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         static const struct open_flags uselib_flags = {
 119                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 120                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 121                 .intent = LOOKUP_OPEN
 122         };
 123
 124         if (IS_ERR(tmp))
 125                 goto out;
 126
 127         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 128         putname(tmp);
 129         error = PTR_ERR(file);
 130         if (IS_ERR(file))
 131                 goto out;
 132
 133         error = -EINVAL;
 134         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 135                 goto exit;
 136
 137         error = -EACCES;
 138         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 139                 goto exit;
 140
 141         fsnotify_open(file);
 142
 143         error = -ENOEXEC;
 144         if(file->f_op) {
 145                 struct linux_binfmt * fmt;
 146
 147                 read_lock(&binfmt_lock);
 148                 list_for_each_entry(fmt, &formats, lh) {
 149                         if (!fmt->load_shlib)
 150                                 continue;
 151                         if (!try_module_get(fmt->module))
 152                                 continue;
 153                         read_unlock(&binfmt_lock);
 154                         error = fmt->load_shlib(file);
 155                         read_lock(&binfmt_lock);
 156                         put_binfmt(fmt);
 157                         if (error != -ENOEXEC)
 158                                 break;
 159                 }
 160                 read_unlock(&binfmt_lock);
 161         }
 162 exit:
 163         fput(file);
 164 out:
 165         return error;
 166 }
 167
 168 #ifdef CONFIG_MMU
 169
 170 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 171 {
 172         struct mm_struct *mm = current->mm;
 173         long diff = (long)(pages - bprm->vma_pages);
 174
 175         if (!mm || !diff)
 176                 return;
 177
 178         bprm->vma_pages = pages;
 179
 180 #ifdef SPLIT_RSS_COUNTING
 181         add_mm_counter(mm, MM_ANONPAGES, diff);
 182 #else
 183         spin_lock(&mm->page_table_lock);
 184         add_mm_counter(mm, MM_ANONPAGES, diff);
 185         spin_unlock(&mm->page_table_lock);
 186 #endif
 187 }
 188
 189 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 190                 int write)
 191 {
 192         struct page *page;
 193         int ret;
 194
 195 #ifdef CONFIG_STACK_GROWSUP
 196         if (write) {
 197                 ret = expand_stack_downwards(bprm->vma, pos);
 198                 if (ret < 0)
 199                         return NULL;
 200         }
 201 #endif
 202         ret = get_user_pages(current, bprm->mm, pos,
 203                         1, write, 1, &page, NULL);
 204         if (ret <= 0)
 205                 return NULL;
 206
 207         if (write) {
 208                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 209                 struct rlimit *rlim;
 210
 211                 acct_arg_size(bprm, size / PAGE_SIZE);
 212
 213                 /*
 214                  * We've historically supported up to 32 pages (ARG_MAX)
 215                  * of argument strings even with small stacks
 216                  */
 217                 if (size <= ARG_MAX)
 218                         return page;
 219
 220                 /*
 221                  * Limit to 1/4-th the stack size for the argv+env strings.
 222                  * This ensures that:
 223                  *  - the remaining binfmt code will not run out of stack space,
 224                  *  - the program will have a reasonable amount of stack left
 225                  *    to work from.
 226                  */
 227                 rlim = current->signal->rlim;
 228                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 229                         put_page(page);
 230                         return NULL;
 231                 }
 232         }
 233
 234         return page;
 235 }
 236
 237 static void put_arg_page(struct page *page)
 238 {
 239         put_page(page);
 240 }
 241
 242 static void free_arg_page(struct linux_binprm *bprm, int i)
 243 {
 244 }
 245
 246 static void free_arg_pages(struct linux_binprm *bprm)
 247 {
 248 }
 249
 250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 251                 struct page *page)
 252 {
 253         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 254 }
 255
 256 static int __bprm_mm_init(struct linux_binprm *bprm)
 257 {
 258         int err;
 259         struct vm_area_struct *vma = NULL;
 260         struct mm_struct *mm = bprm->mm;
 261
 262         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 263         if (!vma)
 264                 return -ENOMEM;
 265
 266         down_write(&mm->mmap_sem);
 267         vma->vm_mm = mm;
 268
 269         /*
 270          * Place the stack at the largest stack address the architecture
 271          * supports. Later, we'll move this to an appropriate place. We don't
 272          * use STACK_TOP because that can depend on attributes which aren't
 273          * configured yet.
 274          */
 275         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 276         vma->vm_end = STACK_TOP_MAX;
 277         vma->vm_start = vma->vm_end - PAGE_SIZE;
 278         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 279         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 280         INIT_LIST_HEAD(&vma->anon_vma_chain);
 281
 282         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 283         if (err)
 284                 goto err;
 285
 286         err = insert_vm_struct(mm, vma);
 287         if (err)
 288                 goto err;
 289
 290         mm->stack_vm = mm->total_vm = 1;
 291         up_write(&mm->mmap_sem);
 292         bprm->p = vma->vm_end - sizeof(void *);
 293         return 0;
 294 err:
 295         up_write(&mm->mmap_sem);
 296         bprm->vma = NULL;
 297         kmem_cache_free(vm_area_cachep, vma);
 298         return err;
 299 }
 300
 301 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 302 {
 303         return len <= MAX_ARG_STRLEN;
 304 }
 305
 306 #else
 307
 308 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 309 {
 310 }
 311
 312 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 313                 int write)
 314 {
 315         struct page *page;
 316
 317         page = bprm->page[pos / PAGE_SIZE];
 318         if (!page && write) {
 319                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 320                 if (!page)
 321                         return NULL;
 322                 bprm->page[pos / PAGE_SIZE] = page;
 323         }
 324
 325         return page;
 326 }
 327
 328 static void put_arg_page(struct page *page)
 329 {
 330 }
 331
 332 static void free_arg_page(struct linux_binprm *bprm, int i)
 333 {
 334         if (bprm->page[i]) {
 335                 __free_page(bprm->page[i]);
 336                 bprm->page[i] = NULL;
 337         }
 338 }
 339
 340 static void free_arg_pages(struct linux_binprm *bprm)
 341 {
 342         int i;
 343
 344         for (i = 0; i < MAX_ARG_PAGES; i++)
 345                 free_arg_page(bprm, i);
 346 }
 347
 348 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 349                 struct page *page)
 350 {
 351 }
 352
 353 static int __bprm_mm_init(struct linux_binprm *bprm)
 354 {
 355         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 356         return 0;
 357 }
 358
 359 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 360 {
 361         return len <= bprm->p;
 362 }
 363
 364 #endif /* CONFIG_MMU */
 365
 366 /*
 367  * Create a new mm_struct and populate it with a temporary stack
 368  * vm_area_struct.  We don't have enough context at this point to set the stack
 369  * flags, permissions, and offset, so we use temporary values.  We'll update
 370  * them later in setup_arg_pages().
 371  */
 372 int bprm_mm_init(struct linux_binprm *bprm)
 373 {
 374         int err;
 375         struct mm_struct *mm = NULL;
 376
 377         bprm->mm = mm = mm_alloc();
 378         err = -ENOMEM;
 379         if (!mm)
 380                 goto err;
 381
 382         err = init_new_context(current, mm);
 383         if (err)
 384                 goto err;
 385
 386         err = __bprm_mm_init(bprm);
 387         if (err)
 388                 goto err;
 389
 390         return 0;
 391
 392 err:
 393         if (mm) {
 394                 bprm->mm = NULL;
 395                 mmdrop(mm);
 396         }
 397
 398         return err;
 399 }
 400
 401 /*
 402  * count() counts the number of strings in array ARGV.
 403  */
 404 static int count(const char __user * const __user * argv, int max)
 405 {
 406         int i = 0;
 407
 408         if (argv != NULL) {
 409                 for (;;) {
 410                         const char __user * p;
 411
 412                         if (get_user(p, argv))
 413                                 return -EFAULT;
 414                         if (!p)
 415                                 break;
 416                         argv++;
 417                         if (i++ >= max)
 418                                 return -E2BIG;
 419
 420                         if (fatal_signal_pending(current))
 421                                 return -ERESTARTNOHAND;
 422                         cond_resched();
 423                 }
 424         }
 425         return i;
 426 }
 427
 428 /*
 429  * 'copy_strings()' copies argument/environment strings from the old
 430  * processes's memory to the new process's stack.  The call to get_user_pages()
 431  * ensures the destination page is created and not swapped out.
 432  */
 433 static int copy_strings(int argc, const char __user *const __user *argv,
 434                         struct linux_binprm *bprm)
 435 {
 436         struct page *kmapped_page = NULL;
 437         char *kaddr = NULL;
 438         unsigned long kpos = 0;
 439         int ret;
 440
 441         while (argc-- > 0) {
 442                 const char __user *str;
 443                 int len;
 444                 unsigned long pos;
 445
 446                 if (get_user(str, argv+argc) ||
 447                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 448                         ret = -EFAULT;
 449                         goto out;
 450                 }
 451
 452                 if (!valid_arg_len(bprm, len)) {
 453                         ret = -E2BIG;
 454                         goto out;
 455                 }
 456
 457                 /* We're going to work our way backwords. */
 458                 pos = bprm->p;
 459                 str += len;
 460                 bprm->p -= len;
 461
 462                 while (len > 0) {
 463                         int offset, bytes_to_copy;
 464
 465                         if (fatal_signal_pending(current)) {
 466                                 ret = -ERESTARTNOHAND;
 467                                 goto out;
 468                         }
 469                         cond_resched();
 470
 471                         offset = pos % PAGE_SIZE;
 472                         if (offset == 0)
 473                                 offset = PAGE_SIZE;
 474
 475                         bytes_to_copy = offset;
 476                         if (bytes_to_copy > len)
 477                                 bytes_to_copy = len;
 478
 479                         offset -= bytes_to_copy;
 480                         pos -= bytes_to_copy;
 481                         str -= bytes_to_copy;
 482                         len -= bytes_to_copy;
 483
 484                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 485                                 struct page *page;
 486
 487                                 page = get_arg_page(bprm, pos, 1);
 488                                 if (!page) {
 489                                         ret = -E2BIG;
 490                                         goto out;
 491                                 }
 492
 493                                 if (kmapped_page) {
 494                                         flush_kernel_dcache_page(kmapped_page);
 495                                         kunmap(kmapped_page);
 496                                         put_arg_page(kmapped_page);
 497                                 }
 498                                 kmapped_page = page;
 499                                 kaddr = kmap(kmapped_page);
 500                                 kpos = pos & PAGE_MASK;
 501                                 flush_arg_page(bprm, kpos, kmapped_page);
 502                         }
 503                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 504                                 ret = -EFAULT;
 505                                 goto out;
 506                         }
 507                 }
 508         }
 509         ret = 0;
 510 out:
 511         if (kmapped_page) {
 512                 flush_kernel_dcache_page(kmapped_page);
 513                 kunmap(kmapped_page);
 514                 put_arg_page(kmapped_page);
 515         }
 516         return ret;
 517 }
 518
 519 /*
 520  * Like copy_strings, but get argv and its values from kernel memory.
 521  */
 522 int copy_strings_kernel(int argc, const char *const *argv,
 523                         struct linux_binprm *bprm)
 524 {
 525         int r;
 526         mm_segment_t oldfs = get_fs();
 527         set_fs(KERNEL_DS);
 528         r = copy_strings(argc, (const char __user *const  __user *)argv, bprm);
 529         set_fs(oldfs);
 530         return r;
 531 }
 532 EXPORT_SYMBOL(copy_strings_kernel);
 533
 534 #ifdef CONFIG_MMU
 535
 536 /*
 537  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 538  * the binfmt code determines where the new stack should reside, we shift it to
 539  * its final location.  The process proceeds as follows:
 540  *
 541  * 1) Use shift to calculate the new vma endpoints.
 542  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 543  *    arguments passed to subsequent functions are consistent.
 544  * 3) Move vma's page tables to the new range.
 545  * 4) Free up any cleared pgd range.
 546  * 5) Shrink the vma to cover only the new range.
 547  */
 548 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 549 {
 550         struct mm_struct *mm = vma->vm_mm;
 551         unsigned long old_start = vma->vm_start;
 552         unsigned long old_end = vma->vm_end;
 553         unsigned long length = old_end - old_start;
 554         unsigned long new_start = old_start - shift;
 555         unsigned long new_end = old_end - shift;
 556         struct mmu_gather *tlb;
 557
 558         BUG_ON(new_start > new_end);
 559
 560         /*
 561          * ensure there are no vmas between where we want to go
 562          * and where we are
 563          */
 564         if (vma != find_vma(mm, new_start))
 565                 return -EFAULT;
 566
 567         /*
 568          * cover the whole range: [new_start, old_end)
 569          */
 570         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 571                 return -ENOMEM;
 572
 573         /*
 574          * move the page tables downwards, on failure we rely on
 575          * process cleanup to remove whatever mess we made.
 576          */
 577         if (length != move_page_tables(vma, old_start,
 578                                        vma, new_start, length))
 579                 return -ENOMEM;
 580
 581         lru_add_drain();
 582         tlb = tlb_gather_mmu(mm, 0);
 583         if (new_end > old_start) {
 584                 /*
 585                  * when the old and new regions overlap clear from new_end.
 586                  */
 587                 free_pgd_range(tlb, new_end, old_end, new_end,
 588                         vma->vm_next ? vma->vm_next->vm_start : 0);
 589         } else {
 590                 /*
 591                  * otherwise, clean from old_start; this is done to not touch
 592                  * the address space in [new_end, old_start) some architectures
 593                  * have constraints on va-space that make this illegal (IA64) -
 594                  * for the others its just a little faster.
 595                  */
 596                 free_pgd_range(tlb, old_start, old_end, new_end,
 597                         vma->vm_next ? vma->vm_next->vm_start : 0);
 598         }
 599         tlb_finish_mmu(tlb, new_end, old_end);
 600
 601         /*
 602          * Shrink the vma to just the new range.  Always succeeds.
 603          */
 604         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 605
 606         return 0;
 607 }
 608
 609 /*
 610  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 611  * the stack is optionally relocated, and some extra space is added.
 612  */
 613 int setup_arg_pages(struct linux_binprm *bprm,
 614                     unsigned long stack_top,
 615                     int executable_stack)
 616 {
 617         unsigned long ret;
 618         unsigned long stack_shift;
 619         struct mm_struct *mm = current->mm;
 620         struct vm_area_struct *vma = bprm->vma;
 621         struct vm_area_struct *prev = NULL;
 622         unsigned long vm_flags;
 623         unsigned long stack_base;
 624         unsigned long stack_size;
 625         unsigned long stack_expand;
 626         unsigned long rlim_stack;
 627
 628 #ifdef CONFIG_STACK_GROWSUP
 629         /* Limit stack size to 1GB */
 630         stack_base = rlimit_max(RLIMIT_STACK);
 631         if (stack_base > (1 << 30))
 632                 stack_base = 1 << 30;
 633
 634         /* Make sure we didn't let the argument array grow too large. */
 635         if (vma->vm_end - vma->vm_start > stack_base)
 636                 return -ENOMEM;
 637
 638         stack_base = PAGE_ALIGN(stack_top - stack_base);
 639
 640         stack_shift = vma->vm_start - stack_base;
 641         mm->arg_start = bprm->p - stack_shift;
 642         bprm->p = vma->vm_end - stack_shift;
 643 #else
 644         stack_top = arch_align_stack(stack_top);
 645         stack_top = PAGE_ALIGN(stack_top);
 646
 647         if (unlikely(stack_top < mmap_min_addr) ||
 648             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 649                 return -ENOMEM;
 650
 651         stack_shift = vma->vm_end - stack_top;
 652
 653         bprm->p -= stack_shift;
 654         mm->arg_start = bprm->p;
 655 #endif
 656
 657         if (bprm->loader)
 658                 bprm->loader -= stack_shift;
 659         bprm->exec -= stack_shift;
 660
 661         down_write(&mm->mmap_sem);
 662         vm_flags = VM_STACK_FLAGS;
 663
 664         /*
 665          * Adjust stack execute permissions; explicitly enable for
 666          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 667          * (arch default) otherwise.
 668          */
 669         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 670                 vm_flags |= VM_EXEC;
 671         else if (executable_stack == EXSTACK_DISABLE_X)
 672                 vm_flags &= ~VM_EXEC;
 673         vm_flags |= mm->def_flags;
 674         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 675
 676         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 677                         vm_flags);
 678         if (ret)
 679                 goto out_unlock;
 680         BUG_ON(prev != vma);
 681
 682         /* Move stack pages down in memory. */
 683         if (stack_shift) {
 684                 ret = shift_arg_pages(vma, stack_shift);
 685                 if (ret)
 686                         goto out_unlock;
 687         }
 688
 689         /* mprotect_fixup is overkill to remove the temporary stack flags */
 690         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 691
 692         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 693         stack_size = vma->vm_end - vma->vm_start;
 694         /*
 695          * Align this down to a page boundary as expand_stack
 696          * will align it up.
 697          */
 698         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 699 #ifdef CONFIG_STACK_GROWSUP
 700         if (stack_size + stack_expand > rlim_stack)
 701                 stack_base = vma->vm_start + rlim_stack;
 702         else
 703                 stack_base = vma->vm_end + stack_expand;
 704 #else
 705         if (stack_size + stack_expand > rlim_stack)
 706                 stack_base = vma->vm_end - rlim_stack;
 707         else
 708                 stack_base = vma->vm_start - stack_expand;
 709 #endif
 710         current->mm->start_stack = bprm->p;
 711         ret = expand_stack(vma, stack_base);
 712         if (ret)
 713                 ret = -EFAULT;
 714
 715 out_unlock:
 716         up_write(&mm->mmap_sem);
 717         return ret;
 718 }
 719 EXPORT_SYMBOL(setup_arg_pages);
 720
 721 #endif /* CONFIG_MMU */
 722
 723 struct file *open_exec(const char *name)
 724 {
 725         struct file *file;
 726         int err;
 727         static const struct open_flags open_exec_flags = {
 728                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 729                 .acc_mode = MAY_EXEC | MAY_OPEN,
 730                 .intent = LOOKUP_OPEN
 731         };
 732
 733         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 734         if (IS_ERR(file))
 735                 goto out;
 736
 737         err = -EACCES;
 738         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 739                 goto exit;
 740
 741         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 742                 goto exit;
 743
 744         fsnotify_open(file);
 745
 746         err = deny_write_access(file);
 747         if (err)
 748                 goto exit;
 749
 750 out:
 751         return file;
 752
 753 exit:
 754         fput(file);
 755         return ERR_PTR(err);
 756 }
 757 EXPORT_SYMBOL(open_exec);
 758
 759 int kernel_read(struct file *file, loff_t offset,
 760                 char *addr, unsigned long count)
 761 {
 762         mm_segment_t old_fs;
 763         loff_t pos = offset;
 764         int result;
 765
 766         old_fs = get_fs();
 767         set_fs(get_ds());
 768         /* The cast to a user pointer is valid due to the set_fs() */
 769         result = vfs_read(file, (void __user *)addr, count, &pos);
 770         set_fs(old_fs);
 771         return result;
 772 }
 773
 774 EXPORT_SYMBOL(kernel_read);
 775
 776 static int exec_mmap(struct mm_struct *mm)
 777 {
 778         struct task_struct *tsk;
 779         struct mm_struct * old_mm, *active_mm;
 780
 781         /* Notify parent that we're no longer interested in the old VM */
 782         tsk = current;
 783         old_mm = current->mm;
 784         sync_mm_rss(tsk, old_mm);
 785         mm_release(tsk, old_mm);
 786
 787         if (old_mm) {
 788                 /*
 789                  * Make sure that if there is a core dump in progress
 790                  * for the old mm, we get out and die instead of going
 791                  * through with the exec.  We must hold mmap_sem around
 792                  * checking core_state and changing tsk->mm.
 793                  */
 794                 down_read(&old_mm->mmap_sem);
 795                 if (unlikely(old_mm->core_state)) {
 796                         up_read(&old_mm->mmap_sem);
 797                         return -EINTR;
 798                 }
 799         }
 800         task_lock(tsk);
 801         active_mm = tsk->active_mm;
 802         tsk->mm = mm;
 803         tsk->active_mm = mm;
 804         activate_mm(active_mm, mm);
 805         if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
 806                 atomic_dec(&old_mm->oom_disable_count);
 807                 atomic_inc(&tsk->mm->oom_disable_count);
 808         }
 809         task_unlock(tsk);
 810         arch_pick_mmap_layout(mm);
 811         if (old_mm) {
 812                 up_read(&old_mm->mmap_sem);
 813                 BUG_ON(active_mm != old_mm);
 814                 mm_update_next_owner(old_mm);
 815                 mmput(old_mm);
 816                 return 0;
 817         }
 818         mmdrop(active_mm);
 819         return 0;
 820 }
 821
 822 /*
 823  * This function makes sure the current process has its own signal table,
 824  * so that flush_signal_handlers can later reset the handlers without
 825  * disturbing other processes.  (Other processes might share the signal
 826  * table via the CLONE_SIGHAND option to clone().)
 827  */
 828 static int de_thread(struct task_struct *tsk)
 829 {
 830         struct signal_struct *sig = tsk->signal;
 831         struct sighand_struct *oldsighand = tsk->sighand;
 832         spinlock_t *lock = &oldsighand->siglock;
 833
 834         if (thread_group_empty(tsk))
 835                 goto no_thread_group;
 836
 837         /*
 838          * Kill all other threads in the thread group.
 839          */
 840         spin_lock_irq(lock);
 841         if (signal_group_exit(sig)) {
 842                 /*
 843                  * Another group action in progress, just
 844                  * return so that the signal is processed.
 845                  */
 846                 spin_unlock_irq(lock);
 847                 return -EAGAIN;
 848         }
 849
 850         sig->group_exit_task = tsk;
 851         sig->notify_count = zap_other_threads(tsk);
 852         if (!thread_group_leader(tsk))
 853                 sig->notify_count--;
 854
 855         while (sig->notify_count) {
 856                 __set_current_state(TASK_UNINTERRUPTIBLE);
 857                 spin_unlock_irq(lock);
 858                 schedule();
 859                 spin_lock_irq(lock);
 860         }
 861         spin_unlock_irq(lock);
 862
 863         /*
 864          * At this point all other threads have exited, all we have to
 865          * do is to wait for the thread group leader to become inactive,
 866          * and to assume its PID:
 867          */
 868         if (!thread_group_leader(tsk)) {
 869                 struct task_struct *leader = tsk->group_leader;
 870
 871                 sig->notify_count = -1; /* for exit_notify() */
 872                 for (;;) {
 873                         write_lock_irq(&tasklist_lock);
 874                         if (likely(leader->exit_state))
 875                                 break;
 876                         __set_current_state(TASK_UNINTERRUPTIBLE);
 877                         write_unlock_irq(&tasklist_lock);
 878                         schedule();
 879                 }
 880
 881                 /*
 882                  * The only record we have of the real-time age of a
 883                  * process, regardless of execs it's done, is start_time.
 884                  * All the past CPU time is accumulated in signal_struct
 885                  * from sister threads now dead.  But in this non-leader
 886                  * exec, nothing survives from the original leader thread,
 887                  * whose birth marks the true age of this process now.
 888                  * When we take on its identity by switching to its PID, we
 889                  * also take its birthdate (always earlier than our own).
 890                  */
 891                 tsk->start_time = leader->start_time;
 892
 893                 BUG_ON(!same_thread_group(leader, tsk));
 894                 BUG_ON(has_group_leader_pid(tsk));
 895                 /*
 896                  * An exec() starts a new thread group with the
 897                  * TGID of the previous thread group. Rehash the
 898                  * two threads with a switched PID, and release
 899                  * the former thread group leader:
 900                  */
 901
 902                 /* Become a process group leader with the old leader's pid.
 903                  * The old leader becomes a thread of the this thread group.
 904                  * Note: The old leader also uses this pid until release_task
 905                  *       is called.  Odd but simple and correct.
 906                  */
 907                 detach_pid(tsk, PIDTYPE_PID);
 908                 tsk->pid = leader->pid;
 909                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 910                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 911                 transfer_pid(leader, tsk, PIDTYPE_SID);
 912
 913                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 914                 list_replace_init(&leader->sibling, &tsk->sibling);
 915
 916                 tsk->group_leader = tsk;
 917                 leader->group_leader = tsk;
 918
 919                 tsk->exit_signal = SIGCHLD;
 920
 921                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 922                 leader->exit_state = EXIT_DEAD;
 923                 write_unlock_irq(&tasklist_lock);
 924
 925                 release_task(leader);
 926         }
 927
 928         sig->group_exit_task = NULL;
 929         sig->notify_count = 0;
 930
 931 no_thread_group:
 932         if (current->mm)
 933                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 934
 935         exit_itimers(sig);
 936         flush_itimer_signals();
 937
 938         if (atomic_read(&oldsighand->count) != 1) {
 939                 struct sighand_struct *newsighand;
 940                 /*
 941                  * This ->sighand is shared with the CLONE_SIGHAND
 942                  * but not CLONE_THREAD task, switch to the new one.
 943                  */
 944                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 945                 if (!newsighand)
 946                         return -ENOMEM;
 947
 948                 atomic_set(&newsighand->count, 1);
 949                 memcpy(newsighand->action, oldsighand->action,
 950                        sizeof(newsighand->action));
 951
 952                 write_lock_irq(&tasklist_lock);
 953                 spin_lock(&oldsighand->siglock);
 954                 rcu_assign_pointer(tsk->sighand, newsighand);
 955                 spin_unlock(&oldsighand->siglock);
 956                 write_unlock_irq(&tasklist_lock);
 957
 958                 __cleanup_sighand(oldsighand);
 959         }
 960
 961         BUG_ON(!thread_group_leader(tsk));
 962         return 0;
 963 }
 964
 965 /*
 966  * These functions flushes out all traces of the currently running executable
 967  * so that a new one can be started
 968  */
 969 static void flush_old_files(struct files_struct * files)
 970 {
 971         long j = -1;
 972         struct fdtable *fdt;
 973
 974         spin_lock(&files->file_lock);
 975         for (;;) {
 976                 unsigned long set, i;
 977
 978                 j++;
 979                 i = j * __NFDBITS;
 980                 fdt = files_fdtable(files);
 981                 if (i >= fdt->max_fds)
 982                         break;
 983                 set = fdt->close_on_exec->fds_bits[j];
 984                 if (!set)
 985                         continue;
 986                 fdt->close_on_exec->fds_bits[j] = 0;
 987                 spin_unlock(&files->file_lock);
 988                 for ( ; set ; i++,set >>= 1) {
 989                         if (set & 1) {
 990                                 sys_close(i);
 991                         }
 992                 }
 993                 spin_lock(&files->file_lock);
 994
 995         }
 996         spin_unlock(&files->file_lock);
 997 }
 998
 999 char *get_task_comm(char *buf, struct task_struct *tsk)
1000 {
1001         /* buf must be at least sizeof(tsk->comm) in size */
1002         task_lock(tsk);
1003         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1004         task_unlock(tsk);
1005         return buf;
1006 }
1007
1008 void set_task_comm(struct task_struct *tsk, char *buf)
1009 {
1010         task_lock(tsk);
1011
1012         /*
1013          * Threads may access current->comm without holding
1014          * the task lock, so write the string carefully.
1015          * Readers without a lock may see incomplete new
1016          * names but are safe from non-terminating string reads.
1017          */
1018         memset(tsk->comm, 0, TASK_COMM_LEN);
1019         wmb();
1020         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1021         task_unlock(tsk);
1022         perf_event_comm(tsk);
1023 }
1024
1025 int flush_old_exec(struct linux_binprm * bprm)
1026 {
1027         int retval;
1028
1029         /*
1030          * Make sure we have a private signal table and that
1031          * we are unassociated from the previous thread group.
1032          */
1033         retval = de_thread(current);
1034         if (retval)
1035                 goto out;
1036
1037         set_mm_exe_file(bprm->mm, bprm->file);
1038
1039         /*
1040          * Release all of the old mmap stuff
1041          */
1042         acct_arg_size(bprm, 0);
1043         retval = exec_mmap(bprm->mm);
1044         if (retval)
1045                 goto out;
1046
1047         bprm->mm = NULL;                /* We're using it now */
1048
1049         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1050         flush_thread();
1051         current->personality &= ~bprm->per_clear;
1052
1053         return 0;
1054
1055 out:
1056         return retval;
1057 }
1058 EXPORT_SYMBOL(flush_old_exec);
1059
1060 void setup_new_exec(struct linux_binprm * bprm)
1061 {
1062         int i, ch;
1063         const char *name;
1064         char tcomm[sizeof(current->comm)];
1065
1066         arch_pick_mmap_layout(current->mm);
1067
1068         /* This is the point of no return */
1069         current->sas_ss_sp = current->sas_ss_size = 0;
1070
1071         if (current_euid() == current_uid() && current_egid() == current_gid())
1072                 set_dumpable(current->mm, 1);
1073         else
1074                 set_dumpable(current->mm, suid_dumpable);
1075
1076         name = bprm->filename;
1077
1078         /* Copies the binary name from after last slash */
1079         for (i=0; (ch = *(name++)) != '\0';) {
1080                 if (ch == '/')
1081                         i = 0; /* overwrite what we wrote */
1082                 else
1083                         if (i < (sizeof(tcomm) - 1))
1084                                 tcomm[i++] = ch;
1085         }
1086         tcomm[i] = '\0';
1087         set_task_comm(current, tcomm);
1088
1089         /* Set the new mm task size. We have to do that late because it may
1090          * depend on TIF_32BIT which is only updated in flush_thread() on
1091          * some architectures like powerpc
1092          */
1093         current->mm->task_size = TASK_SIZE;
1094
1095         /* install the new credentials */
1096         if (bprm->cred->uid != current_euid() ||
1097             bprm->cred->gid != current_egid()) {
1098                 current->pdeath_signal = 0;
1099         } else if (file_permission(bprm->file, MAY_READ) ||
1100                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1101                 set_dumpable(current->mm, suid_dumpable);
1102         }
1103
1104         /*
1105          * Flush performance counters when crossing a
1106          * security domain:
1107          */
1108         if (!get_dumpable(current->mm))
1109                 perf_event_exit_task(current);
1110
1111         /* An exec changes our domain. We are no longer part of the thread
1112            group */
1113
1114         current->self_exec_id++;
1115
1116         flush_signal_handlers(current, 0);
1117         flush_old_files(current->files);
1118 }
1119 EXPORT_SYMBOL(setup_new_exec);
1120
1121 /*
1122  * Prepare credentials and lock ->cred_guard_mutex.
1123  * install_exec_creds() commits the new creds and drops the lock.
1124  * Or, if exec fails before, free_bprm() should release ->cred and
1125  * and unlock.
1126  */
1127 int prepare_bprm_creds(struct linux_binprm *bprm)
1128 {
1129         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1130                 return -ERESTARTNOINTR;
1131
1132         bprm->cred = prepare_exec_creds();
1133         if (likely(bprm->cred))
1134                 return 0;
1135
1136         mutex_unlock(&current->signal->cred_guard_mutex);
1137         return -ENOMEM;
1138 }
1139
1140 void free_bprm(struct linux_binprm *bprm)
1141 {
1142         free_arg_pages(bprm);
1143         if (bprm->cred) {
1144                 mutex_unlock(&current->signal->cred_guard_mutex);
1145                 abort_creds(bprm->cred);
1146         }
1147         kfree(bprm);
1148 }
1149
1150 /*
1151  * install the new credentials for this executable
1152  */
1153 void install_exec_creds(struct linux_binprm *bprm)
1154 {
1155         security_bprm_committing_creds(bprm);
1156
1157         commit_creds(bprm->cred);
1158         bprm->cred = NULL;
1159         /*
1160          * cred_guard_mutex must be held at least to this point to prevent
1161          * ptrace_attach() from altering our determination of the task's
1162          * credentials; any time after this it may be unlocked.
1163          */
1164         security_bprm_committed_creds(bprm);
1165         mutex_unlock(&current->signal->cred_guard_mutex);
1166 }
1167 EXPORT_SYMBOL(install_exec_creds);
1168
1169 /*
1170  * determine how safe it is to execute the proposed program
1171  * - the caller must hold ->cred_guard_mutex to protect against
1172  *   PTRACE_ATTACH
1173  */
1174 int check_unsafe_exec(struct linux_binprm *bprm)
1175 {
1176         struct task_struct *p = current, *t;
1177         unsigned n_fs;
1178         int res = 0;
1179
1180         bprm->unsafe = tracehook_unsafe_exec(p);
1181
1182         n_fs = 1;
1183         spin_lock(&p->fs->lock);
1184         rcu_read_lock();
1185         for (t = next_thread(p); t != p; t = next_thread(t)) {
1186                 if (t->fs == p->fs)
1187                         n_fs++;
1188         }
1189         rcu_read_unlock();
1190
1191         if (p->fs->users > n_fs) {
1192                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1193         } else {
1194                 res = -EAGAIN;
1195                 if (!p->fs->in_exec) {
1196                         p->fs->in_exec = 1;
1197                         res = 1;
1198                 }
1199         }
1200         spin_unlock(&p->fs->lock);
1201
1202         return res;
1203 }
1204
1205 /*
1206  * Fill the binprm structure from the inode.
1207  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1208  *
1209  * This may be called multiple times for binary chains (scripts for example).
1210  */
1211 int prepare_binprm(struct linux_binprm *bprm)
1212 {
1213         umode_t mode;
1214         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1215         int retval;
1216
1217         mode = inode->i_mode;
1218         if (bprm->file->f_op == NULL)
1219                 return -EACCES;
1220
1221         /* clear any previous set[ug]id data from a previous binary */
1222         bprm->cred->euid = current_euid();
1223         bprm->cred->egid = current_egid();
1224
1225         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1226                 /* Set-uid? */
1227                 if (mode & S_ISUID) {
1228                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1229                         bprm->cred->euid = inode->i_uid;
1230                 }
1231
1232                 /* Set-gid? */
1233                 /*
1234                  * If setgid is set but no group execute bit then this
1235                  * is a candidate for mandatory locking, not a setgid
1236                  * executable.
1237                  */
1238                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1239                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1240                         bprm->cred->egid = inode->i_gid;
1241                 }
1242         }
1243
1244         /* fill in binprm security blob */
1245         retval = security_bprm_set_creds(bprm);
1246         if (retval)
1247                 return retval;
1248         bprm->cred_prepared = 1;
1249
1250         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1251         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1252 }
1253
1254 EXPORT_SYMBOL(prepare_binprm);
1255
1256 /*
1257  * Arguments are '\0' separated strings found at the location bprm->p
1258  * points to; chop off the first by relocating brpm->p to right after
1259  * the first '\0' encountered.
1260  */
1261 int remove_arg_zero(struct linux_binprm *bprm)
1262 {
1263         int ret = 0;
1264         unsigned long offset;
1265         char *kaddr;
1266         struct page *page;
1267
1268         if (!bprm->argc)
1269                 return 0;
1270
1271         do {
1272                 offset = bprm->p & ~PAGE_MASK;
1273                 page = get_arg_page(bprm, bprm->p, 0);
1274                 if (!page) {
1275                         ret = -EFAULT;
1276                         goto out;
1277                 }
1278                 kaddr = kmap_atomic(page, KM_USER0);
1279
1280                 for (; offset < PAGE_SIZE && kaddr[offset];
1281                                 offset++, bprm->p++)
1282                         ;
1283
1284                 kunmap_atomic(kaddr, KM_USER0);
1285                 put_arg_page(page);
1286
1287                 if (offset == PAGE_SIZE)
1288                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1289         } while (offset == PAGE_SIZE);
1290
1291         bprm->p++;
1292         bprm->argc--;
1293         ret = 0;
1294
1295 out:
1296         return ret;
1297 }
1298 EXPORT_SYMBOL(remove_arg_zero);
1299
1300 /*
1301  * cycle the list of binary formats handler, until one recognizes the image
1302  */
1303 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1304 {
1305         unsigned int depth = bprm->recursion_depth;
1306         int try,retval;
1307         struct linux_binfmt *fmt;
1308
1309         retval = security_bprm_check(bprm);
1310         if (retval)
1311                 return retval;
1312
1313         /* kernel module loader fixup */
1314         /* so we don't try to load run modprobe in kernel space. */
1315         set_fs(USER_DS);
1316
1317         retval = audit_bprm(bprm);
1318         if (retval)
1319                 return retval;
1320
1321         retval = -ENOENT;
1322         for (try=0; try<2; try++) {
1323                 read_lock(&binfmt_lock);
1324                 list_for_each_entry(fmt, &formats, lh) {
1325                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1326                         if (!fn)
1327                                 continue;
1328                         if (!try_module_get(fmt->module))
1329                                 continue;
1330                         read_unlock(&binfmt_lock);
1331                         retval = fn(bprm, regs);
1332                         /*
1333                          * Restore the depth counter to its starting value
1334                          * in this call, so we don't have to rely on every
1335                          * load_binary function to restore it on return.
1336                          */
1337                         bprm->recursion_depth = depth;
1338                         if (retval >= 0) {
1339                                 if (depth == 0)
1340                                         tracehook_report_exec(fmt, bprm, regs);
1341                                 put_binfmt(fmt);
1342                                 allow_write_access(bprm->file);
1343                                 if (bprm->file)
1344                                         fput(bprm->file);
1345                                 bprm->file = NULL;
1346                                 current->did_exec = 1;
1347                                 proc_exec_connector(current);
1348                                 return retval;
1349                         }
1350                         read_lock(&binfmt_lock);
1351                         put_binfmt(fmt);
1352                         if (retval != -ENOEXEC || bprm->mm == NULL)
1353                                 break;
1354                         if (!bprm->file) {
1355                                 read_unlock(&binfmt_lock);
1356                                 return retval;
1357                         }
1358                 }
1359                 read_unlock(&binfmt_lock);
1360                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1361                         break;
1362 #ifdef CONFIG_MODULES
1363                 } else {
1364 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1365                         if (printable(bprm->buf[0]) &&
1366                             printable(bprm->buf[1]) &&
1367                             printable(bprm->buf[2]) &&
1368                             printable(bprm->buf[3]))
1369                                 break; /* -ENOEXEC */
1370                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1371 #endif
1372                 }
1373         }
1374         return retval;
1375 }
1376
1377 EXPORT_SYMBOL(search_binary_handler);
1378
1379 /*
1380  * sys_execve() executes a new program.
1381  */
1382 int do_execve(const char * filename,
1383         const char __user *const __user *argv,
1384         const char __user *const __user *envp,
1385         struct pt_regs * regs)
1386 {
1387         struct linux_binprm *bprm;
1388         struct file *file;
1389         struct files_struct *displaced;
1390         bool clear_in_exec;
1391         int retval;
1392
1393         retval = unshare_files(&displaced);
1394         if (retval)
1395                 goto out_ret;
1396
1397         retval = -ENOMEM;
1398         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1399         if (!bprm)
1400                 goto out_files;
1401
1402         retval = prepare_bprm_creds(bprm);
1403         if (retval)
1404                 goto out_free;
1405
1406         retval = check_unsafe_exec(bprm);
1407         if (retval < 0)
1408                 goto out_free;
1409         clear_in_exec = retval;
1410         current->in_execve = 1;
1411
1412         file = open_exec(filename);
1413         retval = PTR_ERR(file);
1414         if (IS_ERR(file))
1415                 goto out_unmark;
1416
1417         sched_exec();
1418
1419         bprm->file = file;
1420         bprm->filename = filename;
1421         bprm->interp = filename;
1422
1423         retval = bprm_mm_init(bprm);
1424         if (retval)
1425                 goto out_file;
1426
1427         bprm->argc = count(argv, MAX_ARG_STRINGS);
1428         if ((retval = bprm->argc) < 0)
1429                 goto out;
1430
1431         bprm->envc = count(envp, MAX_ARG_STRINGS);
1432         if ((retval = bprm->envc) < 0)
1433                 goto out;
1434
1435         retval = prepare_binprm(bprm);
1436         if (retval < 0)
1437                 goto out;
1438
1439         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1440         if (retval < 0)
1441                 goto out;
1442
1443         bprm->exec = bprm->p;
1444         retval = copy_strings(bprm->envc, envp, bprm);
1445         if (retval < 0)
1446                 goto out;
1447
1448         retval = copy_strings(bprm->argc, argv, bprm);
1449         if (retval < 0)
1450                 goto out;
1451
1452         retval = search_binary_handler(bprm,regs);
1453         if (retval < 0)
1454                 goto out;
1455
1456         /* execve succeeded */
1457         current->fs->in_exec = 0;
1458         current->in_execve = 0;
1459         acct_update_integrals(current);
1460         free_bprm(bprm);
1461         if (displaced)
1462                 put_files_struct(displaced);
1463         return retval;
1464
1465 out:
1466         if (bprm->mm) {
1467                 acct_arg_size(bprm, 0);
1468                 mmput(bprm->mm);
1469         }
1470
1471 out_file:
1472         if (bprm->file) {
1473                 allow_write_access(bprm->file);
1474                 fput(bprm->file);
1475         }
1476
1477 out_unmark:
1478         if (clear_in_exec)
1479                 current->fs->in_exec = 0;
1480         current->in_execve = 0;
1481
1482 out_free:
1483         free_bprm(bprm);
1484
1485 out_files:
1486         if (displaced)
1487                 reset_files_struct(displaced);
1488 out_ret:
1489         return retval;
1490 }
1491
1492 void set_binfmt(struct linux_binfmt *new)
1493 {
1494         struct mm_struct *mm = current->mm;
1495
1496         if (mm->binfmt)
1497                 module_put(mm->binfmt->module);
1498
1499         mm->binfmt = new;
1500         if (new)
1501                 __module_get(new->module);
1502 }
1503
1504 EXPORT_SYMBOL(set_binfmt);
1505
1506 static int expand_corename(struct core_name *cn)
1507 {
1508         char *old_corename = cn->corename;
1509
1510         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1511         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1512
1513         if (!cn->corename) {
1514                 kfree(old_corename);
1515                 return -ENOMEM;
1516         }
1517
1518         return 0;
1519 }
1520
1521 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1522 {
1523         char *cur;
1524         int need;
1525         int ret;
1526         va_list arg;
1527
1528         va_start(arg, fmt);
1529         need = vsnprintf(NULL, 0, fmt, arg);
1530         va_end(arg);
1531
1532         if (likely(need < cn->size - cn->used - 1))
1533                 goto out_printf;
1534
1535         ret = expand_corename(cn);
1536         if (ret)
1537                 goto expand_fail;
1538
1539 out_printf:
1540         cur = cn->corename + cn->used;
1541         va_start(arg, fmt);
1542         vsnprintf(cur, need + 1, fmt, arg);
1543         va_end(arg);
1544         cn->used += need;
1545         return 0;
1546
1547 expand_fail:
1548         return ret;
1549 }
1550
1551 /* format_corename will inspect the pattern parameter, and output a
1552  * name into corename, which must have space for at least
1553  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1554  */
1555 static int format_corename(struct core_name *cn, long signr)
1556 {
1557         const struct cred *cred = current_cred();
1558         const char *pat_ptr = core_pattern;
1559         int ispipe = (*pat_ptr == '|');
1560         int pid_in_pattern = 0;
1561         int err = 0;
1562
1563         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1564         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1565         cn->used = 0;
1566
1567         if (!cn->corename)
1568                 return -ENOMEM;
1569
1570         /* Repeat as long as we have more pattern to process and more output
1571            space */
1572         while (*pat_ptr) {
1573                 if (*pat_ptr != '%') {
1574                         if (*pat_ptr == 0)
1575                                 goto out;
1576                         err = cn_printf(cn, "%c", *pat_ptr++);
1577                 } else {
1578                         switch (*++pat_ptr) {
1579                         /* single % at the end, drop that */
1580                         case 0:
1581                                 goto out;
1582                         /* Double percent, output one percent */
1583                         case '%':
1584                                 err = cn_printf(cn, "%c", '%');
1585                                 break;
1586                         /* pid */
1587                         case 'p':
1588                                 pid_in_pattern = 1;
1589                                 err = cn_printf(cn, "%d",
1590                                               task_tgid_vnr(current));
1591                                 break;
1592                         /* uid */
1593                         case 'u':
1594                                 err = cn_printf(cn, "%d", cred->uid);
1595                                 break;
1596                         /* gid */
1597                         case 'g':
1598                                 err = cn_printf(cn, "%d", cred->gid);
1599                                 break;
1600                         /* signal that caused the coredump */
1601                         case 's':
1602                                 err = cn_printf(cn, "%ld", signr);
1603                                 break;
1604                         /* UNIX time of coredump */
1605                         case 't': {
1606                                 struct timeval tv;
1607                                 do_gettimeofday(&tv);
1608                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1609                                 break;
1610                         }
1611                         /* hostname */
1612                         case 'h':
1613                                 down_read(&uts_sem);
1614                                 err = cn_printf(cn, "%s",
1615                                               utsname()->nodename);
1616                                 up_read(&uts_sem);
1617                                 break;
1618                         /* executable */
1619                         case 'e':
1620                                 err = cn_printf(cn, "%s", current->comm);
1621                                 break;
1622                         /* core limit size */
1623                         case 'c':
1624                                 err = cn_printf(cn, "%lu",
1625                                               rlimit(RLIMIT_CORE));
1626                                 break;
1627                         default:
1628                                 break;
1629                         }
1630                         ++pat_ptr;
1631                 }
1632
1633                 if (err)
1634                         return err;
1635         }
1636
1637         /* Backward compatibility with core_uses_pid:
1638          *
1639          * If core_pattern does not include a %p (as is the default)
1640          * and core_uses_pid is set, then .%pid will be appended to
1641          * the filename. Do not do this for piped commands. */
1642         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1643                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1644                 if (err)
1645                         return err;
1646         }
1647 out:
1648         return ispipe;
1649 }
1650
1651 static int zap_process(struct task_struct *start, int exit_code)
1652 {
1653         struct task_struct *t;
1654         int nr = 0;
1655
1656         start->signal->flags = SIGNAL_GROUP_EXIT;
1657         start->signal->group_exit_code = exit_code;
1658         start->signal->group_stop_count = 0;
1659
1660         t = start;
1661         do {
1662                 if (t != current && t->mm) {
1663                         sigaddset(&t->pending.signal, SIGKILL);
1664                         signal_wake_up(t, 1);
1665                         nr++;
1666                 }
1667         } while_each_thread(start, t);
1668
1669         return nr;
1670 }
1671
1672 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1673                                 struct core_state *core_state, int exit_code)
1674 {
1675         struct task_struct *g, *p;
1676         unsigned long flags;
1677         int nr = -EAGAIN;
1678
1679         spin_lock_irq(&tsk->sighand->siglock);
1680         if (!signal_group_exit(tsk->signal)) {
1681                 mm->core_state = core_state;
1682                 nr = zap_process(tsk, exit_code);
1683         }
1684         spin_unlock_irq(&tsk->sighand->siglock);
1685         if (unlikely(nr < 0))
1686                 return nr;
1687
1688         if (atomic_read(&mm->mm_users) == nr + 1)
1689                 goto done;
1690         /*
1691          * We should find and kill all tasks which use this mm, and we should
1692          * count them correctly into ->nr_threads. We don't take tasklist
1693          * lock, but this is safe wrt:
1694          *
1695          * fork:
1696          *      None of sub-threads can fork after zap_process(leader). All
1697          *      processes which were created before this point should be
1698          *      visible to zap_threads() because copy_process() adds the new
1699          *      process to the tail of init_task.tasks list, and lock/unlock
1700          *      of ->siglock provides a memory barrier.
1701          *
1702          * do_exit:
1703          *      The caller holds mm->mmap_sem. This means that the task which
1704          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1705          *      its ->mm.
1706          *
1707          * de_thread:
1708          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1709          *      we must see either old or new leader, this does not matter.
1710          *      However, it can change p->sighand, so lock_task_sighand(p)
1711          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1712          *      it can't fail.
1713          *
1714          *      Note also that "g" can be the old leader with ->mm == NULL
1715          *      and already unhashed and thus removed from ->thread_group.
1716          *      This is OK, __unhash_process()->list_del_rcu() does not
1717          *      clear the ->next pointer, we will find the new leader via
1718          *      next_thread().
1719          */
1720         rcu_read_lock();
1721         for_each_process(g) {
1722                 if (g == tsk->group_leader)
1723                         continue;
1724                 if (g->flags & PF_KTHREAD)
1725                         continue;
1726                 p = g;
1727                 do {
1728                         if (p->mm) {
1729                                 if (unlikely(p->mm == mm)) {
1730                                         lock_task_sighand(p, &flags);
1731                                         nr += zap_process(p, exit_code);
1732                                         unlock_task_sighand(p, &flags);
1733                                 }
1734                                 break;
1735                         }
1736                 } while_each_thread(g, p);
1737         }
1738         rcu_read_unlock();
1739 done:
1740         atomic_set(&core_state->nr_threads, nr);
1741         return nr;
1742 }
1743
1744 static int coredump_wait(int exit_code, struct core_state *core_state)
1745 {
1746         struct task_struct *tsk = current;
1747         struct mm_struct *mm = tsk->mm;
1748         struct completion *vfork_done;
1749         int core_waiters = -EBUSY;
1750
1751         init_completion(&core_state->startup);
1752         core_state->dumper.task = tsk;
1753         core_state->dumper.next = NULL;
1754
1755         down_write(&mm->mmap_sem);
1756         if (!mm->core_state)
1757                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1758         up_write(&mm->mmap_sem);
1759
1760         if (unlikely(core_waiters < 0))
1761                 goto fail;
1762
1763         /*
1764          * Make sure nobody is waiting for us to release the VM,
1765          * otherwise we can deadlock when we wait on each other
1766          */
1767         vfork_done = tsk->vfork_done;
1768         if (vfork_done) {
1769                 tsk->vfork_done = NULL;
1770                 complete(vfork_done);
1771         }
1772
1773         if (core_waiters)
1774                 wait_for_completion(&core_state->startup);
1775 fail:
1776         return core_waiters;
1777 }
1778
1779 static void coredump_finish(struct mm_struct *mm)
1780 {
1781         struct core_thread *curr, *next;
1782         struct task_struct *task;
1783
1784         next = mm->core_state->dumper.next;
1785         while ((curr = next) != NULL) {
1786                 next = curr->next;
1787                 task = curr->task;
1788                 /*
1789                  * see exit_mm(), curr->task must not see
1790                  * ->task == NULL before we read ->next.
1791                  */
1792                 smp_mb();
1793                 curr->task = NULL;
1794                 wake_up_process(task);
1795         }
1796
1797         mm->core_state = NULL;
1798 }
1799
1800 /*
1801  * set_dumpable converts traditional three-value dumpable to two flags and
1802  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1803  * these bits are not changed atomically.  So get_dumpable can observe the
1804  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1805  * return either old dumpable or new one by paying attention to the order of
1806  * modifying the bits.
1807  *
1808  * dumpable |   mm->flags (binary)
1809  * old  new | initial interim  final
1810  * ---------+-----------------------
1811  *  0    1  |   00      01      01
1812  *  0    2  |   00      10(*)   11
1813  *  1    0  |   01      00      00
1814  *  1    2  |   01      11      11
1815  *  2    0  |   11      10(*)   00
1816  *  2    1  |   11      11      01
1817  *
1818  * (*) get_dumpable regards interim value of 10 as 11.
1819  */
1820 void set_dumpable(struct mm_struct *mm, int value)
1821 {
1822         switch (value) {
1823         case 0:
1824                 clear_bit(MMF_DUMPABLE, &mm->flags);
1825                 smp_wmb();
1826                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1827                 break;
1828         case 1:
1829                 set_bit(MMF_DUMPABLE, &mm->flags);
1830                 smp_wmb();
1831                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1832                 break;
1833         case 2:
1834                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1835                 smp_wmb();
1836                 set_bit(MMF_DUMPABLE, &mm->flags);
1837                 break;
1838         }
1839 }
1840
1841 static int __get_dumpable(unsigned long mm_flags)
1842 {
1843         int ret;
1844
1845         ret = mm_flags & MMF_DUMPABLE_MASK;
1846         return (ret >= 2) ? 2 : ret;
1847 }
1848
1849 int get_dumpable(struct mm_struct *mm)
1850 {
1851         return __get_dumpable(mm->flags);
1852 }
1853
1854 static void wait_for_dump_helpers(struct file *file)
1855 {
1856         struct pipe_inode_info *pipe;
1857
1858         pipe = file->f_path.dentry->d_inode->i_pipe;
1859
1860         pipe_lock(pipe);
1861         pipe->readers++;
1862         pipe->writers--;
1863
1864         while ((pipe->readers > 1) && (!signal_pending(current))) {
1865                 wake_up_interruptible_sync(&pipe->wait);
1866                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1867                 pipe_wait(pipe);
1868         }
1869
1870         pipe->readers--;
1871         pipe->writers++;
1872         pipe_unlock(pipe);
1873
1874 }
1875
1876
1877 /*
1878  * umh_pipe_setup
1879  * helper function to customize the process used
1880  * to collect the core in userspace.  Specifically
1881  * it sets up a pipe and installs it as fd 0 (stdin)
1882  * for the process.  Returns 0 on success, or
1883  * PTR_ERR on failure.
1884  * Note that it also sets the core limit to 1.  This
1885  * is a special value that we use to trap recursive
1886  * core dumps
1887  */
1888 static int umh_pipe_setup(struct subprocess_info *info)
1889 {
1890         struct file *rp, *wp;
1891         struct fdtable *fdt;
1892         struct coredump_params *cp = (struct coredump_params *)info->data;
1893         struct files_struct *cf = current->files;
1894
1895         wp = create_write_pipe(0);
1896         if (IS_ERR(wp))
1897                 return PTR_ERR(wp);
1898
1899         rp = create_read_pipe(wp, 0);
1900         if (IS_ERR(rp)) {
1901                 free_write_pipe(wp);
1902                 return PTR_ERR(rp);
1903         }
1904
1905         cp->file = wp;
1906
1907         sys_close(0);
1908         fd_install(0, rp);
1909         spin_lock(&cf->file_lock);
1910         fdt = files_fdtable(cf);
1911         FD_SET(0, fdt->open_fds);
1912         FD_CLR(0, fdt->close_on_exec);
1913         spin_unlock(&cf->file_lock);
1914
1915         /* and disallow core files too */
1916         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1917
1918         return 0;
1919 }
1920
1921 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1922 {
1923         struct core_state core_state;
1924         struct core_name cn;
1925         struct mm_struct *mm = current->mm;
1926         struct linux_binfmt * binfmt;
1927         const struct cred *old_cred;
1928         struct cred *cred;
1929         int retval = 0;
1930         int flag = 0;
1931         int ispipe;
1932         static atomic_t core_dump_count = ATOMIC_INIT(0);
1933         struct coredump_params cprm = {
1934                 .signr = signr,
1935                 .regs = regs,
1936                 .limit = rlimit(RLIMIT_CORE),
1937                 /*
1938                  * We must use the same mm->flags while dumping core to avoid
1939                  * inconsistency of bit flags, since this flag is not protected
1940                  * by any locks.
1941                  */
1942                 .mm_flags = mm->flags,
1943         };
1944
1945         audit_core_dumps(signr);
1946
1947         binfmt = mm->binfmt;
1948         if (!binfmt || !binfmt->core_dump)
1949                 goto fail;
1950         if (!__get_dumpable(cprm.mm_flags))
1951                 goto fail;
1952
1953         cred = prepare_creds();
1954         if (!cred)
1955                 goto fail;
1956         /*
1957          *      We cannot trust fsuid as being the "true" uid of the
1958          *      process nor do we know its entire history. We only know it
1959          *      was tainted so we dump it as root in mode 2.
1960          */
1961         if (__get_dumpable(cprm.mm_flags) == 2) {
1962                 /* Setuid core dump mode */
1963                 flag = O_EXCL;          /* Stop rewrite attacks */
1964                 cred->fsuid = 0;        /* Dump root private */
1965         }
1966
1967         retval = coredump_wait(exit_code, &core_state);
1968         if (retval < 0)
1969                 goto fail_creds;
1970
1971         old_cred = override_creds(cred);
1972
1973         /*
1974          * Clear any false indication of pending signals that might
1975          * be seen by the filesystem code called to write the core file.
1976          */
1977         clear_thread_flag(TIF_SIGPENDING);
1978
1979         ispipe = format_corename(&cn, signr);
1980
1981         if (ispipe == -ENOMEM) {
1982                 printk(KERN_WARNING "format_corename failed\n");
1983                 printk(KERN_WARNING "Aborting core\n");
1984                 goto fail_corename;
1985         }
1986
1987         if (ispipe) {
1988                 int dump_count;
1989                 char **helper_argv;
1990
1991                 if (cprm.limit == 1) {
1992                         /*
1993                          * Normally core limits are irrelevant to pipes, since
1994                          * we're not writing to the file system, but we use
1995                          * cprm.limit of 1 here as a speacial value. Any
1996                          * non-1 limit gets set to RLIM_INFINITY below, but
1997                          * a limit of 0 skips the dump.  This is a consistent
1998                          * way to catch recursive crashes.  We can still crash
1999                          * if the core_pattern binary sets RLIM_CORE =  !1
2000                          * but it runs as root, and can do lots of stupid things
2001                          * Note that we use task_tgid_vnr here to grab the pid
2002                          * of the process group leader.  That way we get the
2003                          * right pid if a thread in a multi-threaded
2004                          * core_pattern process dies.
2005                          */
2006                         printk(KERN_WARNING
2007                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2008                                 task_tgid_vnr(current), current->comm);
2009                         printk(KERN_WARNING "Aborting core\n");
2010                         goto fail_unlock;
2011                 }
2012                 cprm.limit = RLIM_INFINITY;
2013
2014                 dump_count = atomic_inc_return(&core_dump_count);
2015                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2016                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2017                                task_tgid_vnr(current), current->comm);
2018                         printk(KERN_WARNING "Skipping core dump\n");
2019                         goto fail_dropcount;
2020                 }
2021
2022                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2023                 if (!helper_argv) {
2024                         printk(KERN_WARNING "%s failed to allocate memory\n",
2025                                __func__);
2026                         goto fail_dropcount;
2027                 }
2028
2029                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2030                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2031                                         NULL, &cprm);
2032                 argv_free(helper_argv);
2033                 if (retval) {
2034                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2035                                cn.corename);
2036                         goto close_fail;
2037                 }
2038         } else {
2039                 struct inode *inode;
2040
2041                 if (cprm.limit < binfmt->min_coredump)
2042                         goto fail_unlock;
2043
2044                 cprm.file = filp_open(cn.corename,
2045                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2046                                  0600);
2047                 if (IS_ERR(cprm.file))
2048                         goto fail_unlock;
2049
2050                 inode = cprm.file->f_path.dentry->d_inode;
2051                 if (inode->i_nlink > 1)
2052                         goto close_fail;
2053                 if (d_unhashed(cprm.file->f_path.dentry))
2054                         goto close_fail;
2055                 /*
2056                  * AK: actually i see no reason to not allow this for named
2057                  * pipes etc, but keep the previous behaviour for now.
2058                  */
2059                 if (!S_ISREG(inode->i_mode))
2060                         goto close_fail;
2061                 /*
2062                  * Dont allow local users get cute and trick others to coredump
2063                  * into their pre-created files.
2064                  */
2065                 if (inode->i_uid != current_fsuid())
2066                         goto close_fail;
2067                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2068                         goto close_fail;
2069                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2070                         goto close_fail;
2071         }
2072
2073         retval = binfmt->core_dump(&cprm);
2074         if (retval)
2075                 current->signal->group_exit_code |= 0x80;
2076
2077         if (ispipe && core_pipe_limit)
2078                 wait_for_dump_helpers(cprm.file);
2079 close_fail:
2080         if (cprm.file)
2081                 filp_close(cprm.file, NULL);
2082 fail_dropcount:
2083         if (ispipe)
2084                 atomic_dec(&core_dump_count);
2085 fail_unlock:
2086         kfree(cn.corename);
2087 fail_corename:
2088         coredump_finish(mm);
2089         revert_creds(old_cred);
2090 fail_creds:
2091         put_cred(cred);
2092 fail:
2093         return;
2094 }
2095
2096 /*
2097  * Core dumping helper functions.  These are the only things you should
2098  * do on a core-file: use only these functions to write out all the
2099  * necessary info.
2100  */
2101 int dump_write(struct file *file, const void *addr, int nr)
2102 {
2103         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2104 }
2105 EXPORT_SYMBOL(dump_write);
2106
2107 int dump_seek(struct file *file, loff_t off)
2108 {
2109         int ret = 1;
2110
2111         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2112                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2113                         return 0;
2114         } else {
2115                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2116
2117                 if (!buf)
2118                         return 0;
2119                 while (off > 0) {
2120                         unsigned long n = off;
2121
2122                         if (n > PAGE_SIZE)
2123                                 n = PAGE_SIZE;
2124                         if (!dump_write(file, buf, n)) {
2125                                 ret = 0;
2126                                 break;
2127                         }
2128                         off -= n;
2129                 }
2130                 free_page((unsigned long)buf);
2131         }
2132         return ret;
2133 }
2134 EXPORT_SYMBOL(dump_seek);