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/swap.h>
  32 #include <linux/string.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.h>
  45 #include <linux/proc_fs.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57 #include <linux/oom.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 struct core_name {
  70         char *corename;
  71         int used, size;
  72 };
  73 static atomic_t call_count = ATOMIC_INIT(1);
  74
  75 /* The maximal length of core_pattern is also specified in sysctl.c */
  76
  77 static LIST_HEAD(formats);
  78 static DEFINE_RWLOCK(binfmt_lock);
  79
  80 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  81 {
  82         if (!fmt)
  83                 return -EINVAL;
  84         write_lock(&binfmt_lock);
  85         insert ? list_add(&fmt->lh, &formats) :
  86                  list_add_tail(&fmt->lh, &formats);
  87         write_unlock(&binfmt_lock);
  88         return 0;
  89 }
  90
  91 EXPORT_SYMBOL(__register_binfmt);
  92
  93 void unregister_binfmt(struct linux_binfmt * fmt)
  94 {
  95         write_lock(&binfmt_lock);
  96         list_del(&fmt->lh);
  97         write_unlock(&binfmt_lock);
  98 }
  99
 100 EXPORT_SYMBOL(unregister_binfmt);
 101
 102 static inline void put_binfmt(struct linux_binfmt * fmt)
 103 {
 104         module_put(fmt->module);
 105 }
 106
 107 /*
 108  * Note that a shared library must be both readable and executable due to
 109  * security reasons.
 110  *
 111  * Also note that we take the address to load from from the file itself.
 112  */
 113 SYSCALL_DEFINE1(uselib, const char __user *, library)
 114 {
 115         struct file *file;
 116         char *tmp = getname(library);
 117         int error = PTR_ERR(tmp);
 118
 119         if (IS_ERR(tmp))
 120                 goto out;
 121
 122         file = do_filp_open(AT_FDCWD, tmp,
 123                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 124                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 125         putname(tmp);
 126         error = PTR_ERR(file);
 127         if (IS_ERR(file))
 128                 goto out;
 129
 130         error = -EINVAL;
 131         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 132                 goto exit;
 133
 134         error = -EACCES;
 135         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 136                 goto exit;
 137
 138         fsnotify_open(file);
 139
 140         error = -ENOEXEC;
 141         if(file->f_op) {
 142                 struct linux_binfmt * fmt;
 143
 144                 read_lock(&binfmt_lock);
 145                 list_for_each_entry(fmt, &formats, lh) {
 146                         if (!fmt->load_shlib)
 147                                 continue;
 148                         if (!try_module_get(fmt->module))
 149                                 continue;
 150                         read_unlock(&binfmt_lock);
 151                         error = fmt->load_shlib(file);
 152                         read_lock(&binfmt_lock);
 153                         put_binfmt(fmt);
 154                         if (error != -ENOEXEC)
 155                                 break;
 156                 }
 157                 read_unlock(&binfmt_lock);
 158         }
 159 exit:
 160         fput(file);
 161 out:
 162         return error;
 163 }
 164
 165 #ifdef CONFIG_MMU
 166
 167 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 168                 int write)
 169 {
 170         struct page *page;
 171         int ret;
 172
 173 #ifdef CONFIG_STACK_GROWSUP
 174         if (write) {
 175                 ret = expand_stack_downwards(bprm->vma, pos);
 176                 if (ret < 0)
 177                         return NULL;
 178         }
 179 #endif
 180         ret = get_user_pages(current, bprm->mm, pos,
 181                         1, write, 1, &page, NULL);
 182         if (ret <= 0)
 183                 return NULL;
 184
 185         if (write) {
 186                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 187                 struct rlimit *rlim;
 188
 189                 /*
 190                  * We've historically supported up to 32 pages (ARG_MAX)
 191                  * of argument strings even with small stacks
 192                  */
 193                 if (size <= ARG_MAX)
 194                         return page;
 195
 196                 /*
 197                  * Limit to 1/4-th the stack size for the argv+env strings.
 198                  * This ensures that:
 199                  *  - the remaining binfmt code will not run out of stack space,
 200                  *  - the program will have a reasonable amount of stack left
 201                  *    to work from.
 202                  */
 203                 rlim = current->signal->rlim;
 204                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 205                         put_page(page);
 206                         return NULL;
 207                 }
 208         }
 209
 210         return page;
 211 }
 212
 213 static void put_arg_page(struct page *page)
 214 {
 215         put_page(page);
 216 }
 217
 218 static void free_arg_page(struct linux_binprm *bprm, int i)
 219 {
 220 }
 221
 222 static void free_arg_pages(struct linux_binprm *bprm)
 223 {
 224 }
 225
 226 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 227                 struct page *page)
 228 {
 229         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 230 }
 231
 232 static int __bprm_mm_init(struct linux_binprm *bprm)
 233 {
 234         int err;
 235         struct vm_area_struct *vma = NULL;
 236         struct mm_struct *mm = bprm->mm;
 237
 238         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 239         if (!vma)
 240                 return -ENOMEM;
 241
 242         down_write(&mm->mmap_sem);
 243         vma->vm_mm = mm;
 244
 245         /*
 246          * Place the stack at the largest stack address the architecture
 247          * supports. Later, we'll move this to an appropriate place. We don't
 248          * use STACK_TOP because that can depend on attributes which aren't
 249          * configured yet.
 250          */
 251         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 252         vma->vm_end = STACK_TOP_MAX;
 253         vma->vm_start = vma->vm_end - PAGE_SIZE;
 254         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 255         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 256         INIT_LIST_HEAD(&vma->anon_vma_chain);
 257         err = insert_vm_struct(mm, vma);
 258         if (err)
 259                 goto err;
 260
 261         mm->stack_vm = mm->total_vm = 1;
 262         up_write(&mm->mmap_sem);
 263         bprm->p = vma->vm_end - sizeof(void *);
 264         return 0;
 265 err:
 266         up_write(&mm->mmap_sem);
 267         bprm->vma = NULL;
 268         kmem_cache_free(vm_area_cachep, vma);
 269         return err;
 270 }
 271
 272 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 273 {
 274         return len <= MAX_ARG_STRLEN;
 275 }
 276
 277 #else
 278
 279 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 280                 int write)
 281 {
 282         struct page *page;
 283
 284         page = bprm->page[pos / PAGE_SIZE];
 285         if (!page && write) {
 286                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 287                 if (!page)
 288                         return NULL;
 289                 bprm->page[pos / PAGE_SIZE] = page;
 290         }
 291
 292         return page;
 293 }
 294
 295 static void put_arg_page(struct page *page)
 296 {
 297 }
 298
 299 static void free_arg_page(struct linux_binprm *bprm, int i)
 300 {
 301         if (bprm->page[i]) {
 302                 __free_page(bprm->page[i]);
 303                 bprm->page[i] = NULL;
 304         }
 305 }
 306
 307 static void free_arg_pages(struct linux_binprm *bprm)
 308 {
 309         int i;
 310
 311         for (i = 0; i < MAX_ARG_PAGES; i++)
 312                 free_arg_page(bprm, i);
 313 }
 314
 315 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 316                 struct page *page)
 317 {
 318 }
 319
 320 static int __bprm_mm_init(struct linux_binprm *bprm)
 321 {
 322         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 323         return 0;
 324 }
 325
 326 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 327 {
 328         return len <= bprm->p;
 329 }
 330
 331 #endif /* CONFIG_MMU */
 332
 333 /*
 334  * Create a new mm_struct and populate it with a temporary stack
 335  * vm_area_struct.  We don't have enough context at this point to set the stack
 336  * flags, permissions, and offset, so we use temporary values.  We'll update
 337  * them later in setup_arg_pages().
 338  */
 339 int bprm_mm_init(struct linux_binprm *bprm)
 340 {
 341         int err;
 342         struct mm_struct *mm = NULL;
 343
 344         bprm->mm = mm = mm_alloc();
 345         err = -ENOMEM;
 346         if (!mm)
 347                 goto err;
 348
 349         err = init_new_context(current, mm);
 350         if (err)
 351                 goto err;
 352
 353         err = __bprm_mm_init(bprm);
 354         if (err)
 355                 goto err;
 356
 357         return 0;
 358
 359 err:
 360         if (mm) {
 361                 bprm->mm = NULL;
 362                 mmdrop(mm);
 363         }
 364
 365         return err;
 366 }
 367
 368 /*
 369  * count() counts the number of strings in array ARGV.
 370  */
 371 static int count(const char __user * const __user * argv, int max)
 372 {
 373         int i = 0;
 374
 375         if (argv != NULL) {
 376                 for (;;) {
 377                         const char __user * p;
 378
 379                         if (get_user(p, argv))
 380                                 return -EFAULT;
 381                         if (!p)
 382                                 break;
 383                         argv++;
 384                         if (i++ >= max)
 385                                 return -E2BIG;
 386
 387                         if (fatal_signal_pending(current))
 388                                 return -ERESTARTNOHAND;
 389                         cond_resched();
 390                 }
 391         }
 392         return i;
 393 }
 394
 395 /*
 396  * 'copy_strings()' copies argument/environment strings from the old
 397  * processes's memory to the new process's stack.  The call to get_user_pages()
 398  * ensures the destination page is created and not swapped out.
 399  */
 400 static int copy_strings(int argc, const char __user *const __user *argv,
 401                         struct linux_binprm *bprm)
 402 {
 403         struct page *kmapped_page = NULL;
 404         char *kaddr = NULL;
 405         unsigned long kpos = 0;
 406         int ret;
 407
 408         while (argc-- > 0) {
 409                 const char __user *str;
 410                 int len;
 411                 unsigned long pos;
 412
 413                 if (get_user(str, argv+argc) ||
 414                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 415                         ret = -EFAULT;
 416                         goto out;
 417                 }
 418
 419                 if (!valid_arg_len(bprm, len)) {
 420                         ret = -E2BIG;
 421                         goto out;
 422                 }
 423
 424                 /* We're going to work our way backwords. */
 425                 pos = bprm->p;
 426                 str += len;
 427                 bprm->p -= len;
 428
 429                 while (len > 0) {
 430                         int offset, bytes_to_copy;
 431
 432                         if (fatal_signal_pending(current)) {
 433                                 ret = -ERESTARTNOHAND;
 434                                 goto out;
 435                         }
 436                         cond_resched();
 437
 438                         offset = pos % PAGE_SIZE;
 439                         if (offset == 0)
 440                                 offset = PAGE_SIZE;
 441
 442                         bytes_to_copy = offset;
 443                         if (bytes_to_copy > len)
 444                                 bytes_to_copy = len;
 445
 446                         offset -= bytes_to_copy;
 447                         pos -= bytes_to_copy;
 448                         str -= bytes_to_copy;
 449                         len -= bytes_to_copy;
 450
 451                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 452                                 struct page *page;
 453
 454                                 page = get_arg_page(bprm, pos, 1);
 455                                 if (!page) {
 456                                         ret = -E2BIG;
 457                                         goto out;
 458                                 }
 459
 460                                 if (kmapped_page) {
 461                                         flush_kernel_dcache_page(kmapped_page);
 462                                         kunmap(kmapped_page);
 463                                         put_arg_page(kmapped_page);
 464                                 }
 465                                 kmapped_page = page;
 466                                 kaddr = kmap(kmapped_page);
 467                                 kpos = pos & PAGE_MASK;
 468                                 flush_arg_page(bprm, kpos, kmapped_page);
 469                         }
 470                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 471                                 ret = -EFAULT;
 472                                 goto out;
 473                         }
 474                 }
 475         }
 476         ret = 0;
 477 out:
 478         if (kmapped_page) {
 479                 flush_kernel_dcache_page(kmapped_page);
 480                 kunmap(kmapped_page);
 481                 put_arg_page(kmapped_page);
 482         }
 483         return ret;
 484 }
 485
 486 /*
 487  * Like copy_strings, but get argv and its values from kernel memory.
 488  */
 489 int copy_strings_kernel(int argc, const char *const *argv,
 490                         struct linux_binprm *bprm)
 491 {
 492         int r;
 493         mm_segment_t oldfs = get_fs();
 494         set_fs(KERNEL_DS);
 495         r = copy_strings(argc, (const char __user *const  __user *)argv, bprm);
 496         set_fs(oldfs);
 497         return r;
 498 }
 499 EXPORT_SYMBOL(copy_strings_kernel);
 500
 501 #ifdef CONFIG_MMU
 502
 503 /*
 504  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 505  * the binfmt code determines where the new stack should reside, we shift it to
 506  * its final location.  The process proceeds as follows:
 507  *
 508  * 1) Use shift to calculate the new vma endpoints.
 509  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 510  *    arguments passed to subsequent functions are consistent.
 511  * 3) Move vma's page tables to the new range.
 512  * 4) Free up any cleared pgd range.
 513  * 5) Shrink the vma to cover only the new range.
 514  */
 515 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 516 {
 517         struct mm_struct *mm = vma->vm_mm;
 518         unsigned long old_start = vma->vm_start;
 519         unsigned long old_end = vma->vm_end;
 520         unsigned long length = old_end - old_start;
 521         unsigned long new_start = old_start - shift;
 522         unsigned long new_end = old_end - shift;
 523         struct mmu_gather *tlb;
 524
 525         BUG_ON(new_start > new_end);
 526
 527         /*
 528          * ensure there are no vmas between where we want to go
 529          * and where we are
 530          */
 531         if (vma != find_vma(mm, new_start))
 532                 return -EFAULT;
 533
 534         /*
 535          * cover the whole range: [new_start, old_end)
 536          */
 537         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 538                 return -ENOMEM;
 539
 540         /*
 541          * move the page tables downwards, on failure we rely on
 542          * process cleanup to remove whatever mess we made.
 543          */
 544         if (length != move_page_tables(vma, old_start,
 545                                        vma, new_start, length))
 546                 return -ENOMEM;
 547
 548         lru_add_drain();
 549         tlb = tlb_gather_mmu(mm, 0);
 550         if (new_end > old_start) {
 551                 /*
 552                  * when the old and new regions overlap clear from new_end.
 553                  */
 554                 free_pgd_range(tlb, new_end, old_end, new_end,
 555                         vma->vm_next ? vma->vm_next->vm_start : 0);
 556         } else {
 557                 /*
 558                  * otherwise, clean from old_start; this is done to not touch
 559                  * the address space in [new_end, old_start) some architectures
 560                  * have constraints on va-space that make this illegal (IA64) -
 561                  * for the others its just a little faster.
 562                  */
 563                 free_pgd_range(tlb, old_start, old_end, new_end,
 564                         vma->vm_next ? vma->vm_next->vm_start : 0);
 565         }
 566         tlb_finish_mmu(tlb, new_end, old_end);
 567
 568         /*
 569          * Shrink the vma to just the new range.  Always succeeds.
 570          */
 571         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 572
 573         return 0;
 574 }
 575
 576 /*
 577  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 578  * the stack is optionally relocated, and some extra space is added.
 579  */
 580 int setup_arg_pages(struct linux_binprm *bprm,
 581                     unsigned long stack_top,
 582                     int executable_stack)
 583 {
 584         unsigned long ret;
 585         unsigned long stack_shift;
 586         struct mm_struct *mm = current->mm;
 587         struct vm_area_struct *vma = bprm->vma;
 588         struct vm_area_struct *prev = NULL;
 589         unsigned long vm_flags;
 590         unsigned long stack_base;
 591         unsigned long stack_size;
 592         unsigned long stack_expand;
 593         unsigned long rlim_stack;
 594
 595 #ifdef CONFIG_STACK_GROWSUP
 596         /* Limit stack size to 1GB */
 597         stack_base = rlimit_max(RLIMIT_STACK);
 598         if (stack_base > (1 << 30))
 599                 stack_base = 1 << 30;
 600
 601         /* Make sure we didn't let the argument array grow too large. */
 602         if (vma->vm_end - vma->vm_start > stack_base)
 603                 return -ENOMEM;
 604
 605         stack_base = PAGE_ALIGN(stack_top - stack_base);
 606
 607         stack_shift = vma->vm_start - stack_base;
 608         mm->arg_start = bprm->p - stack_shift;
 609         bprm->p = vma->vm_end - stack_shift;
 610 #else
 611         stack_top = arch_align_stack(stack_top);
 612         stack_top = PAGE_ALIGN(stack_top);
 613
 614         if (unlikely(stack_top < mmap_min_addr) ||
 615             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 616                 return -ENOMEM;
 617
 618         stack_shift = vma->vm_end - stack_top;
 619
 620         bprm->p -= stack_shift;
 621         mm->arg_start = bprm->p;
 622 #endif
 623
 624         if (bprm->loader)
 625                 bprm->loader -= stack_shift;
 626         bprm->exec -= stack_shift;
 627
 628         down_write(&mm->mmap_sem);
 629         vm_flags = VM_STACK_FLAGS;
 630
 631         /*
 632          * Adjust stack execute permissions; explicitly enable for
 633          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 634          * (arch default) otherwise.
 635          */
 636         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 637                 vm_flags |= VM_EXEC;
 638         else if (executable_stack == EXSTACK_DISABLE_X)
 639                 vm_flags &= ~VM_EXEC;
 640         vm_flags |= mm->def_flags;
 641         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 642
 643         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 644                         vm_flags);
 645         if (ret)
 646                 goto out_unlock;
 647         BUG_ON(prev != vma);
 648
 649         /* Move stack pages down in memory. */
 650         if (stack_shift) {
 651                 ret = shift_arg_pages(vma, stack_shift);
 652                 if (ret)
 653                         goto out_unlock;
 654         }
 655
 656         /* mprotect_fixup is overkill to remove the temporary stack flags */
 657         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 658
 659         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 660         stack_size = vma->vm_end - vma->vm_start;
 661         /*
 662          * Align this down to a page boundary as expand_stack
 663          * will align it up.
 664          */
 665         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 666 #ifdef CONFIG_STACK_GROWSUP
 667         if (stack_size + stack_expand > rlim_stack)
 668                 stack_base = vma->vm_start + rlim_stack;
 669         else
 670                 stack_base = vma->vm_end + stack_expand;
 671 #else
 672         if (stack_size + stack_expand > rlim_stack)
 673                 stack_base = vma->vm_end - rlim_stack;
 674         else
 675                 stack_base = vma->vm_start - stack_expand;
 676 #endif
 677         current->mm->start_stack = bprm->p;
 678         ret = expand_stack(vma, stack_base);
 679         if (ret)
 680                 ret = -EFAULT;
 681
 682 out_unlock:
 683         up_write(&mm->mmap_sem);
 684         return ret;
 685 }
 686 EXPORT_SYMBOL(setup_arg_pages);
 687
 688 #endif /* CONFIG_MMU */
 689
 690 struct file *open_exec(const char *name)
 691 {
 692         struct file *file;
 693         int err;
 694
 695         file = do_filp_open(AT_FDCWD, name,
 696                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 697                                 MAY_EXEC | MAY_OPEN);
 698         if (IS_ERR(file))
 699                 goto out;
 700
 701         err = -EACCES;
 702         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 703                 goto exit;
 704
 705         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 706                 goto exit;
 707
 708         fsnotify_open(file);
 709
 710         err = deny_write_access(file);
 711         if (err)
 712                 goto exit;
 713
 714 out:
 715         return file;
 716
 717 exit:
 718         fput(file);
 719         return ERR_PTR(err);
 720 }
 721 EXPORT_SYMBOL(open_exec);
 722
 723 int kernel_read(struct file *file, loff_t offset,
 724                 char *addr, unsigned long count)
 725 {
 726         mm_segment_t old_fs;
 727         loff_t pos = offset;
 728         int result;
 729
 730         old_fs = get_fs();
 731         set_fs(get_ds());
 732         /* The cast to a user pointer is valid due to the set_fs() */
 733         result = vfs_read(file, (void __user *)addr, count, &pos);
 734         set_fs(old_fs);
 735         return result;
 736 }
 737
 738 EXPORT_SYMBOL(kernel_read);
 739
 740 static int exec_mmap(struct mm_struct *mm)
 741 {
 742         struct task_struct *tsk;
 743         struct mm_struct * old_mm, *active_mm;
 744
 745         /* Notify parent that we're no longer interested in the old VM */
 746         tsk = current;
 747         old_mm = current->mm;
 748         sync_mm_rss(tsk, old_mm);
 749         mm_release(tsk, old_mm);
 750
 751         if (old_mm) {
 752                 /*
 753                  * Make sure that if there is a core dump in progress
 754                  * for the old mm, we get out and die instead of going
 755                  * through with the exec.  We must hold mmap_sem around
 756                  * checking core_state and changing tsk->mm.
 757                  */
 758                 down_read(&old_mm->mmap_sem);
 759                 if (unlikely(old_mm->core_state)) {
 760                         up_read(&old_mm->mmap_sem);
 761                         return -EINTR;
 762                 }
 763         }
 764         task_lock(tsk);
 765         active_mm = tsk->active_mm;
 766         tsk->mm = mm;
 767         tsk->active_mm = mm;
 768         activate_mm(active_mm, mm);
 769         if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
 770                 atomic_dec(&old_mm->oom_disable_count);
 771                 atomic_inc(&tsk->mm->oom_disable_count);
 772         }
 773         task_unlock(tsk);
 774         arch_pick_mmap_layout(mm);
 775         if (old_mm) {
 776                 up_read(&old_mm->mmap_sem);
 777                 BUG_ON(active_mm != old_mm);
 778                 mm_update_next_owner(old_mm);
 779                 mmput(old_mm);
 780                 return 0;
 781         }
 782         mmdrop(active_mm);
 783         return 0;
 784 }
 785
 786 /*
 787  * This function makes sure the current process has its own signal table,
 788  * so that flush_signal_handlers can later reset the handlers without
 789  * disturbing other processes.  (Other processes might share the signal
 790  * table via the CLONE_SIGHAND option to clone().)
 791  */
 792 static int de_thread(struct task_struct *tsk)
 793 {
 794         struct signal_struct *sig = tsk->signal;
 795         struct sighand_struct *oldsighand = tsk->sighand;
 796         spinlock_t *lock = &oldsighand->siglock;
 797
 798         if (thread_group_empty(tsk))
 799                 goto no_thread_group;
 800
 801         /*
 802          * Kill all other threads in the thread group.
 803          */
 804         spin_lock_irq(lock);
 805         if (signal_group_exit(sig)) {
 806                 /*
 807                  * Another group action in progress, just
 808                  * return so that the signal is processed.
 809                  */
 810                 spin_unlock_irq(lock);
 811                 return -EAGAIN;
 812         }
 813
 814         sig->group_exit_task = tsk;
 815         sig->notify_count = zap_other_threads(tsk);
 816         if (!thread_group_leader(tsk))
 817                 sig->notify_count--;
 818
 819         while (sig->notify_count) {
 820                 __set_current_state(TASK_UNINTERRUPTIBLE);
 821                 spin_unlock_irq(lock);
 822                 schedule();
 823                 spin_lock_irq(lock);
 824         }
 825         spin_unlock_irq(lock);
 826
 827         /*
 828          * At this point all other threads have exited, all we have to
 829          * do is to wait for the thread group leader to become inactive,
 830          * and to assume its PID:
 831          */
 832         if (!thread_group_leader(tsk)) {
 833                 struct task_struct *leader = tsk->group_leader;
 834
 835                 sig->notify_count = -1; /* for exit_notify() */
 836                 for (;;) {
 837                         write_lock_irq(&tasklist_lock);
 838                         if (likely(leader->exit_state))
 839                                 break;
 840                         __set_current_state(TASK_UNINTERRUPTIBLE);
 841                         write_unlock_irq(&tasklist_lock);
 842                         schedule();
 843                 }
 844
 845                 /*
 846                  * The only record we have of the real-time age of a
 847                  * process, regardless of execs it's done, is start_time.
 848                  * All the past CPU time is accumulated in signal_struct
 849                  * from sister threads now dead.  But in this non-leader
 850                  * exec, nothing survives from the original leader thread,
 851                  * whose birth marks the true age of this process now.
 852                  * When we take on its identity by switching to its PID, we
 853                  * also take its birthdate (always earlier than our own).
 854                  */
 855                 tsk->start_time = leader->start_time;
 856
 857                 BUG_ON(!same_thread_group(leader, tsk));
 858                 BUG_ON(has_group_leader_pid(tsk));
 859                 /*
 860                  * An exec() starts a new thread group with the
 861                  * TGID of the previous thread group. Rehash the
 862                  * two threads with a switched PID, and release
 863                  * the former thread group leader:
 864                  */
 865
 866                 /* Become a process group leader with the old leader's pid.
 867                  * The old leader becomes a thread of the this thread group.
 868                  * Note: The old leader also uses this pid until release_task
 869                  *       is called.  Odd but simple and correct.
 870                  */
 871                 detach_pid(tsk, PIDTYPE_PID);
 872                 tsk->pid = leader->pid;
 873                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 874                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 875                 transfer_pid(leader, tsk, PIDTYPE_SID);
 876
 877                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 878                 list_replace_init(&leader->sibling, &tsk->sibling);
 879
 880                 tsk->group_leader = tsk;
 881                 leader->group_leader = tsk;
 882
 883                 tsk->exit_signal = SIGCHLD;
 884
 885                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 886                 leader->exit_state = EXIT_DEAD;
 887                 write_unlock_irq(&tasklist_lock);
 888
 889                 release_task(leader);
 890         }
 891
 892         sig->group_exit_task = NULL;
 893         sig->notify_count = 0;
 894
 895 no_thread_group:
 896         if (current->mm)
 897                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 898
 899         exit_itimers(sig);
 900         flush_itimer_signals();
 901
 902         if (atomic_read(&oldsighand->count) != 1) {
 903                 struct sighand_struct *newsighand;
 904                 /*
 905                  * This ->sighand is shared with the CLONE_SIGHAND
 906                  * but not CLONE_THREAD task, switch to the new one.
 907                  */
 908                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 909                 if (!newsighand)
 910                         return -ENOMEM;
 911
 912                 atomic_set(&newsighand->count, 1);
 913                 memcpy(newsighand->action, oldsighand->action,
 914                        sizeof(newsighand->action));
 915
 916                 write_lock_irq(&tasklist_lock);
 917                 spin_lock(&oldsighand->siglock);
 918                 rcu_assign_pointer(tsk->sighand, newsighand);
 919                 spin_unlock(&oldsighand->siglock);
 920                 write_unlock_irq(&tasklist_lock);
 921
 922                 __cleanup_sighand(oldsighand);
 923         }
 924
 925         BUG_ON(!thread_group_leader(tsk));
 926         return 0;
 927 }
 928
 929 /*
 930  * These functions flushes out all traces of the currently running executable
 931  * so that a new one can be started
 932  */
 933 static void flush_old_files(struct files_struct * files)
 934 {
 935         long j = -1;
 936         struct fdtable *fdt;
 937
 938         spin_lock(&files->file_lock);
 939         for (;;) {
 940                 unsigned long set, i;
 941
 942                 j++;
 943                 i = j * __NFDBITS;
 944                 fdt = files_fdtable(files);
 945                 if (i >= fdt->max_fds)
 946                         break;
 947                 set = fdt->close_on_exec->fds_bits[j];
 948                 if (!set)
 949                         continue;
 950                 fdt->close_on_exec->fds_bits[j] = 0;
 951                 spin_unlock(&files->file_lock);
 952                 for ( ; set ; i++,set >>= 1) {
 953                         if (set & 1) {
 954                                 sys_close(i);
 955                         }
 956                 }
 957                 spin_lock(&files->file_lock);
 958
 959         }
 960         spin_unlock(&files->file_lock);
 961 }
 962
 963 char *get_task_comm(char *buf, struct task_struct *tsk)
 964 {
 965         /* buf must be at least sizeof(tsk->comm) in size */
 966         task_lock(tsk);
 967         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 968         task_unlock(tsk);
 969         return buf;
 970 }
 971
 972 void set_task_comm(struct task_struct *tsk, char *buf)
 973 {
 974         task_lock(tsk);
 975
 976         /*
 977          * Threads may access current->comm without holding
 978          * the task lock, so write the string carefully.
 979          * Readers without a lock may see incomplete new
 980          * names but are safe from non-terminating string reads.
 981          */
 982         memset(tsk->comm, 0, TASK_COMM_LEN);
 983         wmb();
 984         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 985         task_unlock(tsk);
 986         perf_event_comm(tsk);
 987 }
 988
 989 int flush_old_exec(struct linux_binprm * bprm)
 990 {
 991         int retval;
 992
 993         /*
 994          * Make sure we have a private signal table and that
 995          * we are unassociated from the previous thread group.
 996          */
 997         retval = de_thread(current);
 998         if (retval)
 999                 goto out;
1000
1001         set_mm_exe_file(bprm->mm, bprm->file);
1002
1003         /*
1004          * Release all of the old mmap stuff
1005          */
1006         retval = exec_mmap(bprm->mm);
1007         if (retval)
1008                 goto out;
1009
1010         bprm->mm = NULL;                /* We're using it now */
1011
1012         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1013         flush_thread();
1014         current->personality &= ~bprm->per_clear;
1015
1016         return 0;
1017
1018 out:
1019         return retval;
1020 }
1021 EXPORT_SYMBOL(flush_old_exec);
1022
1023 void setup_new_exec(struct linux_binprm * bprm)
1024 {
1025         int i, ch;
1026         const char *name;
1027         char tcomm[sizeof(current->comm)];
1028
1029         arch_pick_mmap_layout(current->mm);
1030
1031         /* This is the point of no return */
1032         current->sas_ss_sp = current->sas_ss_size = 0;
1033
1034         if (current_euid() == current_uid() && current_egid() == current_gid())
1035                 set_dumpable(current->mm, 1);
1036         else
1037                 set_dumpable(current->mm, suid_dumpable);
1038
1039         name = bprm->filename;
1040
1041         /* Copies the binary name from after last slash */
1042         for (i=0; (ch = *(name++)) != '\0';) {
1043                 if (ch == '/')
1044                         i = 0; /* overwrite what we wrote */
1045                 else
1046                         if (i < (sizeof(tcomm) - 1))
1047                                 tcomm[i++] = ch;
1048         }
1049         tcomm[i] = '\0';
1050         set_task_comm(current, tcomm);
1051
1052         /* Set the new mm task size. We have to do that late because it may
1053          * depend on TIF_32BIT which is only updated in flush_thread() on
1054          * some architectures like powerpc
1055          */
1056         current->mm->task_size = TASK_SIZE;
1057
1058         /* install the new credentials */
1059         if (bprm->cred->uid != current_euid() ||
1060             bprm->cred->gid != current_egid()) {
1061                 current->pdeath_signal = 0;
1062         } else if (file_permission(bprm->file, MAY_READ) ||
1063                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1064                 set_dumpable(current->mm, suid_dumpable);
1065         }
1066
1067         /*
1068          * Flush performance counters when crossing a
1069          * security domain:
1070          */
1071         if (!get_dumpable(current->mm))
1072                 perf_event_exit_task(current);
1073
1074         /* An exec changes our domain. We are no longer part of the thread
1075            group */
1076
1077         current->self_exec_id++;
1078
1079         flush_signal_handlers(current, 0);
1080         flush_old_files(current->files);
1081 }
1082 EXPORT_SYMBOL(setup_new_exec);
1083
1084 /*
1085  * Prepare credentials and lock ->cred_guard_mutex.
1086  * install_exec_creds() commits the new creds and drops the lock.
1087  * Or, if exec fails before, free_bprm() should release ->cred and
1088  * and unlock.
1089  */
1090 int prepare_bprm_creds(struct linux_binprm *bprm)
1091 {
1092         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1093                 return -ERESTARTNOINTR;
1094
1095         bprm->cred = prepare_exec_creds();
1096         if (likely(bprm->cred))
1097                 return 0;
1098
1099         mutex_unlock(&current->signal->cred_guard_mutex);
1100         return -ENOMEM;
1101 }
1102
1103 void free_bprm(struct linux_binprm *bprm)
1104 {
1105         free_arg_pages(bprm);
1106         if (bprm->cred) {
1107                 mutex_unlock(&current->signal->cred_guard_mutex);
1108                 abort_creds(bprm->cred);
1109         }
1110         kfree(bprm);
1111 }
1112
1113 /*
1114  * install the new credentials for this executable
1115  */
1116 void install_exec_creds(struct linux_binprm *bprm)
1117 {
1118         security_bprm_committing_creds(bprm);
1119
1120         commit_creds(bprm->cred);
1121         bprm->cred = NULL;
1122         /*
1123          * cred_guard_mutex must be held at least to this point to prevent
1124          * ptrace_attach() from altering our determination of the task's
1125          * credentials; any time after this it may be unlocked.
1126          */
1127         security_bprm_committed_creds(bprm);
1128         mutex_unlock(&current->signal->cred_guard_mutex);
1129 }
1130 EXPORT_SYMBOL(install_exec_creds);
1131
1132 /*
1133  * determine how safe it is to execute the proposed program
1134  * - the caller must hold ->cred_guard_mutex to protect against
1135  *   PTRACE_ATTACH
1136  */
1137 int check_unsafe_exec(struct linux_binprm *bprm)
1138 {
1139         struct task_struct *p = current, *t;
1140         unsigned n_fs;
1141         int res = 0;
1142
1143         bprm->unsafe = tracehook_unsafe_exec(p);
1144
1145         n_fs = 1;
1146         spin_lock(&p->fs->lock);
1147         rcu_read_lock();
1148         for (t = next_thread(p); t != p; t = next_thread(t)) {
1149                 if (t->fs == p->fs)
1150                         n_fs++;
1151         }
1152         rcu_read_unlock();
1153
1154         if (p->fs->users > n_fs) {
1155                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1156         } else {
1157                 res = -EAGAIN;
1158                 if (!p->fs->in_exec) {
1159                         p->fs->in_exec = 1;
1160                         res = 1;
1161                 }
1162         }
1163         spin_unlock(&p->fs->lock);
1164
1165         return res;
1166 }
1167
1168 /*
1169  * Fill the binprm structure from the inode.
1170  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1171  *
1172  * This may be called multiple times for binary chains (scripts for example).
1173  */
1174 int prepare_binprm(struct linux_binprm *bprm)
1175 {
1176         umode_t mode;
1177         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1178         int retval;
1179
1180         mode = inode->i_mode;
1181         if (bprm->file->f_op == NULL)
1182                 return -EACCES;
1183
1184         /* clear any previous set[ug]id data from a previous binary */
1185         bprm->cred->euid = current_euid();
1186         bprm->cred->egid = current_egid();
1187
1188         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1189                 /* Set-uid? */
1190                 if (mode & S_ISUID) {
1191                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1192                         bprm->cred->euid = inode->i_uid;
1193                 }
1194
1195                 /* Set-gid? */
1196                 /*
1197                  * If setgid is set but no group execute bit then this
1198                  * is a candidate for mandatory locking, not a setgid
1199                  * executable.
1200                  */
1201                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1202                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1203                         bprm->cred->egid = inode->i_gid;
1204                 }
1205         }
1206
1207         /* fill in binprm security blob */
1208         retval = security_bprm_set_creds(bprm);
1209         if (retval)
1210                 return retval;
1211         bprm->cred_prepared = 1;
1212
1213         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1214         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1215 }
1216
1217 EXPORT_SYMBOL(prepare_binprm);
1218
1219 /*
1220  * Arguments are '\0' separated strings found at the location bprm->p
1221  * points to; chop off the first by relocating brpm->p to right after
1222  * the first '\0' encountered.
1223  */
1224 int remove_arg_zero(struct linux_binprm *bprm)
1225 {
1226         int ret = 0;
1227         unsigned long offset;
1228         char *kaddr;
1229         struct page *page;
1230
1231         if (!bprm->argc)
1232                 return 0;
1233
1234         do {
1235                 offset = bprm->p & ~PAGE_MASK;
1236                 page = get_arg_page(bprm, bprm->p, 0);
1237                 if (!page) {
1238                         ret = -EFAULT;
1239                         goto out;
1240                 }
1241                 kaddr = kmap_atomic(page, KM_USER0);
1242
1243                 for (; offset < PAGE_SIZE && kaddr[offset];
1244                                 offset++, bprm->p++)
1245                         ;
1246
1247                 kunmap_atomic(kaddr, KM_USER0);
1248                 put_arg_page(page);
1249
1250                 if (offset == PAGE_SIZE)
1251                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1252         } while (offset == PAGE_SIZE);
1253
1254         bprm->p++;
1255         bprm->argc--;
1256         ret = 0;
1257
1258 out:
1259         return ret;
1260 }
1261 EXPORT_SYMBOL(remove_arg_zero);
1262
1263 /*
1264  * cycle the list of binary formats handler, until one recognizes the image
1265  */
1266 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1267 {
1268         unsigned int depth = bprm->recursion_depth;
1269         int try,retval;
1270         struct linux_binfmt *fmt;
1271
1272         retval = security_bprm_check(bprm);
1273         if (retval)
1274                 return retval;
1275
1276         /* kernel module loader fixup */
1277         /* so we don't try to load run modprobe in kernel space. */
1278         set_fs(USER_DS);
1279
1280         retval = audit_bprm(bprm);
1281         if (retval)
1282                 return retval;
1283
1284         retval = -ENOENT;
1285         for (try=0; try<2; try++) {
1286                 read_lock(&binfmt_lock);
1287                 list_for_each_entry(fmt, &formats, lh) {
1288                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1289                         if (!fn)
1290                                 continue;
1291                         if (!try_module_get(fmt->module))
1292                                 continue;
1293                         read_unlock(&binfmt_lock);
1294                         retval = fn(bprm, regs);
1295                         /*
1296                          * Restore the depth counter to its starting value
1297                          * in this call, so we don't have to rely on every
1298                          * load_binary function to restore it on return.
1299                          */
1300                         bprm->recursion_depth = depth;
1301                         if (retval >= 0) {
1302                                 if (depth == 0)
1303                                         tracehook_report_exec(fmt, bprm, regs);
1304                                 put_binfmt(fmt);
1305                                 allow_write_access(bprm->file);
1306                                 if (bprm->file)
1307                                         fput(bprm->file);
1308                                 bprm->file = NULL;
1309                                 current->did_exec = 1;
1310                                 proc_exec_connector(current);
1311                                 return retval;
1312                         }
1313                         read_lock(&binfmt_lock);
1314                         put_binfmt(fmt);
1315                         if (retval != -ENOEXEC || bprm->mm == NULL)
1316                                 break;
1317                         if (!bprm->file) {
1318                                 read_unlock(&binfmt_lock);
1319                                 return retval;
1320                         }
1321                 }
1322                 read_unlock(&binfmt_lock);
1323                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1324                         break;
1325 #ifdef CONFIG_MODULES
1326                 } else {
1327 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1328                         if (printable(bprm->buf[0]) &&
1329                             printable(bprm->buf[1]) &&
1330                             printable(bprm->buf[2]) &&
1331                             printable(bprm->buf[3]))
1332                                 break; /* -ENOEXEC */
1333                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1334 #endif
1335                 }
1336         }
1337         return retval;
1338 }
1339
1340 EXPORT_SYMBOL(search_binary_handler);
1341
1342 /*
1343  * sys_execve() executes a new program.
1344  */
1345 int do_execve(const char * filename,
1346         const char __user *const __user *argv,
1347         const char __user *const __user *envp,
1348         struct pt_regs * regs)
1349 {
1350         struct linux_binprm *bprm;
1351         struct file *file;
1352         struct files_struct *displaced;
1353         bool clear_in_exec;
1354         int retval;
1355
1356         retval = unshare_files(&displaced);
1357         if (retval)
1358                 goto out_ret;
1359
1360         retval = -ENOMEM;
1361         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1362         if (!bprm)
1363                 goto out_files;
1364
1365         retval = prepare_bprm_creds(bprm);
1366         if (retval)
1367                 goto out_free;
1368
1369         retval = check_unsafe_exec(bprm);
1370         if (retval < 0)
1371                 goto out_free;
1372         clear_in_exec = retval;
1373         current->in_execve = 1;
1374
1375         file = open_exec(filename);
1376         retval = PTR_ERR(file);
1377         if (IS_ERR(file))
1378                 goto out_unmark;
1379
1380         sched_exec();
1381
1382         bprm->file = file;
1383         bprm->filename = filename;
1384         bprm->interp = filename;
1385
1386         retval = bprm_mm_init(bprm);
1387         if (retval)
1388                 goto out_file;
1389
1390         bprm->argc = count(argv, MAX_ARG_STRINGS);
1391         if ((retval = bprm->argc) < 0)
1392                 goto out;
1393
1394         bprm->envc = count(envp, MAX_ARG_STRINGS);
1395         if ((retval = bprm->envc) < 0)
1396                 goto out;
1397
1398         retval = prepare_binprm(bprm);
1399         if (retval < 0)
1400                 goto out;
1401
1402         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1403         if (retval < 0)
1404                 goto out;
1405
1406         bprm->exec = bprm->p;
1407         retval = copy_strings(bprm->envc, envp, bprm);
1408         if (retval < 0)
1409                 goto out;
1410
1411         retval = copy_strings(bprm->argc, argv, bprm);
1412         if (retval < 0)
1413                 goto out;
1414
1415         retval = search_binary_handler(bprm,regs);
1416         if (retval < 0)
1417                 goto out;
1418
1419         /* execve succeeded */
1420         current->fs->in_exec = 0;
1421         current->in_execve = 0;
1422         acct_update_integrals(current);
1423         free_bprm(bprm);
1424         if (displaced)
1425                 put_files_struct(displaced);
1426         return retval;
1427
1428 out:
1429         if (bprm->mm)
1430                 mmput (bprm->mm);
1431
1432 out_file:
1433         if (bprm->file) {
1434                 allow_write_access(bprm->file);
1435                 fput(bprm->file);
1436         }
1437
1438 out_unmark:
1439         if (clear_in_exec)
1440                 current->fs->in_exec = 0;
1441         current->in_execve = 0;
1442
1443 out_free:
1444         free_bprm(bprm);
1445
1446 out_files:
1447         if (displaced)
1448                 reset_files_struct(displaced);
1449 out_ret:
1450         return retval;
1451 }
1452
1453 void set_binfmt(struct linux_binfmt *new)
1454 {
1455         struct mm_struct *mm = current->mm;
1456
1457         if (mm->binfmt)
1458                 module_put(mm->binfmt->module);
1459
1460         mm->binfmt = new;
1461         if (new)
1462                 __module_get(new->module);
1463 }
1464
1465 EXPORT_SYMBOL(set_binfmt);
1466
1467 static int expand_corename(struct core_name *cn)
1468 {
1469         char *old_corename = cn->corename;
1470
1471         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1472         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1473
1474         if (!cn->corename) {
1475                 kfree(old_corename);
1476                 return -ENOMEM;
1477         }
1478
1479         return 0;
1480 }
1481
1482 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1483 {
1484         char *cur;
1485         int need;
1486         int ret;
1487         va_list arg;
1488
1489         va_start(arg, fmt);
1490         need = vsnprintf(NULL, 0, fmt, arg);
1491         va_end(arg);
1492
1493         if (likely(need < cn->size - cn->used - 1))
1494                 goto out_printf;
1495
1496         ret = expand_corename(cn);
1497         if (ret)
1498                 goto expand_fail;
1499
1500 out_printf:
1501         cur = cn->corename + cn->used;
1502         va_start(arg, fmt);
1503         vsnprintf(cur, need + 1, fmt, arg);
1504         va_end(arg);
1505         cn->used += need;
1506         return 0;
1507
1508 expand_fail:
1509         return ret;
1510 }
1511
1512 /* format_corename will inspect the pattern parameter, and output a
1513  * name into corename, which must have space for at least
1514  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1515  */
1516 static int format_corename(struct core_name *cn, long signr)
1517 {
1518         const struct cred *cred = current_cred();
1519         const char *pat_ptr = core_pattern;
1520         int ispipe = (*pat_ptr == '|');
1521         int pid_in_pattern = 0;
1522         int err = 0;
1523
1524         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1525         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1526         cn->used = 0;
1527
1528         if (!cn->corename)
1529                 return -ENOMEM;
1530
1531         /* Repeat as long as we have more pattern to process and more output
1532            space */
1533         while (*pat_ptr) {
1534                 if (*pat_ptr != '%') {
1535                         if (*pat_ptr == 0)
1536                                 goto out;
1537                         err = cn_printf(cn, "%c", *pat_ptr++);
1538                 } else {
1539                         switch (*++pat_ptr) {
1540                         /* single % at the end, drop that */
1541                         case 0:
1542                                 goto out;
1543                         /* Double percent, output one percent */
1544                         case '%':
1545                                 err = cn_printf(cn, "%c", '%');
1546                                 break;
1547                         /* pid */
1548                         case 'p':
1549                                 pid_in_pattern = 1;
1550                                 err = cn_printf(cn, "%d",
1551                                               task_tgid_vnr(current));
1552                                 break;
1553                         /* uid */
1554                         case 'u':
1555                                 err = cn_printf(cn, "%d", cred->uid);
1556                                 break;
1557                         /* gid */
1558                         case 'g':
1559                                 err = cn_printf(cn, "%d", cred->gid);
1560                                 break;
1561                         /* signal that caused the coredump */
1562                         case 's':
1563                                 err = cn_printf(cn, "%ld", signr);
1564                                 break;
1565                         /* UNIX time of coredump */
1566                         case 't': {
1567                                 struct timeval tv;
1568                                 do_gettimeofday(&tv);
1569                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1570                                 break;
1571                         }
1572                         /* hostname */
1573                         case 'h':
1574                                 down_read(&uts_sem);
1575                                 err = cn_printf(cn, "%s",
1576                                               utsname()->nodename);
1577                                 up_read(&uts_sem);
1578                                 break;
1579                         /* executable */
1580                         case 'e':
1581                                 err = cn_printf(cn, "%s", current->comm);
1582                                 break;
1583                         /* core limit size */
1584                         case 'c':
1585                                 err = cn_printf(cn, "%lu",
1586                                               rlimit(RLIMIT_CORE));
1587                                 break;
1588                         default:
1589                                 break;
1590                         }
1591                         ++pat_ptr;
1592                 }
1593
1594                 if (err)
1595                         return err;
1596         }
1597
1598         /* Backward compatibility with core_uses_pid:
1599          *
1600          * If core_pattern does not include a %p (as is the default)
1601          * and core_uses_pid is set, then .%pid will be appended to
1602          * the filename. Do not do this for piped commands. */
1603         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1604                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1605                 if (err)
1606                         return err;
1607         }
1608 out:
1609         return ispipe;
1610 }
1611
1612 static int zap_process(struct task_struct *start, int exit_code)
1613 {
1614         struct task_struct *t;
1615         int nr = 0;
1616
1617         start->signal->flags = SIGNAL_GROUP_EXIT;
1618         start->signal->group_exit_code = exit_code;
1619         start->signal->group_stop_count = 0;
1620
1621         t = start;
1622         do {
1623                 if (t != current && t->mm) {
1624                         sigaddset(&t->pending.signal, SIGKILL);
1625                         signal_wake_up(t, 1);
1626                         nr++;
1627                 }
1628         } while_each_thread(start, t);
1629
1630         return nr;
1631 }
1632
1633 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1634                                 struct core_state *core_state, int exit_code)
1635 {
1636         struct task_struct *g, *p;
1637         unsigned long flags;
1638         int nr = -EAGAIN;
1639
1640         spin_lock_irq(&tsk->sighand->siglock);
1641         if (!signal_group_exit(tsk->signal)) {
1642                 mm->core_state = core_state;
1643                 nr = zap_process(tsk, exit_code);
1644         }
1645         spin_unlock_irq(&tsk->sighand->siglock);
1646         if (unlikely(nr < 0))
1647                 return nr;
1648
1649         if (atomic_read(&mm->mm_users) == nr + 1)
1650                 goto done;
1651         /*
1652          * We should find and kill all tasks which use this mm, and we should
1653          * count them correctly into ->nr_threads. We don't take tasklist
1654          * lock, but this is safe wrt:
1655          *
1656          * fork:
1657          *      None of sub-threads can fork after zap_process(leader). All
1658          *      processes which were created before this point should be
1659          *      visible to zap_threads() because copy_process() adds the new
1660          *      process to the tail of init_task.tasks list, and lock/unlock
1661          *      of ->siglock provides a memory barrier.
1662          *
1663          * do_exit:
1664          *      The caller holds mm->mmap_sem. This means that the task which
1665          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1666          *      its ->mm.
1667          *
1668          * de_thread:
1669          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1670          *      we must see either old or new leader, this does not matter.
1671          *      However, it can change p->sighand, so lock_task_sighand(p)
1672          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1673          *      it can't fail.
1674          *
1675          *      Note also that "g" can be the old leader with ->mm == NULL
1676          *      and already unhashed and thus removed from ->thread_group.
1677          *      This is OK, __unhash_process()->list_del_rcu() does not
1678          *      clear the ->next pointer, we will find the new leader via
1679          *      next_thread().
1680          */
1681         rcu_read_lock();
1682         for_each_process(g) {
1683                 if (g == tsk->group_leader)
1684                         continue;
1685                 if (g->flags & PF_KTHREAD)
1686                         continue;
1687                 p = g;
1688                 do {
1689                         if (p->mm) {
1690                                 if (unlikely(p->mm == mm)) {
1691                                         lock_task_sighand(p, &flags);
1692                                         nr += zap_process(p, exit_code);
1693                                         unlock_task_sighand(p, &flags);
1694                                 }
1695                                 break;
1696                         }
1697                 } while_each_thread(g, p);
1698         }
1699         rcu_read_unlock();
1700 done:
1701         atomic_set(&core_state->nr_threads, nr);
1702         return nr;
1703 }
1704
1705 static int coredump_wait(int exit_code, struct core_state *core_state)
1706 {
1707         struct task_struct *tsk = current;
1708         struct mm_struct *mm = tsk->mm;
1709         struct completion *vfork_done;
1710         int core_waiters = -EBUSY;
1711
1712         init_completion(&core_state->startup);
1713         core_state->dumper.task = tsk;
1714         core_state->dumper.next = NULL;
1715
1716         down_write(&mm->mmap_sem);
1717         if (!mm->core_state)
1718                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1719         up_write(&mm->mmap_sem);
1720
1721         if (unlikely(core_waiters < 0))
1722                 goto fail;
1723
1724         /*
1725          * Make sure nobody is waiting for us to release the VM,
1726          * otherwise we can deadlock when we wait on each other
1727          */
1728         vfork_done = tsk->vfork_done;
1729         if (vfork_done) {
1730                 tsk->vfork_done = NULL;
1731                 complete(vfork_done);
1732         }
1733
1734         if (core_waiters)
1735                 wait_for_completion(&core_state->startup);
1736 fail:
1737         return core_waiters;
1738 }
1739
1740 static void coredump_finish(struct mm_struct *mm)
1741 {
1742         struct core_thread *curr, *next;
1743         struct task_struct *task;
1744
1745         next = mm->core_state->dumper.next;
1746         while ((curr = next) != NULL) {
1747                 next = curr->next;
1748                 task = curr->task;
1749                 /*
1750                  * see exit_mm(), curr->task must not see
1751                  * ->task == NULL before we read ->next.
1752                  */
1753                 smp_mb();
1754                 curr->task = NULL;
1755                 wake_up_process(task);
1756         }
1757
1758         mm->core_state = NULL;
1759 }
1760
1761 /*
1762  * set_dumpable converts traditional three-value dumpable to two flags and
1763  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1764  * these bits are not changed atomically.  So get_dumpable can observe the
1765  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1766  * return either old dumpable or new one by paying attention to the order of
1767  * modifying the bits.
1768  *
1769  * dumpable |   mm->flags (binary)
1770  * old  new | initial interim  final
1771  * ---------+-----------------------
1772  *  0    1  |   00      01      01
1773  *  0    2  |   00      10(*)   11
1774  *  1    0  |   01      00      00
1775  *  1    2  |   01      11      11
1776  *  2    0  |   11      10(*)   00
1777  *  2    1  |   11      11      01
1778  *
1779  * (*) get_dumpable regards interim value of 10 as 11.
1780  */
1781 void set_dumpable(struct mm_struct *mm, int value)
1782 {
1783         switch (value) {
1784         case 0:
1785                 clear_bit(MMF_DUMPABLE, &mm->flags);
1786                 smp_wmb();
1787                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1788                 break;
1789         case 1:
1790                 set_bit(MMF_DUMPABLE, &mm->flags);
1791                 smp_wmb();
1792                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1793                 break;
1794         case 2:
1795                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1796                 smp_wmb();
1797                 set_bit(MMF_DUMPABLE, &mm->flags);
1798                 break;
1799         }
1800 }
1801
1802 static int __get_dumpable(unsigned long mm_flags)
1803 {
1804         int ret;
1805
1806         ret = mm_flags & MMF_DUMPABLE_MASK;
1807         return (ret >= 2) ? 2 : ret;
1808 }
1809
1810 int get_dumpable(struct mm_struct *mm)
1811 {
1812         return __get_dumpable(mm->flags);
1813 }
1814
1815 static void wait_for_dump_helpers(struct file *file)
1816 {
1817         struct pipe_inode_info *pipe;
1818
1819         pipe = file->f_path.dentry->d_inode->i_pipe;
1820
1821         pipe_lock(pipe);
1822         pipe->readers++;
1823         pipe->writers--;
1824
1825         while ((pipe->readers > 1) && (!signal_pending(current))) {
1826                 wake_up_interruptible_sync(&pipe->wait);
1827                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1828                 pipe_wait(pipe);
1829         }
1830
1831         pipe->readers--;
1832         pipe->writers++;
1833         pipe_unlock(pipe);
1834
1835 }
1836
1837
1838 /*
1839  * uhm_pipe_setup
1840  * helper function to customize the process used
1841  * to collect the core in userspace.  Specifically
1842  * it sets up a pipe and installs it as fd 0 (stdin)
1843  * for the process.  Returns 0 on success, or
1844  * PTR_ERR on failure.
1845  * Note that it also sets the core limit to 1.  This
1846  * is a special value that we use to trap recursive
1847  * core dumps
1848  */
1849 static int umh_pipe_setup(struct subprocess_info *info)
1850 {
1851         struct file *rp, *wp;
1852         struct fdtable *fdt;
1853         struct coredump_params *cp = (struct coredump_params *)info->data;
1854         struct files_struct *cf = current->files;
1855
1856         wp = create_write_pipe(0);
1857         if (IS_ERR(wp))
1858                 return PTR_ERR(wp);
1859
1860         rp = create_read_pipe(wp, 0);
1861         if (IS_ERR(rp)) {
1862                 free_write_pipe(wp);
1863                 return PTR_ERR(rp);
1864         }
1865
1866         cp->file = wp;
1867
1868         sys_close(0);
1869         fd_install(0, rp);
1870         spin_lock(&cf->file_lock);
1871         fdt = files_fdtable(cf);
1872         FD_SET(0, fdt->open_fds);
1873         FD_CLR(0, fdt->close_on_exec);
1874         spin_unlock(&cf->file_lock);
1875
1876         /* and disallow core files too */
1877         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1878
1879         return 0;
1880 }
1881
1882 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1883 {
1884         struct core_state core_state;
1885         struct core_name cn;
1886         struct mm_struct *mm = current->mm;
1887         struct linux_binfmt * binfmt;
1888         const struct cred *old_cred;
1889         struct cred *cred;
1890         int retval = 0;
1891         int flag = 0;
1892         int ispipe;
1893         static atomic_t core_dump_count = ATOMIC_INIT(0);
1894         struct coredump_params cprm = {
1895                 .signr = signr,
1896                 .regs = regs,
1897                 .limit = rlimit(RLIMIT_CORE),
1898                 /*
1899                  * We must use the same mm->flags while dumping core to avoid
1900                  * inconsistency of bit flags, since this flag is not protected
1901                  * by any locks.
1902                  */
1903                 .mm_flags = mm->flags,
1904         };
1905
1906         audit_core_dumps(signr);
1907
1908         binfmt = mm->binfmt;
1909         if (!binfmt || !binfmt->core_dump)
1910                 goto fail;
1911         if (!__get_dumpable(cprm.mm_flags))
1912                 goto fail;
1913
1914         cred = prepare_creds();
1915         if (!cred)
1916                 goto fail;
1917         /*
1918          *      We cannot trust fsuid as being the "true" uid of the
1919          *      process nor do we know its entire history. We only know it
1920          *      was tainted so we dump it as root in mode 2.
1921          */
1922         if (__get_dumpable(cprm.mm_flags) == 2) {
1923                 /* Setuid core dump mode */
1924                 flag = O_EXCL;          /* Stop rewrite attacks */
1925                 cred->fsuid = 0;        /* Dump root private */
1926         }
1927
1928         retval = coredump_wait(exit_code, &core_state);
1929         if (retval < 0)
1930                 goto fail_creds;
1931
1932         old_cred = override_creds(cred);
1933
1934         /*
1935          * Clear any false indication of pending signals that might
1936          * be seen by the filesystem code called to write the core file.
1937          */
1938         clear_thread_flag(TIF_SIGPENDING);
1939
1940         ispipe = format_corename(&cn, signr);
1941
1942         if (ispipe == -ENOMEM) {
1943                 printk(KERN_WARNING "format_corename failed\n");
1944                 printk(KERN_WARNING "Aborting core\n");
1945                 goto fail_corename;
1946         }
1947
1948         if (ispipe) {
1949                 int dump_count;
1950                 char **helper_argv;
1951
1952                 if (cprm.limit == 1) {
1953                         /*
1954                          * Normally core limits are irrelevant to pipes, since
1955                          * we're not writing to the file system, but we use
1956                          * cprm.limit of 1 here as a speacial value. Any
1957                          * non-1 limit gets set to RLIM_INFINITY below, but
1958                          * a limit of 0 skips the dump.  This is a consistent
1959                          * way to catch recursive crashes.  We can still crash
1960                          * if the core_pattern binary sets RLIM_CORE =  !1
1961                          * but it runs as root, and can do lots of stupid things
1962                          * Note that we use task_tgid_vnr here to grab the pid
1963                          * of the process group leader.  That way we get the
1964                          * right pid if a thread in a multi-threaded
1965                          * core_pattern process dies.
1966                          */
1967                         printk(KERN_WARNING
1968                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1969                                 task_tgid_vnr(current), current->comm);
1970                         printk(KERN_WARNING "Aborting core\n");
1971                         goto fail_unlock;
1972                 }
1973                 cprm.limit = RLIM_INFINITY;
1974
1975                 dump_count = atomic_inc_return(&core_dump_count);
1976                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1977                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1978                                task_tgid_vnr(current), current->comm);
1979                         printk(KERN_WARNING "Skipping core dump\n");
1980                         goto fail_dropcount;
1981                 }
1982
1983                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
1984                 if (!helper_argv) {
1985                         printk(KERN_WARNING "%s failed to allocate memory\n",
1986                                __func__);
1987                         goto fail_dropcount;
1988                 }
1989
1990                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1991                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1992                                         NULL, &cprm);
1993                 argv_free(helper_argv);
1994                 if (retval) {
1995                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1996                                cn.corename);
1997                         goto close_fail;
1998                 }
1999         } else {
2000                 struct inode *inode;
2001
2002                 if (cprm.limit < binfmt->min_coredump)
2003                         goto fail_unlock;
2004
2005                 cprm.file = filp_open(cn.corename,
2006                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2007                                  0600);
2008                 if (IS_ERR(cprm.file))
2009                         goto fail_unlock;
2010
2011                 inode = cprm.file->f_path.dentry->d_inode;
2012                 if (inode->i_nlink > 1)
2013                         goto close_fail;
2014                 if (d_unhashed(cprm.file->f_path.dentry))
2015                         goto close_fail;
2016                 /*
2017                  * AK: actually i see no reason to not allow this for named
2018                  * pipes etc, but keep the previous behaviour for now.
2019                  */
2020                 if (!S_ISREG(inode->i_mode))
2021                         goto close_fail;
2022                 /*
2023                  * Dont allow local users get cute and trick others to coredump
2024                  * into their pre-created files.
2025                  */
2026                 if (inode->i_uid != current_fsuid())
2027                         goto close_fail;
2028                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2029                         goto close_fail;
2030                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2031                         goto close_fail;
2032         }
2033
2034         retval = binfmt->core_dump(&cprm);
2035         if (retval)
2036                 current->signal->group_exit_code |= 0x80;
2037
2038         if (ispipe && core_pipe_limit)
2039                 wait_for_dump_helpers(cprm.file);
2040 close_fail:
2041         if (cprm.file)
2042                 filp_close(cprm.file, NULL);
2043 fail_dropcount:
2044         if (ispipe)
2045                 atomic_dec(&core_dump_count);
2046 fail_unlock:
2047         kfree(cn.corename);
2048 fail_corename:
2049         coredump_finish(mm);
2050         revert_creds(old_cred);
2051 fail_creds:
2052         put_cred(cred);
2053 fail:
2054         return;
2055 }
2056
2057 /*
2058  * Core dumping helper functions.  These are the only things you should
2059  * do on a core-file: use only these functions to write out all the
2060  * necessary info.
2061  */
2062 int dump_write(struct file *file, const void *addr, int nr)
2063 {
2064         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2065 }
2066 EXPORT_SYMBOL(dump_write);
2067
2068 int dump_seek(struct file *file, loff_t off)
2069 {
2070         int ret = 1;
2071
2072         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2073                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2074                         return 0;
2075         } else {
2076                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2077
2078                 if (!buf)
2079                         return 0;
2080                 while (off > 0) {
2081                         unsigned long n = off;
2082
2083                         if (n > PAGE_SIZE)
2084                                 n = PAGE_SIZE;
2085                         if (!dump_write(file, buf, n)) {
2086                                 ret = 0;
2087                                 break;
2088                         }
2089                         off -= n;
2090                 }
2091                 free_page((unsigned long)buf);
2092         }
2093         return ret;
2094 }
2095 EXPORT_SYMBOL(dump_seek);