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/smp_lock.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/proc_fs.h>
  47 #include <linux/mount.h>
  48 #include <linux/security.h>
  49 #include <linux/syscalls.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/tracehook.h>
  54 #include <linux/kmod.h>
  55 #include <linux/fsnotify.h>
  56 #include <linux/fs_struct.h>
  57 #include <linux/pipe_fs_i.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62 #include "internal.h"
  63
  64 int core_uses_pid;
  65 char core_pattern[CORENAME_MAX_SIZE] = "core";
  66 unsigned int core_pipe_limit;
  67 int suid_dumpable = 0;
  68
  69 /* The maximal length of core_pattern is also specified in sysctl.c */
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         if (!fmt)
  77                 return -EINVAL;
  78         write_lock(&binfmt_lock);
  79         insert ? list_add(&fmt->lh, &formats) :
  80                  list_add_tail(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82         return 0;
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 SYSCALL_DEFINE1(uselib, const char __user *, library)
 108 {
 109         struct file *file;
 110         char *tmp = getname(library);
 111         int error = PTR_ERR(tmp);
 112
 113         if (IS_ERR(tmp))
 114                 goto out;
 115
 116         file = do_filp_open(AT_FDCWD, tmp,
 117                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 118                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 119         putname(tmp);
 120         error = PTR_ERR(file);
 121         if (IS_ERR(file))
 122                 goto out;
 123
 124         error = -EINVAL;
 125         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 126                 goto exit;
 127
 128         error = -EACCES;
 129         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 130                 goto exit;
 131
 132         fsnotify_open(file->f_path.dentry);
 133
 134         error = -ENOEXEC;
 135         if(file->f_op) {
 136                 struct linux_binfmt * fmt;
 137
 138                 read_lock(&binfmt_lock);
 139                 list_for_each_entry(fmt, &formats, lh) {
 140                         if (!fmt->load_shlib)
 141                                 continue;
 142                         if (!try_module_get(fmt->module))
 143                                 continue;
 144                         read_unlock(&binfmt_lock);
 145                         error = fmt->load_shlib(file);
 146                         read_lock(&binfmt_lock);
 147                         put_binfmt(fmt);
 148                         if (error != -ENOEXEC)
 149                                 break;
 150                 }
 151                 read_unlock(&binfmt_lock);
 152         }
 153 exit:
 154         fput(file);
 155 out:
 156         return error;
 157 }
 158
 159 #ifdef CONFIG_MMU
 160
 161 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 162 {
 163         struct mm_struct *mm = current->mm;
 164         long diff = (long)(pages - bprm->vma_pages);
 165
 166         if (!mm || !diff)
 167                 return;
 168
 169         bprm->vma_pages = pages;
 170
 171         down_write(&mm->mmap_sem);
 172         mm->total_vm += diff;
 173         up_write(&mm->mmap_sem);
 174 }
 175
 176 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 177                 int write)
 178 {
 179         struct page *page;
 180         int ret;
 181
 182 #ifdef CONFIG_STACK_GROWSUP
 183         if (write) {
 184                 ret = expand_stack_downwards(bprm->vma, pos);
 185                 if (ret < 0)
 186                         return NULL;
 187         }
 188 #endif
 189         ret = get_user_pages(current, bprm->mm, pos,
 190                         1, write, 1, &page, NULL);
 191         if (ret <= 0)
 192                 return NULL;
 193
 194         if (write) {
 195                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 196                 struct rlimit *rlim;
 197
 198                 acct_arg_size(bprm, size / PAGE_SIZE);
 199
 200                 /*
 201                  * We've historically supported up to 32 pages (ARG_MAX)
 202                  * of argument strings even with small stacks
 203                  */
 204                 if (size <= ARG_MAX)
 205                         return page;
 206
 207                 /*
 208                  * Limit to 1/4-th the stack size for the argv+env strings.
 209                  * This ensures that:
 210                  *  - the remaining binfmt code will not run out of stack space,
 211                  *  - the program will have a reasonable amount of stack left
 212                  *    to work from.
 213                  */
 214                 rlim = current->signal->rlim;
 215                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 216                         put_page(page);
 217                         return NULL;
 218                 }
 219         }
 220
 221         return page;
 222 }
 223
 224 static void put_arg_page(struct page *page)
 225 {
 226         put_page(page);
 227 }
 228
 229 static void free_arg_page(struct linux_binprm *bprm, int i)
 230 {
 231 }
 232
 233 static void free_arg_pages(struct linux_binprm *bprm)
 234 {
 235 }
 236
 237 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 238                 struct page *page)
 239 {
 240         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 241 }
 242
 243 static int __bprm_mm_init(struct linux_binprm *bprm)
 244 {
 245         int err;
 246         struct vm_area_struct *vma = NULL;
 247         struct mm_struct *mm = bprm->mm;
 248
 249         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 250         if (!vma)
 251                 return -ENOMEM;
 252
 253         down_write(&mm->mmap_sem);
 254         vma->vm_mm = mm;
 255
 256         /*
 257          * Place the stack at the largest stack address the architecture
 258          * supports. Later, we'll move this to an appropriate place. We don't
 259          * use STACK_TOP because that can depend on attributes which aren't
 260          * configured yet.
 261          */
 262         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 263         vma->vm_end = STACK_TOP_MAX;
 264         vma->vm_start = vma->vm_end - PAGE_SIZE;
 265         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 266         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 267         INIT_LIST_HEAD(&vma->anon_vma_chain);
 268
 269         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 270         if (err)
 271                 goto err;
 272
 273         err = insert_vm_struct(mm, vma);
 274         if (err)
 275                 goto err;
 276
 277         mm->stack_vm = mm->total_vm = 1;
 278         up_write(&mm->mmap_sem);
 279         bprm->p = vma->vm_end - sizeof(void *);
 280         return 0;
 281 err:
 282         up_write(&mm->mmap_sem);
 283         bprm->vma = NULL;
 284         kmem_cache_free(vm_area_cachep, vma);
 285         return err;
 286 }
 287
 288 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 289 {
 290         return len <= MAX_ARG_STRLEN;
 291 }
 292
 293 #else
 294
 295 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 296 {
 297 }
 298
 299 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 300                 int write)
 301 {
 302         struct page *page;
 303
 304         page = bprm->page[pos / PAGE_SIZE];
 305         if (!page && write) {
 306                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 307                 if (!page)
 308                         return NULL;
 309                 bprm->page[pos / PAGE_SIZE] = page;
 310         }
 311
 312         return page;
 313 }
 314
 315 static void put_arg_page(struct page *page)
 316 {
 317 }
 318
 319 static void free_arg_page(struct linux_binprm *bprm, int i)
 320 {
 321         if (bprm->page[i]) {
 322                 __free_page(bprm->page[i]);
 323                 bprm->page[i] = NULL;
 324         }
 325 }
 326
 327 static void free_arg_pages(struct linux_binprm *bprm)
 328 {
 329         int i;
 330
 331         for (i = 0; i < MAX_ARG_PAGES; i++)
 332                 free_arg_page(bprm, i);
 333 }
 334
 335 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 336                 struct page *page)
 337 {
 338 }
 339
 340 static int __bprm_mm_init(struct linux_binprm *bprm)
 341 {
 342         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 343         return 0;
 344 }
 345
 346 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 347 {
 348         return len <= bprm->p;
 349 }
 350
 351 #endif /* CONFIG_MMU */
 352
 353 /*
 354  * Create a new mm_struct and populate it with a temporary stack
 355  * vm_area_struct.  We don't have enough context at this point to set the stack
 356  * flags, permissions, and offset, so we use temporary values.  We'll update
 357  * them later in setup_arg_pages().
 358  */
 359 int bprm_mm_init(struct linux_binprm *bprm)
 360 {
 361         int err;
 362         struct mm_struct *mm = NULL;
 363
 364         bprm->mm = mm = mm_alloc();
 365         err = -ENOMEM;
 366         if (!mm)
 367                 goto err;
 368
 369         err = init_new_context(current, mm);
 370         if (err)
 371                 goto err;
 372
 373         err = __bprm_mm_init(bprm);
 374         if (err)
 375                 goto err;
 376
 377         return 0;
 378
 379 err:
 380         if (mm) {
 381                 bprm->mm = NULL;
 382                 mmdrop(mm);
 383         }
 384
 385         return err;
 386 }
 387
 388 /*
 389  * count() counts the number of strings in array ARGV.
 390  */
 391 static int count(char __user * __user * argv, int max)
 392 {
 393         int i = 0;
 394
 395         if (argv != NULL) {
 396                 for (;;) {
 397                         char __user * p;
 398
 399                         if (get_user(p, argv))
 400                                 return -EFAULT;
 401                         if (!p)
 402                                 break;
 403                         argv++;
 404                         if (i++ >= max)
 405                                 return -E2BIG;
 406
 407                         if (fatal_signal_pending(current))
 408                                 return -ERESTARTNOHAND;
 409                         cond_resched();
 410                 }
 411         }
 412         return i;
 413 }
 414
 415 /*
 416  * 'copy_strings()' copies argument/environment strings from the old
 417  * processes's memory to the new process's stack.  The call to get_user_pages()
 418  * ensures the destination page is created and not swapped out.
 419  */
 420 static int copy_strings(int argc, char __user * __user * argv,
 421                         struct linux_binprm *bprm)
 422 {
 423         struct page *kmapped_page = NULL;
 424         char *kaddr = NULL;
 425         unsigned long kpos = 0;
 426         int ret;
 427
 428         while (argc-- > 0) {
 429                 char __user *str;
 430                 int len;
 431                 unsigned long pos;
 432
 433                 if (get_user(str, argv+argc) ||
 434                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 435                         ret = -EFAULT;
 436                         goto out;
 437                 }
 438
 439                 if (!valid_arg_len(bprm, len)) {
 440                         ret = -E2BIG;
 441                         goto out;
 442                 }
 443
 444                 /* We're going to work our way backwords. */
 445                 pos = bprm->p;
 446                 str += len;
 447                 bprm->p -= len;
 448
 449                 while (len > 0) {
 450                         int offset, bytes_to_copy;
 451
 452                         if (fatal_signal_pending(current)) {
 453                                 ret = -ERESTARTNOHAND;
 454                                 goto out;
 455                         }
 456                         cond_resched();
 457
 458                         offset = pos % PAGE_SIZE;
 459                         if (offset == 0)
 460                                 offset = PAGE_SIZE;
 461
 462                         bytes_to_copy = offset;
 463                         if (bytes_to_copy > len)
 464                                 bytes_to_copy = len;
 465
 466                         offset -= bytes_to_copy;
 467                         pos -= bytes_to_copy;
 468                         str -= bytes_to_copy;
 469                         len -= bytes_to_copy;
 470
 471                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 472                                 struct page *page;
 473
 474                                 page = get_arg_page(bprm, pos, 1);
 475                                 if (!page) {
 476                                         ret = -E2BIG;
 477                                         goto out;
 478                                 }
 479
 480                                 if (kmapped_page) {
 481                                         flush_kernel_dcache_page(kmapped_page);
 482                                         kunmap(kmapped_page);
 483                                         put_arg_page(kmapped_page);
 484                                 }
 485                                 kmapped_page = page;
 486                                 kaddr = kmap(kmapped_page);
 487                                 kpos = pos & PAGE_MASK;
 488                                 flush_arg_page(bprm, kpos, kmapped_page);
 489                         }
 490                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 491                                 ret = -EFAULT;
 492                                 goto out;
 493                         }
 494                 }
 495         }
 496         ret = 0;
 497 out:
 498         if (kmapped_page) {
 499                 flush_kernel_dcache_page(kmapped_page);
 500                 kunmap(kmapped_page);
 501                 put_arg_page(kmapped_page);
 502         }
 503         return ret;
 504 }
 505
 506 /*
 507  * Like copy_strings, but get argv and its values from kernel memory.
 508  */
 509 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 510 {
 511         int r;
 512         mm_segment_t oldfs = get_fs();
 513         set_fs(KERNEL_DS);
 514         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 515         set_fs(oldfs);
 516         return r;
 517 }
 518 EXPORT_SYMBOL(copy_strings_kernel);
 519
 520 #ifdef CONFIG_MMU
 521
 522 /*
 523  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 524  * the binfmt code determines where the new stack should reside, we shift it to
 525  * its final location.  The process proceeds as follows:
 526  *
 527  * 1) Use shift to calculate the new vma endpoints.
 528  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 529  *    arguments passed to subsequent functions are consistent.
 530  * 3) Move vma's page tables to the new range.
 531  * 4) Free up any cleared pgd range.
 532  * 5) Shrink the vma to cover only the new range.
 533  */
 534 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 535 {
 536         struct mm_struct *mm = vma->vm_mm;
 537         unsigned long old_start = vma->vm_start;
 538         unsigned long old_end = vma->vm_end;
 539         unsigned long length = old_end - old_start;
 540         unsigned long new_start = old_start - shift;
 541         unsigned long new_end = old_end - shift;
 542         struct mmu_gather *tlb;
 543
 544         BUG_ON(new_start > new_end);
 545
 546         /*
 547          * ensure there are no vmas between where we want to go
 548          * and where we are
 549          */
 550         if (vma != find_vma(mm, new_start))
 551                 return -EFAULT;
 552
 553         /*
 554          * cover the whole range: [new_start, old_end)
 555          */
 556         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 557                 return -ENOMEM;
 558
 559         /*
 560          * move the page tables downwards, on failure we rely on
 561          * process cleanup to remove whatever mess we made.
 562          */
 563         if (length != move_page_tables(vma, old_start,
 564                                        vma, new_start, length))
 565                 return -ENOMEM;
 566
 567         lru_add_drain();
 568         tlb = tlb_gather_mmu(mm, 0);
 569         if (new_end > old_start) {
 570                 /*
 571                  * when the old and new regions overlap clear from new_end.
 572                  */
 573                 free_pgd_range(tlb, new_end, old_end, new_end,
 574                         vma->vm_next ? vma->vm_next->vm_start : 0);
 575         } else {
 576                 /*
 577                  * otherwise, clean from old_start; this is done to not touch
 578                  * the address space in [new_end, old_start) some architectures
 579                  * have constraints on va-space that make this illegal (IA64) -
 580                  * for the others its just a little faster.
 581                  */
 582                 free_pgd_range(tlb, old_start, old_end, new_end,
 583                         vma->vm_next ? vma->vm_next->vm_start : 0);
 584         }
 585         tlb_finish_mmu(tlb, new_end, old_end);
 586
 587         /*
 588          * Shrink the vma to just the new range.  Always succeeds.
 589          */
 590         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 591
 592         return 0;
 593 }
 594
 595 /*
 596  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 597  * the stack is optionally relocated, and some extra space is added.
 598  */
 599 int setup_arg_pages(struct linux_binprm *bprm,
 600                     unsigned long stack_top,
 601                     int executable_stack)
 602 {
 603         unsigned long ret;
 604         unsigned long stack_shift;
 605         struct mm_struct *mm = current->mm;
 606         struct vm_area_struct *vma = bprm->vma;
 607         struct vm_area_struct *prev = NULL;
 608         unsigned long vm_flags;
 609         unsigned long stack_base;
 610         unsigned long stack_size;
 611         unsigned long stack_expand;
 612         unsigned long rlim_stack;
 613
 614 #ifdef CONFIG_STACK_GROWSUP
 615         /* Limit stack size to 1GB */
 616         stack_base = rlimit_max(RLIMIT_STACK);
 617         if (stack_base > (1 << 30))
 618                 stack_base = 1 << 30;
 619
 620         /* Make sure we didn't let the argument array grow too large. */
 621         if (vma->vm_end - vma->vm_start > stack_base)
 622                 return -ENOMEM;
 623
 624         stack_base = PAGE_ALIGN(stack_top - stack_base);
 625
 626         stack_shift = vma->vm_start - stack_base;
 627         mm->arg_start = bprm->p - stack_shift;
 628         bprm->p = vma->vm_end - stack_shift;
 629 #else
 630         stack_top = arch_align_stack(stack_top);
 631         stack_top = PAGE_ALIGN(stack_top);
 632
 633         if (unlikely(stack_top < mmap_min_addr) ||
 634             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 635                 return -ENOMEM;
 636
 637         stack_shift = vma->vm_end - stack_top;
 638
 639         bprm->p -= stack_shift;
 640         mm->arg_start = bprm->p;
 641 #endif
 642
 643         if (bprm->loader)
 644                 bprm->loader -= stack_shift;
 645         bprm->exec -= stack_shift;
 646
 647         down_write(&mm->mmap_sem);
 648         vm_flags = VM_STACK_FLAGS;
 649
 650         /*
 651          * Adjust stack execute permissions; explicitly enable for
 652          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 653          * (arch default) otherwise.
 654          */
 655         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 656                 vm_flags |= VM_EXEC;
 657         else if (executable_stack == EXSTACK_DISABLE_X)
 658                 vm_flags &= ~VM_EXEC;
 659         vm_flags |= mm->def_flags;
 660         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 661
 662         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 663                         vm_flags);
 664         if (ret)
 665                 goto out_unlock;
 666         BUG_ON(prev != vma);
 667
 668         /* Move stack pages down in memory. */
 669         if (stack_shift) {
 670                 ret = shift_arg_pages(vma, stack_shift);
 671                 if (ret)
 672                         goto out_unlock;
 673         }
 674
 675         /* mprotect_fixup is overkill to remove the temporary stack flags */
 676         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 677
 678         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 679         stack_size = vma->vm_end - vma->vm_start;
 680         /*
 681          * Align this down to a page boundary as expand_stack
 682          * will align it up.
 683          */
 684         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 685 #ifdef CONFIG_STACK_GROWSUP
 686         if (stack_size + stack_expand > rlim_stack)
 687                 stack_base = vma->vm_start + rlim_stack;
 688         else
 689                 stack_base = vma->vm_end + stack_expand;
 690 #else
 691         if (stack_size + stack_expand > rlim_stack)
 692                 stack_base = vma->vm_end - rlim_stack;
 693         else
 694                 stack_base = vma->vm_start - stack_expand;
 695 #endif
 696         ret = expand_stack(vma, stack_base);
 697         if (ret)
 698                 ret = -EFAULT;
 699
 700 out_unlock:
 701         up_write(&mm->mmap_sem);
 702         return ret;
 703 }
 704 EXPORT_SYMBOL(setup_arg_pages);
 705
 706 #endif /* CONFIG_MMU */
 707
 708 struct file *open_exec(const char *name)
 709 {
 710         struct file *file;
 711         int err;
 712
 713         file = do_filp_open(AT_FDCWD, name,
 714                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 715                                 MAY_EXEC | MAY_OPEN);
 716         if (IS_ERR(file))
 717                 goto out;
 718
 719         err = -EACCES;
 720         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 721                 goto exit;
 722
 723         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 724                 goto exit;
 725
 726         fsnotify_open(file->f_path.dentry);
 727
 728         err = deny_write_access(file);
 729         if (err)
 730                 goto exit;
 731
 732 out:
 733         return file;
 734
 735 exit:
 736         fput(file);
 737         return ERR_PTR(err);
 738 }
 739 EXPORT_SYMBOL(open_exec);
 740
 741 int kernel_read(struct file *file, loff_t offset,
 742                 char *addr, unsigned long count)
 743 {
 744         mm_segment_t old_fs;
 745         loff_t pos = offset;
 746         int result;
 747
 748         old_fs = get_fs();
 749         set_fs(get_ds());
 750         /* The cast to a user pointer is valid due to the set_fs() */
 751         result = vfs_read(file, (void __user *)addr, count, &pos);
 752         set_fs(old_fs);
 753         return result;
 754 }
 755
 756 EXPORT_SYMBOL(kernel_read);
 757
 758 static int exec_mmap(struct mm_struct *mm)
 759 {
 760         struct task_struct *tsk;
 761         struct mm_struct * old_mm, *active_mm;
 762
 763         /* Notify parent that we're no longer interested in the old VM */
 764         tsk = current;
 765         old_mm = current->mm;
 766         sync_mm_rss(tsk, old_mm);
 767         mm_release(tsk, old_mm);
 768
 769         if (old_mm) {
 770                 /*
 771                  * Make sure that if there is a core dump in progress
 772                  * for the old mm, we get out and die instead of going
 773                  * through with the exec.  We must hold mmap_sem around
 774                  * checking core_state and changing tsk->mm.
 775                  */
 776                 down_read(&old_mm->mmap_sem);
 777                 if (unlikely(old_mm->core_state)) {
 778                         up_read(&old_mm->mmap_sem);
 779                         return -EINTR;
 780                 }
 781         }
 782         task_lock(tsk);
 783         active_mm = tsk->active_mm;
 784         tsk->mm = mm;
 785         tsk->active_mm = mm;
 786         activate_mm(active_mm, mm);
 787         task_unlock(tsk);
 788         arch_pick_mmap_layout(mm);
 789         if (old_mm) {
 790                 up_read(&old_mm->mmap_sem);
 791                 BUG_ON(active_mm != old_mm);
 792                 mm_update_next_owner(old_mm);
 793                 mmput(old_mm);
 794                 return 0;
 795         }
 796         mmdrop(active_mm);
 797         return 0;
 798 }
 799
 800 /*
 801  * This function makes sure the current process has its own signal table,
 802  * so that flush_signal_handlers can later reset the handlers without
 803  * disturbing other processes.  (Other processes might share the signal
 804  * table via the CLONE_SIGHAND option to clone().)
 805  */
 806 static int de_thread(struct task_struct *tsk)
 807 {
 808         struct signal_struct *sig = tsk->signal;
 809         struct sighand_struct *oldsighand = tsk->sighand;
 810         spinlock_t *lock = &oldsighand->siglock;
 811
 812         if (thread_group_empty(tsk))
 813                 goto no_thread_group;
 814
 815         /*
 816          * Kill all other threads in the thread group.
 817          */
 818         spin_lock_irq(lock);
 819         if (signal_group_exit(sig)) {
 820                 /*
 821                  * Another group action in progress, just
 822                  * return so that the signal is processed.
 823                  */
 824                 spin_unlock_irq(lock);
 825                 return -EAGAIN;
 826         }
 827
 828         sig->group_exit_task = tsk;
 829         sig->notify_count = zap_other_threads(tsk);
 830         if (!thread_group_leader(tsk))
 831                 sig->notify_count--;
 832
 833         while (sig->notify_count) {
 834                 __set_current_state(TASK_UNINTERRUPTIBLE);
 835                 spin_unlock_irq(lock);
 836                 schedule();
 837                 spin_lock_irq(lock);
 838         }
 839         spin_unlock_irq(lock);
 840
 841         /*
 842          * At this point all other threads have exited, all we have to
 843          * do is to wait for the thread group leader to become inactive,
 844          * and to assume its PID:
 845          */
 846         if (!thread_group_leader(tsk)) {
 847                 struct task_struct *leader = tsk->group_leader;
 848
 849                 sig->notify_count = -1; /* for exit_notify() */
 850                 for (;;) {
 851                         write_lock_irq(&tasklist_lock);
 852                         if (likely(leader->exit_state))
 853                                 break;
 854                         __set_current_state(TASK_UNINTERRUPTIBLE);
 855                         write_unlock_irq(&tasklist_lock);
 856                         schedule();
 857                 }
 858
 859                 /*
 860                  * The only record we have of the real-time age of a
 861                  * process, regardless of execs it's done, is start_time.
 862                  * All the past CPU time is accumulated in signal_struct
 863                  * from sister threads now dead.  But in this non-leader
 864                  * exec, nothing survives from the original leader thread,
 865                  * whose birth marks the true age of this process now.
 866                  * When we take on its identity by switching to its PID, we
 867                  * also take its birthdate (always earlier than our own).
 868                  */
 869                 tsk->start_time = leader->start_time;
 870
 871                 BUG_ON(!same_thread_group(leader, tsk));
 872                 BUG_ON(has_group_leader_pid(tsk));
 873                 /*
 874                  * An exec() starts a new thread group with the
 875                  * TGID of the previous thread group. Rehash the
 876                  * two threads with a switched PID, and release
 877                  * the former thread group leader:
 878                  */
 879
 880                 /* Become a process group leader with the old leader's pid.
 881                  * The old leader becomes a thread of the this thread group.
 882                  * Note: The old leader also uses this pid until release_task
 883                  *       is called.  Odd but simple and correct.
 884                  */
 885                 detach_pid(tsk, PIDTYPE_PID);
 886                 tsk->pid = leader->pid;
 887                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 888                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 889                 transfer_pid(leader, tsk, PIDTYPE_SID);
 890
 891                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 892                 list_replace_init(&leader->sibling, &tsk->sibling);
 893
 894                 tsk->group_leader = tsk;
 895                 leader->group_leader = tsk;
 896
 897                 tsk->exit_signal = SIGCHLD;
 898
 899                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 900                 leader->exit_state = EXIT_DEAD;
 901                 write_unlock_irq(&tasklist_lock);
 902
 903                 release_task(leader);
 904         }
 905
 906         sig->group_exit_task = NULL;
 907         sig->notify_count = 0;
 908
 909 no_thread_group:
 910         if (current->mm)
 911                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 912
 913         exit_itimers(sig);
 914         flush_itimer_signals();
 915
 916         if (atomic_read(&oldsighand->count) != 1) {
 917                 struct sighand_struct *newsighand;
 918                 /*
 919                  * This ->sighand is shared with the CLONE_SIGHAND
 920                  * but not CLONE_THREAD task, switch to the new one.
 921                  */
 922                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 923                 if (!newsighand)
 924                         return -ENOMEM;
 925
 926                 atomic_set(&newsighand->count, 1);
 927                 memcpy(newsighand->action, oldsighand->action,
 928                        sizeof(newsighand->action));
 929
 930                 write_lock_irq(&tasklist_lock);
 931                 spin_lock(&oldsighand->siglock);
 932                 rcu_assign_pointer(tsk->sighand, newsighand);
 933                 spin_unlock(&oldsighand->siglock);
 934                 write_unlock_irq(&tasklist_lock);
 935
 936                 __cleanup_sighand(oldsighand);
 937         }
 938
 939         BUG_ON(!thread_group_leader(tsk));
 940         return 0;
 941 }
 942
 943 /*
 944  * These functions flushes out all traces of the currently running executable
 945  * so that a new one can be started
 946  */
 947 static void flush_old_files(struct files_struct * files)
 948 {
 949         long j = -1;
 950         struct fdtable *fdt;
 951
 952         spin_lock(&files->file_lock);
 953         for (;;) {
 954                 unsigned long set, i;
 955
 956                 j++;
 957                 i = j * __NFDBITS;
 958                 fdt = files_fdtable(files);
 959                 if (i >= fdt->max_fds)
 960                         break;
 961                 set = fdt->close_on_exec->fds_bits[j];
 962                 if (!set)
 963                         continue;
 964                 fdt->close_on_exec->fds_bits[j] = 0;
 965                 spin_unlock(&files->file_lock);
 966                 for ( ; set ; i++,set >>= 1) {
 967                         if (set & 1) {
 968                                 sys_close(i);
 969                         }
 970                 }
 971                 spin_lock(&files->file_lock);
 972
 973         }
 974         spin_unlock(&files->file_lock);
 975 }
 976
 977 char *get_task_comm(char *buf, struct task_struct *tsk)
 978 {
 979         /* buf must be at least sizeof(tsk->comm) in size */
 980         task_lock(tsk);
 981         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 982         task_unlock(tsk);
 983         return buf;
 984 }
 985
 986 void set_task_comm(struct task_struct *tsk, char *buf)
 987 {
 988         task_lock(tsk);
 989
 990         /*
 991          * Threads may access current->comm without holding
 992          * the task lock, so write the string carefully.
 993          * Readers without a lock may see incomplete new
 994          * names but are safe from non-terminating string reads.
 995          */
 996         memset(tsk->comm, 0, TASK_COMM_LEN);
 997         wmb();
 998         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 999         task_unlock(tsk);
1000         perf_event_comm(tsk);
1001 }
1002
1003 int flush_old_exec(struct linux_binprm * bprm)
1004 {
1005         int retval;
1006
1007         /*
1008          * Make sure we have a private signal table and that
1009          * we are unassociated from the previous thread group.
1010          */
1011         retval = de_thread(current);
1012         if (retval)
1013                 goto out;
1014
1015         set_mm_exe_file(bprm->mm, bprm->file);
1016
1017         /*
1018          * Release all of the old mmap stuff
1019          */
1020         acct_arg_size(bprm, 0);
1021         retval = exec_mmap(bprm->mm);
1022         if (retval)
1023                 goto out;
1024
1025         bprm->mm = NULL;                /* We're using it now */
1026
1027         current->flags &= ~PF_RANDOMIZE;
1028         flush_thread();
1029         current->personality &= ~bprm->per_clear;
1030
1031         return 0;
1032
1033 out:
1034         return retval;
1035 }
1036 EXPORT_SYMBOL(flush_old_exec);
1037
1038 void setup_new_exec(struct linux_binprm * bprm)
1039 {
1040         int i, ch;
1041         char * name;
1042         char tcomm[sizeof(current->comm)];
1043
1044         arch_pick_mmap_layout(current->mm);
1045
1046         /* This is the point of no return */
1047         current->sas_ss_sp = current->sas_ss_size = 0;
1048
1049         if (current_euid() == current_uid() && current_egid() == current_gid())
1050                 set_dumpable(current->mm, 1);
1051         else
1052                 set_dumpable(current->mm, suid_dumpable);
1053
1054         name = bprm->filename;
1055
1056         /* Copies the binary name from after last slash */
1057         for (i=0; (ch = *(name++)) != '\0';) {
1058                 if (ch == '/')
1059                         i = 0; /* overwrite what we wrote */
1060                 else
1061                         if (i < (sizeof(tcomm) - 1))
1062                                 tcomm[i++] = ch;
1063         }
1064         tcomm[i] = '\0';
1065         set_task_comm(current, tcomm);
1066
1067         /* Set the new mm task size. We have to do that late because it may
1068          * depend on TIF_32BIT which is only updated in flush_thread() on
1069          * some architectures like powerpc
1070          */
1071         current->mm->task_size = TASK_SIZE;
1072
1073         /* install the new credentials */
1074         if (bprm->cred->uid != current_euid() ||
1075             bprm->cred->gid != current_egid()) {
1076                 current->pdeath_signal = 0;
1077         } else if (file_permission(bprm->file, MAY_READ) ||
1078                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1079                 set_dumpable(current->mm, suid_dumpable);
1080         }
1081
1082         /*
1083          * Flush performance counters when crossing a
1084          * security domain:
1085          */
1086         if (!get_dumpable(current->mm))
1087                 perf_event_exit_task(current);
1088
1089         /* An exec changes our domain. We are no longer part of the thread
1090            group */
1091
1092         current->self_exec_id++;
1093
1094         flush_signal_handlers(current, 0);
1095         flush_old_files(current->files);
1096 }
1097 EXPORT_SYMBOL(setup_new_exec);
1098
1099 /*
1100  * Prepare credentials and lock ->cred_guard_mutex.
1101  * install_exec_creds() commits the new creds and drops the lock.
1102  * Or, if exec fails before, free_bprm() should release ->cred and
1103  * and unlock.
1104  */
1105 int prepare_bprm_creds(struct linux_binprm *bprm)
1106 {
1107         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1108                 return -ERESTARTNOINTR;
1109
1110         bprm->cred = prepare_exec_creds();
1111         if (likely(bprm->cred))
1112                 return 0;
1113
1114         mutex_unlock(&current->cred_guard_mutex);
1115         return -ENOMEM;
1116 }
1117
1118 void free_bprm(struct linux_binprm *bprm)
1119 {
1120         free_arg_pages(bprm);
1121         if (bprm->cred) {
1122                 mutex_unlock(&current->cred_guard_mutex);
1123                 abort_creds(bprm->cred);
1124         }
1125         kfree(bprm);
1126 }
1127
1128 /*
1129  * install the new credentials for this executable
1130  */
1131 void install_exec_creds(struct linux_binprm *bprm)
1132 {
1133         security_bprm_committing_creds(bprm);
1134
1135         commit_creds(bprm->cred);
1136         bprm->cred = NULL;
1137         /*
1138          * cred_guard_mutex must be held at least to this point to prevent
1139          * ptrace_attach() from altering our determination of the task's
1140          * credentials; any time after this it may be unlocked.
1141          */
1142         security_bprm_committed_creds(bprm);
1143         mutex_unlock(&current->cred_guard_mutex);
1144 }
1145 EXPORT_SYMBOL(install_exec_creds);
1146
1147 /*
1148  * determine how safe it is to execute the proposed program
1149  * - the caller must hold current->cred_guard_mutex to protect against
1150  *   PTRACE_ATTACH
1151  */
1152 int check_unsafe_exec(struct linux_binprm *bprm)
1153 {
1154         struct task_struct *p = current, *t;
1155         unsigned n_fs;
1156         int res = 0;
1157
1158         bprm->unsafe = tracehook_unsafe_exec(p);
1159
1160         n_fs = 1;
1161         write_lock(&p->fs->lock);
1162         rcu_read_lock();
1163         for (t = next_thread(p); t != p; t = next_thread(t)) {
1164                 if (t->fs == p->fs)
1165                         n_fs++;
1166         }
1167         rcu_read_unlock();
1168
1169         if (p->fs->users > n_fs) {
1170                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1171         } else {
1172                 res = -EAGAIN;
1173                 if (!p->fs->in_exec) {
1174                         p->fs->in_exec = 1;
1175                         res = 1;
1176                 }
1177         }
1178         write_unlock(&p->fs->lock);
1179
1180         return res;
1181 }
1182
1183 /*
1184  * Fill the binprm structure from the inode.
1185  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1186  *
1187  * This may be called multiple times for binary chains (scripts for example).
1188  */
1189 int prepare_binprm(struct linux_binprm *bprm)
1190 {
1191         umode_t mode;
1192         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1193         int retval;
1194
1195         mode = inode->i_mode;
1196         if (bprm->file->f_op == NULL)
1197                 return -EACCES;
1198
1199         /* clear any previous set[ug]id data from a previous binary */
1200         bprm->cred->euid = current_euid();
1201         bprm->cred->egid = current_egid();
1202
1203         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1204                 /* Set-uid? */
1205                 if (mode & S_ISUID) {
1206                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1207                         bprm->cred->euid = inode->i_uid;
1208                 }
1209
1210                 /* Set-gid? */
1211                 /*
1212                  * If setgid is set but no group execute bit then this
1213                  * is a candidate for mandatory locking, not a setgid
1214                  * executable.
1215                  */
1216                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1217                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1218                         bprm->cred->egid = inode->i_gid;
1219                 }
1220         }
1221
1222         /* fill in binprm security blob */
1223         retval = security_bprm_set_creds(bprm);
1224         if (retval)
1225                 return retval;
1226         bprm->cred_prepared = 1;
1227
1228         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1229         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1230 }
1231
1232 EXPORT_SYMBOL(prepare_binprm);
1233
1234 /*
1235  * Arguments are '\0' separated strings found at the location bprm->p
1236  * points to; chop off the first by relocating brpm->p to right after
1237  * the first '\0' encountered.
1238  */
1239 int remove_arg_zero(struct linux_binprm *bprm)
1240 {
1241         int ret = 0;
1242         unsigned long offset;
1243         char *kaddr;
1244         struct page *page;
1245
1246         if (!bprm->argc)
1247                 return 0;
1248
1249         do {
1250                 offset = bprm->p & ~PAGE_MASK;
1251                 page = get_arg_page(bprm, bprm->p, 0);
1252                 if (!page) {
1253                         ret = -EFAULT;
1254                         goto out;
1255                 }
1256                 kaddr = kmap_atomic(page, KM_USER0);
1257
1258                 for (; offset < PAGE_SIZE && kaddr[offset];
1259                                 offset++, bprm->p++)
1260                         ;
1261
1262                 kunmap_atomic(kaddr, KM_USER0);
1263                 put_arg_page(page);
1264
1265                 if (offset == PAGE_SIZE)
1266                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1267         } while (offset == PAGE_SIZE);
1268
1269         bprm->p++;
1270         bprm->argc--;
1271         ret = 0;
1272
1273 out:
1274         return ret;
1275 }
1276 EXPORT_SYMBOL(remove_arg_zero);
1277
1278 /*
1279  * cycle the list of binary formats handler, until one recognizes the image
1280  */
1281 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1282 {
1283         unsigned int depth = bprm->recursion_depth;
1284         int try,retval;
1285         struct linux_binfmt *fmt;
1286
1287         retval = security_bprm_check(bprm);
1288         if (retval)
1289                 return retval;
1290
1291         /* kernel module loader fixup */
1292         /* so we don't try to load run modprobe in kernel space. */
1293         set_fs(USER_DS);
1294
1295         retval = audit_bprm(bprm);
1296         if (retval)
1297                 return retval;
1298
1299         retval = -ENOENT;
1300         for (try=0; try<2; try++) {
1301                 read_lock(&binfmt_lock);
1302                 list_for_each_entry(fmt, &formats, lh) {
1303                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1304                         if (!fn)
1305                                 continue;
1306                         if (!try_module_get(fmt->module))
1307                                 continue;
1308                         read_unlock(&binfmt_lock);
1309                         retval = fn(bprm, regs);
1310                         /*
1311                          * Restore the depth counter to its starting value
1312                          * in this call, so we don't have to rely on every
1313                          * load_binary function to restore it on return.
1314                          */
1315                         bprm->recursion_depth = depth;
1316                         if (retval >= 0) {
1317                                 if (depth == 0)
1318                                         tracehook_report_exec(fmt, bprm, regs);
1319                                 put_binfmt(fmt);
1320                                 allow_write_access(bprm->file);
1321                                 if (bprm->file)
1322                                         fput(bprm->file);
1323                                 bprm->file = NULL;
1324                                 current->did_exec = 1;
1325                                 proc_exec_connector(current);
1326                                 return retval;
1327                         }
1328                         read_lock(&binfmt_lock);
1329                         put_binfmt(fmt);
1330                         if (retval != -ENOEXEC || bprm->mm == NULL)
1331                                 break;
1332                         if (!bprm->file) {
1333                                 read_unlock(&binfmt_lock);
1334                                 return retval;
1335                         }
1336                 }
1337                 read_unlock(&binfmt_lock);
1338                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1339                         break;
1340 #ifdef CONFIG_MODULES
1341                 } else {
1342 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1343                         if (printable(bprm->buf[0]) &&
1344                             printable(bprm->buf[1]) &&
1345                             printable(bprm->buf[2]) &&
1346                             printable(bprm->buf[3]))
1347                                 break; /* -ENOEXEC */
1348                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1349 #endif
1350                 }
1351         }
1352         return retval;
1353 }
1354
1355 EXPORT_SYMBOL(search_binary_handler);
1356
1357 /*
1358  * sys_execve() executes a new program.
1359  */
1360 int do_execve(char * filename,
1361         char __user *__user *argv,
1362         char __user *__user *envp,
1363         struct pt_regs * regs)
1364 {
1365         struct linux_binprm *bprm;
1366         struct file *file;
1367         struct files_struct *displaced;
1368         bool clear_in_exec;
1369         int retval;
1370
1371         retval = unshare_files(&displaced);
1372         if (retval)
1373                 goto out_ret;
1374
1375         retval = -ENOMEM;
1376         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1377         if (!bprm)
1378                 goto out_files;
1379
1380         retval = prepare_bprm_creds(bprm);
1381         if (retval)
1382                 goto out_free;
1383
1384         retval = check_unsafe_exec(bprm);
1385         if (retval < 0)
1386                 goto out_free;
1387         clear_in_exec = retval;
1388         current->in_execve = 1;
1389
1390         file = open_exec(filename);
1391         retval = PTR_ERR(file);
1392         if (IS_ERR(file))
1393                 goto out_unmark;
1394
1395         sched_exec();
1396
1397         bprm->file = file;
1398         bprm->filename = filename;
1399         bprm->interp = filename;
1400
1401         retval = bprm_mm_init(bprm);
1402         if (retval)
1403                 goto out_file;
1404
1405         bprm->argc = count(argv, MAX_ARG_STRINGS);
1406         if ((retval = bprm->argc) < 0)
1407                 goto out;
1408
1409         bprm->envc = count(envp, MAX_ARG_STRINGS);
1410         if ((retval = bprm->envc) < 0)
1411                 goto out;
1412
1413         retval = prepare_binprm(bprm);
1414         if (retval < 0)
1415                 goto out;
1416
1417         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1418         if (retval < 0)
1419                 goto out;
1420
1421         bprm->exec = bprm->p;
1422         retval = copy_strings(bprm->envc, envp, bprm);
1423         if (retval < 0)
1424                 goto out;
1425
1426         retval = copy_strings(bprm->argc, argv, bprm);
1427         if (retval < 0)
1428                 goto out;
1429
1430         current->flags &= ~PF_KTHREAD;
1431         retval = search_binary_handler(bprm,regs);
1432         if (retval < 0)
1433                 goto out;
1434
1435         /* execve succeeded */
1436         current->fs->in_exec = 0;
1437         current->in_execve = 0;
1438         acct_update_integrals(current);
1439         free_bprm(bprm);
1440         if (displaced)
1441                 put_files_struct(displaced);
1442         return retval;
1443
1444 out:
1445         if (bprm->mm) {
1446                 acct_arg_size(bprm, 0);
1447                 mmput(bprm->mm);
1448         }
1449
1450 out_file:
1451         if (bprm->file) {
1452                 allow_write_access(bprm->file);
1453                 fput(bprm->file);
1454         }
1455
1456 out_unmark:
1457         if (clear_in_exec)
1458                 current->fs->in_exec = 0;
1459         current->in_execve = 0;
1460
1461 out_free:
1462         free_bprm(bprm);
1463
1464 out_files:
1465         if (displaced)
1466                 reset_files_struct(displaced);
1467 out_ret:
1468         return retval;
1469 }
1470
1471 void set_binfmt(struct linux_binfmt *new)
1472 {
1473         struct mm_struct *mm = current->mm;
1474
1475         if (mm->binfmt)
1476                 module_put(mm->binfmt->module);
1477
1478         mm->binfmt = new;
1479         if (new)
1480                 __module_get(new->module);
1481 }
1482
1483 EXPORT_SYMBOL(set_binfmt);
1484
1485 /* format_corename will inspect the pattern parameter, and output a
1486  * name into corename, which must have space for at least
1487  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1488  */
1489 static int format_corename(char *corename, long signr)
1490 {
1491         const struct cred *cred = current_cred();
1492         const char *pat_ptr = core_pattern;
1493         int ispipe = (*pat_ptr == '|');
1494         char *out_ptr = corename;
1495         char *const out_end = corename + CORENAME_MAX_SIZE;
1496         int rc;
1497         int pid_in_pattern = 0;
1498
1499         /* Repeat as long as we have more pattern to process and more output
1500            space */
1501         while (*pat_ptr) {
1502                 if (*pat_ptr != '%') {
1503                         if (out_ptr == out_end)
1504                                 goto out;
1505                         *out_ptr++ = *pat_ptr++;
1506                 } else {
1507                         switch (*++pat_ptr) {
1508                         case 0:
1509                                 goto out;
1510                         /* Double percent, output one percent */
1511                         case '%':
1512                                 if (out_ptr == out_end)
1513                                         goto out;
1514                                 *out_ptr++ = '%';
1515                                 break;
1516                         /* pid */
1517                         case 'p':
1518                                 pid_in_pattern = 1;
1519                                 rc = snprintf(out_ptr, out_end - out_ptr,
1520                                               "%d", task_tgid_vnr(current));
1521                                 if (rc > out_end - out_ptr)
1522                                         goto out;
1523                                 out_ptr += rc;
1524                                 break;
1525                         /* uid */
1526                         case 'u':
1527                                 rc = snprintf(out_ptr, out_end - out_ptr,
1528                                               "%d", cred->uid);
1529                                 if (rc > out_end - out_ptr)
1530                                         goto out;
1531                                 out_ptr += rc;
1532                                 break;
1533                         /* gid */
1534                         case 'g':
1535                                 rc = snprintf(out_ptr, out_end - out_ptr,
1536                                               "%d", cred->gid);
1537                                 if (rc > out_end - out_ptr)
1538                                         goto out;
1539                                 out_ptr += rc;
1540                                 break;
1541                         /* signal that caused the coredump */
1542                         case 's':
1543                                 rc = snprintf(out_ptr, out_end - out_ptr,
1544                                               "%ld", signr);
1545                                 if (rc > out_end - out_ptr)
1546                                         goto out;
1547                                 out_ptr += rc;
1548                                 break;
1549                         /* UNIX time of coredump */
1550                         case 't': {
1551                                 struct timeval tv;
1552                                 do_gettimeofday(&tv);
1553                                 rc = snprintf(out_ptr, out_end - out_ptr,
1554                                               "%lu", tv.tv_sec);
1555                                 if (rc > out_end - out_ptr)
1556                                         goto out;
1557                                 out_ptr += rc;
1558                                 break;
1559                         }
1560                         /* hostname */
1561                         case 'h':
1562                                 down_read(&uts_sem);
1563                                 rc = snprintf(out_ptr, out_end - out_ptr,
1564                                               "%s", utsname()->nodename);
1565                                 up_read(&uts_sem);
1566                                 if (rc > out_end - out_ptr)
1567                                         goto out;
1568                                 out_ptr += rc;
1569                                 break;
1570                         /* executable */
1571                         case 'e':
1572                                 rc = snprintf(out_ptr, out_end - out_ptr,
1573                                               "%s", current->comm);
1574                                 if (rc > out_end - out_ptr)
1575                                         goto out;
1576                                 out_ptr += rc;
1577                                 break;
1578                         /* core limit size */
1579                         case 'c':
1580                                 rc = snprintf(out_ptr, out_end - out_ptr,
1581                                               "%lu", rlimit(RLIMIT_CORE));
1582                                 if (rc > out_end - out_ptr)
1583                                         goto out;
1584                                 out_ptr += rc;
1585                                 break;
1586                         default:
1587                                 break;
1588                         }
1589                         ++pat_ptr;
1590                 }
1591         }
1592         /* Backward compatibility with core_uses_pid:
1593          *
1594          * If core_pattern does not include a %p (as is the default)
1595          * and core_uses_pid is set, then .%pid will be appended to
1596          * the filename. Do not do this for piped commands. */
1597         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1598                 rc = snprintf(out_ptr, out_end - out_ptr,
1599                               ".%d", task_tgid_vnr(current));
1600                 if (rc > out_end - out_ptr)
1601                         goto out;
1602                 out_ptr += rc;
1603         }
1604 out:
1605         *out_ptr = 0;
1606         return ispipe;
1607 }
1608
1609 static int zap_process(struct task_struct *start, int exit_code)
1610 {
1611         struct task_struct *t;
1612         int nr = 0;
1613
1614         start->signal->flags = SIGNAL_GROUP_EXIT;
1615         start->signal->group_exit_code = exit_code;
1616         start->signal->group_stop_count = 0;
1617
1618         t = start;
1619         do {
1620                 if (t != current && t->mm) {
1621                         sigaddset(&t->pending.signal, SIGKILL);
1622                         signal_wake_up(t, 1);
1623                         nr++;
1624                 }
1625         } while_each_thread(start, t);
1626
1627         return nr;
1628 }
1629
1630 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1631                                 struct core_state *core_state, int exit_code)
1632 {
1633         struct task_struct *g, *p;
1634         unsigned long flags;
1635         int nr = -EAGAIN;
1636
1637         spin_lock_irq(&tsk->sighand->siglock);
1638         if (!signal_group_exit(tsk->signal)) {
1639                 mm->core_state = core_state;
1640                 nr = zap_process(tsk, exit_code);
1641         }
1642         spin_unlock_irq(&tsk->sighand->siglock);
1643         if (unlikely(nr < 0))
1644                 return nr;
1645
1646         if (atomic_read(&mm->mm_users) == nr + 1)
1647                 goto done;
1648         /*
1649          * We should find and kill all tasks which use this mm, and we should
1650          * count them correctly into ->nr_threads. We don't take tasklist
1651          * lock, but this is safe wrt:
1652          *
1653          * fork:
1654          *      None of sub-threads can fork after zap_process(leader). All
1655          *      processes which were created before this point should be
1656          *      visible to zap_threads() because copy_process() adds the new
1657          *      process to the tail of init_task.tasks list, and lock/unlock
1658          *      of ->siglock provides a memory barrier.
1659          *
1660          * do_exit:
1661          *      The caller holds mm->mmap_sem. This means that the task which
1662          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1663          *      its ->mm.
1664          *
1665          * de_thread:
1666          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1667          *      we must see either old or new leader, this does not matter.
1668          *      However, it can change p->sighand, so lock_task_sighand(p)
1669          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1670          *      it can't fail.
1671          *
1672          *      Note also that "g" can be the old leader with ->mm == NULL
1673          *      and already unhashed and thus removed from ->thread_group.
1674          *      This is OK, __unhash_process()->list_del_rcu() does not
1675          *      clear the ->next pointer, we will find the new leader via
1676          *      next_thread().
1677          */
1678         rcu_read_lock();
1679         for_each_process(g) {
1680                 if (g == tsk->group_leader)
1681                         continue;
1682                 if (g->flags & PF_KTHREAD)
1683                         continue;
1684                 p = g;
1685                 do {
1686                         if (p->mm) {
1687                                 if (unlikely(p->mm == mm)) {
1688                                         lock_task_sighand(p, &flags);
1689                                         nr += zap_process(p, exit_code);
1690                                         unlock_task_sighand(p, &flags);
1691                                 }
1692                                 break;
1693                         }
1694                 } while_each_thread(g, p);
1695         }
1696         rcu_read_unlock();
1697 done:
1698         atomic_set(&core_state->nr_threads, nr);
1699         return nr;
1700 }
1701
1702 static int coredump_wait(int exit_code, struct core_state *core_state)
1703 {
1704         struct task_struct *tsk = current;
1705         struct mm_struct *mm = tsk->mm;
1706         struct completion *vfork_done;
1707         int core_waiters = -EBUSY;
1708
1709         init_completion(&core_state->startup);
1710         core_state->dumper.task = tsk;
1711         core_state->dumper.next = NULL;
1712
1713         down_write(&mm->mmap_sem);
1714         if (!mm->core_state)
1715                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1716         up_write(&mm->mmap_sem);
1717
1718         if (unlikely(core_waiters < 0))
1719                 goto fail;
1720
1721         /*
1722          * Make sure nobody is waiting for us to release the VM,
1723          * otherwise we can deadlock when we wait on each other
1724          */
1725         vfork_done = tsk->vfork_done;
1726         if (vfork_done) {
1727                 tsk->vfork_done = NULL;
1728                 complete(vfork_done);
1729         }
1730
1731         if (core_waiters)
1732                 wait_for_completion(&core_state->startup);
1733 fail:
1734         return core_waiters;
1735 }
1736
1737 static void coredump_finish(struct mm_struct *mm)
1738 {
1739         struct core_thread *curr, *next;
1740         struct task_struct *task;
1741
1742         next = mm->core_state->dumper.next;
1743         while ((curr = next) != NULL) {
1744                 next = curr->next;
1745                 task = curr->task;
1746                 /*
1747                  * see exit_mm(), curr->task must not see
1748                  * ->task == NULL before we read ->next.
1749                  */
1750                 smp_mb();
1751                 curr->task = NULL;
1752                 wake_up_process(task);
1753         }
1754
1755         mm->core_state = NULL;
1756 }
1757
1758 /*
1759  * set_dumpable converts traditional three-value dumpable to two flags and
1760  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1761  * these bits are not changed atomically.  So get_dumpable can observe the
1762  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1763  * return either old dumpable or new one by paying attention to the order of
1764  * modifying the bits.
1765  *
1766  * dumpable |   mm->flags (binary)
1767  * old  new | initial interim  final
1768  * ---------+-----------------------
1769  *  0    1  |   00      01      01
1770  *  0    2  |   00      10(*)   11
1771  *  1    0  |   01      00      00
1772  *  1    2  |   01      11      11
1773  *  2    0  |   11      10(*)   00
1774  *  2    1  |   11      11      01
1775  *
1776  * (*) get_dumpable regards interim value of 10 as 11.
1777  */
1778 void set_dumpable(struct mm_struct *mm, int value)
1779 {
1780         switch (value) {
1781         case 0:
1782                 clear_bit(MMF_DUMPABLE, &mm->flags);
1783                 smp_wmb();
1784                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1785                 break;
1786         case 1:
1787                 set_bit(MMF_DUMPABLE, &mm->flags);
1788                 smp_wmb();
1789                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1790                 break;
1791         case 2:
1792                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1793                 smp_wmb();
1794                 set_bit(MMF_DUMPABLE, &mm->flags);
1795                 break;
1796         }
1797 }
1798
1799 static int __get_dumpable(unsigned long mm_flags)
1800 {
1801         int ret;
1802
1803         ret = mm_flags & MMF_DUMPABLE_MASK;
1804         return (ret >= 2) ? 2 : ret;
1805 }
1806
1807 int get_dumpable(struct mm_struct *mm)
1808 {
1809         return __get_dumpable(mm->flags);
1810 }
1811
1812 static void wait_for_dump_helpers(struct file *file)
1813 {
1814         struct pipe_inode_info *pipe;
1815
1816         pipe = file->f_path.dentry->d_inode->i_pipe;
1817
1818         pipe_lock(pipe);
1819         pipe->readers++;
1820         pipe->writers--;
1821
1822         while ((pipe->readers > 1) && (!signal_pending(current))) {
1823                 wake_up_interruptible_sync(&pipe->wait);
1824                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1825                 pipe_wait(pipe);
1826         }
1827
1828         pipe->readers--;
1829         pipe->writers++;
1830         pipe_unlock(pipe);
1831
1832 }
1833
1834
1835 /*
1836  * uhm_pipe_setup
1837  * helper function to customize the process used
1838  * to collect the core in userspace.  Specifically
1839  * it sets up a pipe and installs it as fd 0 (stdin)
1840  * for the process.  Returns 0 on success, or
1841  * PTR_ERR on failure.
1842  * Note that it also sets the core limit to 1.  This
1843  * is a special value that we use to trap recursive
1844  * core dumps
1845  */
1846 static int umh_pipe_setup(struct subprocess_info *info)
1847 {
1848         struct file *rp, *wp;
1849         struct fdtable *fdt;
1850         struct coredump_params *cp = (struct coredump_params *)info->data;
1851         struct files_struct *cf = current->files;
1852
1853         wp = create_write_pipe(0);
1854         if (IS_ERR(wp))
1855                 return PTR_ERR(wp);
1856
1857         rp = create_read_pipe(wp, 0);
1858         if (IS_ERR(rp)) {
1859                 free_write_pipe(wp);
1860                 return PTR_ERR(rp);
1861         }
1862
1863         cp->file = wp;
1864
1865         sys_close(0);
1866         fd_install(0, rp);
1867         spin_lock(&cf->file_lock);
1868         fdt = files_fdtable(cf);
1869         FD_SET(0, fdt->open_fds);
1870         FD_CLR(0, fdt->close_on_exec);
1871         spin_unlock(&cf->file_lock);
1872
1873         /* and disallow core files too */
1874         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1875
1876         return 0;
1877 }
1878
1879 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1880 {
1881         struct core_state core_state;
1882         char corename[CORENAME_MAX_SIZE + 1];
1883         struct mm_struct *mm = current->mm;
1884         struct linux_binfmt * binfmt;
1885         const struct cred *old_cred;
1886         struct cred *cred;
1887         int retval = 0;
1888         int flag = 0;
1889         int ispipe;
1890         static atomic_t core_dump_count = ATOMIC_INIT(0);
1891         struct coredump_params cprm = {
1892                 .signr = signr,
1893                 .regs = regs,
1894                 .limit = rlimit(RLIMIT_CORE),
1895                 /*
1896                  * We must use the same mm->flags while dumping core to avoid
1897                  * inconsistency of bit flags, since this flag is not protected
1898                  * by any locks.
1899                  */
1900                 .mm_flags = mm->flags,
1901         };
1902
1903         audit_core_dumps(signr);
1904
1905         binfmt = mm->binfmt;
1906         if (!binfmt || !binfmt->core_dump)
1907                 goto fail;
1908         if (!__get_dumpable(cprm.mm_flags))
1909                 goto fail;
1910
1911         cred = prepare_creds();
1912         if (!cred)
1913                 goto fail;
1914         /*
1915          *      We cannot trust fsuid as being the "true" uid of the
1916          *      process nor do we know its entire history. We only know it
1917          *      was tainted so we dump it as root in mode 2.
1918          */
1919         if (__get_dumpable(cprm.mm_flags) == 2) {
1920                 /* Setuid core dump mode */
1921                 flag = O_EXCL;          /* Stop rewrite attacks */
1922                 cred->fsuid = 0;        /* Dump root private */
1923         }
1924
1925         retval = coredump_wait(exit_code, &core_state);
1926         if (retval < 0)
1927                 goto fail_creds;
1928
1929         old_cred = override_creds(cred);
1930
1931         /*
1932          * Clear any false indication of pending signals that might
1933          * be seen by the filesystem code called to write the core file.
1934          */
1935         clear_thread_flag(TIF_SIGPENDING);
1936
1937         /*
1938          * lock_kernel() because format_corename() is controlled by sysctl, which
1939          * uses lock_kernel()
1940          */
1941         lock_kernel();
1942         ispipe = format_corename(corename, signr);
1943         unlock_kernel();
1944
1945         if (ispipe) {
1946                 int dump_count;
1947                 char **helper_argv;
1948
1949                 if (cprm.limit == 1) {
1950                         /*
1951                          * Normally core limits are irrelevant to pipes, since
1952                          * we're not writing to the file system, but we use
1953                          * cprm.limit of 1 here as a speacial value. Any
1954                          * non-1 limit gets set to RLIM_INFINITY below, but
1955                          * a limit of 0 skips the dump.  This is a consistent
1956                          * way to catch recursive crashes.  We can still crash
1957                          * if the core_pattern binary sets RLIM_CORE =  !1
1958                          * but it runs as root, and can do lots of stupid things
1959                          * Note that we use task_tgid_vnr here to grab the pid
1960                          * of the process group leader.  That way we get the
1961                          * right pid if a thread in a multi-threaded
1962                          * core_pattern process dies.
1963                          */
1964                         printk(KERN_WARNING
1965                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1966                                 task_tgid_vnr(current), current->comm);
1967                         printk(KERN_WARNING "Aborting core\n");
1968                         goto fail_unlock;
1969                 }
1970                 cprm.limit = RLIM_INFINITY;
1971
1972                 dump_count = atomic_inc_return(&core_dump_count);
1973                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1974                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1975                                task_tgid_vnr(current), current->comm);
1976                         printk(KERN_WARNING "Skipping core dump\n");
1977                         goto fail_dropcount;
1978                 }
1979
1980                 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1981                 if (!helper_argv) {
1982                         printk(KERN_WARNING "%s failed to allocate memory\n",
1983                                __func__);
1984                         goto fail_dropcount;
1985                 }
1986
1987                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1988                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1989                                         NULL, &cprm);
1990                 argv_free(helper_argv);
1991                 if (retval) {
1992                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1993                                corename);
1994                         goto close_fail;
1995                 }
1996         } else {
1997                 struct inode *inode;
1998
1999                 if (cprm.limit < binfmt->min_coredump)
2000                         goto fail_unlock;
2001
2002                 cprm.file = filp_open(corename,
2003                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2004                                  0600);
2005                 if (IS_ERR(cprm.file))
2006                         goto fail_unlock;
2007
2008                 inode = cprm.file->f_path.dentry->d_inode;
2009                 if (inode->i_nlink > 1)
2010                         goto close_fail;
2011                 if (d_unhashed(cprm.file->f_path.dentry))
2012                         goto close_fail;
2013                 /*
2014                  * AK: actually i see no reason to not allow this for named
2015                  * pipes etc, but keep the previous behaviour for now.
2016                  */
2017                 if (!S_ISREG(inode->i_mode))
2018                         goto close_fail;
2019                 /*
2020                  * Dont allow local users get cute and trick others to coredump
2021                  * into their pre-created files.
2022                  */
2023                 if (inode->i_uid != current_fsuid())
2024                         goto close_fail;
2025                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2026                         goto close_fail;
2027                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2028                         goto close_fail;
2029         }
2030
2031         retval = binfmt->core_dump(&cprm);
2032         if (retval)
2033                 current->signal->group_exit_code |= 0x80;
2034
2035         if (ispipe && core_pipe_limit)
2036                 wait_for_dump_helpers(cprm.file);
2037 close_fail:
2038         if (cprm.file)
2039                 filp_close(cprm.file, NULL);
2040 fail_dropcount:
2041         if (ispipe)
2042                 atomic_dec(&core_dump_count);
2043 fail_unlock:
2044         coredump_finish(mm);
2045         revert_creds(old_cred);
2046 fail_creds:
2047         put_cred(cred);
2048 fail:
2049         return;
2050 }