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
  63 #include <trace/events/task.h>
  64 #include "internal.h"
  65 #include "coredump.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         struct filename *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                 const compat_uptr_t __user *compat;
 395 #endif
 396         } ptr;
 397 };
 398
 399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 400 {
 401         const char __user *native;
 402
 403 #ifdef CONFIG_COMPAT
 404         if (unlikely(argv.is_compat)) {
 405                 compat_uptr_t compat;
 406
 407                 if (get_user(compat, argv.ptr.compat + nr))
 408                         return ERR_PTR(-EFAULT);
 409
 410                 return compat_ptr(compat);
 411         }
 412 #endif
 413
 414         if (get_user(native, argv.ptr.native + nr))
 415                 return ERR_PTR(-EFAULT);
 416
 417         return native;
 418 }
 419
 420 /*
 421  * count() counts the number of strings in array ARGV.
 422  */
 423 static int count(struct user_arg_ptr argv, int max)
 424 {
 425         int i = 0;
 426
 427         if (argv.ptr.native != NULL) {
 428                 for (;;) {
 429                         const char __user *p = get_user_arg_ptr(argv, i);
 430
 431                         if (!p)
 432                                 break;
 433
 434                         if (IS_ERR(p))
 435                                 return -EFAULT;
 436
 437                         if (i++ >= max)
 438                                 return -E2BIG;
 439
 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, false))
 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         struct filename tmp = { .name = name };
 755         static const struct open_flags open_exec_flags = {
 756                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 757                 .acc_mode = MAY_EXEC | MAY_OPEN,
 758                 .intent = LOOKUP_OPEN
 759         };
 760
 761         file = do_filp_open(AT_FDCWD, &tmp, &open_exec_flags, LOOKUP_FOLLOW);
 762         if (IS_ERR(file))
 763                 goto out;
 764
 765         err = -EACCES;
 766         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 767                 goto exit;
 768
 769         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 770                 goto exit;
 771
 772         fsnotify_open(file);
 773
 774         err = deny_write_access(file);
 775         if (err)
 776                 goto exit;
 777
 778 out:
 779         return file;
 780
 781 exit:
 782         fput(file);
 783         return ERR_PTR(err);
 784 }
 785 EXPORT_SYMBOL(open_exec);
 786
 787 int kernel_read(struct file *file, loff_t offset,
 788                 char *addr, unsigned long count)
 789 {
 790         mm_segment_t old_fs;
 791         loff_t pos = offset;
 792         int result;
 793
 794         old_fs = get_fs();
 795         set_fs(get_ds());
 796         /* The cast to a user pointer is valid due to the set_fs() */
 797         result = vfs_read(file, (void __user *)addr, count, &pos);
 798         set_fs(old_fs);
 799         return result;
 800 }
 801
 802 EXPORT_SYMBOL(kernel_read);
 803
 804 static int exec_mmap(struct mm_struct *mm)
 805 {
 806         struct task_struct *tsk;
 807         struct mm_struct * old_mm, *active_mm;
 808
 809         /* Notify parent that we're no longer interested in the old VM */
 810         tsk = current;
 811         old_mm = current->mm;
 812         mm_release(tsk, old_mm);
 813
 814         if (old_mm) {
 815                 sync_mm_rss(old_mm);
 816                 /*
 817                  * Make sure that if there is a core dump in progress
 818                  * for the old mm, we get out and die instead of going
 819                  * through with the exec.  We must hold mmap_sem around
 820                  * checking core_state and changing tsk->mm.
 821                  */
 822                 down_read(&old_mm->mmap_sem);
 823                 if (unlikely(old_mm->core_state)) {
 824                         up_read(&old_mm->mmap_sem);
 825                         return -EINTR;
 826                 }
 827         }
 828         task_lock(tsk);
 829         active_mm = tsk->active_mm;
 830         tsk->mm = mm;
 831         tsk->active_mm = mm;
 832         activate_mm(active_mm, mm);
 833         task_unlock(tsk);
 834         arch_pick_mmap_layout(mm);
 835         if (old_mm) {
 836                 up_read(&old_mm->mmap_sem);
 837                 BUG_ON(active_mm != old_mm);
 838                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 839                 mm_update_next_owner(old_mm);
 840                 mmput(old_mm);
 841                 return 0;
 842         }
 843         mmdrop(active_mm);
 844         return 0;
 845 }
 846
 847 /*
 848  * This function makes sure the current process has its own signal table,
 849  * so that flush_signal_handlers can later reset the handlers without
 850  * disturbing other processes.  (Other processes might share the signal
 851  * table via the CLONE_SIGHAND option to clone().)
 852  */
 853 static int de_thread(struct task_struct *tsk)
 854 {
 855         struct signal_struct *sig = tsk->signal;
 856         struct sighand_struct *oldsighand = tsk->sighand;
 857         spinlock_t *lock = &oldsighand->siglock;
 858
 859         if (thread_group_empty(tsk))
 860                 goto no_thread_group;
 861
 862         /*
 863          * Kill all other threads in the thread group.
 864          */
 865         spin_lock_irq(lock);
 866         if (signal_group_exit(sig)) {
 867                 /*
 868                  * Another group action in progress, just
 869                  * return so that the signal is processed.
 870                  */
 871                 spin_unlock_irq(lock);
 872                 return -EAGAIN;
 873         }
 874
 875         sig->group_exit_task = tsk;
 876         sig->notify_count = zap_other_threads(tsk);
 877         if (!thread_group_leader(tsk))
 878                 sig->notify_count--;
 879
 880         while (sig->notify_count) {
 881                 __set_current_state(TASK_KILLABLE);
 882                 spin_unlock_irq(lock);
 883                 schedule();
 884                 if (unlikely(__fatal_signal_pending(tsk)))
 885                         goto killed;
 886                 spin_lock_irq(lock);
 887         }
 888         spin_unlock_irq(lock);
 889
 890         /*
 891          * At this point all other threads have exited, all we have to
 892          * do is to wait for the thread group leader to become inactive,
 893          * and to assume its PID:
 894          */
 895         if (!thread_group_leader(tsk)) {
 896                 struct task_struct *leader = tsk->group_leader;
 897
 898                 sig->notify_count = -1; /* for exit_notify() */
 899                 for (;;) {
 900                         write_lock_irq(&tasklist_lock);
 901                         if (likely(leader->exit_state))
 902                                 break;
 903                         __set_current_state(TASK_KILLABLE);
 904                         write_unlock_irq(&tasklist_lock);
 905                         schedule();
 906                         if (unlikely(__fatal_signal_pending(tsk)))
 907                                 goto killed;
 908                 }
 909
 910                 /*
 911                  * The only record we have of the real-time age of a
 912                  * process, regardless of execs it's done, is start_time.
 913                  * All the past CPU time is accumulated in signal_struct
 914                  * from sister threads now dead.  But in this non-leader
 915                  * exec, nothing survives from the original leader thread,
 916                  * whose birth marks the true age of this process now.
 917                  * When we take on its identity by switching to its PID, we
 918                  * also take its birthdate (always earlier than our own).
 919                  */
 920                 tsk->start_time = leader->start_time;
 921
 922                 BUG_ON(!same_thread_group(leader, tsk));
 923                 BUG_ON(has_group_leader_pid(tsk));
 924                 /*
 925                  * An exec() starts a new thread group with the
 926                  * TGID of the previous thread group. Rehash the
 927                  * two threads with a switched PID, and release
 928                  * the former thread group leader:
 929                  */
 930
 931                 /* Become a process group leader with the old leader's pid.
 932                  * The old leader becomes a thread of the this thread group.
 933                  * Note: The old leader also uses this pid until release_task
 934                  *       is called.  Odd but simple and correct.
 935                  */
 936                 detach_pid(tsk, PIDTYPE_PID);
 937                 tsk->pid = leader->pid;
 938                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 939                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 940                 transfer_pid(leader, tsk, PIDTYPE_SID);
 941
 942                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 943                 list_replace_init(&leader->sibling, &tsk->sibling);
 944
 945                 tsk->group_leader = tsk;
 946                 leader->group_leader = tsk;
 947
 948                 tsk->exit_signal = SIGCHLD;
 949                 leader->exit_signal = -1;
 950
 951                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 952                 leader->exit_state = EXIT_DEAD;
 953
 954                 /*
 955                  * We are going to release_task()->ptrace_unlink() silently,
 956                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 957                  * the tracer wont't block again waiting for this thread.
 958                  */
 959                 if (unlikely(leader->ptrace))
 960                         __wake_up_parent(leader, leader->parent);
 961                 write_unlock_irq(&tasklist_lock);
 962
 963                 release_task(leader);
 964         }
 965
 966         sig->group_exit_task = NULL;
 967         sig->notify_count = 0;
 968
 969 no_thread_group:
 970         /* we have changed execution domain */
 971         tsk->exit_signal = SIGCHLD;
 972
 973         exit_itimers(sig);
 974         flush_itimer_signals();
 975
 976         if (atomic_read(&oldsighand->count) != 1) {
 977                 struct sighand_struct *newsighand;
 978                 /*
 979                  * This ->sighand is shared with the CLONE_SIGHAND
 980                  * but not CLONE_THREAD task, switch to the new one.
 981                  */
 982                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 983                 if (!newsighand)
 984                         return -ENOMEM;
 985
 986                 atomic_set(&newsighand->count, 1);
 987                 memcpy(newsighand->action, oldsighand->action,
 988                        sizeof(newsighand->action));
 989
 990                 write_lock_irq(&tasklist_lock);
 991                 spin_lock(&oldsighand->siglock);
 992                 rcu_assign_pointer(tsk->sighand, newsighand);
 993                 spin_unlock(&oldsighand->siglock);
 994                 write_unlock_irq(&tasklist_lock);
 995
 996                 __cleanup_sighand(oldsighand);
 997         }
 998
 999         BUG_ON(!thread_group_leader(tsk));
1000         return 0;
1001
1002 killed:
1003         /* protects against exit_notify() and __exit_signal() */
1004         read_lock(&tasklist_lock);
1005         sig->group_exit_task = NULL;
1006         sig->notify_count = 0;
1007         read_unlock(&tasklist_lock);
1008         return -EAGAIN;
1009 }
1010
1011 char *get_task_comm(char *buf, struct task_struct *tsk)
1012 {
1013         /* buf must be at least sizeof(tsk->comm) in size */
1014         task_lock(tsk);
1015         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1016         task_unlock(tsk);
1017         return buf;
1018 }
1019 EXPORT_SYMBOL_GPL(get_task_comm);
1020
1021 /*
1022  * These functions flushes out all traces of the currently running executable
1023  * so that a new one can be started
1024  */
1025
1026 void set_task_comm(struct task_struct *tsk, char *buf)
1027 {
1028         task_lock(tsk);
1029
1030         trace_task_rename(tsk, buf);
1031
1032         /*
1033          * Threads may access current->comm without holding
1034          * the task lock, so write the string carefully.
1035          * Readers without a lock may see incomplete new
1036          * names but are safe from non-terminating string reads.
1037          */
1038         memset(tsk->comm, 0, TASK_COMM_LEN);
1039         wmb();
1040         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1041         task_unlock(tsk);
1042         perf_event_comm(tsk);
1043 }
1044
1045 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1046 {
1047         int i, ch;
1048
1049         /* Copies the binary name from after last slash */
1050         for (i = 0; (ch = *(fn++)) != '\0';) {
1051                 if (ch == '/')
1052                         i = 0; /* overwrite what we wrote */
1053                 else
1054                         if (i < len - 1)
1055                                 tcomm[i++] = ch;
1056         }
1057         tcomm[i] = '\0';
1058 }
1059
1060 int flush_old_exec(struct linux_binprm * bprm)
1061 {
1062         int retval;
1063
1064         /*
1065          * Make sure we have a private signal table and that
1066          * we are unassociated from the previous thread group.
1067          */
1068         retval = de_thread(current);
1069         if (retval)
1070                 goto out;
1071
1072         set_mm_exe_file(bprm->mm, bprm->file);
1073
1074         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1075         /*
1076          * Release all of the old mmap stuff
1077          */
1078         acct_arg_size(bprm, 0);
1079         retval = exec_mmap(bprm->mm);
1080         if (retval)
1081                 goto out;
1082
1083         bprm->mm = NULL;                /* We're using it now */
1084
1085         set_fs(USER_DS);
1086         current->flags &=
1087                 ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | PF_NOFREEZE);
1088         flush_thread();
1089         current->personality &= ~bprm->per_clear;
1090
1091         return 0;
1092
1093 out:
1094         return retval;
1095 }
1096 EXPORT_SYMBOL(flush_old_exec);
1097
1098 void would_dump(struct linux_binprm *bprm, struct file *file)
1099 {
1100         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1101                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1102 }
1103 EXPORT_SYMBOL(would_dump);
1104
1105 void setup_new_exec(struct linux_binprm * bprm)
1106 {
1107         arch_pick_mmap_layout(current->mm);
1108
1109         /* This is the point of no return */
1110         current->sas_ss_sp = current->sas_ss_size = 0;
1111
1112         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1113                 set_dumpable(current->mm, SUID_DUMPABLE_ENABLED);
1114         else
1115                 set_dumpable(current->mm, suid_dumpable);
1116
1117         set_task_comm(current, bprm->tcomm);
1118
1119         /* Set the new mm task size. We have to do that late because it may
1120          * depend on TIF_32BIT which is only updated in flush_thread() on
1121          * some architectures like powerpc
1122          */
1123         current->mm->task_size = TASK_SIZE;
1124
1125         /* install the new credentials */
1126         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1127             !gid_eq(bprm->cred->gid, current_egid())) {
1128                 current->pdeath_signal = 0;
1129         } else {
1130                 would_dump(bprm, bprm->file);
1131                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1132                         set_dumpable(current->mm, suid_dumpable);
1133         }
1134
1135         /*
1136          * Flush performance counters when crossing a
1137          * security domain:
1138          */
1139         if (!get_dumpable(current->mm))
1140                 perf_event_exit_task(current);
1141
1142         /* An exec changes our domain. We are no longer part of the thread
1143            group */
1144
1145         current->self_exec_id++;
1146
1147         flush_signal_handlers(current, 0);
1148         do_close_on_exec(current->files);
1149 }
1150 EXPORT_SYMBOL(setup_new_exec);
1151
1152 /*
1153  * Prepare credentials and lock ->cred_guard_mutex.
1154  * install_exec_creds() commits the new creds and drops the lock.
1155  * Or, if exec fails before, free_bprm() should release ->cred and
1156  * and unlock.
1157  */
1158 int prepare_bprm_creds(struct linux_binprm *bprm)
1159 {
1160         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1161                 return -ERESTARTNOINTR;
1162
1163         bprm->cred = prepare_exec_creds();
1164         if (likely(bprm->cred))
1165                 return 0;
1166
1167         mutex_unlock(&current->signal->cred_guard_mutex);
1168         return -ENOMEM;
1169 }
1170
1171 void free_bprm(struct linux_binprm *bprm)
1172 {
1173         free_arg_pages(bprm);
1174         if (bprm->cred) {
1175                 mutex_unlock(&current->signal->cred_guard_mutex);
1176                 abort_creds(bprm->cred);
1177         }
1178         kfree(bprm);
1179 }
1180
1181 /*
1182  * install the new credentials for this executable
1183  */
1184 void install_exec_creds(struct linux_binprm *bprm)
1185 {
1186         security_bprm_committing_creds(bprm);
1187
1188         commit_creds(bprm->cred);
1189         bprm->cred = NULL;
1190         /*
1191          * cred_guard_mutex must be held at least to this point to prevent
1192          * ptrace_attach() from altering our determination of the task's
1193          * credentials; any time after this it may be unlocked.
1194          */
1195         security_bprm_committed_creds(bprm);
1196         mutex_unlock(&current->signal->cred_guard_mutex);
1197 }
1198 EXPORT_SYMBOL(install_exec_creds);
1199
1200 /*
1201  * determine how safe it is to execute the proposed program
1202  * - the caller must hold ->cred_guard_mutex to protect against
1203  *   PTRACE_ATTACH
1204  */
1205 static int check_unsafe_exec(struct linux_binprm *bprm)
1206 {
1207         struct task_struct *p = current, *t;
1208         unsigned n_fs;
1209         int res = 0;
1210
1211         if (p->ptrace) {
1212                 if (p->ptrace & PT_PTRACE_CAP)
1213                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1214                 else
1215                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1216         }
1217
1218         /*
1219          * This isn't strictly necessary, but it makes it harder for LSMs to
1220          * mess up.
1221          */
1222         if (current->no_new_privs)
1223                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1224
1225         n_fs = 1;
1226         spin_lock(&p->fs->lock);
1227         rcu_read_lock();
1228         for (t = next_thread(p); t != p; t = next_thread(t)) {
1229                 if (t->fs == p->fs)
1230                         n_fs++;
1231         }
1232         rcu_read_unlock();
1233
1234         if (p->fs->users > n_fs) {
1235                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1236         } else {
1237                 res = -EAGAIN;
1238                 if (!p->fs->in_exec) {
1239                         p->fs->in_exec = 1;
1240                         res = 1;
1241                 }
1242         }
1243         spin_unlock(&p->fs->lock);
1244
1245         return res;
1246 }
1247
1248 /*
1249  * Fill the binprm structure from the inode.
1250  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1251  *
1252  * This may be called multiple times for binary chains (scripts for example).
1253  */
1254 int prepare_binprm(struct linux_binprm *bprm)
1255 {
1256         umode_t mode;
1257         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1258         int retval;
1259
1260         mode = inode->i_mode;
1261         if (bprm->file->f_op == NULL)
1262                 return -EACCES;
1263
1264         /* clear any previous set[ug]id data from a previous binary */
1265         bprm->cred->euid = current_euid();
1266         bprm->cred->egid = current_egid();
1267
1268         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1269             !current->no_new_privs) {
1270                 /* Set-uid? */
1271                 if (mode & S_ISUID) {
1272                         if (!kuid_has_mapping(bprm->cred->user_ns, inode->i_uid))
1273                                 return -EPERM;
1274                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1275                         bprm->cred->euid = inode->i_uid;
1276
1277                 }
1278
1279                 /* Set-gid? */
1280                 /*
1281                  * If setgid is set but no group execute bit then this
1282                  * is a candidate for mandatory locking, not a setgid
1283                  * executable.
1284                  */
1285                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1286                         if (!kgid_has_mapping(bprm->cred->user_ns, inode->i_gid))
1287                                 return -EPERM;
1288                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1289                         bprm->cred->egid = inode->i_gid;
1290                 }
1291         }
1292
1293         /* fill in binprm security blob */
1294         retval = security_bprm_set_creds(bprm);
1295         if (retval)
1296                 return retval;
1297         bprm->cred_prepared = 1;
1298
1299         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1300         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1301 }
1302
1303 EXPORT_SYMBOL(prepare_binprm);
1304
1305 /*
1306  * Arguments are '\0' separated strings found at the location bprm->p
1307  * points to; chop off the first by relocating brpm->p to right after
1308  * the first '\0' encountered.
1309  */
1310 int remove_arg_zero(struct linux_binprm *bprm)
1311 {
1312         int ret = 0;
1313         unsigned long offset;
1314         char *kaddr;
1315         struct page *page;
1316
1317         if (!bprm->argc)
1318                 return 0;
1319
1320         do {
1321                 offset = bprm->p & ~PAGE_MASK;
1322                 page = get_arg_page(bprm, bprm->p, 0);
1323                 if (!page) {
1324                         ret = -EFAULT;
1325                         goto out;
1326                 }
1327                 kaddr = kmap_atomic(page);
1328
1329                 for (; offset < PAGE_SIZE && kaddr[offset];
1330                                 offset++, bprm->p++)
1331                         ;
1332
1333                 kunmap_atomic(kaddr);
1334                 put_arg_page(page);
1335
1336                 if (offset == PAGE_SIZE)
1337                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1338         } while (offset == PAGE_SIZE);
1339
1340         bprm->p++;
1341         bprm->argc--;
1342         ret = 0;
1343
1344 out:
1345         return ret;
1346 }
1347 EXPORT_SYMBOL(remove_arg_zero);
1348
1349 /*
1350  * cycle the list of binary formats handler, until one recognizes the image
1351  */
1352 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1353 {
1354         unsigned int depth = bprm->recursion_depth;
1355         int try,retval;
1356         struct linux_binfmt *fmt;
1357         pid_t old_pid, old_vpid;
1358
1359         retval = security_bprm_check(bprm);
1360         if (retval)
1361                 return retval;
1362
1363         retval = audit_bprm(bprm);
1364         if (retval)
1365                 return retval;
1366
1367         /* Need to fetch pid before load_binary changes it */
1368         old_pid = current->pid;
1369         rcu_read_lock();
1370         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1371         rcu_read_unlock();
1372
1373         retval = -ENOENT;
1374         for (try=0; try<2; try++) {
1375                 read_lock(&binfmt_lock);
1376                 list_for_each_entry(fmt, &formats, lh) {
1377                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1378                         if (!fn)
1379                                 continue;
1380                         if (!try_module_get(fmt->module))
1381                                 continue;
1382                         read_unlock(&binfmt_lock);
1383                         retval = fn(bprm, regs);
1384                         /*
1385                          * Restore the depth counter to its starting value
1386                          * in this call, so we don't have to rely on every
1387                          * load_binary function to restore it on return.
1388                          */
1389                         bprm->recursion_depth = depth;
1390                         if (retval >= 0) {
1391                                 if (depth == 0) {
1392                                         trace_sched_process_exec(current, old_pid, bprm);
1393                                         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1394                                 }
1395                                 put_binfmt(fmt);
1396                                 allow_write_access(bprm->file);
1397                                 if (bprm->file)
1398                                         fput(bprm->file);
1399                                 bprm->file = NULL;
1400                                 current->did_exec = 1;
1401                                 proc_exec_connector(current);
1402                                 return retval;
1403                         }
1404                         read_lock(&binfmt_lock);
1405                         put_binfmt(fmt);
1406                         if (retval != -ENOEXEC || bprm->mm == NULL)
1407                                 break;
1408                         if (!bprm->file) {
1409                                 read_unlock(&binfmt_lock);
1410                                 return retval;
1411                         }
1412                 }
1413                 read_unlock(&binfmt_lock);
1414 #ifdef CONFIG_MODULES
1415                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1416                         break;
1417                 } else {
1418 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1419                         if (printable(bprm->buf[0]) &&
1420                             printable(bprm->buf[1]) &&
1421                             printable(bprm->buf[2]) &&
1422                             printable(bprm->buf[3]))
1423                                 break; /* -ENOEXEC */
1424                         if (try)
1425                                 break; /* -ENOEXEC */
1426                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1427                 }
1428 #else
1429                 break;
1430 #endif
1431         }
1432         return retval;
1433 }
1434
1435 EXPORT_SYMBOL(search_binary_handler);
1436
1437 /*
1438  * sys_execve() executes a new program.
1439  */
1440 static int do_execve_common(const char *filename,
1441                                 struct user_arg_ptr argv,
1442                                 struct user_arg_ptr envp,
1443                                 struct pt_regs *regs)
1444 {
1445         struct linux_binprm *bprm;
1446         struct file *file;
1447         struct files_struct *displaced;
1448         bool clear_in_exec;
1449         int retval;
1450         const struct cred *cred = current_cred();
1451
1452         /*
1453          * We move the actual failure in case of RLIMIT_NPROC excess from
1454          * set*uid() to execve() because too many poorly written programs
1455          * don't check setuid() return code.  Here we additionally recheck
1456          * whether NPROC limit is still exceeded.
1457          */
1458         if ((current->flags & PF_NPROC_EXCEEDED) &&
1459             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1460                 retval = -EAGAIN;
1461                 goto out_ret;
1462         }
1463
1464         /* We're below the limit (still or again), so we don't want to make
1465          * further execve() calls fail. */
1466         current->flags &= ~PF_NPROC_EXCEEDED;
1467
1468         retval = unshare_files(&displaced);
1469         if (retval)
1470                 goto out_ret;
1471
1472         retval = -ENOMEM;
1473         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1474         if (!bprm)
1475                 goto out_files;
1476
1477         retval = prepare_bprm_creds(bprm);
1478         if (retval)
1479                 goto out_free;
1480
1481         retval = check_unsafe_exec(bprm);
1482         if (retval < 0)
1483                 goto out_free;
1484         clear_in_exec = retval;
1485         current->in_execve = 1;
1486
1487         file = open_exec(filename);
1488         retval = PTR_ERR(file);
1489         if (IS_ERR(file))
1490                 goto out_unmark;
1491
1492         sched_exec();
1493
1494         bprm->file = file;
1495         bprm->filename = filename;
1496         bprm->interp = filename;
1497
1498         retval = bprm_mm_init(bprm);
1499         if (retval)
1500                 goto out_file;
1501
1502         bprm->argc = count(argv, MAX_ARG_STRINGS);
1503         if ((retval = bprm->argc) < 0)
1504                 goto out;
1505
1506         bprm->envc = count(envp, MAX_ARG_STRINGS);
1507         if ((retval = bprm->envc) < 0)
1508                 goto out;
1509
1510         retval = prepare_binprm(bprm);
1511         if (retval < 0)
1512                 goto out;
1513
1514         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1515         if (retval < 0)
1516                 goto out;
1517
1518         bprm->exec = bprm->p;
1519         retval = copy_strings(bprm->envc, envp, bprm);
1520         if (retval < 0)
1521                 goto out;
1522
1523         retval = copy_strings(bprm->argc, argv, bprm);
1524         if (retval < 0)
1525                 goto out;
1526
1527         retval = search_binary_handler(bprm,regs);
1528         if (retval < 0)
1529                 goto out;
1530
1531         /* execve succeeded */
1532         current->fs->in_exec = 0;
1533         current->in_execve = 0;
1534         acct_update_integrals(current);
1535         free_bprm(bprm);
1536         if (displaced)
1537                 put_files_struct(displaced);
1538         return retval;
1539
1540 out:
1541         if (bprm->mm) {
1542                 acct_arg_size(bprm, 0);
1543                 mmput(bprm->mm);
1544         }
1545
1546 out_file:
1547         if (bprm->file) {
1548                 allow_write_access(bprm->file);
1549                 fput(bprm->file);
1550         }
1551
1552 out_unmark:
1553         if (clear_in_exec)
1554                 current->fs->in_exec = 0;
1555         current->in_execve = 0;
1556
1557 out_free:
1558         free_bprm(bprm);
1559
1560 out_files:
1561         if (displaced)
1562                 reset_files_struct(displaced);
1563 out_ret:
1564         return retval;
1565 }
1566
1567 int do_execve(const char *filename,
1568         const char __user *const __user *__argv,
1569         const char __user *const __user *__envp,
1570         struct pt_regs *regs)
1571 {
1572         struct user_arg_ptr argv = { .ptr.native = __argv };
1573         struct user_arg_ptr envp = { .ptr.native = __envp };
1574         return do_execve_common(filename, argv, envp, regs);
1575 }
1576
1577 #ifdef CONFIG_COMPAT
1578 int compat_do_execve(const char *filename,
1579         const compat_uptr_t __user *__argv,
1580         const compat_uptr_t __user *__envp,
1581         struct pt_regs *regs)
1582 {
1583         struct user_arg_ptr argv = {
1584                 .is_compat = true,
1585                 .ptr.compat = __argv,
1586         };
1587         struct user_arg_ptr envp = {
1588                 .is_compat = true,
1589                 .ptr.compat = __envp,
1590         };
1591         return do_execve_common(filename, argv, envp, regs);
1592 }
1593 #endif
1594
1595 void set_binfmt(struct linux_binfmt *new)
1596 {
1597         struct mm_struct *mm = current->mm;
1598
1599         if (mm->binfmt)
1600                 module_put(mm->binfmt->module);
1601
1602         mm->binfmt = new;
1603         if (new)
1604                 __module_get(new->module);
1605 }
1606
1607 EXPORT_SYMBOL(set_binfmt);
1608
1609 /*
1610  * set_dumpable converts traditional three-value dumpable to two flags and
1611  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1612  * these bits are not changed atomically.  So get_dumpable can observe the
1613  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1614  * return either old dumpable or new one by paying attention to the order of
1615  * modifying the bits.
1616  *
1617  * dumpable |   mm->flags (binary)
1618  * old  new | initial interim  final
1619  * ---------+-----------------------
1620  *  0    1  |   00      01      01
1621  *  0    2  |   00      10(*)   11
1622  *  1    0  |   01      00      00
1623  *  1    2  |   01      11      11
1624  *  2    0  |   11      10(*)   00
1625  *  2    1  |   11      11      01
1626  *
1627  * (*) get_dumpable regards interim value of 10 as 11.
1628  */
1629 void set_dumpable(struct mm_struct *mm, int value)
1630 {
1631         switch (value) {
1632         case SUID_DUMPABLE_DISABLED:
1633                 clear_bit(MMF_DUMPABLE, &mm->flags);
1634                 smp_wmb();
1635                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1636                 break;
1637         case SUID_DUMPABLE_ENABLED:
1638                 set_bit(MMF_DUMPABLE, &mm->flags);
1639                 smp_wmb();
1640                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1641                 break;
1642         case SUID_DUMPABLE_SAFE:
1643                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1644                 smp_wmb();
1645                 set_bit(MMF_DUMPABLE, &mm->flags);
1646                 break;
1647         }
1648 }
1649
1650 int __get_dumpable(unsigned long mm_flags)
1651 {
1652         int ret;
1653
1654         ret = mm_flags & MMF_DUMPABLE_MASK;
1655         return (ret > SUID_DUMPABLE_ENABLED) ? SUID_DUMPABLE_SAFE : ret;
1656 }
1657
1658 int get_dumpable(struct mm_struct *mm)
1659 {
1660         return __get_dumpable(mm->flags);
1661 }
1662
1663 #ifdef __ARCH_WANT_SYS_EXECVE
1664 SYSCALL_DEFINE3(execve,
1665                 const char __user *, filename,
1666                 const char __user *const __user *, argv,
1667                 const char __user *const __user *, envp)
1668 {
1669         struct filename *path = getname(filename);
1670         int error = PTR_ERR(path);
1671         if (!IS_ERR(path)) {
1672                 error = do_execve(path->name, argv, envp, current_pt_regs());
1673                 putname(path);
1674         }
1675         return error;
1676 }
1677 #ifdef CONFIG_COMPAT
1678 asmlinkage long compat_sys_execve(const char __user * filename,
1679         const compat_uptr_t __user * argv,
1680         const compat_uptr_t __user * envp)
1681 {
1682         struct filename *path = getname(filename);
1683         int error = PTR_ERR(path);
1684         if (!IS_ERR(path)) {
1685                 error = compat_do_execve(path->name, argv, envp,
1686                                                         current_pt_regs());
1687                 putname(path);
1688         }
1689         return error;
1690 }
1691 #endif
1692 #endif
1693
1694 #ifdef __ARCH_WANT_KERNEL_EXECVE
1695 int kernel_execve(const char *filename,
1696                   const char *const argv[],
1697                   const char *const envp[])
1698 {
1699         struct pt_regs *p = current_pt_regs();
1700         int ret;
1701
1702         ret = do_execve(filename,
1703                         (const char __user *const __user *)argv,
1704                         (const char __user *const __user *)envp, p);
1705         if (ret < 0)
1706                 return ret;
1707
1708         /*
1709          * We were successful.  We won't be returning to our caller, but
1710          * instead to user space by manipulating the kernel stack.
1711          */
1712         ret_from_kernel_execve(p);
1713 }
1714 #endif