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/mman.h>
  28 #include <linux/a.out.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/init.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/highmem.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/key.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/pid_namespace.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/signalfd.h>
  54
  55 #include <asm/uaccess.h>
  56 #include <asm/mmu_context.h>
  57 #include <asm/tlb.h>
  58
  59 #ifdef CONFIG_KMOD
  60 #include <linux/kmod.h>
  61 #endif
  62
  63 int core_uses_pid;
  64 char core_pattern[CORENAME_MAX_SIZE] = "core";
  65 int suid_dumpable = 0;
  66
  67 EXPORT_SYMBOL(suid_dumpable);
  68 /* The maximal length of core_pattern is also specified in sysctl.c */
  69
  70 static struct linux_binfmt *formats;
  71 static DEFINE_RWLOCK(binfmt_lock);
  72
  73 int register_binfmt(struct linux_binfmt * fmt)
  74 {
  75         struct linux_binfmt ** tmp = &formats;
  76
  77         if (!fmt)
  78                 return -EINVAL;
  79         if (fmt->next)
  80                 return -EBUSY;
  81         write_lock(&binfmt_lock);
  82         while (*tmp) {
  83                 if (fmt == *tmp) {
  84                         write_unlock(&binfmt_lock);
  85                         return -EBUSY;
  86                 }
  87                 tmp = &(*tmp)->next;
  88         }
  89         fmt->next = formats;
  90         formats = fmt;
  91         write_unlock(&binfmt_lock);
  92         return 0;
  93 }
  94
  95 EXPORT_SYMBOL(register_binfmt);
  96
  97 int unregister_binfmt(struct linux_binfmt * fmt)
  98 {
  99         struct linux_binfmt ** tmp = &formats;
 100
 101         write_lock(&binfmt_lock);
 102         while (*tmp) {
 103                 if (fmt == *tmp) {
 104                         *tmp = fmt->next;
 105                         fmt->next = NULL;
 106                         write_unlock(&binfmt_lock);
 107                         return 0;
 108                 }
 109                 tmp = &(*tmp)->next;
 110         }
 111         write_unlock(&binfmt_lock);
 112         return -EINVAL;
 113 }
 114
 115 EXPORT_SYMBOL(unregister_binfmt);
 116
 117 static inline void put_binfmt(struct linux_binfmt * fmt)
 118 {
 119         module_put(fmt->module);
 120 }
 121
 122 /*
 123  * Note that a shared library must be both readable and executable due to
 124  * security reasons.
 125  *
 126  * Also note that we take the address to load from from the file itself.
 127  */
 128 asmlinkage long sys_uselib(const char __user * library)
 129 {
 130         struct file * file;
 131         struct nameidata nd;
 132         int error;
 133
 134         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 135         if (error)
 136                 goto out;
 137
 138         error = -EACCES;
 139         if (nd.mnt->mnt_flags & MNT_NOEXEC)
 140                 goto exit;
 141         error = -EINVAL;
 142         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 143                 goto exit;
 144
 145         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 146         if (error)
 147                 goto exit;
 148
 149         file = nameidata_to_filp(&nd, O_RDONLY);
 150         error = PTR_ERR(file);
 151         if (IS_ERR(file))
 152                 goto out;
 153
 154         error = -ENOEXEC;
 155         if(file->f_op) {
 156                 struct linux_binfmt * fmt;
 157
 158                 read_lock(&binfmt_lock);
 159                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 160                         if (!fmt->load_shlib)
 161                                 continue;
 162                         if (!try_module_get(fmt->module))
 163                                 continue;
 164                         read_unlock(&binfmt_lock);
 165                         error = fmt->load_shlib(file);
 166                         read_lock(&binfmt_lock);
 167                         put_binfmt(fmt);
 168                         if (error != -ENOEXEC)
 169                                 break;
 170                 }
 171                 read_unlock(&binfmt_lock);
 172         }
 173         fput(file);
 174 out:
 175         return error;
 176 exit:
 177         release_open_intent(&nd);
 178         path_release(&nd);
 179         goto out;
 180 }
 181
 182 #ifdef CONFIG_MMU
 183
 184 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 185                 int write)
 186 {
 187         struct page *page;
 188         int ret;
 189
 190 #ifdef CONFIG_STACK_GROWSUP
 191         if (write) {
 192                 ret = expand_stack_downwards(bprm->vma, pos);
 193                 if (ret < 0)
 194                         return NULL;
 195         }
 196 #endif
 197         ret = get_user_pages(current, bprm->mm, pos,
 198                         1, write, 1, &page, NULL);
 199         if (ret <= 0)
 200                 return NULL;
 201
 202         if (write) {
 203                 struct rlimit *rlim = current->signal->rlim;
 204                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 205
 206                 /*
 207                  * Limit to 1/4-th the stack size for the argv+env strings.
 208                  * This ensures that:
 209                  *  - the remaining binfmt code will not run out of stack space,
 210                  *  - the program will have a reasonable amount of stack left
 211                  *    to work from.
 212                  */
 213                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 214                         put_page(page);
 215                         return NULL;
 216                 }
 217         }
 218
 219         return page;
 220 }
 221
 222 static void put_arg_page(struct page *page)
 223 {
 224         put_page(page);
 225 }
 226
 227 static void free_arg_page(struct linux_binprm *bprm, int i)
 228 {
 229 }
 230
 231 static void free_arg_pages(struct linux_binprm *bprm)
 232 {
 233 }
 234
 235 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 236                 struct page *page)
 237 {
 238         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 239 }
 240
 241 static int __bprm_mm_init(struct linux_binprm *bprm)
 242 {
 243         int err = -ENOMEM;
 244         struct vm_area_struct *vma = NULL;
 245         struct mm_struct *mm = bprm->mm;
 246
 247         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 248         if (!vma)
 249                 goto err;
 250
 251         down_write(&mm->mmap_sem);
 252         vma->vm_mm = mm;
 253
 254         /*
 255          * Place the stack at the largest stack address the architecture
 256          * supports. Later, we'll move this to an appropriate place. We don't
 257          * use STACK_TOP because that can depend on attributes which aren't
 258          * configured yet.
 259          */
 260         vma->vm_end = STACK_TOP_MAX;
 261         vma->vm_start = vma->vm_end - PAGE_SIZE;
 262
 263         vma->vm_flags = VM_STACK_FLAGS;
 264         vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
 265         err = insert_vm_struct(mm, vma);
 266         if (err) {
 267                 up_write(&mm->mmap_sem);
 268                 goto err;
 269         }
 270
 271         mm->stack_vm = mm->total_vm = 1;
 272         up_write(&mm->mmap_sem);
 273
 274         bprm->p = vma->vm_end - sizeof(void *);
 275
 276         return 0;
 277
 278 err:
 279         if (vma) {
 280                 bprm->vma = NULL;
 281                 kmem_cache_free(vm_area_cachep, vma);
 282         }
 283
 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 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 295                 int write)
 296 {
 297         struct page *page;
 298
 299         page = bprm->page[pos / PAGE_SIZE];
 300         if (!page && write) {
 301                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 302                 if (!page)
 303                         return NULL;
 304                 bprm->page[pos / PAGE_SIZE] = page;
 305         }
 306
 307         return page;
 308 }
 309
 310 static void put_arg_page(struct page *page)
 311 {
 312 }
 313
 314 static void free_arg_page(struct linux_binprm *bprm, int i)
 315 {
 316         if (bprm->page[i]) {
 317                 __free_page(bprm->page[i]);
 318                 bprm->page[i] = NULL;
 319         }
 320 }
 321
 322 static void free_arg_pages(struct linux_binprm *bprm)
 323 {
 324         int i;
 325
 326         for (i = 0; i < MAX_ARG_PAGES; i++)
 327                 free_arg_page(bprm, i);
 328 }
 329
 330 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 331                 struct page *page)
 332 {
 333 }
 334
 335 static int __bprm_mm_init(struct linux_binprm *bprm)
 336 {
 337         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 338         return 0;
 339 }
 340
 341 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 342 {
 343         return len <= bprm->p;
 344 }
 345
 346 #endif /* CONFIG_MMU */
 347
 348 /*
 349  * Create a new mm_struct and populate it with a temporary stack
 350  * vm_area_struct.  We don't have enough context at this point to set the stack
 351  * flags, permissions, and offset, so we use temporary values.  We'll update
 352  * them later in setup_arg_pages().
 353  */
 354 int bprm_mm_init(struct linux_binprm *bprm)
 355 {
 356         int err;
 357         struct mm_struct *mm = NULL;
 358
 359         bprm->mm = mm = mm_alloc();
 360         err = -ENOMEM;
 361         if (!mm)
 362                 goto err;
 363
 364         err = init_new_context(current, mm);
 365         if (err)
 366                 goto err;
 367
 368         err = __bprm_mm_init(bprm);
 369         if (err)
 370                 goto err;
 371
 372         return 0;
 373
 374 err:
 375         if (mm) {
 376                 bprm->mm = NULL;
 377                 mmdrop(mm);
 378         }
 379
 380         return err;
 381 }
 382
 383 /*
 384  * count() counts the number of strings in array ARGV.
 385  */
 386 static int count(char __user * __user * argv, int max)
 387 {
 388         int i = 0;
 389
 390         if (argv != NULL) {
 391                 for (;;) {
 392                         char __user * p;
 393
 394                         if (get_user(p, argv))
 395                                 return -EFAULT;
 396                         if (!p)
 397                                 break;
 398                         argv++;
 399                         if(++i > max)
 400                                 return -E2BIG;
 401                         cond_resched();
 402                 }
 403         }
 404         return i;
 405 }
 406
 407 /*
 408  * 'copy_strings()' copies argument/environment strings from the old
 409  * processes's memory to the new process's stack.  The call to get_user_pages()
 410  * ensures the destination page is created and not swapped out.
 411  */
 412 static int copy_strings(int argc, char __user * __user * argv,
 413                         struct linux_binprm *bprm)
 414 {
 415         struct page *kmapped_page = NULL;
 416         char *kaddr = NULL;
 417         unsigned long kpos = 0;
 418         int ret;
 419
 420         while (argc-- > 0) {
 421                 char __user *str;
 422                 int len;
 423                 unsigned long pos;
 424
 425                 if (get_user(str, argv+argc) ||
 426                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 427                         ret = -EFAULT;
 428                         goto out;
 429                 }
 430
 431                 if (!valid_arg_len(bprm, len)) {
 432                         ret = -E2BIG;
 433                         goto out;
 434                 }
 435
 436                 /* We're going to work our way backwords. */
 437                 pos = bprm->p;
 438                 str += len;
 439                 bprm->p -= len;
 440
 441                 while (len > 0) {
 442                         int offset, bytes_to_copy;
 443
 444                         offset = pos % PAGE_SIZE;
 445                         if (offset == 0)
 446                                 offset = PAGE_SIZE;
 447
 448                         bytes_to_copy = offset;
 449                         if (bytes_to_copy > len)
 450                                 bytes_to_copy = len;
 451
 452                         offset -= bytes_to_copy;
 453                         pos -= bytes_to_copy;
 454                         str -= bytes_to_copy;
 455                         len -= bytes_to_copy;
 456
 457                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 458                                 struct page *page;
 459
 460                                 page = get_arg_page(bprm, pos, 1);
 461                                 if (!page) {
 462                                         ret = -E2BIG;
 463                                         goto out;
 464                                 }
 465
 466                                 if (kmapped_page) {
 467                                         flush_kernel_dcache_page(kmapped_page);
 468                                         kunmap(kmapped_page);
 469                                         put_arg_page(kmapped_page);
 470                                 }
 471                                 kmapped_page = page;
 472                                 kaddr = kmap(kmapped_page);
 473                                 kpos = pos & PAGE_MASK;
 474                                 flush_arg_page(bprm, kpos, kmapped_page);
 475                         }
 476                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 477                                 ret = -EFAULT;
 478                                 goto out;
 479                         }
 480                 }
 481         }
 482         ret = 0;
 483 out:
 484         if (kmapped_page) {
 485                 flush_kernel_dcache_page(kmapped_page);
 486                 kunmap(kmapped_page);
 487                 put_arg_page(kmapped_page);
 488         }
 489         return ret;
 490 }
 491
 492 /*
 493  * Like copy_strings, but get argv and its values from kernel memory.
 494  */
 495 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 496 {
 497         int r;
 498         mm_segment_t oldfs = get_fs();
 499         set_fs(KERNEL_DS);
 500         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 501         set_fs(oldfs);
 502         return r;
 503 }
 504 EXPORT_SYMBOL(copy_strings_kernel);
 505
 506 #ifdef CONFIG_MMU
 507
 508 /*
 509  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 510  * the binfmt code determines where the new stack should reside, we shift it to
 511  * its final location.  The process proceeds as follows:
 512  *
 513  * 1) Use shift to calculate the new vma endpoints.
 514  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 515  *    arguments passed to subsequent functions are consistent.
 516  * 3) Move vma's page tables to the new range.
 517  * 4) Free up any cleared pgd range.
 518  * 5) Shrink the vma to cover only the new range.
 519  */
 520 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 521 {
 522         struct mm_struct *mm = vma->vm_mm;
 523         unsigned long old_start = vma->vm_start;
 524         unsigned long old_end = vma->vm_end;
 525         unsigned long length = old_end - old_start;
 526         unsigned long new_start = old_start - shift;
 527         unsigned long new_end = old_end - shift;
 528         struct mmu_gather *tlb;
 529
 530         BUG_ON(new_start > new_end);
 531
 532         /*
 533          * ensure there are no vmas between where we want to go
 534          * and where we are
 535          */
 536         if (vma != find_vma(mm, new_start))
 537                 return -EFAULT;
 538
 539         /*
 540          * cover the whole range: [new_start, old_end)
 541          */
 542         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 543
 544         /*
 545          * move the page tables downwards, on failure we rely on
 546          * process cleanup to remove whatever mess we made.
 547          */
 548         if (length != move_page_tables(vma, old_start,
 549                                        vma, new_start, length))
 550                 return -ENOMEM;
 551
 552         lru_add_drain();
 553         tlb = tlb_gather_mmu(mm, 0);
 554         if (new_end > old_start) {
 555                 /*
 556                  * when the old and new regions overlap clear from new_end.
 557                  */
 558                 free_pgd_range(&tlb, new_end, old_end, new_end,
 559                         vma->vm_next ? vma->vm_next->vm_start : 0);
 560         } else {
 561                 /*
 562                  * otherwise, clean from old_start; this is done to not touch
 563                  * the address space in [new_end, old_start) some architectures
 564                  * have constraints on va-space that make this illegal (IA64) -
 565                  * for the others its just a little faster.
 566                  */
 567                 free_pgd_range(&tlb, old_start, old_end, new_end,
 568                         vma->vm_next ? vma->vm_next->vm_start : 0);
 569         }
 570         tlb_finish_mmu(tlb, new_end, old_end);
 571
 572         /*
 573          * shrink the vma to just the new range.
 574          */
 575         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 576
 577         return 0;
 578 }
 579
 580 #define EXTRA_STACK_VM_PAGES    20      /* random */
 581
 582 /*
 583  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 584  * the stack is optionally relocated, and some extra space is added.
 585  */
 586 int setup_arg_pages(struct linux_binprm *bprm,
 587                     unsigned long stack_top,
 588                     int executable_stack)
 589 {
 590         unsigned long ret;
 591         unsigned long stack_shift;
 592         struct mm_struct *mm = current->mm;
 593         struct vm_area_struct *vma = bprm->vma;
 594         struct vm_area_struct *prev = NULL;
 595         unsigned long vm_flags;
 596         unsigned long stack_base;
 597
 598 #ifdef CONFIG_STACK_GROWSUP
 599         /* Limit stack size to 1GB */
 600         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 601         if (stack_base > (1 << 30))
 602                 stack_base = 1 << 30;
 603
 604         /* Make sure we didn't let the argument array grow too large. */
 605         if (vma->vm_end - vma->vm_start > stack_base)
 606                 return -ENOMEM;
 607
 608         stack_base = PAGE_ALIGN(stack_top - stack_base);
 609
 610         stack_shift = vma->vm_start - stack_base;
 611         mm->arg_start = bprm->p - stack_shift;
 612         bprm->p = vma->vm_end - stack_shift;
 613 #else
 614         stack_top = arch_align_stack(stack_top);
 615         stack_top = PAGE_ALIGN(stack_top);
 616         stack_shift = vma->vm_end - stack_top;
 617
 618         bprm->p -= stack_shift;
 619         mm->arg_start = bprm->p;
 620 #endif
 621
 622         if (bprm->loader)
 623                 bprm->loader -= stack_shift;
 624         bprm->exec -= stack_shift;
 625
 626         down_write(&mm->mmap_sem);
 627         vm_flags = vma->vm_flags;
 628
 629         /*
 630          * Adjust stack execute permissions; explicitly enable for
 631          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 632          * (arch default) otherwise.
 633          */
 634         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 635                 vm_flags |= VM_EXEC;
 636         else if (executable_stack == EXSTACK_DISABLE_X)
 637                 vm_flags &= ~VM_EXEC;
 638         vm_flags |= mm->def_flags;
 639
 640         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 641                         vm_flags);
 642         if (ret)
 643                 goto out_unlock;
 644         BUG_ON(prev != vma);
 645
 646         /* Move stack pages down in memory. */
 647         if (stack_shift) {
 648                 ret = shift_arg_pages(vma, stack_shift);
 649                 if (ret) {
 650                         up_write(&mm->mmap_sem);
 651                         return ret;
 652                 }
 653         }
 654
 655 #ifdef CONFIG_STACK_GROWSUP
 656         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 657 #else
 658         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 659 #endif
 660         ret = expand_stack(vma, stack_base);
 661         if (ret)
 662                 ret = -EFAULT;
 663
 664 out_unlock:
 665         up_write(&mm->mmap_sem);
 666         return 0;
 667 }
 668 EXPORT_SYMBOL(setup_arg_pages);
 669
 670 #endif /* CONFIG_MMU */
 671
 672 struct file *open_exec(const char *name)
 673 {
 674         struct nameidata nd;
 675         int err;
 676         struct file *file;
 677
 678         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 679         file = ERR_PTR(err);
 680
 681         if (!err) {
 682                 struct inode *inode = nd.dentry->d_inode;
 683                 file = ERR_PTR(-EACCES);
 684                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 685                     S_ISREG(inode->i_mode)) {
 686                         int err = vfs_permission(&nd, MAY_EXEC);
 687                         file = ERR_PTR(err);
 688                         if (!err) {
 689                                 file = nameidata_to_filp(&nd, O_RDONLY);
 690                                 if (!IS_ERR(file)) {
 691                                         err = deny_write_access(file);
 692                                         if (err) {
 693                                                 fput(file);
 694                                                 file = ERR_PTR(err);
 695                                         }
 696                                 }
 697 out:
 698                                 return file;
 699                         }
 700                 }
 701                 release_open_intent(&nd);
 702                 path_release(&nd);
 703         }
 704         goto out;
 705 }
 706
 707 EXPORT_SYMBOL(open_exec);
 708
 709 int kernel_read(struct file *file, unsigned long offset,
 710         char *addr, unsigned long count)
 711 {
 712         mm_segment_t old_fs;
 713         loff_t pos = offset;
 714         int result;
 715
 716         old_fs = get_fs();
 717         set_fs(get_ds());
 718         /* The cast to a user pointer is valid due to the set_fs() */
 719         result = vfs_read(file, (void __user *)addr, count, &pos);
 720         set_fs(old_fs);
 721         return result;
 722 }
 723
 724 EXPORT_SYMBOL(kernel_read);
 725
 726 static int exec_mmap(struct mm_struct *mm)
 727 {
 728         struct task_struct *tsk;
 729         struct mm_struct * old_mm, *active_mm;
 730
 731         /* Notify parent that we're no longer interested in the old VM */
 732         tsk = current;
 733         old_mm = current->mm;
 734         mm_release(tsk, old_mm);
 735
 736         if (old_mm) {
 737                 /*
 738                  * Make sure that if there is a core dump in progress
 739                  * for the old mm, we get out and die instead of going
 740                  * through with the exec.  We must hold mmap_sem around
 741                  * checking core_waiters and changing tsk->mm.  The
 742                  * core-inducing thread will increment core_waiters for
 743                  * each thread whose ->mm == old_mm.
 744                  */
 745                 down_read(&old_mm->mmap_sem);
 746                 if (unlikely(old_mm->core_waiters)) {
 747                         up_read(&old_mm->mmap_sem);
 748                         return -EINTR;
 749                 }
 750         }
 751         task_lock(tsk);
 752         active_mm = tsk->active_mm;
 753         tsk->mm = mm;
 754         tsk->active_mm = mm;
 755         activate_mm(active_mm, mm);
 756         task_unlock(tsk);
 757         arch_pick_mmap_layout(mm);
 758         if (old_mm) {
 759                 up_read(&old_mm->mmap_sem);
 760                 BUG_ON(active_mm != old_mm);
 761                 mmput(old_mm);
 762                 return 0;
 763         }
 764         mmdrop(active_mm);
 765         return 0;
 766 }
 767
 768 /*
 769  * This function makes sure the current process has its own signal table,
 770  * so that flush_signal_handlers can later reset the handlers without
 771  * disturbing other processes.  (Other processes might share the signal
 772  * table via the CLONE_SIGHAND option to clone().)
 773  */
 774 static int de_thread(struct task_struct *tsk)
 775 {
 776         struct signal_struct *sig = tsk->signal;
 777         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 778         spinlock_t *lock = &oldsighand->siglock;
 779         struct task_struct *leader = NULL;
 780         int count;
 781
 782         /*
 783          * If we don't share sighandlers, then we aren't sharing anything
 784          * and we can just re-use it all.
 785          */
 786         if (atomic_read(&oldsighand->count) <= 1) {
 787                 signalfd_detach(tsk);
 788                 exit_itimers(sig);
 789                 return 0;
 790         }
 791
 792         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 793         if (!newsighand)
 794                 return -ENOMEM;
 795
 796         if (thread_group_empty(tsk))
 797                 goto no_thread_group;
 798
 799         /*
 800          * Kill all other threads in the thread group.
 801          * We must hold tasklist_lock to call zap_other_threads.
 802          */
 803         read_lock(&tasklist_lock);
 804         spin_lock_irq(lock);
 805         if (sig->flags & SIGNAL_GROUP_EXIT) {
 806                 /*
 807                  * Another group action in progress, just
 808                  * return so that the signal is processed.
 809                  */
 810                 spin_unlock_irq(lock);
 811                 read_unlock(&tasklist_lock);
 812                 kmem_cache_free(sighand_cachep, newsighand);
 813                 return -EAGAIN;
 814         }
 815
 816         /*
 817          * child_reaper ignores SIGKILL, change it now.
 818          * Reparenting needs write_lock on tasklist_lock,
 819          * so it is safe to do it under read_lock.
 820          */
 821         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 822                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 823
 824         zap_other_threads(tsk);
 825         read_unlock(&tasklist_lock);
 826
 827         /*
 828          * Account for the thread group leader hanging around:
 829          */
 830         count = 1;
 831         if (!thread_group_leader(tsk)) {
 832                 count = 2;
 833                 /*
 834                  * The SIGALRM timer survives the exec, but needs to point
 835                  * at us as the new group leader now.  We have a race with
 836                  * a timer firing now getting the old leader, so we need to
 837                  * synchronize with any firing (by calling del_timer_sync)
 838                  * before we can safely let the old group leader die.
 839                  */
 840                 sig->tsk = tsk;
 841                 spin_unlock_irq(lock);
 842                 if (hrtimer_cancel(&sig->real_timer))
 843                         hrtimer_restart(&sig->real_timer);
 844                 spin_lock_irq(lock);
 845         }
 846         while (atomic_read(&sig->count) > count) {
 847                 sig->group_exit_task = tsk;
 848                 sig->notify_count = count;
 849                 __set_current_state(TASK_UNINTERRUPTIBLE);
 850                 spin_unlock_irq(lock);
 851                 schedule();
 852                 spin_lock_irq(lock);
 853         }
 854         sig->group_exit_task = NULL;
 855         sig->notify_count = 0;
 856         spin_unlock_irq(lock);
 857
 858         /*
 859          * At this point all other threads have exited, all we have to
 860          * do is to wait for the thread group leader to become inactive,
 861          * and to assume its PID:
 862          */
 863         if (!thread_group_leader(tsk)) {
 864                 /*
 865                  * Wait for the thread group leader to be a zombie.
 866                  * It should already be zombie at this point, most
 867                  * of the time.
 868                  */
 869                 leader = tsk->group_leader;
 870                 while (leader->exit_state != EXIT_ZOMBIE)
 871                         yield();
 872
 873                 /*
 874                  * The only record we have of the real-time age of a
 875                  * process, regardless of execs it's done, is start_time.
 876                  * All the past CPU time is accumulated in signal_struct
 877                  * from sister threads now dead.  But in this non-leader
 878                  * exec, nothing survives from the original leader thread,
 879                  * whose birth marks the true age of this process now.
 880                  * When we take on its identity by switching to its PID, we
 881                  * also take its birthdate (always earlier than our own).
 882                  */
 883                 tsk->start_time = leader->start_time;
 884
 885                 write_lock_irq(&tasklist_lock);
 886
 887                 BUG_ON(leader->tgid != tsk->tgid);
 888                 BUG_ON(tsk->pid == tsk->tgid);
 889                 /*
 890                  * An exec() starts a new thread group with the
 891                  * TGID of the previous thread group. Rehash the
 892                  * two threads with a switched PID, and release
 893                  * the former thread group leader:
 894                  */
 895
 896                 /* Become a process group leader with the old leader's pid.
 897                  * The old leader becomes a thread of the this thread group.
 898                  * Note: The old leader also uses this pid until release_task
 899                  *       is called.  Odd but simple and correct.
 900                  */
 901                 detach_pid(tsk, PIDTYPE_PID);
 902                 tsk->pid = leader->pid;
 903                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
 904                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 905                 transfer_pid(leader, tsk, PIDTYPE_SID);
 906                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 907
 908                 tsk->group_leader = tsk;
 909                 leader->group_leader = tsk;
 910
 911                 tsk->exit_signal = SIGCHLD;
 912
 913                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 914                 leader->exit_state = EXIT_DEAD;
 915
 916                 write_unlock_irq(&tasklist_lock);
 917         }
 918
 919         /*
 920          * There may be one thread left which is just exiting,
 921          * but it's safe to stop telling the group to kill themselves.
 922          */
 923         sig->flags = 0;
 924
 925 no_thread_group:
 926         signalfd_detach(tsk);
 927         exit_itimers(sig);
 928         if (leader)
 929                 release_task(leader);
 930
 931         if (atomic_read(&oldsighand->count) == 1) {
 932                 /*
 933                  * Now that we nuked the rest of the thread group,
 934                  * it turns out we are not sharing sighand any more either.
 935                  * So we can just keep it.
 936                  */
 937                 kmem_cache_free(sighand_cachep, newsighand);
 938         } else {
 939                 /*
 940                  * Move our state over to newsighand and switch it in.
 941                  */
 942                 atomic_set(&newsighand->count, 1);
 943                 memcpy(newsighand->action, oldsighand->action,
 944                        sizeof(newsighand->action));
 945
 946                 write_lock_irq(&tasklist_lock);
 947                 spin_lock(&oldsighand->siglock);
 948                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 949
 950                 rcu_assign_pointer(tsk->sighand, newsighand);
 951                 recalc_sigpending();
 952
 953                 spin_unlock(&newsighand->siglock);
 954                 spin_unlock(&oldsighand->siglock);
 955                 write_unlock_irq(&tasklist_lock);
 956
 957                 __cleanup_sighand(oldsighand);
 958         }
 959
 960         BUG_ON(!thread_group_leader(tsk));
 961         return 0;
 962 }
 963
 964 /*
 965  * These functions flushes out all traces of the currently running executable
 966  * so that a new one can be started
 967  */
 968
 969 static void flush_old_files(struct files_struct * files)
 970 {
 971         long j = -1;
 972         struct fdtable *fdt;
 973
 974         spin_lock(&files->file_lock);
 975         for (;;) {
 976                 unsigned long set, i;
 977
 978                 j++;
 979                 i = j * __NFDBITS;
 980                 fdt = files_fdtable(files);
 981                 if (i >= fdt->max_fds)
 982                         break;
 983                 set = fdt->close_on_exec->fds_bits[j];
 984                 if (!set)
 985                         continue;
 986                 fdt->close_on_exec->fds_bits[j] = 0;
 987                 spin_unlock(&files->file_lock);
 988                 for ( ; set ; i++,set >>= 1) {
 989                         if (set & 1) {
 990                                 sys_close(i);
 991                         }
 992                 }
 993                 spin_lock(&files->file_lock);
 994
 995         }
 996         spin_unlock(&files->file_lock);
 997 }
 998
 999 void get_task_comm(char *buf, struct task_struct *tsk)
1000 {
1001         /* buf must be at least sizeof(tsk->comm) in size */
1002         task_lock(tsk);
1003         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1004         task_unlock(tsk);
1005 }
1006
1007 void set_task_comm(struct task_struct *tsk, char *buf)
1008 {
1009         task_lock(tsk);
1010         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1011         task_unlock(tsk);
1012 }
1013
1014 int flush_old_exec(struct linux_binprm * bprm)
1015 {
1016         char * name;
1017         int i, ch, retval;
1018         struct files_struct *files;
1019         char tcomm[sizeof(current->comm)];
1020
1021         /*
1022          * Make sure we have a private signal table and that
1023          * we are unassociated from the previous thread group.
1024          */
1025         retval = de_thread(current);
1026         if (retval)
1027                 goto out;
1028
1029         /*
1030          * Make sure we have private file handles. Ask the
1031          * fork helper to do the work for us and the exit
1032          * helper to do the cleanup of the old one.
1033          */
1034         files = current->files;         /* refcounted so safe to hold */
1035         retval = unshare_files();
1036         if (retval)
1037                 goto out;
1038         /*
1039          * Release all of the old mmap stuff
1040          */
1041         retval = exec_mmap(bprm->mm);
1042         if (retval)
1043                 goto mmap_failed;
1044
1045         bprm->mm = NULL;                /* We're using it now */
1046
1047         /* This is the point of no return */
1048         put_files_struct(files);
1049
1050         current->sas_ss_sp = current->sas_ss_size = 0;
1051
1052         if (current->euid == current->uid && current->egid == current->gid)
1053                 set_dumpable(current->mm, 1);
1054         else
1055                 set_dumpable(current->mm, suid_dumpable);
1056
1057         name = bprm->filename;
1058
1059         /* Copies the binary name from after last slash */
1060         for (i=0; (ch = *(name++)) != '\0';) {
1061                 if (ch == '/')
1062                         i = 0; /* overwrite what we wrote */
1063                 else
1064                         if (i < (sizeof(tcomm) - 1))
1065                                 tcomm[i++] = ch;
1066         }
1067         tcomm[i] = '\0';
1068         set_task_comm(current, tcomm);
1069
1070         current->flags &= ~PF_RANDOMIZE;
1071         flush_thread();
1072
1073         /* Set the new mm task size. We have to do that late because it may
1074          * depend on TIF_32BIT which is only updated in flush_thread() on
1075          * some architectures like powerpc
1076          */
1077         current->mm->task_size = TASK_SIZE;
1078
1079         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1080                 suid_keys(current);
1081                 set_dumpable(current->mm, suid_dumpable);
1082                 current->pdeath_signal = 0;
1083         } else if (file_permission(bprm->file, MAY_READ) ||
1084                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1085                 suid_keys(current);
1086                 set_dumpable(current->mm, suid_dumpable);
1087         }
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         return 0;
1098
1099 mmap_failed:
1100         reset_files_struct(current, files);
1101 out:
1102         return retval;
1103 }
1104
1105 EXPORT_SYMBOL(flush_old_exec);
1106
1107 /*
1108  * Fill the binprm structure from the inode.
1109  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1110  */
1111 int prepare_binprm(struct linux_binprm *bprm)
1112 {
1113         int mode;
1114         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1115         int retval;
1116
1117         mode = inode->i_mode;
1118         if (bprm->file->f_op == NULL)
1119                 return -EACCES;
1120
1121         bprm->e_uid = current->euid;
1122         bprm->e_gid = current->egid;
1123
1124         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1125                 /* Set-uid? */
1126                 if (mode & S_ISUID) {
1127                         current->personality &= ~PER_CLEAR_ON_SETID;
1128                         bprm->e_uid = inode->i_uid;
1129                 }
1130
1131                 /* Set-gid? */
1132                 /*
1133                  * If setgid is set but no group execute bit then this
1134                  * is a candidate for mandatory locking, not a setgid
1135                  * executable.
1136                  */
1137                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1138                         current->personality &= ~PER_CLEAR_ON_SETID;
1139                         bprm->e_gid = inode->i_gid;
1140                 }
1141         }
1142
1143         /* fill in binprm security blob */
1144         retval = security_bprm_set(bprm);
1145         if (retval)
1146                 return retval;
1147
1148         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1149         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1150 }
1151
1152 EXPORT_SYMBOL(prepare_binprm);
1153
1154 static int unsafe_exec(struct task_struct *p)
1155 {
1156         int unsafe = 0;
1157         if (p->ptrace & PT_PTRACED) {
1158                 if (p->ptrace & PT_PTRACE_CAP)
1159                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1160                 else
1161                         unsafe |= LSM_UNSAFE_PTRACE;
1162         }
1163         if (atomic_read(&p->fs->count) > 1 ||
1164             atomic_read(&p->files->count) > 1 ||
1165             atomic_read(&p->sighand->count) > 1)
1166                 unsafe |= LSM_UNSAFE_SHARE;
1167
1168         return unsafe;
1169 }
1170
1171 void compute_creds(struct linux_binprm *bprm)
1172 {
1173         int unsafe;
1174
1175         if (bprm->e_uid != current->uid) {
1176                 suid_keys(current);
1177                 current->pdeath_signal = 0;
1178         }
1179         exec_keys(current);
1180
1181         task_lock(current);
1182         unsafe = unsafe_exec(current);
1183         security_bprm_apply_creds(bprm, unsafe);
1184         task_unlock(current);
1185         security_bprm_post_apply_creds(bprm);
1186 }
1187 EXPORT_SYMBOL(compute_creds);
1188
1189 /*
1190  * Arguments are '\0' separated strings found at the location bprm->p
1191  * points to; chop off the first by relocating brpm->p to right after
1192  * the first '\0' encountered.
1193  */
1194 int remove_arg_zero(struct linux_binprm *bprm)
1195 {
1196         int ret = 0;
1197         unsigned long offset;
1198         char *kaddr;
1199         struct page *page;
1200
1201         if (!bprm->argc)
1202                 return 0;
1203
1204         do {
1205                 offset = bprm->p & ~PAGE_MASK;
1206                 page = get_arg_page(bprm, bprm->p, 0);
1207                 if (!page) {
1208                         ret = -EFAULT;
1209                         goto out;
1210                 }
1211                 kaddr = kmap_atomic(page, KM_USER0);
1212
1213                 for (; offset < PAGE_SIZE && kaddr[offset];
1214                                 offset++, bprm->p++)
1215                         ;
1216
1217                 kunmap_atomic(kaddr, KM_USER0);
1218                 put_arg_page(page);
1219
1220                 if (offset == PAGE_SIZE)
1221                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1222         } while (offset == PAGE_SIZE);
1223
1224         bprm->p++;
1225         bprm->argc--;
1226         ret = 0;
1227
1228 out:
1229         return ret;
1230 }
1231 EXPORT_SYMBOL(remove_arg_zero);
1232
1233 /*
1234  * cycle the list of binary formats handler, until one recognizes the image
1235  */
1236 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1237 {
1238         int try,retval;
1239         struct linux_binfmt *fmt;
1240 #ifdef __alpha__
1241         /* handle /sbin/loader.. */
1242         {
1243             struct exec * eh = (struct exec *) bprm->buf;
1244
1245             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1246                 (eh->fh.f_flags & 0x3000) == 0x3000)
1247             {
1248                 struct file * file;
1249                 unsigned long loader;
1250
1251                 allow_write_access(bprm->file);
1252                 fput(bprm->file);
1253                 bprm->file = NULL;
1254
1255                 loader = bprm->vma->vm_end - sizeof(void *);
1256
1257                 file = open_exec("/sbin/loader");
1258                 retval = PTR_ERR(file);
1259                 if (IS_ERR(file))
1260                         return retval;
1261
1262                 /* Remember if the application is TASO.  */
1263                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1264
1265                 bprm->file = file;
1266                 bprm->loader = loader;
1267                 retval = prepare_binprm(bprm);
1268                 if (retval<0)
1269                         return retval;
1270                 /* should call search_binary_handler recursively here,
1271                    but it does not matter */
1272             }
1273         }
1274 #endif
1275         retval = security_bprm_check(bprm);
1276         if (retval)
1277                 return retval;
1278
1279         /* kernel module loader fixup */
1280         /* so we don't try to load run modprobe in kernel space. */
1281         set_fs(USER_DS);
1282
1283         retval = audit_bprm(bprm);
1284         if (retval)
1285                 return retval;
1286
1287         retval = -ENOENT;
1288         for (try=0; try<2; try++) {
1289                 read_lock(&binfmt_lock);
1290                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1291                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1292                         if (!fn)
1293                                 continue;
1294                         if (!try_module_get(fmt->module))
1295                                 continue;
1296                         read_unlock(&binfmt_lock);
1297                         retval = fn(bprm, regs);
1298                         if (retval >= 0) {
1299                                 put_binfmt(fmt);
1300                                 allow_write_access(bprm->file);
1301                                 if (bprm->file)
1302                                         fput(bprm->file);
1303                                 bprm->file = NULL;
1304                                 current->did_exec = 1;
1305                                 proc_exec_connector(current);
1306                                 return retval;
1307                         }
1308                         read_lock(&binfmt_lock);
1309                         put_binfmt(fmt);
1310                         if (retval != -ENOEXEC || bprm->mm == NULL)
1311                                 break;
1312                         if (!bprm->file) {
1313                                 read_unlock(&binfmt_lock);
1314                                 return retval;
1315                         }
1316                 }
1317                 read_unlock(&binfmt_lock);
1318                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1319                         break;
1320 #ifdef CONFIG_KMOD
1321                 }else{
1322 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1323                         if (printable(bprm->buf[0]) &&
1324                             printable(bprm->buf[1]) &&
1325                             printable(bprm->buf[2]) &&
1326                             printable(bprm->buf[3]))
1327                                 break; /* -ENOEXEC */
1328                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1329 #endif
1330                 }
1331         }
1332         return retval;
1333 }
1334
1335 EXPORT_SYMBOL(search_binary_handler);
1336
1337 /*
1338  * sys_execve() executes a new program.
1339  */
1340 int do_execve(char * filename,
1341         char __user *__user *argv,
1342         char __user *__user *envp,
1343         struct pt_regs * regs)
1344 {
1345         struct linux_binprm *bprm;
1346         struct file *file;
1347         unsigned long env_p;
1348         int retval;
1349
1350         retval = -ENOMEM;
1351         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1352         if (!bprm)
1353                 goto out_ret;
1354
1355         file = open_exec(filename);
1356         retval = PTR_ERR(file);
1357         if (IS_ERR(file))
1358                 goto out_kfree;
1359
1360         sched_exec();
1361
1362         bprm->file = file;
1363         bprm->filename = filename;
1364         bprm->interp = filename;
1365
1366         retval = bprm_mm_init(bprm);
1367         if (retval)
1368                 goto out_file;
1369
1370         bprm->argc = count(argv, MAX_ARG_STRINGS);
1371         if ((retval = bprm->argc) < 0)
1372                 goto out_mm;
1373
1374         bprm->envc = count(envp, MAX_ARG_STRINGS);
1375         if ((retval = bprm->envc) < 0)
1376                 goto out_mm;
1377
1378         retval = security_bprm_alloc(bprm);
1379         if (retval)
1380                 goto out;
1381
1382         retval = prepare_binprm(bprm);
1383         if (retval < 0)
1384                 goto out;
1385
1386         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1387         if (retval < 0)
1388                 goto out;
1389
1390         bprm->exec = bprm->p;
1391         retval = copy_strings(bprm->envc, envp, bprm);
1392         if (retval < 0)
1393                 goto out;
1394
1395         env_p = bprm->p;
1396         retval = copy_strings(bprm->argc, argv, bprm);
1397         if (retval < 0)
1398                 goto out;
1399         bprm->argv_len = env_p - bprm->p;
1400
1401         retval = search_binary_handler(bprm,regs);
1402         if (retval >= 0) {
1403                 /* execve success */
1404                 free_arg_pages(bprm);
1405                 security_bprm_free(bprm);
1406                 acct_update_integrals(current);
1407                 kfree(bprm);
1408                 return retval;
1409         }
1410
1411 out:
1412         free_arg_pages(bprm);
1413         if (bprm->security)
1414                 security_bprm_free(bprm);
1415
1416 out_mm:
1417         if (bprm->mm)
1418                 mmput (bprm->mm);
1419
1420 out_file:
1421         if (bprm->file) {
1422                 allow_write_access(bprm->file);
1423                 fput(bprm->file);
1424         }
1425 out_kfree:
1426         kfree(bprm);
1427
1428 out_ret:
1429         return retval;
1430 }
1431
1432 int set_binfmt(struct linux_binfmt *new)
1433 {
1434         struct linux_binfmt *old = current->binfmt;
1435
1436         if (new) {
1437                 if (!try_module_get(new->module))
1438                         return -1;
1439         }
1440         current->binfmt = new;
1441         if (old)
1442                 module_put(old->module);
1443         return 0;
1444 }
1445
1446 EXPORT_SYMBOL(set_binfmt);
1447
1448 /* format_corename will inspect the pattern parameter, and output a
1449  * name into corename, which must have space for at least
1450  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1451  */
1452 static int format_corename(char *corename, const char *pattern, long signr)
1453 {
1454         const char *pat_ptr = pattern;
1455         char *out_ptr = corename;
1456         char *const out_end = corename + CORENAME_MAX_SIZE;
1457         int rc;
1458         int pid_in_pattern = 0;
1459         int ispipe = 0;
1460
1461         if (*pattern == '|')
1462                 ispipe = 1;
1463
1464         /* Repeat as long as we have more pattern to process and more output
1465            space */
1466         while (*pat_ptr) {
1467                 if (*pat_ptr != '%') {
1468                         if (out_ptr == out_end)
1469                                 goto out;
1470                         *out_ptr++ = *pat_ptr++;
1471                 } else {
1472                         switch (*++pat_ptr) {
1473                         case 0:
1474                                 goto out;
1475                         /* Double percent, output one percent */
1476                         case '%':
1477                                 if (out_ptr == out_end)
1478                                         goto out;
1479                                 *out_ptr++ = '%';
1480                                 break;
1481                         /* pid */
1482                         case 'p':
1483                                 pid_in_pattern = 1;
1484                                 rc = snprintf(out_ptr, out_end - out_ptr,
1485                                               "%d", current->tgid);
1486                                 if (rc > out_end - out_ptr)
1487                                         goto out;
1488                                 out_ptr += rc;
1489                                 break;
1490                         /* uid */
1491                         case 'u':
1492                                 rc = snprintf(out_ptr, out_end - out_ptr,
1493                                               "%d", current->uid);
1494                                 if (rc > out_end - out_ptr)
1495                                         goto out;
1496                                 out_ptr += rc;
1497                                 break;
1498                         /* gid */
1499                         case 'g':
1500                                 rc = snprintf(out_ptr, out_end - out_ptr,
1501                                               "%d", current->gid);
1502                                 if (rc > out_end - out_ptr)
1503                                         goto out;
1504                                 out_ptr += rc;
1505                                 break;
1506                         /* signal that caused the coredump */
1507                         case 's':
1508                                 rc = snprintf(out_ptr, out_end - out_ptr,
1509                                               "%ld", signr);
1510                                 if (rc > out_end - out_ptr)
1511                                         goto out;
1512                                 out_ptr += rc;
1513                                 break;
1514                         /* UNIX time of coredump */
1515                         case 't': {
1516                                 struct timeval tv;
1517                                 do_gettimeofday(&tv);
1518                                 rc = snprintf(out_ptr, out_end - out_ptr,
1519                                               "%lu", tv.tv_sec);
1520                                 if (rc > out_end - out_ptr)
1521                                         goto out;
1522                                 out_ptr += rc;
1523                                 break;
1524                         }
1525                         /* hostname */
1526                         case 'h':
1527                                 down_read(&uts_sem);
1528                                 rc = snprintf(out_ptr, out_end - out_ptr,
1529                                               "%s", utsname()->nodename);
1530                                 up_read(&uts_sem);
1531                                 if (rc > out_end - out_ptr)
1532                                         goto out;
1533                                 out_ptr += rc;
1534                                 break;
1535                         /* executable */
1536                         case 'e':
1537                                 rc = snprintf(out_ptr, out_end - out_ptr,
1538                                               "%s", current->comm);
1539                                 if (rc > out_end - out_ptr)
1540                                         goto out;
1541                                 out_ptr += rc;
1542                                 break;
1543                         default:
1544                                 break;
1545                         }
1546                         ++pat_ptr;
1547                 }
1548         }
1549         /* Backward compatibility with core_uses_pid:
1550          *
1551          * If core_pattern does not include a %p (as is the default)
1552          * and core_uses_pid is set, then .%pid will be appended to
1553          * the filename. Do not do this for piped commands. */
1554         if (!ispipe && !pid_in_pattern
1555             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1556                 rc = snprintf(out_ptr, out_end - out_ptr,
1557                               ".%d", current->tgid);
1558                 if (rc > out_end - out_ptr)
1559                         goto out;
1560                 out_ptr += rc;
1561         }
1562 out:
1563         *out_ptr = 0;
1564         return ispipe;
1565 }
1566
1567 static void zap_process(struct task_struct *start)
1568 {
1569         struct task_struct *t;
1570
1571         start->signal->flags = SIGNAL_GROUP_EXIT;
1572         start->signal->group_stop_count = 0;
1573
1574         t = start;
1575         do {
1576                 if (t != current && t->mm) {
1577                         t->mm->core_waiters++;
1578                         sigaddset(&t->pending.signal, SIGKILL);
1579                         signal_wake_up(t, 1);
1580                 }
1581         } while ((t = next_thread(t)) != start);
1582 }
1583
1584 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1585                                 int exit_code)
1586 {
1587         struct task_struct *g, *p;
1588         unsigned long flags;
1589         int err = -EAGAIN;
1590
1591         spin_lock_irq(&tsk->sighand->siglock);
1592         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1593                 tsk->signal->group_exit_code = exit_code;
1594                 zap_process(tsk);
1595                 err = 0;
1596         }
1597         spin_unlock_irq(&tsk->sighand->siglock);
1598         if (err)
1599                 return err;
1600
1601         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1602                 goto done;
1603
1604         rcu_read_lock();
1605         for_each_process(g) {
1606                 if (g == tsk->group_leader)
1607                         continue;
1608
1609                 p = g;
1610                 do {
1611                         if (p->mm) {
1612                                 if (p->mm == mm) {
1613                                         /*
1614                                          * p->sighand can't disappear, but
1615                                          * may be changed by de_thread()
1616                                          */
1617                                         lock_task_sighand(p, &flags);
1618                                         zap_process(p);
1619                                         unlock_task_sighand(p, &flags);
1620                                 }
1621                                 break;
1622                         }
1623                 } while ((p = next_thread(p)) != g);
1624         }
1625         rcu_read_unlock();
1626 done:
1627         return mm->core_waiters;
1628 }
1629
1630 static int coredump_wait(int exit_code)
1631 {
1632         struct task_struct *tsk = current;
1633         struct mm_struct *mm = tsk->mm;
1634         struct completion startup_done;
1635         struct completion *vfork_done;
1636         int core_waiters;
1637
1638         init_completion(&mm->core_done);
1639         init_completion(&startup_done);
1640         mm->core_startup_done = &startup_done;
1641
1642         core_waiters = zap_threads(tsk, mm, exit_code);
1643         up_write(&mm->mmap_sem);
1644
1645         if (unlikely(core_waiters < 0))
1646                 goto fail;
1647
1648         /*
1649          * Make sure nobody is waiting for us to release the VM,
1650          * otherwise we can deadlock when we wait on each other
1651          */
1652         vfork_done = tsk->vfork_done;
1653         if (vfork_done) {
1654                 tsk->vfork_done = NULL;
1655                 complete(vfork_done);
1656         }
1657
1658         if (core_waiters)
1659                 wait_for_completion(&startup_done);
1660 fail:
1661         BUG_ON(mm->core_waiters);
1662         return core_waiters;
1663 }
1664
1665 /*
1666  * set_dumpable converts traditional three-value dumpable to two flags and
1667  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1668  * these bits are not changed atomically.  So get_dumpable can observe the
1669  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1670  * return either old dumpable or new one by paying attention to the order of
1671  * modifying the bits.
1672  *
1673  * dumpable |   mm->flags (binary)
1674  * old  new | initial interim  final
1675  * ---------+-----------------------
1676  *  0    1  |   00      01      01
1677  *  0    2  |   00      10(*)   11
1678  *  1    0  |   01      00      00
1679  *  1    2  |   01      11      11
1680  *  2    0  |   11      10(*)   00
1681  *  2    1  |   11      11      01
1682  *
1683  * (*) get_dumpable regards interim value of 10 as 11.
1684  */
1685 void set_dumpable(struct mm_struct *mm, int value)
1686 {
1687         switch (value) {
1688         case 0:
1689                 clear_bit(MMF_DUMPABLE, &mm->flags);
1690                 smp_wmb();
1691                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1692                 break;
1693         case 1:
1694                 set_bit(MMF_DUMPABLE, &mm->flags);
1695                 smp_wmb();
1696                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1697                 break;
1698         case 2:
1699                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1700                 smp_wmb();
1701                 set_bit(MMF_DUMPABLE, &mm->flags);
1702                 break;
1703         }
1704 }
1705 EXPORT_SYMBOL_GPL(set_dumpable);
1706
1707 int get_dumpable(struct mm_struct *mm)
1708 {
1709         int ret;
1710
1711         ret = mm->flags & 0x3;
1712         return (ret >= 2) ? 2 : ret;
1713 }
1714
1715 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1716 {
1717         char corename[CORENAME_MAX_SIZE + 1];
1718         struct mm_struct *mm = current->mm;
1719         struct linux_binfmt * binfmt;
1720         struct inode * inode;
1721         struct file * file;
1722         int retval = 0;
1723         int fsuid = current->fsuid;
1724         int flag = 0;
1725         int ispipe = 0;
1726
1727         audit_core_dumps(signr);
1728
1729         binfmt = current->binfmt;
1730         if (!binfmt || !binfmt->core_dump)
1731                 goto fail;
1732         down_write(&mm->mmap_sem);
1733         if (!get_dumpable(mm)) {
1734                 up_write(&mm->mmap_sem);
1735                 goto fail;
1736         }
1737
1738         /*
1739          *      We cannot trust fsuid as being the "true" uid of the
1740          *      process nor do we know its entire history. We only know it
1741          *      was tainted so we dump it as root in mode 2.
1742          */
1743         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1744                 flag = O_EXCL;          /* Stop rewrite attacks */
1745                 current->fsuid = 0;     /* Dump root private */
1746         }
1747         set_dumpable(mm, 0);
1748
1749         retval = coredump_wait(exit_code);
1750         if (retval < 0)
1751                 goto fail;
1752
1753         /*
1754          * Clear any false indication of pending signals that might
1755          * be seen by the filesystem code called to write the core file.
1756          */
1757         clear_thread_flag(TIF_SIGPENDING);
1758
1759         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1760                 goto fail_unlock;
1761
1762         /*
1763          * lock_kernel() because format_corename() is controlled by sysctl, which
1764          * uses lock_kernel()
1765          */
1766         lock_kernel();
1767         ispipe = format_corename(corename, core_pattern, signr);
1768         unlock_kernel();
1769         if (ispipe) {
1770                 /* SIGPIPE can happen, but it's just never processed */
1771                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1772                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1773                                corename);
1774                         goto fail_unlock;
1775                 }
1776         } else
1777                 file = filp_open(corename,
1778                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1779                                  0600);
1780         if (IS_ERR(file))
1781                 goto fail_unlock;
1782         inode = file->f_path.dentry->d_inode;
1783         if (inode->i_nlink > 1)
1784                 goto close_fail;        /* multiple links - don't dump */
1785         if (!ispipe && d_unhashed(file->f_path.dentry))
1786                 goto close_fail;
1787
1788         /* AK: actually i see no reason to not allow this for named pipes etc.,
1789            but keep the previous behaviour for now. */
1790         if (!ispipe && !S_ISREG(inode->i_mode))
1791                 goto close_fail;
1792         if (!file->f_op)
1793                 goto close_fail;
1794         if (!file->f_op->write)
1795                 goto close_fail;
1796         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1797                 goto close_fail;
1798
1799         retval = binfmt->core_dump(signr, regs, file);
1800
1801         if (retval)
1802                 current->signal->group_exit_code |= 0x80;
1803 close_fail:
1804         filp_close(file, NULL);
1805 fail_unlock:
1806         current->fsuid = fsuid;
1807         complete_all(&mm->core_done);
1808 fail:
1809         return retval;
1810 }