git.oblomov.eu Git - linux-2.6/blob - 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/string.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/key.h>
  38 #include <linux/personality.h>
  39 #include <linux/binfmts.h>
  40 #include <linux/swap.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/ptrace.h>
  47 #include <linux/mount.h>
  48 #include <linux/security.h>
  49 #include <linux/syscalls.h>
  50 #include <linux/rmap.h>
  51 #include <linux/tsacct_kern.h>
  52 #include <linux/cn_proc.h>
  53 #include <linux/audit.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 /* The maximal length of core_pattern is also specified in sysctl.c */
  68
  69 static LIST_HEAD(formats);
  70 static DEFINE_RWLOCK(binfmt_lock);
  71
  72 int register_binfmt(struct linux_binfmt * fmt)
  73 {
  74         if (!fmt)
  75                 return -EINVAL;
  76         write_lock(&binfmt_lock);
  77         list_add(&fmt->lh, &formats);
  78         write_unlock(&binfmt_lock);
  79         return 0;
  80 }
  81
  82 EXPORT_SYMBOL(register_binfmt);
  83
  84 void unregister_binfmt(struct linux_binfmt * fmt)
  85 {
  86         write_lock(&binfmt_lock);
  87         list_del(&fmt->lh);
  88         write_unlock(&binfmt_lock);
  89 }
  90
  91 EXPORT_SYMBOL(unregister_binfmt);
  92
  93 static inline void put_binfmt(struct linux_binfmt * fmt)
  94 {
  95         module_put(fmt->module);
  96 }
  97
  98 /*
  99  * Note that a shared library must be both readable and executable due to
 100  * security reasons.
 101  *
 102  * Also note that we take the address to load from from the file itself.
 103  */
 104 asmlinkage long sys_uselib(const char __user * library)
 105 {
 106         struct file * file;
 107         struct nameidata nd;
 108         int error;
 109
 110         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 111         if (error)
 112                 goto out;
 113
 114         error = -EINVAL;
 115         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 116                 goto exit;
 117
 118         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 119         if (error)
 120                 goto exit;
 121
 122         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 123         error = PTR_ERR(file);
 124         if (IS_ERR(file))
 125                 goto out;
 126
 127         error = -ENOEXEC;
 128         if(file->f_op) {
 129                 struct linux_binfmt * fmt;
 130
 131                 read_lock(&binfmt_lock);
 132                 list_for_each_entry(fmt, &formats, lh) {
 133                         if (!fmt->load_shlib)
 134                                 continue;
 135                         if (!try_module_get(fmt->module))
 136                                 continue;
 137                         read_unlock(&binfmt_lock);
 138                         error = fmt->load_shlib(file);
 139                         read_lock(&binfmt_lock);
 140                         put_binfmt(fmt);
 141                         if (error != -ENOEXEC)
 142                                 break;
 143                 }
 144                 read_unlock(&binfmt_lock);
 145         }
 146         fput(file);
 147 out:
 148         return error;
 149 exit:
 150         release_open_intent(&nd);
 151         path_put(&nd.path);
 152         goto out;
 153 }
 154
 155 #ifdef CONFIG_MMU
 156
 157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 158                 int write)
 159 {
 160         struct page *page;
 161         int ret;
 162
 163 #ifdef CONFIG_STACK_GROWSUP
 164         if (write) {
 165                 ret = expand_stack_downwards(bprm->vma, pos);
 166                 if (ret < 0)
 167                         return NULL;
 168         }
 169 #endif
 170         ret = get_user_pages(current, bprm->mm, pos,
 171                         1, write, 1, &page, NULL);
 172         if (ret <= 0)
 173                 return NULL;
 174
 175         if (write) {
 176                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 177                 struct rlimit *rlim;
 178
 179                 /*
 180                  * We've historically supported up to 32 pages (ARG_MAX)
 181                  * of argument strings even with small stacks
 182                  */
 183                 if (size <= ARG_MAX)
 184                         return page;
 185
 186                 /*
 187                  * Limit to 1/4-th the stack size for the argv+env strings.
 188                  * This ensures that:
 189                  *  - the remaining binfmt code will not run out of stack space,
 190                  *  - the program will have a reasonable amount of stack left
 191                  *    to work from.
 192                  */
 193                 rlim = current->signal->rlim;
 194                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 195                         put_page(page);
 196                         return NULL;
 197                 }
 198         }
 199
 200         return page;
 201 }
 202
 203 static void put_arg_page(struct page *page)
 204 {
 205         put_page(page);
 206 }
 207
 208 static void free_arg_page(struct linux_binprm *bprm, int i)
 209 {
 210 }
 211
 212 static void free_arg_pages(struct linux_binprm *bprm)
 213 {
 214 }
 215
 216 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 217                 struct page *page)
 218 {
 219         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 220 }
 221
 222 static int __bprm_mm_init(struct linux_binprm *bprm)
 223 {
 224         int err = -ENOMEM;
 225         struct vm_area_struct *vma = NULL;
 226         struct mm_struct *mm = bprm->mm;
 227
 228         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 229         if (!vma)
 230                 goto err;
 231
 232         down_write(&mm->mmap_sem);
 233         vma->vm_mm = mm;
 234
 235         /*
 236          * Place the stack at the largest stack address the architecture
 237          * supports. Later, we'll move this to an appropriate place. We don't
 238          * use STACK_TOP because that can depend on attributes which aren't
 239          * configured yet.
 240          */
 241         vma->vm_end = STACK_TOP_MAX;
 242         vma->vm_start = vma->vm_end - PAGE_SIZE;
 243
 244         vma->vm_flags = VM_STACK_FLAGS;
 245         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 246         err = insert_vm_struct(mm, vma);
 247         if (err) {
 248                 up_write(&mm->mmap_sem);
 249                 goto err;
 250         }
 251
 252         mm->stack_vm = mm->total_vm = 1;
 253         up_write(&mm->mmap_sem);
 254
 255         bprm->p = vma->vm_end - sizeof(void *);
 256
 257         return 0;
 258
 259 err:
 260         if (vma) {
 261                 bprm->vma = NULL;
 262                 kmem_cache_free(vm_area_cachep, vma);
 263         }
 264
 265         return err;
 266 }
 267
 268 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 269 {
 270         return len <= MAX_ARG_STRLEN;
 271 }
 272
 273 #else
 274
 275 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 276                 int write)
 277 {
 278         struct page *page;
 279
 280         page = bprm->page[pos / PAGE_SIZE];
 281         if (!page && write) {
 282                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 283                 if (!page)
 284                         return NULL;
 285                 bprm->page[pos / PAGE_SIZE] = page;
 286         }
 287
 288         return page;
 289 }
 290
 291 static void put_arg_page(struct page *page)
 292 {
 293 }
 294
 295 static void free_arg_page(struct linux_binprm *bprm, int i)
 296 {
 297         if (bprm->page[i]) {
 298                 __free_page(bprm->page[i]);
 299                 bprm->page[i] = NULL;
 300         }
 301 }
 302
 303 static void free_arg_pages(struct linux_binprm *bprm)
 304 {
 305         int i;
 306
 307         for (i = 0; i < MAX_ARG_PAGES; i++)
 308                 free_arg_page(bprm, i);
 309 }
 310
 311 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 312                 struct page *page)
 313 {
 314 }
 315
 316 static int __bprm_mm_init(struct linux_binprm *bprm)
 317 {
 318         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 319         return 0;
 320 }
 321
 322 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 323 {
 324         return len <= bprm->p;
 325 }
 326
 327 #endif /* CONFIG_MMU */
 328
 329 /*
 330  * Create a new mm_struct and populate it with a temporary stack
 331  * vm_area_struct.  We don't have enough context at this point to set the stack
 332  * flags, permissions, and offset, so we use temporary values.  We'll update
 333  * them later in setup_arg_pages().
 334  */
 335 int bprm_mm_init(struct linux_binprm *bprm)
 336 {
 337         int err;
 338         struct mm_struct *mm = NULL;
 339
 340         bprm->mm = mm = mm_alloc();
 341         err = -ENOMEM;
 342         if (!mm)
 343                 goto err;
 344
 345         err = init_new_context(current, mm);
 346         if (err)
 347                 goto err;
 348
 349         err = __bprm_mm_init(bprm);
 350         if (err)
 351                 goto err;
 352
 353         return 0;
 354
 355 err:
 356         if (mm) {
 357                 bprm->mm = NULL;
 358                 mmdrop(mm);
 359         }
 360
 361         return err;
 362 }
 363
 364 /*
 365  * count() counts the number of strings in array ARGV.
 366  */
 367 static int count(char __user * __user * argv, int max)
 368 {
 369         int i = 0;
 370
 371         if (argv != NULL) {
 372                 for (;;) {
 373                         char __user * p;
 374
 375                         if (get_user(p, argv))
 376                                 return -EFAULT;
 377                         if (!p)
 378                                 break;
 379                         argv++;
 380                         if(++i > max)
 381                                 return -E2BIG;
 382                         cond_resched();
 383                 }
 384         }
 385         return i;
 386 }
 387
 388 /*
 389  * 'copy_strings()' copies argument/environment strings from the old
 390  * processes's memory to the new process's stack.  The call to get_user_pages()
 391  * ensures the destination page is created and not swapped out.
 392  */
 393 static int copy_strings(int argc, char __user * __user * argv,
 394                         struct linux_binprm *bprm)
 395 {
 396         struct page *kmapped_page = NULL;
 397         char *kaddr = NULL;
 398         unsigned long kpos = 0;
 399         int ret;
 400
 401         while (argc-- > 0) {
 402                 char __user *str;
 403                 int len;
 404                 unsigned long pos;
 405
 406                 if (get_user(str, argv+argc) ||
 407                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 408                         ret = -EFAULT;
 409                         goto out;
 410                 }
 411
 412                 if (!valid_arg_len(bprm, len)) {
 413                         ret = -E2BIG;
 414                         goto out;
 415                 }
 416
 417                 /* We're going to work our way backwords. */
 418                 pos = bprm->p;
 419                 str += len;
 420                 bprm->p -= len;
 421
 422                 while (len > 0) {
 423                         int offset, bytes_to_copy;
 424
 425                         offset = pos % PAGE_SIZE;
 426                         if (offset == 0)
 427                                 offset = PAGE_SIZE;
 428
 429                         bytes_to_copy = offset;
 430                         if (bytes_to_copy > len)
 431                                 bytes_to_copy = len;
 432
 433                         offset -= bytes_to_copy;
 434                         pos -= bytes_to_copy;
 435                         str -= bytes_to_copy;
 436                         len -= bytes_to_copy;
 437
 438                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 439                                 struct page *page;
 440
 441                                 page = get_arg_page(bprm, pos, 1);
 442                                 if (!page) {
 443                                         ret = -E2BIG;
 444                                         goto out;
 445                                 }
 446
 447                                 if (kmapped_page) {
 448                                         flush_kernel_dcache_page(kmapped_page);
 449                                         kunmap(kmapped_page);
 450                                         put_arg_page(kmapped_page);
 451                                 }
 452                                 kmapped_page = page;
 453                                 kaddr = kmap(kmapped_page);
 454                                 kpos = pos & PAGE_MASK;
 455                                 flush_arg_page(bprm, kpos, kmapped_page);
 456                         }
 457                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 458                                 ret = -EFAULT;
 459                                 goto out;
 460                         }
 461                 }
 462         }
 463         ret = 0;
 464 out:
 465         if (kmapped_page) {
 466                 flush_kernel_dcache_page(kmapped_page);
 467                 kunmap(kmapped_page);
 468                 put_arg_page(kmapped_page);
 469         }
 470         return ret;
 471 }
 472
 473 /*
 474  * Like copy_strings, but get argv and its values from kernel memory.
 475  */
 476 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 477 {
 478         int r;
 479         mm_segment_t oldfs = get_fs();
 480         set_fs(KERNEL_DS);
 481         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 482         set_fs(oldfs);
 483         return r;
 484 }
 485 EXPORT_SYMBOL(copy_strings_kernel);
 486
 487 #ifdef CONFIG_MMU
 488
 489 /*
 490  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 491  * the binfmt code determines where the new stack should reside, we shift it to
 492  * its final location.  The process proceeds as follows:
 493  *
 494  * 1) Use shift to calculate the new vma endpoints.
 495  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 496  *    arguments passed to subsequent functions are consistent.
 497  * 3) Move vma's page tables to the new range.
 498  * 4) Free up any cleared pgd range.
 499  * 5) Shrink the vma to cover only the new range.
 500  */
 501 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 502 {
 503         struct mm_struct *mm = vma->vm_mm;
 504         unsigned long old_start = vma->vm_start;
 505         unsigned long old_end = vma->vm_end;
 506         unsigned long length = old_end - old_start;
 507         unsigned long new_start = old_start - shift;
 508         unsigned long new_end = old_end - shift;
 509         struct mmu_gather *tlb;
 510
 511         BUG_ON(new_start > new_end);
 512
 513         /*
 514          * ensure there are no vmas between where we want to go
 515          * and where we are
 516          */
 517         if (vma != find_vma(mm, new_start))
 518                 return -EFAULT;
 519
 520         /*
 521          * cover the whole range: [new_start, old_end)
 522          */
 523         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 524
 525         /*
 526          * move the page tables downwards, on failure we rely on
 527          * process cleanup to remove whatever mess we made.
 528          */
 529         if (length != move_page_tables(vma, old_start,
 530                                        vma, new_start, length))
 531                 return -ENOMEM;
 532
 533         lru_add_drain();
 534         tlb = tlb_gather_mmu(mm, 0);
 535         if (new_end > old_start) {
 536                 /*
 537                  * when the old and new regions overlap clear from new_end.
 538                  */
 539                 free_pgd_range(&tlb, new_end, old_end, new_end,
 540                         vma->vm_next ? vma->vm_next->vm_start : 0);
 541         } else {
 542                 /*
 543                  * otherwise, clean from old_start; this is done to not touch
 544                  * the address space in [new_end, old_start) some architectures
 545                  * have constraints on va-space that make this illegal (IA64) -
 546                  * for the others its just a little faster.
 547                  */
 548                 free_pgd_range(&tlb, old_start, old_end, new_end,
 549                         vma->vm_next ? vma->vm_next->vm_start : 0);
 550         }
 551         tlb_finish_mmu(tlb, new_end, old_end);
 552
 553         /*
 554          * shrink the vma to just the new range.
 555          */
 556         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 557
 558         return 0;
 559 }
 560
 561 #define EXTRA_STACK_VM_PAGES    20      /* random */
 562
 563 /*
 564  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 565  * the stack is optionally relocated, and some extra space is added.
 566  */
 567 int setup_arg_pages(struct linux_binprm *bprm,
 568                     unsigned long stack_top,
 569                     int executable_stack)
 570 {
 571         unsigned long ret;
 572         unsigned long stack_shift;
 573         struct mm_struct *mm = current->mm;
 574         struct vm_area_struct *vma = bprm->vma;
 575         struct vm_area_struct *prev = NULL;
 576         unsigned long vm_flags;
 577         unsigned long stack_base;
 578
 579 #ifdef CONFIG_STACK_GROWSUP
 580         /* Limit stack size to 1GB */
 581         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 582         if (stack_base > (1 << 30))
 583                 stack_base = 1 << 30;
 584
 585         /* Make sure we didn't let the argument array grow too large. */
 586         if (vma->vm_end - vma->vm_start > stack_base)
 587                 return -ENOMEM;
 588
 589         stack_base = PAGE_ALIGN(stack_top - stack_base);
 590
 591         stack_shift = vma->vm_start - stack_base;
 592         mm->arg_start = bprm->p - stack_shift;
 593         bprm->p = vma->vm_end - stack_shift;
 594 #else
 595         stack_top = arch_align_stack(stack_top);
 596         stack_top = PAGE_ALIGN(stack_top);
 597         stack_shift = vma->vm_end - stack_top;
 598
 599         bprm->p -= stack_shift;
 600         mm->arg_start = bprm->p;
 601 #endif
 602
 603         if (bprm->loader)
 604                 bprm->loader -= stack_shift;
 605         bprm->exec -= stack_shift;
 606
 607         down_write(&mm->mmap_sem);
 608         vm_flags = vma->vm_flags;
 609
 610         /*
 611          * Adjust stack execute permissions; explicitly enable for
 612          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 613          * (arch default) otherwise.
 614          */
 615         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 616                 vm_flags |= VM_EXEC;
 617         else if (executable_stack == EXSTACK_DISABLE_X)
 618                 vm_flags &= ~VM_EXEC;
 619         vm_flags |= mm->def_flags;
 620
 621         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 622                         vm_flags);
 623         if (ret)
 624                 goto out_unlock;
 625         BUG_ON(prev != vma);
 626
 627         /* Move stack pages down in memory. */
 628         if (stack_shift) {
 629                 ret = shift_arg_pages(vma, stack_shift);
 630                 if (ret) {
 631                         up_write(&mm->mmap_sem);
 632                         return ret;
 633                 }
 634         }
 635
 636 #ifdef CONFIG_STACK_GROWSUP
 637         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 638 #else
 639         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 640 #endif
 641         ret = expand_stack(vma, stack_base);
 642         if (ret)
 643                 ret = -EFAULT;
 644
 645 out_unlock:
 646         up_write(&mm->mmap_sem);
 647         return 0;
 648 }
 649 EXPORT_SYMBOL(setup_arg_pages);
 650
 651 #endif /* CONFIG_MMU */
 652
 653 struct file *open_exec(const char *name)
 654 {
 655         struct nameidata nd;
 656         int err;
 657         struct file *file;
 658
 659         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 660         file = ERR_PTR(err);
 661
 662         if (!err) {
 663                 struct inode *inode = nd.path.dentry->d_inode;
 664                 file = ERR_PTR(-EACCES);
 665                 if (S_ISREG(inode->i_mode)) {
 666                         int err = vfs_permission(&nd, MAY_EXEC);
 667                         file = ERR_PTR(err);
 668                         if (!err) {
 669                                 file = nameidata_to_filp(&nd,
 670                                                         O_RDONLY|O_LARGEFILE);
 671                                 if (!IS_ERR(file)) {
 672                                         err = deny_write_access(file);
 673                                         if (err) {
 674                                                 fput(file);
 675                                                 file = ERR_PTR(err);
 676                                         }
 677                                 }
 678 out:
 679                                 return file;
 680                         }
 681                 }
 682                 release_open_intent(&nd);
 683                 path_put(&nd.path);
 684         }
 685         goto out;
 686 }
 687
 688 EXPORT_SYMBOL(open_exec);
 689
 690 int kernel_read(struct file *file, unsigned long offset,
 691         char *addr, unsigned long count)
 692 {
 693         mm_segment_t old_fs;
 694         loff_t pos = offset;
 695         int result;
 696
 697         old_fs = get_fs();
 698         set_fs(get_ds());
 699         /* The cast to a user pointer is valid due to the set_fs() */
 700         result = vfs_read(file, (void __user *)addr, count, &pos);
 701         set_fs(old_fs);
 702         return result;
 703 }
 704
 705 EXPORT_SYMBOL(kernel_read);
 706
 707 static int exec_mmap(struct mm_struct *mm)
 708 {
 709         struct task_struct *tsk;
 710         struct mm_struct * old_mm, *active_mm;
 711
 712         /* Notify parent that we're no longer interested in the old VM */
 713         tsk = current;
 714         old_mm = current->mm;
 715         mm_release(tsk, old_mm);
 716
 717         if (old_mm) {
 718                 /*
 719                  * Make sure that if there is a core dump in progress
 720                  * for the old mm, we get out and die instead of going
 721                  * through with the exec.  We must hold mmap_sem around
 722                  * checking core_waiters and changing tsk->mm.  The
 723                  * core-inducing thread will increment core_waiters for
 724                  * each thread whose ->mm == old_mm.
 725                  */
 726                 down_read(&old_mm->mmap_sem);
 727                 if (unlikely(old_mm->core_waiters)) {
 728                         up_read(&old_mm->mmap_sem);
 729                         return -EINTR;
 730                 }
 731         }
 732         task_lock(tsk);
 733         active_mm = tsk->active_mm;
 734         tsk->mm = mm;
 735         tsk->active_mm = mm;
 736         activate_mm(active_mm, mm);
 737         task_unlock(tsk);
 738         arch_pick_mmap_layout(mm);
 739         if (old_mm) {
 740                 up_read(&old_mm->mmap_sem);
 741                 BUG_ON(active_mm != old_mm);
 742                 mmput(old_mm);
 743                 return 0;
 744         }
 745         mmdrop(active_mm);
 746         return 0;
 747 }
 748
 749 /*
 750  * This function makes sure the current process has its own signal table,
 751  * so that flush_signal_handlers can later reset the handlers without
 752  * disturbing other processes.  (Other processes might share the signal
 753  * table via the CLONE_SIGHAND option to clone().)
 754  */
 755 static int de_thread(struct task_struct *tsk)
 756 {
 757         struct signal_struct *sig = tsk->signal;
 758         struct sighand_struct *oldsighand = tsk->sighand;
 759         spinlock_t *lock = &oldsighand->siglock;
 760         struct task_struct *leader = NULL;
 761         int count;
 762
 763         if (thread_group_empty(tsk))
 764                 goto no_thread_group;
 765
 766         /*
 767          * Kill all other threads in the thread group.
 768          * We must hold tasklist_lock to call zap_other_threads.
 769          */
 770         read_lock(&tasklist_lock);
 771         spin_lock_irq(lock);
 772         if (signal_group_exit(sig)) {
 773                 /*
 774                  * Another group action in progress, just
 775                  * return so that the signal is processed.
 776                  */
 777                 spin_unlock_irq(lock);
 778                 read_unlock(&tasklist_lock);
 779                 return -EAGAIN;
 780         }
 781
 782         /*
 783          * child_reaper ignores SIGKILL, change it now.
 784          * Reparenting needs write_lock on tasklist_lock,
 785          * so it is safe to do it under read_lock.
 786          */
 787         if (unlikely(tsk->group_leader == task_child_reaper(tsk)))
 788                 task_active_pid_ns(tsk)->child_reaper = tsk;
 789
 790         sig->group_exit_task = tsk;
 791         zap_other_threads(tsk);
 792         read_unlock(&tasklist_lock);
 793
 794         /* Account for the thread group leader hanging around: */
 795         count = thread_group_leader(tsk) ? 1 : 2;
 796         sig->notify_count = count;
 797         while (atomic_read(&sig->count) > count) {
 798                 __set_current_state(TASK_UNINTERRUPTIBLE);
 799                 spin_unlock_irq(lock);
 800                 schedule();
 801                 spin_lock_irq(lock);
 802         }
 803         spin_unlock_irq(lock);
 804
 805         /*
 806          * At this point all other threads have exited, all we have to
 807          * do is to wait for the thread group leader to become inactive,
 808          * and to assume its PID:
 809          */
 810         if (!thread_group_leader(tsk)) {
 811                 leader = tsk->group_leader;
 812
 813                 sig->notify_count = -1;
 814                 for (;;) {
 815                         write_lock_irq(&tasklist_lock);
 816                         if (likely(leader->exit_state))
 817                                 break;
 818                         __set_current_state(TASK_UNINTERRUPTIBLE);
 819                         write_unlock_irq(&tasklist_lock);
 820                         schedule();
 821                 }
 822
 823                 /*
 824                  * The only record we have of the real-time age of a
 825                  * process, regardless of execs it's done, is start_time.
 826                  * All the past CPU time is accumulated in signal_struct
 827                  * from sister threads now dead.  But in this non-leader
 828                  * exec, nothing survives from the original leader thread,
 829                  * whose birth marks the true age of this process now.
 830                  * When we take on its identity by switching to its PID, we
 831                  * also take its birthdate (always earlier than our own).
 832                  */
 833                 tsk->start_time = leader->start_time;
 834
 835                 BUG_ON(!same_thread_group(leader, tsk));
 836                 BUG_ON(has_group_leader_pid(tsk));
 837                 /*
 838                  * An exec() starts a new thread group with the
 839                  * TGID of the previous thread group. Rehash the
 840                  * two threads with a switched PID, and release
 841                  * the former thread group leader:
 842                  */
 843
 844                 /* Become a process group leader with the old leader's pid.
 845                  * The old leader becomes a thread of the this thread group.
 846                  * Note: The old leader also uses this pid until release_task
 847                  *       is called.  Odd but simple and correct.
 848                  */
 849                 detach_pid(tsk, PIDTYPE_PID);
 850                 tsk->pid = leader->pid;
 851                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 852                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 853                 transfer_pid(leader, tsk, PIDTYPE_SID);
 854                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 855
 856                 tsk->group_leader = tsk;
 857                 leader->group_leader = tsk;
 858
 859                 tsk->exit_signal = SIGCHLD;
 860
 861                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 862                 leader->exit_state = EXIT_DEAD;
 863
 864                 write_unlock_irq(&tasklist_lock);
 865         }
 866
 867         sig->group_exit_task = NULL;
 868         sig->notify_count = 0;
 869
 870 no_thread_group:
 871         exit_itimers(sig);
 872         if (leader)
 873                 release_task(leader);
 874
 875         if (atomic_read(&oldsighand->count) != 1) {
 876                 struct sighand_struct *newsighand;
 877                 /*
 878                  * This ->sighand is shared with the CLONE_SIGHAND
 879                  * but not CLONE_THREAD task, switch to the new one.
 880                  */
 881                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 882                 if (!newsighand)
 883                         return -ENOMEM;
 884
 885                 atomic_set(&newsighand->count, 1);
 886                 memcpy(newsighand->action, oldsighand->action,
 887                        sizeof(newsighand->action));
 888
 889                 write_lock_irq(&tasklist_lock);
 890                 spin_lock(&oldsighand->siglock);
 891                 rcu_assign_pointer(tsk->sighand, newsighand);
 892                 spin_unlock(&oldsighand->siglock);
 893                 write_unlock_irq(&tasklist_lock);
 894
 895                 __cleanup_sighand(oldsighand);
 896         }
 897
 898         BUG_ON(!thread_group_leader(tsk));
 899         return 0;
 900 }
 901
 902 /*
 903  * These functions flushes out all traces of the currently running executable
 904  * so that a new one can be started
 905  */
 906 static void flush_old_files(struct files_struct * files)
 907 {
 908         long j = -1;
 909         struct fdtable *fdt;
 910
 911         spin_lock(&files->file_lock);
 912         for (;;) {
 913                 unsigned long set, i;
 914
 915                 j++;
 916                 i = j * __NFDBITS;
 917                 fdt = files_fdtable(files);
 918                 if (i >= fdt->max_fds)
 919                         break;
 920                 set = fdt->close_on_exec->fds_bits[j];
 921                 if (!set)
 922                         continue;
 923                 fdt->close_on_exec->fds_bits[j] = 0;
 924                 spin_unlock(&files->file_lock);
 925                 for ( ; set ; i++,set >>= 1) {
 926                         if (set & 1) {
 927                                 sys_close(i);
 928                         }
 929                 }
 930                 spin_lock(&files->file_lock);
 931
 932         }
 933         spin_unlock(&files->file_lock);
 934 }
 935
 936 char *get_task_comm(char *buf, struct task_struct *tsk)
 937 {
 938         /* buf must be at least sizeof(tsk->comm) in size */
 939         task_lock(tsk);
 940         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 941         task_unlock(tsk);
 942         return buf;
 943 }
 944
 945 void set_task_comm(struct task_struct *tsk, char *buf)
 946 {
 947         task_lock(tsk);
 948         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 949         task_unlock(tsk);
 950 }
 951
 952 int flush_old_exec(struct linux_binprm * bprm)
 953 {
 954         char * name;
 955         int i, ch, retval;
 956         char tcomm[sizeof(current->comm)];
 957
 958         /*
 959          * Make sure we have a private signal table and that
 960          * we are unassociated from the previous thread group.
 961          */
 962         retval = de_thread(current);
 963         if (retval)
 964                 goto out;
 965
 966         /*
 967          * Release all of the old mmap stuff
 968          */
 969         retval = exec_mmap(bprm->mm);
 970         if (retval)
 971                 goto out;
 972
 973         bprm->mm = NULL;                /* We're using it now */
 974
 975         /* This is the point of no return */
 976         current->sas_ss_sp = current->sas_ss_size = 0;
 977
 978         if (current->euid == current->uid && current->egid == current->gid)
 979                 set_dumpable(current->mm, 1);
 980         else
 981                 set_dumpable(current->mm, suid_dumpable);
 982
 983         name = bprm->filename;
 984
 985         /* Copies the binary name from after last slash */
 986         for (i=0; (ch = *(name++)) != '\0';) {
 987                 if (ch == '/')
 988                         i = 0; /* overwrite what we wrote */
 989                 else
 990                         if (i < (sizeof(tcomm) - 1))
 991                                 tcomm[i++] = ch;
 992         }
 993         tcomm[i] = '\0';
 994         set_task_comm(current, tcomm);
 995
 996         current->flags &= ~PF_RANDOMIZE;
 997         flush_thread();
 998
 999         /* Set the new mm task size. We have to do that late because it may
1000          * depend on TIF_32BIT which is only updated in flush_thread() on
1001          * some architectures like powerpc
1002          */
1003         current->mm->task_size = TASK_SIZE;
1004
1005         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1006                 suid_keys(current);
1007                 set_dumpable(current->mm, suid_dumpable);
1008                 current->pdeath_signal = 0;
1009         } else if (file_permission(bprm->file, MAY_READ) ||
1010                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1011                 suid_keys(current);
1012                 set_dumpable(current->mm, suid_dumpable);
1013         }
1014
1015         /* An exec changes our domain. We are no longer part of the thread
1016            group */
1017
1018         current->self_exec_id++;
1019
1020         flush_signal_handlers(current, 0);
1021         flush_old_files(current->files);
1022
1023         return 0;
1024
1025 out:
1026         return retval;
1027 }
1028
1029 EXPORT_SYMBOL(flush_old_exec);
1030
1031 /*
1032  * Fill the binprm structure from the inode.
1033  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1034  */
1035 int prepare_binprm(struct linux_binprm *bprm)
1036 {
1037         int mode;
1038         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1039         int retval;
1040
1041         mode = inode->i_mode;
1042         if (bprm->file->f_op == NULL)
1043                 return -EACCES;
1044
1045         bprm->e_uid = current->euid;
1046         bprm->e_gid = current->egid;
1047
1048         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1049                 /* Set-uid? */
1050                 if (mode & S_ISUID) {
1051                         current->personality &= ~PER_CLEAR_ON_SETID;
1052                         bprm->e_uid = inode->i_uid;
1053                 }
1054
1055                 /* Set-gid? */
1056                 /*
1057                  * If setgid is set but no group execute bit then this
1058                  * is a candidate for mandatory locking, not a setgid
1059                  * executable.
1060                  */
1061                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1062                         current->personality &= ~PER_CLEAR_ON_SETID;
1063                         bprm->e_gid = inode->i_gid;
1064                 }
1065         }
1066
1067         /* fill in binprm security blob */
1068         retval = security_bprm_set(bprm);
1069         if (retval)
1070                 return retval;
1071
1072         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1073         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1074 }
1075
1076 EXPORT_SYMBOL(prepare_binprm);
1077
1078 static int unsafe_exec(struct task_struct *p)
1079 {
1080         int unsafe = 0;
1081         if (p->ptrace & PT_PTRACED) {
1082                 if (p->ptrace & PT_PTRACE_CAP)
1083                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1084                 else
1085                         unsafe |= LSM_UNSAFE_PTRACE;
1086         }
1087         if (atomic_read(&p->fs->count) > 1 ||
1088             atomic_read(&p->files->count) > 1 ||
1089             atomic_read(&p->sighand->count) > 1)
1090                 unsafe |= LSM_UNSAFE_SHARE;
1091
1092         return unsafe;
1093 }
1094
1095 void compute_creds(struct linux_binprm *bprm)
1096 {
1097         int unsafe;
1098
1099         if (bprm->e_uid != current->uid) {
1100                 suid_keys(current);
1101                 current->pdeath_signal = 0;
1102         }
1103         exec_keys(current);
1104
1105         task_lock(current);
1106         unsafe = unsafe_exec(current);
1107         security_bprm_apply_creds(bprm, unsafe);
1108         task_unlock(current);
1109         security_bprm_post_apply_creds(bprm);
1110 }
1111 EXPORT_SYMBOL(compute_creds);
1112
1113 /*
1114  * Arguments are '\0' separated strings found at the location bprm->p
1115  * points to; chop off the first by relocating brpm->p to right after
1116  * the first '\0' encountered.
1117  */
1118 int remove_arg_zero(struct linux_binprm *bprm)
1119 {
1120         int ret = 0;
1121         unsigned long offset;
1122         char *kaddr;
1123         struct page *page;
1124
1125         if (!bprm->argc)
1126                 return 0;
1127
1128         do {
1129                 offset = bprm->p & ~PAGE_MASK;
1130                 page = get_arg_page(bprm, bprm->p, 0);
1131                 if (!page) {
1132                         ret = -EFAULT;
1133                         goto out;
1134                 }
1135                 kaddr = kmap_atomic(page, KM_USER0);
1136
1137                 for (; offset < PAGE_SIZE && kaddr[offset];
1138                                 offset++, bprm->p++)
1139                         ;
1140
1141                 kunmap_atomic(kaddr, KM_USER0);
1142                 put_arg_page(page);
1143
1144                 if (offset == PAGE_SIZE)
1145                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1146         } while (offset == PAGE_SIZE);
1147
1148         bprm->p++;
1149         bprm->argc--;
1150         ret = 0;
1151
1152 out:
1153         return ret;
1154 }
1155 EXPORT_SYMBOL(remove_arg_zero);
1156
1157 /*
1158  * cycle the list of binary formats handler, until one recognizes the image
1159  */
1160 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1161 {
1162         int try,retval;
1163         struct linux_binfmt *fmt;
1164 #if defined(__alpha__) && defined(CONFIG_ARCH_SUPPORTS_AOUT)
1165         /* handle /sbin/loader.. */
1166         {
1167             struct exec * eh = (struct exec *) bprm->buf;
1168
1169             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1170                 (eh->fh.f_flags & 0x3000) == 0x3000)
1171             {
1172                 struct file * file;
1173                 unsigned long loader;
1174
1175                 allow_write_access(bprm->file);
1176                 fput(bprm->file);
1177                 bprm->file = NULL;
1178
1179                 loader = bprm->vma->vm_end - sizeof(void *);
1180
1181                 file = open_exec("/sbin/loader");
1182                 retval = PTR_ERR(file);
1183                 if (IS_ERR(file))
1184                         return retval;
1185
1186                 /* Remember if the application is TASO.  */
1187                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1188
1189                 bprm->file = file;
1190                 bprm->loader = loader;
1191                 retval = prepare_binprm(bprm);
1192                 if (retval<0)
1193                         return retval;
1194                 /* should call search_binary_handler recursively here,
1195                    but it does not matter */
1196             }
1197         }
1198 #endif
1199         retval = security_bprm_check(bprm);
1200         if (retval)
1201                 return retval;
1202
1203         /* kernel module loader fixup */
1204         /* so we don't try to load run modprobe in kernel space. */
1205         set_fs(USER_DS);
1206
1207         retval = audit_bprm(bprm);
1208         if (retval)
1209                 return retval;
1210
1211         retval = -ENOENT;
1212         for (try=0; try<2; try++) {
1213                 read_lock(&binfmt_lock);
1214                 list_for_each_entry(fmt, &formats, lh) {
1215                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1216                         if (!fn)
1217                                 continue;
1218                         if (!try_module_get(fmt->module))
1219                                 continue;
1220                         read_unlock(&binfmt_lock);
1221                         retval = fn(bprm, regs);
1222                         if (retval >= 0) {
1223                                 put_binfmt(fmt);
1224                                 allow_write_access(bprm->file);
1225                                 if (bprm->file)
1226                                         fput(bprm->file);
1227                                 bprm->file = NULL;
1228                                 current->did_exec = 1;
1229                                 proc_exec_connector(current);
1230                                 return retval;
1231                         }
1232                         read_lock(&binfmt_lock);
1233                         put_binfmt(fmt);
1234                         if (retval != -ENOEXEC || bprm->mm == NULL)
1235                                 break;
1236                         if (!bprm->file) {
1237                                 read_unlock(&binfmt_lock);
1238                                 return retval;
1239                         }
1240                 }
1241                 read_unlock(&binfmt_lock);
1242                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1243                         break;
1244 #ifdef CONFIG_KMOD
1245                 }else{
1246 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1247                         if (printable(bprm->buf[0]) &&
1248                             printable(bprm->buf[1]) &&
1249                             printable(bprm->buf[2]) &&
1250                             printable(bprm->buf[3]))
1251                                 break; /* -ENOEXEC */
1252                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1253 #endif
1254                 }
1255         }
1256         return retval;
1257 }
1258
1259 EXPORT_SYMBOL(search_binary_handler);
1260
1261 /*
1262  * sys_execve() executes a new program.
1263  */
1264 int do_execve(char * filename,
1265         char __user *__user *argv,
1266         char __user *__user *envp,
1267         struct pt_regs * regs)
1268 {
1269         struct linux_binprm *bprm;
1270         struct file *file;
1271         unsigned long env_p;
1272         struct files_struct *displaced;
1273         int retval;
1274
1275         retval = unshare_files(&displaced);
1276         if (retval)
1277                 goto out_ret;
1278
1279         retval = -ENOMEM;
1280         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1281         if (!bprm)
1282                 goto out_files;
1283
1284         file = open_exec(filename);
1285         retval = PTR_ERR(file);
1286         if (IS_ERR(file))
1287                 goto out_kfree;
1288
1289         sched_exec();
1290
1291         bprm->file = file;
1292         bprm->filename = filename;
1293         bprm->interp = filename;
1294
1295         retval = bprm_mm_init(bprm);
1296         if (retval)
1297                 goto out_file;
1298
1299         bprm->argc = count(argv, MAX_ARG_STRINGS);
1300         if ((retval = bprm->argc) < 0)
1301                 goto out_mm;
1302
1303         bprm->envc = count(envp, MAX_ARG_STRINGS);
1304         if ((retval = bprm->envc) < 0)
1305                 goto out_mm;
1306
1307         retval = security_bprm_alloc(bprm);
1308         if (retval)
1309                 goto out;
1310
1311         retval = prepare_binprm(bprm);
1312         if (retval < 0)
1313                 goto out;
1314
1315         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1316         if (retval < 0)
1317                 goto out;
1318
1319         bprm->exec = bprm->p;
1320         retval = copy_strings(bprm->envc, envp, bprm);
1321         if (retval < 0)
1322                 goto out;
1323
1324         env_p = bprm->p;
1325         retval = copy_strings(bprm->argc, argv, bprm);
1326         if (retval < 0)
1327                 goto out;
1328         bprm->argv_len = env_p - bprm->p;
1329
1330         retval = search_binary_handler(bprm,regs);
1331         if (retval >= 0) {
1332                 /* execve success */
1333                 free_arg_pages(bprm);
1334                 security_bprm_free(bprm);
1335                 acct_update_integrals(current);
1336                 kfree(bprm);
1337                 if (displaced)
1338                         put_files_struct(displaced);
1339                 return retval;
1340         }
1341
1342 out:
1343         free_arg_pages(bprm);
1344         if (bprm->security)
1345                 security_bprm_free(bprm);
1346
1347 out_mm:
1348         if (bprm->mm)
1349                 mmput (bprm->mm);
1350
1351 out_file:
1352         if (bprm->file) {
1353                 allow_write_access(bprm->file);
1354                 fput(bprm->file);
1355         }
1356 out_kfree:
1357         kfree(bprm);
1358
1359 out_files:
1360         if (displaced)
1361                 reset_files_struct(displaced);
1362 out_ret:
1363         return retval;
1364 }
1365
1366 int set_binfmt(struct linux_binfmt *new)
1367 {
1368         struct linux_binfmt *old = current->binfmt;
1369
1370         if (new) {
1371                 if (!try_module_get(new->module))
1372                         return -1;
1373         }
1374         current->binfmt = new;
1375         if (old)
1376                 module_put(old->module);
1377         return 0;
1378 }
1379
1380 EXPORT_SYMBOL(set_binfmt);
1381
1382 /* format_corename will inspect the pattern parameter, and output a
1383  * name into corename, which must have space for at least
1384  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1385  */
1386 static int format_corename(char *corename, const char *pattern, long signr)
1387 {
1388         const char *pat_ptr = pattern;
1389         char *out_ptr = corename;
1390         char *const out_end = corename + CORENAME_MAX_SIZE;
1391         int rc;
1392         int pid_in_pattern = 0;
1393         int ispipe = 0;
1394
1395         if (*pattern == '|')
1396                 ispipe = 1;
1397
1398         /* Repeat as long as we have more pattern to process and more output
1399            space */
1400         while (*pat_ptr) {
1401                 if (*pat_ptr != '%') {
1402                         if (out_ptr == out_end)
1403                                 goto out;
1404                         *out_ptr++ = *pat_ptr++;
1405                 } else {
1406                         switch (*++pat_ptr) {
1407                         case 0:
1408                                 goto out;
1409                         /* Double percent, output one percent */
1410                         case '%':
1411                                 if (out_ptr == out_end)
1412                                         goto out;
1413                                 *out_ptr++ = '%';
1414                                 break;
1415                         /* pid */
1416                         case 'p':
1417                                 pid_in_pattern = 1;
1418                                 rc = snprintf(out_ptr, out_end - out_ptr,
1419                                               "%d", task_tgid_vnr(current));
1420                                 if (rc > out_end - out_ptr)
1421                                         goto out;
1422                                 out_ptr += rc;
1423                                 break;
1424                         /* uid */
1425                         case 'u':
1426                                 rc = snprintf(out_ptr, out_end - out_ptr,
1427                                               "%d", current->uid);
1428                                 if (rc > out_end - out_ptr)
1429                                         goto out;
1430                                 out_ptr += rc;
1431                                 break;
1432                         /* gid */
1433                         case 'g':
1434                                 rc = snprintf(out_ptr, out_end - out_ptr,
1435                                               "%d", current->gid);
1436                                 if (rc > out_end - out_ptr)
1437                                         goto out;
1438                                 out_ptr += rc;
1439                                 break;
1440                         /* signal that caused the coredump */
1441                         case 's':
1442                                 rc = snprintf(out_ptr, out_end - out_ptr,
1443                                               "%ld", signr);
1444                                 if (rc > out_end - out_ptr)
1445                                         goto out;
1446                                 out_ptr += rc;
1447                                 break;
1448                         /* UNIX time of coredump */
1449                         case 't': {
1450                                 struct timeval tv;
1451                                 do_gettimeofday(&tv);
1452                                 rc = snprintf(out_ptr, out_end - out_ptr,
1453                                               "%lu", tv.tv_sec);
1454                                 if (rc > out_end - out_ptr)
1455                                         goto out;
1456                                 out_ptr += rc;
1457                                 break;
1458                         }
1459                         /* hostname */
1460                         case 'h':
1461                                 down_read(&uts_sem);
1462                                 rc = snprintf(out_ptr, out_end - out_ptr,
1463                                               "%s", utsname()->nodename);
1464                                 up_read(&uts_sem);
1465                                 if (rc > out_end - out_ptr)
1466                                         goto out;
1467                                 out_ptr += rc;
1468                                 break;
1469                         /* executable */
1470                         case 'e':
1471                                 rc = snprintf(out_ptr, out_end - out_ptr,
1472                                               "%s", current->comm);
1473                                 if (rc > out_end - out_ptr)
1474                                         goto out;
1475                                 out_ptr += rc;
1476                                 break;
1477                         /* core limit size */
1478                         case 'c':
1479                                 rc = snprintf(out_ptr, out_end - out_ptr,
1480                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1481                                 if (rc > out_end - out_ptr)
1482                                         goto out;
1483                                 out_ptr += rc;
1484                                 break;
1485                         default:
1486                                 break;
1487                         }
1488                         ++pat_ptr;
1489                 }
1490         }
1491         /* Backward compatibility with core_uses_pid:
1492          *
1493          * If core_pattern does not include a %p (as is the default)
1494          * and core_uses_pid is set, then .%pid will be appended to
1495          * the filename. Do not do this for piped commands. */
1496         if (!ispipe && !pid_in_pattern
1497             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1498                 rc = snprintf(out_ptr, out_end - out_ptr,
1499                               ".%d", task_tgid_vnr(current));
1500                 if (rc > out_end - out_ptr)
1501                         goto out;
1502                 out_ptr += rc;
1503         }
1504 out:
1505         *out_ptr = 0;
1506         return ispipe;
1507 }
1508
1509 static void zap_process(struct task_struct *start)
1510 {
1511         struct task_struct *t;
1512
1513         start->signal->flags = SIGNAL_GROUP_EXIT;
1514         start->signal->group_stop_count = 0;
1515
1516         t = start;
1517         do {
1518                 if (t != current && t->mm) {
1519                         t->mm->core_waiters++;
1520                         sigaddset(&t->pending.signal, SIGKILL);
1521                         signal_wake_up(t, 1);
1522                 }
1523         } while ((t = next_thread(t)) != start);
1524 }
1525
1526 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1527                                 int exit_code)
1528 {
1529         struct task_struct *g, *p;
1530         unsigned long flags;
1531         int err = -EAGAIN;
1532
1533         spin_lock_irq(&tsk->sighand->siglock);
1534         if (!signal_group_exit(tsk->signal)) {
1535                 tsk->signal->group_exit_code = exit_code;
1536                 zap_process(tsk);
1537                 err = 0;
1538         }
1539         spin_unlock_irq(&tsk->sighand->siglock);
1540         if (err)
1541                 return err;
1542
1543         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1544                 goto done;
1545
1546         rcu_read_lock();
1547         for_each_process(g) {
1548                 if (g == tsk->group_leader)
1549                         continue;
1550
1551                 p = g;
1552                 do {
1553                         if (p->mm) {
1554                                 if (p->mm == mm) {
1555                                         /*
1556                                          * p->sighand can't disappear, but
1557                                          * may be changed by de_thread()
1558                                          */
1559                                         lock_task_sighand(p, &flags);
1560                                         zap_process(p);
1561                                         unlock_task_sighand(p, &flags);
1562                                 }
1563                                 break;
1564                         }
1565                 } while ((p = next_thread(p)) != g);
1566         }
1567         rcu_read_unlock();
1568 done:
1569         return mm->core_waiters;
1570 }
1571
1572 static int coredump_wait(int exit_code)
1573 {
1574         struct task_struct *tsk = current;
1575         struct mm_struct *mm = tsk->mm;
1576         struct completion startup_done;
1577         struct completion *vfork_done;
1578         int core_waiters;
1579
1580         init_completion(&mm->core_done);
1581         init_completion(&startup_done);
1582         mm->core_startup_done = &startup_done;
1583
1584         core_waiters = zap_threads(tsk, mm, exit_code);
1585         up_write(&mm->mmap_sem);
1586
1587         if (unlikely(core_waiters < 0))
1588                 goto fail;
1589
1590         /*
1591          * Make sure nobody is waiting for us to release the VM,
1592          * otherwise we can deadlock when we wait on each other
1593          */
1594         vfork_done = tsk->vfork_done;
1595         if (vfork_done) {
1596                 tsk->vfork_done = NULL;
1597                 complete(vfork_done);
1598         }
1599
1600         if (core_waiters)
1601                 wait_for_completion(&startup_done);
1602 fail:
1603         BUG_ON(mm->core_waiters);
1604         return core_waiters;
1605 }
1606
1607 /*
1608  * set_dumpable converts traditional three-value dumpable to two flags and
1609  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1610  * these bits are not changed atomically.  So get_dumpable can observe the
1611  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1612  * return either old dumpable or new one by paying attention to the order of
1613  * modifying the bits.
1614  *
1615  * dumpable |   mm->flags (binary)
1616  * old  new | initial interim  final
1617  * ---------+-----------------------
1618  *  0    1  |   00      01      01
1619  *  0    2  |   00      10(*)   11
1620  *  1    0  |   01      00      00
1621  *  1    2  |   01      11      11
1622  *  2    0  |   11      10(*)   00
1623  *  2    1  |   11      11      01
1624  *
1625  * (*) get_dumpable regards interim value of 10 as 11.
1626  */
1627 void set_dumpable(struct mm_struct *mm, int value)
1628 {
1629         switch (value) {
1630         case 0:
1631                 clear_bit(MMF_DUMPABLE, &mm->flags);
1632                 smp_wmb();
1633                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1634                 break;
1635         case 1:
1636                 set_bit(MMF_DUMPABLE, &mm->flags);
1637                 smp_wmb();
1638                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1639                 break;
1640         case 2:
1641                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1642                 smp_wmb();
1643                 set_bit(MMF_DUMPABLE, &mm->flags);
1644                 break;
1645         }
1646 }
1647
1648 int get_dumpable(struct mm_struct *mm)
1649 {
1650         int ret;
1651
1652         ret = mm->flags & 0x3;
1653         return (ret >= 2) ? 2 : ret;
1654 }
1655
1656 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1657 {
1658         char corename[CORENAME_MAX_SIZE + 1];
1659         struct mm_struct *mm = current->mm;
1660         struct linux_binfmt * binfmt;
1661         struct inode * inode;
1662         struct file * file;
1663         int retval = 0;
1664         int fsuid = current->fsuid;
1665         int flag = 0;
1666         int ispipe = 0;
1667         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1668         char **helper_argv = NULL;
1669         int helper_argc = 0;
1670         char *delimit;
1671
1672         audit_core_dumps(signr);
1673
1674         binfmt = current->binfmt;
1675         if (!binfmt || !binfmt->core_dump)
1676                 goto fail;
1677         down_write(&mm->mmap_sem);
1678         /*
1679          * If another thread got here first, or we are not dumpable, bail out.
1680          */
1681         if (mm->core_waiters || !get_dumpable(mm)) {
1682                 up_write(&mm->mmap_sem);
1683                 goto fail;
1684         }
1685
1686         /*
1687          *      We cannot trust fsuid as being the "true" uid of the
1688          *      process nor do we know its entire history. We only know it
1689          *      was tainted so we dump it as root in mode 2.
1690          */
1691         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1692                 flag = O_EXCL;          /* Stop rewrite attacks */
1693                 current->fsuid = 0;     /* Dump root private */
1694         }
1695
1696         retval = coredump_wait(exit_code);
1697         if (retval < 0)
1698                 goto fail;
1699
1700         /*
1701          * Clear any false indication of pending signals that might
1702          * be seen by the filesystem code called to write the core file.
1703          */
1704         clear_thread_flag(TIF_SIGPENDING);
1705
1706         /*
1707          * lock_kernel() because format_corename() is controlled by sysctl, which
1708          * uses lock_kernel()
1709          */
1710         lock_kernel();
1711         ispipe = format_corename(corename, core_pattern, signr);
1712         unlock_kernel();
1713         /*
1714          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1715          * to a pipe.  Since we're not writing directly to the filesystem
1716          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1717          * created unless the pipe reader choses to write out the core file
1718          * at which point file size limits and permissions will be imposed
1719          * as it does with any other process
1720          */
1721         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1722                 goto fail_unlock;
1723
1724         if (ispipe) {
1725                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1726                 /* Terminate the string before the first option */
1727                 delimit = strchr(corename, ' ');
1728                 if (delimit)
1729                         *delimit = '\0';
1730                 delimit = strrchr(helper_argv[0], '/');
1731                 if (delimit)
1732                         delimit++;
1733                 else
1734                         delimit = helper_argv[0];
1735                 if (!strcmp(delimit, current->comm)) {
1736                         printk(KERN_NOTICE "Recursive core dump detected, "
1737                                         "aborting\n");
1738                         goto fail_unlock;
1739                 }
1740
1741                 core_limit = RLIM_INFINITY;
1742
1743                 /* SIGPIPE can happen, but it's just never processed */
1744                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1745                                 &file)) {
1746                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1747                                corename);
1748                         goto fail_unlock;
1749                 }
1750         } else
1751                 file = filp_open(corename,
1752                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1753                                  0600);
1754         if (IS_ERR(file))
1755                 goto fail_unlock;
1756         inode = file->f_path.dentry->d_inode;
1757         if (inode->i_nlink > 1)
1758                 goto close_fail;        /* multiple links - don't dump */
1759         if (!ispipe && d_unhashed(file->f_path.dentry))
1760                 goto close_fail;
1761
1762         /* AK: actually i see no reason to not allow this for named pipes etc.,
1763            but keep the previous behaviour for now. */
1764         if (!ispipe && !S_ISREG(inode->i_mode))
1765                 goto close_fail;
1766         /*
1767          * Dont allow local users get cute and trick others to coredump
1768          * into their pre-created files:
1769          */
1770         if (inode->i_uid != current->fsuid)
1771                 goto close_fail;
1772         if (!file->f_op)
1773                 goto close_fail;
1774         if (!file->f_op->write)
1775                 goto close_fail;
1776         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1777                 goto close_fail;
1778
1779         retval = binfmt->core_dump(signr, regs, file, core_limit);
1780
1781         if (retval)
1782                 current->signal->group_exit_code |= 0x80;
1783 close_fail:
1784         filp_close(file, NULL);
1785 fail_unlock:
1786         if (helper_argv)
1787                 argv_free(helper_argv);
1788
1789         current->fsuid = fsuid;
1790         complete_all(&mm->core_done);
1791 fail:
1792         return retval;
1793 }