 323         }
 324 }
 325
 326 static void free_arg_pages(struct linux_binprm *bprm)
 327 {
 328         int i;
 329
 330         for (i = 0; i < MAX_ARG_PAGES; i++)
 331                 free_arg_page(bprm, i);
 332 }
 333
 334 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 335                 struct page *page)
 336 {
 337 }
 338
 339 static int __bprm_mm_init(struct linux_binprm *bprm)
 340 {
 341         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 342         return 0;
 343 }
 344
 345 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 346 {
 347         return len <= bprm->p;
 348 }
 349
 350 #endif /* CONFIG_MMU */
 351
 352 /*
 353  * Create a new mm_struct and populate it with a temporary stack
 354  * vm_area_struct.  We don't have enough context at this point to set the stack
 355  * flags, permissions, and offset, so we use temporary values.  We'll update
 356  * them later in setup_arg_pages().
 357  */
 358 int bprm_mm_init(struct linux_binprm *bprm)
 359 {
 360         int err;
 361         struct mm_struct *mm = NULL;
 362
 363         bprm->mm = mm = mm_alloc();
 364         err = -ENOMEM;
 365         if (!mm)
 366                 goto err;
 367
 368         err = init_new_context(current, mm);
 369         if (err)
 370                 goto err;
 371
 372         err = __bprm_mm_init(bprm);
 373         if (err)
 374                 goto err;
 375
 376         return 0;
 377
 378 err:
 379         if (mm) {
 380                 bprm->mm = NULL;
 381                 mmdrop(mm);
 382         }
 383
 384         return err;
 385 }
 386
 387 struct user_arg_ptr {
 388 #ifdef CONFIG_COMPAT
 389         bool is_compat;
 390 #endif
 391         union {
 392                 const char __user *const __user *native;
 393 #ifdef CONFIG_COMPAT
 394                 compat_uptr_t __user *compat;
 395 #endif
 396         } ptr;
 397 };
 398
 399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 400 {
 401         const char __user *native;
 402
 403 #ifdef CONFIG_COMPAT
 404         if (unlikely(argv.is_compat)) {
 405                 compat_uptr_t compat;
 406
 407                 if (get_user(compat, argv.ptr.compat + nr))
 408                         return ERR_PTR(-EFAULT);
 409
 410                 return compat_ptr(compat);
 411         }
 412 #endif
 413
 414         if (get_user(native, argv.ptr.native + nr))
 415                 return ERR_PTR(-EFAULT);
 416
 417         return native;
 418 }
 419
 420 /*
 421  * count() counts the number of strings in array ARGV.
 422  */
 423 static int count(struct user_arg_ptr argv, int max)
 424 {
 425         int i = 0;
 426
 427         if (argv.ptr.native != NULL) {
 428                 for (;;) {
 429                         const char __user *p = get_user_arg_ptr(argv, i);
 430
 431                         if (!p)
 432                                 break;
 433
 434                         if (IS_ERR(p))
 435                                 return -EFAULT;
 436
 437                         if (i++ >= max)
 438                                 return -E2BIG;
 439
 440                         if (fatal_signal_pending(current))
 441                                 return -ERESTARTNOHAND;
 442                         cond_resched();
 443                 }
 444         }
 445         return i;
 446 }
 447
 448 /*
 449  * 'copy_strings()' copies argument/environment strings from the old
 450  * processes's memory to the new process's stack.  The call to get_user_pages()
 451  * ensures the destination page is created and not swapped out.
 452  */
 453 static int copy_strings(int argc, struct user_arg_ptr argv,
 454                         struct linux_binprm *bprm)
 455 {
 456         struct page *kmapped_page = NULL;
 457         char *kaddr = NULL;
 458         unsigned long kpos = 0;
 459         int ret;
 460
 461         while (argc-- > 0) {
 462                 const char __user *str;
 463                 int len;
 464                 unsigned long pos;
 465
 466                 ret = -EFAULT;
 467                 str = get_user_arg_ptr(argv, argc);
 468                 if (IS_ERR(str))
 469                         goto out;
 470
 471                 len = strnlen_user(str, MAX_ARG_STRLEN);
 472                 if (!len)
 473                         goto out;
 474
 475                 ret = -E2BIG;
 476                 if (!valid_arg_len(bprm, len))
 477                         goto out;
 478
 479                 /* We're going to work our way backwords. */
 480                 pos = bprm->p;
 481                 str += len;
 482                 bprm->p -= len;
 483
 484                 while (len > 0) {
 485                         int offset, bytes_to_copy;
 486
 487                         if (fatal_signal_pending(current)) {
 488                                 ret = -ERESTARTNOHAND;
 489                                 goto out;
 490                         }
 491                         cond_resched();
 492
 493                         offset = pos % PAGE_SIZE;
 494                         if (offset == 0)
 495                                 offset = PAGE_SIZE;
 496
 497                         bytes_to_copy = offset;
 498                         if (bytes_to_copy > len)
 499                                 bytes_to_copy = len;
 500
 501                         offset -= bytes_to_copy;
 502                         pos -= bytes_to_copy;
 503                         str -= bytes_to_copy;
 504                         len -= bytes_to_copy;
 505
 506                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 507                                 struct page *page;
 508
 509                                 page = get_arg_page(bprm, pos, 1);
 510                                 if (!page) {
 511                                         ret = -E2BIG;
 512                                         goto out;
 513                                 }
 514
 515                                 if (kmapped_page) {
 516                                         flush_kernel_dcache_page(kmapped_page);
 517                                         kunmap(kmapped_page);
 518                                         put_arg_page(kmapped_page);
 519                                 }
 520                                 kmapped_page = page;
 521                                 kaddr = kmap(kmapped_page);
 522                                 kpos = pos & PAGE_MASK;
 523                                 flush_arg_page(bprm, kpos, kmapped_page);
 524                         }
 525                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 526                                 ret = -EFAULT;
 527                                 goto out;
 528                         }
 529                 }
 530         }
 531         ret = 0;
 532 out:
 533         if (kmapped_page) {
 534                 flush_kernel_dcache_page(kmapped_page);
 535                 kunmap(kmapped_page);
 536                 put_arg_page(kmapped_page);
 537         }
 538         return ret;
 539 }
 540
 541 /*
 542  * Like copy_strings, but get argv and its values from kernel memory.
 543  */
 544 int copy_strings_kernel(int argc, const char *const *__argv,
 545                         struct linux_binprm *bprm)
 546 {
 547         int r;
 548         mm_segment_t oldfs = get_fs();
 549         struct user_arg_ptr argv = {
 550                 .ptr.native = (const char __user *const  __user *)__argv,
 551         };
 552
 553         set_fs(KERNEL_DS);
 554         r = copy_strings(argc, argv, bprm);
 555         set_fs(oldfs);
 556
 557         return r;
 558 }
 559 EXPORT_SYMBOL(copy_strings_kernel);
 560
 561 #ifdef CONFIG_MMU
 562
 563 /*
 564  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 565  * the binfmt code determines where the new stack should reside, we shift it to
 566  * its final location.  The process proceeds as follows:
 567  *
 568  * 1) Use shift to calculate the new vma endpoints.
 569  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 570  *    arguments passed to subsequent functions are consistent.
 571  * 3) Move vma's page tables to the new range.
 572  * 4) Free up any cleared pgd range.
 573  * 5) Shrink the vma to cover only the new range.
 574  */
 575 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 576 {
 577         struct mm_struct *mm = vma->vm_mm;
 578         unsigned long old_start = vma->vm_start;
 579         unsigned long old_end = vma->vm_end;
 580         unsigned long length = old_end - old_start;
 581         unsigned long new_start = old_start - shift;
 582         unsigned long new_end = old_end - shift;
 583         struct mmu_gather tlb;
 584
 585         BUG_ON(new_start > new_end);
 586
 587         /*
 588          * ensure there are no vmas between where we want to go
 589          * and where we are
 590          */
 591         if (vma != find_vma(mm, new_start))
 592                 return -EFAULT;
 593
 594         /*
 595          * cover the whole range: [new_start, old_end)
 596          */
 597         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 598                 return -ENOMEM;
 599
 600         /*
 601          * move the page tables downwards, on failure we rely on
 602          * process cleanup to remove whatever mess we made.
 603          */
 604         if (length != move_page_tables(vma, old_start,
 605                                        vma, new_start, length))
 606                 return -ENOMEM;
 607
 608         lru_add_drain();
 609         tlb_gather_mmu(&tlb, mm, 0);
 610         if (new_end > old_start) {
 611                 /*
 612                  * when the old and new regions overlap clear from new_end.
 613                  */
 614                 free_pgd_range(&tlb, new_end, old_end, new_end,
 615                         vma->vm_next ? vma->vm_next->vm_start : 0);
 616         } else {
 617                 /*
 618                  * otherwise, clean from old_start; this is done to not touch
 619                  * the address space in [new_end, old_start) some architectures
 620                  * have constraints on va-space that make this illegal (IA64) -
 621                  * for the others its just a little faster.
 622                  */
 623                 free_pgd_range(&tlb, old_start, old_end, new_end,
 624                         vma->vm_next ? vma->vm_next->vm_start : 0);
 625         }
 626         tlb_finish_mmu(&tlb, new_end, old_end);
 627
 628         /*
 629          * Shrink the vma to just the new range.  Always succeeds.
 630          */
 631         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 632
 633         return 0;
 634 }
 635
 636 /*
 637  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 638  * the stack is optionally relocated, and some extra space is added.
 639  */
 640 int setup_arg_pages(struct linux_binprm *bprm,
 641                     unsigned long stack_top,
 642                     int executable_stack)
 643 {
 644         unsigned long ret;
 645         unsigned long stack_shift;
 646         struct mm_struct *mm = current->mm;
 647         struct vm_area_struct *vma = bprm->vma;
 648         struct vm_area_struct *prev = NULL;
 649         unsigned long vm_flags;
 650         unsigned long stack_base;
 651         unsigned long stack_size;
 652         unsigned long stack_expand;
 653         unsigned long rlim_stack;
 654
 655 #ifdef CONFIG_STACK_GROWSUP
 656         /* Limit stack size to 1GB */
 657         stack_base = rlimit_max(RLIMIT_STACK);
 658         if (stack_base > (1 << 30))
 659                 stack_base = 1 << 30;
 660
 661         /* Make sure we didn't let the argument array grow too large. */
 662         if (vma->vm_end - vma->vm_start > stack_base)
 663                 return -ENOMEM;
 664
 665         stack_base = PAGE_ALIGN(stack_top - stack_base);
 666
 667         stack_shift = vma->vm_start - stack_base;
 668         mm->arg_start = bprm->p - stack_shift;
 669         bprm->p = vma->vm_end - stack_shift;
 670 #else
 671         stack_top = arch_align_stack(stack_top);
 672         stack_top = PAGE_ALIGN(stack_top);
 673
 674         if (unlikely(stack_top < mmap_min_addr) ||
 675             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 676                 return -ENOMEM;
 677
 678         stack_shift = vma->vm_end - stack_top;
 679
 680         bprm->p -= stack_shift;
 681         mm->arg_start = bprm->p;
 682 #endif
 683
 684         if (bprm->loader)
 685                 bprm->loader -= stack_shift;
 686         bprm->exec -= stack_shift;
 687
 688         down_write(&mm->mmap_sem);
 689         vm_flags = VM_STACK_FLAGS;
 690
 691         /*
 692          * Adjust stack execute permissions; explicitly enable for
 693          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 694          * (arch default) otherwise.
 695          */
 696         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 697                 vm_flags |= VM_EXEC;
 698         else if (executable_stack == EXSTACK_DISABLE_X)
 699                 vm_flags &= ~VM_EXEC;
 700         vm_flags |= mm->def_flags;
 701         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 702
 703         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 704                         vm_flags);
 705         if (ret)
 706                 goto out_unlock;
 707         BUG_ON(prev != vma);
 708
 709         /* Move stack pages down in memory. */
 710         if (stack_shift) {
 711                 ret = shift_arg_pages(vma, stack_shift);
 712                 if (ret)
 713                         goto out_unlock;
 714         }
 715
 716         /* mprotect_fixup is overkill to remove the temporary stack flags */
 717         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 718
 719         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 720         stack_size = vma->vm_end - vma->vm_start;
 721         /*
 722          * Align this down to a page boundary as expand_stack
 723          * will align it up.
 724          */
 725         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 726 #ifdef CONFIG_STACK_GROWSUP
 727         if (stack_size + stack_expand > rlim_stack)
 728                 stack_base = vma->vm_start + rlim_stack;
 729         else
 730                 stack_base = vma->vm_end + stack_expand;
 731 #else
 732         if (stack_size + stack_expand > rlim_stack)
 733                 stack_base = vma->vm_end - rlim_stack;
 734         else
 735                 stack_base = vma->vm_start - stack_expand;
 736 #endif
 737         current->mm->start_stack = bprm->p;
 738         ret = expand_stack(vma, stack_base);
 739         if (ret)
 740                 ret = -EFAULT;
 741
 742 out_unlock:
 743         up_write(&mm->mmap_sem);
 744         return ret;
 745 }
 746 EXPORT_SYMBOL(setup_arg_pages);
 747
 748 #endif /* CONFIG_MMU */
 749
 750 struct file *open_exec(const char *name)
 751 {
 752         struct file *file;
 753         int err;
 754         static const struct open_flags open_exec_flags = {
 755                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 756                 .acc_mode = MAY_EXEC | MAY_OPEN,
 757                 .intent = LOOKUP_OPEN
 758         };
 759
 760         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 761         if (IS_ERR(file))
 762                 goto out;
 763
 764         err = -EACCES;
 765         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 766                 goto exit;
 767
 768         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 769                 goto exit;
 770
 771         fsnotify_open(file);
 772
 773         err = deny_write_access(file);
 774         if (err)
 775                 goto exit;
 776
 777 out:
 778         return file;
 779
 780 exit:
 781         fput(file);
 782         return ERR_PTR(err);
 783 }
 784 EXPORT_SYMBOL(open_exec);
 785
 786 int kernel_read(struct file *file, loff_t offset,
 787                 char *addr, unsigned long count)
 788 {
 789         mm_segment_t old_fs;
 790         loff_t pos = offset;
 791         int result;
 792
 793         old_fs = get_fs();
 794         set_fs(get_ds());
 795         /* The cast to a user pointer is valid due to the set_fs() */
 796         result = vfs_read(file, (void __user *)addr, count, &pos);
 797         set_fs(old_fs);
 798         return result;
 799 }
 800
 801 EXPORT_SYMBOL(kernel_read);
 802
 803 static int exec_mmap(struct mm_struct *mm)
 804 {
 805         struct task_struct *tsk;
 806         struct mm_struct * old_mm, *active_mm;
 807
 808         /* Notify parent that we're no longer interested in the old VM */
 809         tsk = current;
 810         old_mm = current->mm;
 811         mm_release(tsk, old_mm);
 812
 813         if (old_mm) {
 814                 sync_mm_rss(old_mm);
 815                 /*
 816                  * Make sure that if there is a core dump in progress
 817                  * for the old mm, we get out and die instead of going
 818                  * through with the exec.  We must hold mmap_sem around
 819                  * checking core_state and changing tsk->mm.
 820                  */
 821                 down_read(&old_mm->mmap_sem);
 822                 if (unlikely(old_mm->core_state)) {
 823                         up_read(&old_mm->mmap_sem);
 824                         return -EINTR;
 825                 }
 826         }
 827         task_lock(tsk);
 828         active_mm = tsk->active_mm;
 829         tsk->mm = mm;
 830         tsk->active_mm = mm;
 831         activate_mm(active_mm, mm);
 832         task_unlock(tsk);
 833         arch_pick_mmap_layout(mm);
 834         if (old_mm) {
 835                 up_read(&old_mm->mmap_sem);
 836                 BUG_ON(active_mm != old_mm);
 837                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 838                 mm_update_next_owner(old_mm);
 839                 mmput(old_mm);
 840                 return 0;
 841         }
 842         mmdrop(active_mm);
 843         return 0;
 844 }
 845
 846 /*
 847  * This function makes sure the current process has its own signal table,
 848  * so that flush_signal_handlers can later reset the handlers without
 849  * disturbing other processes.  (Other processes might share the signal
 850  * table via the CLONE_SIGHAND option to clone().)
 851  */
 852 static int de_thread(struct task_struct *tsk)
 853 {
 854         struct signal_struct *sig = tsk->signal;
 855         struct sighand_struct *oldsighand = tsk->sighand;
 856         spinlock_t *lock = &oldsighand->siglock;
 857
 858         if (thread_group_empty(tsk))
 859                 goto no_thread_group;
 860
 861         /*
 862          * Kill all other threads in the thread group.
 863          */
 864         spin_lock_irq(lock);
 865         if (signal_group_exit(sig)) {
 866                 /*
 867                  * Another group action in progress, just
 868                  * return so that the signal is processed.
 869                  */
 870                 spin_unlock_irq(lock);
 871                 return -EAGAIN;
 872         }
 873
 874         sig->group_exit_task = tsk;
 875         sig->notify_count = zap_other_threads(tsk);
 876         if (!thread_group_leader(tsk))
 877                 sig->notify_count--;
 878
 879         while (sig->notify_count) {
 880                 __set_current_state(TASK_UNINTERRUPTIBLE);
 881                 spin_unlock_irq(lock);
 882                 schedule();
 883                 spin_lock_irq(lock);
 884         }
 885         spin_unlock_irq(lock);
 886
 887         /*
 888          * At this point all other threads have exited, all we have to
 889          * do is to wait for the thread group leader to become inactive,
 890          * and to assume its PID:
 891          */
 892         if (!thread_group_leader(tsk)) {
 893                 struct task_struct *leader = tsk->group_leader;
 894
 895                 sig->notify_count = -1; /* for exit_notify() */
 896                 for (;;) {
 897                         write_lock_irq(&tasklist_lock);
 898                         if (likely(leader->exit_state))
 899                                 break;
 900                         __set_current_state(TASK_UNINTERRUPTIBLE);
 901                         write_unlock_irq(&tasklist_lock);
 902                         schedule();
 903                 }
 904
 905                 /*
 906                  * The only record we have of the real-time age of a
 907                  * process, regardless of execs it's done, is start_time.
 908                  * All the past CPU time is accumulated in signal_struct
 909                  * from sister threads now dead.  But in this non-leader
 910                  * exec, nothing survives from the original leader thread,
 911                  * whose birth marks the true age of this process now.
 912                  * When we take on its identity by switching to its PID, we
 913                  * also take its birthdate (always earlier than our own).
 914                  */
 915                 tsk->start_time = leader->start_time;
 916
 917                 BUG_ON(!same_thread_group(leader, tsk));
 918                 BUG_ON(has_group_leader_pid(tsk));
 919                 /*
 920                  * An exec() starts a new thread group with the
 921                  * TGID of the previous thread group. Rehash the
 922                  * two threads with a switched PID, and release
 923                  * the former thread group leader:
 924                  */
 925
 926                 /* Become a process group leader with the old leader's pid.
 927                  * The old leader becomes a thread of the this thread group.
 928                  * Note: The old leader also uses this pid until release_task
 929                  *       is called.  Odd but simple and correct.
 930                  */
 931                 detach_pid(tsk, PIDTYPE_PID);
 932                 tsk->pid = leader->pid;
 933                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 934                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 935                 transfer_pid(leader, tsk, PIDTYPE_SID);
 936
 937                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 938                 list_replace_init(&leader->sibling, &tsk->sibling);
 939
 940                 tsk->group_leader = tsk;
 941                 leader->group_leader = tsk;
 942
 943                 tsk->exit_signal = SIGCHLD;
 944                 leader->exit_signal = -1;
 945
 946                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 947                 leader->exit_state = EXIT_DEAD;
 948
 949                 /*
 950                  * We are going to release_task()->ptrace_unlink() silently,
 951                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 952                  * the tracer wont't block again waiting for this thread.
 953                  */
 954                 if (unlikely(leader->ptrace))
 955                         __wake_up_parent(leader, leader->parent);
 956                 write_unlock_irq(&tasklist_lock);
 957
 958                 release_task(leader);
 959         }
 960
 961         sig->group_exit_task = NULL;
 962         sig->notify_count = 0;
 963
 964 no_thread_group:
 965         /* we have changed execution domain */
 966         tsk->exit_signal = SIGCHLD;
 967
 968         exit_itimers(sig);
 969         flush_itimer_signals();
 970
 971         if (atomic_read(&oldsighand->count) != 1) {
 972                 struct sighand_struct *newsighand;
 973                 /*
 974                  * This ->sighand is shared with the CLONE_SIGHAND
 975                  * but not CLONE_THREAD task, switch to the new one.
 976                  */
 977                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 978                 if (!newsighand)
 979                         return -ENOMEM;
 980
 981                 atomic_set(&newsighand->count, 1);
 982                 memcpy(newsighand->action, oldsighand->action,
 983                        sizeof(newsighand->action));
 984
 985                 write_lock_irq(&tasklist_lock);
 986                 spin_lock(&oldsighand->siglock);
 987                 rcu_assign_pointer(tsk->sighand, newsighand);
 988                 spin_unlock(&oldsighand->siglock);
 989                 write_unlock_irq(&tasklist_lock);
 990
 991                 __cleanup_sighand(oldsighand);
 992         }
 993
 994         BUG_ON(!thread_group_leader(tsk));
 995         return 0;
 996 }
 997
 998 char *get_task_comm(char *buf, struct task_struct *tsk)
 999 {
1000         /* buf must be at least sizeof(tsk->comm) in size */
1001         task_lock(tsk);
1002         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1003         task_unlock(tsk);
1004         return buf;
1005 }
1006 EXPORT_SYMBOL_GPL(get_task_comm);
1007
1008 /*
1009  * These functions flushes out all traces of the currently running executable
1010  * so that a new one can be started
1011  */
1012
1013 void set_task_comm(struct task_struct *tsk, char *buf)
1014 {
1015         task_lock(tsk);
1016
1017         trace_task_rename(tsk, buf);
1018
1019         /*
1020          * Threads may access current->comm without holding
1021          * the task lock, so write the string carefully.
1022          * Readers without a lock may see incomplete new
1023          * names but are safe from non-terminating string reads.
1024          */
1025         memset(tsk->comm, 0, TASK_COMM_LEN);
1026         wmb();
1027         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1028         task_unlock(tsk);
1029         perf_event_comm(tsk);
1030 }
1031
1032 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1033 {
1034         int i, ch;
1035
1036         /* Copies the binary name from after last slash */
1037         for (i = 0; (ch = *(fn++)) != '\0';) {
1038                 if (ch == '/')
1039                         i = 0; /* overwrite what we wrote */
1040                 else
1041                         if (i < len - 1)
1042                                 tcomm[i++] = ch;
1043         }
1044         tcomm[i] = '\0';
1045 }
1046
1047 int flush_old_exec(struct linux_binprm * bprm)
1048 {
1049         int retval;
1050
1051         /*
1052          * Make sure we have a private signal table and that
1053          * we are unassociated from the previous thread group.
1054          */
1055         retval = de_thread(current);
1056         if (retval)
1057                 goto out;
1058
1059         set_mm_exe_file(bprm->mm, bprm->file);
1060
1061         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1062         /*
1063          * Release all of the old mmap stuff
1064          */
1065         acct_arg_size(bprm, 0);
1066         retval = exec_mmap(bprm->mm);
1067         if (retval)
1068                 goto out;
1069
1070         bprm->mm = NULL;                /* We're using it now */
1071
1072         set_fs(USER_DS);
1073         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD);
1074         flush_thread();
1075         current->personality &= ~bprm->per_clear;
1076
1077         return 0;
1078
1079 out:
1080         return retval;
1081 }
1082 EXPORT_SYMBOL(flush_old_exec);
1083
1084 void would_dump(struct linux_binprm *bprm, struct file *file)
1085 {
1086         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1087                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1088 }
1089 EXPORT_SYMBOL(would_dump);
1090
1091 void setup_new_exec(struct linux_binprm * bprm)
1092 {
1093         arch_pick_mmap_layout(current->mm);
1094
1095         /* This is the point of no return */
1096         current->sas_ss_sp = current->sas_ss_size = 0;
1097
1098         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1099                 set_dumpable(current->mm, 1);
1100         else
1101                 set_dumpable(current->mm, suid_dumpable);
1102
1103         set_task_comm(current, bprm->tcomm);
1104
1105         /* Set the new mm task size. We have to do that late because it may
1106          * depend on TIF_32BIT which is only updated in flush_thread() on
1107          * some architectures like powerpc
1108          */
1109         current->mm->task_size = TASK_SIZE;
1110
1111         /* install the new credentials */
1112         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1113             !gid_eq(bprm->cred->gid, current_egid())) {
1114                 current->pdeath_signal = 0;
1115         } else {
1116                 would_dump(bprm, bprm->file);
1117                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1118                         set_dumpable(current->mm, suid_dumpable);
1119         }
1120
1121         /*
1122          * Flush performance counters when crossing a
1123          * security domain:
1124          */
1125         if (!get_dumpable(current->mm))
1126                 perf_event_exit_task(current);
1127
1128         /* An exec changes our domain. We are no longer part of the thread
1129            group */
1130
1131         current->self_exec_id++;
1132
1133         flush_signal_handlers(current, 0);
1134         do_close_on_exec(current->files);
1135 }
1136 EXPORT_SYMBOL(setup_new_exec);
1137
1138 /*
1139  * Prepare credentials and lock ->cred_guard_mutex.
1140  * install_exec_creds() commits the new creds and drops the lock.
1141  * Or, if exec fails before, free_bprm() should release ->cred and
1142  * and unlock.
1143  */
1144 int prepare_bprm_creds(struct linux_binprm *bprm)
1145 {
1146         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1147                 return -ERESTARTNOINTR;
1148
1149         bprm->cred = prepare_exec_creds();
1150         if (likely(bprm->cred))
1151                 return 0;
1152
1153         mutex_unlock(&current->signal->cred_guard_mutex);
1154         return -ENOMEM;
1155 }
1156
1157 void free_bprm(struct linux_binprm *bprm)
1158 {
1159         free_arg_pages(bprm);
1160         if (bprm->cred) {
1161                 mutex_unlock(&current->signal->cred_guard_mutex);
1162                 abort_creds(bprm->cred);
1163         }
1164         kfree(bprm);
1165 }
1166
1167 /*
1168  * install the new credentials for this executable
1169  */
1170 void install_exec_creds(struct linux_binprm *bprm)
1171 {
1172         security_bprm_committing_creds(bprm);
1173
1174         commit_creds(bprm->cred);
1175         bprm->cred = NULL;
1176         /*
1177          * cred_guard_mutex must be held at least to this point to prevent
1178          * ptrace_attach() from altering our determination of the task's
1179          * credentials; any time after this it may be unlocked.
1180          */
1181         security_bprm_committed_creds(bprm);
1182         mutex_unlock(&current->signal->cred_guard_mutex);
1183 }
1184 EXPORT_SYMBOL(install_exec_creds);
1185
1186 /*
1187  * determine how safe it is to execute the proposed program
1188  * - the caller must hold ->cred_guard_mutex to protect against
1189  *   PTRACE_ATTACH
1190  */
1191 static int check_unsafe_exec(struct linux_binprm *bprm)
1192 {
1193         struct task_struct *p = current, *t;
1194         unsigned n_fs;
1195         int res = 0;
1196
1197         if (p->ptrace) {
1198                 if (p->ptrace & PT_PTRACE_CAP)
1199                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1200                 else
1201                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1202         }
1203
1204         /*
1205          * This isn't strictly necessary, but it makes it harder for LSMs to
1206          * mess up.
1207          */
1208         if (current->no_new_privs)
1209                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1210
1211         n_fs = 1;
1212         spin_lock(&p->fs->lock);
1213         rcu_read_lock();
1214         for (t = next_thread(p); t != p; t = next_thread(t)) {
1215                 if (t->fs == p->fs)
1216                         n_fs++;
1217         }
1218         rcu_read_unlock();
1219
1220         if (p->fs->users > n_fs) {
1221                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1222         } else {
1223                 res = -EAGAIN;
1224                 if (!p->fs->in_exec) {
1225                         p->fs->in_exec = 1;
1226                         res = 1;
1227                 }
1228         }
1229         spin_unlock(&p->fs->lock);
1230
1231         return res;
1232 }
1233
1234 /*
1235  * Fill the binprm structure from the inode.
1236  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1237  *
1238  * This may be called multiple times for binary chains (scripts for example).
1239  */
1240 int prepare_binprm(struct linux_binprm *bprm)
1241 {
1242         umode_t mode;
1243         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1244         int retval;
1245
1246         mode = inode->i_mode;
1247         if (bprm->file->f_op == NULL)
1248                 return -EACCES;
1249
1250         /* clear any previous set[ug]id data from a previous binary */
1251         bprm->cred->euid = current_euid();
1252         bprm->cred->egid = current_egid();
1253
1254         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1255             !current->no_new_privs) {
1256                 /* Set-uid? */
1257                 if (mode & S_ISUID) {
1258                         if (!kuid_has_mapping(bprm->cred->user_ns, inode->i_uid))
1259                                 return -EPERM;
1260                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1261                         bprm->cred->euid = inode->i_uid;
1262
1263                 }
1264
1265                 /* Set-gid? */
1266                 /*
1267                  * If setgid is set but no group execute bit then this
1268                  * is a candidate for mandatory locking, not a setgid
1269                  * executable.
1270                  */
1271                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1272                         if (!kgid_has_mapping(bprm->cred->user_ns, inode->i_gid))
1273                                 return -EPERM;
1274                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1275                         bprm->cred->egid = inode->i_gid;
1276                 }
1277         }
1278
1279         /* fill in binprm security blob */
1280         retval = security_bprm_set_creds(bprm);
1281         if (retval)
1282                 return retval;
1283         bprm->cred_prepared = 1;
1284
1285         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1286         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1287 }
1288
1289 EXPORT_SYMBOL(prepare_binprm);
1290
1291 /*
1292  * Arguments are '\0' separated strings found at the location bprm->p
1293  * points to; chop off the first by relocating brpm->p to right after
1294  * the first '\0' encountered.
1295  */
1296 int remove_arg_zero(struct linux_binprm *bprm)
1297 {
1298         int ret = 0;
1299         unsigned long offset;
1300         char *kaddr;
1301         struct page *page;
1302
1303         if (!bprm->argc)
1304                 return 0;
1305
1306         do {
1307                 offset = bprm->p & ~PAGE_MASK;
1308                 page = get_arg_page(bprm, bprm->p, 0);
1309                 if (!page) {
1310                         ret = -EFAULT;
1311                         goto out;
1312                 }
1313                 kaddr = kmap_atomic(page);
1314
1315                 for (; offset < PAGE_SIZE && kaddr[offset];
1316                                 offset++, bprm->p++)
1317                         ;
1318
1319                 kunmap_atomic(kaddr);
1320                 put_arg_page(page);
1321
1322                 if (offset == PAGE_SIZE)
1323                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1324         } while (offset == PAGE_SIZE);
1325
1326         bprm->p++;
1327         bprm->argc--;
1328         ret = 0;
1329
1330 out:
1331         return ret;
1332 }
1333 EXPORT_SYMBOL(remove_arg_zero);
1334
1335 /*
1336  * cycle the list of binary formats handler, until one recognizes the image
1337  */
1338 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1339 {
1340         unsigned int depth = bprm->recursion_depth;
1341         int try,retval;
1342         struct linux_binfmt *fmt;
1343         pid_t old_pid, old_vpid;
1344
1345         retval = security_bprm_check(bprm);
1346         if (retval)
1347                 return retval;
1348
1349         retval = audit_bprm(bprm);
1350         if (retval)
1351                 return retval;
1352
1353         /* Need to fetch pid before load_binary changes it */
1354         old_pid = current->pid;
1355         rcu_read_lock();
1356         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1357         rcu_read_unlock();
1358
1359         retval = -ENOENT;
1360         for (try=0; try<2; try++) {
1361                 read_lock(&binfmt_lock);
1362                 list_for_each_entry(fmt, &formats, lh) {
1363                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1364                         if (!fn)
1365                                 continue;
1366                         if (!try_module_get(fmt->module))
1367                                 continue;
1368                         read_unlock(&binfmt_lock);
1369                         retval = fn(bprm, regs);
1370                         /*
1371                          * Restore the depth counter to its starting value
1372                          * in this call, so we don't have to rely on every
1373                          * load_binary function to restore it on return.
1374                          */
1375                         bprm->recursion_depth = depth;
1376                         if (retval >= 0) {
1377                                 if (depth == 0) {
1378                                         trace_sched_process_exec(current, old_pid, bprm);
1379                                         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1380                                 }
1381                                 put_binfmt(fmt);
1382                                 allow_write_access(bprm->file);
1383                                 if (bprm->file)
1384                                         fput(bprm->file);
1385                                 bprm->file = NULL;
1386                                 current->did_exec = 1;
1387                                 proc_exec_connector(current);
1388                                 return retval;
1389                         }
1390                         read_lock(&binfmt_lock);
1391                         put_binfmt(fmt);
1392                         if (retval != -ENOEXEC || bprm->mm == NULL)
1393                                 break;
1394                         if (!bprm->file) {
1395                                 read_unlock(&binfmt_lock);
1396                                 return retval;
1397                         }
1398                 }
1399                 read_unlock(&binfmt_lock);
1400 #ifdef CONFIG_MODULES
1401                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1402                         break;
1403                 } else {
1404 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1405                         if (printable(bprm->buf[0]) &&
1406                             printable(bprm->buf[1]) &&
1407                             printable(bprm->buf[2]) &&
1408                             printable(bprm->buf[3]))
1409                                 break; /* -ENOEXEC */
1410                         if (try)
1411                                 break; /* -ENOEXEC */
1412                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1413                 }
1414 #else
1415                 break;
1416 #endif
1417         }
1418         return retval;
1419 }
1420
1421 EXPORT_SYMBOL(search_binary_handler);
1422
1423 /*
1424  * sys_execve() executes a new program.
1425  */
1426 static int do_execve_common(const char *filename,
1427                                 struct user_arg_ptr argv,
1428                                 struct user_arg_ptr envp,
1429                                 struct pt_regs *regs)
1430 {
1431         struct linux_binprm *bprm;
1432         struct file *file;
1433         struct files_struct *displaced;
1434         bool clear_in_exec;
1435         int retval;
1436         const struct cred *cred = current_cred();
1437
1438         /*
1439          * We move the actual failure in case of RLIMIT_NPROC excess from
1440          * set*uid() to execve() because too many poorly written programs
1441          * don't check setuid() return code.  Here we additionally recheck
1442          * whether NPROC limit is still exceeded.
1443          */
1444         if ((current->flags & PF_NPROC_EXCEEDED) &&
1445             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1446                 retval = -EAGAIN;
1447                 goto out_ret;
1448         }
1449
1450         /* We're below the limit (still or again), so we don't want to make
1451          * further execve() calls fail. */
1452         current->flags &= ~PF_NPROC_EXCEEDED;
1453
1454         retval = unshare_files(&displaced);
1455         if (retval)
1456                 goto out_ret;
1457
1458         retval = -ENOMEM;
1459         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1460         if (!bprm)
1461                 goto out_files;
1462
1463         retval = prepare_bprm_creds(bprm);
1464         if (retval)
1465                 goto out_free;
1466
1467         retval = check_unsafe_exec(bprm);
1468         if (retval < 0)
1469                 goto out_free;
1470         clear_in_exec = retval;
1471         current->in_execve = 1;
1472
1473         file = open_exec(filename);
1474         retval = PTR_ERR(file);
1475         if (IS_ERR(file))
1476                 goto out_unmark;
1477
1478         sched_exec();
1479
1480         bprm->file = file;
1481         bprm->filename = filename;
1482         bprm->interp = filename;
1483
1484         retval = bprm_mm_init(bprm);
1485         if (retval)
1486                 goto out_file;
1487
1488         bprm->argc = count(argv, MAX_ARG_STRINGS);
1489         if ((retval = bprm->argc) < 0)
1490                 goto out;
1491
1492         bprm->envc = count(envp, MAX_ARG_STRINGS);
1493         if ((retval = bprm->envc) < 0)
1494                 goto out;
1495
1496         retval = prepare_binprm(bprm);
1497         if (retval < 0)
1498                 goto out;
1499
1500         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1501         if (retval < 0)
1502                 goto out;
1503
1504         bprm->exec = bprm->p;
1505         retval = copy_strings(bprm->envc, envp, bprm);
1506         if (retval < 0)
1507                 goto out;
1508
1509         retval = copy_strings(bprm->argc, argv, bprm);
1510         if (retval < 0)
1511                 goto out;
1512
1513         retval = search_binary_handler(bprm,regs);
1514         if (retval < 0)
1515                 goto out;
1516
1517         /* execve succeeded */
1518         current->fs->in_exec = 0;
1519         current->in_execve = 0;
1520         acct_update_integrals(current);
1521         free_bprm(bprm);
1522         if (displaced)
1523                 put_files_struct(displaced);
1524         return retval;
1525
1526 out:
1527         if (bprm->mm) {
1528                 acct_arg_size(bprm, 0);
1529                 mmput(bprm->mm);
1530         }
1531
1532 out_file:
1533         if (bprm->file) {
1534                 allow_write_access(bprm->file);
1535                 fput(bprm->file);
1536         }
1537
1538 out_unmark:
1539         if (clear_in_exec)
1540                 current->fs->in_exec = 0;
1541         current->in_execve = 0;
1542
1543 out_free:
1544         free_bprm(bprm);
1545
1546 out_files:
1547         if (displaced)
1548                 reset_files_struct(displaced);
1549 out_ret:
1550         return retval;
1551 }
1552
1553 int do_execve(const char *filename,
1554         const char __user *const __user *__argv,
1555         const char __user *const __user *__envp,
1556         struct pt_regs *regs)
1557 {
1558         struct user_arg_ptr argv = { .ptr.native = __argv };
1559         struct user_arg_ptr envp = { .ptr.native = __envp };
1560         return do_execve_common(filename, argv, envp, regs);
1561 }
1562
1563 #ifdef CONFIG_COMPAT
1564 int compat_do_execve(char *filename,
1565         compat_uptr_t __user *__argv,
1566         compat_uptr_t __user *__envp,
1567         struct pt_regs *regs)
1568 {
1569         struct user_arg_ptr argv = {
1570                 .is_compat = true,
1571                 .ptr.compat = __argv,
1572         };
1573         struct user_arg_ptr envp = {
1574                 .is_compat = true,
1575                 .ptr.compat = __envp,
1576         };
1577         return do_execve_common(filename, argv, envp, regs);
1578 }
1579 #endif
1580
1581 void set_binfmt(struct linux_binfmt *new)
1582 {
1583         struct mm_struct *mm = current->mm;
1584
1585         if (mm->binfmt)
1586                 module_put(mm->binfmt->module);
1587
1588         mm->binfmt = new;
1589         if (new)
1590                 __module_get(new->module);
1591 }
1592
1593 EXPORT_SYMBOL(set_binfmt);
1594
1595 /*
1596  * set_dumpable converts traditional three-value dumpable to two flags and
1597  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1598  * these bits are not changed atomically.  So get_dumpable can observe the
1599  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1600  * return either old dumpable or new one by paying attention to the order of
1601  * modifying the bits.
1602  *
1603  * dumpable |   mm->flags (binary)
1604  * old  new | initial interim  final
1605  * ---------+-----------------------
1606  *  0    1  |   00      01      01
1607  *  0    2  |   00      10(*)   11
1608  *  1    0  |   01      00      00
1609  *  1    2  |   01      11      11
1610  *  2    0  |   11      10(*)   00
1611  *  2    1  |   11      11      01
1612  *
1613  * (*) get_dumpable regards interim value of 10 as 11.
1614  */
1615 void set_dumpable(struct mm_struct *mm, int value)
1616 {
1617         switch (value) {
1618         case SUID_DUMPABLE_DISABLED:
1619                 clear_bit(MMF_DUMPABLE, &mm->flags);
1620                 smp_wmb();
1621                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1622                 break;
1623         case SUID_DUMPABLE_ENABLED:
1624                 set_bit(MMF_DUMPABLE, &mm->flags);
1625                 smp_wmb();
1626                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1627                 break;
1628         case SUID_DUMPABLE_SAFE:
1629                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1630                 smp_wmb();
1631                 set_bit(MMF_DUMPABLE, &mm->flags);
1632                 break;
1633         }
1634 }
1635
1636 int __get_dumpable(unsigned long mm_flags)
1637 {
1638         int ret;
1639
1640         ret = mm_flags & MMF_DUMPABLE_MASK;
1641         return (ret > SUID_DUMPABLE_ENABLED) ? SUID_DUMPABLE_SAFE : ret;
1642 }
1643
1644 int get_dumpable(struct mm_struct *mm)
1645 {
1646         return __get_dumpable(mm->flags);
1647 }