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          * Tell all the sighand listeners that this sighand has
 784          * been detached. The signalfd_detach() function grabs the
 785          * sighand lock, if signal listeners are present on the sighand.
 786          */
 787         signalfd_detach(tsk);
 788
 789         /*
 790          * If we don't share sighandlers, then we aren't sharing anything
 791          * and we can just re-use it all.
 792          */
 793         if (atomic_read(&oldsighand->count) <= 1) {
 794                 BUG_ON(atomic_read(&sig->count) != 1);
 795                 exit_itimers(sig);
 796                 return 0;
 797         }
 798
 799         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 800         if (!newsighand)
 801                 return -ENOMEM;
 802
 803         if (thread_group_empty(tsk))
 804                 goto no_thread_group;
 805
 806         /*
 807          * Kill all other threads in the thread group.
 808          * We must hold tasklist_lock to call zap_other_threads.
 809          */
 810         read_lock(&tasklist_lock);
 811         spin_lock_irq(lock);
 812         if (sig->flags & SIGNAL_GROUP_EXIT) {
 813                 /*
 814                  * Another group action in progress, just
 815                  * return so that the signal is processed.
 816                  */
 817                 spin_unlock_irq(lock);
 818                 read_unlock(&tasklist_lock);
 819                 kmem_cache_free(sighand_cachep, newsighand);
 820                 return -EAGAIN;
 821         }
 822
 823         /*
 824          * child_reaper ignores SIGKILL, change it now.
 825          * Reparenting needs write_lock on tasklist_lock,
 826          * so it is safe to do it under read_lock.
 827          */
 828         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 829                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 830
 831         zap_other_threads(tsk);
 832         read_unlock(&tasklist_lock);
 833
 834         /*
 835          * Account for the thread group leader hanging around:
 836          */
 837         count = 1;
 838         if (!thread_group_leader(tsk)) {
 839                 count = 2;
 840                 /*
 841                  * The SIGALRM timer survives the exec, but needs to point
 842                  * at us as the new group leader now.  We have a race with
 843                  * a timer firing now getting the old leader, so we need to
 844                  * synchronize with any firing (by calling del_timer_sync)
 845                  * before we can safely let the old group leader die.
 846                  */
 847                 sig->tsk = tsk;
 848                 spin_unlock_irq(lock);
 849                 if (hrtimer_cancel(&sig->real_timer))
 850                         hrtimer_restart(&sig->real_timer);
 851                 spin_lock_irq(lock);
 852         }
 853         while (atomic_read(&sig->count) > count) {
 854                 sig->group_exit_task = tsk;
 855                 sig->notify_count = count;
 856                 __set_current_state(TASK_UNINTERRUPTIBLE);
 857                 spin_unlock_irq(lock);
 858                 schedule();
 859                 spin_lock_irq(lock);
 860         }
 861         sig->group_exit_task = NULL;
 862         sig->notify_count = 0;
 863         spin_unlock_irq(lock);
 864
 865         /*
 866          * At this point all other threads have exited, all we have to
 867          * do is to wait for the thread group leader to become inactive,
 868          * and to assume its PID:
 869          */
 870         if (!thread_group_leader(tsk)) {
 871                 /*
 872                  * Wait for the thread group leader to be a zombie.
 873                  * It should already be zombie at this point, most
 874                  * of the time.
 875                  */
 876                 leader = tsk->group_leader;
 877                 while (leader->exit_state != EXIT_ZOMBIE)
 878                         yield();
 879
 880                 /*
 881                  * The only record we have of the real-time age of a
 882                  * process, regardless of execs it's done, is start_time.
 883                  * All the past CPU time is accumulated in signal_struct
 884                  * from sister threads now dead.  But in this non-leader
 885                  * exec, nothing survives from the original leader thread,
 886                  * whose birth marks the true age of this process now.
 887                  * When we take on its identity by switching to its PID, we
 888                  * also take its birthdate (always earlier than our own).
 889                  */
 890                 tsk->start_time = leader->start_time;
 891
 892                 write_lock_irq(&tasklist_lock);
 893
 894                 BUG_ON(leader->tgid != tsk->tgid);
 895                 BUG_ON(tsk->pid == tsk->tgid);
 896                 /*
 897                  * An exec() starts a new thread group with the
 898                  * TGID of the previous thread group. Rehash the
 899                  * two threads with a switched PID, and release
 900                  * the former thread group leader:
 901                  */
 902
 903                 /* Become a process group leader with the old leader's pid.
 904                  * The old leader becomes a thread of the this thread group.
 905                  * Note: The old leader also uses this pid until release_task
 906                  *       is called.  Odd but simple and correct.
 907                  */
 908                 detach_pid(tsk, PIDTYPE_PID);
 909                 tsk->pid = leader->pid;
 910                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
 911                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 912                 transfer_pid(leader, tsk, PIDTYPE_SID);
 913                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 914
 915                 tsk->group_leader = tsk;
 916                 leader->group_leader = tsk;
 917
 918                 tsk->exit_signal = SIGCHLD;
 919
 920                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 921                 leader->exit_state = EXIT_DEAD;
 922
 923                 write_unlock_irq(&tasklist_lock);
 924         }
 925
 926         /*
 927          * There may be one thread left which is just exiting,
 928          * but it's safe to stop telling the group to kill themselves.
 929          */
 930         sig->flags = 0;
 931
 932 no_thread_group:
 933         exit_itimers(sig);
 934         if (leader)
 935                 release_task(leader);
 936
 937         BUG_ON(atomic_read(&sig->count) != 1);
 938
 939         if (atomic_read(&oldsighand->count) == 1) {
 940                 /*
 941                  * Now that we nuked the rest of the thread group,
 942                  * it turns out we are not sharing sighand any more either.
 943                  * So we can just keep it.
 944                  */
 945                 kmem_cache_free(sighand_cachep, newsighand);
 946         } else {
 947                 /*
 948                  * Move our state over to newsighand and switch it in.
 949                  */
 950                 atomic_set(&newsighand->count, 1);
 951                 memcpy(newsighand->action, oldsighand->action,
 952                        sizeof(newsighand->action));
 953
 954                 write_lock_irq(&tasklist_lock);
 955                 spin_lock(&oldsighand->siglock);
 956                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 957
 958                 rcu_assign_pointer(tsk->sighand, newsighand);
 959                 recalc_sigpending();
 960
 961                 spin_unlock(&newsighand->siglock);
 962                 spin_unlock(&oldsighand->siglock);
 963                 write_unlock_irq(&tasklist_lock);
 964
 965                 __cleanup_sighand(oldsighand);
 966         }
 967
 968         BUG_ON(!thread_group_leader(tsk));
 969         return 0;
 970 }
 971
 972 /*
 973  * These functions flushes out all traces of the currently running executable
 974  * so that a new one can be started
 975  */
 976
 977 static void flush_old_files(struct files_struct * files)
 978 {
 979         long j = -1;
 980         struct fdtable *fdt;
 981
 982         spin_lock(&files->file_lock);
 983         for (;;) {
 984                 unsigned long set, i;
 985
 986                 j++;
 987                 i = j * __NFDBITS;
 988                 fdt = files_fdtable(files);
 989                 if (i >= fdt->max_fds)
 990                         break;
 991                 set = fdt->close_on_exec->fds_bits[j];
 992                 if (!set)
 993                         continue;
 994                 fdt->close_on_exec->fds_bits[j] = 0;
 995                 spin_unlock(&files->file_lock);
 996                 for ( ; set ; i++,set >>= 1) {
 997                         if (set & 1) {
 998                                 sys_close(i);
 999                         }
1000                 }
1001                 spin_lock(&files->file_lock);
1002
1003         }
1004         spin_unlock(&files->file_lock);
1005 }
1006
1007 void get_task_comm(char *buf, struct task_struct *tsk)
1008 {
1009         /* buf must be at least sizeof(tsk->comm) in size */
1010         task_lock(tsk);
1011         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1012         task_unlock(tsk);
1013 }
1014
1015 void set_task_comm(struct task_struct *tsk, char *buf)
1016 {
1017         task_lock(tsk);
1018         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1019         task_unlock(tsk);
1020 }
1021
1022 int flush_old_exec(struct linux_binprm * bprm)
1023 {
1024         char * name;
1025         int i, ch, retval;
1026         struct files_struct *files;
1027         char tcomm[sizeof(current->comm)];
1028
1029         /*
1030          * Make sure we have a private signal table and that
1031          * we are unassociated from the previous thread group.
1032          */
1033         retval = de_thread(current);
1034         if (retval)
1035                 goto out;
1036
1037         /*
1038          * Make sure we have private file handles. Ask the
1039          * fork helper to do the work for us and the exit
1040          * helper to do the cleanup of the old one.
1041          */
1042         files = current->files;         /* refcounted so safe to hold */
1043         retval = unshare_files();
1044         if (retval)
1045                 goto out;
1046         /*
1047          * Release all of the old mmap stuff
1048          */
1049         retval = exec_mmap(bprm->mm);
1050         if (retval)
1051                 goto mmap_failed;
1052
1053         bprm->mm = NULL;                /* We're using it now */
1054
1055         /* This is the point of no return */
1056         put_files_struct(files);
1057
1058         current->sas_ss_sp = current->sas_ss_size = 0;
1059
1060         if (current->euid == current->uid && current->egid == current->gid)
1061                 set_dumpable(current->mm, 1);
1062         else
1063                 set_dumpable(current->mm, suid_dumpable);
1064
1065         name = bprm->filename;
1066
1067         /* Copies the binary name from after last slash */
1068         for (i=0; (ch = *(name++)) != '\0';) {
1069                 if (ch == '/')
1070                         i = 0; /* overwrite what we wrote */
1071                 else
1072                         if (i < (sizeof(tcomm) - 1))
1073                                 tcomm[i++] = ch;
1074         }
1075         tcomm[i] = '\0';
1076         set_task_comm(current, tcomm);
1077
1078         current->flags &= ~PF_RANDOMIZE;
1079         flush_thread();
1080
1081         /* Set the new mm task size. We have to do that late because it may
1082          * depend on TIF_32BIT which is only updated in flush_thread() on
1083          * some architectures like powerpc
1084          */
1085         current->mm->task_size = TASK_SIZE;
1086
1087         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
1088             file_permission(bprm->file, MAY_READ) ||
1089             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1090                 suid_keys(current);
1091                 set_dumpable(current->mm, suid_dumpable);
1092         }
1093
1094         /* An exec changes our domain. We are no longer part of the thread
1095            group */
1096
1097         current->self_exec_id++;
1098
1099         flush_signal_handlers(current, 0);
1100         flush_old_files(current->files);
1101
1102         return 0;
1103
1104 mmap_failed:
1105         reset_files_struct(current, files);
1106 out:
1107         return retval;
1108 }
1109
1110 EXPORT_SYMBOL(flush_old_exec);
1111
1112 /*
1113  * Fill the binprm structure from the inode.
1114  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1115  */
1116 int prepare_binprm(struct linux_binprm *bprm)
1117 {
1118         int mode;
1119         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1120         int retval;
1121
1122         mode = inode->i_mode;
1123         if (bprm->file->f_op == NULL)
1124                 return -EACCES;
1125
1126         bprm->e_uid = current->euid;
1127         bprm->e_gid = current->egid;
1128
1129         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1130                 /* Set-uid? */
1131                 if (mode & S_ISUID) {
1132                         current->personality &= ~PER_CLEAR_ON_SETID;
1133                         bprm->e_uid = inode->i_uid;
1134                 }
1135
1136                 /* Set-gid? */
1137                 /*
1138                  * If setgid is set but no group execute bit then this
1139                  * is a candidate for mandatory locking, not a setgid
1140                  * executable.
1141                  */
1142                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1143                         current->personality &= ~PER_CLEAR_ON_SETID;
1144                         bprm->e_gid = inode->i_gid;
1145                 }
1146         }
1147
1148         /* fill in binprm security blob */
1149         retval = security_bprm_set(bprm);
1150         if (retval)
1151                 return retval;
1152
1153         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1154         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1155 }
1156
1157 EXPORT_SYMBOL(prepare_binprm);
1158
1159 static int unsafe_exec(struct task_struct *p)
1160 {
1161         int unsafe = 0;
1162         if (p->ptrace & PT_PTRACED) {
1163                 if (p->ptrace & PT_PTRACE_CAP)
1164                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1165                 else
1166                         unsafe |= LSM_UNSAFE_PTRACE;
1167         }
1168         if (atomic_read(&p->fs->count) > 1 ||
1169             atomic_read(&p->files->count) > 1 ||
1170             atomic_read(&p->sighand->count) > 1)
1171                 unsafe |= LSM_UNSAFE_SHARE;
1172
1173         return unsafe;
1174 }
1175
1176 void compute_creds(struct linux_binprm *bprm)
1177 {
1178         int unsafe;
1179
1180         if (bprm->e_uid != current->uid)
1181                 suid_keys(current);
1182         exec_keys(current);
1183
1184         task_lock(current);
1185         unsafe = unsafe_exec(current);
1186         security_bprm_apply_creds(bprm, unsafe);
1187         task_unlock(current);
1188         security_bprm_post_apply_creds(bprm);
1189 }
1190 EXPORT_SYMBOL(compute_creds);
1191
1192 /*
1193  * Arguments are '\0' separated strings found at the location bprm->p
1194  * points to; chop off the first by relocating brpm->p to right after
1195  * the first '\0' encountered.
1196  */
1197 int remove_arg_zero(struct linux_binprm *bprm)
1198 {
1199         int ret = 0;
1200         unsigned long offset;
1201         char *kaddr;
1202         struct page *page;
1203
1204         if (!bprm->argc)
1205                 return 0;
1206
1207         do {
1208                 offset = bprm->p & ~PAGE_MASK;
1209                 page = get_arg_page(bprm, bprm->p, 0);
1210                 if (!page) {
1211                         ret = -EFAULT;
1212                         goto out;
1213                 }
1214                 kaddr = kmap_atomic(page, KM_USER0);
1215
1216                 for (; offset < PAGE_SIZE && kaddr[offset];
1217                                 offset++, bprm->p++)
1218                         ;
1219
1220                 kunmap_atomic(kaddr, KM_USER0);
1221                 put_arg_page(page);
1222
1223                 if (offset == PAGE_SIZE)
1224                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1225         } while (offset == PAGE_SIZE);
1226
1227         bprm->p++;
1228         bprm->argc--;
1229         ret = 0;
1230
1231 out:
1232         return ret;
1233 }
1234 EXPORT_SYMBOL(remove_arg_zero);
1235
1236 /*
1237  * cycle the list of binary formats handler, until one recognizes the image
1238  */
1239 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1240 {
1241         int try,retval;
1242         struct linux_binfmt *fmt;
1243 #ifdef __alpha__
1244         /* handle /sbin/loader.. */
1245         {
1246             struct exec * eh = (struct exec *) bprm->buf;
1247
1248             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1249                 (eh->fh.f_flags & 0x3000) == 0x3000)
1250             {
1251                 struct file * file;
1252                 unsigned long loader;
1253
1254                 allow_write_access(bprm->file);
1255                 fput(bprm->file);
1256                 bprm->file = NULL;
1257
1258                 loader = bprm->vma->vm_end - sizeof(void *);
1259
1260                 file = open_exec("/sbin/loader");
1261                 retval = PTR_ERR(file);
1262                 if (IS_ERR(file))
1263                         return retval;
1264
1265                 /* Remember if the application is TASO.  */
1266                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1267
1268                 bprm->file = file;
1269                 bprm->loader = loader;
1270                 retval = prepare_binprm(bprm);
1271                 if (retval<0)
1272                         return retval;
1273                 /* should call search_binary_handler recursively here,
1274                    but it does not matter */
1275             }
1276         }
1277 #endif
1278         retval = security_bprm_check(bprm);
1279         if (retval)
1280                 return retval;
1281
1282         /* kernel module loader fixup */
1283         /* so we don't try to load run modprobe in kernel space. */
1284         set_fs(USER_DS);
1285
1286         retval = audit_bprm(bprm);
1287         if (retval)
1288                 return retval;
1289
1290         retval = -ENOENT;
1291         for (try=0; try<2; try++) {
1292                 read_lock(&binfmt_lock);
1293                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1294                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1295                         if (!fn)
1296                                 continue;
1297                         if (!try_module_get(fmt->module))
1298                                 continue;
1299                         read_unlock(&binfmt_lock);
1300                         retval = fn(bprm, regs);
1301                         if (retval >= 0) {
1302                                 put_binfmt(fmt);
1303                                 allow_write_access(bprm->file);
1304                                 if (bprm->file)
1305                                         fput(bprm->file);
1306                                 bprm->file = NULL;
1307                                 current->did_exec = 1;
1308                                 proc_exec_connector(current);
1309                                 return retval;
1310                         }
1311                         read_lock(&binfmt_lock);
1312                         put_binfmt(fmt);
1313                         if (retval != -ENOEXEC || bprm->mm == NULL)
1314                                 break;
1315                         if (!bprm->file) {
1316                                 read_unlock(&binfmt_lock);
1317                                 return retval;
1318                         }
1319                 }
1320                 read_unlock(&binfmt_lock);
1321                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1322                         break;
1323 #ifdef CONFIG_KMOD
1324                 }else{
1325 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1326                         if (printable(bprm->buf[0]) &&
1327                             printable(bprm->buf[1]) &&
1328                             printable(bprm->buf[2]) &&
1329                             printable(bprm->buf[3]))
1330                                 break; /* -ENOEXEC */
1331                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1332 #endif
1333                 }
1334         }
1335         return retval;
1336 }
1337
1338 EXPORT_SYMBOL(search_binary_handler);
1339
1340 /*
1341  * sys_execve() executes a new program.
1342  */
1343 int do_execve(char * filename,
1344         char __user *__user *argv,
1345         char __user *__user *envp,
1346         struct pt_regs * regs)
1347 {
1348         struct linux_binprm *bprm;
1349         struct file *file;
1350         unsigned long env_p;
1351         int retval;
1352
1353         retval = -ENOMEM;
1354         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1355         if (!bprm)
1356                 goto out_ret;
1357
1358         file = open_exec(filename);
1359         retval = PTR_ERR(file);
1360         if (IS_ERR(file))
1361                 goto out_kfree;
1362
1363         sched_exec();
1364
1365         bprm->file = file;
1366         bprm->filename = filename;
1367         bprm->interp = filename;
1368
1369         retval = bprm_mm_init(bprm);
1370         if (retval)
1371                 goto out_file;
1372
1373         bprm->argc = count(argv, MAX_ARG_STRINGS);
1374         if ((retval = bprm->argc) < 0)
1375                 goto out_mm;
1376
1377         bprm->envc = count(envp, MAX_ARG_STRINGS);
1378         if ((retval = bprm->envc) < 0)
1379                 goto out_mm;
1380
1381         retval = security_bprm_alloc(bprm);
1382         if (retval)
1383                 goto out;
1384
1385         retval = prepare_binprm(bprm);
1386         if (retval < 0)
1387                 goto out;
1388
1389         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1390         if (retval < 0)
1391                 goto out;
1392
1393         bprm->exec = bprm->p;
1394         retval = copy_strings(bprm->envc, envp, bprm);
1395         if (retval < 0)
1396                 goto out;
1397
1398         env_p = bprm->p;
1399         retval = copy_strings(bprm->argc, argv, bprm);
1400         if (retval < 0)
1401                 goto out;
1402         bprm->argv_len = env_p - bprm->p;
1403
1404         retval = search_binary_handler(bprm,regs);
1405         if (retval >= 0) {
1406                 /* execve success */
1407                 free_arg_pages(bprm);
1408                 security_bprm_free(bprm);
1409                 acct_update_integrals(current);
1410                 kfree(bprm);
1411                 return retval;
1412         }
1413
1414 out:
1415         free_arg_pages(bprm);
1416         if (bprm->security)
1417                 security_bprm_free(bprm);
1418
1419 out_mm:
1420         if (bprm->mm)
1421                 mmput (bprm->mm);
1422
1423 out_file:
1424         if (bprm->file) {
1425                 allow_write_access(bprm->file);
1426                 fput(bprm->file);
1427         }
1428 out_kfree:
1429         kfree(bprm);
1430
1431 out_ret:
1432         return retval;
1433 }
1434
1435 int set_binfmt(struct linux_binfmt *new)
1436 {
1437         struct linux_binfmt *old = current->binfmt;
1438
1439         if (new) {
1440                 if (!try_module_get(new->module))
1441                         return -1;
1442         }
1443         current->binfmt = new;
1444         if (old)
1445                 module_put(old->module);
1446         return 0;
1447 }
1448
1449 EXPORT_SYMBOL(set_binfmt);
1450
1451 /* format_corename will inspect the pattern parameter, and output a
1452  * name into corename, which must have space for at least
1453  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1454  */
1455 static int format_corename(char *corename, const char *pattern, long signr)
1456 {
1457         const char *pat_ptr = pattern;
1458         char *out_ptr = corename;
1459         char *const out_end = corename + CORENAME_MAX_SIZE;
1460         int rc;
1461         int pid_in_pattern = 0;
1462         int ispipe = 0;
1463
1464         if (*pattern == '|')
1465                 ispipe = 1;
1466
1467         /* Repeat as long as we have more pattern to process and more output
1468            space */
1469         while (*pat_ptr) {
1470                 if (*pat_ptr != '%') {
1471                         if (out_ptr == out_end)
1472                                 goto out;
1473                         *out_ptr++ = *pat_ptr++;
1474                 } else {
1475                         switch (*++pat_ptr) {
1476                         case 0:
1477                                 goto out;
1478                         /* Double percent, output one percent */
1479                         case '%':
1480                                 if (out_ptr == out_end)
1481                                         goto out;
1482                                 *out_ptr++ = '%';
1483                                 break;
1484                         /* pid */
1485                         case 'p':
1486                                 pid_in_pattern = 1;
1487                                 rc = snprintf(out_ptr, out_end - out_ptr,
1488                                               "%d", current->tgid);
1489                                 if (rc > out_end - out_ptr)
1490                                         goto out;
1491                                 out_ptr += rc;
1492                                 break;
1493                         /* uid */
1494                         case 'u':
1495                                 rc = snprintf(out_ptr, out_end - out_ptr,
1496                                               "%d", current->uid);
1497                                 if (rc > out_end - out_ptr)
1498                                         goto out;
1499                                 out_ptr += rc;
1500                                 break;
1501                         /* gid */
1502                         case 'g':
1503                                 rc = snprintf(out_ptr, out_end - out_ptr,
1504                                               "%d", current->gid);
1505                                 if (rc > out_end - out_ptr)
1506                                         goto out;
1507                                 out_ptr += rc;
1508                                 break;
1509                         /* signal that caused the coredump */
1510                         case 's':
1511                                 rc = snprintf(out_ptr, out_end - out_ptr,
1512                                               "%ld", signr);
1513                                 if (rc > out_end - out_ptr)
1514                                         goto out;
1515                                 out_ptr += rc;
1516                                 break;
1517                         /* UNIX time of coredump */
1518                         case 't': {
1519                                 struct timeval tv;
1520                                 do_gettimeofday(&tv);
1521                                 rc = snprintf(out_ptr, out_end - out_ptr,
1522                                               "%lu", tv.tv_sec);
1523                                 if (rc > out_end - out_ptr)
1524                                         goto out;
1525                                 out_ptr += rc;
1526                                 break;
1527                         }
1528                         /* hostname */
1529                         case 'h':
1530                                 down_read(&uts_sem);
1531                                 rc = snprintf(out_ptr, out_end - out_ptr,
1532                                               "%s", utsname()->nodename);
1533                                 up_read(&uts_sem);
1534                                 if (rc > out_end - out_ptr)
1535                                         goto out;
1536                                 out_ptr += rc;
1537                                 break;
1538                         /* executable */
1539                         case 'e':
1540                                 rc = snprintf(out_ptr, out_end - out_ptr,
1541                                               "%s", current->comm);
1542                                 if (rc > out_end - out_ptr)
1543                                         goto out;
1544                                 out_ptr += rc;
1545                                 break;
1546                         default:
1547                                 break;
1548                         }
1549                         ++pat_ptr;
1550                 }
1551         }
1552         /* Backward compatibility with core_uses_pid:
1553          *
1554          * If core_pattern does not include a %p (as is the default)
1555          * and core_uses_pid is set, then .%pid will be appended to
1556          * the filename. Do not do this for piped commands. */
1557         if (!ispipe && !pid_in_pattern
1558             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1559                 rc = snprintf(out_ptr, out_end - out_ptr,
1560                               ".%d", current->tgid);
1561                 if (rc > out_end - out_ptr)
1562                         goto out;
1563                 out_ptr += rc;
1564         }
1565 out:
1566         *out_ptr = 0;
1567         return ispipe;
1568 }
1569
1570 static void zap_process(struct task_struct *start)
1571 {
1572         struct task_struct *t;
1573
1574         start->signal->flags = SIGNAL_GROUP_EXIT;
1575         start->signal->group_stop_count = 0;
1576
1577         t = start;
1578         do {
1579                 if (t != current && t->mm) {
1580                         t->mm->core_waiters++;
1581                         sigaddset(&t->pending.signal, SIGKILL);
1582                         signal_wake_up(t, 1);
1583                 }
1584         } while ((t = next_thread(t)) != start);
1585 }
1586
1587 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1588                                 int exit_code)
1589 {
1590         struct task_struct *g, *p;
1591         unsigned long flags;
1592         int err = -EAGAIN;
1593
1594         spin_lock_irq(&tsk->sighand->siglock);
1595         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1596                 tsk->signal->group_exit_code = exit_code;
1597                 zap_process(tsk);
1598                 err = 0;
1599         }
1600         spin_unlock_irq(&tsk->sighand->siglock);
1601         if (err)
1602                 return err;
1603
1604         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1605                 goto done;
1606
1607         rcu_read_lock();
1608         for_each_process(g) {
1609                 if (g == tsk->group_leader)
1610                         continue;
1611
1612                 p = g;
1613                 do {
1614                         if (p->mm) {
1615                                 if (p->mm == mm) {
1616                                         /*
1617                                          * p->sighand can't disappear, but
1618                                          * may be changed by de_thread()
1619                                          */
1620                                         lock_task_sighand(p, &flags);
1621                                         zap_process(p);
1622                                         unlock_task_sighand(p, &flags);
1623                                 }
1624                                 break;
1625                         }
1626                 } while ((p = next_thread(p)) != g);
1627         }
1628         rcu_read_unlock();
1629 done:
1630         return mm->core_waiters;
1631 }
1632
1633 static int coredump_wait(int exit_code)
1634 {
1635         struct task_struct *tsk = current;
1636         struct mm_struct *mm = tsk->mm;
1637         struct completion startup_done;
1638         struct completion *vfork_done;
1639         int core_waiters;
1640
1641         init_completion(&mm->core_done);
1642         init_completion(&startup_done);
1643         mm->core_startup_done = &startup_done;
1644
1645         core_waiters = zap_threads(tsk, mm, exit_code);
1646         up_write(&mm->mmap_sem);
1647
1648         if (unlikely(core_waiters < 0))
1649                 goto fail;
1650
1651         /*
1652          * Make sure nobody is waiting for us to release the VM,
1653          * otherwise we can deadlock when we wait on each other
1654          */
1655         vfork_done = tsk->vfork_done;
1656         if (vfork_done) {
1657                 tsk->vfork_done = NULL;
1658                 complete(vfork_done);
1659         }
1660
1661         if (core_waiters)
1662                 wait_for_completion(&startup_done);
1663 fail:
1664         BUG_ON(mm->core_waiters);
1665         return core_waiters;
1666 }
1667
1668 /*
1669  * set_dumpable converts traditional three-value dumpable to two flags and
1670  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1671  * these bits are not changed atomically.  So get_dumpable can observe the
1672  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1673  * return either old dumpable or new one by paying attention to the order of
1674  * modifying the bits.
1675  *
1676  * dumpable |   mm->flags (binary)
1677  * old  new | initial interim  final
1678  * ---------+-----------------------
1679  *  0    1  |   00      01      01
1680  *  0    2  |   00      10(*)   11
1681  *  1    0  |   01      00      00
1682  *  1    2  |   01      11      11
1683  *  2    0  |   11      10(*)   00
1684  *  2    1  |   11      11      01
1685  *
1686  * (*) get_dumpable regards interim value of 10 as 11.
1687  */
1688 void set_dumpable(struct mm_struct *mm, int value)
1689 {
1690         switch (value) {
1691         case 0:
1692                 clear_bit(MMF_DUMPABLE, &mm->flags);
1693                 smp_wmb();
1694                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1695                 break;
1696         case 1:
1697                 set_bit(MMF_DUMPABLE, &mm->flags);
1698                 smp_wmb();
1699                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1700                 break;
1701         case 2:
1702                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1703                 smp_wmb();
1704                 set_bit(MMF_DUMPABLE, &mm->flags);
1705                 break;
1706         }
1707 }
1708 EXPORT_SYMBOL_GPL(set_dumpable);
1709
1710 int get_dumpable(struct mm_struct *mm)
1711 {
1712         int ret;
1713
1714         ret = mm->flags & 0x3;
1715         return (ret >= 2) ? 2 : ret;
1716 }
1717
1718 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1719 {
1720         char corename[CORENAME_MAX_SIZE + 1];
1721         struct mm_struct *mm = current->mm;
1722         struct linux_binfmt * binfmt;
1723         struct inode * inode;
1724         struct file * file;
1725         int retval = 0;
1726         int fsuid = current->fsuid;
1727         int flag = 0;
1728         int ispipe = 0;
1729
1730         audit_core_dumps(signr);
1731
1732         binfmt = current->binfmt;
1733         if (!binfmt || !binfmt->core_dump)
1734                 goto fail;
1735         down_write(&mm->mmap_sem);
1736         if (!get_dumpable(mm)) {
1737                 up_write(&mm->mmap_sem);
1738                 goto fail;
1739         }
1740
1741         /*
1742          *      We cannot trust fsuid as being the "true" uid of the
1743          *      process nor do we know its entire history. We only know it
1744          *      was tainted so we dump it as root in mode 2.
1745          */
1746         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1747                 flag = O_EXCL;          /* Stop rewrite attacks */
1748                 current->fsuid = 0;     /* Dump root private */
1749         }
1750         set_dumpable(mm, 0);
1751
1752         retval = coredump_wait(exit_code);
1753         if (retval < 0)
1754                 goto fail;
1755
1756         /*
1757          * Clear any false indication of pending signals that might
1758          * be seen by the filesystem code called to write the core file.
1759          */
1760         clear_thread_flag(TIF_SIGPENDING);
1761
1762         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1763                 goto fail_unlock;
1764
1765         /*
1766          * lock_kernel() because format_corename() is controlled by sysctl, which
1767          * uses lock_kernel()
1768          */
1769         lock_kernel();
1770         ispipe = format_corename(corename, core_pattern, signr);
1771         unlock_kernel();
1772         if (ispipe) {
1773                 /* SIGPIPE can happen, but it's just never processed */
1774                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1775                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1776                                corename);
1777                         goto fail_unlock;
1778                 }
1779         } else
1780                 file = filp_open(corename,
1781                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1782                                  0600);
1783         if (IS_ERR(file))
1784                 goto fail_unlock;
1785         inode = file->f_path.dentry->d_inode;
1786         if (inode->i_nlink > 1)
1787                 goto close_fail;        /* multiple links - don't dump */
1788         if (!ispipe && d_unhashed(file->f_path.dentry))
1789                 goto close_fail;
1790
1791         /* AK: actually i see no reason to not allow this for named pipes etc.,
1792            but keep the previous behaviour for now. */
1793         if (!ispipe && !S_ISREG(inode->i_mode))
1794                 goto close_fail;
1795         if (!file->f_op)
1796                 goto close_fail;
1797         if (!file->f_op->write)
1798                 goto close_fail;
1799         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1800                 goto close_fail;
1801
1802         retval = binfmt->core_dump(signr, regs, file);
1803
1804         if (retval)
1805                 current->signal->group_exit_code |= 0x80;
1806 close_fail:
1807         filp_close(file, NULL);
1808 fail_unlock:
1809         current->fsuid = fsuid;
1810         complete_all(&mm->core_done);
1811 fail:
1812         return retval;
1813 }