git.oblomov.eu Git - linux-2.6/blob - fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54
  55 #include <asm/uaccess.h>
  56 #include <asm/mmu_context.h>
  57 #include <asm/tlb.h>
  58 #include "internal.h"
  59
  60 #ifdef __alpha__
  61 /* for /sbin/loader handling in search_binary_handler() */
  62 #include <linux/a.out.h>
  63 #endif
  64
  65 int core_uses_pid;
  66 char core_pattern[CORENAME_MAX_SIZE] = "core";
  67 int suid_dumpable = 0;
  68
  69 /* The maximal length of core_pattern is also specified in sysctl.c */
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 int register_binfmt(struct linux_binfmt * fmt)
  75 {
  76         if (!fmt)
  77                 return -EINVAL;
  78         write_lock(&binfmt_lock);
  79         list_add(&fmt->lh, &formats);
  80         write_unlock(&binfmt_lock);
  81         return 0;
  82 }
  83
  84 EXPORT_SYMBOL(register_binfmt);
  85
  86 void unregister_binfmt(struct linux_binfmt * fmt)
  87 {
  88         write_lock(&binfmt_lock);
  89         list_del(&fmt->lh);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(unregister_binfmt);
  94
  95 static inline void put_binfmt(struct linux_binfmt * fmt)
  96 {
  97         module_put(fmt->module);
  98 }
  99
 100 /*
 101  * Note that a shared library must be both readable and executable due to
 102  * security reasons.
 103  *
 104  * Also note that we take the address to load from from the file itself.
 105  */
 106 asmlinkage long sys_uselib(const char __user * library)
 107 {
 108         struct file *file;
 109         struct nameidata nd;
 110         char *tmp = getname(library);
 111         int error = PTR_ERR(tmp);
 112
 113         if (!IS_ERR(tmp)) {
 114                 error = path_lookup_open(AT_FDCWD, tmp,
 115                                          LOOKUP_FOLLOW, &nd,
 116                                          FMODE_READ|FMODE_EXEC);
 117                 putname(tmp);
 118         }
 119         if (error)
 120                 goto out;
 121
 122         error = -EINVAL;
 123         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 124                 goto exit;
 125
 126         error = -EACCES;
 127         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
 128                 goto exit;
 129
 130         error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
 131         if (error)
 132                 goto exit;
 133
 134         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 135         error = PTR_ERR(file);
 136         if (IS_ERR(file))
 137                 goto out;
 138
 139         error = -ENOEXEC;
 140         if(file->f_op) {
 141                 struct linux_binfmt * fmt;
 142
 143                 read_lock(&binfmt_lock);
 144                 list_for_each_entry(fmt, &formats, lh) {
 145                         if (!fmt->load_shlib)
 146                                 continue;
 147                         if (!try_module_get(fmt->module))
 148                                 continue;
 149                         read_unlock(&binfmt_lock);
 150                         error = fmt->load_shlib(file);
 151                         read_lock(&binfmt_lock);
 152                         put_binfmt(fmt);
 153                         if (error != -ENOEXEC)
 154                                 break;
 155                 }
 156                 read_unlock(&binfmt_lock);
 157         }
 158         fput(file);
 159 out:
 160         return error;
 161 exit:
 162         release_open_intent(&nd);
 163         path_put(&nd.path);
 164         goto out;
 165 }
 166
 167 #ifdef CONFIG_MMU
 168
 169 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 170                 int write)
 171 {
 172         struct page *page;
 173         int ret;
 174
 175 #ifdef CONFIG_STACK_GROWSUP
 176         if (write) {
 177                 ret = expand_stack_downwards(bprm->vma, pos);
 178                 if (ret < 0)
 179                         return NULL;
 180         }
 181 #endif
 182         ret = get_user_pages(current, bprm->mm, pos,
 183                         1, write, 1, &page, NULL);
 184         if (ret <= 0)
 185                 return NULL;
 186
 187         if (write) {
 188                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 189                 struct rlimit *rlim;
 190
 191                 /*
 192                  * We've historically supported up to 32 pages (ARG_MAX)
 193                  * of argument strings even with small stacks
 194                  */
 195                 if (size <= ARG_MAX)
 196                         return page;
 197
 198                 /*
 199                  * Limit to 1/4-th the stack size for the argv+env strings.
 200                  * This ensures that:
 201                  *  - the remaining binfmt code will not run out of stack space,
 202                  *  - the program will have a reasonable amount of stack left
 203                  *    to work from.
 204                  */
 205                 rlim = current->signal->rlim;
 206                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 207                         put_page(page);
 208                         return NULL;
 209                 }
 210         }
 211
 212         return page;
 213 }
 214
 215 static void put_arg_page(struct page *page)
 216 {
 217         put_page(page);
 218 }
 219
 220 static void free_arg_page(struct linux_binprm *bprm, int i)
 221 {
 222 }
 223
 224 static void free_arg_pages(struct linux_binprm *bprm)
 225 {
 226 }
 227
 228 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 229                 struct page *page)
 230 {
 231         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 232 }
 233
 234 static int __bprm_mm_init(struct linux_binprm *bprm)
 235 {
 236         int err = -ENOMEM;
 237         struct vm_area_struct *vma = NULL;
 238         struct mm_struct *mm = bprm->mm;
 239
 240         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 241         if (!vma)
 242                 goto err;
 243
 244         down_write(&mm->mmap_sem);
 245         vma->vm_mm = mm;
 246
 247         /*
 248          * Place the stack at the largest stack address the architecture
 249          * supports. Later, we'll move this to an appropriate place. We don't
 250          * use STACK_TOP because that can depend on attributes which aren't
 251          * configured yet.
 252          */
 253         vma->vm_end = STACK_TOP_MAX;
 254         vma->vm_start = vma->vm_end - PAGE_SIZE;
 255
 256         vma->vm_flags = VM_STACK_FLAGS;
 257         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 258         err = insert_vm_struct(mm, vma);
 259         if (err) {
 260                 up_write(&mm->mmap_sem);
 261                 goto err;
 262         }
 263
 264         mm->stack_vm = mm->total_vm = 1;
 265         up_write(&mm->mmap_sem);
 266
 267         bprm->p = vma->vm_end - sizeof(void *);
 268
 269         return 0;
 270
 271 err:
 272         if (vma) {
 273                 bprm->vma = NULL;
 274                 kmem_cache_free(vm_area_cachep, vma);
 275         }
 276
 277         return err;
 278 }
 279
 280 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 281 {
 282         return len <= MAX_ARG_STRLEN;
 283 }
 284
 285 #else
 286
 287 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 288                 int write)
 289 {
 290         struct page *page;
 291
 292         page = bprm->page[pos / PAGE_SIZE];
 293         if (!page && write) {
 294                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 295                 if (!page)
 296                         return NULL;
 297                 bprm->page[pos / PAGE_SIZE] = page;
 298         }
 299
 300         return page;
 301 }
 302
 303 static void put_arg_page(struct page *page)
 304 {
 305 }
 306
 307 static void free_arg_page(struct linux_binprm *bprm, int i)
 308 {
 309         if (bprm->page[i]) {
 310                 __free_page(bprm->page[i]);
 311                 bprm->page[i] = NULL;
 312         }
 313 }
 314
 315 static void free_arg_pages(struct linux_binprm *bprm)
 316 {
 317         int i;
 318
 319         for (i = 0; i < MAX_ARG_PAGES; i++)
 320                 free_arg_page(bprm, i);
 321 }
 322
 323 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 324                 struct page *page)
 325 {
 326 }
 327
 328 static int __bprm_mm_init(struct linux_binprm *bprm)
 329 {
 330         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 331         return 0;
 332 }
 333
 334 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 335 {
 336         return len <= bprm->p;
 337 }
 338
 339 #endif /* CONFIG_MMU */
 340
 341 /*
 342  * Create a new mm_struct and populate it with a temporary stack
 343  * vm_area_struct.  We don't have enough context at this point to set the stack
 344  * flags, permissions, and offset, so we use temporary values.  We'll update
 345  * them later in setup_arg_pages().
 346  */
 347 int bprm_mm_init(struct linux_binprm *bprm)
 348 {
 349         int err;
 350         struct mm_struct *mm = NULL;
 351
 352         bprm->mm = mm = mm_alloc();
 353         err = -ENOMEM;
 354         if (!mm)
 355                 goto err;
 356
 357         err = init_new_context(current, mm);
 358         if (err)
 359                 goto err;
 360
 361         err = __bprm_mm_init(bprm);
 362         if (err)
 363                 goto err;
 364
 365         return 0;
 366
 367 err:
 368         if (mm) {
 369                 bprm->mm = NULL;
 370                 mmdrop(mm);
 371         }
 372
 373         return err;
 374 }
 375
 376 /*
 377  * count() counts the number of strings in array ARGV.
 378  */
 379 static int count(char __user * __user * argv, int max)
 380 {
 381         int i = 0;
 382
 383         if (argv != NULL) {
 384                 for (;;) {
 385                         char __user * p;
 386
 387                         if (get_user(p, argv))
 388                                 return -EFAULT;
 389                         if (!p)
 390                                 break;
 391                         argv++;
 392                         if (i++ >= max)
 393                                 return -E2BIG;
 394                         cond_resched();
 395                 }
 396         }
 397         return i;
 398 }
 399
 400 /*
 401  * 'copy_strings()' copies argument/environment strings from the old
 402  * processes's memory to the new process's stack.  The call to get_user_pages()
 403  * ensures the destination page is created and not swapped out.
 404  */
 405 static int copy_strings(int argc, char __user * __user * argv,
 406                         struct linux_binprm *bprm)
 407 {
 408         struct page *kmapped_page = NULL;
 409         char *kaddr = NULL;
 410         unsigned long kpos = 0;
 411         int ret;
 412
 413         while (argc-- > 0) {
 414                 char __user *str;
 415                 int len;
 416                 unsigned long pos;
 417
 418                 if (get_user(str, argv+argc) ||
 419                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 420                         ret = -EFAULT;
 421                         goto out;
 422                 }
 423
 424                 if (!valid_arg_len(bprm, len)) {
 425                         ret = -E2BIG;
 426                         goto out;
 427                 }
 428
 429                 /* We're going to work our way backwords. */
 430                 pos = bprm->p;
 431                 str += len;
 432                 bprm->p -= len;
 433
 434                 while (len > 0) {
 435                         int offset, bytes_to_copy;
 436
 437                         offset = pos % PAGE_SIZE;
 438                         if (offset == 0)
 439                                 offset = PAGE_SIZE;
 440
 441                         bytes_to_copy = offset;
 442                         if (bytes_to_copy > len)
 443                                 bytes_to_copy = len;
 444
 445                         offset -= bytes_to_copy;
 446                         pos -= bytes_to_copy;
 447                         str -= bytes_to_copy;
 448                         len -= bytes_to_copy;
 449
 450                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 451                                 struct page *page;
 452
 453                                 page = get_arg_page(bprm, pos, 1);
 454                                 if (!page) {
 455                                         ret = -E2BIG;
 456                                         goto out;
 457                                 }
 458
 459                                 if (kmapped_page) {
 460                                         flush_kernel_dcache_page(kmapped_page);
 461                                         kunmap(kmapped_page);
 462                                         put_arg_page(kmapped_page);
 463                                 }
 464                                 kmapped_page = page;
 465                                 kaddr = kmap(kmapped_page);
 466                                 kpos = pos & PAGE_MASK;
 467                                 flush_arg_page(bprm, kpos, kmapped_page);
 468                         }
 469                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 470                                 ret = -EFAULT;
 471                                 goto out;
 472                         }
 473                 }
 474         }
 475         ret = 0;
 476 out:
 477         if (kmapped_page) {
 478                 flush_kernel_dcache_page(kmapped_page);
 479                 kunmap(kmapped_page);
 480                 put_arg_page(kmapped_page);
 481         }
 482         return ret;
 483 }
 484
 485 /*
 486  * Like copy_strings, but get argv and its values from kernel memory.
 487  */
 488 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 489 {
 490         int r;
 491         mm_segment_t oldfs = get_fs();
 492         set_fs(KERNEL_DS);
 493         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 494         set_fs(oldfs);
 495         return r;
 496 }
 497 EXPORT_SYMBOL(copy_strings_kernel);
 498
 499 #ifdef CONFIG_MMU
 500
 501 /*
 502  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 503  * the binfmt code determines where the new stack should reside, we shift it to
 504  * its final location.  The process proceeds as follows:
 505  *
 506  * 1) Use shift to calculate the new vma endpoints.
 507  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 508  *    arguments passed to subsequent functions are consistent.
 509  * 3) Move vma's page tables to the new range.
 510  * 4) Free up any cleared pgd range.
 511  * 5) Shrink the vma to cover only the new range.
 512  */
 513 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 514 {
 515         struct mm_struct *mm = vma->vm_mm;
 516         unsigned long old_start = vma->vm_start;
 517         unsigned long old_end = vma->vm_end;
 518         unsigned long length = old_end - old_start;
 519         unsigned long new_start = old_start - shift;
 520         unsigned long new_end = old_end - shift;
 521         struct mmu_gather *tlb;
 522
 523         BUG_ON(new_start > new_end);
 524
 525         /*
 526          * ensure there are no vmas between where we want to go
 527          * and where we are
 528          */
 529         if (vma != find_vma(mm, new_start))
 530                 return -EFAULT;
 531
 532         /*
 533          * cover the whole range: [new_start, old_end)
 534          */
 535         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 536
 537         /*
 538          * move the page tables downwards, on failure we rely on
 539          * process cleanup to remove whatever mess we made.
 540          */
 541         if (length != move_page_tables(vma, old_start,
 542                                        vma, new_start, length))
 543                 return -ENOMEM;
 544
 545         lru_add_drain();
 546         tlb = tlb_gather_mmu(mm, 0);
 547         if (new_end > old_start) {
 548                 /*
 549                  * when the old and new regions overlap clear from new_end.
 550                  */
 551                 free_pgd_range(tlb, new_end, old_end, new_end,
 552                         vma->vm_next ? vma->vm_next->vm_start : 0);
 553         } else {
 554                 /*
 555                  * otherwise, clean from old_start; this is done to not touch
 556                  * the address space in [new_end, old_start) some architectures
 557                  * have constraints on va-space that make this illegal (IA64) -
 558                  * for the others its just a little faster.
 559                  */
 560                 free_pgd_range(tlb, old_start, old_end, new_end,
 561                         vma->vm_next ? vma->vm_next->vm_start : 0);
 562         }
 563         tlb_finish_mmu(tlb, new_end, old_end);
 564
 565         /*
 566          * shrink the vma to just the new range.
 567          */
 568         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 569
 570         return 0;
 571 }
 572
 573 #define EXTRA_STACK_VM_PAGES    20      /* random */
 574
 575 /*
 576  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 577  * the stack is optionally relocated, and some extra space is added.
 578  */
 579 int setup_arg_pages(struct linux_binprm *bprm,
 580                     unsigned long stack_top,
 581                     int executable_stack)
 582 {
 583         unsigned long ret;
 584         unsigned long stack_shift;
 585         struct mm_struct *mm = current->mm;
 586         struct vm_area_struct *vma = bprm->vma;
 587         struct vm_area_struct *prev = NULL;
 588         unsigned long vm_flags;
 589         unsigned long stack_base;
 590
 591 #ifdef CONFIG_STACK_GROWSUP
 592         /* Limit stack size to 1GB */
 593         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 594         if (stack_base > (1 << 30))
 595                 stack_base = 1 << 30;
 596
 597         /* Make sure we didn't let the argument array grow too large. */
 598         if (vma->vm_end - vma->vm_start > stack_base)
 599                 return -ENOMEM;
 600
 601         stack_base = PAGE_ALIGN(stack_top - stack_base);
 602
 603         stack_shift = vma->vm_start - stack_base;
 604         mm->arg_start = bprm->p - stack_shift;
 605         bprm->p = vma->vm_end - stack_shift;
 606 #else
 607         stack_top = arch_align_stack(stack_top);
 608         stack_top = PAGE_ALIGN(stack_top);
 609         stack_shift = vma->vm_end - stack_top;
 610
 611         bprm->p -= stack_shift;
 612         mm->arg_start = bprm->p;
 613 #endif
 614
 615         if (bprm->loader)
 616                 bprm->loader -= stack_shift;
 617         bprm->exec -= stack_shift;
 618
 619         down_write(&mm->mmap_sem);
 620         vm_flags = VM_STACK_FLAGS;
 621
 622         /*
 623          * Adjust stack execute permissions; explicitly enable for
 624          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 625          * (arch default) otherwise.
 626          */
 627         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 628                 vm_flags |= VM_EXEC;
 629         else if (executable_stack == EXSTACK_DISABLE_X)
 630                 vm_flags &= ~VM_EXEC;
 631         vm_flags |= mm->def_flags;
 632
 633         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 634                         vm_flags);
 635         if (ret)
 636                 goto out_unlock;
 637         BUG_ON(prev != vma);
 638
 639         /* Move stack pages down in memory. */
 640         if (stack_shift) {
 641                 ret = shift_arg_pages(vma, stack_shift);
 642                 if (ret) {
 643                         up_write(&mm->mmap_sem);
 644                         return ret;
 645                 }
 646         }
 647
 648 #ifdef CONFIG_STACK_GROWSUP
 649         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 650 #else
 651         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 652 #endif
 653         ret = expand_stack(vma, stack_base);
 654         if (ret)
 655                 ret = -EFAULT;
 656
 657 out_unlock:
 658         up_write(&mm->mmap_sem);
 659         return 0;
 660 }
 661 EXPORT_SYMBOL(setup_arg_pages);
 662
 663 #endif /* CONFIG_MMU */
 664
 665 struct file *open_exec(const char *name)
 666 {
 667         struct nameidata nd;
 668         struct file *file;
 669         int err;
 670
 671         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
 672                                 FMODE_READ|FMODE_EXEC);
 673         if (err)
 674                 goto out;
 675
 676         err = -EACCES;
 677         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 678                 goto out_path_put;
 679
 680         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
 681                 goto out_path_put;
 682
 683         err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
 684         if (err)
 685                 goto out_path_put;
 686
 687         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 688         if (IS_ERR(file))
 689                 return file;
 690
 691         err = deny_write_access(file);
 692         if (err) {
 693                 fput(file);
 694                 goto out;
 695         }
 696
 697         return file;
 698
 699  out_path_put:
 700         release_open_intent(&nd);
 701         path_put(&nd.path);
 702  out:
 703         return ERR_PTR(err);
 704 }
 705 EXPORT_SYMBOL(open_exec);
 706
 707 int kernel_read(struct file *file, unsigned long offset,
 708         char *addr, unsigned long count)
 709 {
 710         mm_segment_t old_fs;
 711         loff_t pos = offset;
 712         int result;
 713
 714         old_fs = get_fs();
 715         set_fs(get_ds());
 716         /* The cast to a user pointer is valid due to the set_fs() */
 717         result = vfs_read(file, (void __user *)addr, count, &pos);
 718         set_fs(old_fs);
 719         return result;
 720 }
 721
 722 EXPORT_SYMBOL(kernel_read);
 723
 724 static int exec_mmap(struct mm_struct *mm)
 725 {
 726         struct task_struct *tsk;
 727         struct mm_struct * old_mm, *active_mm;
 728
 729         /* Notify parent that we're no longer interested in the old VM */
 730         tsk = current;
 731         old_mm = current->mm;
 732         mm_release(tsk, old_mm);
 733
 734         if (old_mm) {
 735                 /*
 736                  * Make sure that if there is a core dump in progress
 737                  * for the old mm, we get out and die instead of going
 738                  * through with the exec.  We must hold mmap_sem around
 739                  * checking core_state and changing tsk->mm.
 740                  */
 741                 down_read(&old_mm->mmap_sem);
 742                 if (unlikely(old_mm->core_state)) {
 743                         up_read(&old_mm->mmap_sem);
 744                         return -EINTR;
 745                 }
 746         }
 747         task_lock(tsk);
 748         active_mm = tsk->active_mm;
 749         tsk->mm = mm;
 750         tsk->active_mm = mm;
 751         activate_mm(active_mm, mm);
 752         task_unlock(tsk);
 753         arch_pick_mmap_layout(mm);
 754         if (old_mm) {
 755                 up_read(&old_mm->mmap_sem);
 756                 BUG_ON(active_mm != old_mm);
 757                 mm_update_next_owner(old_mm);
 758                 mmput(old_mm);
 759                 return 0;
 760         }
 761         mmdrop(active_mm);
 762         return 0;
 763 }
 764
 765 /*
 766  * This function makes sure the current process has its own signal table,
 767  * so that flush_signal_handlers can later reset the handlers without
 768  * disturbing other processes.  (Other processes might share the signal
 769  * table via the CLONE_SIGHAND option to clone().)
 770  */
 771 static int de_thread(struct task_struct *tsk)
 772 {
 773         struct signal_struct *sig = tsk->signal;
 774         struct sighand_struct *oldsighand = tsk->sighand;
 775         spinlock_t *lock = &oldsighand->siglock;
 776         int count;
 777
 778         if (thread_group_empty(tsk))
 779                 goto no_thread_group;
 780
 781         /*
 782          * Kill all other threads in the thread group.
 783          */
 784         spin_lock_irq(lock);
 785         if (signal_group_exit(sig)) {
 786                 /*
 787                  * Another group action in progress, just
 788                  * return so that the signal is processed.
 789                  */
 790                 spin_unlock_irq(lock);
 791                 return -EAGAIN;
 792         }
 793         sig->group_exit_task = tsk;
 794         zap_other_threads(tsk);
 795
 796         /* Account for the thread group leader hanging around: */
 797         count = thread_group_leader(tsk) ? 1 : 2;
 798         sig->notify_count = count;
 799         while (atomic_read(&sig->count) > count) {
 800                 __set_current_state(TASK_UNINTERRUPTIBLE);
 801                 spin_unlock_irq(lock);
 802                 schedule();
 803                 spin_lock_irq(lock);
 804         }
 805         spin_unlock_irq(lock);
 806
 807         /*
 808          * At this point all other threads have exited, all we have to
 809          * do is to wait for the thread group leader to become inactive,
 810          * and to assume its PID:
 811          */
 812         if (!thread_group_leader(tsk)) {
 813                 struct task_struct *leader = tsk->group_leader;
 814
 815                 sig->notify_count = -1; /* for exit_notify() */
 816                 for (;;) {
 817                         write_lock_irq(&tasklist_lock);
 818                         if (likely(leader->exit_state))
 819                                 break;
 820                         __set_current_state(TASK_UNINTERRUPTIBLE);
 821                         write_unlock_irq(&tasklist_lock);
 822                         schedule();
 823                 }
 824
 825                 /*
 826                  * The only record we have of the real-time age of a
 827                  * process, regardless of execs it's done, is start_time.
 828                  * All the past CPU time is accumulated in signal_struct
 829                  * from sister threads now dead.  But in this non-leader
 830                  * exec, nothing survives from the original leader thread,
 831                  * whose birth marks the true age of this process now.
 832                  * When we take on its identity by switching to its PID, we
 833                  * also take its birthdate (always earlier than our own).
 834                  */
 835                 tsk->start_time = leader->start_time;
 836
 837                 BUG_ON(!same_thread_group(leader, tsk));
 838                 BUG_ON(has_group_leader_pid(tsk));
 839                 /*
 840                  * An exec() starts a new thread group with the
 841                  * TGID of the previous thread group. Rehash the
 842                  * two threads with a switched PID, and release
 843                  * the former thread group leader:
 844                  */
 845
 846                 /* Become a process group leader with the old leader's pid.
 847                  * The old leader becomes a thread of the this thread group.
 848                  * Note: The old leader also uses this pid until release_task
 849                  *       is called.  Odd but simple and correct.
 850                  */
 851                 detach_pid(tsk, PIDTYPE_PID);
 852                 tsk->pid = leader->pid;
 853                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 854                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 855                 transfer_pid(leader, tsk, PIDTYPE_SID);
 856                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 857
 858                 tsk->group_leader = tsk;
 859                 leader->group_leader = tsk;
 860
 861                 tsk->exit_signal = SIGCHLD;
 862
 863                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 864                 leader->exit_state = EXIT_DEAD;
 865                 write_unlock_irq(&tasklist_lock);
 866
 867                 release_task(leader);
 868         }
 869
 870         sig->group_exit_task = NULL;
 871         sig->notify_count = 0;
 872
 873 no_thread_group:
 874         exit_itimers(sig);
 875         flush_itimer_signals();
 876
 877         if (atomic_read(&oldsighand->count) != 1) {
 878                 struct sighand_struct *newsighand;
 879                 /*
 880                  * This ->sighand is shared with the CLONE_SIGHAND
 881                  * but not CLONE_THREAD task, switch to the new one.
 882                  */
 883                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 884                 if (!newsighand)
 885                         return -ENOMEM;
 886
 887                 atomic_set(&newsighand->count, 1);
 888                 memcpy(newsighand->action, oldsighand->action,
 889                        sizeof(newsighand->action));
 890
 891                 write_lock_irq(&tasklist_lock);
 892                 spin_lock(&oldsighand->siglock);
 893                 rcu_assign_pointer(tsk->sighand, newsighand);
 894                 spin_unlock(&oldsighand->siglock);
 895                 write_unlock_irq(&tasklist_lock);
 896
 897                 __cleanup_sighand(oldsighand);
 898         }
 899
 900         BUG_ON(!thread_group_leader(tsk));
 901         return 0;
 902 }
 903
 904 /*
 905  * These functions flushes out all traces of the currently running executable
 906  * so that a new one can be started
 907  */
 908 static void flush_old_files(struct files_struct * files)
 909 {
 910         long j = -1;
 911         struct fdtable *fdt;
 912
 913         spin_lock(&files->file_lock);
 914         for (;;) {
 915                 unsigned long set, i;
 916
 917                 j++;
 918                 i = j * __NFDBITS;
 919                 fdt = files_fdtable(files);
 920                 if (i >= fdt->max_fds)
 921                         break;
 922                 set = fdt->close_on_exec->fds_bits[j];
 923                 if (!set)
 924                         continue;
 925                 fdt->close_on_exec->fds_bits[j] = 0;
 926                 spin_unlock(&files->file_lock);
 927                 for ( ; set ; i++,set >>= 1) {
 928                         if (set & 1) {
 929                                 sys_close(i);
 930                         }
 931                 }
 932                 spin_lock(&files->file_lock);
 933
 934         }
 935         spin_unlock(&files->file_lock);
 936 }
 937
 938 char *get_task_comm(char *buf, struct task_struct *tsk)
 939 {
 940         /* buf must be at least sizeof(tsk->comm) in size */
 941         task_lock(tsk);
 942         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 943         task_unlock(tsk);
 944         return buf;
 945 }
 946
 947 void set_task_comm(struct task_struct *tsk, char *buf)
 948 {
 949         task_lock(tsk);
 950         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 951         task_unlock(tsk);
 952 }
 953
 954 int flush_old_exec(struct linux_binprm * bprm)
 955 {
 956         char * name;
 957         int i, ch, retval;
 958         char tcomm[sizeof(current->comm)];
 959
 960         /*
 961          * Make sure we have a private signal table and that
 962          * we are unassociated from the previous thread group.
 963          */
 964         retval = de_thread(current);
 965         if (retval)
 966                 goto out;
 967
 968         set_mm_exe_file(bprm->mm, bprm->file);
 969
 970         /*
 971          * Release all of the old mmap stuff
 972          */
 973         retval = exec_mmap(bprm->mm);
 974         if (retval)
 975                 goto out;
 976
 977         bprm->mm = NULL;                /* We're using it now */
 978
 979         /* This is the point of no return */
 980         current->sas_ss_sp = current->sas_ss_size = 0;
 981
 982         if (current_euid() == current_uid() && current_egid() == current_gid())
 983                 set_dumpable(current->mm, 1);
 984         else
 985                 set_dumpable(current->mm, suid_dumpable);
 986
 987         name = bprm->filename;
 988
 989         /* Copies the binary name from after last slash */
 990         for (i=0; (ch = *(name++)) != '\0';) {
 991                 if (ch == '/')
 992                         i = 0; /* overwrite what we wrote */
 993                 else
 994                         if (i < (sizeof(tcomm) - 1))
 995                                 tcomm[i++] = ch;
 996         }
 997         tcomm[i] = '\0';
 998         set_task_comm(current, tcomm);
 999
1000         current->flags &= ~PF_RANDOMIZE;
1001         flush_thread();
1002
1003         /* Set the new mm task size. We have to do that late because it may
1004          * depend on TIF_32BIT which is only updated in flush_thread() on
1005          * some architectures like powerpc
1006          */
1007         current->mm->task_size = TASK_SIZE;
1008
1009         /* install the new credentials */
1010         if (bprm->cred->uid != current_euid() ||
1011             bprm->cred->gid != current_egid()) {
1012                 current->pdeath_signal = 0;
1013         } else if (file_permission(bprm->file, MAY_READ) ||
1014                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1015                 set_dumpable(current->mm, suid_dumpable);
1016         }
1017
1018         current->personality &= ~bprm->per_clear;
1019
1020         /* An exec changes our domain. We are no longer part of the thread
1021            group */
1022
1023         current->self_exec_id++;
1024
1025         flush_signal_handlers(current, 0);
1026         flush_old_files(current->files);
1027
1028         return 0;
1029
1030 out:
1031         return retval;
1032 }
1033
1034 EXPORT_SYMBOL(flush_old_exec);
1035
1036 /*
1037  * install the new credentials for this executable
1038  */
1039 void install_exec_creds(struct linux_binprm *bprm)
1040 {
1041         security_bprm_committing_creds(bprm);
1042
1043         commit_creds(bprm->cred);
1044         bprm->cred = NULL;
1045
1046         /* cred_exec_mutex must be held at least to this point to prevent
1047          * ptrace_attach() from altering our determination of the task's
1048          * credentials; any time after this it may be unlocked */
1049
1050         security_bprm_committed_creds(bprm);
1051 }
1052 EXPORT_SYMBOL(install_exec_creds);
1053
1054 /*
1055  * determine how safe it is to execute the proposed program
1056  * - the caller must hold current->cred_exec_mutex to protect against
1057  *   PTRACE_ATTACH
1058  */
1059 void check_unsafe_exec(struct linux_binprm *bprm)
1060 {
1061         struct task_struct *p = current;
1062
1063         bprm->unsafe = tracehook_unsafe_exec(p);
1064
1065         if (atomic_read(&p->fs->count) > 1 ||
1066             atomic_read(&p->files->count) > 1 ||
1067             atomic_read(&p->sighand->count) > 1)
1068                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1069 }
1070
1071 /*
1072  * Fill the binprm structure from the inode.
1073  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1074  *
1075  * This may be called multiple times for binary chains (scripts for example).
1076  */
1077 int prepare_binprm(struct linux_binprm *bprm)
1078 {
1079         umode_t mode;
1080         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1081         int retval;
1082
1083         mode = inode->i_mode;
1084         if (bprm->file->f_op == NULL)
1085                 return -EACCES;
1086
1087         /* clear any previous set[ug]id data from a previous binary */
1088         bprm->cred->euid = current_euid();
1089         bprm->cred->egid = current_egid();
1090
1091         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1092                 /* Set-uid? */
1093                 if (mode & S_ISUID) {
1094                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1095                         bprm->cred->euid = inode->i_uid;
1096                 }
1097
1098                 /* Set-gid? */
1099                 /*
1100                  * If setgid is set but no group execute bit then this
1101                  * is a candidate for mandatory locking, not a setgid
1102                  * executable.
1103                  */
1104                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1105                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1106                         bprm->cred->egid = inode->i_gid;
1107                 }
1108         }
1109
1110         /* fill in binprm security blob */
1111         retval = security_bprm_set_creds(bprm);
1112         if (retval)
1113                 return retval;
1114         bprm->cred_prepared = 1;
1115
1116         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1117         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1118 }
1119
1120 EXPORT_SYMBOL(prepare_binprm);
1121
1122 /*
1123  * Arguments are '\0' separated strings found at the location bprm->p
1124  * points to; chop off the first by relocating brpm->p to right after
1125  * the first '\0' encountered.
1126  */
1127 int remove_arg_zero(struct linux_binprm *bprm)
1128 {
1129         int ret = 0;
1130         unsigned long offset;
1131         char *kaddr;
1132         struct page *page;
1133
1134         if (!bprm->argc)
1135                 return 0;
1136
1137         do {
1138                 offset = bprm->p & ~PAGE_MASK;
1139                 page = get_arg_page(bprm, bprm->p, 0);
1140                 if (!page) {
1141                         ret = -EFAULT;
1142                         goto out;
1143                 }
1144                 kaddr = kmap_atomic(page, KM_USER0);
1145
1146                 for (; offset < PAGE_SIZE && kaddr[offset];
1147                                 offset++, bprm->p++)
1148                         ;
1149
1150                 kunmap_atomic(kaddr, KM_USER0);
1151                 put_arg_page(page);
1152
1153                 if (offset == PAGE_SIZE)
1154                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1155         } while (offset == PAGE_SIZE);
1156
1157         bprm->p++;
1158         bprm->argc--;
1159         ret = 0;
1160
1161 out:
1162         return ret;
1163 }
1164 EXPORT_SYMBOL(remove_arg_zero);
1165
1166 /*
1167  * cycle the list of binary formats handler, until one recognizes the image
1168  */
1169 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1170 {
1171         unsigned int depth = bprm->recursion_depth;
1172         int try,retval;
1173         struct linux_binfmt *fmt;
1174 #ifdef __alpha__
1175         /* handle /sbin/loader.. */
1176         {
1177             struct exec * eh = (struct exec *) bprm->buf;
1178
1179             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1180                 (eh->fh.f_flags & 0x3000) == 0x3000)
1181             {
1182                 struct file * file;
1183                 unsigned long loader;
1184
1185                 allow_write_access(bprm->file);
1186                 fput(bprm->file);
1187                 bprm->file = NULL;
1188
1189                 loader = bprm->vma->vm_end - sizeof(void *);
1190
1191                 file = open_exec("/sbin/loader");
1192                 retval = PTR_ERR(file);
1193                 if (IS_ERR(file))
1194                         return retval;
1195
1196                 /* Remember if the application is TASO.  */
1197                 bprm->taso = eh->ah.entry < 0x100000000UL;
1198
1199                 bprm->file = file;
1200                 bprm->loader = loader;
1201                 retval = prepare_binprm(bprm);
1202                 if (retval<0)
1203                         return retval;
1204                 /* should call search_binary_handler recursively here,
1205                    but it does not matter */
1206             }
1207         }
1208 #endif
1209         retval = security_bprm_check(bprm);
1210         if (retval)
1211                 return retval;
1212
1213         /* kernel module loader fixup */
1214         /* so we don't try to load run modprobe in kernel space. */
1215         set_fs(USER_DS);
1216
1217         retval = audit_bprm(bprm);
1218         if (retval)
1219                 return retval;
1220
1221         retval = -ENOENT;
1222         for (try=0; try<2; try++) {
1223                 read_lock(&binfmt_lock);
1224                 list_for_each_entry(fmt, &formats, lh) {
1225                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1226                         if (!fn)
1227                                 continue;
1228                         if (!try_module_get(fmt->module))
1229                                 continue;
1230                         read_unlock(&binfmt_lock);
1231                         retval = fn(bprm, regs);
1232                         /*
1233                          * Restore the depth counter to its starting value
1234                          * in this call, so we don't have to rely on every
1235                          * load_binary function to restore it on return.
1236                          */
1237                         bprm->recursion_depth = depth;
1238                         if (retval >= 0) {
1239                                 if (depth == 0)
1240                                         tracehook_report_exec(fmt, bprm, regs);
1241                                 put_binfmt(fmt);
1242                                 allow_write_access(bprm->file);
1243                                 if (bprm->file)
1244                                         fput(bprm->file);
1245                                 bprm->file = NULL;
1246                                 current->did_exec = 1;
1247                                 proc_exec_connector(current);
1248                                 return retval;
1249                         }
1250                         read_lock(&binfmt_lock);
1251                         put_binfmt(fmt);
1252                         if (retval != -ENOEXEC || bprm->mm == NULL)
1253                                 break;
1254                         if (!bprm->file) {
1255                                 read_unlock(&binfmt_lock);
1256                                 return retval;
1257                         }
1258                 }
1259                 read_unlock(&binfmt_lock);
1260                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1261                         break;
1262 #ifdef CONFIG_MODULES
1263                 } else {
1264 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1265                         if (printable(bprm->buf[0]) &&
1266                             printable(bprm->buf[1]) &&
1267                             printable(bprm->buf[2]) &&
1268                             printable(bprm->buf[3]))
1269                                 break; /* -ENOEXEC */
1270                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1271 #endif
1272                 }
1273         }
1274         return retval;
1275 }
1276
1277 EXPORT_SYMBOL(search_binary_handler);
1278
1279 void free_bprm(struct linux_binprm *bprm)
1280 {
1281         free_arg_pages(bprm);
1282         if (bprm->cred)
1283                 abort_creds(bprm->cred);
1284         kfree(bprm);
1285 }
1286
1287 /*
1288  * sys_execve() executes a new program.
1289  */
1290 int do_execve(char * filename,
1291         char __user *__user *argv,
1292         char __user *__user *envp,
1293         struct pt_regs * regs)
1294 {
1295         struct linux_binprm *bprm;
1296         struct file *file;
1297         struct files_struct *displaced;
1298         int retval;
1299
1300         retval = unshare_files(&displaced);
1301         if (retval)
1302                 goto out_ret;
1303
1304         retval = -ENOMEM;
1305         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1306         if (!bprm)
1307                 goto out_files;
1308
1309         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1310         if (retval < 0)
1311                 goto out_free;
1312
1313         retval = -ENOMEM;
1314         bprm->cred = prepare_exec_creds();
1315         if (!bprm->cred)
1316                 goto out_unlock;
1317         check_unsafe_exec(bprm);
1318
1319         file = open_exec(filename);
1320         retval = PTR_ERR(file);
1321         if (IS_ERR(file))
1322                 goto out_unlock;
1323
1324         sched_exec();
1325
1326         bprm->file = file;
1327         bprm->filename = filename;
1328         bprm->interp = filename;
1329
1330         retval = bprm_mm_init(bprm);
1331         if (retval)
1332                 goto out_file;
1333
1334         bprm->argc = count(argv, MAX_ARG_STRINGS);
1335         if ((retval = bprm->argc) < 0)
1336                 goto out;
1337
1338         bprm->envc = count(envp, MAX_ARG_STRINGS);
1339         if ((retval = bprm->envc) < 0)
1340                 goto out;
1341
1342         retval = prepare_binprm(bprm);
1343         if (retval < 0)
1344                 goto out;
1345
1346         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1347         if (retval < 0)
1348                 goto out;
1349
1350         bprm->exec = bprm->p;
1351         retval = copy_strings(bprm->envc, envp, bprm);
1352         if (retval < 0)
1353                 goto out;
1354
1355         retval = copy_strings(bprm->argc, argv, bprm);
1356         if (retval < 0)
1357                 goto out;
1358
1359         current->flags &= ~PF_KTHREAD;
1360         retval = search_binary_handler(bprm,regs);
1361         if (retval < 0)
1362                 goto out;
1363
1364         /* execve succeeded */
1365         mutex_unlock(&current->cred_exec_mutex);
1366         acct_update_integrals(current);
1367         free_bprm(bprm);
1368         if (displaced)
1369                 put_files_struct(displaced);
1370         return retval;
1371
1372 out:
1373         if (bprm->mm)
1374                 mmput (bprm->mm);
1375
1376 out_file:
1377         if (bprm->file) {
1378                 allow_write_access(bprm->file);
1379                 fput(bprm->file);
1380         }
1381
1382 out_unlock:
1383         mutex_unlock(&current->cred_exec_mutex);
1384
1385 out_free:
1386         free_bprm(bprm);
1387
1388 out_files:
1389         if (displaced)
1390                 reset_files_struct(displaced);
1391 out_ret:
1392         return retval;
1393 }
1394
1395 int set_binfmt(struct linux_binfmt *new)
1396 {
1397         struct linux_binfmt *old = current->binfmt;
1398
1399         if (new) {
1400                 if (!try_module_get(new->module))
1401                         return -1;
1402         }
1403         current->binfmt = new;
1404         if (old)
1405                 module_put(old->module);
1406         return 0;
1407 }
1408
1409 EXPORT_SYMBOL(set_binfmt);
1410
1411 /* format_corename will inspect the pattern parameter, and output a
1412  * name into corename, which must have space for at least
1413  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1414  */
1415 static int format_corename(char *corename, long signr)
1416 {
1417         const struct cred *cred = current_cred();
1418         const char *pat_ptr = core_pattern;
1419         int ispipe = (*pat_ptr == '|');
1420         char *out_ptr = corename;
1421         char *const out_end = corename + CORENAME_MAX_SIZE;
1422         int rc;
1423         int pid_in_pattern = 0;
1424
1425         /* Repeat as long as we have more pattern to process and more output
1426            space */
1427         while (*pat_ptr) {
1428                 if (*pat_ptr != '%') {
1429                         if (out_ptr == out_end)
1430                                 goto out;
1431                         *out_ptr++ = *pat_ptr++;
1432                 } else {
1433                         switch (*++pat_ptr) {
1434                         case 0:
1435                                 goto out;
1436                         /* Double percent, output one percent */
1437                         case '%':
1438                                 if (out_ptr == out_end)
1439                                         goto out;
1440                                 *out_ptr++ = '%';
1441                                 break;
1442                         /* pid */
1443                         case 'p':
1444                                 pid_in_pattern = 1;
1445                                 rc = snprintf(out_ptr, out_end - out_ptr,
1446                                               "%d", task_tgid_vnr(current));
1447                                 if (rc > out_end - out_ptr)
1448                                         goto out;
1449                                 out_ptr += rc;
1450                                 break;
1451                         /* uid */
1452                         case 'u':
1453                                 rc = snprintf(out_ptr, out_end - out_ptr,
1454                                               "%d", cred->uid);
1455                                 if (rc > out_end - out_ptr)
1456                                         goto out;
1457                                 out_ptr += rc;
1458                                 break;
1459                         /* gid */
1460                         case 'g':
1461                                 rc = snprintf(out_ptr, out_end - out_ptr,
1462                                               "%d", cred->gid);
1463                                 if (rc > out_end - out_ptr)
1464                                         goto out;
1465                                 out_ptr += rc;
1466                                 break;
1467                         /* signal that caused the coredump */
1468                         case 's':
1469                                 rc = snprintf(out_ptr, out_end - out_ptr,
1470                                               "%ld", signr);
1471                                 if (rc > out_end - out_ptr)
1472                                         goto out;
1473                                 out_ptr += rc;
1474                                 break;
1475                         /* UNIX time of coredump */
1476                         case 't': {
1477                                 struct timeval tv;
1478                                 do_gettimeofday(&tv);
1479                                 rc = snprintf(out_ptr, out_end - out_ptr,
1480                                               "%lu", tv.tv_sec);
1481                                 if (rc > out_end - out_ptr)
1482                                         goto out;
1483                                 out_ptr += rc;
1484                                 break;
1485                         }
1486                         /* hostname */
1487                         case 'h':
1488                                 down_read(&uts_sem);
1489                                 rc = snprintf(out_ptr, out_end - out_ptr,
1490                                               "%s", utsname()->nodename);
1491                                 up_read(&uts_sem);
1492                                 if (rc > out_end - out_ptr)
1493                                         goto out;
1494                                 out_ptr += rc;
1495                                 break;
1496                         /* executable */
1497                         case 'e':
1498                                 rc = snprintf(out_ptr, out_end - out_ptr,
1499                                               "%s", current->comm);
1500                                 if (rc > out_end - out_ptr)
1501                                         goto out;
1502                                 out_ptr += rc;
1503                                 break;
1504                         /* core limit size */
1505                         case 'c':
1506                                 rc = snprintf(out_ptr, out_end - out_ptr,
1507                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1508                                 if (rc > out_end - out_ptr)
1509                                         goto out;
1510                                 out_ptr += rc;
1511                                 break;
1512                         default:
1513                                 break;
1514                         }
1515                         ++pat_ptr;
1516                 }
1517         }
1518         /* Backward compatibility with core_uses_pid:
1519          *
1520          * If core_pattern does not include a %p (as is the default)
1521          * and core_uses_pid is set, then .%pid will be appended to
1522          * the filename. Do not do this for piped commands. */
1523         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1524                 rc = snprintf(out_ptr, out_end - out_ptr,
1525                               ".%d", task_tgid_vnr(current));
1526                 if (rc > out_end - out_ptr)
1527                         goto out;
1528                 out_ptr += rc;
1529         }
1530 out:
1531         *out_ptr = 0;
1532         return ispipe;
1533 }
1534
1535 static int zap_process(struct task_struct *start)
1536 {
1537         struct task_struct *t;
1538         int nr = 0;
1539
1540         start->signal->flags = SIGNAL_GROUP_EXIT;
1541         start->signal->group_stop_count = 0;
1542
1543         t = start;
1544         do {
1545                 if (t != current && t->mm) {
1546                         sigaddset(&t->pending.signal, SIGKILL);
1547                         signal_wake_up(t, 1);
1548                         nr++;
1549                 }
1550         } while_each_thread(start, t);
1551
1552         return nr;
1553 }
1554
1555 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1556                                 struct core_state *core_state, int exit_code)
1557 {
1558         struct task_struct *g, *p;
1559         unsigned long flags;
1560         int nr = -EAGAIN;
1561
1562         spin_lock_irq(&tsk->sighand->siglock);
1563         if (!signal_group_exit(tsk->signal)) {
1564                 mm->core_state = core_state;
1565                 tsk->signal->group_exit_code = exit_code;
1566                 nr = zap_process(tsk);
1567         }
1568         spin_unlock_irq(&tsk->sighand->siglock);
1569         if (unlikely(nr < 0))
1570                 return nr;
1571
1572         if (atomic_read(&mm->mm_users) == nr + 1)
1573                 goto done;
1574         /*
1575          * We should find and kill all tasks which use this mm, and we should
1576          * count them correctly into ->nr_threads. We don't take tasklist
1577          * lock, but this is safe wrt:
1578          *
1579          * fork:
1580          *      None of sub-threads can fork after zap_process(leader). All
1581          *      processes which were created before this point should be
1582          *      visible to zap_threads() because copy_process() adds the new
1583          *      process to the tail of init_task.tasks list, and lock/unlock
1584          *      of ->siglock provides a memory barrier.
1585          *
1586          * do_exit:
1587          *      The caller holds mm->mmap_sem. This means that the task which
1588          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1589          *      its ->mm.
1590          *
1591          * de_thread:
1592          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1593          *      we must see either old or new leader, this does not matter.
1594          *      However, it can change p->sighand, so lock_task_sighand(p)
1595          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1596          *      it can't fail.
1597          *
1598          *      Note also that "g" can be the old leader with ->mm == NULL
1599          *      and already unhashed and thus removed from ->thread_group.
1600          *      This is OK, __unhash_process()->list_del_rcu() does not
1601          *      clear the ->next pointer, we will find the new leader via
1602          *      next_thread().
1603          */
1604         rcu_read_lock();
1605         for_each_process(g) {
1606                 if (g == tsk->group_leader)
1607                         continue;
1608                 if (g->flags & PF_KTHREAD)
1609                         continue;
1610                 p = g;
1611                 do {
1612                         if (p->mm) {
1613                                 if (unlikely(p->mm == mm)) {
1614                                         lock_task_sighand(p, &flags);
1615                                         nr += zap_process(p);
1616                                         unlock_task_sighand(p, &flags);
1617                                 }
1618                                 break;
1619                         }
1620                 } while_each_thread(g, p);
1621         }
1622         rcu_read_unlock();
1623 done:
1624         atomic_set(&core_state->nr_threads, nr);
1625         return nr;
1626 }
1627
1628 static int coredump_wait(int exit_code, struct core_state *core_state)
1629 {
1630         struct task_struct *tsk = current;
1631         struct mm_struct *mm = tsk->mm;
1632         struct completion *vfork_done;
1633         int core_waiters;
1634
1635         init_completion(&core_state->startup);
1636         core_state->dumper.task = tsk;
1637         core_state->dumper.next = NULL;
1638         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1639         up_write(&mm->mmap_sem);
1640
1641         if (unlikely(core_waiters < 0))
1642                 goto fail;
1643
1644         /*
1645          * Make sure nobody is waiting for us to release the VM,
1646          * otherwise we can deadlock when we wait on each other
1647          */
1648         vfork_done = tsk->vfork_done;
1649         if (vfork_done) {
1650                 tsk->vfork_done = NULL;
1651                 complete(vfork_done);
1652         }
1653
1654         if (core_waiters)
1655                 wait_for_completion(&core_state->startup);
1656 fail:
1657         return core_waiters;
1658 }
1659
1660 static void coredump_finish(struct mm_struct *mm)
1661 {
1662         struct core_thread *curr, *next;
1663         struct task_struct *task;
1664
1665         next = mm->core_state->dumper.next;
1666         while ((curr = next) != NULL) {
1667                 next = curr->next;
1668                 task = curr->task;
1669                 /*
1670                  * see exit_mm(), curr->task must not see
1671                  * ->task == NULL before we read ->next.
1672                  */
1673                 smp_mb();
1674                 curr->task = NULL;
1675                 wake_up_process(task);
1676         }
1677
1678         mm->core_state = NULL;
1679 }
1680
1681 /*
1682  * set_dumpable converts traditional three-value dumpable to two flags and
1683  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1684  * these bits are not changed atomically.  So get_dumpable can observe the
1685  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1686  * return either old dumpable or new one by paying attention to the order of
1687  * modifying the bits.
1688  *
1689  * dumpable |   mm->flags (binary)
1690  * old  new | initial interim  final
1691  * ---------+-----------------------
1692  *  0    1  |   00      01      01
1693  *  0    2  |   00      10(*)   11
1694  *  1    0  |   01      00      00
1695  *  1    2  |   01      11      11
1696  *  2    0  |   11      10(*)   00
1697  *  2    1  |   11      11      01
1698  *
1699  * (*) get_dumpable regards interim value of 10 as 11.
1700  */
1701 void set_dumpable(struct mm_struct *mm, int value)
1702 {
1703         switch (value) {
1704         case 0:
1705                 clear_bit(MMF_DUMPABLE, &mm->flags);
1706                 smp_wmb();
1707                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1708                 break;
1709         case 1:
1710                 set_bit(MMF_DUMPABLE, &mm->flags);
1711                 smp_wmb();
1712                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1713                 break;
1714         case 2:
1715                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1716                 smp_wmb();
1717                 set_bit(MMF_DUMPABLE, &mm->flags);
1718                 break;
1719         }
1720 }
1721
1722 int get_dumpable(struct mm_struct *mm)
1723 {
1724         int ret;
1725
1726         ret = mm->flags & 0x3;
1727         return (ret >= 2) ? 2 : ret;
1728 }
1729
1730 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1731 {
1732         struct core_state core_state;
1733         char corename[CORENAME_MAX_SIZE + 1];
1734         struct mm_struct *mm = current->mm;
1735         struct linux_binfmt * binfmt;
1736         struct inode * inode;
1737         struct file * file;
1738         const struct cred *old_cred;
1739         struct cred *cred;
1740         int retval = 0;
1741         int flag = 0;
1742         int ispipe = 0;
1743         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1744         char **helper_argv = NULL;
1745         int helper_argc = 0;
1746         char *delimit;
1747
1748         audit_core_dumps(signr);
1749
1750         binfmt = current->binfmt;
1751         if (!binfmt || !binfmt->core_dump)
1752                 goto fail;
1753
1754         cred = prepare_creds();
1755         if (!cred) {
1756                 retval = -ENOMEM;
1757                 goto fail;
1758         }
1759
1760         down_write(&mm->mmap_sem);
1761         /*
1762          * If another thread got here first, or we are not dumpable, bail out.
1763          */
1764         if (mm->core_state || !get_dumpable(mm)) {
1765                 up_write(&mm->mmap_sem);
1766                 put_cred(cred);
1767                 goto fail;
1768         }
1769
1770         /*
1771          *      We cannot trust fsuid as being the "true" uid of the
1772          *      process nor do we know its entire history. We only know it
1773          *      was tainted so we dump it as root in mode 2.
1774          */
1775         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1776                 flag = O_EXCL;          /* Stop rewrite attacks */
1777                 cred->fsuid = 0;        /* Dump root private */
1778         }
1779
1780         retval = coredump_wait(exit_code, &core_state);
1781         if (retval < 0) {
1782                 put_cred(cred);
1783                 goto fail;
1784         }
1785
1786         old_cred = override_creds(cred);
1787
1788         /*
1789          * Clear any false indication of pending signals that might
1790          * be seen by the filesystem code called to write the core file.
1791          */
1792         clear_thread_flag(TIF_SIGPENDING);
1793
1794         /*
1795          * lock_kernel() because format_corename() is controlled by sysctl, which
1796          * uses lock_kernel()
1797          */
1798         lock_kernel();
1799         ispipe = format_corename(corename, signr);
1800         unlock_kernel();
1801         /*
1802          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1803          * to a pipe.  Since we're not writing directly to the filesystem
1804          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1805          * created unless the pipe reader choses to write out the core file
1806          * at which point file size limits and permissions will be imposed
1807          * as it does with any other process
1808          */
1809         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1810                 goto fail_unlock;
1811
1812         if (ispipe) {
1813                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1814                 /* Terminate the string before the first option */
1815                 delimit = strchr(corename, ' ');
1816                 if (delimit)
1817                         *delimit = '\0';
1818                 delimit = strrchr(helper_argv[0], '/');
1819                 if (delimit)
1820                         delimit++;
1821                 else
1822                         delimit = helper_argv[0];
1823                 if (!strcmp(delimit, current->comm)) {
1824                         printk(KERN_NOTICE "Recursive core dump detected, "
1825                                         "aborting\n");
1826                         goto fail_unlock;
1827                 }
1828
1829                 core_limit = RLIM_INFINITY;
1830
1831                 /* SIGPIPE can happen, but it's just never processed */
1832                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1833                                 &file)) {
1834                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1835                                corename);
1836                         goto fail_unlock;
1837                 }
1838         } else
1839                 file = filp_open(corename,
1840                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1841                                  0600);
1842         if (IS_ERR(file))
1843                 goto fail_unlock;
1844         inode = file->f_path.dentry->d_inode;
1845         if (inode->i_nlink > 1)
1846                 goto close_fail;        /* multiple links - don't dump */
1847         if (!ispipe && d_unhashed(file->f_path.dentry))
1848                 goto close_fail;
1849
1850         /* AK: actually i see no reason to not allow this for named pipes etc.,
1851            but keep the previous behaviour for now. */
1852         if (!ispipe && !S_ISREG(inode->i_mode))
1853                 goto close_fail;
1854         /*
1855          * Dont allow local users get cute and trick others to coredump
1856          * into their pre-created files:
1857          */
1858         if (inode->i_uid != current_fsuid())
1859                 goto close_fail;
1860         if (!file->f_op)
1861                 goto close_fail;
1862         if (!file->f_op->write)
1863                 goto close_fail;
1864         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1865                 goto close_fail;
1866
1867         retval = binfmt->core_dump(signr, regs, file, core_limit);
1868
1869         if (retval)
1870                 current->signal->group_exit_code |= 0x80;
1871 close_fail:
1872         filp_close(file, NULL);
1873 fail_unlock:
1874         if (helper_argv)
1875                 argv_free(helper_argv);
1876
1877         revert_creds(old_cred);
1878         put_cred(cred);
1879         coredump_finish(mm);
1880 fail:
1881         return retval;
1882 }