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/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/perf_counter.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/proc_fs.h>
  47 #include <linux/mount.h>
  48 #include <linux/security.h>
  49 #include <linux/syscalls.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/tracehook.h>
  54 #include <linux/kmod.h>
  55
  56 #include <asm/uaccess.h>
  57 #include <asm/mmu_context.h>
  58 #include <asm/tlb.h>
  59 #include "internal.h"
  60
  61 #ifdef __alpha__
  62 /* for /sbin/loader handling in search_binary_handler() */
  63 #include <linux/a.out.h>
  64 #endif
  65
  66 int core_uses_pid;
  67 char core_pattern[CORENAME_MAX_SIZE] = "core";
  68 int suid_dumpable = 0;
  69
  70 /* The maximal length of core_pattern is also specified in sysctl.c */
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 int register_binfmt(struct linux_binfmt * fmt)
  76 {
  77         if (!fmt)
  78                 return -EINVAL;
  79         write_lock(&binfmt_lock);
  80         list_add(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82         return 0;
  83 }
  84
  85 EXPORT_SYMBOL(register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 asmlinkage long sys_uselib(const char __user * library)
 108 {
 109         struct file *file;
 110         struct nameidata nd;
 111         char *tmp = getname(library);
 112         int error = PTR_ERR(tmp);
 113
 114         if (!IS_ERR(tmp)) {
 115                 error = path_lookup_open(AT_FDCWD, tmp,
 116                                          LOOKUP_FOLLOW, &nd,
 117                                          FMODE_READ|FMODE_EXEC);
 118                 putname(tmp);
 119         }
 120         if (error)
 121                 goto out;
 122
 123         error = -EINVAL;
 124         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 125                 goto exit;
 126
 127         error = -EACCES;
 128         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
 129                 goto exit;
 130
 131         error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
 132         if (error)
 133                 goto exit;
 134
 135         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 136         error = PTR_ERR(file);
 137         if (IS_ERR(file))
 138                 goto out;
 139
 140         error = -ENOEXEC;
 141         if(file->f_op) {
 142                 struct linux_binfmt * fmt;
 143
 144                 read_lock(&binfmt_lock);
 145                 list_for_each_entry(fmt, &formats, lh) {
 146                         if (!fmt->load_shlib)
 147                                 continue;
 148                         if (!try_module_get(fmt->module))
 149                                 continue;
 150                         read_unlock(&binfmt_lock);
 151                         error = fmt->load_shlib(file);
 152                         read_lock(&binfmt_lock);
 153                         put_binfmt(fmt);
 154                         if (error != -ENOEXEC)
 155                                 break;
 156                 }
 157                 read_unlock(&binfmt_lock);
 158         }
 159         fput(file);
 160 out:
 161         return error;
 162 exit:
 163         release_open_intent(&nd);
 164         path_put(&nd.path);
 165         goto out;
 166 }
 167
 168 #ifdef CONFIG_MMU
 169
 170 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 171                 int write)
 172 {
 173         struct page *page;
 174         int ret;
 175
 176 #ifdef CONFIG_STACK_GROWSUP
 177         if (write) {
 178                 ret = expand_stack_downwards(bprm->vma, pos);
 179                 if (ret < 0)
 180                         return NULL;
 181         }
 182 #endif
 183         ret = get_user_pages(current, bprm->mm, pos,
 184                         1, write, 1, &page, NULL);
 185         if (ret <= 0)
 186                 return NULL;
 187
 188         if (write) {
 189                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 190                 struct rlimit *rlim;
 191
 192                 /*
 193                  * We've historically supported up to 32 pages (ARG_MAX)
 194                  * of argument strings even with small stacks
 195                  */
 196                 if (size <= ARG_MAX)
 197                         return page;
 198
 199                 /*
 200                  * Limit to 1/4-th the stack size for the argv+env strings.
 201                  * This ensures that:
 202                  *  - the remaining binfmt code will not run out of stack space,
 203                  *  - the program will have a reasonable amount of stack left
 204                  *    to work from.
 205                  */
 206                 rlim = current->signal->rlim;
 207                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 208                         put_page(page);
 209                         return NULL;
 210                 }
 211         }
 212
 213         return page;
 214 }
 215
 216 static void put_arg_page(struct page *page)
 217 {
 218         put_page(page);
 219 }
 220
 221 static void free_arg_page(struct linux_binprm *bprm, int i)
 222 {
 223 }
 224
 225 static void free_arg_pages(struct linux_binprm *bprm)
 226 {
 227 }
 228
 229 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 230                 struct page *page)
 231 {
 232         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 233 }
 234
 235 static int __bprm_mm_init(struct linux_binprm *bprm)
 236 {
 237         int err = -ENOMEM;
 238         struct vm_area_struct *vma = NULL;
 239         struct mm_struct *mm = bprm->mm;
 240
 241         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 242         if (!vma)
 243                 goto err;
 244
 245         down_write(&mm->mmap_sem);
 246         vma->vm_mm = mm;
 247
 248         /*
 249          * Place the stack at the largest stack address the architecture
 250          * supports. Later, we'll move this to an appropriate place. We don't
 251          * use STACK_TOP because that can depend on attributes which aren't
 252          * configured yet.
 253          */
 254         vma->vm_end = STACK_TOP_MAX;
 255         vma->vm_start = vma->vm_end - PAGE_SIZE;
 256
 257         vma->vm_flags = VM_STACK_FLAGS;
 258         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 259         err = insert_vm_struct(mm, vma);
 260         if (err) {
 261                 up_write(&mm->mmap_sem);
 262                 goto err;
 263         }
 264
 265         mm->stack_vm = mm->total_vm = 1;
 266         up_write(&mm->mmap_sem);
 267
 268         bprm->p = vma->vm_end - sizeof(void *);
 269
 270         return 0;
 271
 272 err:
 273         if (vma) {
 274                 bprm->vma = NULL;
 275                 kmem_cache_free(vm_area_cachep, vma);
 276         }
 277
 278         return err;
 279 }
 280
 281 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 282 {
 283         return len <= MAX_ARG_STRLEN;
 284 }
 285
 286 #else
 287
 288 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 289                 int write)
 290 {
 291         struct page *page;
 292
 293         page = bprm->page[pos / PAGE_SIZE];
 294         if (!page && write) {
 295                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 296                 if (!page)
 297                         return NULL;
 298                 bprm->page[pos / PAGE_SIZE] = page;
 299         }
 300
 301         return page;
 302 }
 303
 304 static void put_arg_page(struct page *page)
 305 {
 306 }
 307
 308 static void free_arg_page(struct linux_binprm *bprm, int i)
 309 {
 310         if (bprm->page[i]) {
 311                 __free_page(bprm->page[i]);
 312                 bprm->page[i] = NULL;
 313         }
 314 }
 315
 316 static void free_arg_pages(struct linux_binprm *bprm)
 317 {
 318         int i;
 319
 320         for (i = 0; i < MAX_ARG_PAGES; i++)
 321                 free_arg_page(bprm, i);
 322 }
 323
 324 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 325                 struct page *page)
 326 {
 327 }
 328
 329 static int __bprm_mm_init(struct linux_binprm *bprm)
 330 {
 331         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 332         return 0;
 333 }
 334
 335 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 336 {
 337         return len <= bprm->p;
 338 }
 339
 340 #endif /* CONFIG_MMU */
 341
 342 /*
 343  * Create a new mm_struct and populate it with a temporary stack
 344  * vm_area_struct.  We don't have enough context at this point to set the stack
 345  * flags, permissions, and offset, so we use temporary values.  We'll update
 346  * them later in setup_arg_pages().
 347  */
 348 int bprm_mm_init(struct linux_binprm *bprm)
 349 {
 350         int err;
 351         struct mm_struct *mm = NULL;
 352
 353         bprm->mm = mm = mm_alloc();
 354         err = -ENOMEM;
 355         if (!mm)
 356                 goto err;
 357
 358         err = init_new_context(current, mm);
 359         if (err)
 360                 goto err;
 361
 362         err = __bprm_mm_init(bprm);
 363         if (err)
 364                 goto err;
 365
 366         return 0;
 367
 368 err:
 369         if (mm) {
 370                 bprm->mm = NULL;
 371                 mmdrop(mm);
 372         }
 373
 374         return err;
 375 }
 376
 377 /*
 378  * count() counts the number of strings in array ARGV.
 379  */
 380 static int count(char __user * __user * argv, int max)
 381 {
 382         int i = 0;
 383
 384         if (argv != NULL) {
 385                 for (;;) {
 386                         char __user * p;
 387
 388                         if (get_user(p, argv))
 389                                 return -EFAULT;
 390                         if (!p)
 391                                 break;
 392                         argv++;
 393                         if (i++ >= max)
 394                                 return -E2BIG;
 395                         cond_resched();
 396                 }
 397         }
 398         return i;
 399 }
 400
 401 /*
 402  * 'copy_strings()' copies argument/environment strings from the old
 403  * processes's memory to the new process's stack.  The call to get_user_pages()
 404  * ensures the destination page is created and not swapped out.
 405  */
 406 static int copy_strings(int argc, char __user * __user * argv,
 407                         struct linux_binprm *bprm)
 408 {
 409         struct page *kmapped_page = NULL;
 410         char *kaddr = NULL;
 411         unsigned long kpos = 0;
 412         int ret;
 413
 414         while (argc-- > 0) {
 415                 char __user *str;
 416                 int len;
 417                 unsigned long pos;
 418
 419                 if (get_user(str, argv+argc) ||
 420                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 421                         ret = -EFAULT;
 422                         goto out;
 423                 }
 424
 425                 if (!valid_arg_len(bprm, len)) {
 426                         ret = -E2BIG;
 427                         goto out;
 428                 }
 429
 430                 /* We're going to work our way backwords. */
 431                 pos = bprm->p;
 432                 str += len;
 433                 bprm->p -= len;
 434
 435                 while (len > 0) {
 436                         int offset, bytes_to_copy;
 437
 438                         offset = pos % PAGE_SIZE;
 439                         if (offset == 0)
 440                                 offset = PAGE_SIZE;
 441
 442                         bytes_to_copy = offset;
 443                         if (bytes_to_copy > len)
 444                                 bytes_to_copy = len;
 445
 446                         offset -= bytes_to_copy;
 447                         pos -= bytes_to_copy;
 448                         str -= bytes_to_copy;
 449                         len -= bytes_to_copy;
 450
 451                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 452                                 struct page *page;
 453
 454                                 page = get_arg_page(bprm, pos, 1);
 455                                 if (!page) {
 456                                         ret = -E2BIG;
 457                                         goto out;
 458                                 }
 459
 460                                 if (kmapped_page) {
 461                                         flush_kernel_dcache_page(kmapped_page);
 462                                         kunmap(kmapped_page);
 463                                         put_arg_page(kmapped_page);
 464                                 }
 465                                 kmapped_page = page;
 466                                 kaddr = kmap(kmapped_page);
 467                                 kpos = pos & PAGE_MASK;
 468                                 flush_arg_page(bprm, kpos, kmapped_page);
 469                         }
 470                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 471                                 ret = -EFAULT;
 472                                 goto out;
 473                         }
 474                 }
 475         }
 476         ret = 0;
 477 out:
 478         if (kmapped_page) {
 479                 flush_kernel_dcache_page(kmapped_page);
 480                 kunmap(kmapped_page);
 481                 put_arg_page(kmapped_page);
 482         }
 483         return ret;
 484 }
 485
 486 /*
 487  * Like copy_strings, but get argv and its values from kernel memory.
 488  */
 489 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 490 {
 491         int r;
 492         mm_segment_t oldfs = get_fs();
 493         set_fs(KERNEL_DS);
 494         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 495         set_fs(oldfs);
 496         return r;
 497 }
 498 EXPORT_SYMBOL(copy_strings_kernel);
 499
 500 #ifdef CONFIG_MMU
 501
 502 /*
 503  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 504  * the binfmt code determines where the new stack should reside, we shift it to
 505  * its final location.  The process proceeds as follows:
 506  *
 507  * 1) Use shift to calculate the new vma endpoints.
 508  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 509  *    arguments passed to subsequent functions are consistent.
 510  * 3) Move vma's page tables to the new range.
 511  * 4) Free up any cleared pgd range.
 512  * 5) Shrink the vma to cover only the new range.
 513  */
 514 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 515 {
 516         struct mm_struct *mm = vma->vm_mm;
 517         unsigned long old_start = vma->vm_start;
 518         unsigned long old_end = vma->vm_end;
 519         unsigned long length = old_end - old_start;
 520         unsigned long new_start = old_start - shift;
 521         unsigned long new_end = old_end - shift;
 522         struct mmu_gather *tlb;
 523
 524         BUG_ON(new_start > new_end);
 525
 526         /*
 527          * ensure there are no vmas between where we want to go
 528          * and where we are
 529          */
 530         if (vma != find_vma(mm, new_start))
 531                 return -EFAULT;
 532
 533         /*
 534          * cover the whole range: [new_start, old_end)
 535          */
 536         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 537
 538         /*
 539          * move the page tables downwards, on failure we rely on
 540          * process cleanup to remove whatever mess we made.
 541          */
 542         if (length != move_page_tables(vma, old_start,
 543                                        vma, new_start, length))
 544                 return -ENOMEM;
 545
 546         lru_add_drain();
 547         tlb = tlb_gather_mmu(mm, 0);
 548         if (new_end > old_start) {
 549                 /*
 550                  * when the old and new regions overlap clear from new_end.
 551                  */
 552                 free_pgd_range(tlb, new_end, old_end, new_end,
 553                         vma->vm_next ? vma->vm_next->vm_start : 0);
 554         } else {
 555                 /*
 556                  * otherwise, clean from old_start; this is done to not touch
 557                  * the address space in [new_end, old_start) some architectures
 558                  * have constraints on va-space that make this illegal (IA64) -
 559                  * for the others its just a little faster.
 560                  */
 561                 free_pgd_range(tlb, old_start, old_end, new_end,
 562                         vma->vm_next ? vma->vm_next->vm_start : 0);
 563         }
 564         tlb_finish_mmu(tlb, new_end, old_end);
 565
 566         /*
 567          * shrink the vma to just the new range.
 568          */
 569         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 570
 571         return 0;
 572 }
 573
 574 #define EXTRA_STACK_VM_PAGES    20      /* random */
 575
 576 /*
 577  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 578  * the stack is optionally relocated, and some extra space is added.
 579  */
 580 int setup_arg_pages(struct linux_binprm *bprm,
 581                     unsigned long stack_top,
 582                     int executable_stack)
 583 {
 584         unsigned long ret;
 585         unsigned long stack_shift;
 586         struct mm_struct *mm = current->mm;
 587         struct vm_area_struct *vma = bprm->vma;
 588         struct vm_area_struct *prev = NULL;
 589         unsigned long vm_flags;
 590         unsigned long stack_base;
 591
 592 #ifdef CONFIG_STACK_GROWSUP
 593         /* Limit stack size to 1GB */
 594         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 595         if (stack_base > (1 << 30))
 596                 stack_base = 1 << 30;
 597
 598         /* Make sure we didn't let the argument array grow too large. */
 599         if (vma->vm_end - vma->vm_start > stack_base)
 600                 return -ENOMEM;
 601
 602         stack_base = PAGE_ALIGN(stack_top - stack_base);
 603
 604         stack_shift = vma->vm_start - stack_base;
 605         mm->arg_start = bprm->p - stack_shift;
 606         bprm->p = vma->vm_end - stack_shift;
 607 #else
 608         stack_top = arch_align_stack(stack_top);
 609         stack_top = PAGE_ALIGN(stack_top);
 610         stack_shift = vma->vm_end - stack_top;
 611
 612         bprm->p -= stack_shift;
 613         mm->arg_start = bprm->p;
 614 #endif
 615
 616         if (bprm->loader)
 617                 bprm->loader -= stack_shift;
 618         bprm->exec -= stack_shift;
 619
 620         down_write(&mm->mmap_sem);
 621         vm_flags = VM_STACK_FLAGS;
 622
 623         /*
 624          * Adjust stack execute permissions; explicitly enable for
 625          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 626          * (arch default) otherwise.
 627          */
 628         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 629                 vm_flags |= VM_EXEC;
 630         else if (executable_stack == EXSTACK_DISABLE_X)
 631                 vm_flags &= ~VM_EXEC;
 632         vm_flags |= mm->def_flags;
 633
 634         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 635                         vm_flags);
 636         if (ret)
 637                 goto out_unlock;
 638         BUG_ON(prev != vma);
 639
 640         /* Move stack pages down in memory. */
 641         if (stack_shift) {
 642                 ret = shift_arg_pages(vma, stack_shift);
 643                 if (ret) {
 644                         up_write(&mm->mmap_sem);
 645                         return ret;
 646                 }
 647         }
 648
 649 #ifdef CONFIG_STACK_GROWSUP
 650         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 651 #else
 652         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 653 #endif
 654         ret = expand_stack(vma, stack_base);
 655         if (ret)
 656                 ret = -EFAULT;
 657
 658 out_unlock:
 659         up_write(&mm->mmap_sem);
 660         return 0;
 661 }
 662 EXPORT_SYMBOL(setup_arg_pages);
 663
 664 #endif /* CONFIG_MMU */
 665
 666 struct file *open_exec(const char *name)
 667 {
 668         struct nameidata nd;
 669         struct file *file;
 670         int err;
 671
 672         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
 673                                 FMODE_READ|FMODE_EXEC);
 674         if (err)
 675                 goto out;
 676
 677         err = -EACCES;
 678         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 679                 goto out_path_put;
 680
 681         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
 682                 goto out_path_put;
 683
 684         err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
 685         if (err)
 686                 goto out_path_put;
 687
 688         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 689         if (IS_ERR(file))
 690                 return file;
 691
 692         err = deny_write_access(file);
 693         if (err) {
 694                 fput(file);
 695                 goto out;
 696         }
 697
 698         return file;
 699
 700  out_path_put:
 701         release_open_intent(&nd);
 702         path_put(&nd.path);
 703  out:
 704         return ERR_PTR(err);
 705 }
 706 EXPORT_SYMBOL(open_exec);
 707
 708 int kernel_read(struct file *file, unsigned long offset,
 709         char *addr, unsigned long count)
 710 {
 711         mm_segment_t old_fs;
 712         loff_t pos = offset;
 713         int result;
 714
 715         old_fs = get_fs();
 716         set_fs(get_ds());
 717         /* The cast to a user pointer is valid due to the set_fs() */
 718         result = vfs_read(file, (void __user *)addr, count, &pos);
 719         set_fs(old_fs);
 720         return result;
 721 }
 722
 723 EXPORT_SYMBOL(kernel_read);
 724
 725 static int exec_mmap(struct mm_struct *mm)
 726 {
 727         struct task_struct *tsk;
 728         struct mm_struct * old_mm, *active_mm;
 729
 730         /* Notify parent that we're no longer interested in the old VM */
 731         tsk = current;
 732         old_mm = current->mm;
 733         mm_release(tsk, old_mm);
 734
 735         if (old_mm) {
 736                 /*
 737                  * Make sure that if there is a core dump in progress
 738                  * for the old mm, we get out and die instead of going
 739                  * through with the exec.  We must hold mmap_sem around
 740                  * checking core_state and changing tsk->mm.
 741                  */
 742                 down_read(&old_mm->mmap_sem);
 743                 if (unlikely(old_mm->core_state)) {
 744                         up_read(&old_mm->mmap_sem);
 745                         return -EINTR;
 746                 }
 747         }
 748         task_lock(tsk);
 749         active_mm = tsk->active_mm;
 750         tsk->mm = mm;
 751         tsk->active_mm = mm;
 752         activate_mm(active_mm, mm);
 753         task_unlock(tsk);
 754         arch_pick_mmap_layout(mm);
 755         if (old_mm) {
 756                 up_read(&old_mm->mmap_sem);
 757                 BUG_ON(active_mm != old_mm);
 758                 mm_update_next_owner(old_mm);
 759                 mmput(old_mm);
 760                 return 0;
 761         }
 762         mmdrop(active_mm);
 763         return 0;
 764 }
 765
 766 /*
 767  * This function makes sure the current process has its own signal table,
 768  * so that flush_signal_handlers can later reset the handlers without
 769  * disturbing other processes.  (Other processes might share the signal
 770  * table via the CLONE_SIGHAND option to clone().)
 771  */
 772 static int de_thread(struct task_struct *tsk)
 773 {
 774         struct signal_struct *sig = tsk->signal;
 775         struct sighand_struct *oldsighand = tsk->sighand;
 776         spinlock_t *lock = &oldsighand->siglock;
 777         struct task_struct *leader = NULL;
 778         int count;
 779
 780         if (thread_group_empty(tsk))
 781                 goto no_thread_group;
 782
 783         /*
 784          * Kill all other threads in the thread group.
 785          */
 786         spin_lock_irq(lock);
 787         if (signal_group_exit(sig)) {
 788                 /*
 789                  * Another group action in progress, just
 790                  * return so that the signal is processed.
 791                  */
 792                 spin_unlock_irq(lock);
 793                 return -EAGAIN;
 794         }
 795         sig->group_exit_task = tsk;
 796         zap_other_threads(tsk);
 797
 798         /* Account for the thread group leader hanging around: */
 799         count = thread_group_leader(tsk) ? 1 : 2;
 800         sig->notify_count = count;
 801         while (atomic_read(&sig->count) > count) {
 802                 __set_current_state(TASK_UNINTERRUPTIBLE);
 803                 spin_unlock_irq(lock);
 804                 schedule();
 805                 spin_lock_irq(lock);
 806         }
 807         spin_unlock_irq(lock);
 808
 809         /*
 810          * At this point all other threads have exited, all we have to
 811          * do is to wait for the thread group leader to become inactive,
 812          * and to assume its PID:
 813          */
 814         if (!thread_group_leader(tsk)) {
 815                 leader = tsk->group_leader;
 816
 817                 sig->notify_count = -1; /* for exit_notify() */
 818                 for (;;) {
 819                         write_lock_irq(&tasklist_lock);
 820                         if (likely(leader->exit_state))
 821                                 break;
 822                         __set_current_state(TASK_UNINTERRUPTIBLE);
 823                         write_unlock_irq(&tasklist_lock);
 824                         schedule();
 825                 }
 826
 827                 /*
 828                  * The only record we have of the real-time age of a
 829                  * process, regardless of execs it's done, is start_time.
 830                  * All the past CPU time is accumulated in signal_struct
 831                  * from sister threads now dead.  But in this non-leader
 832                  * exec, nothing survives from the original leader thread,
 833                  * whose birth marks the true age of this process now.
 834                  * When we take on its identity by switching to its PID, we
 835                  * also take its birthdate (always earlier than our own).
 836                  */
 837                 tsk->start_time = leader->start_time;
 838
 839                 BUG_ON(!same_thread_group(leader, tsk));
 840                 BUG_ON(has_group_leader_pid(tsk));
 841                 /*
 842                  * An exec() starts a new thread group with the
 843                  * TGID of the previous thread group. Rehash the
 844                  * two threads with a switched PID, and release
 845                  * the former thread group leader:
 846                  */
 847
 848                 /* Become a process group leader with the old leader's pid.
 849                  * The old leader becomes a thread of the this thread group.
 850                  * Note: The old leader also uses this pid until release_task
 851                  *       is called.  Odd but simple and correct.
 852                  */
 853                 detach_pid(tsk, PIDTYPE_PID);
 854                 tsk->pid = leader->pid;
 855                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 856                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 857                 transfer_pid(leader, tsk, PIDTYPE_SID);
 858                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 859
 860                 tsk->group_leader = tsk;
 861                 leader->group_leader = tsk;
 862
 863                 tsk->exit_signal = SIGCHLD;
 864
 865                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 866                 leader->exit_state = EXIT_DEAD;
 867
 868                 write_unlock_irq(&tasklist_lock);
 869         }
 870
 871         sig->group_exit_task = NULL;
 872         sig->notify_count = 0;
 873
 874 no_thread_group:
 875         exit_itimers(sig);
 876         flush_itimer_signals();
 877         if (leader)
 878                 release_task(leader);
 879
 880         if (atomic_read(&oldsighand->count) != 1) {
 881                 struct sighand_struct *newsighand;
 882                 /*
 883                  * This ->sighand is shared with the CLONE_SIGHAND
 884                  * but not CLONE_THREAD task, switch to the new one.
 885                  */
 886                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 887                 if (!newsighand)
 888                         return -ENOMEM;
 889
 890                 atomic_set(&newsighand->count, 1);
 891                 memcpy(newsighand->action, oldsighand->action,
 892                        sizeof(newsighand->action));
 893
 894                 write_lock_irq(&tasklist_lock);
 895                 spin_lock(&oldsighand->siglock);
 896                 rcu_assign_pointer(tsk->sighand, newsighand);
 897                 spin_unlock(&oldsighand->siglock);
 898                 write_unlock_irq(&tasklist_lock);
 899
 900                 __cleanup_sighand(oldsighand);
 901         }
 902
 903         BUG_ON(!thread_group_leader(tsk));
 904         return 0;
 905 }
 906
 907 /*
 908  * These functions flushes out all traces of the currently running executable
 909  * so that a new one can be started
 910  */
 911 static void flush_old_files(struct files_struct * files)
 912 {
 913         long j = -1;
 914         struct fdtable *fdt;
 915
 916         spin_lock(&files->file_lock);
 917         for (;;) {
 918                 unsigned long set, i;
 919
 920                 j++;
 921                 i = j * __NFDBITS;
 922                 fdt = files_fdtable(files);
 923                 if (i >= fdt->max_fds)
 924                         break;
 925                 set = fdt->close_on_exec->fds_bits[j];
 926                 if (!set)
 927                         continue;
 928                 fdt->close_on_exec->fds_bits[j] = 0;
 929                 spin_unlock(&files->file_lock);
 930                 for ( ; set ; i++,set >>= 1) {
 931                         if (set & 1) {
 932                                 sys_close(i);
 933                         }
 934                 }
 935                 spin_lock(&files->file_lock);
 936
 937         }
 938         spin_unlock(&files->file_lock);
 939 }
 940
 941 char *get_task_comm(char *buf, struct task_struct *tsk)
 942 {
 943         /* buf must be at least sizeof(tsk->comm) in size */
 944         task_lock(tsk);
 945         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 946         task_unlock(tsk);
 947         return buf;
 948 }
 949
 950 void set_task_comm(struct task_struct *tsk, char *buf)
 951 {
 952         task_lock(tsk);
 953         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 954         task_unlock(tsk);
 955 }
 956
 957 int flush_old_exec(struct linux_binprm * bprm)
 958 {
 959         char * name;
 960         int i, ch, retval;
 961         char tcomm[sizeof(current->comm)];
 962
 963         /*
 964          * Make sure we have a private signal table and that
 965          * we are unassociated from the previous thread group.
 966          */
 967         retval = de_thread(current);
 968         if (retval)
 969                 goto out;
 970
 971         set_mm_exe_file(bprm->mm, bprm->file);
 972
 973         /*
 974          * Release all of the old mmap stuff
 975          */
 976         retval = exec_mmap(bprm->mm);
 977         if (retval)
 978                 goto out;
 979
 980         bprm->mm = NULL;                /* We're using it now */
 981
 982         /* This is the point of no return */
 983         current->sas_ss_sp = current->sas_ss_size = 0;
 984
 985         if (current_euid() == current_uid() && current_egid() == current_gid())
 986                 set_dumpable(current->mm, 1);
 987         else
 988                 set_dumpable(current->mm, suid_dumpable);
 989
 990         name = bprm->filename;
 991
 992         /* Copies the binary name from after last slash */
 993         for (i=0; (ch = *(name++)) != '\0';) {
 994                 if (ch == '/')
 995                         i = 0; /* overwrite what we wrote */
 996                 else
 997                         if (i < (sizeof(tcomm) - 1))
 998                                 tcomm[i++] = ch;
 999         }
1000         tcomm[i] = '\0';
1001         set_task_comm(current, tcomm);
1002
1003         current->flags &= ~PF_RANDOMIZE;
1004         flush_thread();
1005
1006         /* Set the new mm task size. We have to do that late because it may
1007          * depend on TIF_32BIT which is only updated in flush_thread() on
1008          * some architectures like powerpc
1009          */
1010         current->mm->task_size = TASK_SIZE;
1011
1012         /* install the new credentials */
1013         if (bprm->cred->uid != current_euid() ||
1014             bprm->cred->gid != current_egid()) {
1015                 current->pdeath_signal = 0;
1016         } else if (file_permission(bprm->file, MAY_READ) ||
1017                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1018                 set_dumpable(current->mm, suid_dumpable);
1019         }
1020
1021         current->personality &= ~bprm->per_clear;
1022
1023         /*
1024          * Flush performance counters when crossing a
1025          * security domain:
1026          */
1027         if (!get_dumpable(current->mm))
1028                 perf_counter_exit_task(current);
1029
1030         /* An exec changes our domain. We are no longer part of the thread
1031            group */
1032
1033         current->self_exec_id++;
1034
1035         flush_signal_handlers(current, 0);
1036         flush_old_files(current->files);
1037
1038         return 0;
1039
1040 out:
1041         return retval;
1042 }
1043
1044 EXPORT_SYMBOL(flush_old_exec);
1045
1046 /*
1047  * install the new credentials for this executable
1048  */
1049 void install_exec_creds(struct linux_binprm *bprm)
1050 {
1051         security_bprm_committing_creds(bprm);
1052
1053         commit_creds(bprm->cred);
1054         bprm->cred = NULL;
1055
1056         /* cred_exec_mutex must be held at least to this point to prevent
1057          * ptrace_attach() from altering our determination of the task's
1058          * credentials; any time after this it may be unlocked */
1059
1060         security_bprm_committed_creds(bprm);
1061 }
1062 EXPORT_SYMBOL(install_exec_creds);
1063
1064 /*
1065  * determine how safe it is to execute the proposed program
1066  * - the caller must hold current->cred_exec_mutex to protect against
1067  *   PTRACE_ATTACH
1068  */
1069 void check_unsafe_exec(struct linux_binprm *bprm)
1070 {
1071         struct task_struct *p = current;
1072
1073         bprm->unsafe = tracehook_unsafe_exec(p);
1074
1075         if (atomic_read(&p->fs->count) > 1 ||
1076             atomic_read(&p->files->count) > 1 ||
1077             atomic_read(&p->sighand->count) > 1)
1078                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1079 }
1080
1081 /*
1082  * Fill the binprm structure from the inode.
1083  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1084  *
1085  * This may be called multiple times for binary chains (scripts for example).
1086  */
1087 int prepare_binprm(struct linux_binprm *bprm)
1088 {
1089         umode_t mode;
1090         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1091         int retval;
1092
1093         mode = inode->i_mode;
1094         if (bprm->file->f_op == NULL)
1095                 return -EACCES;
1096
1097         /* clear any previous set[ug]id data from a previous binary */
1098         bprm->cred->euid = current_euid();
1099         bprm->cred->egid = current_egid();
1100
1101         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1102                 /* Set-uid? */
1103                 if (mode & S_ISUID) {
1104                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1105                         bprm->cred->euid = inode->i_uid;
1106                 }
1107
1108                 /* Set-gid? */
1109                 /*
1110                  * If setgid is set but no group execute bit then this
1111                  * is a candidate for mandatory locking, not a setgid
1112                  * executable.
1113                  */
1114                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1115                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1116                         bprm->cred->egid = inode->i_gid;
1117                 }
1118         }
1119
1120         /* fill in binprm security blob */
1121         retval = security_bprm_set_creds(bprm);
1122         if (retval)
1123                 return retval;
1124         bprm->cred_prepared = 1;
1125
1126         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1127         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1128 }
1129
1130 EXPORT_SYMBOL(prepare_binprm);
1131
1132 /*
1133  * Arguments are '\0' separated strings found at the location bprm->p
1134  * points to; chop off the first by relocating brpm->p to right after
1135  * the first '\0' encountered.
1136  */
1137 int remove_arg_zero(struct linux_binprm *bprm)
1138 {
1139         int ret = 0;
1140         unsigned long offset;
1141         char *kaddr;
1142         struct page *page;
1143
1144         if (!bprm->argc)
1145                 return 0;
1146
1147         do {
1148                 offset = bprm->p & ~PAGE_MASK;
1149                 page = get_arg_page(bprm, bprm->p, 0);
1150                 if (!page) {
1151                         ret = -EFAULT;
1152                         goto out;
1153                 }
1154                 kaddr = kmap_atomic(page, KM_USER0);
1155
1156                 for (; offset < PAGE_SIZE && kaddr[offset];
1157                                 offset++, bprm->p++)
1158                         ;
1159
1160                 kunmap_atomic(kaddr, KM_USER0);
1161                 put_arg_page(page);
1162
1163                 if (offset == PAGE_SIZE)
1164                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1165         } while (offset == PAGE_SIZE);
1166
1167         bprm->p++;
1168         bprm->argc--;
1169         ret = 0;
1170
1171 out:
1172         return ret;
1173 }
1174 EXPORT_SYMBOL(remove_arg_zero);
1175
1176 /*
1177  * cycle the list of binary formats handler, until one recognizes the image
1178  */
1179 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1180 {
1181         unsigned int depth = bprm->recursion_depth;
1182         int try,retval;
1183         struct linux_binfmt *fmt;
1184 #ifdef __alpha__
1185         /* handle /sbin/loader.. */
1186         {
1187             struct exec * eh = (struct exec *) bprm->buf;
1188
1189             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1190                 (eh->fh.f_flags & 0x3000) == 0x3000)
1191             {
1192                 struct file * file;
1193                 unsigned long loader;
1194
1195                 allow_write_access(bprm->file);
1196                 fput(bprm->file);
1197                 bprm->file = NULL;
1198
1199                 loader = bprm->vma->vm_end - sizeof(void *);
1200
1201                 file = open_exec("/sbin/loader");
1202                 retval = PTR_ERR(file);
1203                 if (IS_ERR(file))
1204                         return retval;
1205
1206                 /* Remember if the application is TASO.  */
1207                 bprm->taso = eh->ah.entry < 0x100000000UL;
1208
1209                 bprm->file = file;
1210                 bprm->loader = loader;
1211                 retval = prepare_binprm(bprm);
1212                 if (retval<0)
1213                         return retval;
1214                 /* should call search_binary_handler recursively here,
1215                    but it does not matter */
1216             }
1217         }
1218 #endif
1219         retval = security_bprm_check(bprm);
1220         if (retval)
1221                 return retval;
1222
1223         /* kernel module loader fixup */
1224         /* so we don't try to load run modprobe in kernel space. */
1225         set_fs(USER_DS);
1226
1227         retval = audit_bprm(bprm);
1228         if (retval)
1229                 return retval;
1230
1231         retval = -ENOENT;
1232         for (try=0; try<2; try++) {
1233                 read_lock(&binfmt_lock);
1234                 list_for_each_entry(fmt, &formats, lh) {
1235                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1236                         if (!fn)
1237                                 continue;
1238                         if (!try_module_get(fmt->module))
1239                                 continue;
1240                         read_unlock(&binfmt_lock);
1241                         retval = fn(bprm, regs);
1242                         /*
1243                          * Restore the depth counter to its starting value
1244                          * in this call, so we don't have to rely on every
1245                          * load_binary function to restore it on return.
1246                          */
1247                         bprm->recursion_depth = depth;
1248                         if (retval >= 0) {
1249                                 if (depth == 0)
1250                                         tracehook_report_exec(fmt, bprm, regs);
1251                                 put_binfmt(fmt);
1252                                 allow_write_access(bprm->file);
1253                                 if (bprm->file)
1254                                         fput(bprm->file);
1255                                 bprm->file = NULL;
1256                                 current->did_exec = 1;
1257                                 proc_exec_connector(current);
1258                                 return retval;
1259                         }
1260                         read_lock(&binfmt_lock);
1261                         put_binfmt(fmt);
1262                         if (retval != -ENOEXEC || bprm->mm == NULL)
1263                                 break;
1264                         if (!bprm->file) {
1265                                 read_unlock(&binfmt_lock);
1266                                 return retval;
1267                         }
1268                 }
1269                 read_unlock(&binfmt_lock);
1270                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1271                         break;
1272 #ifdef CONFIG_MODULES
1273                 } else {
1274 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1275                         if (printable(bprm->buf[0]) &&
1276                             printable(bprm->buf[1]) &&
1277                             printable(bprm->buf[2]) &&
1278                             printable(bprm->buf[3]))
1279                                 break; /* -ENOEXEC */
1280                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1281 #endif
1282                 }
1283         }
1284         return retval;
1285 }
1286
1287 EXPORT_SYMBOL(search_binary_handler);
1288
1289 void free_bprm(struct linux_binprm *bprm)
1290 {
1291         free_arg_pages(bprm);
1292         if (bprm->cred)
1293                 abort_creds(bprm->cred);
1294         kfree(bprm);
1295 }
1296
1297 /*
1298  * sys_execve() executes a new program.
1299  */
1300 int do_execve(char * filename,
1301         char __user *__user *argv,
1302         char __user *__user *envp,
1303         struct pt_regs * regs)
1304 {
1305         struct linux_binprm *bprm;
1306         struct file *file;
1307         struct files_struct *displaced;
1308         int retval;
1309
1310         retval = unshare_files(&displaced);
1311         if (retval)
1312                 goto out_ret;
1313
1314         retval = -ENOMEM;
1315         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1316         if (!bprm)
1317                 goto out_files;
1318
1319         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1320         if (retval < 0)
1321                 goto out_free;
1322
1323         retval = -ENOMEM;
1324         bprm->cred = prepare_exec_creds();
1325         if (!bprm->cred)
1326                 goto out_unlock;
1327         check_unsafe_exec(bprm);
1328
1329         file = open_exec(filename);
1330         retval = PTR_ERR(file);
1331         if (IS_ERR(file))
1332                 goto out_unlock;
1333
1334         sched_exec();
1335
1336         bprm->file = file;
1337         bprm->filename = filename;
1338         bprm->interp = filename;
1339
1340         retval = bprm_mm_init(bprm);
1341         if (retval)
1342                 goto out_file;
1343
1344         bprm->argc = count(argv, MAX_ARG_STRINGS);
1345         if ((retval = bprm->argc) < 0)
1346                 goto out;
1347
1348         bprm->envc = count(envp, MAX_ARG_STRINGS);
1349         if ((retval = bprm->envc) < 0)
1350                 goto out;
1351
1352         retval = prepare_binprm(bprm);
1353         if (retval < 0)
1354                 goto out;
1355
1356         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1357         if (retval < 0)
1358                 goto out;
1359
1360         bprm->exec = bprm->p;
1361         retval = copy_strings(bprm->envc, envp, bprm);
1362         if (retval < 0)
1363                 goto out;
1364
1365         retval = copy_strings(bprm->argc, argv, bprm);
1366         if (retval < 0)
1367                 goto out;
1368
1369         current->flags &= ~PF_KTHREAD;
1370         retval = search_binary_handler(bprm,regs);
1371         if (retval < 0)
1372                 goto out;
1373
1374         /* execve succeeded */
1375         mutex_unlock(&current->cred_exec_mutex);
1376         acct_update_integrals(current);
1377         free_bprm(bprm);
1378         if (displaced)
1379                 put_files_struct(displaced);
1380         return retval;
1381
1382 out:
1383         if (bprm->mm)
1384                 mmput (bprm->mm);
1385
1386 out_file:
1387         if (bprm->file) {
1388                 allow_write_access(bprm->file);
1389                 fput(bprm->file);
1390         }
1391
1392 out_unlock:
1393         mutex_unlock(&current->cred_exec_mutex);
1394
1395 out_free:
1396         free_bprm(bprm);
1397
1398 out_files:
1399         if (displaced)
1400                 reset_files_struct(displaced);
1401 out_ret:
1402         return retval;
1403 }
1404
1405 int set_binfmt(struct linux_binfmt *new)
1406 {
1407         struct linux_binfmt *old = current->binfmt;
1408
1409         if (new) {
1410                 if (!try_module_get(new->module))
1411                         return -1;
1412         }
1413         current->binfmt = new;
1414         if (old)
1415                 module_put(old->module);
1416         return 0;
1417 }
1418
1419 EXPORT_SYMBOL(set_binfmt);
1420
1421 /* format_corename will inspect the pattern parameter, and output a
1422  * name into corename, which must have space for at least
1423  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1424  */
1425 static int format_corename(char *corename, long signr)
1426 {
1427         const struct cred *cred = current_cred();
1428         const char *pat_ptr = core_pattern;
1429         int ispipe = (*pat_ptr == '|');
1430         char *out_ptr = corename;
1431         char *const out_end = corename + CORENAME_MAX_SIZE;
1432         int rc;
1433         int pid_in_pattern = 0;
1434
1435         /* Repeat as long as we have more pattern to process and more output
1436            space */
1437         while (*pat_ptr) {
1438                 if (*pat_ptr != '%') {
1439                         if (out_ptr == out_end)
1440                                 goto out;
1441                         *out_ptr++ = *pat_ptr++;
1442                 } else {
1443                         switch (*++pat_ptr) {
1444                         case 0:
1445                                 goto out;
1446                         /* Double percent, output one percent */
1447                         case '%':
1448                                 if (out_ptr == out_end)
1449                                         goto out;
1450                                 *out_ptr++ = '%';
1451                                 break;
1452                         /* pid */
1453                         case 'p':
1454                                 pid_in_pattern = 1;
1455                                 rc = snprintf(out_ptr, out_end - out_ptr,
1456                                               "%d", task_tgid_vnr(current));
1457                                 if (rc > out_end - out_ptr)
1458                                         goto out;
1459                                 out_ptr += rc;
1460                                 break;
1461                         /* uid */
1462                         case 'u':
1463                                 rc = snprintf(out_ptr, out_end - out_ptr,
1464                                               "%d", cred->uid);
1465                                 if (rc > out_end - out_ptr)
1466                                         goto out;
1467                                 out_ptr += rc;
1468                                 break;
1469                         /* gid */
1470                         case 'g':
1471                                 rc = snprintf(out_ptr, out_end - out_ptr,
1472                                               "%d", cred->gid);
1473                                 if (rc > out_end - out_ptr)
1474                                         goto out;
1475                                 out_ptr += rc;
1476                                 break;
1477                         /* signal that caused the coredump */
1478                         case 's':
1479                                 rc = snprintf(out_ptr, out_end - out_ptr,
1480                                               "%ld", signr);
1481                                 if (rc > out_end - out_ptr)
1482                                         goto out;
1483                                 out_ptr += rc;
1484                                 break;
1485                         /* UNIX time of coredump */
1486                         case 't': {
1487                                 struct timeval tv;
1488                                 do_gettimeofday(&tv);
1489                                 rc = snprintf(out_ptr, out_end - out_ptr,
1490                                               "%lu", tv.tv_sec);
1491                                 if (rc > out_end - out_ptr)
1492                                         goto out;
1493                                 out_ptr += rc;
1494                                 break;
1495                         }
1496                         /* hostname */
1497                         case 'h':
1498                                 down_read(&uts_sem);
1499                                 rc = snprintf(out_ptr, out_end - out_ptr,
1500                                               "%s", utsname()->nodename);
1501                                 up_read(&uts_sem);
1502                                 if (rc > out_end - out_ptr)
1503                                         goto out;
1504                                 out_ptr += rc;
1505                                 break;
1506                         /* executable */
1507                         case 'e':
1508                                 rc = snprintf(out_ptr, out_end - out_ptr,
1509                                               "%s", current->comm);
1510                                 if (rc > out_end - out_ptr)
1511                                         goto out;
1512                                 out_ptr += rc;
1513                                 break;
1514                         /* core limit size */
1515                         case 'c':
1516                                 rc = snprintf(out_ptr, out_end - out_ptr,
1517                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1518                                 if (rc > out_end - out_ptr)
1519                                         goto out;
1520                                 out_ptr += rc;
1521                                 break;
1522                         default:
1523                                 break;
1524                         }
1525                         ++pat_ptr;
1526                 }
1527         }
1528         /* Backward compatibility with core_uses_pid:
1529          *
1530          * If core_pattern does not include a %p (as is the default)
1531          * and core_uses_pid is set, then .%pid will be appended to
1532          * the filename. Do not do this for piped commands. */
1533         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1534                 rc = snprintf(out_ptr, out_end - out_ptr,
1535                               ".%d", task_tgid_vnr(current));
1536                 if (rc > out_end - out_ptr)
1537                         goto out;
1538                 out_ptr += rc;
1539         }
1540 out:
1541         *out_ptr = 0;
1542         return ispipe;
1543 }
1544
1545 static int zap_process(struct task_struct *start)
1546 {
1547         struct task_struct *t;
1548         int nr = 0;
1549
1550         start->signal->flags = SIGNAL_GROUP_EXIT;
1551         start->signal->group_stop_count = 0;
1552
1553         t = start;
1554         do {
1555                 if (t != current && t->mm) {
1556                         sigaddset(&t->pending.signal, SIGKILL);
1557                         signal_wake_up(t, 1);
1558                         nr++;
1559                 }
1560         } while_each_thread(start, t);
1561
1562         return nr;
1563 }
1564
1565 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1566                                 struct core_state *core_state, int exit_code)
1567 {
1568         struct task_struct *g, *p;
1569         unsigned long flags;
1570         int nr = -EAGAIN;
1571
1572         spin_lock_irq(&tsk->sighand->siglock);
1573         if (!signal_group_exit(tsk->signal)) {
1574                 mm->core_state = core_state;
1575                 tsk->signal->group_exit_code = exit_code;
1576                 nr = zap_process(tsk);
1577         }
1578         spin_unlock_irq(&tsk->sighand->siglock);
1579         if (unlikely(nr < 0))
1580                 return nr;
1581
1582         if (atomic_read(&mm->mm_users) == nr + 1)
1583                 goto done;
1584         /*
1585          * We should find and kill all tasks which use this mm, and we should
1586          * count them correctly into ->nr_threads. We don't take tasklist
1587          * lock, but this is safe wrt:
1588          *
1589          * fork:
1590          *      None of sub-threads can fork after zap_process(leader). All
1591          *      processes which were created before this point should be
1592          *      visible to zap_threads() because copy_process() adds the new
1593          *      process to the tail of init_task.tasks list, and lock/unlock
1594          *      of ->siglock provides a memory barrier.
1595          *
1596          * do_exit:
1597          *      The caller holds mm->mmap_sem. This means that the task which
1598          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1599          *      its ->mm.
1600          *
1601          * de_thread:
1602          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1603          *      we must see either old or new leader, this does not matter.
1604          *      However, it can change p->sighand, so lock_task_sighand(p)
1605          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1606          *      it can't fail.
1607          *
1608          *      Note also that "g" can be the old leader with ->mm == NULL
1609          *      and already unhashed and thus removed from ->thread_group.
1610          *      This is OK, __unhash_process()->list_del_rcu() does not
1611          *      clear the ->next pointer, we will find the new leader via
1612          *      next_thread().
1613          */
1614         rcu_read_lock();
1615         for_each_process(g) {
1616                 if (g == tsk->group_leader)
1617                         continue;
1618                 if (g->flags & PF_KTHREAD)
1619                         continue;
1620                 p = g;
1621                 do {
1622                         if (p->mm) {
1623                                 if (unlikely(p->mm == mm)) {
1624                                         lock_task_sighand(p, &flags);
1625                                         nr += zap_process(p);
1626                                         unlock_task_sighand(p, &flags);
1627                                 }
1628                                 break;
1629                         }
1630                 } while_each_thread(g, p);
1631         }
1632         rcu_read_unlock();
1633 done:
1634         atomic_set(&core_state->nr_threads, nr);
1635         return nr;
1636 }
1637
1638 static int coredump_wait(int exit_code, struct core_state *core_state)
1639 {
1640         struct task_struct *tsk = current;
1641         struct mm_struct *mm = tsk->mm;
1642         struct completion *vfork_done;
1643         int core_waiters;
1644
1645         init_completion(&core_state->startup);
1646         core_state->dumper.task = tsk;
1647         core_state->dumper.next = NULL;
1648         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1649         up_write(&mm->mmap_sem);
1650
1651         if (unlikely(core_waiters < 0))
1652                 goto fail;
1653
1654         /*
1655          * Make sure nobody is waiting for us to release the VM,
1656          * otherwise we can deadlock when we wait on each other
1657          */
1658         vfork_done = tsk->vfork_done;
1659         if (vfork_done) {
1660                 tsk->vfork_done = NULL;
1661                 complete(vfork_done);
1662         }
1663
1664         if (core_waiters)
1665                 wait_for_completion(&core_state->startup);
1666 fail:
1667         return core_waiters;
1668 }
1669
1670 static void coredump_finish(struct mm_struct *mm)
1671 {
1672         struct core_thread *curr, *next;
1673         struct task_struct *task;
1674
1675         next = mm->core_state->dumper.next;
1676         while ((curr = next) != NULL) {
1677                 next = curr->next;
1678                 task = curr->task;
1679                 /*
1680                  * see exit_mm(), curr->task must not see
1681                  * ->task == NULL before we read ->next.
1682                  */
1683                 smp_mb();
1684                 curr->task = NULL;
1685                 wake_up_process(task);
1686         }
1687
1688         mm->core_state = NULL;
1689 }
1690
1691 /*
1692  * set_dumpable converts traditional three-value dumpable to two flags and
1693  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1694  * these bits are not changed atomically.  So get_dumpable can observe the
1695  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1696  * return either old dumpable or new one by paying attention to the order of
1697  * modifying the bits.
1698  *
1699  * dumpable |   mm->flags (binary)
1700  * old  new | initial interim  final
1701  * ---------+-----------------------
1702  *  0    1  |   00      01      01
1703  *  0    2  |   00      10(*)   11
1704  *  1    0  |   01      00      00
1705  *  1    2  |   01      11      11
1706  *  2    0  |   11      10(*)   00
1707  *  2    1  |   11      11      01
1708  *
1709  * (*) get_dumpable regards interim value of 10 as 11.
1710  */
1711 void set_dumpable(struct mm_struct *mm, int value)
1712 {
1713         switch (value) {
1714         case 0:
1715                 clear_bit(MMF_DUMPABLE, &mm->flags);
1716                 smp_wmb();
1717                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1718                 break;
1719         case 1:
1720                 set_bit(MMF_DUMPABLE, &mm->flags);
1721                 smp_wmb();
1722                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1723                 break;
1724         case 2:
1725                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1726                 smp_wmb();
1727                 set_bit(MMF_DUMPABLE, &mm->flags);
1728                 break;
1729         }
1730 }
1731
1732 int get_dumpable(struct mm_struct *mm)
1733 {
1734         int ret;
1735
1736         ret = mm->flags & 0x3;
1737         return (ret >= 2) ? 2 : ret;
1738 }
1739
1740 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1741 {
1742         struct core_state core_state;
1743         char corename[CORENAME_MAX_SIZE + 1];
1744         struct mm_struct *mm = current->mm;
1745         struct linux_binfmt * binfmt;
1746         struct inode * inode;
1747         struct file * file;
1748         const struct cred *old_cred;
1749         struct cred *cred;
1750         int retval = 0;
1751         int flag = 0;
1752         int ispipe = 0;
1753         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1754         char **helper_argv = NULL;
1755         int helper_argc = 0;
1756         char *delimit;
1757
1758         audit_core_dumps(signr);
1759
1760         binfmt = current->binfmt;
1761         if (!binfmt || !binfmt->core_dump)
1762                 goto fail;
1763
1764         cred = prepare_creds();
1765         if (!cred) {
1766                 retval = -ENOMEM;
1767                 goto fail;
1768         }
1769
1770         down_write(&mm->mmap_sem);
1771         /*
1772          * If another thread got here first, or we are not dumpable, bail out.
1773          */
1774         if (mm->core_state || !get_dumpable(mm)) {
1775                 up_write(&mm->mmap_sem);
1776                 put_cred(cred);
1777                 goto fail;
1778         }
1779
1780         /*
1781          *      We cannot trust fsuid as being the "true" uid of the
1782          *      process nor do we know its entire history. We only know it
1783          *      was tainted so we dump it as root in mode 2.
1784          */
1785         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1786                 flag = O_EXCL;          /* Stop rewrite attacks */
1787                 cred->fsuid = 0;        /* Dump root private */
1788         }
1789
1790         retval = coredump_wait(exit_code, &core_state);
1791         if (retval < 0) {
1792                 put_cred(cred);
1793                 goto fail;
1794         }
1795
1796         old_cred = override_creds(cred);
1797
1798         /*
1799          * Clear any false indication of pending signals that might
1800          * be seen by the filesystem code called to write the core file.
1801          */
1802         clear_thread_flag(TIF_SIGPENDING);
1803
1804         /*
1805          * lock_kernel() because format_corename() is controlled by sysctl, which
1806          * uses lock_kernel()
1807          */
1808         lock_kernel();
1809         ispipe = format_corename(corename, signr);
1810         unlock_kernel();
1811         /*
1812          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1813          * to a pipe.  Since we're not writing directly to the filesystem
1814          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1815          * created unless the pipe reader choses to write out the core file
1816          * at which point file size limits and permissions will be imposed
1817          * as it does with any other process
1818          */
1819         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1820                 goto fail_unlock;
1821
1822         if (ispipe) {
1823                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1824                 /* Terminate the string before the first option */
1825                 delimit = strchr(corename, ' ');
1826                 if (delimit)
1827                         *delimit = '\0';
1828                 delimit = strrchr(helper_argv[0], '/');
1829                 if (delimit)
1830                         delimit++;
1831                 else
1832                         delimit = helper_argv[0];
1833                 if (!strcmp(delimit, current->comm)) {
1834                         printk(KERN_NOTICE "Recursive core dump detected, "
1835                                         "aborting\n");
1836                         goto fail_unlock;
1837                 }
1838
1839                 core_limit = RLIM_INFINITY;
1840
1841                 /* SIGPIPE can happen, but it's just never processed */
1842                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1843                                 &file)) {
1844                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1845                                corename);
1846                         goto fail_unlock;
1847                 }
1848         } else
1849                 file = filp_open(corename,
1850                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1851                                  0600);
1852         if (IS_ERR(file))
1853                 goto fail_unlock;
1854         inode = file->f_path.dentry->d_inode;
1855         if (inode->i_nlink > 1)
1856                 goto close_fail;        /* multiple links - don't dump */
1857         if (!ispipe && d_unhashed(file->f_path.dentry))
1858                 goto close_fail;
1859
1860         /* AK: actually i see no reason to not allow this for named pipes etc.,
1861            but keep the previous behaviour for now. */
1862         if (!ispipe && !S_ISREG(inode->i_mode))
1863                 goto close_fail;
1864         /*
1865          * Dont allow local users get cute and trick others to coredump
1866          * into their pre-created files:
1867          */
1868         if (inode->i_uid != current_fsuid())
1869                 goto close_fail;
1870         if (!file->f_op)
1871                 goto close_fail;
1872         if (!file->f_op->write)
1873                 goto close_fail;
1874         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1875                 goto close_fail;
1876
1877         retval = binfmt->core_dump(signr, regs, file, core_limit);
1878
1879         if (retval)
1880                 current->signal->group_exit_code |= 0x80;
1881 close_fail:
1882         filp_close(file, NULL);
1883 fail_unlock:
1884         if (helper_argv)
1885                 argv_free(helper_argv);
1886
1887         revert_creds(old_cred);
1888         put_cred(cred);
1889         coredump_finish(mm);
1890 fail:
1891         return retval;
1892 }