fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/mman.h>
  28 #include <linux/a.out.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/init.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/highmem.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/key.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/pid_namespace.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53
  54 #include <asm/uaccess.h>
  55 #include <asm/mmu_context.h>
  56 #include <asm/tlb.h>
  57
  58 #ifdef CONFIG_KMOD
  59 #include <linux/kmod.h>
  60 #endif
  61
  62 int core_uses_pid;
  63 char core_pattern[CORENAME_MAX_SIZE] = "core";
  64 int suid_dumpable = 0;
  65
  66 EXPORT_SYMBOL(suid_dumpable);
  67 /* The maximal length of core_pattern is also specified in sysctl.c */
  68
  69 static struct linux_binfmt *formats;
  70 static DEFINE_RWLOCK(binfmt_lock);
  71
  72 int register_binfmt(struct linux_binfmt * fmt)
  73 {
  74         struct linux_binfmt ** tmp = &formats;
  75
  76         if (!fmt)
  77                 return -EINVAL;
  78         if (fmt->next)
  79                 return -EBUSY;
  80         write_lock(&binfmt_lock);
  81         while (*tmp) {
  82                 if (fmt == *tmp) {
  83                         write_unlock(&binfmt_lock);
  84                         return -EBUSY;
  85                 }
  86                 tmp = &(*tmp)->next;
  87         }
  88         fmt->next = formats;
  89         formats = fmt;
  90         write_unlock(&binfmt_lock);
  91         return 0;
  92 }
  93
  94 EXPORT_SYMBOL(register_binfmt);
  95
  96 int unregister_binfmt(struct linux_binfmt * fmt)
  97 {
  98         struct linux_binfmt ** tmp = &formats;
  99
 100         write_lock(&binfmt_lock);
 101         while (*tmp) {
 102                 if (fmt == *tmp) {
 103                         *tmp = fmt->next;
 104                         fmt->next = NULL;
 105                         write_unlock(&binfmt_lock);
 106                         return 0;
 107                 }
 108                 tmp = &(*tmp)->next;
 109         }
 110         write_unlock(&binfmt_lock);
 111         return -EINVAL;
 112 }
 113
 114 EXPORT_SYMBOL(unregister_binfmt);
 115
 116 static inline void put_binfmt(struct linux_binfmt * fmt)
 117 {
 118         module_put(fmt->module);
 119 }
 120
 121 /*
 122  * Note that a shared library must be both readable and executable due to
 123  * security reasons.
 124  *
 125  * Also note that we take the address to load from from the file itself.
 126  */
 127 asmlinkage long sys_uselib(const char __user * library)
 128 {
 129         struct file * file;
 130         struct nameidata nd;
 131         int error;
 132
 133         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 134         if (error)
 135                 goto out;
 136
 137         error = -EACCES;
 138         if (nd.mnt->mnt_flags & MNT_NOEXEC)
 139                 goto exit;
 140         error = -EINVAL;
 141         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 142                 goto exit;
 143
 144         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 145         if (error)
 146                 goto exit;
 147
 148         file = nameidata_to_filp(&nd, O_RDONLY);
 149         error = PTR_ERR(file);
 150         if (IS_ERR(file))
 151                 goto out;
 152
 153         error = -ENOEXEC;
 154         if(file->f_op) {
 155                 struct linux_binfmt * fmt;
 156
 157                 read_lock(&binfmt_lock);
 158                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 159                         if (!fmt->load_shlib)
 160                                 continue;
 161                         if (!try_module_get(fmt->module))
 162                                 continue;
 163                         read_unlock(&binfmt_lock);
 164                         error = fmt->load_shlib(file);
 165                         read_lock(&binfmt_lock);
 166                         put_binfmt(fmt);
 167                         if (error != -ENOEXEC)
 168                                 break;
 169                 }
 170                 read_unlock(&binfmt_lock);
 171         }
 172         fput(file);
 173 out:
 174         return error;
 175 exit:
 176         release_open_intent(&nd);
 177         path_release(&nd);
 178         goto out;
 179 }
 180
 181 #ifdef CONFIG_MMU
 182
 183 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 184                 int write)
 185 {
 186         struct page *page;
 187         int ret;
 188
 189 #ifdef CONFIG_STACK_GROWSUP
 190         if (write) {
 191                 ret = expand_stack_downwards(bprm->vma, pos);
 192                 if (ret < 0)
 193                         return NULL;
 194         }
 195 #endif
 196         ret = get_user_pages(current, bprm->mm, pos,
 197                         1, write, 1, &page, NULL);
 198         if (ret <= 0)
 199                 return NULL;
 200
 201         if (write) {
 202                 struct rlimit *rlim = current->signal->rlim;
 203                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 204
 205                 /*
 206                  * Limit to 1/4-th the stack size for the argv+env strings.
 207                  * This ensures that:
 208                  *  - the remaining binfmt code will not run out of stack space,
 209                  *  - the program will have a reasonable amount of stack left
 210                  *    to work from.
 211                  */
 212                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 213                         put_page(page);
 214                         return NULL;
 215                 }
 216         }
 217
 218         return page;
 219 }
 220
 221 static void put_arg_page(struct page *page)
 222 {
 223         put_page(page);
 224 }
 225
 226 static void free_arg_page(struct linux_binprm *bprm, int i)
 227 {
 228 }
 229
 230 static void free_arg_pages(struct linux_binprm *bprm)
 231 {
 232 }
 233
 234 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 235                 struct page *page)
 236 {
 237         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 238 }
 239
 240 static int __bprm_mm_init(struct linux_binprm *bprm)
 241 {
 242         int err = -ENOMEM;
 243         struct vm_area_struct *vma = NULL;
 244         struct mm_struct *mm = bprm->mm;
 245
 246         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 247         if (!vma)
 248                 goto err;
 249
 250         down_write(&mm->mmap_sem);
 251         vma->vm_mm = mm;
 252
 253         /*
 254          * Place the stack at the largest stack address the architecture
 255          * supports. Later, we'll move this to an appropriate place. We don't
 256          * use STACK_TOP because that can depend on attributes which aren't
 257          * configured yet.
 258          */
 259         vma->vm_end = STACK_TOP_MAX;
 260         vma->vm_start = vma->vm_end - PAGE_SIZE;
 261
 262         vma->vm_flags = VM_STACK_FLAGS;
 263         vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
 264         err = insert_vm_struct(mm, vma);
 265         if (err) {
 266                 up_write(&mm->mmap_sem);
 267                 goto err;
 268         }
 269
 270         mm->stack_vm = mm->total_vm = 1;
 271         up_write(&mm->mmap_sem);
 272
 273         bprm->p = vma->vm_end - sizeof(void *);
 274
 275         return 0;
 276
 277 err:
 278         if (vma) {
 279                 bprm->vma = NULL;
 280                 kmem_cache_free(vm_area_cachep, vma);
 281         }
 282
 283         return err;
 284 }
 285
 286 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 287 {
 288         return len <= MAX_ARG_STRLEN;
 289 }
 290
 291 #else
 292
 293 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 294                 int write)
 295 {
 296         struct page *page;
 297
 298         page = bprm->page[pos / PAGE_SIZE];
 299         if (!page && write) {
 300                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 301                 if (!page)
 302                         return NULL;
 303                 bprm->page[pos / PAGE_SIZE] = page;
 304         }
 305
 306         return page;
 307 }
 308
 309 static void put_arg_page(struct page *page)
 310 {
 311 }
 312
 313 static void free_arg_page(struct linux_binprm *bprm, int i)
 314 {
 315         if (bprm->page[i]) {
 316                 __free_page(bprm->page[i]);
 317                 bprm->page[i] = NULL;
 318         }
 319 }
 320
 321 static void free_arg_pages(struct linux_binprm *bprm)
 322 {
 323         int i;
 324
 325         for (i = 0; i < MAX_ARG_PAGES; i++)
 326                 free_arg_page(bprm, i);
 327 }
 328
 329 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 330                 struct page *page)
 331 {
 332 }
 333
 334 static int __bprm_mm_init(struct linux_binprm *bprm)
 335 {
 336         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 337         return 0;
 338 }
 339
 340 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 341 {
 342         return len <= bprm->p;
 343 }
 344
 345 #endif /* CONFIG_MMU */
 346
 347 /*
 348  * Create a new mm_struct and populate it with a temporary stack
 349  * vm_area_struct.  We don't have enough context at this point to set the stack
 350  * flags, permissions, and offset, so we use temporary values.  We'll update
 351  * them later in setup_arg_pages().
 352  */
 353 int bprm_mm_init(struct linux_binprm *bprm)
 354 {
 355         int err;
 356         struct mm_struct *mm = NULL;
 357
 358         bprm->mm = mm = mm_alloc();
 359         err = -ENOMEM;
 360         if (!mm)
 361                 goto err;
 362
 363         err = init_new_context(current, mm);
 364         if (err)
 365                 goto err;
 366
 367         err = __bprm_mm_init(bprm);
 368         if (err)
 369                 goto err;
 370
 371         return 0;
 372
 373 err:
 374         if (mm) {
 375                 bprm->mm = NULL;
 376                 mmdrop(mm);
 377         }
 378
 379         return err;
 380 }
 381
 382 /*
 383  * count() counts the number of strings in array ARGV.
 384  */
 385 static int count(char __user * __user * argv, int max)
 386 {
 387         int i = 0;
 388
 389         if (argv != NULL) {
 390                 for (;;) {
 391                         char __user * p;
 392
 393                         if (get_user(p, argv))
 394                                 return -EFAULT;
 395                         if (!p)
 396                                 break;
 397                         argv++;
 398                         if(++i > max)
 399                                 return -E2BIG;
 400                         cond_resched();
 401                 }
 402         }
 403         return i;
 404 }
 405
 406 /*
 407  * 'copy_strings()' copies argument/environment strings from the old
 408  * processes's memory to the new process's stack.  The call to get_user_pages()
 409  * ensures the destination page is created and not swapped out.
 410  */
 411 static int copy_strings(int argc, char __user * __user * argv,
 412                         struct linux_binprm *bprm)
 413 {
 414         struct page *kmapped_page = NULL;
 415         char *kaddr = NULL;
 416         unsigned long kpos = 0;
 417         int ret;
 418
 419         while (argc-- > 0) {
 420                 char __user *str;
 421                 int len;
 422                 unsigned long pos;
 423
 424                 if (get_user(str, argv+argc) ||
 425                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 426                         ret = -EFAULT;
 427                         goto out;
 428                 }
 429
 430                 if (!valid_arg_len(bprm, len)) {
 431                         ret = -E2BIG;
 432                         goto out;
 433                 }
 434
 435                 /* We're going to work our way backwords. */
 436                 pos = bprm->p;
 437                 str += len;
 438                 bprm->p -= len;
 439
 440                 while (len > 0) {
 441                         int offset, bytes_to_copy;
 442
 443                         offset = pos % PAGE_SIZE;
 444                         if (offset == 0)
 445                                 offset = PAGE_SIZE;
 446
 447                         bytes_to_copy = offset;
 448                         if (bytes_to_copy > len)
 449                                 bytes_to_copy = len;
 450
 451                         offset -= bytes_to_copy;
 452                         pos -= bytes_to_copy;
 453                         str -= bytes_to_copy;
 454                         len -= bytes_to_copy;
 455
 456                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 457                                 struct page *page;
 458
 459                                 page = get_arg_page(bprm, pos, 1);
 460                                 if (!page) {
 461                                         ret = -E2BIG;
 462                                         goto out;
 463                                 }
 464
 465                                 if (kmapped_page) {
 466                                         flush_kernel_dcache_page(kmapped_page);
 467                                         kunmap(kmapped_page);
 468                                         put_arg_page(kmapped_page);
 469                                 }
 470                                 kmapped_page = page;
 471                                 kaddr = kmap(kmapped_page);
 472                                 kpos = pos & PAGE_MASK;
 473                                 flush_arg_page(bprm, kpos, kmapped_page);
 474                         }
 475                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 476                                 ret = -EFAULT;
 477                                 goto out;
 478                         }
 479                 }
 480         }
 481         ret = 0;
 482 out:
 483         if (kmapped_page) {
 484                 flush_kernel_dcache_page(kmapped_page);
 485                 kunmap(kmapped_page);
 486                 put_arg_page(kmapped_page);
 487         }
 488         return ret;
 489 }
 490
 491 /*
 492  * Like copy_strings, but get argv and its values from kernel memory.
 493  */
 494 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 495 {
 496         int r;
 497         mm_segment_t oldfs = get_fs();
 498         set_fs(KERNEL_DS);
 499         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 500         set_fs(oldfs);
 501         return r;
 502 }
 503 EXPORT_SYMBOL(copy_strings_kernel);
 504
 505 #ifdef CONFIG_MMU
 506
 507 /*
 508  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 509  * the binfmt code determines where the new stack should reside, we shift it to
 510  * its final location.  The process proceeds as follows:
 511  *
 512  * 1) Use shift to calculate the new vma endpoints.
 513  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 514  *    arguments passed to subsequent functions are consistent.
 515  * 3) Move vma's page tables to the new range.
 516  * 4) Free up any cleared pgd range.
 517  * 5) Shrink the vma to cover only the new range.
 518  */
 519 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 520 {
 521         struct mm_struct *mm = vma->vm_mm;
 522         unsigned long old_start = vma->vm_start;
 523         unsigned long old_end = vma->vm_end;
 524         unsigned long length = old_end - old_start;
 525         unsigned long new_start = old_start - shift;
 526         unsigned long new_end = old_end - shift;
 527         struct mmu_gather *tlb;
 528
 529         BUG_ON(new_start > new_end);
 530
 531         /*
 532          * ensure there are no vmas between where we want to go
 533          * and where we are
 534          */
 535         if (vma != find_vma(mm, new_start))
 536                 return -EFAULT;
 537
 538         /*
 539          * cover the whole range: [new_start, old_end)
 540          */
 541         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 542
 543         /*
 544          * move the page tables downwards, on failure we rely on
 545          * process cleanup to remove whatever mess we made.
 546          */
 547         if (length != move_page_tables(vma, old_start,
 548                                        vma, new_start, length))
 549                 return -ENOMEM;
 550
 551         lru_add_drain();
 552         tlb = tlb_gather_mmu(mm, 0);
 553         if (new_end > old_start) {
 554                 /*
 555                  * when the old and new regions overlap clear from new_end.
 556                  */
 557                 free_pgd_range(&tlb, new_end, old_end, new_end,
 558                         vma->vm_next ? vma->vm_next->vm_start : 0);
 559         } else {
 560                 /*
 561                  * otherwise, clean from old_start; this is done to not touch
 562                  * the address space in [new_end, old_start) some architectures
 563                  * have constraints on va-space that make this illegal (IA64) -
 564                  * for the others its just a little faster.
 565                  */
 566                 free_pgd_range(&tlb, old_start, old_end, new_end,
 567                         vma->vm_next ? vma->vm_next->vm_start : 0);
 568         }
 569         tlb_finish_mmu(tlb, new_end, old_end);
 570
 571         /*
 572          * shrink the vma to just the new range.
 573          */
 574         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 575
 576         return 0;
 577 }
 578
 579 #define EXTRA_STACK_VM_PAGES    20      /* random */
 580
 581 /*
 582  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 583  * the stack is optionally relocated, and some extra space is added.
 584  */
 585 int setup_arg_pages(struct linux_binprm *bprm,
 586                     unsigned long stack_top,
 587                     int executable_stack)
 588 {
 589         unsigned long ret;
 590         unsigned long stack_shift;
 591         struct mm_struct *mm = current->mm;
 592         struct vm_area_struct *vma = bprm->vma;
 593         struct vm_area_struct *prev = NULL;
 594         unsigned long vm_flags;
 595         unsigned long stack_base;
 596
 597 #ifdef CONFIG_STACK_GROWSUP
 598         /* Limit stack size to 1GB */
 599         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 600         if (stack_base > (1 << 30))
 601                 stack_base = 1 << 30;
 602
 603         /* Make sure we didn't let the argument array grow too large. */
 604         if (vma->vm_end - vma->vm_start > stack_base)
 605                 return -ENOMEM;
 606
 607         stack_base = PAGE_ALIGN(stack_top - stack_base);
 608
 609         stack_shift = vma->vm_start - stack_base;
 610         mm->arg_start = bprm->p - stack_shift;
 611         bprm->p = vma->vm_end - stack_shift;
 612 #else
 613         stack_top = arch_align_stack(stack_top);
 614         stack_top = PAGE_ALIGN(stack_top);
 615         stack_shift = vma->vm_end - stack_top;
 616
 617         bprm->p -= stack_shift;
 618         mm->arg_start = bprm->p;
 619 #endif
 620
 621         if (bprm->loader)
 622                 bprm->loader -= stack_shift;
 623         bprm->exec -= stack_shift;
 624
 625         down_write(&mm->mmap_sem);
 626         vm_flags = vma->vm_flags;
 627
 628         /*
 629          * Adjust stack execute permissions; explicitly enable for
 630          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 631          * (arch default) otherwise.
 632          */
 633         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 634                 vm_flags |= VM_EXEC;
 635         else if (executable_stack == EXSTACK_DISABLE_X)
 636                 vm_flags &= ~VM_EXEC;
 637         vm_flags |= mm->def_flags;
 638
 639         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 640                         vm_flags);
 641         if (ret)
 642                 goto out_unlock;
 643         BUG_ON(prev != vma);
 644
 645         /* Move stack pages down in memory. */
 646         if (stack_shift) {
 647                 ret = shift_arg_pages(vma, stack_shift);
 648                 if (ret) {
 649                         up_write(&mm->mmap_sem);
 650                         return ret;
 651                 }
 652         }
 653
 654 #ifdef CONFIG_STACK_GROWSUP
 655         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 656 #else
 657         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 658 #endif
 659         ret = expand_stack(vma, stack_base);
 660         if (ret)
 661                 ret = -EFAULT;
 662
 663 out_unlock:
 664         up_write(&mm->mmap_sem);
 665         return 0;
 666 }
 667 EXPORT_SYMBOL(setup_arg_pages);
 668
 669 #endif /* CONFIG_MMU */
 670
 671 struct file *open_exec(const char *name)
 672 {
 673         struct nameidata nd;
 674         int err;
 675         struct file *file;
 676
 677         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 678         file = ERR_PTR(err);
 679
 680         if (!err) {
 681                 struct inode *inode = nd.dentry->d_inode;
 682                 file = ERR_PTR(-EACCES);
 683                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 684                     S_ISREG(inode->i_mode)) {
 685                         int err = vfs_permission(&nd, MAY_EXEC);
 686                         file = ERR_PTR(err);
 687                         if (!err) {
 688                                 file = nameidata_to_filp(&nd, O_RDONLY);
 689                                 if (!IS_ERR(file)) {
 690                                         err = deny_write_access(file);
 691                                         if (err) {
 692                                                 fput(file);
 693                                                 file = ERR_PTR(err);
 694                                         }
 695                                 }
 696 out:
 697                                 return file;
 698                         }
 699                 }
 700                 release_open_intent(&nd);
 701                 path_release(&nd);
 702         }
 703         goto out;
 704 }
 705
 706 EXPORT_SYMBOL(open_exec);
 707
 708 int kernel_read(struct file *file, unsigned long offset,
 709         char *addr, unsigned long count)
 710 {
 711         mm_segment_t old_fs;
 712         loff_t pos = offset;
 713         int result;
 714
 715         old_fs = get_fs();
 716         set_fs(get_ds());
 717         /* The cast to a user pointer is valid due to the set_fs() */
 718         result = vfs_read(file, (void __user *)addr, count, &pos);
 719         set_fs(old_fs);
 720         return result;
 721 }
 722
 723 EXPORT_SYMBOL(kernel_read);
 724
 725 static int exec_mmap(struct mm_struct *mm)
 726 {
 727         struct task_struct *tsk;
 728         struct mm_struct * old_mm, *active_mm;
 729
 730         /* Notify parent that we're no longer interested in the old VM */
 731         tsk = current;
 732         old_mm = current->mm;
 733         mm_release(tsk, old_mm);
 734
 735         if (old_mm) {
 736                 /*
 737                  * Make sure that if there is a core dump in progress
 738                  * for the old mm, we get out and die instead of going
 739                  * through with the exec.  We must hold mmap_sem around
 740                  * checking core_waiters and changing tsk->mm.  The
 741                  * core-inducing thread will increment core_waiters for
 742                  * each thread whose ->mm == old_mm.
 743                  */
 744                 down_read(&old_mm->mmap_sem);
 745                 if (unlikely(old_mm->core_waiters)) {
 746                         up_read(&old_mm->mmap_sem);
 747                         return -EINTR;
 748                 }
 749         }
 750         task_lock(tsk);
 751         active_mm = tsk->active_mm;
 752         tsk->mm = mm;
 753         tsk->active_mm = mm;
 754         activate_mm(active_mm, mm);
 755         task_unlock(tsk);
 756         arch_pick_mmap_layout(mm);
 757         if (old_mm) {
 758                 up_read(&old_mm->mmap_sem);
 759                 BUG_ON(active_mm != old_mm);
 760                 mmput(old_mm);
 761                 return 0;
 762         }
 763         mmdrop(active_mm);
 764         return 0;
 765 }
 766
 767 /*
 768  * This function makes sure the current process has its own signal table,
 769  * so that flush_signal_handlers can later reset the handlers without
 770  * disturbing other processes.  (Other processes might share the signal
 771  * table via the CLONE_SIGHAND option to clone().)
 772  */
 773 static int de_thread(struct task_struct *tsk)
 774 {
 775         struct signal_struct *sig = tsk->signal;
 776         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 777         spinlock_t *lock = &oldsighand->siglock;
 778         struct task_struct *leader = NULL;
 779         int count;
 780
 781         /*
 782          * If we don't share sighandlers, then we aren't sharing anything
 783          * and we can just re-use it all.
 784          */
 785         if (atomic_read(&oldsighand->count) <= 1) {
 786                 exit_itimers(sig);
 787                 return 0;
 788         }
 789
 790         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 791         if (!newsighand)
 792                 return -ENOMEM;
 793
 794         if (thread_group_empty(tsk))
 795                 goto no_thread_group;
 796
 797         /*
 798          * Kill all other threads in the thread group.
 799          * We must hold tasklist_lock to call zap_other_threads.
 800          */
 801         read_lock(&tasklist_lock);
 802         spin_lock_irq(lock);
 803         if (sig->flags & SIGNAL_GROUP_EXIT) {
 804                 /*
 805                  * Another group action in progress, just
 806                  * return so that the signal is processed.
 807                  */
 808                 spin_unlock_irq(lock);
 809                 read_unlock(&tasklist_lock);
 810                 kmem_cache_free(sighand_cachep, newsighand);
 811                 return -EAGAIN;
 812         }
 813
 814         /*
 815          * child_reaper ignores SIGKILL, change it now.
 816          * Reparenting needs write_lock on tasklist_lock,
 817          * so it is safe to do it under read_lock.
 818          */
 819         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 820                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 821
 822         zap_other_threads(tsk);
 823         read_unlock(&tasklist_lock);
 824
 825         /*
 826          * Account for the thread group leader hanging around:
 827          */
 828         count = 1;
 829         if (!thread_group_leader(tsk)) {
 830                 count = 2;
 831                 /*
 832                  * The SIGALRM timer survives the exec, but needs to point
 833                  * at us as the new group leader now.  We have a race with
 834                  * a timer firing now getting the old leader, so we need to
 835                  * synchronize with any firing (by calling del_timer_sync)
 836                  * before we can safely let the old group leader die.
 837                  */
 838                 sig->tsk = tsk;
 839                 spin_unlock_irq(lock);
 840                 if (hrtimer_cancel(&sig->real_timer))
 841                         hrtimer_restart(&sig->real_timer);
 842                 spin_lock_irq(lock);
 843         }
 844         while (atomic_read(&sig->count) > count) {
 845                 sig->group_exit_task = tsk;
 846                 sig->notify_count = count;
 847                 __set_current_state(TASK_UNINTERRUPTIBLE);
 848                 spin_unlock_irq(lock);
 849                 schedule();
 850                 spin_lock_irq(lock);
 851         }
 852         sig->group_exit_task = NULL;
 853         sig->notify_count = 0;
 854         spin_unlock_irq(lock);
 855
 856         /*
 857          * At this point all other threads have exited, all we have to
 858          * do is to wait for the thread group leader to become inactive,
 859          * and to assume its PID:
 860          */
 861         if (!thread_group_leader(tsk)) {
 862                 /*
 863                  * Wait for the thread group leader to be a zombie.
 864                  * It should already be zombie at this point, most
 865                  * of the time.
 866                  */
 867                 leader = tsk->group_leader;
 868                 while (leader->exit_state != EXIT_ZOMBIE)
 869                         yield();
 870
 871                 /*
 872                  * The only record we have of the real-time age of a
 873                  * process, regardless of execs it's done, is start_time.
 874                  * All the past CPU time is accumulated in signal_struct
 875                  * from sister threads now dead.  But in this non-leader
 876                  * exec, nothing survives from the original leader thread,
 877                  * whose birth marks the true age of this process now.
 878                  * When we take on its identity by switching to its PID, we
 879                  * also take its birthdate (always earlier than our own).
 880                  */
 881                 tsk->start_time = leader->start_time;
 882
 883                 write_lock_irq(&tasklist_lock);
 884
 885                 BUG_ON(leader->tgid != tsk->tgid);
 886                 BUG_ON(tsk->pid == tsk->tgid);
 887                 /*
 888                  * An exec() starts a new thread group with the
 889                  * TGID of the previous thread group. Rehash the
 890                  * two threads with a switched PID, and release
 891                  * the former thread group leader:
 892                  */
 893
 894                 /* Become a process group leader with the old leader's pid.
 895                  * The old leader becomes a thread of the this thread group.
 896                  * Note: The old leader also uses this pid until release_task
 897                  *       is called.  Odd but simple and correct.
 898                  */
 899                 detach_pid(tsk, PIDTYPE_PID);
 900                 tsk->pid = leader->pid;
 901                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
 902                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 903                 transfer_pid(leader, tsk, PIDTYPE_SID);
 904                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 905
 906                 tsk->group_leader = tsk;
 907                 leader->group_leader = tsk;
 908
 909                 tsk->exit_signal = SIGCHLD;
 910
 911                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 912                 leader->exit_state = EXIT_DEAD;
 913
 914                 write_unlock_irq(&tasklist_lock);
 915         }
 916
 917         /*
 918          * There may be one thread left which is just exiting,
 919          * but it's safe to stop telling the group to kill themselves.
 920          */
 921         sig->flags = 0;
 922
 923 no_thread_group:
 924         exit_itimers(sig);
 925         if (leader)
 926                 release_task(leader);
 927
 928         if (atomic_read(&oldsighand->count) == 1) {
 929                 /*
 930                  * Now that we nuked the rest of the thread group,
 931                  * it turns out we are not sharing sighand any more either.
 932                  * So we can just keep it.
 933                  */
 934                 kmem_cache_free(sighand_cachep, newsighand);
 935         } else {
 936                 /*
 937                  * Move our state over to newsighand and switch it in.
 938                  */
 939                 atomic_set(&newsighand->count, 1);
 940                 memcpy(newsighand->action, oldsighand->action,
 941                        sizeof(newsighand->action));
 942
 943                 write_lock_irq(&tasklist_lock);
 944                 spin_lock(&oldsighand->siglock);
 945                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 946
 947                 rcu_assign_pointer(tsk->sighand, newsighand);
 948                 recalc_sigpending();
 949
 950                 spin_unlock(&newsighand->siglock);
 951                 spin_unlock(&oldsighand->siglock);
 952                 write_unlock_irq(&tasklist_lock);
 953
 954                 __cleanup_sighand(oldsighand);
 955         }
 956
 957         BUG_ON(!thread_group_leader(tsk));
 958         return 0;
 959 }
 960
 961 /*
 962  * These functions flushes out all traces of the currently running executable
 963  * so that a new one can be started
 964  */
 965
 966 static void flush_old_files(struct files_struct * files)
 967 {
 968         long j = -1;
 969         struct fdtable *fdt;
 970
 971         spin_lock(&files->file_lock);
 972         for (;;) {
 973                 unsigned long set, i;
 974
 975                 j++;
 976                 i = j * __NFDBITS;
 977                 fdt = files_fdtable(files);
 978                 if (i >= fdt->max_fds)
 979                         break;
 980                 set = fdt->close_on_exec->fds_bits[j];
 981                 if (!set)
 982                         continue;
 983                 fdt->close_on_exec->fds_bits[j] = 0;
 984                 spin_unlock(&files->file_lock);
 985                 for ( ; set ; i++,set >>= 1) {
 986                         if (set & 1) {
 987                                 sys_close(i);
 988                         }
 989                 }
 990                 spin_lock(&files->file_lock);
 991
 992         }
 993         spin_unlock(&files->file_lock);
 994 }
 995
 996 void get_task_comm(char *buf, struct task_struct *tsk)
 997 {
 998         /* buf must be at least sizeof(tsk->comm) in size */
 999         task_lock(tsk);
1000         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1001         task_unlock(tsk);
1002 }
1003
1004 void set_task_comm(struct task_struct *tsk, char *buf)
1005 {
1006         task_lock(tsk);
1007         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1008         task_unlock(tsk);
1009 }
1010
1011 int flush_old_exec(struct linux_binprm * bprm)
1012 {
1013         char * name;
1014         int i, ch, retval;
1015         struct files_struct *files;
1016         char tcomm[sizeof(current->comm)];
1017
1018         /*
1019          * Make sure we have a private signal table and that
1020          * we are unassociated from the previous thread group.
1021          */
1022         retval = de_thread(current);
1023         if (retval)
1024                 goto out;
1025
1026         /*
1027          * Make sure we have private file handles. Ask the
1028          * fork helper to do the work for us and the exit
1029          * helper to do the cleanup of the old one.
1030          */
1031         files = current->files;         /* refcounted so safe to hold */
1032         retval = unshare_files();
1033         if (retval)
1034                 goto out;
1035         /*
1036          * Release all of the old mmap stuff
1037          */
1038         retval = exec_mmap(bprm->mm);
1039         if (retval)
1040                 goto mmap_failed;
1041
1042         bprm->mm = NULL;                /* We're using it now */
1043
1044         /* This is the point of no return */
1045         put_files_struct(files);
1046
1047         current->sas_ss_sp = current->sas_ss_size = 0;
1048
1049         if (current->euid == current->uid && current->egid == current->gid)
1050                 set_dumpable(current->mm, 1);
1051         else
1052                 set_dumpable(current->mm, suid_dumpable);
1053
1054         name = bprm->filename;
1055
1056         /* Copies the binary name from after last slash */
1057         for (i=0; (ch = *(name++)) != '\0';) {
1058                 if (ch == '/')
1059                         i = 0; /* overwrite what we wrote */
1060                 else
1061                         if (i < (sizeof(tcomm) - 1))
1062                                 tcomm[i++] = ch;
1063         }
1064         tcomm[i] = '\0';
1065         set_task_comm(current, tcomm);
1066
1067         current->flags &= ~PF_RANDOMIZE;
1068         flush_thread();
1069
1070         /* Set the new mm task size. We have to do that late because it may
1071          * depend on TIF_32BIT which is only updated in flush_thread() on
1072          * some architectures like powerpc
1073          */
1074         current->mm->task_size = TASK_SIZE;
1075
1076         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1077                 suid_keys(current);
1078                 set_dumpable(current->mm, suid_dumpable);
1079                 current->pdeath_signal = 0;
1080         } else if (file_permission(bprm->file, MAY_READ) ||
1081                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1082                 suid_keys(current);
1083                 set_dumpable(current->mm, suid_dumpable);
1084         }
1085
1086         /* An exec changes our domain. We are no longer part of the thread
1087            group */
1088
1089         current->self_exec_id++;
1090
1091         flush_signal_handlers(current, 0);
1092         flush_old_files(current->files);
1093
1094         return 0;
1095
1096 mmap_failed:
1097         reset_files_struct(current, files);
1098 out:
1099         return retval;
1100 }
1101
1102 EXPORT_SYMBOL(flush_old_exec);
1103
1104 /*
1105  * Fill the binprm structure from the inode.
1106  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1107  */
1108 int prepare_binprm(struct linux_binprm *bprm)
1109 {
1110         int mode;
1111         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1112         int retval;
1113
1114         mode = inode->i_mode;
1115         if (bprm->file->f_op == NULL)
1116                 return -EACCES;
1117
1118         bprm->e_uid = current->euid;
1119         bprm->e_gid = current->egid;
1120
1121         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1122                 /* Set-uid? */
1123                 if (mode & S_ISUID) {
1124                         current->personality &= ~PER_CLEAR_ON_SETID;
1125                         bprm->e_uid = inode->i_uid;
1126                 }
1127
1128                 /* Set-gid? */
1129                 /*
1130                  * If setgid is set but no group execute bit then this
1131                  * is a candidate for mandatory locking, not a setgid
1132                  * executable.
1133                  */
1134                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1135                         current->personality &= ~PER_CLEAR_ON_SETID;
1136                         bprm->e_gid = inode->i_gid;
1137                 }
1138         }
1139
1140         /* fill in binprm security blob */
1141         retval = security_bprm_set(bprm);
1142         if (retval)
1143                 return retval;
1144
1145         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1146         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1147 }
1148
1149 EXPORT_SYMBOL(prepare_binprm);
1150
1151 static int unsafe_exec(struct task_struct *p)
1152 {
1153         int unsafe = 0;
1154         if (p->ptrace & PT_PTRACED) {
1155                 if (p->ptrace & PT_PTRACE_CAP)
1156                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1157                 else
1158                         unsafe |= LSM_UNSAFE_PTRACE;
1159         }
1160         if (atomic_read(&p->fs->count) > 1 ||
1161             atomic_read(&p->files->count) > 1 ||
1162             atomic_read(&p->sighand->count) > 1)
1163                 unsafe |= LSM_UNSAFE_SHARE;
1164
1165         return unsafe;
1166 }
1167
1168 void compute_creds(struct linux_binprm *bprm)
1169 {
1170         int unsafe;
1171
1172         if (bprm->e_uid != current->uid) {
1173                 suid_keys(current);
1174                 current->pdeath_signal = 0;
1175         }
1176         exec_keys(current);
1177
1178         task_lock(current);
1179         unsafe = unsafe_exec(current);
1180         security_bprm_apply_creds(bprm, unsafe);
1181         task_unlock(current);
1182         security_bprm_post_apply_creds(bprm);
1183 }
1184 EXPORT_SYMBOL(compute_creds);
1185
1186 /*
1187  * Arguments are '\0' separated strings found at the location bprm->p
1188  * points to; chop off the first by relocating brpm->p to right after
1189  * the first '\0' encountered.
1190  */
1191 int remove_arg_zero(struct linux_binprm *bprm)
1192 {
1193         int ret = 0;
1194         unsigned long offset;
1195         char *kaddr;
1196         struct page *page;
1197
1198         if (!bprm->argc)
1199                 return 0;
1200
1201         do {
1202                 offset = bprm->p & ~PAGE_MASK;
1203                 page = get_arg_page(bprm, bprm->p, 0);
1204                 if (!page) {
1205                         ret = -EFAULT;
1206                         goto out;
1207                 }
1208                 kaddr = kmap_atomic(page, KM_USER0);
1209
1210                 for (; offset < PAGE_SIZE && kaddr[offset];
1211                                 offset++, bprm->p++)
1212                         ;
1213
1214                 kunmap_atomic(kaddr, KM_USER0);
1215                 put_arg_page(page);
1216
1217                 if (offset == PAGE_SIZE)
1218                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1219         } while (offset == PAGE_SIZE);
1220
1221         bprm->p++;
1222         bprm->argc--;
1223         ret = 0;
1224
1225 out:
1226         return ret;
1227 }
1228 EXPORT_SYMBOL(remove_arg_zero);
1229
1230 /*
1231  * cycle the list of binary formats handler, until one recognizes the image
1232  */
1233 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1234 {
1235         int try,retval;
1236         struct linux_binfmt *fmt;
1237 #ifdef __alpha__
1238         /* handle /sbin/loader.. */
1239         {
1240             struct exec * eh = (struct exec *) bprm->buf;
1241
1242             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1243                 (eh->fh.f_flags & 0x3000) == 0x3000)
1244             {
1245                 struct file * file;
1246                 unsigned long loader;
1247
1248                 allow_write_access(bprm->file);
1249                 fput(bprm->file);
1250                 bprm->file = NULL;
1251
1252                 loader = bprm->vma->vm_end - sizeof(void *);
1253
1254                 file = open_exec("/sbin/loader");
1255                 retval = PTR_ERR(file);
1256                 if (IS_ERR(file))
1257                         return retval;
1258
1259                 /* Remember if the application is TASO.  */
1260                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1261
1262                 bprm->file = file;
1263                 bprm->loader = loader;
1264                 retval = prepare_binprm(bprm);
1265                 if (retval<0)
1266                         return retval;
1267                 /* should call search_binary_handler recursively here,
1268                    but it does not matter */
1269             }
1270         }
1271 #endif
1272         retval = security_bprm_check(bprm);
1273         if (retval)
1274                 return retval;
1275
1276         /* kernel module loader fixup */
1277         /* so we don't try to load run modprobe in kernel space. */
1278         set_fs(USER_DS);
1279
1280         retval = audit_bprm(bprm);
1281         if (retval)
1282                 return retval;
1283
1284         retval = -ENOENT;
1285         for (try=0; try<2; try++) {
1286                 read_lock(&binfmt_lock);
1287                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1288                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1289                         if (!fn)
1290                                 continue;
1291                         if (!try_module_get(fmt->module))
1292                                 continue;
1293                         read_unlock(&binfmt_lock);
1294                         retval = fn(bprm, regs);
1295                         if (retval >= 0) {
1296                                 put_binfmt(fmt);
1297                                 allow_write_access(bprm->file);
1298                                 if (bprm->file)
1299                                         fput(bprm->file);
1300                                 bprm->file = NULL;
1301                                 current->did_exec = 1;
1302                                 proc_exec_connector(current);
1303                                 return retval;
1304                         }
1305                         read_lock(&binfmt_lock);
1306                         put_binfmt(fmt);
1307                         if (retval != -ENOEXEC || bprm->mm == NULL)
1308                                 break;
1309                         if (!bprm->file) {
1310                                 read_unlock(&binfmt_lock);
1311                                 return retval;
1312                         }
1313                 }
1314                 read_unlock(&binfmt_lock);
1315                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1316                         break;
1317 #ifdef CONFIG_KMOD
1318                 }else{
1319 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1320                         if (printable(bprm->buf[0]) &&
1321                             printable(bprm->buf[1]) &&
1322                             printable(bprm->buf[2]) &&
1323                             printable(bprm->buf[3]))
1324                                 break; /* -ENOEXEC */
1325                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1326 #endif
1327                 }
1328         }
1329         return retval;
1330 }
1331
1332 EXPORT_SYMBOL(search_binary_handler);
1333
1334 /*
1335  * sys_execve() executes a new program.
1336  */
1337 int do_execve(char * filename,
1338         char __user *__user *argv,
1339         char __user *__user *envp,
1340         struct pt_regs * regs)
1341 {
1342         struct linux_binprm *bprm;
1343         struct file *file;
1344         unsigned long env_p;
1345         int retval;
1346
1347         retval = -ENOMEM;
1348         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1349         if (!bprm)
1350                 goto out_ret;
1351
1352         file = open_exec(filename);
1353         retval = PTR_ERR(file);
1354         if (IS_ERR(file))
1355                 goto out_kfree;
1356
1357         sched_exec();
1358
1359         bprm->file = file;
1360         bprm->filename = filename;
1361         bprm->interp = filename;
1362
1363         retval = bprm_mm_init(bprm);
1364         if (retval)
1365                 goto out_file;
1366
1367         bprm->argc = count(argv, MAX_ARG_STRINGS);
1368         if ((retval = bprm->argc) < 0)
1369                 goto out_mm;
1370
1371         bprm->envc = count(envp, MAX_ARG_STRINGS);
1372         if ((retval = bprm->envc) < 0)
1373                 goto out_mm;
1374
1375         retval = security_bprm_alloc(bprm);
1376         if (retval)
1377                 goto out;
1378
1379         retval = prepare_binprm(bprm);
1380         if (retval < 0)
1381                 goto out;
1382
1383         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1384         if (retval < 0)
1385                 goto out;
1386
1387         bprm->exec = bprm->p;
1388         retval = copy_strings(bprm->envc, envp, bprm);
1389         if (retval < 0)
1390                 goto out;
1391
1392         env_p = bprm->p;
1393         retval = copy_strings(bprm->argc, argv, bprm);
1394         if (retval < 0)
1395                 goto out;
1396         bprm->argv_len = env_p - bprm->p;
1397
1398         retval = search_binary_handler(bprm,regs);
1399         if (retval >= 0) {
1400                 /* execve success */
1401                 free_arg_pages(bprm);
1402                 security_bprm_free(bprm);
1403                 acct_update_integrals(current);
1404                 kfree(bprm);
1405                 return retval;
1406         }
1407
1408 out:
1409         free_arg_pages(bprm);
1410         if (bprm->security)
1411                 security_bprm_free(bprm);
1412
1413 out_mm:
1414         if (bprm->mm)
1415                 mmput (bprm->mm);
1416
1417 out_file:
1418         if (bprm->file) {
1419                 allow_write_access(bprm->file);
1420                 fput(bprm->file);
1421         }
1422 out_kfree:
1423         kfree(bprm);
1424
1425 out_ret:
1426         return retval;
1427 }
1428
1429 int set_binfmt(struct linux_binfmt *new)
1430 {
1431         struct linux_binfmt *old = current->binfmt;
1432
1433         if (new) {
1434                 if (!try_module_get(new->module))
1435                         return -1;
1436         }
1437         current->binfmt = new;
1438         if (old)
1439                 module_put(old->module);
1440         return 0;
1441 }
1442
1443 EXPORT_SYMBOL(set_binfmt);
1444
1445 /* format_corename will inspect the pattern parameter, and output a
1446  * name into corename, which must have space for at least
1447  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1448  */
1449 static int format_corename(char *corename, const char *pattern, long signr)
1450 {
1451         const char *pat_ptr = pattern;
1452         char *out_ptr = corename;
1453         char *const out_end = corename + CORENAME_MAX_SIZE;
1454         int rc;
1455         int pid_in_pattern = 0;
1456         int ispipe = 0;
1457
1458         if (*pattern == '|')
1459                 ispipe = 1;
1460
1461         /* Repeat as long as we have more pattern to process and more output
1462            space */
1463         while (*pat_ptr) {
1464                 if (*pat_ptr != '%') {
1465                         if (out_ptr == out_end)
1466                                 goto out;
1467                         *out_ptr++ = *pat_ptr++;
1468                 } else {
1469                         switch (*++pat_ptr) {
1470                         case 0:
1471                                 goto out;
1472                         /* Double percent, output one percent */
1473                         case '%':
1474                                 if (out_ptr == out_end)
1475                                         goto out;
1476                                 *out_ptr++ = '%';
1477                                 break;
1478                         /* pid */
1479                         case 'p':
1480                                 pid_in_pattern = 1;
1481                                 rc = snprintf(out_ptr, out_end - out_ptr,
1482                                               "%d", current->tgid);
1483                                 if (rc > out_end - out_ptr)
1484                                         goto out;
1485                                 out_ptr += rc;
1486                                 break;
1487                         /* uid */
1488                         case 'u':
1489                                 rc = snprintf(out_ptr, out_end - out_ptr,
1490                                               "%d", current->uid);
1491                                 if (rc > out_end - out_ptr)
1492                                         goto out;
1493                                 out_ptr += rc;
1494                                 break;
1495                         /* gid */
1496                         case 'g':
1497                                 rc = snprintf(out_ptr, out_end - out_ptr,
1498                                               "%d", current->gid);
1499                                 if (rc > out_end - out_ptr)
1500                                         goto out;
1501                                 out_ptr += rc;
1502                                 break;
1503                         /* signal that caused the coredump */
1504                         case 's':
1505                                 rc = snprintf(out_ptr, out_end - out_ptr,
1506                                               "%ld", signr);
1507                                 if (rc > out_end - out_ptr)
1508                                         goto out;
1509                                 out_ptr += rc;
1510                                 break;
1511                         /* UNIX time of coredump */
1512                         case 't': {
1513                                 struct timeval tv;
1514                                 do_gettimeofday(&tv);
1515                                 rc = snprintf(out_ptr, out_end - out_ptr,
1516                                               "%lu", tv.tv_sec);
1517                                 if (rc > out_end - out_ptr)
1518                                         goto out;
1519                                 out_ptr += rc;
1520                                 break;
1521                         }
1522                         /* hostname */
1523                         case 'h':
1524                                 down_read(&uts_sem);
1525                                 rc = snprintf(out_ptr, out_end - out_ptr,
1526                                               "%s", utsname()->nodename);
1527                                 up_read(&uts_sem);
1528                                 if (rc > out_end - out_ptr)
1529                                         goto out;
1530                                 out_ptr += rc;
1531                                 break;
1532                         /* executable */
1533                         case 'e':
1534                                 rc = snprintf(out_ptr, out_end - out_ptr,
1535                                               "%s", current->comm);
1536                                 if (rc > out_end - out_ptr)
1537                                         goto out;
1538                                 out_ptr += rc;
1539                                 break;
1540                         default:
1541                                 break;
1542                         }
1543                         ++pat_ptr;
1544                 }
1545         }
1546         /* Backward compatibility with core_uses_pid:
1547          *
1548          * If core_pattern does not include a %p (as is the default)
1549          * and core_uses_pid is set, then .%pid will be appended to
1550          * the filename. Do not do this for piped commands. */
1551         if (!ispipe && !pid_in_pattern
1552             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1553                 rc = snprintf(out_ptr, out_end - out_ptr,
1554                               ".%d", current->tgid);
1555                 if (rc > out_end - out_ptr)
1556                         goto out;
1557                 out_ptr += rc;
1558         }
1559 out:
1560         *out_ptr = 0;
1561         return ispipe;
1562 }
1563
1564 static void zap_process(struct task_struct *start)
1565 {
1566         struct task_struct *t;
1567
1568         start->signal->flags = SIGNAL_GROUP_EXIT;
1569         start->signal->group_stop_count = 0;
1570
1571         t = start;
1572         do {
1573                 if (t != current && t->mm) {
1574                         t->mm->core_waiters++;
1575                         sigaddset(&t->pending.signal, SIGKILL);
1576                         signal_wake_up(t, 1);
1577                 }
1578         } while ((t = next_thread(t)) != start);
1579 }
1580
1581 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1582                                 int exit_code)
1583 {
1584         struct task_struct *g, *p;
1585         unsigned long flags;
1586         int err = -EAGAIN;
1587
1588         spin_lock_irq(&tsk->sighand->siglock);
1589         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1590                 tsk->signal->group_exit_code = exit_code;
1591                 zap_process(tsk);
1592                 err = 0;
1593         }
1594         spin_unlock_irq(&tsk->sighand->siglock);
1595         if (err)
1596                 return err;
1597
1598         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1599                 goto done;
1600
1601         rcu_read_lock();
1602         for_each_process(g) {
1603                 if (g == tsk->group_leader)
1604                         continue;
1605
1606                 p = g;
1607                 do {
1608                         if (p->mm) {
1609                                 if (p->mm == mm) {
1610                                         /*
1611                                          * p->sighand can't disappear, but
1612                                          * may be changed by de_thread()
1613                                          */
1614                                         lock_task_sighand(p, &flags);
1615                                         zap_process(p);
1616                                         unlock_task_sighand(p, &flags);
1617                                 }
1618                                 break;
1619                         }
1620                 } while ((p = next_thread(p)) != g);
1621         }
1622         rcu_read_unlock();
1623 done:
1624         return mm->core_waiters;
1625 }
1626
1627 static int coredump_wait(int exit_code)
1628 {
1629         struct task_struct *tsk = current;
1630         struct mm_struct *mm = tsk->mm;
1631         struct completion startup_done;
1632         struct completion *vfork_done;
1633         int core_waiters;
1634
1635         init_completion(&mm->core_done);
1636         init_completion(&startup_done);
1637         mm->core_startup_done = &startup_done;
1638
1639         core_waiters = zap_threads(tsk, mm, exit_code);
1640         up_write(&mm->mmap_sem);
1641
1642         if (unlikely(core_waiters < 0))
1643                 goto fail;
1644
1645         /*
1646          * Make sure nobody is waiting for us to release the VM,
1647          * otherwise we can deadlock when we wait on each other
1648          */
1649         vfork_done = tsk->vfork_done;
1650         if (vfork_done) {
1651                 tsk->vfork_done = NULL;
1652                 complete(vfork_done);
1653         }
1654
1655         if (core_waiters)
1656                 wait_for_completion(&startup_done);
1657 fail:
1658         BUG_ON(mm->core_waiters);
1659         return core_waiters;
1660 }
1661
1662 /*
1663  * set_dumpable converts traditional three-value dumpable to two flags and
1664  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1665  * these bits are not changed atomically.  So get_dumpable can observe the
1666  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1667  * return either old dumpable or new one by paying attention to the order of
1668  * modifying the bits.
1669  *
1670  * dumpable |   mm->flags (binary)
1671  * old  new | initial interim  final
1672  * ---------+-----------------------
1673  *  0    1  |   00      01      01
1674  *  0    2  |   00      10(*)   11
1675  *  1    0  |   01      00      00
1676  *  1    2  |   01      11      11
1677  *  2    0  |   11      10(*)   00
1678  *  2    1  |   11      11      01
1679  *
1680  * (*) get_dumpable regards interim value of 10 as 11.
1681  */
1682 void set_dumpable(struct mm_struct *mm, int value)
1683 {
1684         switch (value) {
1685         case 0:
1686                 clear_bit(MMF_DUMPABLE, &mm->flags);
1687                 smp_wmb();
1688                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1689                 break;
1690         case 1:
1691                 set_bit(MMF_DUMPABLE, &mm->flags);
1692                 smp_wmb();
1693                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1694                 break;
1695         case 2:
1696                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1697                 smp_wmb();
1698                 set_bit(MMF_DUMPABLE, &mm->flags);
1699                 break;
1700         }
1701 }
1702 EXPORT_SYMBOL_GPL(set_dumpable);
1703
1704 int get_dumpable(struct mm_struct *mm)
1705 {
1706         int ret;
1707
1708         ret = mm->flags & 0x3;
1709         return (ret >= 2) ? 2 : ret;
1710 }
1711
1712 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1713 {
1714         char corename[CORENAME_MAX_SIZE + 1];
1715         struct mm_struct *mm = current->mm;
1716         struct linux_binfmt * binfmt;
1717         struct inode * inode;
1718         struct file * file;
1719         int retval = 0;
1720         int fsuid = current->fsuid;
1721         int flag = 0;
1722         int ispipe = 0;
1723
1724         audit_core_dumps(signr);
1725
1726         binfmt = current->binfmt;
1727         if (!binfmt || !binfmt->core_dump)
1728                 goto fail;
1729         down_write(&mm->mmap_sem);
1730         if (!get_dumpable(mm)) {
1731                 up_write(&mm->mmap_sem);
1732                 goto fail;
1733         }
1734
1735         /*
1736          *      We cannot trust fsuid as being the "true" uid of the
1737          *      process nor do we know its entire history. We only know it
1738          *      was tainted so we dump it as root in mode 2.
1739          */
1740         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1741                 flag = O_EXCL;          /* Stop rewrite attacks */
1742                 current->fsuid = 0;     /* Dump root private */
1743         }
1744         set_dumpable(mm, 0);
1745
1746         retval = coredump_wait(exit_code);
1747         if (retval < 0)
1748                 goto fail;
1749
1750         /*
1751          * Clear any false indication of pending signals that might
1752          * be seen by the filesystem code called to write the core file.
1753          */
1754         clear_thread_flag(TIF_SIGPENDING);
1755
1756         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1757                 goto fail_unlock;
1758
1759         /*
1760          * lock_kernel() because format_corename() is controlled by sysctl, which
1761          * uses lock_kernel()
1762          */
1763         lock_kernel();
1764         ispipe = format_corename(corename, core_pattern, signr);
1765         unlock_kernel();
1766         if (ispipe) {
1767                 /* SIGPIPE can happen, but it's just never processed */
1768                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1769                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1770                                corename);
1771                         goto fail_unlock;
1772                 }
1773         } else
1774                 file = filp_open(corename,
1775                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1776                                  0600);
1777         if (IS_ERR(file))
1778                 goto fail_unlock;
1779         inode = file->f_path.dentry->d_inode;
1780         if (inode->i_nlink > 1)
1781                 goto close_fail;        /* multiple links - don't dump */
1782         if (!ispipe && d_unhashed(file->f_path.dentry))
1783                 goto close_fail;
1784
1785         /* AK: actually i see no reason to not allow this for named pipes etc.,
1786            but keep the previous behaviour for now. */
1787         if (!ispipe && !S_ISREG(inode->i_mode))
1788                 goto close_fail;
1789         if (!file->f_op)
1790                 goto close_fail;
1791         if (!file->f_op->write)
1792                 goto close_fail;
1793         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1794                 goto close_fail;
1795
1796         retval = binfmt->core_dump(signr, regs, file);
1797
1798         if (retval)
1799                 current->signal->group_exit_code |= 0x80;
1800 close_fail:
1801         filp_close(file, NULL);
1802 fail_unlock:
1803         current->fsuid = fsuid;
1804         complete_all(&mm->core_done);
1805 fail:
1806         return retval;
1807 }