SAFE public projects git trees. - safe/jmp/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 EXPORT_SYMBOL(suid_dumpable);
  68 /* The maximal length of core_pattern is also specified in sysctl.c */
  69
  70 static LIST_HEAD(formats);
  71 static DEFINE_RWLOCK(binfmt_lock);
  72
  73 int register_binfmt(struct linux_binfmt * fmt)
  74 {
  75         if (!fmt)
  76                 return -EINVAL;
  77         write_lock(&binfmt_lock);
  78         list_add(&fmt->lh, &formats);
  79         write_unlock(&binfmt_lock);
  80         return 0;
  81 }
  82
  83 EXPORT_SYMBOL(register_binfmt);
  84
  85 void unregister_binfmt(struct linux_binfmt * fmt)
  86 {
  87         write_lock(&binfmt_lock);
  88         list_del(&fmt->lh);
  89         write_unlock(&binfmt_lock);
  90 }
  91
  92 EXPORT_SYMBOL(unregister_binfmt);
  93
  94 static inline void put_binfmt(struct linux_binfmt * fmt)
  95 {
  96         module_put(fmt->module);
  97 }
  98
  99 /*
 100  * Note that a shared library must be both readable and executable due to
 101  * security reasons.
 102  *
 103  * Also note that we take the address to load from from the file itself.
 104  */
 105 asmlinkage long sys_uselib(const char __user * library)
 106 {
 107         struct file * file;
 108         struct nameidata nd;
 109         int error;
 110
 111         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 112         if (error)
 113                 goto out;
 114
 115         error = -EINVAL;
 116         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 117                 goto exit;
 118
 119         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 120         if (error)
 121                 goto exit;
 122
 123         file = nameidata_to_filp(&nd, O_RDONLY);
 124         error = PTR_ERR(file);
 125         if (IS_ERR(file))
 126                 goto out;
 127
 128         error = -ENOEXEC;
 129         if(file->f_op) {
 130                 struct linux_binfmt * fmt;
 131
 132                 read_lock(&binfmt_lock);
 133                 list_for_each_entry(fmt, &formats, lh) {
 134                         if (!fmt->load_shlib)
 135                                 continue;
 136                         if (!try_module_get(fmt->module))
 137                                 continue;
 138                         read_unlock(&binfmt_lock);
 139                         error = fmt->load_shlib(file);
 140                         read_lock(&binfmt_lock);
 141                         put_binfmt(fmt);
 142                         if (error != -ENOEXEC)
 143                                 break;
 144                 }
 145                 read_unlock(&binfmt_lock);
 146         }
 147         fput(file);
 148 out:
 149         return error;
 150 exit:
 151         release_open_intent(&nd);
 152         path_release(&nd);
 153         goto out;
 154 }
 155
 156 #ifdef CONFIG_MMU
 157
 158 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 159                 int write)
 160 {
 161         struct page *page;
 162         int ret;
 163
 164 #ifdef CONFIG_STACK_GROWSUP
 165         if (write) {
 166                 ret = expand_stack_downwards(bprm->vma, pos);
 167                 if (ret < 0)
 168                         return NULL;
 169         }
 170 #endif
 171         ret = get_user_pages(current, bprm->mm, pos,
 172                         1, write, 1, &page, NULL);
 173         if (ret <= 0)
 174                 return NULL;
 175
 176         if (write) {
 177                 struct rlimit *rlim = current->signal->rlim;
 178                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 179
 180                 /*
 181                  * Limit to 1/4-th the stack size for the argv+env strings.
 182                  * This ensures that:
 183                  *  - the remaining binfmt code will not run out of stack space,
 184                  *  - the program will have a reasonable amount of stack left
 185                  *    to work from.
 186                  */
 187                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 188                         put_page(page);
 189                         return NULL;
 190                 }
 191         }
 192
 193         return page;
 194 }
 195
 196 static void put_arg_page(struct page *page)
 197 {
 198         put_page(page);
 199 }
 200
 201 static void free_arg_page(struct linux_binprm *bprm, int i)
 202 {
 203 }
 204
 205 static void free_arg_pages(struct linux_binprm *bprm)
 206 {
 207 }
 208
 209 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 210                 struct page *page)
 211 {
 212         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 213 }
 214
 215 static int __bprm_mm_init(struct linux_binprm *bprm)
 216 {
 217         int err = -ENOMEM;
 218         struct vm_area_struct *vma = NULL;
 219         struct mm_struct *mm = bprm->mm;
 220
 221         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 222         if (!vma)
 223                 goto err;
 224
 225         down_write(&mm->mmap_sem);
 226         vma->vm_mm = mm;
 227
 228         /*
 229          * Place the stack at the largest stack address the architecture
 230          * supports. Later, we'll move this to an appropriate place. We don't
 231          * use STACK_TOP because that can depend on attributes which aren't
 232          * configured yet.
 233          */
 234         vma->vm_end = STACK_TOP_MAX;
 235         vma->vm_start = vma->vm_end - PAGE_SIZE;
 236
 237         vma->vm_flags = VM_STACK_FLAGS;
 238         vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
 239         err = insert_vm_struct(mm, vma);
 240         if (err) {
 241                 up_write(&mm->mmap_sem);
 242                 goto err;
 243         }
 244
 245         mm->stack_vm = mm->total_vm = 1;
 246         up_write(&mm->mmap_sem);
 247
 248         bprm->p = vma->vm_end - sizeof(void *);
 249
 250         return 0;
 251
 252 err:
 253         if (vma) {
 254                 bprm->vma = NULL;
 255                 kmem_cache_free(vm_area_cachep, vma);
 256         }
 257
 258         return err;
 259 }
 260
 261 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 262 {
 263         return len <= MAX_ARG_STRLEN;
 264 }
 265
 266 #else
 267
 268 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 269                 int write)
 270 {
 271         struct page *page;
 272
 273         page = bprm->page[pos / PAGE_SIZE];
 274         if (!page && write) {
 275                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 276                 if (!page)
 277                         return NULL;
 278                 bprm->page[pos / PAGE_SIZE] = page;
 279         }
 280
 281         return page;
 282 }
 283
 284 static void put_arg_page(struct page *page)
 285 {
 286 }
 287
 288 static void free_arg_page(struct linux_binprm *bprm, int i)
 289 {
 290         if (bprm->page[i]) {
 291                 __free_page(bprm->page[i]);
 292                 bprm->page[i] = NULL;
 293         }
 294 }
 295
 296 static void free_arg_pages(struct linux_binprm *bprm)
 297 {
 298         int i;
 299
 300         for (i = 0; i < MAX_ARG_PAGES; i++)
 301                 free_arg_page(bprm, i);
 302 }
 303
 304 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 305                 struct page *page)
 306 {
 307 }
 308
 309 static int __bprm_mm_init(struct linux_binprm *bprm)
 310 {
 311         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 312         return 0;
 313 }
 314
 315 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 316 {
 317         return len <= bprm->p;
 318 }
 319
 320 #endif /* CONFIG_MMU */
 321
 322 /*
 323  * Create a new mm_struct and populate it with a temporary stack
 324  * vm_area_struct.  We don't have enough context at this point to set the stack
 325  * flags, permissions, and offset, so we use temporary values.  We'll update
 326  * them later in setup_arg_pages().
 327  */
 328 int bprm_mm_init(struct linux_binprm *bprm)
 329 {
 330         int err;
 331         struct mm_struct *mm = NULL;
 332
 333         bprm->mm = mm = mm_alloc();
 334         err = -ENOMEM;
 335         if (!mm)
 336                 goto err;
 337
 338         err = init_new_context(current, mm);
 339         if (err)
 340                 goto err;
 341
 342         err = __bprm_mm_init(bprm);
 343         if (err)
 344                 goto err;
 345
 346         return 0;
 347
 348 err:
 349         if (mm) {
 350                 bprm->mm = NULL;
 351                 mmdrop(mm);
 352         }
 353
 354         return err;
 355 }
 356
 357 /*
 358  * count() counts the number of strings in array ARGV.
 359  */
 360 static int count(char __user * __user * argv, int max)
 361 {
 362         int i = 0;
 363
 364         if (argv != NULL) {
 365                 for (;;) {
 366                         char __user * p;
 367
 368                         if (get_user(p, argv))
 369                                 return -EFAULT;
 370                         if (!p)
 371                                 break;
 372                         argv++;
 373                         if(++i > max)
 374                                 return -E2BIG;
 375                         cond_resched();
 376                 }
 377         }
 378         return i;
 379 }
 380
 381 /*
 382  * 'copy_strings()' copies argument/environment strings from the old
 383  * processes's memory to the new process's stack.  The call to get_user_pages()
 384  * ensures the destination page is created and not swapped out.
 385  */
 386 static int copy_strings(int argc, char __user * __user * argv,
 387                         struct linux_binprm *bprm)
 388 {
 389         struct page *kmapped_page = NULL;
 390         char *kaddr = NULL;
 391         unsigned long kpos = 0;
 392         int ret;
 393
 394         while (argc-- > 0) {
 395                 char __user *str;
 396                 int len;
 397                 unsigned long pos;
 398
 399                 if (get_user(str, argv+argc) ||
 400                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 401                         ret = -EFAULT;
 402                         goto out;
 403                 }
 404
 405                 if (!valid_arg_len(bprm, len)) {
 406                         ret = -E2BIG;
 407                         goto out;
 408                 }
 409
 410                 /* We're going to work our way backwords. */
 411                 pos = bprm->p;
 412                 str += len;
 413                 bprm->p -= len;
 414
 415                 while (len > 0) {
 416                         int offset, bytes_to_copy;
 417
 418                         offset = pos % PAGE_SIZE;
 419                         if (offset == 0)
 420                                 offset = PAGE_SIZE;
 421
 422                         bytes_to_copy = offset;
 423                         if (bytes_to_copy > len)
 424                                 bytes_to_copy = len;
 425
 426                         offset -= bytes_to_copy;
 427                         pos -= bytes_to_copy;
 428                         str -= bytes_to_copy;
 429                         len -= bytes_to_copy;
 430
 431                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 432                                 struct page *page;
 433
 434                                 page = get_arg_page(bprm, pos, 1);
 435                                 if (!page) {
 436                                         ret = -E2BIG;
 437                                         goto out;
 438                                 }
 439
 440                                 if (kmapped_page) {
 441                                         flush_kernel_dcache_page(kmapped_page);
 442                                         kunmap(kmapped_page);
 443                                         put_arg_page(kmapped_page);
 444                                 }
 445                                 kmapped_page = page;
 446                                 kaddr = kmap(kmapped_page);
 447                                 kpos = pos & PAGE_MASK;
 448                                 flush_arg_page(bprm, kpos, kmapped_page);
 449                         }
 450                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 451                                 ret = -EFAULT;
 452                                 goto out;
 453                         }
 454                 }
 455         }
 456         ret = 0;
 457 out:
 458         if (kmapped_page) {
 459                 flush_kernel_dcache_page(kmapped_page);
 460                 kunmap(kmapped_page);
 461                 put_arg_page(kmapped_page);
 462         }
 463         return ret;
 464 }
 465
 466 /*
 467  * Like copy_strings, but get argv and its values from kernel memory.
 468  */
 469 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 470 {
 471         int r;
 472         mm_segment_t oldfs = get_fs();
 473         set_fs(KERNEL_DS);
 474         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 475         set_fs(oldfs);
 476         return r;
 477 }
 478 EXPORT_SYMBOL(copy_strings_kernel);
 479
 480 #ifdef CONFIG_MMU
 481
 482 /*
 483  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 484  * the binfmt code determines where the new stack should reside, we shift it to
 485  * its final location.  The process proceeds as follows:
 486  *
 487  * 1) Use shift to calculate the new vma endpoints.
 488  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 489  *    arguments passed to subsequent functions are consistent.
 490  * 3) Move vma's page tables to the new range.
 491  * 4) Free up any cleared pgd range.
 492  * 5) Shrink the vma to cover only the new range.
 493  */
 494 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 495 {
 496         struct mm_struct *mm = vma->vm_mm;
 497         unsigned long old_start = vma->vm_start;
 498         unsigned long old_end = vma->vm_end;
 499         unsigned long length = old_end - old_start;
 500         unsigned long new_start = old_start - shift;
 501         unsigned long new_end = old_end - shift;
 502         struct mmu_gather *tlb;
 503
 504         BUG_ON(new_start > new_end);
 505
 506         /*
 507          * ensure there are no vmas between where we want to go
 508          * and where we are
 509          */
 510         if (vma != find_vma(mm, new_start))
 511                 return -EFAULT;
 512
 513         /*
 514          * cover the whole range: [new_start, old_end)
 515          */
 516         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 517
 518         /*
 519          * move the page tables downwards, on failure we rely on
 520          * process cleanup to remove whatever mess we made.
 521          */
 522         if (length != move_page_tables(vma, old_start,
 523                                        vma, new_start, length))
 524                 return -ENOMEM;
 525
 526         lru_add_drain();
 527         tlb = tlb_gather_mmu(mm, 0);
 528         if (new_end > old_start) {
 529                 /*
 530                  * when the old and new regions overlap clear from new_end.
 531                  */
 532                 free_pgd_range(&tlb, new_end, old_end, new_end,
 533                         vma->vm_next ? vma->vm_next->vm_start : 0);
 534         } else {
 535                 /*
 536                  * otherwise, clean from old_start; this is done to not touch
 537                  * the address space in [new_end, old_start) some architectures
 538                  * have constraints on va-space that make this illegal (IA64) -
 539                  * for the others its just a little faster.
 540                  */
 541                 free_pgd_range(&tlb, old_start, old_end, new_end,
 542                         vma->vm_next ? vma->vm_next->vm_start : 0);
 543         }
 544         tlb_finish_mmu(tlb, new_end, old_end);
 545
 546         /*
 547          * shrink the vma to just the new range.
 548          */
 549         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 550
 551         return 0;
 552 }
 553
 554 #define EXTRA_STACK_VM_PAGES    20      /* random */
 555
 556 /*
 557  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 558  * the stack is optionally relocated, and some extra space is added.
 559  */
 560 int setup_arg_pages(struct linux_binprm *bprm,
 561                     unsigned long stack_top,
 562                     int executable_stack)
 563 {
 564         unsigned long ret;
 565         unsigned long stack_shift;
 566         struct mm_struct *mm = current->mm;
 567         struct vm_area_struct *vma = bprm->vma;
 568         struct vm_area_struct *prev = NULL;
 569         unsigned long vm_flags;
 570         unsigned long stack_base;
 571
 572 #ifdef CONFIG_STACK_GROWSUP
 573         /* Limit stack size to 1GB */
 574         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 575         if (stack_base > (1 << 30))
 576                 stack_base = 1 << 30;
 577
 578         /* Make sure we didn't let the argument array grow too large. */
 579         if (vma->vm_end - vma->vm_start > stack_base)
 580                 return -ENOMEM;
 581
 582         stack_base = PAGE_ALIGN(stack_top - stack_base);
 583
 584         stack_shift = vma->vm_start - stack_base;
 585         mm->arg_start = bprm->p - stack_shift;
 586         bprm->p = vma->vm_end - stack_shift;
 587 #else
 588         stack_top = arch_align_stack(stack_top);
 589         stack_top = PAGE_ALIGN(stack_top);
 590         stack_shift = vma->vm_end - stack_top;
 591
 592         bprm->p -= stack_shift;
 593         mm->arg_start = bprm->p;
 594 #endif
 595
 596         if (bprm->loader)
 597                 bprm->loader -= stack_shift;
 598         bprm->exec -= stack_shift;
 599
 600         down_write(&mm->mmap_sem);
 601         vm_flags = vma->vm_flags;
 602
 603         /*
 604          * Adjust stack execute permissions; explicitly enable for
 605          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 606          * (arch default) otherwise.
 607          */
 608         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 609                 vm_flags |= VM_EXEC;
 610         else if (executable_stack == EXSTACK_DISABLE_X)
 611                 vm_flags &= ~VM_EXEC;
 612         vm_flags |= mm->def_flags;
 613
 614         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 615                         vm_flags);
 616         if (ret)
 617                 goto out_unlock;
 618         BUG_ON(prev != vma);
 619
 620         /* Move stack pages down in memory. */
 621         if (stack_shift) {
 622                 ret = shift_arg_pages(vma, stack_shift);
 623                 if (ret) {
 624                         up_write(&mm->mmap_sem);
 625                         return ret;
 626                 }
 627         }
 628
 629 #ifdef CONFIG_STACK_GROWSUP
 630         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 631 #else
 632         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 633 #endif
 634         ret = expand_stack(vma, stack_base);
 635         if (ret)
 636                 ret = -EFAULT;
 637
 638 out_unlock:
 639         up_write(&mm->mmap_sem);
 640         return 0;
 641 }
 642 EXPORT_SYMBOL(setup_arg_pages);
 643
 644 #endif /* CONFIG_MMU */
 645
 646 struct file *open_exec(const char *name)
 647 {
 648         struct nameidata nd;
 649         int err;
 650         struct file *file;
 651
 652         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 653         file = ERR_PTR(err);
 654
 655         if (!err) {
 656                 struct inode *inode = nd.dentry->d_inode;
 657                 file = ERR_PTR(-EACCES);
 658                 if (S_ISREG(inode->i_mode)) {
 659                         int err = vfs_permission(&nd, MAY_EXEC);
 660                         file = ERR_PTR(err);
 661                         if (!err) {
 662                                 file = nameidata_to_filp(&nd, O_RDONLY);
 663                                 if (!IS_ERR(file)) {
 664                                         err = deny_write_access(file);
 665                                         if (err) {
 666                                                 fput(file);
 667                                                 file = ERR_PTR(err);
 668                                         }
 669                                 }
 670 out:
 671                                 return file;
 672                         }
 673                 }
 674                 release_open_intent(&nd);
 675                 path_release(&nd);
 676         }
 677         goto out;
 678 }
 679
 680 EXPORT_SYMBOL(open_exec);
 681
 682 int kernel_read(struct file *file, unsigned long offset,
 683         char *addr, unsigned long count)
 684 {
 685         mm_segment_t old_fs;
 686         loff_t pos = offset;
 687         int result;
 688
 689         old_fs = get_fs();
 690         set_fs(get_ds());
 691         /* The cast to a user pointer is valid due to the set_fs() */
 692         result = vfs_read(file, (void __user *)addr, count, &pos);
 693         set_fs(old_fs);
 694         return result;
 695 }
 696
 697 EXPORT_SYMBOL(kernel_read);
 698
 699 static int exec_mmap(struct mm_struct *mm)
 700 {
 701         struct task_struct *tsk;
 702         struct mm_struct * old_mm, *active_mm;
 703
 704         /* Notify parent that we're no longer interested in the old VM */
 705         tsk = current;
 706         old_mm = current->mm;
 707         mm_release(tsk, old_mm);
 708
 709         if (old_mm) {
 710                 /*
 711                  * Make sure that if there is a core dump in progress
 712                  * for the old mm, we get out and die instead of going
 713                  * through with the exec.  We must hold mmap_sem around
 714                  * checking core_waiters and changing tsk->mm.  The
 715                  * core-inducing thread will increment core_waiters for
 716                  * each thread whose ->mm == old_mm.
 717                  */
 718                 down_read(&old_mm->mmap_sem);
 719                 if (unlikely(old_mm->core_waiters)) {
 720                         up_read(&old_mm->mmap_sem);
 721                         return -EINTR;
 722                 }
 723         }
 724         task_lock(tsk);
 725         active_mm = tsk->active_mm;
 726         tsk->mm = mm;
 727         tsk->active_mm = mm;
 728         activate_mm(active_mm, mm);
 729         task_unlock(tsk);
 730         arch_pick_mmap_layout(mm);
 731         if (old_mm) {
 732                 up_read(&old_mm->mmap_sem);
 733                 BUG_ON(active_mm != old_mm);
 734                 mmput(old_mm);
 735                 return 0;
 736         }
 737         mmdrop(active_mm);
 738         return 0;
 739 }
 740
 741 /*
 742  * This function makes sure the current process has its own signal table,
 743  * so that flush_signal_handlers can later reset the handlers without
 744  * disturbing other processes.  (Other processes might share the signal
 745  * table via the CLONE_SIGHAND option to clone().)
 746  */
 747 static int de_thread(struct task_struct *tsk)
 748 {
 749         struct signal_struct *sig = tsk->signal;
 750         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 751         spinlock_t *lock = &oldsighand->siglock;
 752         struct task_struct *leader = NULL;
 753         int count;
 754
 755         /*
 756          * If we don't share sighandlers, then we aren't sharing anything
 757          * and we can just re-use it all.
 758          */
 759         if (atomic_read(&oldsighand->count) <= 1) {
 760                 exit_itimers(sig);
 761                 return 0;
 762         }
 763
 764         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 765         if (!newsighand)
 766                 return -ENOMEM;
 767
 768         if (thread_group_empty(tsk))
 769                 goto no_thread_group;
 770
 771         /*
 772          * Kill all other threads in the thread group.
 773          * We must hold tasklist_lock to call zap_other_threads.
 774          */
 775         read_lock(&tasklist_lock);
 776         spin_lock_irq(lock);
 777         if (sig->flags & SIGNAL_GROUP_EXIT) {
 778                 /*
 779                  * Another group action in progress, just
 780                  * return so that the signal is processed.
 781                  */
 782                 spin_unlock_irq(lock);
 783                 read_unlock(&tasklist_lock);
 784                 kmem_cache_free(sighand_cachep, newsighand);
 785                 return -EAGAIN;
 786         }
 787
 788         /*
 789          * child_reaper ignores SIGKILL, change it now.
 790          * Reparenting needs write_lock on tasklist_lock,
 791          * so it is safe to do it under read_lock.
 792          */
 793         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 794                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 795
 796         zap_other_threads(tsk);
 797         read_unlock(&tasklist_lock);
 798
 799         /*
 800          * Account for the thread group leader hanging around:
 801          */
 802         count = 1;
 803         if (!thread_group_leader(tsk)) {
 804                 count = 2;
 805                 /*
 806                  * The SIGALRM timer survives the exec, but needs to point
 807                  * at us as the new group leader now.  We have a race with
 808                  * a timer firing now getting the old leader, so we need to
 809                  * synchronize with any firing (by calling del_timer_sync)
 810                  * before we can safely let the old group leader die.
 811                  */
 812                 sig->tsk = tsk;
 813                 spin_unlock_irq(lock);
 814                 if (hrtimer_cancel(&sig->real_timer))
 815                         hrtimer_restart(&sig->real_timer);
 816                 spin_lock_irq(lock);
 817         }
 818         while (atomic_read(&sig->count) > count) {
 819                 sig->group_exit_task = tsk;
 820                 sig->notify_count = count;
 821                 __set_current_state(TASK_UNINTERRUPTIBLE);
 822                 spin_unlock_irq(lock);
 823                 schedule();
 824                 spin_lock_irq(lock);
 825         }
 826         sig->group_exit_task = NULL;
 827         sig->notify_count = 0;
 828         spin_unlock_irq(lock);
 829
 830         /*
 831          * At this point all other threads have exited, all we have to
 832          * do is to wait for the thread group leader to become inactive,
 833          * and to assume its PID:
 834          */
 835         if (!thread_group_leader(tsk)) {
 836                 /*
 837                  * Wait for the thread group leader to be a zombie.
 838                  * It should already be zombie at this point, most
 839                  * of the time.
 840                  */
 841                 leader = tsk->group_leader;
 842                 while (leader->exit_state != EXIT_ZOMBIE)
 843                         yield();
 844
 845                 /*
 846                  * The only record we have of the real-time age of a
 847                  * process, regardless of execs it's done, is start_time.
 848                  * All the past CPU time is accumulated in signal_struct
 849                  * from sister threads now dead.  But in this non-leader
 850                  * exec, nothing survives from the original leader thread,
 851                  * whose birth marks the true age of this process now.
 852                  * When we take on its identity by switching to its PID, we
 853                  * also take its birthdate (always earlier than our own).
 854                  */
 855                 tsk->start_time = leader->start_time;
 856
 857                 write_lock_irq(&tasklist_lock);
 858
 859                 BUG_ON(leader->tgid != tsk->tgid);
 860                 BUG_ON(tsk->pid == tsk->tgid);
 861                 /*
 862                  * An exec() starts a new thread group with the
 863                  * TGID of the previous thread group. Rehash the
 864                  * two threads with a switched PID, and release
 865                  * the former thread group leader:
 866                  */
 867
 868                 /* Become a process group leader with the old leader's pid.
 869                  * The old leader becomes a thread of the this thread group.
 870                  * Note: The old leader also uses this pid until release_task
 871                  *       is called.  Odd but simple and correct.
 872                  */
 873                 detach_pid(tsk, PIDTYPE_PID);
 874                 tsk->pid = leader->pid;
 875                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
 876                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 877                 transfer_pid(leader, tsk, PIDTYPE_SID);
 878                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 879
 880                 tsk->group_leader = tsk;
 881                 leader->group_leader = tsk;
 882
 883                 tsk->exit_signal = SIGCHLD;
 884
 885                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 886                 leader->exit_state = EXIT_DEAD;
 887
 888                 write_unlock_irq(&tasklist_lock);
 889         }
 890
 891         /*
 892          * There may be one thread left which is just exiting,
 893          * but it's safe to stop telling the group to kill themselves.
 894          */
 895         sig->flags = 0;
 896
 897 no_thread_group:
 898         exit_itimers(sig);
 899         if (leader)
 900                 release_task(leader);
 901
 902         if (atomic_read(&oldsighand->count) == 1) {
 903                 /*
 904                  * Now that we nuked the rest of the thread group,
 905                  * it turns out we are not sharing sighand any more either.
 906                  * So we can just keep it.
 907                  */
 908                 kmem_cache_free(sighand_cachep, newsighand);
 909         } else {
 910                 /*
 911                  * Move our state over to newsighand and switch it in.
 912                  */
 913                 atomic_set(&newsighand->count, 1);
 914                 memcpy(newsighand->action, oldsighand->action,
 915                        sizeof(newsighand->action));
 916
 917                 write_lock_irq(&tasklist_lock);
 918                 spin_lock(&oldsighand->siglock);
 919                 rcu_assign_pointer(tsk->sighand, newsighand);
 920                 spin_unlock(&oldsighand->siglock);
 921                 write_unlock_irq(&tasklist_lock);
 922
 923                 __cleanup_sighand(oldsighand);
 924         }
 925
 926         BUG_ON(!thread_group_leader(tsk));
 927         return 0;
 928 }
 929
 930 /*
 931  * These functions flushes out all traces of the currently running executable
 932  * so that a new one can be started
 933  */
 934 static void flush_old_files(struct files_struct * files)
 935 {
 936         long j = -1;
 937         struct fdtable *fdt;
 938
 939         spin_lock(&files->file_lock);
 940         for (;;) {
 941                 unsigned long set, i;
 942
 943                 j++;
 944                 i = j * __NFDBITS;
 945                 fdt = files_fdtable(files);
 946                 if (i >= fdt->max_fds)
 947                         break;
 948                 set = fdt->close_on_exec->fds_bits[j];
 949                 if (!set)
 950                         continue;
 951                 fdt->close_on_exec->fds_bits[j] = 0;
 952                 spin_unlock(&files->file_lock);
 953                 for ( ; set ; i++,set >>= 1) {
 954                         if (set & 1) {
 955                                 sys_close(i);
 956                         }
 957                 }
 958                 spin_lock(&files->file_lock);
 959
 960         }
 961         spin_unlock(&files->file_lock);
 962 }
 963
 964 void get_task_comm(char *buf, struct task_struct *tsk)
 965 {
 966         /* buf must be at least sizeof(tsk->comm) in size */
 967         task_lock(tsk);
 968         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 969         task_unlock(tsk);
 970 }
 971
 972 void set_task_comm(struct task_struct *tsk, char *buf)
 973 {
 974         task_lock(tsk);
 975         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 976         task_unlock(tsk);
 977 }
 978
 979 int flush_old_exec(struct linux_binprm * bprm)
 980 {
 981         char * name;
 982         int i, ch, retval;
 983         struct files_struct *files;
 984         char tcomm[sizeof(current->comm)];
 985
 986         /*
 987          * Make sure we have a private signal table and that
 988          * we are unassociated from the previous thread group.
 989          */
 990         retval = de_thread(current);
 991         if (retval)
 992                 goto out;
 993
 994         /*
 995          * Make sure we have private file handles. Ask the
 996          * fork helper to do the work for us and the exit
 997          * helper to do the cleanup of the old one.
 998          */
 999         files = current->files;         /* refcounted so safe to hold */
1000         retval = unshare_files();
1001         if (retval)
1002                 goto out;
1003         /*
1004          * Release all of the old mmap stuff
1005          */
1006         retval = exec_mmap(bprm->mm);
1007         if (retval)
1008                 goto mmap_failed;
1009
1010         bprm->mm = NULL;                /* We're using it now */
1011
1012         /* This is the point of no return */
1013         put_files_struct(files);
1014
1015         current->sas_ss_sp = current->sas_ss_size = 0;
1016
1017         if (current->euid == current->uid && current->egid == current->gid)
1018                 set_dumpable(current->mm, 1);
1019         else
1020                 set_dumpable(current->mm, suid_dumpable);
1021
1022         name = bprm->filename;
1023
1024         /* Copies the binary name from after last slash */
1025         for (i=0; (ch = *(name++)) != '\0';) {
1026                 if (ch == '/')
1027                         i = 0; /* overwrite what we wrote */
1028                 else
1029                         if (i < (sizeof(tcomm) - 1))
1030                                 tcomm[i++] = ch;
1031         }
1032         tcomm[i] = '\0';
1033         set_task_comm(current, tcomm);
1034
1035         current->flags &= ~PF_RANDOMIZE;
1036         flush_thread();
1037
1038         /* Set the new mm task size. We have to do that late because it may
1039          * depend on TIF_32BIT which is only updated in flush_thread() on
1040          * some architectures like powerpc
1041          */
1042         current->mm->task_size = TASK_SIZE;
1043
1044         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1045                 suid_keys(current);
1046                 set_dumpable(current->mm, suid_dumpable);
1047                 current->pdeath_signal = 0;
1048         } else if (file_permission(bprm->file, MAY_READ) ||
1049                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1050                 suid_keys(current);
1051                 set_dumpable(current->mm, suid_dumpable);
1052         }
1053
1054         /* An exec changes our domain. We are no longer part of the thread
1055            group */
1056
1057         current->self_exec_id++;
1058
1059         flush_signal_handlers(current, 0);
1060         flush_old_files(current->files);
1061
1062         return 0;
1063
1064 mmap_failed:
1065         reset_files_struct(current, files);
1066 out:
1067         return retval;
1068 }
1069
1070 EXPORT_SYMBOL(flush_old_exec);
1071
1072 /*
1073  * Fill the binprm structure from the inode.
1074  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1075  */
1076 int prepare_binprm(struct linux_binprm *bprm)
1077 {
1078         int mode;
1079         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1080         int retval;
1081
1082         mode = inode->i_mode;
1083         if (bprm->file->f_op == NULL)
1084                 return -EACCES;
1085
1086         bprm->e_uid = current->euid;
1087         bprm->e_gid = current->egid;
1088
1089         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1090                 /* Set-uid? */
1091                 if (mode & S_ISUID) {
1092                         current->personality &= ~PER_CLEAR_ON_SETID;
1093                         bprm->e_uid = inode->i_uid;
1094                 }
1095
1096                 /* Set-gid? */
1097                 /*
1098                  * If setgid is set but no group execute bit then this
1099                  * is a candidate for mandatory locking, not a setgid
1100                  * executable.
1101                  */
1102                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1103                         current->personality &= ~PER_CLEAR_ON_SETID;
1104                         bprm->e_gid = inode->i_gid;
1105                 }
1106         }
1107
1108         /* fill in binprm security blob */
1109         retval = security_bprm_set(bprm);
1110         if (retval)
1111                 return retval;
1112
1113         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1114         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1115 }
1116
1117 EXPORT_SYMBOL(prepare_binprm);
1118
1119 static int unsafe_exec(struct task_struct *p)
1120 {
1121         int unsafe = 0;
1122         if (p->ptrace & PT_PTRACED) {
1123                 if (p->ptrace & PT_PTRACE_CAP)
1124                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1125                 else
1126                         unsafe |= LSM_UNSAFE_PTRACE;
1127         }
1128         if (atomic_read(&p->fs->count) > 1 ||
1129             atomic_read(&p->files->count) > 1 ||
1130             atomic_read(&p->sighand->count) > 1)
1131                 unsafe |= LSM_UNSAFE_SHARE;
1132
1133         return unsafe;
1134 }
1135
1136 void compute_creds(struct linux_binprm *bprm)
1137 {
1138         int unsafe;
1139
1140         if (bprm->e_uid != current->uid) {
1141                 suid_keys(current);
1142                 current->pdeath_signal = 0;
1143         }
1144         exec_keys(current);
1145
1146         task_lock(current);
1147         unsafe = unsafe_exec(current);
1148         security_bprm_apply_creds(bprm, unsafe);
1149         task_unlock(current);
1150         security_bprm_post_apply_creds(bprm);
1151 }
1152 EXPORT_SYMBOL(compute_creds);
1153
1154 /*
1155  * Arguments are '\0' separated strings found at the location bprm->p
1156  * points to; chop off the first by relocating brpm->p to right after
1157  * the first '\0' encountered.
1158  */
1159 int remove_arg_zero(struct linux_binprm *bprm)
1160 {
1161         int ret = 0;
1162         unsigned long offset;
1163         char *kaddr;
1164         struct page *page;
1165
1166         if (!bprm->argc)
1167                 return 0;
1168
1169         do {
1170                 offset = bprm->p & ~PAGE_MASK;
1171                 page = get_arg_page(bprm, bprm->p, 0);
1172                 if (!page) {
1173                         ret = -EFAULT;
1174                         goto out;
1175                 }
1176                 kaddr = kmap_atomic(page, KM_USER0);
1177
1178                 for (; offset < PAGE_SIZE && kaddr[offset];
1179                                 offset++, bprm->p++)
1180                         ;
1181
1182                 kunmap_atomic(kaddr, KM_USER0);
1183                 put_arg_page(page);
1184
1185                 if (offset == PAGE_SIZE)
1186                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1187         } while (offset == PAGE_SIZE);
1188
1189         bprm->p++;
1190         bprm->argc--;
1191         ret = 0;
1192
1193 out:
1194         return ret;
1195 }
1196 EXPORT_SYMBOL(remove_arg_zero);
1197
1198 /*
1199  * cycle the list of binary formats handler, until one recognizes the image
1200  */
1201 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1202 {
1203         int try,retval;
1204         struct linux_binfmt *fmt;
1205 #ifdef __alpha__
1206         /* handle /sbin/loader.. */
1207         {
1208             struct exec * eh = (struct exec *) bprm->buf;
1209
1210             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1211                 (eh->fh.f_flags & 0x3000) == 0x3000)
1212             {
1213                 struct file * file;
1214                 unsigned long loader;
1215
1216                 allow_write_access(bprm->file);
1217                 fput(bprm->file);
1218                 bprm->file = NULL;
1219
1220                 loader = bprm->vma->vm_end - sizeof(void *);
1221
1222                 file = open_exec("/sbin/loader");
1223                 retval = PTR_ERR(file);
1224                 if (IS_ERR(file))
1225                         return retval;
1226
1227                 /* Remember if the application is TASO.  */
1228                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1229
1230                 bprm->file = file;
1231                 bprm->loader = loader;
1232                 retval = prepare_binprm(bprm);
1233                 if (retval<0)
1234                         return retval;
1235                 /* should call search_binary_handler recursively here,
1236                    but it does not matter */
1237             }
1238         }
1239 #endif
1240         retval = security_bprm_check(bprm);
1241         if (retval)
1242                 return retval;
1243
1244         /* kernel module loader fixup */
1245         /* so we don't try to load run modprobe in kernel space. */
1246         set_fs(USER_DS);
1247
1248         retval = audit_bprm(bprm);
1249         if (retval)
1250                 return retval;
1251
1252         retval = -ENOENT;
1253         for (try=0; try<2; try++) {
1254                 read_lock(&binfmt_lock);
1255                 list_for_each_entry(fmt, &formats, lh) {
1256                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1257                         if (!fn)
1258                                 continue;
1259                         if (!try_module_get(fmt->module))
1260                                 continue;
1261                         read_unlock(&binfmt_lock);
1262                         retval = fn(bprm, regs);
1263                         if (retval >= 0) {
1264                                 put_binfmt(fmt);
1265                                 allow_write_access(bprm->file);
1266                                 if (bprm->file)
1267                                         fput(bprm->file);
1268                                 bprm->file = NULL;
1269                                 current->did_exec = 1;
1270                                 proc_exec_connector(current);
1271                                 return retval;
1272                         }
1273                         read_lock(&binfmt_lock);
1274                         put_binfmt(fmt);
1275                         if (retval != -ENOEXEC || bprm->mm == NULL)
1276                                 break;
1277                         if (!bprm->file) {
1278                                 read_unlock(&binfmt_lock);
1279                                 return retval;
1280                         }
1281                 }
1282                 read_unlock(&binfmt_lock);
1283                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1284                         break;
1285 #ifdef CONFIG_KMOD
1286                 }else{
1287 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1288                         if (printable(bprm->buf[0]) &&
1289                             printable(bprm->buf[1]) &&
1290                             printable(bprm->buf[2]) &&
1291                             printable(bprm->buf[3]))
1292                                 break; /* -ENOEXEC */
1293                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1294 #endif
1295                 }
1296         }
1297         return retval;
1298 }
1299
1300 EXPORT_SYMBOL(search_binary_handler);
1301
1302 /*
1303  * sys_execve() executes a new program.
1304  */
1305 int do_execve(char * filename,
1306         char __user *__user *argv,
1307         char __user *__user *envp,
1308         struct pt_regs * regs)
1309 {
1310         struct linux_binprm *bprm;
1311         struct file *file;
1312         unsigned long env_p;
1313         int retval;
1314
1315         retval = -ENOMEM;
1316         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1317         if (!bprm)
1318                 goto out_ret;
1319
1320         file = open_exec(filename);
1321         retval = PTR_ERR(file);
1322         if (IS_ERR(file))
1323                 goto out_kfree;
1324
1325         sched_exec();
1326
1327         bprm->file = file;
1328         bprm->filename = filename;
1329         bprm->interp = filename;
1330
1331         retval = bprm_mm_init(bprm);
1332         if (retval)
1333                 goto out_file;
1334
1335         bprm->argc = count(argv, MAX_ARG_STRINGS);
1336         if ((retval = bprm->argc) < 0)
1337                 goto out_mm;
1338
1339         bprm->envc = count(envp, MAX_ARG_STRINGS);
1340         if ((retval = bprm->envc) < 0)
1341                 goto out_mm;
1342
1343         retval = security_bprm_alloc(bprm);
1344         if (retval)
1345                 goto out;
1346
1347         retval = prepare_binprm(bprm);
1348         if (retval < 0)
1349                 goto out;
1350
1351         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1352         if (retval < 0)
1353                 goto out;
1354
1355         bprm->exec = bprm->p;
1356         retval = copy_strings(bprm->envc, envp, bprm);
1357         if (retval < 0)
1358                 goto out;
1359
1360         env_p = bprm->p;
1361         retval = copy_strings(bprm->argc, argv, bprm);
1362         if (retval < 0)
1363                 goto out;
1364         bprm->argv_len = env_p - bprm->p;
1365
1366         retval = search_binary_handler(bprm,regs);
1367         if (retval >= 0) {
1368                 /* execve success */
1369                 free_arg_pages(bprm);
1370                 security_bprm_free(bprm);
1371                 acct_update_integrals(current);
1372                 kfree(bprm);
1373                 return retval;
1374         }
1375
1376 out:
1377         free_arg_pages(bprm);
1378         if (bprm->security)
1379                 security_bprm_free(bprm);
1380
1381 out_mm:
1382         if (bprm->mm)
1383                 mmput (bprm->mm);
1384
1385 out_file:
1386         if (bprm->file) {
1387                 allow_write_access(bprm->file);
1388                 fput(bprm->file);
1389         }
1390 out_kfree:
1391         kfree(bprm);
1392
1393 out_ret:
1394         return retval;
1395 }
1396
1397 int set_binfmt(struct linux_binfmt *new)
1398 {
1399         struct linux_binfmt *old = current->binfmt;
1400
1401         if (new) {
1402                 if (!try_module_get(new->module))
1403                         return -1;
1404         }
1405         current->binfmt = new;
1406         if (old)
1407                 module_put(old->module);
1408         return 0;
1409 }
1410
1411 EXPORT_SYMBOL(set_binfmt);
1412
1413 /* format_corename will inspect the pattern parameter, and output a
1414  * name into corename, which must have space for at least
1415  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1416  */
1417 static int format_corename(char *corename, const char *pattern, long signr)
1418 {
1419         const char *pat_ptr = pattern;
1420         char *out_ptr = corename;
1421         char *const out_end = corename + CORENAME_MAX_SIZE;
1422         int rc;
1423         int pid_in_pattern = 0;
1424         int ispipe = 0;
1425
1426         if (*pattern == '|')
1427                 ispipe = 1;
1428
1429         /* Repeat as long as we have more pattern to process and more output
1430            space */
1431         while (*pat_ptr) {
1432                 if (*pat_ptr != '%') {
1433                         if (out_ptr == out_end)
1434                                 goto out;
1435                         *out_ptr++ = *pat_ptr++;
1436                 } else {
1437                         switch (*++pat_ptr) {
1438                         case 0:
1439                                 goto out;
1440                         /* Double percent, output one percent */
1441                         case '%':
1442                                 if (out_ptr == out_end)
1443                                         goto out;
1444                                 *out_ptr++ = '%';
1445                                 break;
1446                         /* pid */
1447                         case 'p':
1448                                 pid_in_pattern = 1;
1449                                 rc = snprintf(out_ptr, out_end - out_ptr,
1450                                               "%d", current->tgid);
1451                                 if (rc > out_end - out_ptr)
1452                                         goto out;
1453                                 out_ptr += rc;
1454                                 break;
1455                         /* uid */
1456                         case 'u':
1457                                 rc = snprintf(out_ptr, out_end - out_ptr,
1458                                               "%d", current->uid);
1459                                 if (rc > out_end - out_ptr)
1460                                         goto out;
1461                                 out_ptr += rc;
1462                                 break;
1463                         /* gid */
1464                         case 'g':
1465                                 rc = snprintf(out_ptr, out_end - out_ptr,
1466                                               "%d", current->gid);
1467                                 if (rc > out_end - out_ptr)
1468                                         goto out;
1469                                 out_ptr += rc;
1470                                 break;
1471                         /* signal that caused the coredump */
1472                         case 's':
1473                                 rc = snprintf(out_ptr, out_end - out_ptr,
1474                                               "%ld", signr);
1475                                 if (rc > out_end - out_ptr)
1476                                         goto out;
1477                                 out_ptr += rc;
1478                                 break;
1479                         /* UNIX time of coredump */
1480                         case 't': {
1481                                 struct timeval tv;
1482                                 do_gettimeofday(&tv);
1483                                 rc = snprintf(out_ptr, out_end - out_ptr,
1484                                               "%lu", tv.tv_sec);
1485                                 if (rc > out_end - out_ptr)
1486                                         goto out;
1487                                 out_ptr += rc;
1488                                 break;
1489                         }
1490                         /* hostname */
1491                         case 'h':
1492                                 down_read(&uts_sem);
1493                                 rc = snprintf(out_ptr, out_end - out_ptr,
1494                                               "%s", utsname()->nodename);
1495                                 up_read(&uts_sem);
1496                                 if (rc > out_end - out_ptr)
1497                                         goto out;
1498                                 out_ptr += rc;
1499                                 break;
1500                         /* executable */
1501                         case 'e':
1502                                 rc = snprintf(out_ptr, out_end - out_ptr,
1503                                               "%s", current->comm);
1504                                 if (rc > out_end - out_ptr)
1505                                         goto out;
1506                                 out_ptr += rc;
1507                                 break;
1508                         /* core limit size */
1509                         case 'c':
1510                                 rc = snprintf(out_ptr, out_end - out_ptr,
1511                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1512                                 if (rc > out_end - out_ptr)
1513                                         goto out;
1514                                 out_ptr += rc;
1515                                 break;
1516                         default:
1517                                 break;
1518                         }
1519                         ++pat_ptr;
1520                 }
1521         }
1522         /* Backward compatibility with core_uses_pid:
1523          *
1524          * If core_pattern does not include a %p (as is the default)
1525          * and core_uses_pid is set, then .%pid will be appended to
1526          * the filename. Do not do this for piped commands. */
1527         if (!ispipe && !pid_in_pattern
1528             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1529                 rc = snprintf(out_ptr, out_end - out_ptr,
1530                               ".%d", current->tgid);
1531                 if (rc > out_end - out_ptr)
1532                         goto out;
1533                 out_ptr += rc;
1534         }
1535 out:
1536         *out_ptr = 0;
1537         return ispipe;
1538 }
1539
1540 static void zap_process(struct task_struct *start)
1541 {
1542         struct task_struct *t;
1543
1544         start->signal->flags = SIGNAL_GROUP_EXIT;
1545         start->signal->group_stop_count = 0;
1546
1547         t = start;
1548         do {
1549                 if (t != current && t->mm) {
1550                         t->mm->core_waiters++;
1551                         sigaddset(&t->pending.signal, SIGKILL);
1552                         signal_wake_up(t, 1);
1553                 }
1554         } while ((t = next_thread(t)) != start);
1555 }
1556
1557 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1558                                 int exit_code)
1559 {
1560         struct task_struct *g, *p;
1561         unsigned long flags;
1562         int err = -EAGAIN;
1563
1564         spin_lock_irq(&tsk->sighand->siglock);
1565         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1566                 tsk->signal->group_exit_code = exit_code;
1567                 zap_process(tsk);
1568                 err = 0;
1569         }
1570         spin_unlock_irq(&tsk->sighand->siglock);
1571         if (err)
1572                 return err;
1573
1574         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1575                 goto done;
1576
1577         rcu_read_lock();
1578         for_each_process(g) {
1579                 if (g == tsk->group_leader)
1580                         continue;
1581
1582                 p = g;
1583                 do {
1584                         if (p->mm) {
1585                                 if (p->mm == mm) {
1586                                         /*
1587                                          * p->sighand can't disappear, but
1588                                          * may be changed by de_thread()
1589                                          */
1590                                         lock_task_sighand(p, &flags);
1591                                         zap_process(p);
1592                                         unlock_task_sighand(p, &flags);
1593                                 }
1594                                 break;
1595                         }
1596                 } while ((p = next_thread(p)) != g);
1597         }
1598         rcu_read_unlock();
1599 done:
1600         return mm->core_waiters;
1601 }
1602
1603 static int coredump_wait(int exit_code)
1604 {
1605         struct task_struct *tsk = current;
1606         struct mm_struct *mm = tsk->mm;
1607         struct completion startup_done;
1608         struct completion *vfork_done;
1609         int core_waiters;
1610
1611         init_completion(&mm->core_done);
1612         init_completion(&startup_done);
1613         mm->core_startup_done = &startup_done;
1614
1615         core_waiters = zap_threads(tsk, mm, exit_code);
1616         up_write(&mm->mmap_sem);
1617
1618         if (unlikely(core_waiters < 0))
1619                 goto fail;
1620
1621         /*
1622          * Make sure nobody is waiting for us to release the VM,
1623          * otherwise we can deadlock when we wait on each other
1624          */
1625         vfork_done = tsk->vfork_done;
1626         if (vfork_done) {
1627                 tsk->vfork_done = NULL;
1628                 complete(vfork_done);
1629         }
1630
1631         if (core_waiters)
1632                 wait_for_completion(&startup_done);
1633 fail:
1634         BUG_ON(mm->core_waiters);
1635         return core_waiters;
1636 }
1637
1638 /*
1639  * set_dumpable converts traditional three-value dumpable to two flags and
1640  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1641  * these bits are not changed atomically.  So get_dumpable can observe the
1642  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1643  * return either old dumpable or new one by paying attention to the order of
1644  * modifying the bits.
1645  *
1646  * dumpable |   mm->flags (binary)
1647  * old  new | initial interim  final
1648  * ---------+-----------------------
1649  *  0    1  |   00      01      01
1650  *  0    2  |   00      10(*)   11
1651  *  1    0  |   01      00      00
1652  *  1    2  |   01      11      11
1653  *  2    0  |   11      10(*)   00
1654  *  2    1  |   11      11      01
1655  *
1656  * (*) get_dumpable regards interim value of 10 as 11.
1657  */
1658 void set_dumpable(struct mm_struct *mm, int value)
1659 {
1660         switch (value) {
1661         case 0:
1662                 clear_bit(MMF_DUMPABLE, &mm->flags);
1663                 smp_wmb();
1664                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1665                 break;
1666         case 1:
1667                 set_bit(MMF_DUMPABLE, &mm->flags);
1668                 smp_wmb();
1669                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1670                 break;
1671         case 2:
1672                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1673                 smp_wmb();
1674                 set_bit(MMF_DUMPABLE, &mm->flags);
1675                 break;
1676         }
1677 }
1678 EXPORT_SYMBOL_GPL(set_dumpable);
1679
1680 int get_dumpable(struct mm_struct *mm)
1681 {
1682         int ret;
1683
1684         ret = mm->flags & 0x3;
1685         return (ret >= 2) ? 2 : ret;
1686 }
1687
1688 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1689 {
1690         char corename[CORENAME_MAX_SIZE + 1];
1691         struct mm_struct *mm = current->mm;
1692         struct linux_binfmt * binfmt;
1693         struct inode * inode;
1694         struct file * file;
1695         int retval = 0;
1696         int fsuid = current->fsuid;
1697         int flag = 0;
1698         int ispipe = 0;
1699         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1700         char **helper_argv = NULL;
1701         int helper_argc = 0;
1702         char *delimit;
1703
1704         audit_core_dumps(signr);
1705
1706         binfmt = current->binfmt;
1707         if (!binfmt || !binfmt->core_dump)
1708                 goto fail;
1709         down_write(&mm->mmap_sem);
1710         if (!get_dumpable(mm)) {
1711                 up_write(&mm->mmap_sem);
1712                 goto fail;
1713         }
1714
1715         /*
1716          *      We cannot trust fsuid as being the "true" uid of the
1717          *      process nor do we know its entire history. We only know it
1718          *      was tainted so we dump it as root in mode 2.
1719          */
1720         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1721                 flag = O_EXCL;          /* Stop rewrite attacks */
1722                 current->fsuid = 0;     /* Dump root private */
1723         }
1724         set_dumpable(mm, 0);
1725
1726         retval = coredump_wait(exit_code);
1727         if (retval < 0)
1728                 goto fail;
1729
1730         /*
1731          * Clear any false indication of pending signals that might
1732          * be seen by the filesystem code called to write the core file.
1733          */
1734         clear_thread_flag(TIF_SIGPENDING);
1735
1736         /*
1737          * lock_kernel() because format_corename() is controlled by sysctl, which
1738          * uses lock_kernel()
1739          */
1740         lock_kernel();
1741         ispipe = format_corename(corename, core_pattern, signr);
1742         unlock_kernel();
1743         /*
1744          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1745          * to a pipe.  Since we're not writing directly to the filesystem
1746          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1747          * created unless the pipe reader choses to write out the core file
1748          * at which point file size limits and permissions will be imposed
1749          * as it does with any other process
1750          */
1751         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1752                 goto fail_unlock;
1753
1754         if (ispipe) {
1755                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1756                 /* Terminate the string before the first option */
1757                 delimit = strchr(corename, ' ');
1758                 if (delimit)
1759                         *delimit = '\0';
1760                 delimit = strrchr(helper_argv[0], '/');
1761                 if (delimit)
1762                         delimit++;
1763                 else
1764                         delimit = helper_argv[0];
1765                 if (!strcmp(delimit, current->comm)) {
1766                         printk(KERN_NOTICE "Recursive core dump detected, "
1767                                         "aborting\n");
1768                         goto fail_unlock;
1769                 }
1770
1771                 core_limit = RLIM_INFINITY;
1772
1773                 /* SIGPIPE can happen, but it's just never processed */
1774                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1775                                 &file)) {
1776                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1777                                corename);
1778                         goto fail_unlock;
1779                 }
1780         } else
1781                 file = filp_open(corename,
1782                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1783                                  0600);
1784         if (IS_ERR(file))
1785                 goto fail_unlock;
1786         inode = file->f_path.dentry->d_inode;
1787         if (inode->i_nlink > 1)
1788                 goto close_fail;        /* multiple links - don't dump */
1789         if (!ispipe && d_unhashed(file->f_path.dentry))
1790                 goto close_fail;
1791
1792         /* AK: actually i see no reason to not allow this for named pipes etc.,
1793            but keep the previous behaviour for now. */
1794         if (!ispipe && !S_ISREG(inode->i_mode))
1795                 goto close_fail;
1796         if (!file->f_op)
1797                 goto close_fail;
1798         if (!file->f_op->write)
1799                 goto close_fail;
1800         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1801                 goto close_fail;
1802
1803         retval = binfmt->core_dump(signr, regs, file, core_limit);
1804
1805         if (retval)
1806                 current->signal->group_exit_code |= 0x80;
1807 close_fail:
1808         filp_close(file, NULL);
1809 fail_unlock:
1810         if (helper_argv)
1811                 argv_free(helper_argv);
1812
1813         current->fsuid = fsuid;
1814         complete_all(&mm->core_done);
1815 fail:
1816         return retval;
1817 }