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_event.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 #include <linux/fsnotify.h>
  56 #include <linux/fs_struct.h>
  57 #include <linux/pipe_fs_i.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62 #include "internal.h"
  63
  64 int core_uses_pid;
  65 char core_pattern[CORENAME_MAX_SIZE] = "core";
  66 unsigned int core_pipe_limit;
  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, int insert)
  75 {
  76         if (!fmt)
  77                 return -EINVAL;
  78         write_lock(&binfmt_lock);
  79         insert ? list_add(&fmt->lh, &formats) :
  80                  list_add_tail(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82         return 0;
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 SYSCALL_DEFINE1(uselib, const char __user *, library)
 108 {
 109         struct file *file;
 110         char *tmp = getname(library);
 111         int error = PTR_ERR(tmp);
 112
 113         if (IS_ERR(tmp))
 114                 goto out;
 115
 116         file = do_filp_open(AT_FDCWD, tmp,
 117                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 118                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 119         putname(tmp);
 120         error = PTR_ERR(file);
 121         if (IS_ERR(file))
 122                 goto out;
 123
 124         error = -EINVAL;
 125         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 126                 goto exit;
 127
 128         error = -EACCES;
 129         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 130                 goto exit;
 131
 132         fsnotify_open(file->f_path.dentry);
 133
 134         error = -ENOEXEC;
 135         if(file->f_op) {
 136                 struct linux_binfmt * fmt;
 137
 138                 read_lock(&binfmt_lock);
 139                 list_for_each_entry(fmt, &formats, lh) {
 140                         if (!fmt->load_shlib)
 141                                 continue;
 142                         if (!try_module_get(fmt->module))
 143                                 continue;
 144                         read_unlock(&binfmt_lock);
 145                         error = fmt->load_shlib(file);
 146                         read_lock(&binfmt_lock);
 147                         put_binfmt(fmt);
 148                         if (error != -ENOEXEC)
 149                                 break;
 150                 }
 151                 read_unlock(&binfmt_lock);
 152         }
 153 exit:
 154         fput(file);
 155 out:
 156         return error;
 157 }
 158
 159 #ifdef CONFIG_MMU
 160
 161 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 162                 int write)
 163 {
 164         struct page *page;
 165         int ret;
 166
 167 #ifdef CONFIG_STACK_GROWSUP
 168         if (write) {
 169                 ret = expand_stack_downwards(bprm->vma, pos);
 170                 if (ret < 0)
 171                         return NULL;
 172         }
 173 #endif
 174         ret = get_user_pages(current, bprm->mm, pos,
 175                         1, write, 1, &page, NULL);
 176         if (ret <= 0)
 177                 return NULL;
 178
 179         if (write) {
 180                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 181                 struct rlimit *rlim;
 182
 183                 /*
 184                  * We've historically supported up to 32 pages (ARG_MAX)
 185                  * of argument strings even with small stacks
 186                  */
 187                 if (size <= ARG_MAX)
 188                         return page;
 189
 190                 /*
 191                  * Limit to 1/4-th the stack size for the argv+env strings.
 192                  * This ensures that:
 193                  *  - the remaining binfmt code will not run out of stack space,
 194                  *  - the program will have a reasonable amount of stack left
 195                  *    to work from.
 196                  */
 197                 rlim = current->signal->rlim;
 198                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 199                         put_page(page);
 200                         return NULL;
 201                 }
 202         }
 203
 204         return page;
 205 }
 206
 207 static void put_arg_page(struct page *page)
 208 {
 209         put_page(page);
 210 }
 211
 212 static void free_arg_page(struct linux_binprm *bprm, int i)
 213 {
 214 }
 215
 216 static void free_arg_pages(struct linux_binprm *bprm)
 217 {
 218 }
 219
 220 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 221                 struct page *page)
 222 {
 223         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 224 }
 225
 226 static int __bprm_mm_init(struct linux_binprm *bprm)
 227 {
 228         int err;
 229         struct vm_area_struct *vma = NULL;
 230         struct mm_struct *mm = bprm->mm;
 231
 232         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 233         if (!vma)
 234                 return -ENOMEM;
 235
 236         down_write(&mm->mmap_sem);
 237         vma->vm_mm = mm;
 238
 239         /*
 240          * Place the stack at the largest stack address the architecture
 241          * supports. Later, we'll move this to an appropriate place. We don't
 242          * use STACK_TOP because that can depend on attributes which aren't
 243          * configured yet.
 244          */
 245         vma->vm_end = STACK_TOP_MAX;
 246         vma->vm_start = vma->vm_end - PAGE_SIZE;
 247         vma->vm_flags = VM_STACK_FLAGS;
 248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 249         err = insert_vm_struct(mm, vma);
 250         if (err)
 251                 goto err;
 252
 253         mm->stack_vm = mm->total_vm = 1;
 254         up_write(&mm->mmap_sem);
 255         bprm->p = vma->vm_end - sizeof(void *);
 256         return 0;
 257 err:
 258         up_write(&mm->mmap_sem);
 259         bprm->vma = NULL;
 260         kmem_cache_free(vm_area_cachep, vma);
 261         return err;
 262 }
 263
 264 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 265 {
 266         return len <= MAX_ARG_STRLEN;
 267 }
 268
 269 #else
 270
 271 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 272                 int write)
 273 {
 274         struct page *page;
 275
 276         page = bprm->page[pos / PAGE_SIZE];
 277         if (!page && write) {
 278                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 279                 if (!page)
 280                         return NULL;
 281                 bprm->page[pos / PAGE_SIZE] = page;
 282         }
 283
 284         return page;
 285 }
 286
 287 static void put_arg_page(struct page *page)
 288 {
 289 }
 290
 291 static void free_arg_page(struct linux_binprm *bprm, int i)
 292 {
 293         if (bprm->page[i]) {
 294                 __free_page(bprm->page[i]);
 295                 bprm->page[i] = NULL;
 296         }
 297 }
 298
 299 static void free_arg_pages(struct linux_binprm *bprm)
 300 {
 301         int i;
 302
 303         for (i = 0; i < MAX_ARG_PAGES; i++)
 304                 free_arg_page(bprm, i);
 305 }
 306
 307 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 308                 struct page *page)
 309 {
 310 }
 311
 312 static int __bprm_mm_init(struct linux_binprm *bprm)
 313 {
 314         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 315         return 0;
 316 }
 317
 318 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 319 {
 320         return len <= bprm->p;
 321 }
 322
 323 #endif /* CONFIG_MMU */
 324
 325 /*
 326  * Create a new mm_struct and populate it with a temporary stack
 327  * vm_area_struct.  We don't have enough context at this point to set the stack
 328  * flags, permissions, and offset, so we use temporary values.  We'll update
 329  * them later in setup_arg_pages().
 330  */
 331 int bprm_mm_init(struct linux_binprm *bprm)
 332 {
 333         int err;
 334         struct mm_struct *mm = NULL;
 335
 336         bprm->mm = mm = mm_alloc();
 337         err = -ENOMEM;
 338         if (!mm)
 339                 goto err;
 340
 341         err = init_new_context(current, mm);
 342         if (err)
 343                 goto err;
 344
 345         err = __bprm_mm_init(bprm);
 346         if (err)
 347                 goto err;
 348
 349         return 0;
 350
 351 err:
 352         if (mm) {
 353                 bprm->mm = NULL;
 354                 mmdrop(mm);
 355         }
 356
 357         return err;
 358 }
 359
 360 /*
 361  * count() counts the number of strings in array ARGV.
 362  */
 363 static int count(char __user * __user * argv, int max)
 364 {
 365         int i = 0;
 366
 367         if (argv != NULL) {
 368                 for (;;) {
 369                         char __user * p;
 370
 371                         if (get_user(p, argv))
 372                                 return -EFAULT;
 373                         if (!p)
 374                                 break;
 375                         argv++;
 376                         if (i++ >= max)
 377                                 return -E2BIG;
 378
 379                         if (fatal_signal_pending(current))
 380                                 return -ERESTARTNOHAND;
 381                         cond_resched();
 382                 }
 383         }
 384         return i;
 385 }
 386
 387 /*
 388  * 'copy_strings()' copies argument/environment strings from the old
 389  * processes's memory to the new process's stack.  The call to get_user_pages()
 390  * ensures the destination page is created and not swapped out.
 391  */
 392 static int copy_strings(int argc, char __user * __user * argv,
 393                         struct linux_binprm *bprm)
 394 {
 395         struct page *kmapped_page = NULL;
 396         char *kaddr = NULL;
 397         unsigned long kpos = 0;
 398         int ret;
 399
 400         while (argc-- > 0) {
 401                 char __user *str;
 402                 int len;
 403                 unsigned long pos;
 404
 405                 if (get_user(str, argv+argc) ||
 406                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 407                         ret = -EFAULT;
 408                         goto out;
 409                 }
 410
 411                 if (!valid_arg_len(bprm, len)) {
 412                         ret = -E2BIG;
 413                         goto out;
 414                 }
 415
 416                 /* We're going to work our way backwords. */
 417                 pos = bprm->p;
 418                 str += len;
 419                 bprm->p -= len;
 420
 421                 while (len > 0) {
 422                         int offset, bytes_to_copy;
 423
 424                         if (fatal_signal_pending(current)) {
 425                                 ret = -ERESTARTNOHAND;
 426                                 goto out;
 427                         }
 428                         cond_resched();
 429
 430                         offset = pos % PAGE_SIZE;
 431                         if (offset == 0)
 432                                 offset = PAGE_SIZE;
 433
 434                         bytes_to_copy = offset;
 435                         if (bytes_to_copy > len)
 436                                 bytes_to_copy = len;
 437
 438                         offset -= bytes_to_copy;
 439                         pos -= bytes_to_copy;
 440                         str -= bytes_to_copy;
 441                         len -= bytes_to_copy;
 442
 443                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 444                                 struct page *page;
 445
 446                                 page = get_arg_page(bprm, pos, 1);
 447                                 if (!page) {
 448                                         ret = -E2BIG;
 449                                         goto out;
 450                                 }
 451
 452                                 if (kmapped_page) {
 453                                         flush_kernel_dcache_page(kmapped_page);
 454                                         kunmap(kmapped_page);
 455                                         put_arg_page(kmapped_page);
 456                                 }
 457                                 kmapped_page = page;
 458                                 kaddr = kmap(kmapped_page);
 459                                 kpos = pos & PAGE_MASK;
 460                                 flush_arg_page(bprm, kpos, kmapped_page);
 461                         }
 462                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 463                                 ret = -EFAULT;
 464                                 goto out;
 465                         }
 466                 }
 467         }
 468         ret = 0;
 469 out:
 470         if (kmapped_page) {
 471                 flush_kernel_dcache_page(kmapped_page);
 472                 kunmap(kmapped_page);
 473                 put_arg_page(kmapped_page);
 474         }
 475         return ret;
 476 }
 477
 478 /*
 479  * Like copy_strings, but get argv and its values from kernel memory.
 480  */
 481 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 482 {
 483         int r;
 484         mm_segment_t oldfs = get_fs();
 485         set_fs(KERNEL_DS);
 486         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 487         set_fs(oldfs);
 488         return r;
 489 }
 490 EXPORT_SYMBOL(copy_strings_kernel);
 491
 492 #ifdef CONFIG_MMU
 493
 494 /*
 495  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 496  * the binfmt code determines where the new stack should reside, we shift it to
 497  * its final location.  The process proceeds as follows:
 498  *
 499  * 1) Use shift to calculate the new vma endpoints.
 500  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 501  *    arguments passed to subsequent functions are consistent.
 502  * 3) Move vma's page tables to the new range.
 503  * 4) Free up any cleared pgd range.
 504  * 5) Shrink the vma to cover only the new range.
 505  */
 506 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 507 {
 508         struct mm_struct *mm = vma->vm_mm;
 509         unsigned long old_start = vma->vm_start;
 510         unsigned long old_end = vma->vm_end;
 511         unsigned long length = old_end - old_start;
 512         unsigned long new_start = old_start - shift;
 513         unsigned long new_end = old_end - shift;
 514         struct mmu_gather *tlb;
 515
 516         BUG_ON(new_start > new_end);
 517
 518         /*
 519          * ensure there are no vmas between where we want to go
 520          * and where we are
 521          */
 522         if (vma != find_vma(mm, new_start))
 523                 return -EFAULT;
 524
 525         /*
 526          * cover the whole range: [new_start, old_end)
 527          */
 528         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 529
 530         /*
 531          * move the page tables downwards, on failure we rely on
 532          * process cleanup to remove whatever mess we made.
 533          */
 534         if (length != move_page_tables(vma, old_start,
 535                                        vma, new_start, length))
 536                 return -ENOMEM;
 537
 538         lru_add_drain();
 539         tlb = tlb_gather_mmu(mm, 0);
 540         if (new_end > old_start) {
 541                 /*
 542                  * when the old and new regions overlap clear from new_end.
 543                  */
 544                 free_pgd_range(tlb, new_end, old_end, new_end,
 545                         vma->vm_next ? vma->vm_next->vm_start : 0);
 546         } else {
 547                 /*
 548                  * otherwise, clean from old_start; this is done to not touch
 549                  * the address space in [new_end, old_start) some architectures
 550                  * have constraints on va-space that make this illegal (IA64) -
 551                  * for the others its just a little faster.
 552                  */
 553                 free_pgd_range(tlb, old_start, old_end, new_end,
 554                         vma->vm_next ? vma->vm_next->vm_start : 0);
 555         }
 556         tlb_finish_mmu(tlb, new_end, old_end);
 557
 558         /*
 559          * shrink the vma to just the new range.
 560          */
 561         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 562
 563         return 0;
 564 }
 565
 566 #define EXTRA_STACK_VM_PAGES    20      /* random */
 567
 568 /*
 569  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 570  * the stack is optionally relocated, and some extra space is added.
 571  */
 572 int setup_arg_pages(struct linux_binprm *bprm,
 573                     unsigned long stack_top,
 574                     int executable_stack)
 575 {
 576         unsigned long ret;
 577         unsigned long stack_shift;
 578         struct mm_struct *mm = current->mm;
 579         struct vm_area_struct *vma = bprm->vma;
 580         struct vm_area_struct *prev = NULL;
 581         unsigned long vm_flags;
 582         unsigned long stack_base;
 583         unsigned long stack_size;
 584         unsigned long stack_expand;
 585         unsigned long rlim_stack;
 586
 587 #ifdef CONFIG_STACK_GROWSUP
 588         /* Limit stack size to 1GB */
 589         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 590         if (stack_base > (1 << 30))
 591                 stack_base = 1 << 30;
 592
 593         /* Make sure we didn't let the argument array grow too large. */
 594         if (vma->vm_end - vma->vm_start > stack_base)
 595                 return -ENOMEM;
 596
 597         stack_base = PAGE_ALIGN(stack_top - stack_base);
 598
 599         stack_shift = vma->vm_start - stack_base;
 600         mm->arg_start = bprm->p - stack_shift;
 601         bprm->p = vma->vm_end - stack_shift;
 602 #else
 603         stack_top = arch_align_stack(stack_top);
 604         stack_top = PAGE_ALIGN(stack_top);
 605
 606         if (unlikely(stack_top < mmap_min_addr) ||
 607             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 608                 return -ENOMEM;
 609
 610         stack_shift = vma->vm_end - stack_top;
 611
 612         bprm->p -= stack_shift;
 613         mm->arg_start = bprm->p;
 614 #endif
 615
 616         if (bprm->loader)
 617                 bprm->loader -= stack_shift;
 618         bprm->exec -= stack_shift;
 619
 620         down_write(&mm->mmap_sem);
 621         vm_flags = VM_STACK_FLAGS;
 622
 623         /*
 624          * Adjust stack execute permissions; explicitly enable for
 625          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 626          * (arch default) otherwise.
 627          */
 628         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 629                 vm_flags |= VM_EXEC;
 630         else if (executable_stack == EXSTACK_DISABLE_X)
 631                 vm_flags &= ~VM_EXEC;
 632         vm_flags |= mm->def_flags;
 633
 634         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 635                         vm_flags);
 636         if (ret)
 637                 goto out_unlock;
 638         BUG_ON(prev != vma);
 639
 640         /* Move stack pages down in memory. */
 641         if (stack_shift) {
 642                 ret = shift_arg_pages(vma, stack_shift);
 643                 if (ret)
 644                         goto out_unlock;
 645         }
 646
 647         stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 648         stack_size = vma->vm_end - vma->vm_start;
 649         /*
 650          * Align this down to a page boundary as expand_stack
 651          * will align it up.
 652          */
 653         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 654 #ifdef CONFIG_STACK_GROWSUP
 655         if (stack_size + stack_expand > rlim_stack)
 656                 stack_base = vma->vm_start + rlim_stack;
 657         else
 658                 stack_base = vma->vm_end + stack_expand;
 659 #else
 660         if (stack_size + stack_expand > rlim_stack)
 661                 stack_base = vma->vm_end - rlim_stack;
 662         else
 663                 stack_base = vma->vm_start - stack_expand;
 664 #endif
 665         ret = expand_stack(vma, stack_base);
 666         if (ret)
 667                 ret = -EFAULT;
 668
 669 out_unlock:
 670         up_write(&mm->mmap_sem);
 671         return ret;
 672 }
 673 EXPORT_SYMBOL(setup_arg_pages);
 674
 675 #endif /* CONFIG_MMU */
 676
 677 struct file *open_exec(const char *name)
 678 {
 679         struct file *file;
 680         int err;
 681
 682         file = do_filp_open(AT_FDCWD, name,
 683                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 684                                 MAY_EXEC | MAY_OPEN);
 685         if (IS_ERR(file))
 686                 goto out;
 687
 688         err = -EACCES;
 689         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 690                 goto exit;
 691
 692         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 693                 goto exit;
 694
 695         fsnotify_open(file->f_path.dentry);
 696
 697         err = deny_write_access(file);
 698         if (err)
 699                 goto exit;
 700
 701 out:
 702         return file;
 703
 704 exit:
 705         fput(file);
 706         return ERR_PTR(err);
 707 }
 708 EXPORT_SYMBOL(open_exec);
 709
 710 int kernel_read(struct file *file, loff_t offset,
 711                 char *addr, unsigned long count)
 712 {
 713         mm_segment_t old_fs;
 714         loff_t pos = offset;
 715         int result;
 716
 717         old_fs = get_fs();
 718         set_fs(get_ds());
 719         /* The cast to a user pointer is valid due to the set_fs() */
 720         result = vfs_read(file, (void __user *)addr, count, &pos);
 721         set_fs(old_fs);
 722         return result;
 723 }
 724
 725 EXPORT_SYMBOL(kernel_read);
 726
 727 static int exec_mmap(struct mm_struct *mm)
 728 {
 729         struct task_struct *tsk;
 730         struct mm_struct * old_mm, *active_mm;
 731
 732         /* Notify parent that we're no longer interested in the old VM */
 733         tsk = current;
 734         old_mm = current->mm;
 735         mm_release(tsk, old_mm);
 736
 737         if (old_mm) {
 738                 /*
 739                  * Make sure that if there is a core dump in progress
 740                  * for the old mm, we get out and die instead of going
 741                  * through with the exec.  We must hold mmap_sem around
 742                  * checking core_state and changing tsk->mm.
 743                  */
 744                 down_read(&old_mm->mmap_sem);
 745                 if (unlikely(old_mm->core_state)) {
 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                 mm_update_next_owner(old_mm);
 761                 mmput(old_mm);
 762                 return 0;
 763         }
 764         mmdrop(active_mm);
 765         return 0;
 766 }
 767
 768 /*
 769  * This function makes sure the current process has its own signal table,
 770  * so that flush_signal_handlers can later reset the handlers without
 771  * disturbing other processes.  (Other processes might share the signal
 772  * table via the CLONE_SIGHAND option to clone().)
 773  */
 774 static int de_thread(struct task_struct *tsk)
 775 {
 776         struct signal_struct *sig = tsk->signal;
 777         struct sighand_struct *oldsighand = tsk->sighand;
 778         spinlock_t *lock = &oldsighand->siglock;
 779         int count;
 780
 781         if (thread_group_empty(tsk))
 782                 goto no_thread_group;
 783
 784         /*
 785          * Kill all other threads in the thread group.
 786          */
 787         spin_lock_irq(lock);
 788         if (signal_group_exit(sig)) {
 789                 /*
 790                  * Another group action in progress, just
 791                  * return so that the signal is processed.
 792                  */
 793                 spin_unlock_irq(lock);
 794                 return -EAGAIN;
 795         }
 796         sig->group_exit_task = tsk;
 797         zap_other_threads(tsk);
 798
 799         /* Account for the thread group leader hanging around: */
 800         count = thread_group_leader(tsk) ? 1 : 2;
 801         sig->notify_count = count;
 802         while (atomic_read(&sig->count) > count) {
 803                 __set_current_state(TASK_UNINTERRUPTIBLE);
 804                 spin_unlock_irq(lock);
 805                 schedule();
 806                 spin_lock_irq(lock);
 807         }
 808         spin_unlock_irq(lock);
 809
 810         /*
 811          * At this point all other threads have exited, all we have to
 812          * do is to wait for the thread group leader to become inactive,
 813          * and to assume its PID:
 814          */
 815         if (!thread_group_leader(tsk)) {
 816                 struct task_struct *leader = tsk->group_leader;
 817
 818                 sig->notify_count = -1; /* for exit_notify() */
 819                 for (;;) {
 820                         write_lock_irq(&tasklist_lock);
 821                         if (likely(leader->exit_state))
 822                                 break;
 823                         __set_current_state(TASK_UNINTERRUPTIBLE);
 824                         write_unlock_irq(&tasklist_lock);
 825                         schedule();
 826                 }
 827
 828                 /*
 829                  * The only record we have of the real-time age of a
 830                  * process, regardless of execs it's done, is start_time.
 831                  * All the past CPU time is accumulated in signal_struct
 832                  * from sister threads now dead.  But in this non-leader
 833                  * exec, nothing survives from the original leader thread,
 834                  * whose birth marks the true age of this process now.
 835                  * When we take on its identity by switching to its PID, we
 836                  * also take its birthdate (always earlier than our own).
 837                  */
 838                 tsk->start_time = leader->start_time;
 839
 840                 BUG_ON(!same_thread_group(leader, tsk));
 841                 BUG_ON(has_group_leader_pid(tsk));
 842                 /*
 843                  * An exec() starts a new thread group with the
 844                  * TGID of the previous thread group. Rehash the
 845                  * two threads with a switched PID, and release
 846                  * the former thread group leader:
 847                  */
 848
 849                 /* Become a process group leader with the old leader's pid.
 850                  * The old leader becomes a thread of the this thread group.
 851                  * Note: The old leader also uses this pid until release_task
 852                  *       is called.  Odd but simple and correct.
 853                  */
 854                 detach_pid(tsk, PIDTYPE_PID);
 855                 tsk->pid = leader->pid;
 856                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 857                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 858                 transfer_pid(leader, tsk, PIDTYPE_SID);
 859
 860                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 861                 list_replace_init(&leader->sibling, &tsk->sibling);
 862
 863                 tsk->group_leader = tsk;
 864                 leader->group_leader = tsk;
 865
 866                 tsk->exit_signal = SIGCHLD;
 867
 868                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 869                 leader->exit_state = EXIT_DEAD;
 870                 write_unlock_irq(&tasklist_lock);
 871
 872                 release_task(leader);
 873         }
 874
 875         sig->group_exit_task = NULL;
 876         sig->notify_count = 0;
 877
 878 no_thread_group:
 879         if (current->mm)
 880                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 881
 882         exit_itimers(sig);
 883         flush_itimer_signals();
 884
 885         if (atomic_read(&oldsighand->count) != 1) {
 886                 struct sighand_struct *newsighand;
 887                 /*
 888                  * This ->sighand is shared with the CLONE_SIGHAND
 889                  * but not CLONE_THREAD task, switch to the new one.
 890                  */
 891                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 892                 if (!newsighand)
 893                         return -ENOMEM;
 894
 895                 atomic_set(&newsighand->count, 1);
 896                 memcpy(newsighand->action, oldsighand->action,
 897                        sizeof(newsighand->action));
 898
 899                 write_lock_irq(&tasklist_lock);
 900                 spin_lock(&oldsighand->siglock);
 901                 rcu_assign_pointer(tsk->sighand, newsighand);
 902                 spin_unlock(&oldsighand->siglock);
 903                 write_unlock_irq(&tasklist_lock);
 904
 905                 __cleanup_sighand(oldsighand);
 906         }
 907
 908         BUG_ON(!thread_group_leader(tsk));
 909         return 0;
 910 }
 911
 912 /*
 913  * These functions flushes out all traces of the currently running executable
 914  * so that a new one can be started
 915  */
 916 static void flush_old_files(struct files_struct * files)
 917 {
 918         long j = -1;
 919         struct fdtable *fdt;
 920
 921         spin_lock(&files->file_lock);
 922         for (;;) {
 923                 unsigned long set, i;
 924
 925                 j++;
 926                 i = j * __NFDBITS;
 927                 fdt = files_fdtable(files);
 928                 if (i >= fdt->max_fds)
 929                         break;
 930                 set = fdt->close_on_exec->fds_bits[j];
 931                 if (!set)
 932                         continue;
 933                 fdt->close_on_exec->fds_bits[j] = 0;
 934                 spin_unlock(&files->file_lock);
 935                 for ( ; set ; i++,set >>= 1) {
 936                         if (set & 1) {
 937                                 sys_close(i);
 938                         }
 939                 }
 940                 spin_lock(&files->file_lock);
 941
 942         }
 943         spin_unlock(&files->file_lock);
 944 }
 945
 946 char *get_task_comm(char *buf, struct task_struct *tsk)
 947 {
 948         /* buf must be at least sizeof(tsk->comm) in size */
 949         task_lock(tsk);
 950         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 951         task_unlock(tsk);
 952         return buf;
 953 }
 954
 955 void set_task_comm(struct task_struct *tsk, char *buf)
 956 {
 957         task_lock(tsk);
 958
 959         /*
 960          * Threads may access current->comm without holding
 961          * the task lock, so write the string carefully.
 962          * Readers without a lock may see incomplete new
 963          * names but are safe from non-terminating string reads.
 964          */
 965         memset(tsk->comm, 0, TASK_COMM_LEN);
 966         wmb();
 967         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 968         task_unlock(tsk);
 969         perf_event_comm(tsk);
 970 }
 971
 972 int flush_old_exec(struct linux_binprm * bprm)
 973 {
 974         int retval;
 975
 976         /*
 977          * Make sure we have a private signal table and that
 978          * we are unassociated from the previous thread group.
 979          */
 980         retval = de_thread(current);
 981         if (retval)
 982                 goto out;
 983
 984         set_mm_exe_file(bprm->mm, bprm->file);
 985
 986         /*
 987          * Release all of the old mmap stuff
 988          */
 989         retval = exec_mmap(bprm->mm);
 990         if (retval)
 991                 goto out;
 992
 993         bprm->mm = NULL;                /* We're using it now */
 994
 995         current->flags &= ~PF_RANDOMIZE;
 996         flush_thread();
 997         current->personality &= ~bprm->per_clear;
 998
 999         return 0;
1000
1001 out:
1002         return retval;
1003 }
1004 EXPORT_SYMBOL(flush_old_exec);
1005
1006 void setup_new_exec(struct linux_binprm * bprm)
1007 {
1008         int i, ch;
1009         char * name;
1010         char tcomm[sizeof(current->comm)];
1011
1012         arch_pick_mmap_layout(current->mm);
1013
1014         /* This is the point of no return */
1015         current->sas_ss_sp = current->sas_ss_size = 0;
1016
1017         if (current_euid() == current_uid() && current_egid() == current_gid())
1018                 set_dumpable(current->mm, 1);
1019         else
1020                 set_dumpable(current->mm, suid_dumpable);
1021
1022         name = bprm->filename;
1023
1024         /* Copies the binary name from after last slash */
1025         for (i=0; (ch = *(name++)) != '\0';) {
1026                 if (ch == '/')
1027                         i = 0; /* overwrite what we wrote */
1028                 else
1029                         if (i < (sizeof(tcomm) - 1))
1030                                 tcomm[i++] = ch;
1031         }
1032         tcomm[i] = '\0';
1033         set_task_comm(current, tcomm);
1034
1035         /* Set the new mm task size. We have to do that late because it may
1036          * depend on TIF_32BIT which is only updated in flush_thread() on
1037          * some architectures like powerpc
1038          */
1039         current->mm->task_size = TASK_SIZE;
1040
1041         /* install the new credentials */
1042         if (bprm->cred->uid != current_euid() ||
1043             bprm->cred->gid != current_egid()) {
1044                 current->pdeath_signal = 0;
1045         } else if (file_permission(bprm->file, MAY_READ) ||
1046                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1047                 set_dumpable(current->mm, suid_dumpable);
1048         }
1049
1050         /*
1051          * Flush performance counters when crossing a
1052          * security domain:
1053          */
1054         if (!get_dumpable(current->mm))
1055                 perf_event_exit_task(current);
1056
1057         /* An exec changes our domain. We are no longer part of the thread
1058            group */
1059
1060         current->self_exec_id++;
1061
1062         flush_signal_handlers(current, 0);
1063         flush_old_files(current->files);
1064 }
1065 EXPORT_SYMBOL(setup_new_exec);
1066
1067 /*
1068  * Prepare credentials and lock ->cred_guard_mutex.
1069  * install_exec_creds() commits the new creds and drops the lock.
1070  * Or, if exec fails before, free_bprm() should release ->cred and
1071  * and unlock.
1072  */
1073 int prepare_bprm_creds(struct linux_binprm *bprm)
1074 {
1075         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1076                 return -ERESTARTNOINTR;
1077
1078         bprm->cred = prepare_exec_creds();
1079         if (likely(bprm->cred))
1080                 return 0;
1081
1082         mutex_unlock(&current->cred_guard_mutex);
1083         return -ENOMEM;
1084 }
1085
1086 void free_bprm(struct linux_binprm *bprm)
1087 {
1088         free_arg_pages(bprm);
1089         if (bprm->cred) {
1090                 mutex_unlock(&current->cred_guard_mutex);
1091                 abort_creds(bprm->cred);
1092         }
1093         kfree(bprm);
1094 }
1095
1096 /*
1097  * install the new credentials for this executable
1098  */
1099 void install_exec_creds(struct linux_binprm *bprm)
1100 {
1101         security_bprm_committing_creds(bprm);
1102
1103         commit_creds(bprm->cred);
1104         bprm->cred = NULL;
1105         /*
1106          * cred_guard_mutex must be held at least to this point to prevent
1107          * ptrace_attach() from altering our determination of the task's
1108          * credentials; any time after this it may be unlocked.
1109          */
1110         security_bprm_committed_creds(bprm);
1111         mutex_unlock(&current->cred_guard_mutex);
1112 }
1113 EXPORT_SYMBOL(install_exec_creds);
1114
1115 /*
1116  * determine how safe it is to execute the proposed program
1117  * - the caller must hold current->cred_guard_mutex to protect against
1118  *   PTRACE_ATTACH
1119  */
1120 int check_unsafe_exec(struct linux_binprm *bprm)
1121 {
1122         struct task_struct *p = current, *t;
1123         unsigned n_fs;
1124         int res = 0;
1125
1126         bprm->unsafe = tracehook_unsafe_exec(p);
1127
1128         n_fs = 1;
1129         write_lock(&p->fs->lock);
1130         rcu_read_lock();
1131         for (t = next_thread(p); t != p; t = next_thread(t)) {
1132                 if (t->fs == p->fs)
1133                         n_fs++;
1134         }
1135         rcu_read_unlock();
1136
1137         if (p->fs->users > n_fs) {
1138                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1139         } else {
1140                 res = -EAGAIN;
1141                 if (!p->fs->in_exec) {
1142                         p->fs->in_exec = 1;
1143                         res = 1;
1144                 }
1145         }
1146         write_unlock(&p->fs->lock);
1147
1148         return res;
1149 }
1150
1151 /*
1152  * Fill the binprm structure from the inode.
1153  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1154  *
1155  * This may be called multiple times for binary chains (scripts for example).
1156  */
1157 int prepare_binprm(struct linux_binprm *bprm)
1158 {
1159         umode_t mode;
1160         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1161         int retval;
1162
1163         mode = inode->i_mode;
1164         if (bprm->file->f_op == NULL)
1165                 return -EACCES;
1166
1167         /* clear any previous set[ug]id data from a previous binary */
1168         bprm->cred->euid = current_euid();
1169         bprm->cred->egid = current_egid();
1170
1171         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1172                 /* Set-uid? */
1173                 if (mode & S_ISUID) {
1174                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1175                         bprm->cred->euid = inode->i_uid;
1176                 }
1177
1178                 /* Set-gid? */
1179                 /*
1180                  * If setgid is set but no group execute bit then this
1181                  * is a candidate for mandatory locking, not a setgid
1182                  * executable.
1183                  */
1184                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1185                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1186                         bprm->cred->egid = inode->i_gid;
1187                 }
1188         }
1189
1190         /* fill in binprm security blob */
1191         retval = security_bprm_set_creds(bprm);
1192         if (retval)
1193                 return retval;
1194         bprm->cred_prepared = 1;
1195
1196         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1197         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1198 }
1199
1200 EXPORT_SYMBOL(prepare_binprm);
1201
1202 /*
1203  * Arguments are '\0' separated strings found at the location bprm->p
1204  * points to; chop off the first by relocating brpm->p to right after
1205  * the first '\0' encountered.
1206  */
1207 int remove_arg_zero(struct linux_binprm *bprm)
1208 {
1209         int ret = 0;
1210         unsigned long offset;
1211         char *kaddr;
1212         struct page *page;
1213
1214         if (!bprm->argc)
1215                 return 0;
1216
1217         do {
1218                 offset = bprm->p & ~PAGE_MASK;
1219                 page = get_arg_page(bprm, bprm->p, 0);
1220                 if (!page) {
1221                         ret = -EFAULT;
1222                         goto out;
1223                 }
1224                 kaddr = kmap_atomic(page, KM_USER0);
1225
1226                 for (; offset < PAGE_SIZE && kaddr[offset];
1227                                 offset++, bprm->p++)
1228                         ;
1229
1230                 kunmap_atomic(kaddr, KM_USER0);
1231                 put_arg_page(page);
1232
1233                 if (offset == PAGE_SIZE)
1234                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1235         } while (offset == PAGE_SIZE);
1236
1237         bprm->p++;
1238         bprm->argc--;
1239         ret = 0;
1240
1241 out:
1242         return ret;
1243 }
1244 EXPORT_SYMBOL(remove_arg_zero);
1245
1246 /*
1247  * cycle the list of binary formats handler, until one recognizes the image
1248  */
1249 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1250 {
1251         unsigned int depth = bprm->recursion_depth;
1252         int try,retval;
1253         struct linux_binfmt *fmt;
1254
1255         retval = security_bprm_check(bprm);
1256         if (retval)
1257                 return retval;
1258
1259         /* kernel module loader fixup */
1260         /* so we don't try to load run modprobe in kernel space. */
1261         set_fs(USER_DS);
1262
1263         retval = audit_bprm(bprm);
1264         if (retval)
1265                 return retval;
1266
1267         retval = -ENOENT;
1268         for (try=0; try<2; try++) {
1269                 read_lock(&binfmt_lock);
1270                 list_for_each_entry(fmt, &formats, lh) {
1271                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1272                         if (!fn)
1273                                 continue;
1274                         if (!try_module_get(fmt->module))
1275                                 continue;
1276                         read_unlock(&binfmt_lock);
1277                         retval = fn(bprm, regs);
1278                         /*
1279                          * Restore the depth counter to its starting value
1280                          * in this call, so we don't have to rely on every
1281                          * load_binary function to restore it on return.
1282                          */
1283                         bprm->recursion_depth = depth;
1284                         if (retval >= 0) {
1285                                 if (depth == 0)
1286                                         tracehook_report_exec(fmt, bprm, regs);
1287                                 put_binfmt(fmt);
1288                                 allow_write_access(bprm->file);
1289                                 if (bprm->file)
1290                                         fput(bprm->file);
1291                                 bprm->file = NULL;
1292                                 current->did_exec = 1;
1293                                 proc_exec_connector(current);
1294                                 return retval;
1295                         }
1296                         read_lock(&binfmt_lock);
1297                         put_binfmt(fmt);
1298                         if (retval != -ENOEXEC || bprm->mm == NULL)
1299                                 break;
1300                         if (!bprm->file) {
1301                                 read_unlock(&binfmt_lock);
1302                                 return retval;
1303                         }
1304                 }
1305                 read_unlock(&binfmt_lock);
1306                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1307                         break;
1308 #ifdef CONFIG_MODULES
1309                 } else {
1310 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1311                         if (printable(bprm->buf[0]) &&
1312                             printable(bprm->buf[1]) &&
1313                             printable(bprm->buf[2]) &&
1314                             printable(bprm->buf[3]))
1315                                 break; /* -ENOEXEC */
1316                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1317 #endif
1318                 }
1319         }
1320         return retval;
1321 }
1322
1323 EXPORT_SYMBOL(search_binary_handler);
1324
1325 /*
1326  * sys_execve() executes a new program.
1327  */
1328 int do_execve(char * filename,
1329         char __user *__user *argv,
1330         char __user *__user *envp,
1331         struct pt_regs * regs)
1332 {
1333         struct linux_binprm *bprm;
1334         struct file *file;
1335         struct files_struct *displaced;
1336         bool clear_in_exec;
1337         int retval;
1338
1339         retval = unshare_files(&displaced);
1340         if (retval)
1341                 goto out_ret;
1342
1343         retval = -ENOMEM;
1344         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1345         if (!bprm)
1346                 goto out_files;
1347
1348         retval = prepare_bprm_creds(bprm);
1349         if (retval)
1350                 goto out_free;
1351
1352         retval = check_unsafe_exec(bprm);
1353         if (retval < 0)
1354                 goto out_free;
1355         clear_in_exec = retval;
1356         current->in_execve = 1;
1357
1358         file = open_exec(filename);
1359         retval = PTR_ERR(file);
1360         if (IS_ERR(file))
1361                 goto out_unmark;
1362
1363         sched_exec();
1364
1365         bprm->file = file;
1366         bprm->filename = filename;
1367         bprm->interp = filename;
1368
1369         retval = bprm_mm_init(bprm);
1370         if (retval)
1371                 goto out_file;
1372
1373         bprm->argc = count(argv, MAX_ARG_STRINGS);
1374         if ((retval = bprm->argc) < 0)
1375                 goto out;
1376
1377         bprm->envc = count(envp, MAX_ARG_STRINGS);
1378         if ((retval = bprm->envc) < 0)
1379                 goto out;
1380
1381         retval = prepare_binprm(bprm);
1382         if (retval < 0)
1383                 goto out;
1384
1385         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1386         if (retval < 0)
1387                 goto out;
1388
1389         bprm->exec = bprm->p;
1390         retval = copy_strings(bprm->envc, envp, bprm);
1391         if (retval < 0)
1392                 goto out;
1393
1394         retval = copy_strings(bprm->argc, argv, bprm);
1395         if (retval < 0)
1396                 goto out;
1397
1398         current->flags &= ~PF_KTHREAD;
1399         retval = search_binary_handler(bprm,regs);
1400         if (retval < 0)
1401                 goto out;
1402
1403         /* execve succeeded */
1404         current->fs->in_exec = 0;
1405         current->in_execve = 0;
1406         acct_update_integrals(current);
1407         free_bprm(bprm);
1408         if (displaced)
1409                 put_files_struct(displaced);
1410         return retval;
1411
1412 out:
1413         if (bprm->mm)
1414                 mmput (bprm->mm);
1415
1416 out_file:
1417         if (bprm->file) {
1418                 allow_write_access(bprm->file);
1419                 fput(bprm->file);
1420         }
1421
1422 out_unmark:
1423         if (clear_in_exec)
1424                 current->fs->in_exec = 0;
1425         current->in_execve = 0;
1426
1427 out_free:
1428         free_bprm(bprm);
1429
1430 out_files:
1431         if (displaced)
1432                 reset_files_struct(displaced);
1433 out_ret:
1434         return retval;
1435 }
1436
1437 void set_binfmt(struct linux_binfmt *new)
1438 {
1439         struct mm_struct *mm = current->mm;
1440
1441         if (mm->binfmt)
1442                 module_put(mm->binfmt->module);
1443
1444         mm->binfmt = new;
1445         if (new)
1446                 __module_get(new->module);
1447 }
1448
1449 EXPORT_SYMBOL(set_binfmt);
1450
1451 /* format_corename will inspect the pattern parameter, and output a
1452  * name into corename, which must have space for at least
1453  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1454  */
1455 static int format_corename(char *corename, long signr)
1456 {
1457         const struct cred *cred = current_cred();
1458         const char *pat_ptr = core_pattern;
1459         int ispipe = (*pat_ptr == '|');
1460         char *out_ptr = corename;
1461         char *const out_end = corename + CORENAME_MAX_SIZE;
1462         int rc;
1463         int pid_in_pattern = 0;
1464
1465         /* Repeat as long as we have more pattern to process and more output
1466            space */
1467         while (*pat_ptr) {
1468                 if (*pat_ptr != '%') {
1469                         if (out_ptr == out_end)
1470                                 goto out;
1471                         *out_ptr++ = *pat_ptr++;
1472                 } else {
1473                         switch (*++pat_ptr) {
1474                         case 0:
1475                                 goto out;
1476                         /* Double percent, output one percent */
1477                         case '%':
1478                                 if (out_ptr == out_end)
1479                                         goto out;
1480                                 *out_ptr++ = '%';
1481                                 break;
1482                         /* pid */
1483                         case 'p':
1484                                 pid_in_pattern = 1;
1485                                 rc = snprintf(out_ptr, out_end - out_ptr,
1486                                               "%d", task_tgid_vnr(current));
1487                                 if (rc > out_end - out_ptr)
1488                                         goto out;
1489                                 out_ptr += rc;
1490                                 break;
1491                         /* uid */
1492                         case 'u':
1493                                 rc = snprintf(out_ptr, out_end - out_ptr,
1494                                               "%d", cred->uid);
1495                                 if (rc > out_end - out_ptr)
1496                                         goto out;
1497                                 out_ptr += rc;
1498                                 break;
1499                         /* gid */
1500                         case 'g':
1501                                 rc = snprintf(out_ptr, out_end - out_ptr,
1502                                               "%d", cred->gid);
1503                                 if (rc > out_end - out_ptr)
1504                                         goto out;
1505                                 out_ptr += rc;
1506                                 break;
1507                         /* signal that caused the coredump */
1508                         case 's':
1509                                 rc = snprintf(out_ptr, out_end - out_ptr,
1510                                               "%ld", signr);
1511                                 if (rc > out_end - out_ptr)
1512                                         goto out;
1513                                 out_ptr += rc;
1514                                 break;
1515                         /* UNIX time of coredump */
1516                         case 't': {
1517                                 struct timeval tv;
1518                                 do_gettimeofday(&tv);
1519                                 rc = snprintf(out_ptr, out_end - out_ptr,
1520                                               "%lu", tv.tv_sec);
1521                                 if (rc > out_end - out_ptr)
1522                                         goto out;
1523                                 out_ptr += rc;
1524                                 break;
1525                         }
1526                         /* hostname */
1527                         case 'h':
1528                                 down_read(&uts_sem);
1529                                 rc = snprintf(out_ptr, out_end - out_ptr,
1530                                               "%s", utsname()->nodename);
1531                                 up_read(&uts_sem);
1532                                 if (rc > out_end - out_ptr)
1533                                         goto out;
1534                                 out_ptr += rc;
1535                                 break;
1536                         /* executable */
1537                         case 'e':
1538                                 rc = snprintf(out_ptr, out_end - out_ptr,
1539                                               "%s", current->comm);
1540                                 if (rc > out_end - out_ptr)
1541                                         goto out;
1542                                 out_ptr += rc;
1543                                 break;
1544                         /* core limit size */
1545                         case 'c':
1546                                 rc = snprintf(out_ptr, out_end - out_ptr,
1547                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1548                                 if (rc > out_end - out_ptr)
1549                                         goto out;
1550                                 out_ptr += rc;
1551                                 break;
1552                         default:
1553                                 break;
1554                         }
1555                         ++pat_ptr;
1556                 }
1557         }
1558         /* Backward compatibility with core_uses_pid:
1559          *
1560          * If core_pattern does not include a %p (as is the default)
1561          * and core_uses_pid is set, then .%pid will be appended to
1562          * the filename. Do not do this for piped commands. */
1563         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1564                 rc = snprintf(out_ptr, out_end - out_ptr,
1565                               ".%d", task_tgid_vnr(current));
1566                 if (rc > out_end - out_ptr)
1567                         goto out;
1568                 out_ptr += rc;
1569         }
1570 out:
1571         *out_ptr = 0;
1572         return ispipe;
1573 }
1574
1575 static int zap_process(struct task_struct *start)
1576 {
1577         struct task_struct *t;
1578         int nr = 0;
1579
1580         start->signal->flags = SIGNAL_GROUP_EXIT;
1581         start->signal->group_stop_count = 0;
1582
1583         t = start;
1584         do {
1585                 if (t != current && t->mm) {
1586                         sigaddset(&t->pending.signal, SIGKILL);
1587                         signal_wake_up(t, 1);
1588                         nr++;
1589                 }
1590         } while_each_thread(start, t);
1591
1592         return nr;
1593 }
1594
1595 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1596                                 struct core_state *core_state, int exit_code)
1597 {
1598         struct task_struct *g, *p;
1599         unsigned long flags;
1600         int nr = -EAGAIN;
1601
1602         spin_lock_irq(&tsk->sighand->siglock);
1603         if (!signal_group_exit(tsk->signal)) {
1604                 mm->core_state = core_state;
1605                 tsk->signal->group_exit_code = exit_code;
1606                 nr = zap_process(tsk);
1607         }
1608         spin_unlock_irq(&tsk->sighand->siglock);
1609         if (unlikely(nr < 0))
1610                 return nr;
1611
1612         if (atomic_read(&mm->mm_users) == nr + 1)
1613                 goto done;
1614         /*
1615          * We should find and kill all tasks which use this mm, and we should
1616          * count them correctly into ->nr_threads. We don't take tasklist
1617          * lock, but this is safe wrt:
1618          *
1619          * fork:
1620          *      None of sub-threads can fork after zap_process(leader). All
1621          *      processes which were created before this point should be
1622          *      visible to zap_threads() because copy_process() adds the new
1623          *      process to the tail of init_task.tasks list, and lock/unlock
1624          *      of ->siglock provides a memory barrier.
1625          *
1626          * do_exit:
1627          *      The caller holds mm->mmap_sem. This means that the task which
1628          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1629          *      its ->mm.
1630          *
1631          * de_thread:
1632          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1633          *      we must see either old or new leader, this does not matter.
1634          *      However, it can change p->sighand, so lock_task_sighand(p)
1635          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1636          *      it can't fail.
1637          *
1638          *      Note also that "g" can be the old leader with ->mm == NULL
1639          *      and already unhashed and thus removed from ->thread_group.
1640          *      This is OK, __unhash_process()->list_del_rcu() does not
1641          *      clear the ->next pointer, we will find the new leader via
1642          *      next_thread().
1643          */
1644         rcu_read_lock();
1645         for_each_process(g) {
1646                 if (g == tsk->group_leader)
1647                         continue;
1648                 if (g->flags & PF_KTHREAD)
1649                         continue;
1650                 p = g;
1651                 do {
1652                         if (p->mm) {
1653                                 if (unlikely(p->mm == mm)) {
1654                                         lock_task_sighand(p, &flags);
1655                                         nr += zap_process(p);
1656                                         unlock_task_sighand(p, &flags);
1657                                 }
1658                                 break;
1659                         }
1660                 } while_each_thread(g, p);
1661         }
1662         rcu_read_unlock();
1663 done:
1664         atomic_set(&core_state->nr_threads, nr);
1665         return nr;
1666 }
1667
1668 static int coredump_wait(int exit_code, struct core_state *core_state)
1669 {
1670         struct task_struct *tsk = current;
1671         struct mm_struct *mm = tsk->mm;
1672         struct completion *vfork_done;
1673         int core_waiters;
1674
1675         init_completion(&core_state->startup);
1676         core_state->dumper.task = tsk;
1677         core_state->dumper.next = NULL;
1678         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1679         up_write(&mm->mmap_sem);
1680
1681         if (unlikely(core_waiters < 0))
1682                 goto fail;
1683
1684         /*
1685          * Make sure nobody is waiting for us to release the VM,
1686          * otherwise we can deadlock when we wait on each other
1687          */
1688         vfork_done = tsk->vfork_done;
1689         if (vfork_done) {
1690                 tsk->vfork_done = NULL;
1691                 complete(vfork_done);
1692         }
1693
1694         if (core_waiters)
1695                 wait_for_completion(&core_state->startup);
1696 fail:
1697         return core_waiters;
1698 }
1699
1700 static void coredump_finish(struct mm_struct *mm)
1701 {
1702         struct core_thread *curr, *next;
1703         struct task_struct *task;
1704
1705         next = mm->core_state->dumper.next;
1706         while ((curr = next) != NULL) {
1707                 next = curr->next;
1708                 task = curr->task;
1709                 /*
1710                  * see exit_mm(), curr->task must not see
1711                  * ->task == NULL before we read ->next.
1712                  */
1713                 smp_mb();
1714                 curr->task = NULL;
1715                 wake_up_process(task);
1716         }
1717
1718         mm->core_state = NULL;
1719 }
1720
1721 /*
1722  * set_dumpable converts traditional three-value dumpable to two flags and
1723  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1724  * these bits are not changed atomically.  So get_dumpable can observe the
1725  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1726  * return either old dumpable or new one by paying attention to the order of
1727  * modifying the bits.
1728  *
1729  * dumpable |   mm->flags (binary)
1730  * old  new | initial interim  final
1731  * ---------+-----------------------
1732  *  0    1  |   00      01      01
1733  *  0    2  |   00      10(*)   11
1734  *  1    0  |   01      00      00
1735  *  1    2  |   01      11      11
1736  *  2    0  |   11      10(*)   00
1737  *  2    1  |   11      11      01
1738  *
1739  * (*) get_dumpable regards interim value of 10 as 11.
1740  */
1741 void set_dumpable(struct mm_struct *mm, int value)
1742 {
1743         switch (value) {
1744         case 0:
1745                 clear_bit(MMF_DUMPABLE, &mm->flags);
1746                 smp_wmb();
1747                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1748                 break;
1749         case 1:
1750                 set_bit(MMF_DUMPABLE, &mm->flags);
1751                 smp_wmb();
1752                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1753                 break;
1754         case 2:
1755                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1756                 smp_wmb();
1757                 set_bit(MMF_DUMPABLE, &mm->flags);
1758                 break;
1759         }
1760 }
1761
1762 int get_dumpable(struct mm_struct *mm)
1763 {
1764         int ret;
1765
1766         ret = mm->flags & 0x3;
1767         return (ret >= 2) ? 2 : ret;
1768 }
1769
1770 static void wait_for_dump_helpers(struct file *file)
1771 {
1772         struct pipe_inode_info *pipe;
1773
1774         pipe = file->f_path.dentry->d_inode->i_pipe;
1775
1776         pipe_lock(pipe);
1777         pipe->readers++;
1778         pipe->writers--;
1779
1780         while ((pipe->readers > 1) && (!signal_pending(current))) {
1781                 wake_up_interruptible_sync(&pipe->wait);
1782                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1783                 pipe_wait(pipe);
1784         }
1785
1786         pipe->readers--;
1787         pipe->writers++;
1788         pipe_unlock(pipe);
1789
1790 }
1791
1792
1793 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1794 {
1795         struct core_state core_state;
1796         char corename[CORENAME_MAX_SIZE + 1];
1797         struct mm_struct *mm = current->mm;
1798         struct linux_binfmt * binfmt;
1799         struct inode * inode;
1800         const struct cred *old_cred;
1801         struct cred *cred;
1802         int retval = 0;
1803         int flag = 0;
1804         int ispipe = 0;
1805         char **helper_argv = NULL;
1806         int helper_argc = 0;
1807         int dump_count = 0;
1808         static atomic_t core_dump_count = ATOMIC_INIT(0);
1809         struct coredump_params cprm = {
1810                 .signr = signr,
1811                 .regs = regs,
1812                 .limit = current->signal->rlim[RLIMIT_CORE].rlim_cur,
1813         };
1814
1815         audit_core_dumps(signr);
1816
1817         binfmt = mm->binfmt;
1818         if (!binfmt || !binfmt->core_dump)
1819                 goto fail;
1820
1821         cred = prepare_creds();
1822         if (!cred) {
1823                 retval = -ENOMEM;
1824                 goto fail;
1825         }
1826
1827         down_write(&mm->mmap_sem);
1828         /*
1829          * If another thread got here first, or we are not dumpable, bail out.
1830          */
1831         if (mm->core_state || !get_dumpable(mm)) {
1832                 up_write(&mm->mmap_sem);
1833                 put_cred(cred);
1834                 goto fail;
1835         }
1836
1837         /*
1838          *      We cannot trust fsuid as being the "true" uid of the
1839          *      process nor do we know its entire history. We only know it
1840          *      was tainted so we dump it as root in mode 2.
1841          */
1842         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1843                 flag = O_EXCL;          /* Stop rewrite attacks */
1844                 cred->fsuid = 0;        /* Dump root private */
1845         }
1846
1847         retval = coredump_wait(exit_code, &core_state);
1848         if (retval < 0) {
1849                 put_cred(cred);
1850                 goto fail;
1851         }
1852
1853         old_cred = override_creds(cred);
1854
1855         /*
1856          * Clear any false indication of pending signals that might
1857          * be seen by the filesystem code called to write the core file.
1858          */
1859         clear_thread_flag(TIF_SIGPENDING);
1860
1861         /*
1862          * lock_kernel() because format_corename() is controlled by sysctl, which
1863          * uses lock_kernel()
1864          */
1865         lock_kernel();
1866         ispipe = format_corename(corename, signr);
1867         unlock_kernel();
1868
1869         if ((!ispipe) && (cprm.limit < binfmt->min_coredump))
1870                 goto fail_unlock;
1871
1872         if (ispipe) {
1873                 if (cprm.limit == 0) {
1874                         /*
1875                          * Normally core limits are irrelevant to pipes, since
1876                          * we're not writing to the file system, but we use
1877                          * cprm.limit of 0 here as a speacial value. Any
1878                          * non-zero limit gets set to RLIM_INFINITY below, but
1879                          * a limit of 0 skips the dump.  This is a consistent
1880                          * way to catch recursive crashes.  We can still crash
1881                          * if the core_pattern binary sets RLIM_CORE =  !0
1882                          * but it runs as root, and can do lots of stupid things
1883                          * Note that we use task_tgid_vnr here to grab the pid
1884                          * of the process group leader.  That way we get the
1885                          * right pid if a thread in a multi-threaded
1886                          * core_pattern process dies.
1887                          */
1888                         printk(KERN_WARNING
1889                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1890                                 task_tgid_vnr(current), current->comm);
1891                         printk(KERN_WARNING "Aborting core\n");
1892                         goto fail_unlock;
1893                 }
1894
1895                 dump_count = atomic_inc_return(&core_dump_count);
1896                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1897                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1898                                task_tgid_vnr(current), current->comm);
1899                         printk(KERN_WARNING "Skipping core dump\n");
1900                         goto fail_dropcount;
1901                 }
1902
1903                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1904                 if (!helper_argv) {
1905                         printk(KERN_WARNING "%s failed to allocate memory\n",
1906                                __func__);
1907                         goto fail_dropcount;
1908                 }
1909
1910                 cprm.limit = RLIM_INFINITY;
1911
1912                 /* SIGPIPE can happen, but it's just never processed */
1913                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1914                                 &cprm.file)) {
1915                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1916                                corename);
1917                         goto fail_dropcount;
1918                 }
1919         } else
1920                 cprm.file = filp_open(corename,
1921                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1922                                  0600);
1923         if (IS_ERR(cprm.file))
1924                 goto fail_dropcount;
1925         inode = cprm.file->f_path.dentry->d_inode;
1926         if (inode->i_nlink > 1)
1927                 goto close_fail;        /* multiple links - don't dump */
1928         if (!ispipe && d_unhashed(cprm.file->f_path.dentry))
1929                 goto close_fail;
1930
1931         /* AK: actually i see no reason to not allow this for named pipes etc.,
1932            but keep the previous behaviour for now. */
1933         if (!ispipe && !S_ISREG(inode->i_mode))
1934                 goto close_fail;
1935         /*
1936          * Dont allow local users get cute and trick others to coredump
1937          * into their pre-created files:
1938          * Note, this is not relevant for pipes
1939          */
1940         if (!ispipe && (inode->i_uid != current_fsuid()))
1941                 goto close_fail;
1942         if (!cprm.file->f_op)
1943                 goto close_fail;
1944         if (!cprm.file->f_op->write)
1945                 goto close_fail;
1946         if (!ispipe &&
1947             do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0)
1948                 goto close_fail;
1949
1950         retval = binfmt->core_dump(&cprm);
1951
1952         if (retval)
1953                 current->signal->group_exit_code |= 0x80;
1954 close_fail:
1955         if (ispipe && core_pipe_limit)
1956                 wait_for_dump_helpers(cprm.file);
1957         filp_close(cprm.file, NULL);
1958 fail_dropcount:
1959         if (dump_count)
1960                 atomic_dec(&core_dump_count);
1961 fail_unlock:
1962         if (helper_argv)
1963                 argv_free(helper_argv);
1964
1965         revert_creds(old_cred);
1966         put_cred(cred);
1967         coredump_finish(mm);
1968 fail:
1969         return;
1970 }