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/ima.h>
  50 #include <linux/syscalls.h>
  51 #include <linux/tsacct_kern.h>
  52 #include <linux/cn_proc.h>
  53 #include <linux/audit.h>
  54 #include <linux/tracehook.h>
  55 #include <linux/kmod.h>
  56 #include <linux/fsnotify.h>
  57 #include <linux/fs_struct.h>
  58 #include <linux/pipe_fs_i.h>
  59
  60 #include <asm/uaccess.h>
  61 #include <asm/mmu_context.h>
  62 #include <asm/tlb.h>
  63 #include "internal.h"
  64
  65 int core_uses_pid;
  66 char core_pattern[CORENAME_MAX_SIZE] = "core";
  67 unsigned int core_pipe_limit;
  68 int suid_dumpable = 0;
  69
  70 /* The maximal length of core_pattern is also specified in sysctl.c */
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  76 {
  77         if (!fmt)
  78                 return -EINVAL;
  79         write_lock(&binfmt_lock);
  80         insert ? list_add(&fmt->lh, &formats) :
  81                  list_add_tail(&fmt->lh, &formats);
  82         write_unlock(&binfmt_lock);
  83         return 0;
  84 }
  85
  86 EXPORT_SYMBOL(__register_binfmt);
  87
  88 void unregister_binfmt(struct linux_binfmt * fmt)
  89 {
  90         write_lock(&binfmt_lock);
  91         list_del(&fmt->lh);
  92         write_unlock(&binfmt_lock);
  93 }
  94
  95 EXPORT_SYMBOL(unregister_binfmt);
  96
  97 static inline void put_binfmt(struct linux_binfmt * fmt)
  98 {
  99         module_put(fmt->module);
 100 }
 101
 102 /*
 103  * Note that a shared library must be both readable and executable due to
 104  * security reasons.
 105  *
 106  * Also note that we take the address to load from from the file itself.
 107  */
 108 SYSCALL_DEFINE1(uselib, const char __user *, library)
 109 {
 110         struct file *file;
 111         char *tmp = getname(library);
 112         int error = PTR_ERR(tmp);
 113
 114         if (IS_ERR(tmp))
 115                 goto out;
 116
 117         file = do_filp_open(AT_FDCWD, tmp,
 118                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 119                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 120         putname(tmp);
 121         error = PTR_ERR(file);
 122         if (IS_ERR(file))
 123                 goto out;
 124
 125         error = -EINVAL;
 126         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 127                 goto exit;
 128
 129         error = -EACCES;
 130         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 131                 goto exit;
 132
 133         fsnotify_open(file->f_path.dentry);
 134
 135         error = -ENOEXEC;
 136         if(file->f_op) {
 137                 struct linux_binfmt * fmt;
 138
 139                 read_lock(&binfmt_lock);
 140                 list_for_each_entry(fmt, &formats, lh) {
 141                         if (!fmt->load_shlib)
 142                                 continue;
 143                         if (!try_module_get(fmt->module))
 144                                 continue;
 145                         read_unlock(&binfmt_lock);
 146                         error = fmt->load_shlib(file);
 147                         read_lock(&binfmt_lock);
 148                         put_binfmt(fmt);
 149                         if (error != -ENOEXEC)
 150                                 break;
 151                 }
 152                 read_unlock(&binfmt_lock);
 153         }
 154 exit:
 155         fput(file);
 156 out:
 157         return error;
 158 }
 159
 160 #ifdef CONFIG_MMU
 161
 162 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 163                 int write)
 164 {
 165         struct page *page;
 166         int ret;
 167
 168 #ifdef CONFIG_STACK_GROWSUP
 169         if (write) {
 170                 ret = expand_stack_downwards(bprm->vma, pos);
 171                 if (ret < 0)
 172                         return NULL;
 173         }
 174 #endif
 175         ret = get_user_pages(current, bprm->mm, pos,
 176                         1, write, 1, &page, NULL);
 177         if (ret <= 0)
 178                 return NULL;
 179
 180         if (write) {
 181                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 182                 struct rlimit *rlim;
 183
 184                 /*
 185                  * We've historically supported up to 32 pages (ARG_MAX)
 186                  * of argument strings even with small stacks
 187                  */
 188                 if (size <= ARG_MAX)
 189                         return page;
 190
 191                 /*
 192                  * Limit to 1/4-th the stack size for the argv+env strings.
 193                  * This ensures that:
 194                  *  - the remaining binfmt code will not run out of stack space,
 195                  *  - the program will have a reasonable amount of stack left
 196                  *    to work from.
 197                  */
 198                 rlim = current->signal->rlim;
 199                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 200                         put_page(page);
 201                         return NULL;
 202                 }
 203         }
 204
 205         return page;
 206 }
 207
 208 static void put_arg_page(struct page *page)
 209 {
 210         put_page(page);
 211 }
 212
 213 static void free_arg_page(struct linux_binprm *bprm, int i)
 214 {
 215 }
 216
 217 static void free_arg_pages(struct linux_binprm *bprm)
 218 {
 219 }
 220
 221 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 222                 struct page *page)
 223 {
 224         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 225 }
 226
 227 static int __bprm_mm_init(struct linux_binprm *bprm)
 228 {
 229         int err;
 230         struct vm_area_struct *vma = NULL;
 231         struct mm_struct *mm = bprm->mm;
 232
 233         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 234         if (!vma)
 235                 return -ENOMEM;
 236
 237         down_write(&mm->mmap_sem);
 238         vma->vm_mm = mm;
 239
 240         /*
 241          * Place the stack at the largest stack address the architecture
 242          * supports. Later, we'll move this to an appropriate place. We don't
 243          * use STACK_TOP because that can depend on attributes which aren't
 244          * configured yet.
 245          */
 246         vma->vm_end = STACK_TOP_MAX;
 247         vma->vm_start = vma->vm_end - PAGE_SIZE;
 248         vma->vm_flags = VM_STACK_FLAGS;
 249         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 250         err = insert_vm_struct(mm, vma);
 251         if (err)
 252                 goto err;
 253
 254         mm->stack_vm = mm->total_vm = 1;
 255         up_write(&mm->mmap_sem);
 256         bprm->p = vma->vm_end - sizeof(void *);
 257         return 0;
 258 err:
 259         up_write(&mm->mmap_sem);
 260         bprm->vma = NULL;
 261         kmem_cache_free(vm_area_cachep, vma);
 262         return err;
 263 }
 264
 265 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 266 {
 267         return len <= MAX_ARG_STRLEN;
 268 }
 269
 270 #else
 271
 272 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 273                 int write)
 274 {
 275         struct page *page;
 276
 277         page = bprm->page[pos / PAGE_SIZE];
 278         if (!page && write) {
 279                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 280                 if (!page)
 281                         return NULL;
 282                 bprm->page[pos / PAGE_SIZE] = page;
 283         }
 284
 285         return page;
 286 }
 287
 288 static void put_arg_page(struct page *page)
 289 {
 290 }
 291
 292 static void free_arg_page(struct linux_binprm *bprm, int i)
 293 {
 294         if (bprm->page[i]) {
 295                 __free_page(bprm->page[i]);
 296                 bprm->page[i] = NULL;
 297         }
 298 }
 299
 300 static void free_arg_pages(struct linux_binprm *bprm)
 301 {
 302         int i;
 303
 304         for (i = 0; i < MAX_ARG_PAGES; i++)
 305                 free_arg_page(bprm, i);
 306 }
 307
 308 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 309                 struct page *page)
 310 {
 311 }
 312
 313 static int __bprm_mm_init(struct linux_binprm *bprm)
 314 {
 315         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 316         return 0;
 317 }
 318
 319 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 320 {
 321         return len <= bprm->p;
 322 }
 323
 324 #endif /* CONFIG_MMU */
 325
 326 /*
 327  * Create a new mm_struct and populate it with a temporary stack
 328  * vm_area_struct.  We don't have enough context at this point to set the stack
 329  * flags, permissions, and offset, so we use temporary values.  We'll update
 330  * them later in setup_arg_pages().
 331  */
 332 int bprm_mm_init(struct linux_binprm *bprm)
 333 {
 334         int err;
 335         struct mm_struct *mm = NULL;
 336
 337         bprm->mm = mm = mm_alloc();
 338         err = -ENOMEM;
 339         if (!mm)
 340                 goto err;
 341
 342         err = init_new_context(current, mm);
 343         if (err)
 344                 goto err;
 345
 346         err = __bprm_mm_init(bprm);
 347         if (err)
 348                 goto err;
 349
 350         return 0;
 351
 352 err:
 353         if (mm) {
 354                 bprm->mm = NULL;
 355                 mmdrop(mm);
 356         }
 357
 358         return err;
 359 }
 360
 361 /*
 362  * count() counts the number of strings in array ARGV.
 363  */
 364 static int count(char __user * __user * argv, int max)
 365 {
 366         int i = 0;
 367
 368         if (argv != NULL) {
 369                 for (;;) {
 370                         char __user * p;
 371
 372                         if (get_user(p, argv))
 373                                 return -EFAULT;
 374                         if (!p)
 375                                 break;
 376                         argv++;
 377                         if (i++ >= max)
 378                                 return -E2BIG;
 379                         cond_resched();
 380                 }
 381         }
 382         return i;
 383 }
 384
 385 /*
 386  * 'copy_strings()' copies argument/environment strings from the old
 387  * processes's memory to the new process's stack.  The call to get_user_pages()
 388  * ensures the destination page is created and not swapped out.
 389  */
 390 static int copy_strings(int argc, char __user * __user * argv,
 391                         struct linux_binprm *bprm)
 392 {
 393         struct page *kmapped_page = NULL;
 394         char *kaddr = NULL;
 395         unsigned long kpos = 0;
 396         int ret;
 397
 398         while (argc-- > 0) {
 399                 char __user *str;
 400                 int len;
 401                 unsigned long pos;
 402
 403                 if (get_user(str, argv+argc) ||
 404                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 405                         ret = -EFAULT;
 406                         goto out;
 407                 }
 408
 409                 if (!valid_arg_len(bprm, len)) {
 410                         ret = -E2BIG;
 411                         goto out;
 412                 }
 413
 414                 /* We're going to work our way backwords. */
 415                 pos = bprm->p;
 416                 str += len;
 417                 bprm->p -= len;
 418
 419                 while (len > 0) {
 420                         int offset, bytes_to_copy;
 421
 422                         offset = pos % PAGE_SIZE;
 423                         if (offset == 0)
 424                                 offset = PAGE_SIZE;
 425
 426                         bytes_to_copy = offset;
 427                         if (bytes_to_copy > len)
 428                                 bytes_to_copy = len;
 429
 430                         offset -= bytes_to_copy;
 431                         pos -= bytes_to_copy;
 432                         str -= bytes_to_copy;
 433                         len -= bytes_to_copy;
 434
 435                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 436                                 struct page *page;
 437
 438                                 page = get_arg_page(bprm, pos, 1);
 439                                 if (!page) {
 440                                         ret = -E2BIG;
 441                                         goto out;
 442                                 }
 443
 444                                 if (kmapped_page) {
 445                                         flush_kernel_dcache_page(kmapped_page);
 446                                         kunmap(kmapped_page);
 447                                         put_arg_page(kmapped_page);
 448                                 }
 449                                 kmapped_page = page;
 450                                 kaddr = kmap(kmapped_page);
 451                                 kpos = pos & PAGE_MASK;
 452                                 flush_arg_page(bprm, kpos, kmapped_page);
 453                         }
 454                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 455                                 ret = -EFAULT;
 456                                 goto out;
 457                         }
 458                 }
 459         }
 460         ret = 0;
 461 out:
 462         if (kmapped_page) {
 463                 flush_kernel_dcache_page(kmapped_page);
 464                 kunmap(kmapped_page);
 465                 put_arg_page(kmapped_page);
 466         }
 467         return ret;
 468 }
 469
 470 /*
 471  * Like copy_strings, but get argv and its values from kernel memory.
 472  */
 473 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 474 {
 475         int r;
 476         mm_segment_t oldfs = get_fs();
 477         set_fs(KERNEL_DS);
 478         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 479         set_fs(oldfs);
 480         return r;
 481 }
 482 EXPORT_SYMBOL(copy_strings_kernel);
 483
 484 #ifdef CONFIG_MMU
 485
 486 /*
 487  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 488  * the binfmt code determines where the new stack should reside, we shift it to
 489  * its final location.  The process proceeds as follows:
 490  *
 491  * 1) Use shift to calculate the new vma endpoints.
 492  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 493  *    arguments passed to subsequent functions are consistent.
 494  * 3) Move vma's page tables to the new range.
 495  * 4) Free up any cleared pgd range.
 496  * 5) Shrink the vma to cover only the new range.
 497  */
 498 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 499 {
 500         struct mm_struct *mm = vma->vm_mm;
 501         unsigned long old_start = vma->vm_start;
 502         unsigned long old_end = vma->vm_end;
 503         unsigned long length = old_end - old_start;
 504         unsigned long new_start = old_start - shift;
 505         unsigned long new_end = old_end - shift;
 506         struct mmu_gather *tlb;
 507
 508         BUG_ON(new_start > new_end);
 509
 510         /*
 511          * ensure there are no vmas between where we want to go
 512          * and where we are
 513          */
 514         if (vma != find_vma(mm, new_start))
 515                 return -EFAULT;
 516
 517         /*
 518          * cover the whole range: [new_start, old_end)
 519          */
 520         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 521
 522         /*
 523          * move the page tables downwards, on failure we rely on
 524          * process cleanup to remove whatever mess we made.
 525          */
 526         if (length != move_page_tables(vma, old_start,
 527                                        vma, new_start, length))
 528                 return -ENOMEM;
 529
 530         lru_add_drain();
 531         tlb = tlb_gather_mmu(mm, 0);
 532         if (new_end > old_start) {
 533                 /*
 534                  * when the old and new regions overlap clear from new_end.
 535                  */
 536                 free_pgd_range(tlb, new_end, old_end, new_end,
 537                         vma->vm_next ? vma->vm_next->vm_start : 0);
 538         } else {
 539                 /*
 540                  * otherwise, clean from old_start; this is done to not touch
 541                  * the address space in [new_end, old_start) some architectures
 542                  * have constraints on va-space that make this illegal (IA64) -
 543                  * for the others its just a little faster.
 544                  */
 545                 free_pgd_range(tlb, old_start, old_end, new_end,
 546                         vma->vm_next ? vma->vm_next->vm_start : 0);
 547         }
 548         tlb_finish_mmu(tlb, new_end, old_end);
 549
 550         /*
 551          * shrink the vma to just the new range.
 552          */
 553         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 554
 555         return 0;
 556 }
 557
 558 #define EXTRA_STACK_VM_PAGES    20      /* random */
 559
 560 /*
 561  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 562  * the stack is optionally relocated, and some extra space is added.
 563  */
 564 int setup_arg_pages(struct linux_binprm *bprm,
 565                     unsigned long stack_top,
 566                     int executable_stack)
 567 {
 568         unsigned long ret;
 569         unsigned long stack_shift;
 570         struct mm_struct *mm = current->mm;
 571         struct vm_area_struct *vma = bprm->vma;
 572         struct vm_area_struct *prev = NULL;
 573         unsigned long vm_flags;
 574         unsigned long stack_base;
 575
 576 #ifdef CONFIG_STACK_GROWSUP
 577         /* Limit stack size to 1GB */
 578         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 579         if (stack_base > (1 << 30))
 580                 stack_base = 1 << 30;
 581
 582         /* Make sure we didn't let the argument array grow too large. */
 583         if (vma->vm_end - vma->vm_start > stack_base)
 584                 return -ENOMEM;
 585
 586         stack_base = PAGE_ALIGN(stack_top - stack_base);
 587
 588         stack_shift = vma->vm_start - stack_base;
 589         mm->arg_start = bprm->p - stack_shift;
 590         bprm->p = vma->vm_end - stack_shift;
 591 #else
 592         stack_top = arch_align_stack(stack_top);
 593         stack_top = PAGE_ALIGN(stack_top);
 594         stack_shift = vma->vm_end - stack_top;
 595
 596         bprm->p -= stack_shift;
 597         mm->arg_start = bprm->p;
 598 #endif
 599
 600         if (bprm->loader)
 601                 bprm->loader -= stack_shift;
 602         bprm->exec -= stack_shift;
 603
 604         down_write(&mm->mmap_sem);
 605         vm_flags = VM_STACK_FLAGS;
 606
 607         /*
 608          * Adjust stack execute permissions; explicitly enable for
 609          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 610          * (arch default) otherwise.
 611          */
 612         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 613                 vm_flags |= VM_EXEC;
 614         else if (executable_stack == EXSTACK_DISABLE_X)
 615                 vm_flags &= ~VM_EXEC;
 616         vm_flags |= mm->def_flags;
 617
 618         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 619                         vm_flags);
 620         if (ret)
 621                 goto out_unlock;
 622         BUG_ON(prev != vma);
 623
 624         /* Move stack pages down in memory. */
 625         if (stack_shift) {
 626                 ret = shift_arg_pages(vma, stack_shift);
 627                 if (ret)
 628                         goto out_unlock;
 629         }
 630
 631 #ifdef CONFIG_STACK_GROWSUP
 632         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 633 #else
 634         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 635 #endif
 636         ret = expand_stack(vma, stack_base);
 637         if (ret)
 638                 ret = -EFAULT;
 639
 640 out_unlock:
 641         up_write(&mm->mmap_sem);
 642         return ret;
 643 }
 644 EXPORT_SYMBOL(setup_arg_pages);
 645
 646 #endif /* CONFIG_MMU */
 647
 648 struct file *open_exec(const char *name)
 649 {
 650         struct file *file;
 651         int err;
 652
 653         file = do_filp_open(AT_FDCWD, name,
 654                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 655                                 MAY_EXEC | MAY_OPEN);
 656         if (IS_ERR(file))
 657                 goto out;
 658
 659         err = -EACCES;
 660         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 661                 goto exit;
 662
 663         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 664                 goto exit;
 665
 666         fsnotify_open(file->f_path.dentry);
 667
 668         err = deny_write_access(file);
 669         if (err)
 670                 goto exit;
 671
 672 out:
 673         return file;
 674
 675 exit:
 676         fput(file);
 677         return ERR_PTR(err);
 678 }
 679 EXPORT_SYMBOL(open_exec);
 680
 681 int kernel_read(struct file *file, loff_t offset,
 682                 char *addr, unsigned long count)
 683 {
 684         mm_segment_t old_fs;
 685         loff_t pos = offset;
 686         int result;
 687
 688         old_fs = get_fs();
 689         set_fs(get_ds());
 690         /* The cast to a user pointer is valid due to the set_fs() */
 691         result = vfs_read(file, (void __user *)addr, count, &pos);
 692         set_fs(old_fs);
 693         return result;
 694 }
 695
 696 EXPORT_SYMBOL(kernel_read);
 697
 698 static int exec_mmap(struct mm_struct *mm)
 699 {
 700         struct task_struct *tsk;
 701         struct mm_struct * old_mm, *active_mm;
 702
 703         /* Notify parent that we're no longer interested in the old VM */
 704         tsk = current;
 705         old_mm = current->mm;
 706         mm_release(tsk, old_mm);
 707
 708         if (old_mm) {
 709                 /*
 710                  * Make sure that if there is a core dump in progress
 711                  * for the old mm, we get out and die instead of going
 712                  * through with the exec.  We must hold mmap_sem around
 713                  * checking core_state and changing tsk->mm.
 714                  */
 715                 down_read(&old_mm->mmap_sem);
 716                 if (unlikely(old_mm->core_state)) {
 717                         up_read(&old_mm->mmap_sem);
 718                         return -EINTR;
 719                 }
 720         }
 721         task_lock(tsk);
 722         active_mm = tsk->active_mm;
 723         tsk->mm = mm;
 724         tsk->active_mm = mm;
 725         activate_mm(active_mm, mm);
 726         task_unlock(tsk);
 727         arch_pick_mmap_layout(mm);
 728         if (old_mm) {
 729                 up_read(&old_mm->mmap_sem);
 730                 BUG_ON(active_mm != old_mm);
 731                 mm_update_next_owner(old_mm);
 732                 mmput(old_mm);
 733                 return 0;
 734         }
 735         mmdrop(active_mm);
 736         return 0;
 737 }
 738
 739 /*
 740  * This function makes sure the current process has its own signal table,
 741  * so that flush_signal_handlers can later reset the handlers without
 742  * disturbing other processes.  (Other processes might share the signal
 743  * table via the CLONE_SIGHAND option to clone().)
 744  */
 745 static int de_thread(struct task_struct *tsk)
 746 {
 747         struct signal_struct *sig = tsk->signal;
 748         struct sighand_struct *oldsighand = tsk->sighand;
 749         spinlock_t *lock = &oldsighand->siglock;
 750         int count;
 751
 752         if (thread_group_empty(tsk))
 753                 goto no_thread_group;
 754
 755         /*
 756          * Kill all other threads in the thread group.
 757          */
 758         spin_lock_irq(lock);
 759         if (signal_group_exit(sig)) {
 760                 /*
 761                  * Another group action in progress, just
 762                  * return so that the signal is processed.
 763                  */
 764                 spin_unlock_irq(lock);
 765                 return -EAGAIN;
 766         }
 767         sig->group_exit_task = tsk;
 768         zap_other_threads(tsk);
 769
 770         /* Account for the thread group leader hanging around: */
 771         count = thread_group_leader(tsk) ? 1 : 2;
 772         sig->notify_count = count;
 773         while (atomic_read(&sig->count) > count) {
 774                 __set_current_state(TASK_UNINTERRUPTIBLE);
 775                 spin_unlock_irq(lock);
 776                 schedule();
 777                 spin_lock_irq(lock);
 778         }
 779         spin_unlock_irq(lock);
 780
 781         /*
 782          * At this point all other threads have exited, all we have to
 783          * do is to wait for the thread group leader to become inactive,
 784          * and to assume its PID:
 785          */
 786         if (!thread_group_leader(tsk)) {
 787                 struct task_struct *leader = tsk->group_leader;
 788
 789                 sig->notify_count = -1; /* for exit_notify() */
 790                 for (;;) {
 791                         write_lock_irq(&tasklist_lock);
 792                         if (likely(leader->exit_state))
 793                                 break;
 794                         __set_current_state(TASK_UNINTERRUPTIBLE);
 795                         write_unlock_irq(&tasklist_lock);
 796                         schedule();
 797                 }
 798
 799                 /*
 800                  * The only record we have of the real-time age of a
 801                  * process, regardless of execs it's done, is start_time.
 802                  * All the past CPU time is accumulated in signal_struct
 803                  * from sister threads now dead.  But in this non-leader
 804                  * exec, nothing survives from the original leader thread,
 805                  * whose birth marks the true age of this process now.
 806                  * When we take on its identity by switching to its PID, we
 807                  * also take its birthdate (always earlier than our own).
 808                  */
 809                 tsk->start_time = leader->start_time;
 810
 811                 BUG_ON(!same_thread_group(leader, tsk));
 812                 BUG_ON(has_group_leader_pid(tsk));
 813                 /*
 814                  * An exec() starts a new thread group with the
 815                  * TGID of the previous thread group. Rehash the
 816                  * two threads with a switched PID, and release
 817                  * the former thread group leader:
 818                  */
 819
 820                 /* Become a process group leader with the old leader's pid.
 821                  * The old leader becomes a thread of the this thread group.
 822                  * Note: The old leader also uses this pid until release_task
 823                  *       is called.  Odd but simple and correct.
 824                  */
 825                 detach_pid(tsk, PIDTYPE_PID);
 826                 tsk->pid = leader->pid;
 827                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 828                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 829                 transfer_pid(leader, tsk, PIDTYPE_SID);
 830                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 831
 832                 tsk->group_leader = tsk;
 833                 leader->group_leader = tsk;
 834
 835                 tsk->exit_signal = SIGCHLD;
 836
 837                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 838                 leader->exit_state = EXIT_DEAD;
 839                 write_unlock_irq(&tasklist_lock);
 840
 841                 release_task(leader);
 842         }
 843
 844         sig->group_exit_task = NULL;
 845         sig->notify_count = 0;
 846
 847 no_thread_group:
 848         if (current->mm)
 849                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 850
 851         exit_itimers(sig);
 852         flush_itimer_signals();
 853
 854         if (atomic_read(&oldsighand->count) != 1) {
 855                 struct sighand_struct *newsighand;
 856                 /*
 857                  * This ->sighand is shared with the CLONE_SIGHAND
 858                  * but not CLONE_THREAD task, switch to the new one.
 859                  */
 860                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 861                 if (!newsighand)
 862                         return -ENOMEM;
 863
 864                 atomic_set(&newsighand->count, 1);
 865                 memcpy(newsighand->action, oldsighand->action,
 866                        sizeof(newsighand->action));
 867
 868                 write_lock_irq(&tasklist_lock);
 869                 spin_lock(&oldsighand->siglock);
 870                 rcu_assign_pointer(tsk->sighand, newsighand);
 871                 spin_unlock(&oldsighand->siglock);
 872                 write_unlock_irq(&tasklist_lock);
 873
 874                 __cleanup_sighand(oldsighand);
 875         }
 876
 877         BUG_ON(!thread_group_leader(tsk));
 878         return 0;
 879 }
 880
 881 /*
 882  * These functions flushes out all traces of the currently running executable
 883  * so that a new one can be started
 884  */
 885 static void flush_old_files(struct files_struct * files)
 886 {
 887         long j = -1;
 888         struct fdtable *fdt;
 889
 890         spin_lock(&files->file_lock);
 891         for (;;) {
 892                 unsigned long set, i;
 893
 894                 j++;
 895                 i = j * __NFDBITS;
 896                 fdt = files_fdtable(files);
 897                 if (i >= fdt->max_fds)
 898                         break;
 899                 set = fdt->close_on_exec->fds_bits[j];
 900                 if (!set)
 901                         continue;
 902                 fdt->close_on_exec->fds_bits[j] = 0;
 903                 spin_unlock(&files->file_lock);
 904                 for ( ; set ; i++,set >>= 1) {
 905                         if (set & 1) {
 906                                 sys_close(i);
 907                         }
 908                 }
 909                 spin_lock(&files->file_lock);
 910
 911         }
 912         spin_unlock(&files->file_lock);
 913 }
 914
 915 char *get_task_comm(char *buf, struct task_struct *tsk)
 916 {
 917         /* buf must be at least sizeof(tsk->comm) in size */
 918         task_lock(tsk);
 919         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 920         task_unlock(tsk);
 921         return buf;
 922 }
 923
 924 void set_task_comm(struct task_struct *tsk, char *buf)
 925 {
 926         task_lock(tsk);
 927         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 928         task_unlock(tsk);
 929         perf_event_comm(tsk);
 930 }
 931
 932 int flush_old_exec(struct linux_binprm * bprm)
 933 {
 934         int retval;
 935
 936         /*
 937          * Make sure we have a private signal table and that
 938          * we are unassociated from the previous thread group.
 939          */
 940         retval = de_thread(current);
 941         if (retval)
 942                 goto out;
 943
 944         set_mm_exe_file(bprm->mm, bprm->file);
 945
 946         /*
 947          * Release all of the old mmap stuff
 948          */
 949         retval = exec_mmap(bprm->mm);
 950         if (retval)
 951                 goto out;
 952
 953         bprm->mm = NULL;                /* We're using it now */
 954
 955         current->flags &= ~PF_RANDOMIZE;
 956         flush_thread();
 957         current->personality &= ~bprm->per_clear;
 958
 959         return 0;
 960
 961 out:
 962         return retval;
 963 }
 964 EXPORT_SYMBOL(flush_old_exec);
 965
 966 void setup_new_exec(struct linux_binprm * bprm)
 967 {
 968         int i, ch;
 969         char * name;
 970         char tcomm[sizeof(current->comm)];
 971
 972         arch_pick_mmap_layout(current->mm);
 973
 974         /* This is the point of no return */
 975         current->sas_ss_sp = current->sas_ss_size = 0;
 976
 977         if (current_euid() == current_uid() && current_egid() == current_gid())
 978                 set_dumpable(current->mm, 1);
 979         else
 980                 set_dumpable(current->mm, suid_dumpable);
 981
 982         name = bprm->filename;
 983
 984         /* Copies the binary name from after last slash */
 985         for (i=0; (ch = *(name++)) != '\0';) {
 986                 if (ch == '/')
 987                         i = 0; /* overwrite what we wrote */
 988                 else
 989                         if (i < (sizeof(tcomm) - 1))
 990                                 tcomm[i++] = ch;
 991         }
 992         tcomm[i] = '\0';
 993         set_task_comm(current, tcomm);
 994
 995         /* Set the new mm task size. We have to do that late because it may
 996          * depend on TIF_32BIT which is only updated in flush_thread() on
 997          * some architectures like powerpc
 998          */
 999         current->mm->task_size = TASK_SIZE;
1000
1001         /* install the new credentials */
1002         if (bprm->cred->uid != current_euid() ||
1003             bprm->cred->gid != current_egid()) {
1004                 current->pdeath_signal = 0;
1005         } else if (file_permission(bprm->file, MAY_READ) ||
1006                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1007                 set_dumpable(current->mm, suid_dumpable);
1008         }
1009
1010         /*
1011          * Flush performance counters when crossing a
1012          * security domain:
1013          */
1014         if (!get_dumpable(current->mm))
1015                 perf_event_exit_task(current);
1016
1017         /* An exec changes our domain. We are no longer part of the thread
1018            group */
1019
1020         current->self_exec_id++;
1021
1022         flush_signal_handlers(current, 0);
1023         flush_old_files(current->files);
1024 }
1025 EXPORT_SYMBOL(setup_new_exec);
1026
1027 /*
1028  * Prepare credentials and lock ->cred_guard_mutex.
1029  * install_exec_creds() commits the new creds and drops the lock.
1030  * Or, if exec fails before, free_bprm() should release ->cred and
1031  * and unlock.
1032  */
1033 int prepare_bprm_creds(struct linux_binprm *bprm)
1034 {
1035         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1036                 return -ERESTARTNOINTR;
1037
1038         bprm->cred = prepare_exec_creds();
1039         if (likely(bprm->cred))
1040                 return 0;
1041
1042         mutex_unlock(&current->cred_guard_mutex);
1043         return -ENOMEM;
1044 }
1045
1046 void free_bprm(struct linux_binprm *bprm)
1047 {
1048         free_arg_pages(bprm);
1049         if (bprm->cred) {
1050                 mutex_unlock(&current->cred_guard_mutex);
1051                 abort_creds(bprm->cred);
1052         }
1053         kfree(bprm);
1054 }
1055
1056 /*
1057  * install the new credentials for this executable
1058  */
1059 void install_exec_creds(struct linux_binprm *bprm)
1060 {
1061         security_bprm_committing_creds(bprm);
1062
1063         commit_creds(bprm->cred);
1064         bprm->cred = NULL;
1065         /*
1066          * cred_guard_mutex must be held at least to this point to prevent
1067          * ptrace_attach() from altering our determination of the task's
1068          * credentials; any time after this it may be unlocked.
1069          */
1070         security_bprm_committed_creds(bprm);
1071         mutex_unlock(&current->cred_guard_mutex);
1072 }
1073 EXPORT_SYMBOL(install_exec_creds);
1074
1075 /*
1076  * determine how safe it is to execute the proposed program
1077  * - the caller must hold current->cred_guard_mutex to protect against
1078  *   PTRACE_ATTACH
1079  */
1080 int check_unsafe_exec(struct linux_binprm *bprm)
1081 {
1082         struct task_struct *p = current, *t;
1083         unsigned n_fs;
1084         int res = 0;
1085
1086         bprm->unsafe = tracehook_unsafe_exec(p);
1087
1088         n_fs = 1;
1089         write_lock(&p->fs->lock);
1090         rcu_read_lock();
1091         for (t = next_thread(p); t != p; t = next_thread(t)) {
1092                 if (t->fs == p->fs)
1093                         n_fs++;
1094         }
1095         rcu_read_unlock();
1096
1097         if (p->fs->users > n_fs) {
1098                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1099         } else {
1100                 res = -EAGAIN;
1101                 if (!p->fs->in_exec) {
1102                         p->fs->in_exec = 1;
1103                         res = 1;
1104                 }
1105         }
1106         write_unlock(&p->fs->lock);
1107
1108         return res;
1109 }
1110
1111 /*
1112  * Fill the binprm structure from the inode.
1113  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1114  *
1115  * This may be called multiple times for binary chains (scripts for example).
1116  */
1117 int prepare_binprm(struct linux_binprm *bprm)
1118 {
1119         umode_t mode;
1120         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1121         int retval;
1122
1123         mode = inode->i_mode;
1124         if (bprm->file->f_op == NULL)
1125                 return -EACCES;
1126
1127         /* clear any previous set[ug]id data from a previous binary */
1128         bprm->cred->euid = current_euid();
1129         bprm->cred->egid = current_egid();
1130
1131         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1132                 /* Set-uid? */
1133                 if (mode & S_ISUID) {
1134                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1135                         bprm->cred->euid = inode->i_uid;
1136                 }
1137
1138                 /* Set-gid? */
1139                 /*
1140                  * If setgid is set but no group execute bit then this
1141                  * is a candidate for mandatory locking, not a setgid
1142                  * executable.
1143                  */
1144                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1145                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1146                         bprm->cred->egid = inode->i_gid;
1147                 }
1148         }
1149
1150         /* fill in binprm security blob */
1151         retval = security_bprm_set_creds(bprm);
1152         if (retval)
1153                 return retval;
1154         bprm->cred_prepared = 1;
1155
1156         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1157         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1158 }
1159
1160 EXPORT_SYMBOL(prepare_binprm);
1161
1162 /*
1163  * Arguments are '\0' separated strings found at the location bprm->p
1164  * points to; chop off the first by relocating brpm->p to right after
1165  * the first '\0' encountered.
1166  */
1167 int remove_arg_zero(struct linux_binprm *bprm)
1168 {
1169         int ret = 0;
1170         unsigned long offset;
1171         char *kaddr;
1172         struct page *page;
1173
1174         if (!bprm->argc)
1175                 return 0;
1176
1177         do {
1178                 offset = bprm->p & ~PAGE_MASK;
1179                 page = get_arg_page(bprm, bprm->p, 0);
1180                 if (!page) {
1181                         ret = -EFAULT;
1182                         goto out;
1183                 }
1184                 kaddr = kmap_atomic(page, KM_USER0);
1185
1186                 for (; offset < PAGE_SIZE && kaddr[offset];
1187                                 offset++, bprm->p++)
1188                         ;
1189
1190                 kunmap_atomic(kaddr, KM_USER0);
1191                 put_arg_page(page);
1192
1193                 if (offset == PAGE_SIZE)
1194                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1195         } while (offset == PAGE_SIZE);
1196
1197         bprm->p++;
1198         bprm->argc--;
1199         ret = 0;
1200
1201 out:
1202         return ret;
1203 }
1204 EXPORT_SYMBOL(remove_arg_zero);
1205
1206 /*
1207  * cycle the list of binary formats handler, until one recognizes the image
1208  */
1209 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1210 {
1211         unsigned int depth = bprm->recursion_depth;
1212         int try,retval;
1213         struct linux_binfmt *fmt;
1214
1215         retval = security_bprm_check(bprm);
1216         if (retval)
1217                 return retval;
1218         retval = ima_bprm_check(bprm);
1219         if (retval)
1220                 return retval;
1221
1222         /* kernel module loader fixup */
1223         /* so we don't try to load run modprobe in kernel space. */
1224         set_fs(USER_DS);
1225
1226         retval = audit_bprm(bprm);
1227         if (retval)
1228                 return retval;
1229
1230         retval = -ENOENT;
1231         for (try=0; try<2; try++) {
1232                 read_lock(&binfmt_lock);
1233                 list_for_each_entry(fmt, &formats, lh) {
1234                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1235                         if (!fn)
1236                                 continue;
1237                         if (!try_module_get(fmt->module))
1238                                 continue;
1239                         read_unlock(&binfmt_lock);
1240                         retval = fn(bprm, regs);
1241                         /*
1242                          * Restore the depth counter to its starting value
1243                          * in this call, so we don't have to rely on every
1244                          * load_binary function to restore it on return.
1245                          */
1246                         bprm->recursion_depth = depth;
1247                         if (retval >= 0) {
1248                                 if (depth == 0)
1249                                         tracehook_report_exec(fmt, bprm, regs);
1250                                 put_binfmt(fmt);
1251                                 allow_write_access(bprm->file);
1252                                 if (bprm->file)
1253                                         fput(bprm->file);
1254                                 bprm->file = NULL;
1255                                 current->did_exec = 1;
1256                                 proc_exec_connector(current);
1257                                 return retval;
1258                         }
1259                         read_lock(&binfmt_lock);
1260                         put_binfmt(fmt);
1261                         if (retval != -ENOEXEC || bprm->mm == NULL)
1262                                 break;
1263                         if (!bprm->file) {
1264                                 read_unlock(&binfmt_lock);
1265                                 return retval;
1266                         }
1267                 }
1268                 read_unlock(&binfmt_lock);
1269                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1270                         break;
1271 #ifdef CONFIG_MODULES
1272                 } else {
1273 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1274                         if (printable(bprm->buf[0]) &&
1275                             printable(bprm->buf[1]) &&
1276                             printable(bprm->buf[2]) &&
1277                             printable(bprm->buf[3]))
1278                                 break; /* -ENOEXEC */
1279                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1280 #endif
1281                 }
1282         }
1283         return retval;
1284 }
1285
1286 EXPORT_SYMBOL(search_binary_handler);
1287
1288 /*
1289  * sys_execve() executes a new program.
1290  */
1291 int do_execve(char * filename,
1292         char __user *__user *argv,
1293         char __user *__user *envp,
1294         struct pt_regs * regs)
1295 {
1296         struct linux_binprm *bprm;
1297         struct file *file;
1298         struct files_struct *displaced;
1299         bool clear_in_exec;
1300         int retval;
1301
1302         retval = unshare_files(&displaced);
1303         if (retval)
1304                 goto out_ret;
1305
1306         retval = -ENOMEM;
1307         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1308         if (!bprm)
1309                 goto out_files;
1310
1311         retval = prepare_bprm_creds(bprm);
1312         if (retval)
1313                 goto out_free;
1314
1315         retval = check_unsafe_exec(bprm);
1316         if (retval < 0)
1317                 goto out_free;
1318         clear_in_exec = retval;
1319         current->in_execve = 1;
1320
1321         file = open_exec(filename);
1322         retval = PTR_ERR(file);
1323         if (IS_ERR(file))
1324                 goto out_unmark;
1325
1326         sched_exec();
1327
1328         bprm->file = file;
1329         bprm->filename = filename;
1330         bprm->interp = filename;
1331
1332         retval = bprm_mm_init(bprm);
1333         if (retval)
1334                 goto out_file;
1335
1336         bprm->argc = count(argv, MAX_ARG_STRINGS);
1337         if ((retval = bprm->argc) < 0)
1338                 goto out;
1339
1340         bprm->envc = count(envp, MAX_ARG_STRINGS);
1341         if ((retval = bprm->envc) < 0)
1342                 goto out;
1343
1344         retval = prepare_binprm(bprm);
1345         if (retval < 0)
1346                 goto out;
1347
1348         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1349         if (retval < 0)
1350                 goto out;
1351
1352         bprm->exec = bprm->p;
1353         retval = copy_strings(bprm->envc, envp, bprm);
1354         if (retval < 0)
1355                 goto out;
1356
1357         retval = copy_strings(bprm->argc, argv, bprm);
1358         if (retval < 0)
1359                 goto out;
1360
1361         current->flags &= ~PF_KTHREAD;
1362         retval = search_binary_handler(bprm,regs);
1363         if (retval < 0)
1364                 goto out;
1365
1366         current->stack_start = current->mm->start_stack;
1367
1368         /* execve succeeded */
1369         current->fs->in_exec = 0;
1370         current->in_execve = 0;
1371         acct_update_integrals(current);
1372         free_bprm(bprm);
1373         if (displaced)
1374                 put_files_struct(displaced);
1375         return retval;
1376
1377 out:
1378         if (bprm->mm)
1379                 mmput (bprm->mm);
1380
1381 out_file:
1382         if (bprm->file) {
1383                 allow_write_access(bprm->file);
1384                 fput(bprm->file);
1385         }
1386
1387 out_unmark:
1388         if (clear_in_exec)
1389                 current->fs->in_exec = 0;
1390         current->in_execve = 0;
1391
1392 out_free:
1393         free_bprm(bprm);
1394
1395 out_files:
1396         if (displaced)
1397                 reset_files_struct(displaced);
1398 out_ret:
1399         return retval;
1400 }
1401
1402 void set_binfmt(struct linux_binfmt *new)
1403 {
1404         struct mm_struct *mm = current->mm;
1405
1406         if (mm->binfmt)
1407                 module_put(mm->binfmt->module);
1408
1409         mm->binfmt = new;
1410         if (new)
1411                 __module_get(new->module);
1412 }
1413
1414 EXPORT_SYMBOL(set_binfmt);
1415
1416 /* format_corename will inspect the pattern parameter, and output a
1417  * name into corename, which must have space for at least
1418  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1419  */
1420 static int format_corename(char *corename, long signr)
1421 {
1422         const struct cred *cred = current_cred();
1423         const char *pat_ptr = core_pattern;
1424         int ispipe = (*pat_ptr == '|');
1425         char *out_ptr = corename;
1426         char *const out_end = corename + CORENAME_MAX_SIZE;
1427         int rc;
1428         int pid_in_pattern = 0;
1429
1430         /* Repeat as long as we have more pattern to process and more output
1431            space */
1432         while (*pat_ptr) {
1433                 if (*pat_ptr != '%') {
1434                         if (out_ptr == out_end)
1435                                 goto out;
1436                         *out_ptr++ = *pat_ptr++;
1437                 } else {
1438                         switch (*++pat_ptr) {
1439                         case 0:
1440                                 goto out;
1441                         /* Double percent, output one percent */
1442                         case '%':
1443                                 if (out_ptr == out_end)
1444                                         goto out;
1445                                 *out_ptr++ = '%';
1446                                 break;
1447                         /* pid */
1448                         case 'p':
1449                                 pid_in_pattern = 1;
1450                                 rc = snprintf(out_ptr, out_end - out_ptr,
1451                                               "%d", task_tgid_vnr(current));
1452                                 if (rc > out_end - out_ptr)
1453                                         goto out;
1454                                 out_ptr += rc;
1455                                 break;
1456                         /* uid */
1457                         case 'u':
1458                                 rc = snprintf(out_ptr, out_end - out_ptr,
1459                                               "%d", cred->uid);
1460                                 if (rc > out_end - out_ptr)
1461                                         goto out;
1462                                 out_ptr += rc;
1463                                 break;
1464                         /* gid */
1465                         case 'g':
1466                                 rc = snprintf(out_ptr, out_end - out_ptr,
1467                                               "%d", cred->gid);
1468                                 if (rc > out_end - out_ptr)
1469                                         goto out;
1470                                 out_ptr += rc;
1471                                 break;
1472                         /* signal that caused the coredump */
1473                         case 's':
1474                                 rc = snprintf(out_ptr, out_end - out_ptr,
1475                                               "%ld", signr);
1476                                 if (rc > out_end - out_ptr)
1477                                         goto out;
1478                                 out_ptr += rc;
1479                                 break;
1480                         /* UNIX time of coredump */
1481                         case 't': {
1482                                 struct timeval tv;
1483                                 do_gettimeofday(&tv);
1484                                 rc = snprintf(out_ptr, out_end - out_ptr,
1485                                               "%lu", tv.tv_sec);
1486                                 if (rc > out_end - out_ptr)
1487                                         goto out;
1488                                 out_ptr += rc;
1489                                 break;
1490                         }
1491                         /* hostname */
1492                         case 'h':
1493                                 down_read(&uts_sem);
1494                                 rc = snprintf(out_ptr, out_end - out_ptr,
1495                                               "%s", utsname()->nodename);
1496                                 up_read(&uts_sem);
1497                                 if (rc > out_end - out_ptr)
1498                                         goto out;
1499                                 out_ptr += rc;
1500                                 break;
1501                         /* executable */
1502                         case 'e':
1503                                 rc = snprintf(out_ptr, out_end - out_ptr,
1504                                               "%s", current->comm);
1505                                 if (rc > out_end - out_ptr)
1506                                         goto out;
1507                                 out_ptr += rc;
1508                                 break;
1509                         /* core limit size */
1510                         case 'c':
1511                                 rc = snprintf(out_ptr, out_end - out_ptr,
1512                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1513                                 if (rc > out_end - out_ptr)
1514                                         goto out;
1515                                 out_ptr += rc;
1516                                 break;
1517                         default:
1518                                 break;
1519                         }
1520                         ++pat_ptr;
1521                 }
1522         }
1523         /* Backward compatibility with core_uses_pid:
1524          *
1525          * If core_pattern does not include a %p (as is the default)
1526          * and core_uses_pid is set, then .%pid will be appended to
1527          * the filename. Do not do this for piped commands. */
1528         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1529                 rc = snprintf(out_ptr, out_end - out_ptr,
1530                               ".%d", task_tgid_vnr(current));
1531                 if (rc > out_end - out_ptr)
1532                         goto out;
1533                 out_ptr += rc;
1534         }
1535 out:
1536         *out_ptr = 0;
1537         return ispipe;
1538 }
1539
1540 static int zap_process(struct task_struct *start)
1541 {
1542         struct task_struct *t;
1543         int nr = 0;
1544
1545         start->signal->flags = SIGNAL_GROUP_EXIT;
1546         start->signal->group_stop_count = 0;
1547
1548         t = start;
1549         do {
1550                 if (t != current && t->mm) {
1551                         sigaddset(&t->pending.signal, SIGKILL);
1552                         signal_wake_up(t, 1);
1553                         nr++;
1554                 }
1555         } while_each_thread(start, t);
1556
1557         return nr;
1558 }
1559
1560 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1561                                 struct core_state *core_state, int exit_code)
1562 {
1563         struct task_struct *g, *p;
1564         unsigned long flags;
1565         int nr = -EAGAIN;
1566
1567         spin_lock_irq(&tsk->sighand->siglock);
1568         if (!signal_group_exit(tsk->signal)) {
1569                 mm->core_state = core_state;
1570                 tsk->signal->group_exit_code = exit_code;
1571                 nr = zap_process(tsk);
1572         }
1573         spin_unlock_irq(&tsk->sighand->siglock);
1574         if (unlikely(nr < 0))
1575                 return nr;
1576
1577         if (atomic_read(&mm->mm_users) == nr + 1)
1578                 goto done;
1579         /*
1580          * We should find and kill all tasks which use this mm, and we should
1581          * count them correctly into ->nr_threads. We don't take tasklist
1582          * lock, but this is safe wrt:
1583          *
1584          * fork:
1585          *      None of sub-threads can fork after zap_process(leader). All
1586          *      processes which were created before this point should be
1587          *      visible to zap_threads() because copy_process() adds the new
1588          *      process to the tail of init_task.tasks list, and lock/unlock
1589          *      of ->siglock provides a memory barrier.
1590          *
1591          * do_exit:
1592          *      The caller holds mm->mmap_sem. This means that the task which
1593          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1594          *      its ->mm.
1595          *
1596          * de_thread:
1597          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1598          *      we must see either old or new leader, this does not matter.
1599          *      However, it can change p->sighand, so lock_task_sighand(p)
1600          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1601          *      it can't fail.
1602          *
1603          *      Note also that "g" can be the old leader with ->mm == NULL
1604          *      and already unhashed and thus removed from ->thread_group.
1605          *      This is OK, __unhash_process()->list_del_rcu() does not
1606          *      clear the ->next pointer, we will find the new leader via
1607          *      next_thread().
1608          */
1609         rcu_read_lock();
1610         for_each_process(g) {
1611                 if (g == tsk->group_leader)
1612                         continue;
1613                 if (g->flags & PF_KTHREAD)
1614                         continue;
1615                 p = g;
1616                 do {
1617                         if (p->mm) {
1618                                 if (unlikely(p->mm == mm)) {
1619                                         lock_task_sighand(p, &flags);
1620                                         nr += zap_process(p);
1621                                         unlock_task_sighand(p, &flags);
1622                                 }
1623                                 break;
1624                         }
1625                 } while_each_thread(g, p);
1626         }
1627         rcu_read_unlock();
1628 done:
1629         atomic_set(&core_state->nr_threads, nr);
1630         return nr;
1631 }
1632
1633 static int coredump_wait(int exit_code, struct core_state *core_state)
1634 {
1635         struct task_struct *tsk = current;
1636         struct mm_struct *mm = tsk->mm;
1637         struct completion *vfork_done;
1638         int core_waiters;
1639
1640         init_completion(&core_state->startup);
1641         core_state->dumper.task = tsk;
1642         core_state->dumper.next = NULL;
1643         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1644         up_write(&mm->mmap_sem);
1645
1646         if (unlikely(core_waiters < 0))
1647                 goto fail;
1648
1649         /*
1650          * Make sure nobody is waiting for us to release the VM,
1651          * otherwise we can deadlock when we wait on each other
1652          */
1653         vfork_done = tsk->vfork_done;
1654         if (vfork_done) {
1655                 tsk->vfork_done = NULL;
1656                 complete(vfork_done);
1657         }
1658
1659         if (core_waiters)
1660                 wait_for_completion(&core_state->startup);
1661 fail:
1662         return core_waiters;
1663 }
1664
1665 static void coredump_finish(struct mm_struct *mm)
1666 {
1667         struct core_thread *curr, *next;
1668         struct task_struct *task;
1669
1670         next = mm->core_state->dumper.next;
1671         while ((curr = next) != NULL) {
1672                 next = curr->next;
1673                 task = curr->task;
1674                 /*
1675                  * see exit_mm(), curr->task must not see
1676                  * ->task == NULL before we read ->next.
1677                  */
1678                 smp_mb();
1679                 curr->task = NULL;
1680                 wake_up_process(task);
1681         }
1682
1683         mm->core_state = NULL;
1684 }
1685
1686 /*
1687  * set_dumpable converts traditional three-value dumpable to two flags and
1688  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1689  * these bits are not changed atomically.  So get_dumpable can observe the
1690  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1691  * return either old dumpable or new one by paying attention to the order of
1692  * modifying the bits.
1693  *
1694  * dumpable |   mm->flags (binary)
1695  * old  new | initial interim  final
1696  * ---------+-----------------------
1697  *  0    1  |   00      01      01
1698  *  0    2  |   00      10(*)   11
1699  *  1    0  |   01      00      00
1700  *  1    2  |   01      11      11
1701  *  2    0  |   11      10(*)   00
1702  *  2    1  |   11      11      01
1703  *
1704  * (*) get_dumpable regards interim value of 10 as 11.
1705  */
1706 void set_dumpable(struct mm_struct *mm, int value)
1707 {
1708         switch (value) {
1709         case 0:
1710                 clear_bit(MMF_DUMPABLE, &mm->flags);
1711                 smp_wmb();
1712                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1713                 break;
1714         case 1:
1715                 set_bit(MMF_DUMPABLE, &mm->flags);
1716                 smp_wmb();
1717                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1718                 break;
1719         case 2:
1720                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1721                 smp_wmb();
1722                 set_bit(MMF_DUMPABLE, &mm->flags);
1723                 break;
1724         }
1725 }
1726
1727 int get_dumpable(struct mm_struct *mm)
1728 {
1729         int ret;
1730
1731         ret = mm->flags & 0x3;
1732         return (ret >= 2) ? 2 : ret;
1733 }
1734
1735 static void wait_for_dump_helpers(struct file *file)
1736 {
1737         struct pipe_inode_info *pipe;
1738
1739         pipe = file->f_path.dentry->d_inode->i_pipe;
1740
1741         pipe_lock(pipe);
1742         pipe->readers++;
1743         pipe->writers--;
1744
1745         while ((pipe->readers > 1) && (!signal_pending(current))) {
1746                 wake_up_interruptible_sync(&pipe->wait);
1747                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1748                 pipe_wait(pipe);
1749         }
1750
1751         pipe->readers--;
1752         pipe->writers++;
1753         pipe_unlock(pipe);
1754
1755 }
1756
1757
1758 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1759 {
1760         struct core_state core_state;
1761         char corename[CORENAME_MAX_SIZE + 1];
1762         struct mm_struct *mm = current->mm;
1763         struct linux_binfmt * binfmt;
1764         struct inode * inode;
1765         struct file * file;
1766         const struct cred *old_cred;
1767         struct cred *cred;
1768         int retval = 0;
1769         int flag = 0;
1770         int ispipe = 0;
1771         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1772         char **helper_argv = NULL;
1773         int helper_argc = 0;
1774         int dump_count = 0;
1775         static atomic_t core_dump_count = ATOMIC_INIT(0);
1776
1777         audit_core_dumps(signr);
1778
1779         binfmt = mm->binfmt;
1780         if (!binfmt || !binfmt->core_dump)
1781                 goto fail;
1782
1783         cred = prepare_creds();
1784         if (!cred) {
1785                 retval = -ENOMEM;
1786                 goto fail;
1787         }
1788
1789         down_write(&mm->mmap_sem);
1790         /*
1791          * If another thread got here first, or we are not dumpable, bail out.
1792          */
1793         if (mm->core_state || !get_dumpable(mm)) {
1794                 up_write(&mm->mmap_sem);
1795                 put_cred(cred);
1796                 goto fail;
1797         }
1798
1799         /*
1800          *      We cannot trust fsuid as being the "true" uid of the
1801          *      process nor do we know its entire history. We only know it
1802          *      was tainted so we dump it as root in mode 2.
1803          */
1804         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1805                 flag = O_EXCL;          /* Stop rewrite attacks */
1806                 cred->fsuid = 0;        /* Dump root private */
1807         }
1808
1809         retval = coredump_wait(exit_code, &core_state);
1810         if (retval < 0) {
1811                 put_cred(cred);
1812                 goto fail;
1813         }
1814
1815         old_cred = override_creds(cred);
1816
1817         /*
1818          * Clear any false indication of pending signals that might
1819          * be seen by the filesystem code called to write the core file.
1820          */
1821         clear_thread_flag(TIF_SIGPENDING);
1822
1823         /*
1824          * lock_kernel() because format_corename() is controlled by sysctl, which
1825          * uses lock_kernel()
1826          */
1827         lock_kernel();
1828         ispipe = format_corename(corename, signr);
1829         unlock_kernel();
1830
1831         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1832                 goto fail_unlock;
1833
1834         if (ispipe) {
1835                 if (core_limit == 0) {
1836                         /*
1837                          * Normally core limits are irrelevant to pipes, since
1838                          * we're not writing to the file system, but we use
1839                          * core_limit of 0 here as a speacial value. Any
1840                          * non-zero limit gets set to RLIM_INFINITY below, but
1841                          * a limit of 0 skips the dump.  This is a consistent
1842                          * way to catch recursive crashes.  We can still crash
1843                          * if the core_pattern binary sets RLIM_CORE =  !0
1844                          * but it runs as root, and can do lots of stupid things
1845                          * Note that we use task_tgid_vnr here to grab the pid
1846                          * of the process group leader.  That way we get the
1847                          * right pid if a thread in a multi-threaded
1848                          * core_pattern process dies.
1849                          */
1850                         printk(KERN_WARNING
1851                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1852                                 task_tgid_vnr(current), current->comm);
1853                         printk(KERN_WARNING "Aborting core\n");
1854                         goto fail_unlock;
1855                 }
1856
1857                 dump_count = atomic_inc_return(&core_dump_count);
1858                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1859                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1860                                task_tgid_vnr(current), current->comm);
1861                         printk(KERN_WARNING "Skipping core dump\n");
1862                         goto fail_dropcount;
1863                 }
1864
1865                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1866                 if (!helper_argv) {
1867                         printk(KERN_WARNING "%s failed to allocate memory\n",
1868                                __func__);
1869                         goto fail_dropcount;
1870                 }
1871
1872                 core_limit = RLIM_INFINITY;
1873
1874                 /* SIGPIPE can happen, but it's just never processed */
1875                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1876                                 &file)) {
1877                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1878                                corename);
1879                         goto fail_dropcount;
1880                 }
1881         } else
1882                 file = filp_open(corename,
1883                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1884                                  0600);
1885         if (IS_ERR(file))
1886                 goto fail_dropcount;
1887         inode = file->f_path.dentry->d_inode;
1888         if (inode->i_nlink > 1)
1889                 goto close_fail;        /* multiple links - don't dump */
1890         if (!ispipe && d_unhashed(file->f_path.dentry))
1891                 goto close_fail;
1892
1893         /* AK: actually i see no reason to not allow this for named pipes etc.,
1894            but keep the previous behaviour for now. */
1895         if (!ispipe && !S_ISREG(inode->i_mode))
1896                 goto close_fail;
1897         /*
1898          * Dont allow local users get cute and trick others to coredump
1899          * into their pre-created files:
1900          */
1901         if (inode->i_uid != current_fsuid())
1902                 goto close_fail;
1903         if (!file->f_op)
1904                 goto close_fail;
1905         if (!file->f_op->write)
1906                 goto close_fail;
1907         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1908                 goto close_fail;
1909
1910         retval = binfmt->core_dump(signr, regs, file, core_limit);
1911
1912         if (retval)
1913                 current->signal->group_exit_code |= 0x80;
1914 close_fail:
1915         if (ispipe && core_pipe_limit)
1916                 wait_for_dump_helpers(file);
1917         filp_close(file, NULL);
1918 fail_dropcount:
1919         if (dump_count)
1920                 atomic_dec(&core_dump_count);
1921 fail_unlock:
1922         if (helper_argv)
1923                 argv_free(helper_argv);
1924
1925         revert_creds(old_cred);
1926         put_cred(cred);
1927         coredump_finish(mm);
1928 fail:
1929         return;
1930 }