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/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55
  56 #include <asm/uaccess.h>
  57 #include <asm/mmu_context.h>
  58 #include <asm/tlb.h>
  59 #include "internal.h"
  60
  61 int core_uses_pid;
  62 char core_pattern[CORENAME_MAX_SIZE] = "core";
  63 int suid_dumpable = 0;
  64
  65 /* The maximal length of core_pattern is also specified in sysctl.c */
  66
  67 static LIST_HEAD(formats);
  68 static DEFINE_RWLOCK(binfmt_lock);
  69
  70 int register_binfmt(struct linux_binfmt * fmt)
  71 {
  72         if (!fmt)
  73                 return -EINVAL;
  74         write_lock(&binfmt_lock);
  75         list_add(&fmt->lh, &formats);
  76         write_unlock(&binfmt_lock);
  77         return 0;
  78 }
  79
  80 EXPORT_SYMBOL(register_binfmt);
  81
  82 void unregister_binfmt(struct linux_binfmt * fmt)
  83 {
  84         write_lock(&binfmt_lock);
  85         list_del(&fmt->lh);
  86         write_unlock(&binfmt_lock);
  87 }
  88
  89 EXPORT_SYMBOL(unregister_binfmt);
  90
  91 static inline void put_binfmt(struct linux_binfmt * fmt)
  92 {
  93         module_put(fmt->module);
  94 }
  95
  96 /*
  97  * Note that a shared library must be both readable and executable due to
  98  * security reasons.
  99  *
 100  * Also note that we take the address to load from from the file itself.
 101  */
 102 SYSCALL_DEFINE1(uselib, const char __user *, library)
 103 {
 104         struct file *file;
 105         struct nameidata nd;
 106         char *tmp = getname(library);
 107         int error = PTR_ERR(tmp);
 108
 109         if (!IS_ERR(tmp)) {
 110                 error = path_lookup_open(AT_FDCWD, tmp,
 111                                          LOOKUP_FOLLOW, &nd,
 112                                          FMODE_READ|FMODE_EXEC);
 113                 putname(tmp);
 114         }
 115         if (error)
 116                 goto out;
 117
 118         error = -EINVAL;
 119         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 120                 goto exit;
 121
 122         error = -EACCES;
 123         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
 124                 goto exit;
 125
 126         error = inode_permission(nd.path.dentry->d_inode,
 127                                  MAY_READ | MAY_EXEC | MAY_OPEN);
 128         if (error)
 129                 goto exit;
 130
 131         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 132         error = PTR_ERR(file);
 133         if (IS_ERR(file))
 134                 goto out;
 135
 136         fsnotify_open(file->f_path.dentry);
 137
 138         error = -ENOEXEC;
 139         if(file->f_op) {
 140                 struct linux_binfmt * fmt;
 141
 142                 read_lock(&binfmt_lock);
 143                 list_for_each_entry(fmt, &formats, lh) {
 144                         if (!fmt->load_shlib)
 145                                 continue;
 146                         if (!try_module_get(fmt->module))
 147                                 continue;
 148                         read_unlock(&binfmt_lock);
 149                         error = fmt->load_shlib(file);
 150                         read_lock(&binfmt_lock);
 151                         put_binfmt(fmt);
 152                         if (error != -ENOEXEC)
 153                                 break;
 154                 }
 155                 read_unlock(&binfmt_lock);
 156         }
 157         fput(file);
 158 out:
 159         return error;
 160 exit:
 161         release_open_intent(&nd);
 162         path_put(&nd.path);
 163         goto out;
 164 }
 165
 166 #ifdef CONFIG_MMU
 167
 168 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 169                 int write)
 170 {
 171         struct page *page;
 172         int ret;
 173
 174 #ifdef CONFIG_STACK_GROWSUP
 175         if (write) {
 176                 ret = expand_stack_downwards(bprm->vma, pos);
 177                 if (ret < 0)
 178                         return NULL;
 179         }
 180 #endif
 181         ret = get_user_pages(current, bprm->mm, pos,
 182                         1, write, 1, &page, NULL);
 183         if (ret <= 0)
 184                 return NULL;
 185
 186         if (write) {
 187                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 188                 struct rlimit *rlim;
 189
 190                 /*
 191                  * We've historically supported up to 32 pages (ARG_MAX)
 192                  * of argument strings even with small stacks
 193                  */
 194                 if (size <= ARG_MAX)
 195                         return page;
 196
 197                 /*
 198                  * Limit to 1/4-th the stack size for the argv+env strings.
 199                  * This ensures that:
 200                  *  - the remaining binfmt code will not run out of stack space,
 201                  *  - the program will have a reasonable amount of stack left
 202                  *    to work from.
 203                  */
 204                 rlim = current->signal->rlim;
 205                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 206                         put_page(page);
 207                         return NULL;
 208                 }
 209         }
 210
 211         return page;
 212 }
 213
 214 static void put_arg_page(struct page *page)
 215 {
 216         put_page(page);
 217 }
 218
 219 static void free_arg_page(struct linux_binprm *bprm, int i)
 220 {
 221 }
 222
 223 static void free_arg_pages(struct linux_binprm *bprm)
 224 {
 225 }
 226
 227 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 228                 struct page *page)
 229 {
 230         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 231 }
 232
 233 static int __bprm_mm_init(struct linux_binprm *bprm)
 234 {
 235         int err;
 236         struct vm_area_struct *vma = NULL;
 237         struct mm_struct *mm = bprm->mm;
 238
 239         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 240         if (!vma)
 241                 return -ENOMEM;
 242
 243         down_write(&mm->mmap_sem);
 244         vma->vm_mm = mm;
 245
 246         /*
 247          * Place the stack at the largest stack address the architecture
 248          * supports. Later, we'll move this to an appropriate place. We don't
 249          * use STACK_TOP because that can depend on attributes which aren't
 250          * configured yet.
 251          */
 252         vma->vm_end = STACK_TOP_MAX;
 253         vma->vm_start = vma->vm_end - PAGE_SIZE;
 254         vma->vm_flags = VM_STACK_FLAGS;
 255         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 256         err = insert_vm_struct(mm, vma);
 257         if (err)
 258                 goto err;
 259
 260         mm->stack_vm = mm->total_vm = 1;
 261         up_write(&mm->mmap_sem);
 262         bprm->p = vma->vm_end - sizeof(void *);
 263         return 0;
 264 err:
 265         up_write(&mm->mmap_sem);
 266         bprm->vma = NULL;
 267         kmem_cache_free(vm_area_cachep, vma);
 268         return err;
 269 }
 270
 271 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 272 {
 273         return len <= MAX_ARG_STRLEN;
 274 }
 275
 276 #else
 277
 278 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 279                 int write)
 280 {
 281         struct page *page;
 282
 283         page = bprm->page[pos / PAGE_SIZE];
 284         if (!page && write) {
 285                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 286                 if (!page)
 287                         return NULL;
 288                 bprm->page[pos / PAGE_SIZE] = page;
 289         }
 290
 291         return page;
 292 }
 293
 294 static void put_arg_page(struct page *page)
 295 {
 296 }
 297
 298 static void free_arg_page(struct linux_binprm *bprm, int i)
 299 {
 300         if (bprm->page[i]) {
 301                 __free_page(bprm->page[i]);
 302                 bprm->page[i] = NULL;
 303         }
 304 }
 305
 306 static void free_arg_pages(struct linux_binprm *bprm)
 307 {
 308         int i;
 309
 310         for (i = 0; i < MAX_ARG_PAGES; i++)
 311                 free_arg_page(bprm, i);
 312 }
 313
 314 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 315                 struct page *page)
 316 {
 317 }
 318
 319 static int __bprm_mm_init(struct linux_binprm *bprm)
 320 {
 321         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 322         return 0;
 323 }
 324
 325 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 326 {
 327         return len <= bprm->p;
 328 }
 329
 330 #endif /* CONFIG_MMU */
 331
 332 /*
 333  * Create a new mm_struct and populate it with a temporary stack
 334  * vm_area_struct.  We don't have enough context at this point to set the stack
 335  * flags, permissions, and offset, so we use temporary values.  We'll update
 336  * them later in setup_arg_pages().
 337  */
 338 int bprm_mm_init(struct linux_binprm *bprm)
 339 {
 340         int err;
 341         struct mm_struct *mm = NULL;
 342
 343         bprm->mm = mm = mm_alloc();
 344         err = -ENOMEM;
 345         if (!mm)
 346                 goto err;
 347
 348         err = init_new_context(current, mm);
 349         if (err)
 350                 goto err;
 351
 352         err = __bprm_mm_init(bprm);
 353         if (err)
 354                 goto err;
 355
 356         return 0;
 357
 358 err:
 359         if (mm) {
 360                 bprm->mm = NULL;
 361                 mmdrop(mm);
 362         }
 363
 364         return err;
 365 }
 366
 367 /*
 368  * count() counts the number of strings in array ARGV.
 369  */
 370 static int count(char __user * __user * argv, int max)
 371 {
 372         int i = 0;
 373
 374         if (argv != NULL) {
 375                 for (;;) {
 376                         char __user * p;
 377
 378                         if (get_user(p, argv))
 379                                 return -EFAULT;
 380                         if (!p)
 381                                 break;
 382                         argv++;
 383                         if (i++ >= max)
 384                                 return -E2BIG;
 385                         cond_resched();
 386                 }
 387         }
 388         return i;
 389 }
 390
 391 /*
 392  * 'copy_strings()' copies argument/environment strings from the old
 393  * processes's memory to the new process's stack.  The call to get_user_pages()
 394  * ensures the destination page is created and not swapped out.
 395  */
 396 static int copy_strings(int argc, char __user * __user * argv,
 397                         struct linux_binprm *bprm)
 398 {
 399         struct page *kmapped_page = NULL;
 400         char *kaddr = NULL;
 401         unsigned long kpos = 0;
 402         int ret;
 403
 404         while (argc-- > 0) {
 405                 char __user *str;
 406                 int len;
 407                 unsigned long pos;
 408
 409                 if (get_user(str, argv+argc) ||
 410                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 411                         ret = -EFAULT;
 412                         goto out;
 413                 }
 414
 415                 if (!valid_arg_len(bprm, len)) {
 416                         ret = -E2BIG;
 417                         goto out;
 418                 }
 419
 420                 /* We're going to work our way backwords. */
 421                 pos = bprm->p;
 422                 str += len;
 423                 bprm->p -= len;
 424
 425                 while (len > 0) {
 426                         int offset, bytes_to_copy;
 427
 428                         offset = pos % PAGE_SIZE;
 429                         if (offset == 0)
 430                                 offset = PAGE_SIZE;
 431
 432                         bytes_to_copy = offset;
 433                         if (bytes_to_copy > len)
 434                                 bytes_to_copy = len;
 435
 436                         offset -= bytes_to_copy;
 437                         pos -= bytes_to_copy;
 438                         str -= bytes_to_copy;
 439                         len -= bytes_to_copy;
 440
 441                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 442                                 struct page *page;
 443
 444                                 page = get_arg_page(bprm, pos, 1);
 445                                 if (!page) {
 446                                         ret = -E2BIG;
 447                                         goto out;
 448                                 }
 449
 450                                 if (kmapped_page) {
 451                                         flush_kernel_dcache_page(kmapped_page);
 452                                         kunmap(kmapped_page);
 453                                         put_arg_page(kmapped_page);
 454                                 }
 455                                 kmapped_page = page;
 456                                 kaddr = kmap(kmapped_page);
 457                                 kpos = pos & PAGE_MASK;
 458                                 flush_arg_page(bprm, kpos, kmapped_page);
 459                         }
 460                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 461                                 ret = -EFAULT;
 462                                 goto out;
 463                         }
 464                 }
 465         }
 466         ret = 0;
 467 out:
 468         if (kmapped_page) {
 469                 flush_kernel_dcache_page(kmapped_page);
 470                 kunmap(kmapped_page);
 471                 put_arg_page(kmapped_page);
 472         }
 473         return ret;
 474 }
 475
 476 /*
 477  * Like copy_strings, but get argv and its values from kernel memory.
 478  */
 479 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 480 {
 481         int r;
 482         mm_segment_t oldfs = get_fs();
 483         set_fs(KERNEL_DS);
 484         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 485         set_fs(oldfs);
 486         return r;
 487 }
 488 EXPORT_SYMBOL(copy_strings_kernel);
 489
 490 #ifdef CONFIG_MMU
 491
 492 /*
 493  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 494  * the binfmt code determines where the new stack should reside, we shift it to
 495  * its final location.  The process proceeds as follows:
 496  *
 497  * 1) Use shift to calculate the new vma endpoints.
 498  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 499  *    arguments passed to subsequent functions are consistent.
 500  * 3) Move vma's page tables to the new range.
 501  * 4) Free up any cleared pgd range.
 502  * 5) Shrink the vma to cover only the new range.
 503  */
 504 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 505 {
 506         struct mm_struct *mm = vma->vm_mm;
 507         unsigned long old_start = vma->vm_start;
 508         unsigned long old_end = vma->vm_end;
 509         unsigned long length = old_end - old_start;
 510         unsigned long new_start = old_start - shift;
 511         unsigned long new_end = old_end - shift;
 512         struct mmu_gather *tlb;
 513
 514         BUG_ON(new_start > new_end);
 515
 516         /*
 517          * ensure there are no vmas between where we want to go
 518          * and where we are
 519          */
 520         if (vma != find_vma(mm, new_start))
 521                 return -EFAULT;
 522
 523         /*
 524          * cover the whole range: [new_start, old_end)
 525          */
 526         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 527
 528         /*
 529          * move the page tables downwards, on failure we rely on
 530          * process cleanup to remove whatever mess we made.
 531          */
 532         if (length != move_page_tables(vma, old_start,
 533                                        vma, new_start, length))
 534                 return -ENOMEM;
 535
 536         lru_add_drain();
 537         tlb = tlb_gather_mmu(mm, 0);
 538         if (new_end > old_start) {
 539                 /*
 540                  * when the old and new regions overlap clear from new_end.
 541                  */
 542                 free_pgd_range(tlb, new_end, old_end, new_end,
 543                         vma->vm_next ? vma->vm_next->vm_start : 0);
 544         } else {
 545                 /*
 546                  * otherwise, clean from old_start; this is done to not touch
 547                  * the address space in [new_end, old_start) some architectures
 548                  * have constraints on va-space that make this illegal (IA64) -
 549                  * for the others its just a little faster.
 550                  */
 551                 free_pgd_range(tlb, old_start, old_end, new_end,
 552                         vma->vm_next ? vma->vm_next->vm_start : 0);
 553         }
 554         tlb_finish_mmu(tlb, new_end, old_end);
 555
 556         /*
 557          * shrink the vma to just the new range.
 558          */
 559         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 560
 561         return 0;
 562 }
 563
 564 #define EXTRA_STACK_VM_PAGES    20      /* random */
 565
 566 /*
 567  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 568  * the stack is optionally relocated, and some extra space is added.
 569  */
 570 int setup_arg_pages(struct linux_binprm *bprm,
 571                     unsigned long stack_top,
 572                     int executable_stack)
 573 {
 574         unsigned long ret;
 575         unsigned long stack_shift;
 576         struct mm_struct *mm = current->mm;
 577         struct vm_area_struct *vma = bprm->vma;
 578         struct vm_area_struct *prev = NULL;
 579         unsigned long vm_flags;
 580         unsigned long stack_base;
 581
 582 #ifdef CONFIG_STACK_GROWSUP
 583         /* Limit stack size to 1GB */
 584         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 585         if (stack_base > (1 << 30))
 586                 stack_base = 1 << 30;
 587
 588         /* Make sure we didn't let the argument array grow too large. */
 589         if (vma->vm_end - vma->vm_start > stack_base)
 590                 return -ENOMEM;
 591
 592         stack_base = PAGE_ALIGN(stack_top - stack_base);
 593
 594         stack_shift = vma->vm_start - stack_base;
 595         mm->arg_start = bprm->p - stack_shift;
 596         bprm->p = vma->vm_end - stack_shift;
 597 #else
 598         stack_top = arch_align_stack(stack_top);
 599         stack_top = PAGE_ALIGN(stack_top);
 600         stack_shift = vma->vm_end - stack_top;
 601
 602         bprm->p -= stack_shift;
 603         mm->arg_start = bprm->p;
 604 #endif
 605
 606         if (bprm->loader)
 607                 bprm->loader -= stack_shift;
 608         bprm->exec -= stack_shift;
 609
 610         down_write(&mm->mmap_sem);
 611         vm_flags = VM_STACK_FLAGS;
 612
 613         /*
 614          * Adjust stack execute permissions; explicitly enable for
 615          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 616          * (arch default) otherwise.
 617          */
 618         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 619                 vm_flags |= VM_EXEC;
 620         else if (executable_stack == EXSTACK_DISABLE_X)
 621                 vm_flags &= ~VM_EXEC;
 622         vm_flags |= mm->def_flags;
 623
 624         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 625                         vm_flags);
 626         if (ret)
 627                 goto out_unlock;
 628         BUG_ON(prev != vma);
 629
 630         /* Move stack pages down in memory. */
 631         if (stack_shift) {
 632                 ret = shift_arg_pages(vma, stack_shift);
 633                 if (ret) {
 634                         up_write(&mm->mmap_sem);
 635                         return ret;
 636                 }
 637         }
 638
 639 #ifdef CONFIG_STACK_GROWSUP
 640         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 641 #else
 642         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 643 #endif
 644         ret = expand_stack(vma, stack_base);
 645         if (ret)
 646                 ret = -EFAULT;
 647
 648 out_unlock:
 649         up_write(&mm->mmap_sem);
 650         return 0;
 651 }
 652 EXPORT_SYMBOL(setup_arg_pages);
 653
 654 #endif /* CONFIG_MMU */
 655
 656 struct file *open_exec(const char *name)
 657 {
 658         struct nameidata nd;
 659         struct file *file;
 660         int err;
 661
 662         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
 663                                 FMODE_READ|FMODE_EXEC);
 664         if (err)
 665                 goto out;
 666
 667         err = -EACCES;
 668         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 669                 goto out_path_put;
 670
 671         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
 672                 goto out_path_put;
 673
 674         err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
 675         if (err)
 676                 goto out_path_put;
 677
 678         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 679         if (IS_ERR(file))
 680                 return file;
 681
 682         fsnotify_open(file->f_path.dentry);
 683
 684         err = deny_write_access(file);
 685         if (err) {
 686                 fput(file);
 687                 goto out;
 688         }
 689
 690         return file;
 691
 692  out_path_put:
 693         release_open_intent(&nd);
 694         path_put(&nd.path);
 695  out:
 696         return ERR_PTR(err);
 697 }
 698 EXPORT_SYMBOL(open_exec);
 699
 700 int kernel_read(struct file *file, unsigned long offset,
 701         char *addr, unsigned long count)
 702 {
 703         mm_segment_t old_fs;
 704         loff_t pos = offset;
 705         int result;
 706
 707         old_fs = get_fs();
 708         set_fs(get_ds());
 709         /* The cast to a user pointer is valid due to the set_fs() */
 710         result = vfs_read(file, (void __user *)addr, count, &pos);
 711         set_fs(old_fs);
 712         return result;
 713 }
 714
 715 EXPORT_SYMBOL(kernel_read);
 716
 717 static int exec_mmap(struct mm_struct *mm)
 718 {
 719         struct task_struct *tsk;
 720         struct mm_struct * old_mm, *active_mm;
 721
 722         /* Notify parent that we're no longer interested in the old VM */
 723         tsk = current;
 724         old_mm = current->mm;
 725         mm_release(tsk, old_mm);
 726
 727         if (old_mm) {
 728                 /*
 729                  * Make sure that if there is a core dump in progress
 730                  * for the old mm, we get out and die instead of going
 731                  * through with the exec.  We must hold mmap_sem around
 732                  * checking core_state and changing tsk->mm.
 733                  */
 734                 down_read(&old_mm->mmap_sem);
 735                 if (unlikely(old_mm->core_state)) {
 736                         up_read(&old_mm->mmap_sem);
 737                         return -EINTR;
 738                 }
 739         }
 740         task_lock(tsk);
 741         active_mm = tsk->active_mm;
 742         tsk->mm = mm;
 743         tsk->active_mm = mm;
 744         activate_mm(active_mm, mm);
 745         task_unlock(tsk);
 746         arch_pick_mmap_layout(mm);
 747         if (old_mm) {
 748                 up_read(&old_mm->mmap_sem);
 749                 BUG_ON(active_mm != old_mm);
 750                 mm_update_next_owner(old_mm);
 751                 mmput(old_mm);
 752                 return 0;
 753         }
 754         mmdrop(active_mm);
 755         return 0;
 756 }
 757
 758 /*
 759  * This function makes sure the current process has its own signal table,
 760  * so that flush_signal_handlers can later reset the handlers without
 761  * disturbing other processes.  (Other processes might share the signal
 762  * table via the CLONE_SIGHAND option to clone().)
 763  */
 764 static int de_thread(struct task_struct *tsk)
 765 {
 766         struct signal_struct *sig = tsk->signal;
 767         struct sighand_struct *oldsighand = tsk->sighand;
 768         spinlock_t *lock = &oldsighand->siglock;
 769         int count;
 770
 771         if (thread_group_empty(tsk))
 772                 goto no_thread_group;
 773
 774         /*
 775          * Kill all other threads in the thread group.
 776          */
 777         spin_lock_irq(lock);
 778         if (signal_group_exit(sig)) {
 779                 /*
 780                  * Another group action in progress, just
 781                  * return so that the signal is processed.
 782                  */
 783                 spin_unlock_irq(lock);
 784                 return -EAGAIN;
 785         }
 786         sig->group_exit_task = tsk;
 787         zap_other_threads(tsk);
 788
 789         /* Account for the thread group leader hanging around: */
 790         count = thread_group_leader(tsk) ? 1 : 2;
 791         sig->notify_count = count;
 792         while (atomic_read(&sig->count) > count) {
 793                 __set_current_state(TASK_UNINTERRUPTIBLE);
 794                 spin_unlock_irq(lock);
 795                 schedule();
 796                 spin_lock_irq(lock);
 797         }
 798         spin_unlock_irq(lock);
 799
 800         /*
 801          * At this point all other threads have exited, all we have to
 802          * do is to wait for the thread group leader to become inactive,
 803          * and to assume its PID:
 804          */
 805         if (!thread_group_leader(tsk)) {
 806                 struct task_struct *leader = tsk->group_leader;
 807
 808                 sig->notify_count = -1; /* for exit_notify() */
 809                 for (;;) {
 810                         write_lock_irq(&tasklist_lock);
 811                         if (likely(leader->exit_state))
 812                                 break;
 813                         __set_current_state(TASK_UNINTERRUPTIBLE);
 814                         write_unlock_irq(&tasklist_lock);
 815                         schedule();
 816                 }
 817
 818                 /*
 819                  * The only record we have of the real-time age of a
 820                  * process, regardless of execs it's done, is start_time.
 821                  * All the past CPU time is accumulated in signal_struct
 822                  * from sister threads now dead.  But in this non-leader
 823                  * exec, nothing survives from the original leader thread,
 824                  * whose birth marks the true age of this process now.
 825                  * When we take on its identity by switching to its PID, we
 826                  * also take its birthdate (always earlier than our own).
 827                  */
 828                 tsk->start_time = leader->start_time;
 829
 830                 BUG_ON(!same_thread_group(leader, tsk));
 831                 BUG_ON(has_group_leader_pid(tsk));
 832                 /*
 833                  * An exec() starts a new thread group with the
 834                  * TGID of the previous thread group. Rehash the
 835                  * two threads with a switched PID, and release
 836                  * the former thread group leader:
 837                  */
 838
 839                 /* Become a process group leader with the old leader's pid.
 840                  * The old leader becomes a thread of the this thread group.
 841                  * Note: The old leader also uses this pid until release_task
 842                  *       is called.  Odd but simple and correct.
 843                  */
 844                 detach_pid(tsk, PIDTYPE_PID);
 845                 tsk->pid = leader->pid;
 846                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 847                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 848                 transfer_pid(leader, tsk, PIDTYPE_SID);
 849                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 850
 851                 tsk->group_leader = tsk;
 852                 leader->group_leader = tsk;
 853
 854                 tsk->exit_signal = SIGCHLD;
 855
 856                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 857                 leader->exit_state = EXIT_DEAD;
 858                 write_unlock_irq(&tasklist_lock);
 859
 860                 release_task(leader);
 861         }
 862
 863         sig->group_exit_task = NULL;
 864         sig->notify_count = 0;
 865
 866 no_thread_group:
 867         exit_itimers(sig);
 868         flush_itimer_signals();
 869
 870         if (atomic_read(&oldsighand->count) != 1) {
 871                 struct sighand_struct *newsighand;
 872                 /*
 873                  * This ->sighand is shared with the CLONE_SIGHAND
 874                  * but not CLONE_THREAD task, switch to the new one.
 875                  */
 876                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 877                 if (!newsighand)
 878                         return -ENOMEM;
 879
 880                 atomic_set(&newsighand->count, 1);
 881                 memcpy(newsighand->action, oldsighand->action,
 882                        sizeof(newsighand->action));
 883
 884                 write_lock_irq(&tasklist_lock);
 885                 spin_lock(&oldsighand->siglock);
 886                 rcu_assign_pointer(tsk->sighand, newsighand);
 887                 spin_unlock(&oldsighand->siglock);
 888                 write_unlock_irq(&tasklist_lock);
 889
 890                 __cleanup_sighand(oldsighand);
 891         }
 892
 893         BUG_ON(!thread_group_leader(tsk));
 894         return 0;
 895 }
 896
 897 /*
 898  * These functions flushes out all traces of the currently running executable
 899  * so that a new one can be started
 900  */
 901 static void flush_old_files(struct files_struct * files)
 902 {
 903         long j = -1;
 904         struct fdtable *fdt;
 905
 906         spin_lock(&files->file_lock);
 907         for (;;) {
 908                 unsigned long set, i;
 909
 910                 j++;
 911                 i = j * __NFDBITS;
 912                 fdt = files_fdtable(files);
 913                 if (i >= fdt->max_fds)
 914                         break;
 915                 set = fdt->close_on_exec->fds_bits[j];
 916                 if (!set)
 917                         continue;
 918                 fdt->close_on_exec->fds_bits[j] = 0;
 919                 spin_unlock(&files->file_lock);
 920                 for ( ; set ; i++,set >>= 1) {
 921                         if (set & 1) {
 922                                 sys_close(i);
 923                         }
 924                 }
 925                 spin_lock(&files->file_lock);
 926
 927         }
 928         spin_unlock(&files->file_lock);
 929 }
 930
 931 char *get_task_comm(char *buf, struct task_struct *tsk)
 932 {
 933         /* buf must be at least sizeof(tsk->comm) in size */
 934         task_lock(tsk);
 935         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 936         task_unlock(tsk);
 937         return buf;
 938 }
 939
 940 void set_task_comm(struct task_struct *tsk, char *buf)
 941 {
 942         task_lock(tsk);
 943         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 944         task_unlock(tsk);
 945 }
 946
 947 int flush_old_exec(struct linux_binprm * bprm)
 948 {
 949         char * name;
 950         int i, ch, retval;
 951         char tcomm[sizeof(current->comm)];
 952
 953         /*
 954          * Make sure we have a private signal table and that
 955          * we are unassociated from the previous thread group.
 956          */
 957         retval = de_thread(current);
 958         if (retval)
 959                 goto out;
 960
 961         set_mm_exe_file(bprm->mm, bprm->file);
 962
 963         /*
 964          * Release all of the old mmap stuff
 965          */
 966         retval = exec_mmap(bprm->mm);
 967         if (retval)
 968                 goto out;
 969
 970         bprm->mm = NULL;                /* We're using it now */
 971
 972         /* This is the point of no return */
 973         current->sas_ss_sp = current->sas_ss_size = 0;
 974
 975         if (current_euid() == current_uid() && current_egid() == current_gid())
 976                 set_dumpable(current->mm, 1);
 977         else
 978                 set_dumpable(current->mm, suid_dumpable);
 979
 980         name = bprm->filename;
 981
 982         /* Copies the binary name from after last slash */
 983         for (i=0; (ch = *(name++)) != '\0';) {
 984                 if (ch == '/')
 985                         i = 0; /* overwrite what we wrote */
 986                 else
 987                         if (i < (sizeof(tcomm) - 1))
 988                                 tcomm[i++] = ch;
 989         }
 990         tcomm[i] = '\0';
 991         set_task_comm(current, tcomm);
 992
 993         current->flags &= ~PF_RANDOMIZE;
 994         flush_thread();
 995
 996         /* Set the new mm task size. We have to do that late because it may
 997          * depend on TIF_32BIT which is only updated in flush_thread() on
 998          * some architectures like powerpc
 999          */
1000         current->mm->task_size = TASK_SIZE;
1001
1002         /* install the new credentials */
1003         if (bprm->cred->uid != current_euid() ||
1004             bprm->cred->gid != current_egid()) {
1005                 current->pdeath_signal = 0;
1006         } else if (file_permission(bprm->file, MAY_READ) ||
1007                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1008                 set_dumpable(current->mm, suid_dumpable);
1009         }
1010
1011         current->personality &= ~bprm->per_clear;
1012
1013         /* An exec changes our domain. We are no longer part of the thread
1014            group */
1015
1016         current->self_exec_id++;
1017
1018         flush_signal_handlers(current, 0);
1019         flush_old_files(current->files);
1020
1021         return 0;
1022
1023 out:
1024         return retval;
1025 }
1026
1027 EXPORT_SYMBOL(flush_old_exec);
1028
1029 /*
1030  * install the new credentials for this executable
1031  */
1032 void install_exec_creds(struct linux_binprm *bprm)
1033 {
1034         security_bprm_committing_creds(bprm);
1035
1036         commit_creds(bprm->cred);
1037         bprm->cred = NULL;
1038
1039         /* cred_exec_mutex must be held at least to this point to prevent
1040          * ptrace_attach() from altering our determination of the task's
1041          * credentials; any time after this it may be unlocked */
1042
1043         security_bprm_committed_creds(bprm);
1044 }
1045 EXPORT_SYMBOL(install_exec_creds);
1046
1047 /*
1048  * determine how safe it is to execute the proposed program
1049  * - the caller must hold current->cred_exec_mutex to protect against
1050  *   PTRACE_ATTACH
1051  */
1052 void check_unsafe_exec(struct linux_binprm *bprm)
1053 {
1054         struct task_struct *p = current;
1055
1056         bprm->unsafe = tracehook_unsafe_exec(p);
1057
1058         if (atomic_read(&p->fs->count) > 1 ||
1059             atomic_read(&p->files->count) > 1 ||
1060             atomic_read(&p->sighand->count) > 1)
1061                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1062 }
1063
1064 /*
1065  * Fill the binprm structure from the inode.
1066  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1067  *
1068  * This may be called multiple times for binary chains (scripts for example).
1069  */
1070 int prepare_binprm(struct linux_binprm *bprm)
1071 {
1072         umode_t mode;
1073         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1074         int retval;
1075
1076         mode = inode->i_mode;
1077         if (bprm->file->f_op == NULL)
1078                 return -EACCES;
1079
1080         /* clear any previous set[ug]id data from a previous binary */
1081         bprm->cred->euid = current_euid();
1082         bprm->cred->egid = current_egid();
1083
1084         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1085                 /* Set-uid? */
1086                 if (mode & S_ISUID) {
1087                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1088                         bprm->cred->euid = inode->i_uid;
1089                 }
1090
1091                 /* Set-gid? */
1092                 /*
1093                  * If setgid is set but no group execute bit then this
1094                  * is a candidate for mandatory locking, not a setgid
1095                  * executable.
1096                  */
1097                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1098                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1099                         bprm->cred->egid = inode->i_gid;
1100                 }
1101         }
1102
1103         /* fill in binprm security blob */
1104         retval = security_bprm_set_creds(bprm);
1105         if (retval)
1106                 return retval;
1107         bprm->cred_prepared = 1;
1108
1109         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1110         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1111 }
1112
1113 EXPORT_SYMBOL(prepare_binprm);
1114
1115 /*
1116  * Arguments are '\0' separated strings found at the location bprm->p
1117  * points to; chop off the first by relocating brpm->p to right after
1118  * the first '\0' encountered.
1119  */
1120 int remove_arg_zero(struct linux_binprm *bprm)
1121 {
1122         int ret = 0;
1123         unsigned long offset;
1124         char *kaddr;
1125         struct page *page;
1126
1127         if (!bprm->argc)
1128                 return 0;
1129
1130         do {
1131                 offset = bprm->p & ~PAGE_MASK;
1132                 page = get_arg_page(bprm, bprm->p, 0);
1133                 if (!page) {
1134                         ret = -EFAULT;
1135                         goto out;
1136                 }
1137                 kaddr = kmap_atomic(page, KM_USER0);
1138
1139                 for (; offset < PAGE_SIZE && kaddr[offset];
1140                                 offset++, bprm->p++)
1141                         ;
1142
1143                 kunmap_atomic(kaddr, KM_USER0);
1144                 put_arg_page(page);
1145
1146                 if (offset == PAGE_SIZE)
1147                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1148         } while (offset == PAGE_SIZE);
1149
1150         bprm->p++;
1151         bprm->argc--;
1152         ret = 0;
1153
1154 out:
1155         return ret;
1156 }
1157 EXPORT_SYMBOL(remove_arg_zero);
1158
1159 /*
1160  * cycle the list of binary formats handler, until one recognizes the image
1161  */
1162 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1163 {
1164         unsigned int depth = bprm->recursion_depth;
1165         int try,retval;
1166         struct linux_binfmt *fmt;
1167
1168         retval = security_bprm_check(bprm);
1169         if (retval)
1170                 return retval;
1171
1172         /* kernel module loader fixup */
1173         /* so we don't try to load run modprobe in kernel space. */
1174         set_fs(USER_DS);
1175
1176         retval = audit_bprm(bprm);
1177         if (retval)
1178                 return retval;
1179
1180         retval = -ENOENT;
1181         for (try=0; try<2; try++) {
1182                 read_lock(&binfmt_lock);
1183                 list_for_each_entry(fmt, &formats, lh) {
1184                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1185                         if (!fn)
1186                                 continue;
1187                         if (!try_module_get(fmt->module))
1188                                 continue;
1189                         read_unlock(&binfmt_lock);
1190                         retval = fn(bprm, regs);
1191                         /*
1192                          * Restore the depth counter to its starting value
1193                          * in this call, so we don't have to rely on every
1194                          * load_binary function to restore it on return.
1195                          */
1196                         bprm->recursion_depth = depth;
1197                         if (retval >= 0) {
1198                                 if (depth == 0)
1199                                         tracehook_report_exec(fmt, bprm, regs);
1200                                 put_binfmt(fmt);
1201                                 allow_write_access(bprm->file);
1202                                 if (bprm->file)
1203                                         fput(bprm->file);
1204                                 bprm->file = NULL;
1205                                 current->did_exec = 1;
1206                                 proc_exec_connector(current);
1207                                 return retval;
1208                         }
1209                         read_lock(&binfmt_lock);
1210                         put_binfmt(fmt);
1211                         if (retval != -ENOEXEC || bprm->mm == NULL)
1212                                 break;
1213                         if (!bprm->file) {
1214                                 read_unlock(&binfmt_lock);
1215                                 return retval;
1216                         }
1217                 }
1218                 read_unlock(&binfmt_lock);
1219                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1220                         break;
1221 #ifdef CONFIG_MODULES
1222                 } else {
1223 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1224                         if (printable(bprm->buf[0]) &&
1225                             printable(bprm->buf[1]) &&
1226                             printable(bprm->buf[2]) &&
1227                             printable(bprm->buf[3]))
1228                                 break; /* -ENOEXEC */
1229                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1230 #endif
1231                 }
1232         }
1233         return retval;
1234 }
1235
1236 EXPORT_SYMBOL(search_binary_handler);
1237
1238 void free_bprm(struct linux_binprm *bprm)
1239 {
1240         free_arg_pages(bprm);
1241         if (bprm->cred)
1242                 abort_creds(bprm->cred);
1243         kfree(bprm);
1244 }
1245
1246 /*
1247  * sys_execve() executes a new program.
1248  */
1249 int do_execve(char * filename,
1250         char __user *__user *argv,
1251         char __user *__user *envp,
1252         struct pt_regs * regs)
1253 {
1254         struct linux_binprm *bprm;
1255         struct file *file;
1256         struct files_struct *displaced;
1257         int retval;
1258
1259         retval = unshare_files(&displaced);
1260         if (retval)
1261                 goto out_ret;
1262
1263         retval = -ENOMEM;
1264         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1265         if (!bprm)
1266                 goto out_files;
1267
1268         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1269         if (retval < 0)
1270                 goto out_free;
1271
1272         retval = -ENOMEM;
1273         bprm->cred = prepare_exec_creds();
1274         if (!bprm->cred)
1275                 goto out_unlock;
1276         check_unsafe_exec(bprm);
1277
1278         file = open_exec(filename);
1279         retval = PTR_ERR(file);
1280         if (IS_ERR(file))
1281                 goto out_unlock;
1282
1283         sched_exec();
1284
1285         bprm->file = file;
1286         bprm->filename = filename;
1287         bprm->interp = filename;
1288
1289         retval = bprm_mm_init(bprm);
1290         if (retval)
1291                 goto out_file;
1292
1293         bprm->argc = count(argv, MAX_ARG_STRINGS);
1294         if ((retval = bprm->argc) < 0)
1295                 goto out;
1296
1297         bprm->envc = count(envp, MAX_ARG_STRINGS);
1298         if ((retval = bprm->envc) < 0)
1299                 goto out;
1300
1301         retval = prepare_binprm(bprm);
1302         if (retval < 0)
1303                 goto out;
1304
1305         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1306         if (retval < 0)
1307                 goto out;
1308
1309         bprm->exec = bprm->p;
1310         retval = copy_strings(bprm->envc, envp, bprm);
1311         if (retval < 0)
1312                 goto out;
1313
1314         retval = copy_strings(bprm->argc, argv, bprm);
1315         if (retval < 0)
1316                 goto out;
1317
1318         current->flags &= ~PF_KTHREAD;
1319         retval = search_binary_handler(bprm,regs);
1320         if (retval < 0)
1321                 goto out;
1322
1323         /* execve succeeded */
1324         mutex_unlock(&current->cred_exec_mutex);
1325         acct_update_integrals(current);
1326         free_bprm(bprm);
1327         if (displaced)
1328                 put_files_struct(displaced);
1329         return retval;
1330
1331 out:
1332         if (bprm->mm)
1333                 mmput (bprm->mm);
1334
1335 out_file:
1336         if (bprm->file) {
1337                 allow_write_access(bprm->file);
1338                 fput(bprm->file);
1339         }
1340
1341 out_unlock:
1342         mutex_unlock(&current->cred_exec_mutex);
1343
1344 out_free:
1345         free_bprm(bprm);
1346
1347 out_files:
1348         if (displaced)
1349                 reset_files_struct(displaced);
1350 out_ret:
1351         return retval;
1352 }
1353
1354 int set_binfmt(struct linux_binfmt *new)
1355 {
1356         struct linux_binfmt *old = current->binfmt;
1357
1358         if (new) {
1359                 if (!try_module_get(new->module))
1360                         return -1;
1361         }
1362         current->binfmt = new;
1363         if (old)
1364                 module_put(old->module);
1365         return 0;
1366 }
1367
1368 EXPORT_SYMBOL(set_binfmt);
1369
1370 /* format_corename will inspect the pattern parameter, and output a
1371  * name into corename, which must have space for at least
1372  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1373  */
1374 static int format_corename(char *corename, long signr)
1375 {
1376         const struct cred *cred = current_cred();
1377         const char *pat_ptr = core_pattern;
1378         int ispipe = (*pat_ptr == '|');
1379         char *out_ptr = corename;
1380         char *const out_end = corename + CORENAME_MAX_SIZE;
1381         int rc;
1382         int pid_in_pattern = 0;
1383
1384         /* Repeat as long as we have more pattern to process and more output
1385            space */
1386         while (*pat_ptr) {
1387                 if (*pat_ptr != '%') {
1388                         if (out_ptr == out_end)
1389                                 goto out;
1390                         *out_ptr++ = *pat_ptr++;
1391                 } else {
1392                         switch (*++pat_ptr) {
1393                         case 0:
1394                                 goto out;
1395                         /* Double percent, output one percent */
1396                         case '%':
1397                                 if (out_ptr == out_end)
1398                                         goto out;
1399                                 *out_ptr++ = '%';
1400                                 break;
1401                         /* pid */
1402                         case 'p':
1403                                 pid_in_pattern = 1;
1404                                 rc = snprintf(out_ptr, out_end - out_ptr,
1405                                               "%d", task_tgid_vnr(current));
1406                                 if (rc > out_end - out_ptr)
1407                                         goto out;
1408                                 out_ptr += rc;
1409                                 break;
1410                         /* uid */
1411                         case 'u':
1412                                 rc = snprintf(out_ptr, out_end - out_ptr,
1413                                               "%d", cred->uid);
1414                                 if (rc > out_end - out_ptr)
1415                                         goto out;
1416                                 out_ptr += rc;
1417                                 break;
1418                         /* gid */
1419                         case 'g':
1420                                 rc = snprintf(out_ptr, out_end - out_ptr,
1421                                               "%d", cred->gid);
1422                                 if (rc > out_end - out_ptr)
1423                                         goto out;
1424                                 out_ptr += rc;
1425                                 break;
1426                         /* signal that caused the coredump */
1427                         case 's':
1428                                 rc = snprintf(out_ptr, out_end - out_ptr,
1429                                               "%ld", signr);
1430                                 if (rc > out_end - out_ptr)
1431                                         goto out;
1432                                 out_ptr += rc;
1433                                 break;
1434                         /* UNIX time of coredump */
1435                         case 't': {
1436                                 struct timeval tv;
1437                                 do_gettimeofday(&tv);
1438                                 rc = snprintf(out_ptr, out_end - out_ptr,
1439                                               "%lu", tv.tv_sec);
1440                                 if (rc > out_end - out_ptr)
1441                                         goto out;
1442                                 out_ptr += rc;
1443                                 break;
1444                         }
1445                         /* hostname */
1446                         case 'h':
1447                                 down_read(&uts_sem);
1448                                 rc = snprintf(out_ptr, out_end - out_ptr,
1449                                               "%s", utsname()->nodename);
1450                                 up_read(&uts_sem);
1451                                 if (rc > out_end - out_ptr)
1452                                         goto out;
1453                                 out_ptr += rc;
1454                                 break;
1455                         /* executable */
1456                         case 'e':
1457                                 rc = snprintf(out_ptr, out_end - out_ptr,
1458                                               "%s", current->comm);
1459                                 if (rc > out_end - out_ptr)
1460                                         goto out;
1461                                 out_ptr += rc;
1462                                 break;
1463                         /* core limit size */
1464                         case 'c':
1465                                 rc = snprintf(out_ptr, out_end - out_ptr,
1466                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1467                                 if (rc > out_end - out_ptr)
1468                                         goto out;
1469                                 out_ptr += rc;
1470                                 break;
1471                         default:
1472                                 break;
1473                         }
1474                         ++pat_ptr;
1475                 }
1476         }
1477         /* Backward compatibility with core_uses_pid:
1478          *
1479          * If core_pattern does not include a %p (as is the default)
1480          * and core_uses_pid is set, then .%pid will be appended to
1481          * the filename. Do not do this for piped commands. */
1482         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1483                 rc = snprintf(out_ptr, out_end - out_ptr,
1484                               ".%d", task_tgid_vnr(current));
1485                 if (rc > out_end - out_ptr)
1486                         goto out;
1487                 out_ptr += rc;
1488         }
1489 out:
1490         *out_ptr = 0;
1491         return ispipe;
1492 }
1493
1494 static int zap_process(struct task_struct *start)
1495 {
1496         struct task_struct *t;
1497         int nr = 0;
1498
1499         start->signal->flags = SIGNAL_GROUP_EXIT;
1500         start->signal->group_stop_count = 0;
1501
1502         t = start;
1503         do {
1504                 if (t != current && t->mm) {
1505                         sigaddset(&t->pending.signal, SIGKILL);
1506                         signal_wake_up(t, 1);
1507                         nr++;
1508                 }
1509         } while_each_thread(start, t);
1510
1511         return nr;
1512 }
1513
1514 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1515                                 struct core_state *core_state, int exit_code)
1516 {
1517         struct task_struct *g, *p;
1518         unsigned long flags;
1519         int nr = -EAGAIN;
1520
1521         spin_lock_irq(&tsk->sighand->siglock);
1522         if (!signal_group_exit(tsk->signal)) {
1523                 mm->core_state = core_state;
1524                 tsk->signal->group_exit_code = exit_code;
1525                 nr = zap_process(tsk);
1526         }
1527         spin_unlock_irq(&tsk->sighand->siglock);
1528         if (unlikely(nr < 0))
1529                 return nr;
1530
1531         if (atomic_read(&mm->mm_users) == nr + 1)
1532                 goto done;
1533         /*
1534          * We should find and kill all tasks which use this mm, and we should
1535          * count them correctly into ->nr_threads. We don't take tasklist
1536          * lock, but this is safe wrt:
1537          *
1538          * fork:
1539          *      None of sub-threads can fork after zap_process(leader). All
1540          *      processes which were created before this point should be
1541          *      visible to zap_threads() because copy_process() adds the new
1542          *      process to the tail of init_task.tasks list, and lock/unlock
1543          *      of ->siglock provides a memory barrier.
1544          *
1545          * do_exit:
1546          *      The caller holds mm->mmap_sem. This means that the task which
1547          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1548          *      its ->mm.
1549          *
1550          * de_thread:
1551          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1552          *      we must see either old or new leader, this does not matter.
1553          *      However, it can change p->sighand, so lock_task_sighand(p)
1554          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1555          *      it can't fail.
1556          *
1557          *      Note also that "g" can be the old leader with ->mm == NULL
1558          *      and already unhashed and thus removed from ->thread_group.
1559          *      This is OK, __unhash_process()->list_del_rcu() does not
1560          *      clear the ->next pointer, we will find the new leader via
1561          *      next_thread().
1562          */
1563         rcu_read_lock();
1564         for_each_process(g) {
1565                 if (g == tsk->group_leader)
1566                         continue;
1567                 if (g->flags & PF_KTHREAD)
1568                         continue;
1569                 p = g;
1570                 do {
1571                         if (p->mm) {
1572                                 if (unlikely(p->mm == mm)) {
1573                                         lock_task_sighand(p, &flags);
1574                                         nr += zap_process(p);
1575                                         unlock_task_sighand(p, &flags);
1576                                 }
1577                                 break;
1578                         }
1579                 } while_each_thread(g, p);
1580         }
1581         rcu_read_unlock();
1582 done:
1583         atomic_set(&core_state->nr_threads, nr);
1584         return nr;
1585 }
1586
1587 static int coredump_wait(int exit_code, struct core_state *core_state)
1588 {
1589         struct task_struct *tsk = current;
1590         struct mm_struct *mm = tsk->mm;
1591         struct completion *vfork_done;
1592         int core_waiters;
1593
1594         init_completion(&core_state->startup);
1595         core_state->dumper.task = tsk;
1596         core_state->dumper.next = NULL;
1597         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1598         up_write(&mm->mmap_sem);
1599
1600         if (unlikely(core_waiters < 0))
1601                 goto fail;
1602
1603         /*
1604          * Make sure nobody is waiting for us to release the VM,
1605          * otherwise we can deadlock when we wait on each other
1606          */
1607         vfork_done = tsk->vfork_done;
1608         if (vfork_done) {
1609                 tsk->vfork_done = NULL;
1610                 complete(vfork_done);
1611         }
1612
1613         if (core_waiters)
1614                 wait_for_completion(&core_state->startup);
1615 fail:
1616         return core_waiters;
1617 }
1618
1619 static void coredump_finish(struct mm_struct *mm)
1620 {
1621         struct core_thread *curr, *next;
1622         struct task_struct *task;
1623
1624         next = mm->core_state->dumper.next;
1625         while ((curr = next) != NULL) {
1626                 next = curr->next;
1627                 task = curr->task;
1628                 /*
1629                  * see exit_mm(), curr->task must not see
1630                  * ->task == NULL before we read ->next.
1631                  */
1632                 smp_mb();
1633                 curr->task = NULL;
1634                 wake_up_process(task);
1635         }
1636
1637         mm->core_state = NULL;
1638 }
1639
1640 /*
1641  * set_dumpable converts traditional three-value dumpable to two flags and
1642  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1643  * these bits are not changed atomically.  So get_dumpable can observe the
1644  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1645  * return either old dumpable or new one by paying attention to the order of
1646  * modifying the bits.
1647  *
1648  * dumpable |   mm->flags (binary)
1649  * old  new | initial interim  final
1650  * ---------+-----------------------
1651  *  0    1  |   00      01      01
1652  *  0    2  |   00      10(*)   11
1653  *  1    0  |   01      00      00
1654  *  1    2  |   01      11      11
1655  *  2    0  |   11      10(*)   00
1656  *  2    1  |   11      11      01
1657  *
1658  * (*) get_dumpable regards interim value of 10 as 11.
1659  */
1660 void set_dumpable(struct mm_struct *mm, int value)
1661 {
1662         switch (value) {
1663         case 0:
1664                 clear_bit(MMF_DUMPABLE, &mm->flags);
1665                 smp_wmb();
1666                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1667                 break;
1668         case 1:
1669                 set_bit(MMF_DUMPABLE, &mm->flags);
1670                 smp_wmb();
1671                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1672                 break;
1673         case 2:
1674                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1675                 smp_wmb();
1676                 set_bit(MMF_DUMPABLE, &mm->flags);
1677                 break;
1678         }
1679 }
1680
1681 int get_dumpable(struct mm_struct *mm)
1682 {
1683         int ret;
1684
1685         ret = mm->flags & 0x3;
1686         return (ret >= 2) ? 2 : ret;
1687 }
1688
1689 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1690 {
1691         struct core_state core_state;
1692         char corename[CORENAME_MAX_SIZE + 1];
1693         struct mm_struct *mm = current->mm;
1694         struct linux_binfmt * binfmt;
1695         struct inode * inode;
1696         struct file * file;
1697         const struct cred *old_cred;
1698         struct cred *cred;
1699         int retval = 0;
1700         int flag = 0;
1701         int ispipe = 0;
1702         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1703         char **helper_argv = NULL;
1704         int helper_argc = 0;
1705         char *delimit;
1706
1707         audit_core_dumps(signr);
1708
1709         binfmt = current->binfmt;
1710         if (!binfmt || !binfmt->core_dump)
1711                 goto fail;
1712
1713         cred = prepare_creds();
1714         if (!cred) {
1715                 retval = -ENOMEM;
1716                 goto fail;
1717         }
1718
1719         down_write(&mm->mmap_sem);
1720         /*
1721          * If another thread got here first, or we are not dumpable, bail out.
1722          */
1723         if (mm->core_state || !get_dumpable(mm)) {
1724                 up_write(&mm->mmap_sem);
1725                 put_cred(cred);
1726                 goto fail;
1727         }
1728
1729         /*
1730          *      We cannot trust fsuid as being the "true" uid of the
1731          *      process nor do we know its entire history. We only know it
1732          *      was tainted so we dump it as root in mode 2.
1733          */
1734         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1735                 flag = O_EXCL;          /* Stop rewrite attacks */
1736                 cred->fsuid = 0;        /* Dump root private */
1737         }
1738
1739         retval = coredump_wait(exit_code, &core_state);
1740         if (retval < 0) {
1741                 put_cred(cred);
1742                 goto fail;
1743         }
1744
1745         old_cred = override_creds(cred);
1746
1747         /*
1748          * Clear any false indication of pending signals that might
1749          * be seen by the filesystem code called to write the core file.
1750          */
1751         clear_thread_flag(TIF_SIGPENDING);
1752
1753         /*
1754          * lock_kernel() because format_corename() is controlled by sysctl, which
1755          * uses lock_kernel()
1756          */
1757         lock_kernel();
1758         ispipe = format_corename(corename, signr);
1759         unlock_kernel();
1760         /*
1761          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1762          * to a pipe.  Since we're not writing directly to the filesystem
1763          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1764          * created unless the pipe reader choses to write out the core file
1765          * at which point file size limits and permissions will be imposed
1766          * as it does with any other process
1767          */
1768         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1769                 goto fail_unlock;
1770
1771         if (ispipe) {
1772                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1773                 if (!helper_argv) {
1774                         printk(KERN_WARNING "%s failed to allocate memory\n",
1775                                __func__);
1776                         goto fail_unlock;
1777                 }
1778                 /* Terminate the string before the first option */
1779                 delimit = strchr(corename, ' ');
1780                 if (delimit)
1781                         *delimit = '\0';
1782                 delimit = strrchr(helper_argv[0], '/');
1783                 if (delimit)
1784                         delimit++;
1785                 else
1786                         delimit = helper_argv[0];
1787                 if (!strcmp(delimit, current->comm)) {
1788                         printk(KERN_NOTICE "Recursive core dump detected, "
1789                                         "aborting\n");
1790                         goto fail_unlock;
1791                 }
1792
1793                 core_limit = RLIM_INFINITY;
1794
1795                 /* SIGPIPE can happen, but it's just never processed */
1796                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1797                                 &file)) {
1798                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1799                                corename);
1800                         goto fail_unlock;
1801                 }
1802         } else
1803                 file = filp_open(corename,
1804                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1805                                  0600);
1806         if (IS_ERR(file))
1807                 goto fail_unlock;
1808         inode = file->f_path.dentry->d_inode;
1809         if (inode->i_nlink > 1)
1810                 goto close_fail;        /* multiple links - don't dump */
1811         if (!ispipe && d_unhashed(file->f_path.dentry))
1812                 goto close_fail;
1813
1814         /* AK: actually i see no reason to not allow this for named pipes etc.,
1815            but keep the previous behaviour for now. */
1816         if (!ispipe && !S_ISREG(inode->i_mode))
1817                 goto close_fail;
1818         /*
1819          * Dont allow local users get cute and trick others to coredump
1820          * into their pre-created files:
1821          */
1822         if (inode->i_uid != current_fsuid())
1823                 goto close_fail;
1824         if (!file->f_op)
1825                 goto close_fail;
1826         if (!file->f_op->write)
1827                 goto close_fail;
1828         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1829                 goto close_fail;
1830
1831         retval = binfmt->core_dump(signr, regs, file, core_limit);
1832
1833         if (retval)
1834                 current->signal->group_exit_code |= 0x80;
1835 close_fail:
1836         filp_close(file, NULL);
1837 fail_unlock:
1838         if (helper_argv)
1839                 argv_free(helper_argv);
1840
1841         revert_creds(old_cred);
1842         put_cred(cred);
1843         coredump_finish(mm);
1844 fail:
1845         return;
1846 }