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