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