fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_counter.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/proc_fs.h>
  47 #include <linux/mount.h>
  48 #include <linux/security.h>
  49 #include <linux/syscalls.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/tracehook.h>
  54 #include <linux/kmod.h>
  55
  56 #include <asm/uaccess.h>
  57 #include <asm/mmu_context.h>
  58 #include <asm/tlb.h>
  59
  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         struct task_struct *leader = NULL;
 777         int count;
 778
 779         if (thread_group_empty(tsk))
 780                 goto no_thread_group;
 781
 782         /*
 783          * Kill all other threads in the thread group.
 784          */
 785         spin_lock_irq(lock);
 786         if (signal_group_exit(sig)) {
 787                 /*
 788                  * Another group action in progress, just
 789                  * return so that the signal is processed.
 790                  */
 791                 spin_unlock_irq(lock);
 792                 return -EAGAIN;
 793         }
 794         sig->group_exit_task = tsk;
 795         zap_other_threads(tsk);
 796
 797         /* Account for the thread group leader hanging around: */
 798         count = thread_group_leader(tsk) ? 1 : 2;
 799         sig->notify_count = count;
 800         while (atomic_read(&sig->count) > count) {
 801                 __set_current_state(TASK_UNINTERRUPTIBLE);
 802                 spin_unlock_irq(lock);
 803                 schedule();
 804                 spin_lock_irq(lock);
 805         }
 806         spin_unlock_irq(lock);
 807
 808         /*
 809          * At this point all other threads have exited, all we have to
 810          * do is to wait for the thread group leader to become inactive,
 811          * and to assume its PID:
 812          */
 813         if (!thread_group_leader(tsk)) {
 814                 leader = tsk->group_leader;
 815
 816                 sig->notify_count = -1; /* for exit_notify() */
 817                 for (;;) {
 818                         write_lock_irq(&tasklist_lock);
 819                         if (likely(leader->exit_state))
 820                                 break;
 821                         __set_current_state(TASK_UNINTERRUPTIBLE);
 822                         write_unlock_irq(&tasklist_lock);
 823                         schedule();
 824                 }
 825
 826                 /*
 827                  * The only record we have of the real-time age of a
 828                  * process, regardless of execs it's done, is start_time.
 829                  * All the past CPU time is accumulated in signal_struct
 830                  * from sister threads now dead.  But in this non-leader
 831                  * exec, nothing survives from the original leader thread,
 832                  * whose birth marks the true age of this process now.
 833                  * When we take on its identity by switching to its PID, we
 834                  * also take its birthdate (always earlier than our own).
 835                  */
 836                 tsk->start_time = leader->start_time;
 837
 838                 BUG_ON(!same_thread_group(leader, tsk));
 839                 BUG_ON(has_group_leader_pid(tsk));
 840                 /*
 841                  * An exec() starts a new thread group with the
 842                  * TGID of the previous thread group. Rehash the
 843                  * two threads with a switched PID, and release
 844                  * the former thread group leader:
 845                  */
 846
 847                 /* Become a process group leader with the old leader's pid.
 848                  * The old leader becomes a thread of the this thread group.
 849                  * Note: The old leader also uses this pid until release_task
 850                  *       is called.  Odd but simple and correct.
 851                  */
 852                 detach_pid(tsk, PIDTYPE_PID);
 853                 tsk->pid = leader->pid;
 854                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 855                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 856                 transfer_pid(leader, tsk, PIDTYPE_SID);
 857                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 858
 859                 tsk->group_leader = tsk;
 860                 leader->group_leader = tsk;
 861
 862                 tsk->exit_signal = SIGCHLD;
 863
 864                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 865                 leader->exit_state = EXIT_DEAD;
 866
 867                 write_unlock_irq(&tasklist_lock);
 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         if (leader)
 877                 release_task(leader);
 878
 879         if (atomic_read(&oldsighand->count) != 1) {
 880                 struct sighand_struct *newsighand;
 881                 /*
 882                  * This ->sighand is shared with the CLONE_SIGHAND
 883                  * but not CLONE_THREAD task, switch to the new one.
 884                  */
 885                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 886                 if (!newsighand)
 887                         return -ENOMEM;
 888
 889                 atomic_set(&newsighand->count, 1);
 890                 memcpy(newsighand->action, oldsighand->action,
 891                        sizeof(newsighand->action));
 892
 893                 write_lock_irq(&tasklist_lock);
 894                 spin_lock(&oldsighand->siglock);
 895                 rcu_assign_pointer(tsk->sighand, newsighand);
 896                 spin_unlock(&oldsighand->siglock);
 897                 write_unlock_irq(&tasklist_lock);
 898
 899                 __cleanup_sighand(oldsighand);
 900         }
 901
 902         BUG_ON(!thread_group_leader(tsk));
 903         return 0;
 904 }
 905
 906 /*
 907  * These functions flushes out all traces of the currently running executable
 908  * so that a new one can be started
 909  */
 910 static void flush_old_files(struct files_struct * files)
 911 {
 912         long j = -1;
 913         struct fdtable *fdt;
 914
 915         spin_lock(&files->file_lock);
 916         for (;;) {
 917                 unsigned long set, i;
 918
 919                 j++;
 920                 i = j * __NFDBITS;
 921                 fdt = files_fdtable(files);
 922                 if (i >= fdt->max_fds)
 923                         break;
 924                 set = fdt->close_on_exec->fds_bits[j];
 925                 if (!set)
 926                         continue;
 927                 fdt->close_on_exec->fds_bits[j] = 0;
 928                 spin_unlock(&files->file_lock);
 929                 for ( ; set ; i++,set >>= 1) {
 930                         if (set & 1) {
 931                                 sys_close(i);
 932                         }
 933                 }
 934                 spin_lock(&files->file_lock);
 935
 936         }
 937         spin_unlock(&files->file_lock);
 938 }
 939
 940 char *get_task_comm(char *buf, struct task_struct *tsk)
 941 {
 942         /* buf must be at least sizeof(tsk->comm) in size */
 943         task_lock(tsk);
 944         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 945         task_unlock(tsk);
 946         return buf;
 947 }
 948
 949 void set_task_comm(struct task_struct *tsk, char *buf)
 950 {
 951         task_lock(tsk);
 952         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 953         task_unlock(tsk);
 954 }
 955
 956 int flush_old_exec(struct linux_binprm * bprm)
 957 {
 958         char * name;
 959         int i, ch, retval;
 960         char tcomm[sizeof(current->comm)];
 961
 962         /*
 963          * Make sure we have a private signal table and that
 964          * we are unassociated from the previous thread group.
 965          */
 966         retval = de_thread(current);
 967         if (retval)
 968                 goto out;
 969
 970         set_mm_exe_file(bprm->mm, bprm->file);
 971
 972         /*
 973          * Release all of the old mmap stuff
 974          */
 975         retval = exec_mmap(bprm->mm);
 976         if (retval)
 977                 goto out;
 978
 979         bprm->mm = NULL;                /* We're using it now */
 980
 981         /* This is the point of no return */
 982         current->sas_ss_sp = current->sas_ss_size = 0;
 983
 984         if (current->euid == current->uid && current->egid == current->gid)
 985                 set_dumpable(current->mm, 1);
 986         else
 987                 set_dumpable(current->mm, suid_dumpable);
 988
 989         name = bprm->filename;
 990
 991         /* Copies the binary name from after last slash */
 992         for (i=0; (ch = *(name++)) != '\0';) {
 993                 if (ch == '/')
 994                         i = 0; /* overwrite what we wrote */
 995                 else
 996                         if (i < (sizeof(tcomm) - 1))
 997                                 tcomm[i++] = ch;
 998         }
 999         tcomm[i] = '\0';
1000         set_task_comm(current, tcomm);
1001
1002         current->flags &= ~PF_RANDOMIZE;
1003         flush_thread();
1004
1005         /* Set the new mm task size. We have to do that late because it may
1006          * depend on TIF_32BIT which is only updated in flush_thread() on
1007          * some architectures like powerpc
1008          */
1009         current->mm->task_size = TASK_SIZE;
1010
1011         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1012                 suid_keys(current);
1013                 set_dumpable(current->mm, suid_dumpable);
1014                 current->pdeath_signal = 0;
1015         } else if (file_permission(bprm->file, MAY_READ) ||
1016                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1017                 suid_keys(current);
1018                 set_dumpable(current->mm, suid_dumpable);
1019         }
1020
1021         /*
1022          * Flush performance counters when crossing a
1023          * security domain:
1024          */
1025         if (!get_dumpable(current->mm))
1026                 perf_counter_exit_task(current);
1027
1028         /* An exec changes our domain. We are no longer part of the thread
1029            group */
1030
1031         current->self_exec_id++;
1032
1033         flush_signal_handlers(current, 0);
1034         flush_old_files(current->files);
1035
1036         return 0;
1037
1038 out:
1039         return retval;
1040 }
1041
1042 EXPORT_SYMBOL(flush_old_exec);
1043
1044 /*
1045  * Fill the binprm structure from the inode.
1046  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1047  */
1048 int prepare_binprm(struct linux_binprm *bprm)
1049 {
1050         int mode;
1051         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1052         int retval;
1053
1054         mode = inode->i_mode;
1055         if (bprm->file->f_op == NULL)
1056                 return -EACCES;
1057
1058         bprm->e_uid = current->euid;
1059         bprm->e_gid = current->egid;
1060
1061         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1062                 /* Set-uid? */
1063                 if (mode & S_ISUID) {
1064                         current->personality &= ~PER_CLEAR_ON_SETID;
1065                         bprm->e_uid = inode->i_uid;
1066                 }
1067
1068                 /* Set-gid? */
1069                 /*
1070                  * If setgid is set but no group execute bit then this
1071                  * is a candidate for mandatory locking, not a setgid
1072                  * executable.
1073                  */
1074                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1075                         current->personality &= ~PER_CLEAR_ON_SETID;
1076                         bprm->e_gid = inode->i_gid;
1077                 }
1078         }
1079
1080         /* fill in binprm security blob */
1081         retval = security_bprm_set(bprm);
1082         if (retval)
1083                 return retval;
1084
1085         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1086         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1087 }
1088
1089 EXPORT_SYMBOL(prepare_binprm);
1090
1091 static int unsafe_exec(struct task_struct *p)
1092 {
1093         int unsafe = tracehook_unsafe_exec(p);
1094
1095         if (atomic_read(&p->fs->count) > 1 ||
1096             atomic_read(&p->files->count) > 1 ||
1097             atomic_read(&p->sighand->count) > 1)
1098                 unsafe |= LSM_UNSAFE_SHARE;
1099
1100         return unsafe;
1101 }
1102
1103 void compute_creds(struct linux_binprm *bprm)
1104 {
1105         int unsafe;
1106
1107         if (bprm->e_uid != current->uid) {
1108                 suid_keys(current);
1109                 current->pdeath_signal = 0;
1110         }
1111         exec_keys(current);
1112
1113         task_lock(current);
1114         unsafe = unsafe_exec(current);
1115         security_bprm_apply_creds(bprm, unsafe);
1116         task_unlock(current);
1117         security_bprm_post_apply_creds(bprm);
1118 }
1119 EXPORT_SYMBOL(compute_creds);
1120
1121 /*
1122  * Arguments are '\0' separated strings found at the location bprm->p
1123  * points to; chop off the first by relocating brpm->p to right after
1124  * the first '\0' encountered.
1125  */
1126 int remove_arg_zero(struct linux_binprm *bprm)
1127 {
1128         int ret = 0;
1129         unsigned long offset;
1130         char *kaddr;
1131         struct page *page;
1132
1133         if (!bprm->argc)
1134                 return 0;
1135
1136         do {
1137                 offset = bprm->p & ~PAGE_MASK;
1138                 page = get_arg_page(bprm, bprm->p, 0);
1139                 if (!page) {
1140                         ret = -EFAULT;
1141                         goto out;
1142                 }
1143                 kaddr = kmap_atomic(page, KM_USER0);
1144
1145                 for (; offset < PAGE_SIZE && kaddr[offset];
1146                                 offset++, bprm->p++)
1147                         ;
1148
1149                 kunmap_atomic(kaddr, KM_USER0);
1150                 put_arg_page(page);
1151
1152                 if (offset == PAGE_SIZE)
1153                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1154         } while (offset == PAGE_SIZE);
1155
1156         bprm->p++;
1157         bprm->argc--;
1158         ret = 0;
1159
1160 out:
1161         return ret;
1162 }
1163 EXPORT_SYMBOL(remove_arg_zero);
1164
1165 /*
1166  * cycle the list of binary formats handler, until one recognizes the image
1167  */
1168 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1169 {
1170         unsigned int depth = bprm->recursion_depth;
1171         int try,retval;
1172         struct linux_binfmt *fmt;
1173 #ifdef __alpha__
1174         /* handle /sbin/loader.. */
1175         {
1176             struct exec * eh = (struct exec *) bprm->buf;
1177
1178             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1179                 (eh->fh.f_flags & 0x3000) == 0x3000)
1180             {
1181                 struct file * file;
1182                 unsigned long loader;
1183
1184                 allow_write_access(bprm->file);
1185                 fput(bprm->file);
1186                 bprm->file = NULL;
1187
1188                 loader = bprm->vma->vm_end - sizeof(void *);
1189
1190                 file = open_exec("/sbin/loader");
1191                 retval = PTR_ERR(file);
1192                 if (IS_ERR(file))
1193                         return retval;
1194
1195                 /* Remember if the application is TASO.  */
1196                 bprm->taso = eh->ah.entry < 0x100000000UL;
1197
1198                 bprm->file = file;
1199                 bprm->loader = loader;
1200                 retval = prepare_binprm(bprm);
1201                 if (retval<0)
1202                         return retval;
1203                 /* should call search_binary_handler recursively here,
1204                    but it does not matter */
1205             }
1206         }
1207 #endif
1208         retval = security_bprm_check(bprm);
1209         if (retval)
1210                 return retval;
1211
1212         /* kernel module loader fixup */
1213         /* so we don't try to load run modprobe in kernel space. */
1214         set_fs(USER_DS);
1215
1216         retval = audit_bprm(bprm);
1217         if (retval)
1218                 return retval;
1219
1220         retval = -ENOENT;
1221         for (try=0; try<2; try++) {
1222                 read_lock(&binfmt_lock);
1223                 list_for_each_entry(fmt, &formats, lh) {
1224                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1225                         if (!fn)
1226                                 continue;
1227                         if (!try_module_get(fmt->module))
1228                                 continue;
1229                         read_unlock(&binfmt_lock);
1230                         retval = fn(bprm, regs);
1231                         /*
1232                          * Restore the depth counter to its starting value
1233                          * in this call, so we don't have to rely on every
1234                          * load_binary function to restore it on return.
1235                          */
1236                         bprm->recursion_depth = depth;
1237                         if (retval >= 0) {
1238                                 if (depth == 0)
1239                                         tracehook_report_exec(fmt, bprm, regs);
1240                                 put_binfmt(fmt);
1241                                 allow_write_access(bprm->file);
1242                                 if (bprm->file)
1243                                         fput(bprm->file);
1244                                 bprm->file = NULL;
1245                                 current->did_exec = 1;
1246                                 proc_exec_connector(current);
1247                                 return retval;
1248                         }
1249                         read_lock(&binfmt_lock);
1250                         put_binfmt(fmt);
1251                         if (retval != -ENOEXEC || bprm->mm == NULL)
1252                                 break;
1253                         if (!bprm->file) {
1254                                 read_unlock(&binfmt_lock);
1255                                 return retval;
1256                         }
1257                 }
1258                 read_unlock(&binfmt_lock);
1259                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1260                         break;
1261 #ifdef CONFIG_MODULES
1262                 } else {
1263 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1264                         if (printable(bprm->buf[0]) &&
1265                             printable(bprm->buf[1]) &&
1266                             printable(bprm->buf[2]) &&
1267                             printable(bprm->buf[3]))
1268                                 break; /* -ENOEXEC */
1269                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1270 #endif
1271                 }
1272         }
1273         return retval;
1274 }
1275
1276 EXPORT_SYMBOL(search_binary_handler);
1277
1278 void free_bprm(struct linux_binprm *bprm)
1279 {
1280         free_arg_pages(bprm);
1281         kfree(bprm);
1282 }
1283
1284 /*
1285  * sys_execve() executes a new program.
1286  */
1287 int do_execve(char * filename,
1288         char __user *__user *argv,
1289         char __user *__user *envp,
1290         struct pt_regs * regs)
1291 {
1292         struct linux_binprm *bprm;
1293         struct file *file;
1294         struct files_struct *displaced;
1295         int retval;
1296
1297         retval = unshare_files(&displaced);
1298         if (retval)
1299                 goto out_ret;
1300
1301         retval = -ENOMEM;
1302         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1303         if (!bprm)
1304                 goto out_files;
1305
1306         file = open_exec(filename);
1307         retval = PTR_ERR(file);
1308         if (IS_ERR(file))
1309                 goto out_kfree;
1310
1311         sched_exec();
1312
1313         bprm->file = file;
1314         bprm->filename = filename;
1315         bprm->interp = filename;
1316
1317         retval = bprm_mm_init(bprm);
1318         if (retval)
1319                 goto out_file;
1320
1321         bprm->argc = count(argv, MAX_ARG_STRINGS);
1322         if ((retval = bprm->argc) < 0)
1323                 goto out_mm;
1324
1325         bprm->envc = count(envp, MAX_ARG_STRINGS);
1326         if ((retval = bprm->envc) < 0)
1327                 goto out_mm;
1328
1329         retval = security_bprm_alloc(bprm);
1330         if (retval)
1331                 goto out;
1332
1333         retval = prepare_binprm(bprm);
1334         if (retval < 0)
1335                 goto out;
1336
1337         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1338         if (retval < 0)
1339                 goto out;
1340
1341         bprm->exec = bprm->p;
1342         retval = copy_strings(bprm->envc, envp, bprm);
1343         if (retval < 0)
1344                 goto out;
1345
1346         retval = copy_strings(bprm->argc, argv, bprm);
1347         if (retval < 0)
1348                 goto out;
1349
1350         current->flags &= ~PF_KTHREAD;
1351         retval = search_binary_handler(bprm,regs);
1352         if (retval >= 0) {
1353                 /* execve success */
1354                 security_bprm_free(bprm);
1355                 acct_update_integrals(current);
1356                 free_bprm(bprm);
1357                 if (displaced)
1358                         put_files_struct(displaced);
1359                 return retval;
1360         }
1361
1362 out:
1363         if (bprm->security)
1364                 security_bprm_free(bprm);
1365
1366 out_mm:
1367         if (bprm->mm)
1368                 mmput (bprm->mm);
1369
1370 out_file:
1371         if (bprm->file) {
1372                 allow_write_access(bprm->file);
1373                 fput(bprm->file);
1374         }
1375 out_kfree:
1376         free_bprm(bprm);
1377
1378 out_files:
1379         if (displaced)
1380                 reset_files_struct(displaced);
1381 out_ret:
1382         return retval;
1383 }
1384
1385 int set_binfmt(struct linux_binfmt *new)
1386 {
1387         struct linux_binfmt *old = current->binfmt;
1388
1389         if (new) {
1390                 if (!try_module_get(new->module))
1391                         return -1;
1392         }
1393         current->binfmt = new;
1394         if (old)
1395                 module_put(old->module);
1396         return 0;
1397 }
1398
1399 EXPORT_SYMBOL(set_binfmt);
1400
1401 /* format_corename will inspect the pattern parameter, and output a
1402  * name into corename, which must have space for at least
1403  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1404  */
1405 static int format_corename(char *corename, long signr)
1406 {
1407         const char *pat_ptr = core_pattern;
1408         int ispipe = (*pat_ptr == '|');
1409         char *out_ptr = corename;
1410         char *const out_end = corename + CORENAME_MAX_SIZE;
1411         int rc;
1412         int pid_in_pattern = 0;
1413
1414         /* Repeat as long as we have more pattern to process and more output
1415            space */
1416         while (*pat_ptr) {
1417                 if (*pat_ptr != '%') {
1418                         if (out_ptr == out_end)
1419                                 goto out;
1420                         *out_ptr++ = *pat_ptr++;
1421                 } else {
1422                         switch (*++pat_ptr) {
1423                         case 0:
1424                                 goto out;
1425                         /* Double percent, output one percent */
1426                         case '%':
1427                                 if (out_ptr == out_end)
1428                                         goto out;
1429                                 *out_ptr++ = '%';
1430                                 break;
1431                         /* pid */
1432                         case 'p':
1433                                 pid_in_pattern = 1;
1434                                 rc = snprintf(out_ptr, out_end - out_ptr,
1435                                               "%d", task_tgid_vnr(current));
1436                                 if (rc > out_end - out_ptr)
1437                                         goto out;
1438                                 out_ptr += rc;
1439                                 break;
1440                         /* uid */
1441                         case 'u':
1442                                 rc = snprintf(out_ptr, out_end - out_ptr,
1443                                               "%d", current->uid);
1444                                 if (rc > out_end - out_ptr)
1445                                         goto out;
1446                                 out_ptr += rc;
1447                                 break;
1448                         /* gid */
1449                         case 'g':
1450                                 rc = snprintf(out_ptr, out_end - out_ptr,
1451                                               "%d", current->gid);
1452                                 if (rc > out_end - out_ptr)
1453                                         goto out;
1454                                 out_ptr += rc;
1455                                 break;
1456                         /* signal that caused the coredump */
1457                         case 's':
1458                                 rc = snprintf(out_ptr, out_end - out_ptr,
1459                                               "%ld", signr);
1460                                 if (rc > out_end - out_ptr)
1461                                         goto out;
1462                                 out_ptr += rc;
1463                                 break;
1464                         /* UNIX time of coredump */
1465                         case 't': {
1466                                 struct timeval tv;
1467                                 do_gettimeofday(&tv);
1468                                 rc = snprintf(out_ptr, out_end - out_ptr,
1469                                               "%lu", tv.tv_sec);
1470                                 if (rc > out_end - out_ptr)
1471                                         goto out;
1472                                 out_ptr += rc;
1473                                 break;
1474                         }
1475                         /* hostname */
1476                         case 'h':
1477                                 down_read(&uts_sem);
1478                                 rc = snprintf(out_ptr, out_end - out_ptr,
1479                                               "%s", utsname()->nodename);
1480                                 up_read(&uts_sem);
1481                                 if (rc > out_end - out_ptr)
1482                                         goto out;
1483                                 out_ptr += rc;
1484                                 break;
1485                         /* executable */
1486                         case 'e':
1487                                 rc = snprintf(out_ptr, out_end - out_ptr,
1488                                               "%s", current->comm);
1489                                 if (rc > out_end - out_ptr)
1490                                         goto out;
1491                                 out_ptr += rc;
1492                                 break;
1493                         /* core limit size */
1494                         case 'c':
1495                                 rc = snprintf(out_ptr, out_end - out_ptr,
1496                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1497                                 if (rc > out_end - out_ptr)
1498                                         goto out;
1499                                 out_ptr += rc;
1500                                 break;
1501                         default:
1502                                 break;
1503                         }
1504                         ++pat_ptr;
1505                 }
1506         }
1507         /* Backward compatibility with core_uses_pid:
1508          *
1509          * If core_pattern does not include a %p (as is the default)
1510          * and core_uses_pid is set, then .%pid will be appended to
1511          * the filename. Do not do this for piped commands. */
1512         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1513                 rc = snprintf(out_ptr, out_end - out_ptr,
1514                               ".%d", task_tgid_vnr(current));
1515                 if (rc > out_end - out_ptr)
1516                         goto out;
1517                 out_ptr += rc;
1518         }
1519 out:
1520         *out_ptr = 0;
1521         return ispipe;
1522 }
1523
1524 static int zap_process(struct task_struct *start)
1525 {
1526         struct task_struct *t;
1527         int nr = 0;
1528
1529         start->signal->flags = SIGNAL_GROUP_EXIT;
1530         start->signal->group_stop_count = 0;
1531
1532         t = start;
1533         do {
1534                 if (t != current && t->mm) {
1535                         sigaddset(&t->pending.signal, SIGKILL);
1536                         signal_wake_up(t, 1);
1537                         nr++;
1538                 }
1539         } while_each_thread(start, t);
1540
1541         return nr;
1542 }
1543
1544 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1545                                 struct core_state *core_state, int exit_code)
1546 {
1547         struct task_struct *g, *p;
1548         unsigned long flags;
1549         int nr = -EAGAIN;
1550
1551         spin_lock_irq(&tsk->sighand->siglock);
1552         if (!signal_group_exit(tsk->signal)) {
1553                 mm->core_state = core_state;
1554                 tsk->signal->group_exit_code = exit_code;
1555                 nr = zap_process(tsk);
1556         }
1557         spin_unlock_irq(&tsk->sighand->siglock);
1558         if (unlikely(nr < 0))
1559                 return nr;
1560
1561         if (atomic_read(&mm->mm_users) == nr + 1)
1562                 goto done;
1563         /*
1564          * We should find and kill all tasks which use this mm, and we should
1565          * count them correctly into ->nr_threads. We don't take tasklist
1566          * lock, but this is safe wrt:
1567          *
1568          * fork:
1569          *      None of sub-threads can fork after zap_process(leader). All
1570          *      processes which were created before this point should be
1571          *      visible to zap_threads() because copy_process() adds the new
1572          *      process to the tail of init_task.tasks list, and lock/unlock
1573          *      of ->siglock provides a memory barrier.
1574          *
1575          * do_exit:
1576          *      The caller holds mm->mmap_sem. This means that the task which
1577          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1578          *      its ->mm.
1579          *
1580          * de_thread:
1581          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1582          *      we must see either old or new leader, this does not matter.
1583          *      However, it can change p->sighand, so lock_task_sighand(p)
1584          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1585          *      it can't fail.
1586          *
1587          *      Note also that "g" can be the old leader with ->mm == NULL
1588          *      and already unhashed and thus removed from ->thread_group.
1589          *      This is OK, __unhash_process()->list_del_rcu() does not
1590          *      clear the ->next pointer, we will find the new leader via
1591          *      next_thread().
1592          */
1593         rcu_read_lock();
1594         for_each_process(g) {
1595                 if (g == tsk->group_leader)
1596                         continue;
1597                 if (g->flags & PF_KTHREAD)
1598                         continue;
1599                 p = g;
1600                 do {
1601                         if (p->mm) {
1602                                 if (unlikely(p->mm == mm)) {
1603                                         lock_task_sighand(p, &flags);
1604                                         nr += zap_process(p);
1605                                         unlock_task_sighand(p, &flags);
1606                                 }
1607                                 break;
1608                         }
1609                 } while_each_thread(g, p);
1610         }
1611         rcu_read_unlock();
1612 done:
1613         atomic_set(&core_state->nr_threads, nr);
1614         return nr;
1615 }
1616
1617 static int coredump_wait(int exit_code, struct core_state *core_state)
1618 {
1619         struct task_struct *tsk = current;
1620         struct mm_struct *mm = tsk->mm;
1621         struct completion *vfork_done;
1622         int core_waiters;
1623
1624         init_completion(&core_state->startup);
1625         core_state->dumper.task = tsk;
1626         core_state->dumper.next = NULL;
1627         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1628         up_write(&mm->mmap_sem);
1629
1630         if (unlikely(core_waiters < 0))
1631                 goto fail;
1632
1633         /*
1634          * Make sure nobody is waiting for us to release the VM,
1635          * otherwise we can deadlock when we wait on each other
1636          */
1637         vfork_done = tsk->vfork_done;
1638         if (vfork_done) {
1639                 tsk->vfork_done = NULL;
1640                 complete(vfork_done);
1641         }
1642
1643         if (core_waiters)
1644                 wait_for_completion(&core_state->startup);
1645 fail:
1646         return core_waiters;
1647 }
1648
1649 static void coredump_finish(struct mm_struct *mm)
1650 {
1651         struct core_thread *curr, *next;
1652         struct task_struct *task;
1653
1654         next = mm->core_state->dumper.next;
1655         while ((curr = next) != NULL) {
1656                 next = curr->next;
1657                 task = curr->task;
1658                 /*
1659                  * see exit_mm(), curr->task must not see
1660                  * ->task == NULL before we read ->next.
1661                  */
1662                 smp_mb();
1663                 curr->task = NULL;
1664                 wake_up_process(task);
1665         }
1666
1667         mm->core_state = NULL;
1668 }
1669
1670 /*
1671  * set_dumpable converts traditional three-value dumpable to two flags and
1672  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1673  * these bits are not changed atomically.  So get_dumpable can observe the
1674  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1675  * return either old dumpable or new one by paying attention to the order of
1676  * modifying the bits.
1677  *
1678  * dumpable |   mm->flags (binary)
1679  * old  new | initial interim  final
1680  * ---------+-----------------------
1681  *  0    1  |   00      01      01
1682  *  0    2  |   00      10(*)   11
1683  *  1    0  |   01      00      00
1684  *  1    2  |   01      11      11
1685  *  2    0  |   11      10(*)   00
1686  *  2    1  |   11      11      01
1687  *
1688  * (*) get_dumpable regards interim value of 10 as 11.
1689  */
1690 void set_dumpable(struct mm_struct *mm, int value)
1691 {
1692         switch (value) {
1693         case 0:
1694                 clear_bit(MMF_DUMPABLE, &mm->flags);
1695                 smp_wmb();
1696                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1697                 break;
1698         case 1:
1699                 set_bit(MMF_DUMPABLE, &mm->flags);
1700                 smp_wmb();
1701                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1702                 break;
1703         case 2:
1704                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1705                 smp_wmb();
1706                 set_bit(MMF_DUMPABLE, &mm->flags);
1707                 break;
1708         }
1709 }
1710
1711 int get_dumpable(struct mm_struct *mm)
1712 {
1713         int ret;
1714
1715         ret = mm->flags & 0x3;
1716         return (ret >= 2) ? 2 : ret;
1717 }
1718
1719 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1720 {
1721         struct core_state core_state;
1722         char corename[CORENAME_MAX_SIZE + 1];
1723         struct mm_struct *mm = current->mm;
1724         struct linux_binfmt * binfmt;
1725         struct inode * inode;
1726         struct file * file;
1727         int retval = 0;
1728         int fsuid = current->fsuid;
1729         int flag = 0;
1730         int ispipe = 0;
1731         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1732         char **helper_argv = NULL;
1733         int helper_argc = 0;
1734         char *delimit;
1735
1736         audit_core_dumps(signr);
1737
1738         binfmt = current->binfmt;
1739         if (!binfmt || !binfmt->core_dump)
1740                 goto fail;
1741         down_write(&mm->mmap_sem);
1742         /*
1743          * If another thread got here first, or we are not dumpable, bail out.
1744          */
1745         if (mm->core_state || !get_dumpable(mm)) {
1746                 up_write(&mm->mmap_sem);
1747                 goto fail;
1748         }
1749
1750         /*
1751          *      We cannot trust fsuid as being the "true" uid of the
1752          *      process nor do we know its entire history. We only know it
1753          *      was tainted so we dump it as root in mode 2.
1754          */
1755         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1756                 flag = O_EXCL;          /* Stop rewrite attacks */
1757                 current->fsuid = 0;     /* Dump root private */
1758         }
1759
1760         retval = coredump_wait(exit_code, &core_state);
1761         if (retval < 0)
1762                 goto fail;
1763
1764         /*
1765          * Clear any false indication of pending signals that might
1766          * be seen by the filesystem code called to write the core file.
1767          */
1768         clear_thread_flag(TIF_SIGPENDING);
1769
1770         /*
1771          * lock_kernel() because format_corename() is controlled by sysctl, which
1772          * uses lock_kernel()
1773          */
1774         lock_kernel();
1775         ispipe = format_corename(corename, signr);
1776         unlock_kernel();
1777         /*
1778          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1779          * to a pipe.  Since we're not writing directly to the filesystem
1780          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1781          * created unless the pipe reader choses to write out the core file
1782          * at which point file size limits and permissions will be imposed
1783          * as it does with any other process
1784          */
1785         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1786                 goto fail_unlock;
1787
1788         if (ispipe) {
1789                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1790                 /* Terminate the string before the first option */
1791                 delimit = strchr(corename, ' ');
1792                 if (delimit)
1793                         *delimit = '\0';
1794                 delimit = strrchr(helper_argv[0], '/');
1795                 if (delimit)
1796                         delimit++;
1797                 else
1798                         delimit = helper_argv[0];
1799                 if (!strcmp(delimit, current->comm)) {
1800                         printk(KERN_NOTICE "Recursive core dump detected, "
1801                                         "aborting\n");
1802                         goto fail_unlock;
1803                 }
1804
1805                 core_limit = RLIM_INFINITY;
1806
1807                 /* SIGPIPE can happen, but it's just never processed */
1808                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1809                                 &file)) {
1810                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1811                                corename);
1812                         goto fail_unlock;
1813                 }
1814         } else
1815                 file = filp_open(corename,
1816                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1817                                  0600);
1818         if (IS_ERR(file))
1819                 goto fail_unlock;
1820         inode = file->f_path.dentry->d_inode;
1821         if (inode->i_nlink > 1)
1822                 goto close_fail;        /* multiple links - don't dump */
1823         if (!ispipe && d_unhashed(file->f_path.dentry))
1824                 goto close_fail;
1825
1826         /* AK: actually i see no reason to not allow this for named pipes etc.,
1827            but keep the previous behaviour for now. */
1828         if (!ispipe && !S_ISREG(inode->i_mode))
1829                 goto close_fail;
1830         /*
1831          * Dont allow local users get cute and trick others to coredump
1832          * into their pre-created files:
1833          */
1834         if (inode->i_uid != current->fsuid)
1835                 goto close_fail;
1836         if (!file->f_op)
1837                 goto close_fail;
1838         if (!file->f_op->write)
1839                 goto close_fail;
1840         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1841                 goto close_fail;
1842
1843         retval = binfmt->core_dump(signr, regs, file, core_limit);
1844
1845         if (retval)
1846                 current->signal->group_exit_code |= 0x80;
1847 close_fail:
1848         filp_close(file, NULL);
1849 fail_unlock:
1850         if (helper_argv)
1851                 argv_free(helper_argv);
1852
1853         current->fsuid = fsuid;
1854         coredump_finish(mm);
1855 fail:
1856         return retval;
1857 }