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 <trace/events/fs.h>
  60
  61 #include <asm/uaccess.h>
  62 #include <asm/mmu_context.h>
  63 #include <asm/tlb.h>
  64 #include "internal.h"
  65
  66 int core_uses_pid;
  67 char core_pattern[CORENAME_MAX_SIZE] = "core";
  68 unsigned int core_pipe_limit;
  69 int suid_dumpable = 0;
  70
  71 struct core_name {
  72         char *corename;
  73         int used, size;
  74 };
  75 static atomic_t call_count = ATOMIC_INIT(1);
  76
  77 /* The maximal length of core_pattern is also specified in sysctl.c */
  78
  79 static LIST_HEAD(formats);
  80 static DEFINE_RWLOCK(binfmt_lock);
  81
  82 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  83 {
  84         if (!fmt)
  85                 return -EINVAL;
  86         write_lock(&binfmt_lock);
  87         insert ? list_add(&fmt->lh, &formats) :
  88                  list_add_tail(&fmt->lh, &formats);
  89         write_unlock(&binfmt_lock);
  90         return 0;
  91 }
  92
  93 EXPORT_SYMBOL(__register_binfmt);
  94
  95 void unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         write_lock(&binfmt_lock);
  98         list_del(&fmt->lh);
  99         write_unlock(&binfmt_lock);
 100 }
 101
 102 EXPORT_SYMBOL(unregister_binfmt);
 103
 104 static inline void put_binfmt(struct linux_binfmt * fmt)
 105 {
 106         module_put(fmt->module);
 107 }
 108
 109 /*
 110  * Note that a shared library must be both readable and executable due to
 111  * security reasons.
 112  *
 113  * Also note that we take the address to load from from the file itself.
 114  */
 115 SYSCALL_DEFINE1(uselib, const char __user *, library)
 116 {
 117         struct file *file;
 118         char *tmp = getname(library);
 119         int error = PTR_ERR(tmp);
 120
 121         if (IS_ERR(tmp))
 122                 goto out;
 123
 124         file = do_filp_open(AT_FDCWD, tmp,
 125                                 O_LARGEFILE | O_RDONLY | __FMODE_EXEC, 0,
 126                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 127         putname(tmp);
 128         error = PTR_ERR(file);
 129         if (IS_ERR(file))
 130                 goto out;
 131
 132         error = -EINVAL;
 133         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 134                 goto exit;
 135
 136         error = -EACCES;
 137         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 138                 goto exit;
 139
 140         fsnotify_open(file);
 141
 142         error = -ENOEXEC;
 143         if(file->f_op) {
 144                 struct linux_binfmt * fmt;
 145
 146                 read_lock(&binfmt_lock);
 147                 list_for_each_entry(fmt, &formats, lh) {
 148                         if (!fmt->load_shlib)
 149                                 continue;
 150                         if (!try_module_get(fmt->module))
 151                                 continue;
 152                         read_unlock(&binfmt_lock);
 153                         error = fmt->load_shlib(file);
 154                         read_lock(&binfmt_lock);
 155                         put_binfmt(fmt);
 156                         if (error != -ENOEXEC)
 157                                 break;
 158                 }
 159                 read_unlock(&binfmt_lock);
 160         }
 161 exit:
 162         fput(file);
 163 out:
 164         return error;
 165 }
 166
 167 #ifdef CONFIG_MMU
 168
 169 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 170 {
 171         struct mm_struct *mm = current->mm;
 172         long diff = (long)(pages - bprm->vma_pages);
 173
 174         if (!mm || !diff)
 175                 return;
 176
 177         bprm->vma_pages = pages;
 178
 179 #ifdef SPLIT_RSS_COUNTING
 180         add_mm_counter(mm, MM_ANONPAGES, diff);
 181 #else
 182         spin_lock(&mm->page_table_lock);
 183         add_mm_counter(mm, MM_ANONPAGES, diff);
 184         spin_unlock(&mm->page_table_lock);
 185 #endif
 186 }
 187
 188 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 189                 int write)
 190 {
 191         struct page *page;
 192         int ret;
 193
 194 #ifdef CONFIG_STACK_GROWSUP
 195         if (write) {
 196                 ret = expand_stack_downwards(bprm->vma, pos);
 197                 if (ret < 0)
 198                         return NULL;
 199         }
 200 #endif
 201         ret = get_user_pages(current, bprm->mm, pos,
 202                         1, write, 1, &page, NULL);
 203         if (ret <= 0)
 204                 return NULL;
 205
 206         if (write) {
 207                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 208                 struct rlimit *rlim;
 209
 210                 acct_arg_size(bprm, size / PAGE_SIZE);
 211
 212                 /*
 213                  * We've historically supported up to 32 pages (ARG_MAX)
 214                  * of argument strings even with small stacks
 215                  */
 216                 if (size <= ARG_MAX)
 217                         return page;
 218
 219                 /*
 220                  * Limit to 1/4-th the stack size for the argv+env strings.
 221                  * This ensures that:
 222                  *  - the remaining binfmt code will not run out of stack space,
 223                  *  - the program will have a reasonable amount of stack left
 224                  *    to work from.
 225                  */
 226                 rlim = current->signal->rlim;
 227                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 228                         put_page(page);
 229                         return NULL;
 230                 }
 231         }
 232
 233         return page;
 234 }
 235
 236 static void put_arg_page(struct page *page)
 237 {
 238         put_page(page);
 239 }
 240
 241 static void free_arg_page(struct linux_binprm *bprm, int i)
 242 {
 243 }
 244
 245 static void free_arg_pages(struct linux_binprm *bprm)
 246 {
 247 }
 248
 249 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 250                 struct page *page)
 251 {
 252         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 253 }
 254
 255 static int __bprm_mm_init(struct linux_binprm *bprm)
 256 {
 257         int err;
 258         struct vm_area_struct *vma = NULL;
 259         struct mm_struct *mm = bprm->mm;
 260
 261         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 262         if (!vma)
 263                 return -ENOMEM;
 264
 265         down_write(&mm->mmap_sem);
 266         vma->vm_mm = mm;
 267
 268         /*
 269          * Place the stack at the largest stack address the architecture
 270          * supports. Later, we'll move this to an appropriate place. We don't
 271          * use STACK_TOP because that can depend on attributes which aren't
 272          * configured yet.
 273          */
 274         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 275         vma->vm_end = STACK_TOP_MAX;
 276         vma->vm_start = vma->vm_end - PAGE_SIZE;
 277         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 278         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 279         INIT_LIST_HEAD(&vma->anon_vma_chain);
 280
 281         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 282         if (err)
 283                 goto err;
 284
 285         err = insert_vm_struct(mm, vma);
 286         if (err)
 287                 goto err;
 288
 289         mm->stack_vm = mm->total_vm = 1;
 290         up_write(&mm->mmap_sem);
 291         bprm->p = vma->vm_end - sizeof(void *);
 292         return 0;
 293 err:
 294         up_write(&mm->mmap_sem);
 295         bprm->vma = NULL;
 296         kmem_cache_free(vm_area_cachep, vma);
 297         return err;
 298 }
 299
 300 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 301 {
 302         return len <= MAX_ARG_STRLEN;
 303 }
 304
 305 #else
 306
 307 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 308 {
 309 }
 310
 311 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 312                 int write)
 313 {
 314         struct page *page;
 315
 316         page = bprm->page[pos / PAGE_SIZE];
 317         if (!page && write) {
 318                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 319                 if (!page)
 320                         return NULL;
 321                 bprm->page[pos / PAGE_SIZE] = page;
 322         }
 323
 324         return page;
 325 }
 326
 327 static void put_arg_page(struct page *page)
 328 {
 329 }
 330
 331 static void free_arg_page(struct linux_binprm *bprm, int i)
 332 {
 333         if (bprm->page[i]) {
 334                 __free_page(bprm->page[i]);
 335                 bprm->page[i] = NULL;
 336         }
 337 }
 338
 339 static void free_arg_pages(struct linux_binprm *bprm)
 340 {
 341         int i;
 342
 343         for (i = 0; i < MAX_ARG_PAGES; i++)
 344                 free_arg_page(bprm, i);
 345 }
 346
 347 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 348                 struct page *page)
 349 {
 350 }
 351
 352 static int __bprm_mm_init(struct linux_binprm *bprm)
 353 {
 354         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 355         return 0;
 356 }
 357
 358 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 359 {
 360         return len <= bprm->p;
 361 }
 362
 363 #endif /* CONFIG_MMU */
 364
 365 /*
 366  * Create a new mm_struct and populate it with a temporary stack
 367  * vm_area_struct.  We don't have enough context at this point to set the stack
 368  * flags, permissions, and offset, so we use temporary values.  We'll update
 369  * them later in setup_arg_pages().
 370  */
 371 int bprm_mm_init(struct linux_binprm *bprm)
 372 {
 373         int err;
 374         struct mm_struct *mm = NULL;
 375
 376         bprm->mm = mm = mm_alloc();
 377         err = -ENOMEM;
 378         if (!mm)
 379                 goto err;
 380
 381         err = init_new_context(current, mm);
 382         if (err)
 383                 goto err;
 384
 385         err = __bprm_mm_init(bprm);
 386         if (err)
 387                 goto err;
 388
 389         return 0;
 390
 391 err:
 392         if (mm) {
 393                 bprm->mm = NULL;
 394                 mmdrop(mm);
 395         }
 396
 397         return err;
 398 }
 399
 400 /*
 401  * count() counts the number of strings in array ARGV.
 402  */
 403 static int count(const char __user * const __user * argv, int max)
 404 {
 405         int i = 0;
 406
 407         if (argv != NULL) {
 408                 for (;;) {
 409                         const char __user * p;
 410
 411                         if (get_user(p, argv))
 412                                 return -EFAULT;
 413                         if (!p)
 414                                 break;
 415                         argv++;
 416                         if (i++ >= max)
 417                                 return -E2BIG;
 418
 419                         if (fatal_signal_pending(current))
 420                                 return -ERESTARTNOHAND;
 421                         cond_resched();
 422                 }
 423         }
 424         return i;
 425 }
 426
 427 /*
 428  * 'copy_strings()' copies argument/environment strings from the old
 429  * processes's memory to the new process's stack.  The call to get_user_pages()
 430  * ensures the destination page is created and not swapped out.
 431  */
 432 static int copy_strings(int argc, const char __user *const __user *argv,
 433                         struct linux_binprm *bprm)
 434 {
 435         struct page *kmapped_page = NULL;
 436         char *kaddr = NULL;
 437         unsigned long kpos = 0;
 438         int ret;
 439
 440         while (argc-- > 0) {
 441                 const char __user *str;
 442                 int len;
 443                 unsigned long pos;
 444
 445                 if (get_user(str, argv+argc) ||
 446                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 447                         ret = -EFAULT;
 448                         goto out;
 449                 }
 450
 451                 if (!valid_arg_len(bprm, len)) {
 452                         ret = -E2BIG;
 453                         goto out;
 454                 }
 455
 456                 /* We're going to work our way backwords. */
 457                 pos = bprm->p;
 458                 str += len;
 459                 bprm->p -= len;
 460
 461                 while (len > 0) {
 462                         int offset, bytes_to_copy;
 463
 464                         if (fatal_signal_pending(current)) {
 465                                 ret = -ERESTARTNOHAND;
 466                                 goto out;
 467                         }
 468                         cond_resched();
 469
 470                         offset = pos % PAGE_SIZE;
 471                         if (offset == 0)
 472                                 offset = PAGE_SIZE;
 473
 474                         bytes_to_copy = offset;
 475                         if (bytes_to_copy > len)
 476                                 bytes_to_copy = len;
 477
 478                         offset -= bytes_to_copy;
 479                         pos -= bytes_to_copy;
 480                         str -= bytes_to_copy;
 481                         len -= bytes_to_copy;
 482
 483                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 484                                 struct page *page;
 485
 486                                 page = get_arg_page(bprm, pos, 1);
 487                                 if (!page) {
 488                                         ret = -E2BIG;
 489                                         goto out;
 490                                 }
 491
 492                                 if (kmapped_page) {
 493                                         flush_kernel_dcache_page(kmapped_page);
 494                                         kunmap(kmapped_page);
 495                                         put_arg_page(kmapped_page);
 496                                 }
 497                                 kmapped_page = page;
 498                                 kaddr = kmap(kmapped_page);
 499                                 kpos = pos & PAGE_MASK;
 500                                 flush_arg_page(bprm, kpos, kmapped_page);
 501                         }
 502                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 503                                 ret = -EFAULT;
 504                                 goto out;
 505                         }
 506                 }
 507         }
 508         ret = 0;
 509 out:
 510         if (kmapped_page) {
 511                 flush_kernel_dcache_page(kmapped_page);
 512                 kunmap(kmapped_page);
 513                 put_arg_page(kmapped_page);
 514         }
 515         return ret;
 516 }
 517
 518 /*
 519  * Like copy_strings, but get argv and its values from kernel memory.
 520  */
 521 int copy_strings_kernel(int argc, const char *const *argv,
 522                         struct linux_binprm *bprm)
 523 {
 524         int r;
 525         mm_segment_t oldfs = get_fs();
 526         set_fs(KERNEL_DS);
 527         r = copy_strings(argc, (const char __user *const  __user *)argv, bprm);
 528         set_fs(oldfs);
 529         return r;
 530 }
 531 EXPORT_SYMBOL(copy_strings_kernel);
 532
 533 #ifdef CONFIG_MMU
 534
 535 /*
 536  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 537  * the binfmt code determines where the new stack should reside, we shift it to
 538  * its final location.  The process proceeds as follows:
 539  *
 540  * 1) Use shift to calculate the new vma endpoints.
 541  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 542  *    arguments passed to subsequent functions are consistent.
 543  * 3) Move vma's page tables to the new range.
 544  * 4) Free up any cleared pgd range.
 545  * 5) Shrink the vma to cover only the new range.
 546  */
 547 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 548 {
 549         struct mm_struct *mm = vma->vm_mm;
 550         unsigned long old_start = vma->vm_start;
 551         unsigned long old_end = vma->vm_end;
 552         unsigned long length = old_end - old_start;
 553         unsigned long new_start = old_start - shift;
 554         unsigned long new_end = old_end - shift;
 555         struct mmu_gather *tlb;
 556
 557         BUG_ON(new_start > new_end);
 558
 559         /*
 560          * ensure there are no vmas between where we want to go
 561          * and where we are
 562          */
 563         if (vma != find_vma(mm, new_start))
 564                 return -EFAULT;
 565
 566         /*
 567          * cover the whole range: [new_start, old_end)
 568          */
 569         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 570                 return -ENOMEM;
 571
 572         /*
 573          * move the page tables downwards, on failure we rely on
 574          * process cleanup to remove whatever mess we made.
 575          */
 576         if (length != move_page_tables(vma, old_start,
 577                                        vma, new_start, length))
 578                 return -ENOMEM;
 579
 580         lru_add_drain();
 581         tlb = tlb_gather_mmu(mm, 0);
 582         if (new_end > old_start) {
 583                 /*
 584                  * when the old and new regions overlap clear from new_end.
 585                  */
 586                 free_pgd_range(tlb, new_end, old_end, new_end,
 587                         vma->vm_next ? vma->vm_next->vm_start : 0);
 588         } else {
 589                 /*
 590                  * otherwise, clean from old_start; this is done to not touch
 591                  * the address space in [new_end, old_start) some architectures
 592                  * have constraints on va-space that make this illegal (IA64) -
 593                  * for the others its just a little faster.
 594                  */
 595                 free_pgd_range(tlb, old_start, old_end, new_end,
 596                         vma->vm_next ? vma->vm_next->vm_start : 0);
 597         }
 598         tlb_finish_mmu(tlb, new_end, old_end);
 599
 600         /*
 601          * Shrink the vma to just the new range.  Always succeeds.
 602          */
 603         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 604
 605         return 0;
 606 }
 607
 608 /*
 609  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 610  * the stack is optionally relocated, and some extra space is added.
 611  */
 612 int setup_arg_pages(struct linux_binprm *bprm,
 613                     unsigned long stack_top,
 614                     int executable_stack)
 615 {
 616         unsigned long ret;
 617         unsigned long stack_shift;
 618         struct mm_struct *mm = current->mm;
 619         struct vm_area_struct *vma = bprm->vma;
 620         struct vm_area_struct *prev = NULL;
 621         unsigned long vm_flags;
 622         unsigned long stack_base;
 623         unsigned long stack_size;
 624         unsigned long stack_expand;
 625         unsigned long rlim_stack;
 626
 627 #ifdef CONFIG_STACK_GROWSUP
 628         /* Limit stack size to 1GB */
 629         stack_base = rlimit_max(RLIMIT_STACK);
 630         if (stack_base > (1 << 30))
 631                 stack_base = 1 << 30;
 632
 633         /* Make sure we didn't let the argument array grow too large. */
 634         if (vma->vm_end - vma->vm_start > stack_base)
 635                 return -ENOMEM;
 636
 637         stack_base = PAGE_ALIGN(stack_top - stack_base);
 638
 639         stack_shift = vma->vm_start - stack_base;
 640         mm->arg_start = bprm->p - stack_shift;
 641         bprm->p = vma->vm_end - stack_shift;
 642 #else
 643         stack_top = arch_align_stack(stack_top);
 644         stack_top = PAGE_ALIGN(stack_top);
 645
 646         if (unlikely(stack_top < mmap_min_addr) ||
 647             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 648                 return -ENOMEM;
 649
 650         stack_shift = vma->vm_end - stack_top;
 651
 652         bprm->p -= stack_shift;
 653         mm->arg_start = bprm->p;
 654 #endif
 655
 656         if (bprm->loader)
 657                 bprm->loader -= stack_shift;
 658         bprm->exec -= stack_shift;
 659
 660         down_write(&mm->mmap_sem);
 661         vm_flags = VM_STACK_FLAGS;
 662
 663         /*
 664          * Adjust stack execute permissions; explicitly enable for
 665          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 666          * (arch default) otherwise.
 667          */
 668         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 669                 vm_flags |= VM_EXEC;
 670         else if (executable_stack == EXSTACK_DISABLE_X)
 671                 vm_flags &= ~VM_EXEC;
 672         vm_flags |= mm->def_flags;
 673         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 674
 675         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 676                         vm_flags);
 677         if (ret)
 678                 goto out_unlock;
 679         BUG_ON(prev != vma);
 680
 681         /* Move stack pages down in memory. */
 682         if (stack_shift) {
 683                 ret = shift_arg_pages(vma, stack_shift);
 684                 if (ret)
 685                         goto out_unlock;
 686         }
 687
 688         /* mprotect_fixup is overkill to remove the temporary stack flags */
 689         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 690
 691         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 692         stack_size = vma->vm_end - vma->vm_start;
 693         /*
 694          * Align this down to a page boundary as expand_stack
 695          * will align it up.
 696          */
 697         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 698 #ifdef CONFIG_STACK_GROWSUP
 699         if (stack_size + stack_expand > rlim_stack)
 700                 stack_base = vma->vm_start + rlim_stack;
 701         else
 702                 stack_base = vma->vm_end + stack_expand;
 703 #else
 704         if (stack_size + stack_expand > rlim_stack)
 705                 stack_base = vma->vm_end - rlim_stack;
 706         else
 707                 stack_base = vma->vm_start - stack_expand;
 708 #endif
 709         current->mm->start_stack = bprm->p;
 710         ret = expand_stack(vma, stack_base);
 711         if (ret)
 712                 ret = -EFAULT;
 713
 714 out_unlock:
 715         up_write(&mm->mmap_sem);
 716         return ret;
 717 }
 718 EXPORT_SYMBOL(setup_arg_pages);
 719
 720 #endif /* CONFIG_MMU */
 721
 722 struct file *open_exec(const char *name)
 723 {
 724         struct file *file;
 725         int err;
 726
 727         file = do_filp_open(AT_FDCWD, name,
 728                                 O_LARGEFILE | O_RDONLY | __FMODE_EXEC, 0,
 729                                 MAY_EXEC | MAY_OPEN);
 730         if (IS_ERR(file))
 731                 goto out;
 732
 733         err = -EACCES;
 734         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 735                 goto exit;
 736
 737         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 738                 goto exit;
 739
 740         fsnotify_open(file);
 741
 742         trace_open_exec(name);
 743
 744         err = deny_write_access(file);
 745         if (err)
 746                 goto exit;
 747
 748 out:
 749         return file;
 750
 751 exit:
 752         fput(file);
 753         return ERR_PTR(err);
 754 }
 755 EXPORT_SYMBOL(open_exec);
 756
 757 int kernel_read(struct file *file, loff_t offset,
 758                 char *addr, unsigned long count)
 759 {
 760         mm_segment_t old_fs;
 761         loff_t pos = offset;
 762         int result;
 763
 764         old_fs = get_fs();
 765         set_fs(get_ds());
 766         /* The cast to a user pointer is valid due to the set_fs() */
 767         result = vfs_read(file, (void __user *)addr, count, &pos);
 768         set_fs(old_fs);
 769         return result;
 770 }
 771
 772 EXPORT_SYMBOL(kernel_read);
 773
 774 static int exec_mmap(struct mm_struct *mm)
 775 {
 776         struct task_struct *tsk;
 777         struct mm_struct * old_mm, *active_mm;
 778
 779         /* Notify parent that we're no longer interested in the old VM */
 780         tsk = current;
 781         old_mm = current->mm;
 782         sync_mm_rss(tsk, old_mm);
 783         mm_release(tsk, old_mm);
 784
 785         if (old_mm) {
 786                 /*
 787                  * Make sure that if there is a core dump in progress
 788                  * for the old mm, we get out and die instead of going
 789                  * through with the exec.  We must hold mmap_sem around
 790                  * checking core_state and changing tsk->mm.
 791                  */
 792                 down_read(&old_mm->mmap_sem);
 793                 if (unlikely(old_mm->core_state)) {
 794                         up_read(&old_mm->mmap_sem);
 795                         return -EINTR;
 796                 }
 797         }
 798         task_lock(tsk);
 799         active_mm = tsk->active_mm;
 800         tsk->mm = mm;
 801         tsk->active_mm = mm;
 802         activate_mm(active_mm, mm);
 803         if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
 804                 atomic_dec(&old_mm->oom_disable_count);
 805                 atomic_inc(&tsk->mm->oom_disable_count);
 806         }
 807         task_unlock(tsk);
 808         arch_pick_mmap_layout(mm);
 809         if (old_mm) {
 810                 up_read(&old_mm->mmap_sem);
 811                 BUG_ON(active_mm != old_mm);
 812                 mm_update_next_owner(old_mm);
 813                 mmput(old_mm);
 814                 return 0;
 815         }
 816         mmdrop(active_mm);
 817         return 0;
 818 }
 819
 820 /*
 821  * This function makes sure the current process has its own signal table,
 822  * so that flush_signal_handlers can later reset the handlers without
 823  * disturbing other processes.  (Other processes might share the signal
 824  * table via the CLONE_SIGHAND option to clone().)
 825  */
 826 static int de_thread(struct task_struct *tsk)
 827 {
 828         struct signal_struct *sig = tsk->signal;
 829         struct sighand_struct *oldsighand = tsk->sighand;
 830         spinlock_t *lock = &oldsighand->siglock;
 831
 832         if (thread_group_empty(tsk))
 833                 goto no_thread_group;
 834
 835         /*
 836          * Kill all other threads in the thread group.
 837          */
 838         spin_lock_irq(lock);
 839         if (signal_group_exit(sig)) {
 840                 /*
 841                  * Another group action in progress, just
 842                  * return so that the signal is processed.
 843                  */
 844                 spin_unlock_irq(lock);
 845                 return -EAGAIN;
 846         }
 847
 848         sig->group_exit_task = tsk;
 849         sig->notify_count = zap_other_threads(tsk);
 850         if (!thread_group_leader(tsk))
 851                 sig->notify_count--;
 852
 853         while (sig->notify_count) {
 854                 __set_current_state(TASK_UNINTERRUPTIBLE);
 855                 spin_unlock_irq(lock);
 856                 schedule();
 857                 spin_lock_irq(lock);
 858         }
 859         spin_unlock_irq(lock);
 860
 861         /*
 862          * At this point all other threads have exited, all we have to
 863          * do is to wait for the thread group leader to become inactive,
 864          * and to assume its PID:
 865          */
 866         if (!thread_group_leader(tsk)) {
 867                 struct task_struct *leader = tsk->group_leader;
 868
 869                 sig->notify_count = -1; /* for exit_notify() */
 870                 for (;;) {
 871                         write_lock_irq(&tasklist_lock);
 872                         if (likely(leader->exit_state))
 873                                 break;
 874                         __set_current_state(TASK_UNINTERRUPTIBLE);
 875                         write_unlock_irq(&tasklist_lock);
 876                         schedule();
 877                 }
 878
 879                 /*
 880                  * The only record we have of the real-time age of a
 881                  * process, regardless of execs it's done, is start_time.
 882                  * All the past CPU time is accumulated in signal_struct
 883                  * from sister threads now dead.  But in this non-leader
 884                  * exec, nothing survives from the original leader thread,
 885                  * whose birth marks the true age of this process now.
 886                  * When we take on its identity by switching to its PID, we
 887                  * also take its birthdate (always earlier than our own).
 888                  */
 889                 tsk->start_time = leader->start_time;
 890
 891                 BUG_ON(!same_thread_group(leader, tsk));
 892                 BUG_ON(has_group_leader_pid(tsk));
 893                 /*
 894                  * An exec() starts a new thread group with the
 895                  * TGID of the previous thread group. Rehash the
 896                  * two threads with a switched PID, and release
 897                  * the former thread group leader:
 898                  */
 899
 900                 /* Become a process group leader with the old leader's pid.
 901                  * The old leader becomes a thread of the this thread group.
 902                  * Note: The old leader also uses this pid until release_task
 903                  *       is called.  Odd but simple and correct.
 904                  */
 905                 detach_pid(tsk, PIDTYPE_PID);
 906                 tsk->pid = leader->pid;
 907                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 908                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 909                 transfer_pid(leader, tsk, PIDTYPE_SID);
 910
 911                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 912                 list_replace_init(&leader->sibling, &tsk->sibling);
 913
 914                 tsk->group_leader = tsk;
 915                 leader->group_leader = tsk;
 916
 917                 tsk->exit_signal = SIGCHLD;
 918
 919                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 920                 leader->exit_state = EXIT_DEAD;
 921                 write_unlock_irq(&tasklist_lock);
 922
 923                 release_task(leader);
 924         }
 925
 926         sig->group_exit_task = NULL;
 927         sig->notify_count = 0;
 928
 929 no_thread_group:
 930         if (current->mm)
 931                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 932
 933         exit_itimers(sig);
 934         flush_itimer_signals();
 935
 936         if (atomic_read(&oldsighand->count) != 1) {
 937                 struct sighand_struct *newsighand;
 938                 /*
 939                  * This ->sighand is shared with the CLONE_SIGHAND
 940                  * but not CLONE_THREAD task, switch to the new one.
 941                  */
 942                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 943                 if (!newsighand)
 944                         return -ENOMEM;
 945
 946                 atomic_set(&newsighand->count, 1);
 947                 memcpy(newsighand->action, oldsighand->action,
 948                        sizeof(newsighand->action));
 949
 950                 write_lock_irq(&tasklist_lock);
 951                 spin_lock(&oldsighand->siglock);
 952                 rcu_assign_pointer(tsk->sighand, newsighand);
 953                 spin_unlock(&oldsighand->siglock);
 954                 write_unlock_irq(&tasklist_lock);
 955
 956                 __cleanup_sighand(oldsighand);
 957         }
 958
 959         BUG_ON(!thread_group_leader(tsk));
 960         return 0;
 961 }
 962
 963 /*
 964  * These functions flushes out all traces of the currently running executable
 965  * so that a new one can be started
 966  */
 967 static void flush_old_files(struct files_struct * files)
 968 {
 969         long j = -1;
 970         struct fdtable *fdt;
 971
 972         spin_lock(&files->file_lock);
 973         for (;;) {
 974                 unsigned long set, i;
 975
 976                 j++;
 977                 i = j * __NFDBITS;
 978                 fdt = files_fdtable(files);
 979                 if (i >= fdt->max_fds)
 980                         break;
 981                 set = fdt->close_on_exec->fds_bits[j];
 982                 if (!set)
 983                         continue;
 984                 fdt->close_on_exec->fds_bits[j] = 0;
 985                 spin_unlock(&files->file_lock);
 986                 for ( ; set ; i++,set >>= 1) {
 987                         if (set & 1) {
 988                                 sys_close(i);
 989                         }
 990                 }
 991                 spin_lock(&files->file_lock);
 992
 993         }
 994         spin_unlock(&files->file_lock);
 995 }
 996
 997 char *get_task_comm(char *buf, struct task_struct *tsk)
 998 {
 999         /* buf must be at least sizeof(tsk->comm) in size */
1000         task_lock(tsk);
1001         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1002         task_unlock(tsk);
1003         return buf;
1004 }
1005
1006 void set_task_comm(struct task_struct *tsk, char *buf)
1007 {
1008         task_lock(tsk);
1009
1010         /*
1011          * Threads may access current->comm without holding
1012          * the task lock, so write the string carefully.
1013          * Readers without a lock may see incomplete new
1014          * names but are safe from non-terminating string reads.
1015          */
1016         memset(tsk->comm, 0, TASK_COMM_LEN);
1017         wmb();
1018         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1019         task_unlock(tsk);
1020         perf_event_comm(tsk);
1021 }
1022
1023 int flush_old_exec(struct linux_binprm * bprm)
1024 {
1025         int retval;
1026
1027         /*
1028          * Make sure we have a private signal table and that
1029          * we are unassociated from the previous thread group.
1030          */
1031         retval = de_thread(current);
1032         if (retval)
1033                 goto out;
1034
1035         set_mm_exe_file(bprm->mm, bprm->file);
1036
1037         /*
1038          * Release all of the old mmap stuff
1039          */
1040         acct_arg_size(bprm, 0);
1041         retval = exec_mmap(bprm->mm);
1042         if (retval)
1043                 goto out;
1044
1045         bprm->mm = NULL;                /* We're using it now */
1046
1047         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1048         flush_thread();
1049         current->personality &= ~bprm->per_clear;
1050
1051         return 0;
1052
1053 out:
1054         return retval;
1055 }
1056 EXPORT_SYMBOL(flush_old_exec);
1057
1058 void setup_new_exec(struct linux_binprm * bprm)
1059 {
1060         int i, ch;
1061         const char *name;
1062         char tcomm[sizeof(current->comm)];
1063
1064         arch_pick_mmap_layout(current->mm);
1065
1066         /* This is the point of no return */
1067         current->sas_ss_sp = current->sas_ss_size = 0;
1068
1069         if (current_euid() == current_uid() && current_egid() == current_gid())
1070                 set_dumpable(current->mm, 1);
1071         else
1072                 set_dumpable(current->mm, suid_dumpable);
1073
1074         name = bprm->filename;
1075
1076         /* Copies the binary name from after last slash */
1077         for (i=0; (ch = *(name++)) != '\0';) {
1078                 if (ch == '/')
1079                         i = 0; /* overwrite what we wrote */
1080                 else
1081                         if (i < (sizeof(tcomm) - 1))
1082                                 tcomm[i++] = ch;
1083         }
1084         tcomm[i] = '\0';
1085         set_task_comm(current, tcomm);
1086
1087         /* Set the new mm task size. We have to do that late because it may
1088          * depend on TIF_32BIT which is only updated in flush_thread() on
1089          * some architectures like powerpc
1090          */
1091         current->mm->task_size = TASK_SIZE;
1092
1093         /* install the new credentials */
1094         if (bprm->cred->uid != current_euid() ||
1095             bprm->cred->gid != current_egid()) {
1096                 current->pdeath_signal = 0;
1097         } else if (file_permission(bprm->file, MAY_READ) ||
1098                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1099                 set_dumpable(current->mm, suid_dumpable);
1100         }
1101
1102         /*
1103          * Flush performance counters when crossing a
1104          * security domain:
1105          */
1106         if (!get_dumpable(current->mm))
1107                 perf_event_exit_task(current);
1108
1109         /* An exec changes our domain. We are no longer part of the thread
1110            group */
1111
1112         current->self_exec_id++;
1113
1114         flush_signal_handlers(current, 0);
1115         flush_old_files(current->files);
1116 }
1117 EXPORT_SYMBOL(setup_new_exec);
1118
1119 /*
1120  * Prepare credentials and lock ->cred_guard_mutex.
1121  * install_exec_creds() commits the new creds and drops the lock.
1122  * Or, if exec fails before, free_bprm() should release ->cred and
1123  * and unlock.
1124  */
1125 int prepare_bprm_creds(struct linux_binprm *bprm)
1126 {
1127         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1128                 return -ERESTARTNOINTR;
1129
1130         bprm->cred = prepare_exec_creds();
1131         if (likely(bprm->cred))
1132                 return 0;
1133
1134         mutex_unlock(&current->signal->cred_guard_mutex);
1135         return -ENOMEM;
1136 }
1137
1138 void free_bprm(struct linux_binprm *bprm)
1139 {
1140         free_arg_pages(bprm);
1141         if (bprm->cred) {
1142                 mutex_unlock(&current->signal->cred_guard_mutex);
1143                 abort_creds(bprm->cred);
1144         }
1145         kfree(bprm);
1146 }
1147
1148 /*
1149  * install the new credentials for this executable
1150  */
1151 void install_exec_creds(struct linux_binprm *bprm)
1152 {
1153         security_bprm_committing_creds(bprm);
1154
1155         commit_creds(bprm->cred);
1156         bprm->cred = NULL;
1157         /*
1158          * cred_guard_mutex must be held at least to this point to prevent
1159          * ptrace_attach() from altering our determination of the task's
1160          * credentials; any time after this it may be unlocked.
1161          */
1162         security_bprm_committed_creds(bprm);
1163         mutex_unlock(&current->signal->cred_guard_mutex);
1164 }
1165 EXPORT_SYMBOL(install_exec_creds);
1166
1167 /*
1168  * determine how safe it is to execute the proposed program
1169  * - the caller must hold ->cred_guard_mutex to protect against
1170  *   PTRACE_ATTACH
1171  */
1172 int check_unsafe_exec(struct linux_binprm *bprm)
1173 {
1174         struct task_struct *p = current, *t;
1175         unsigned n_fs;
1176         int res = 0;
1177
1178         bprm->unsafe = tracehook_unsafe_exec(p);
1179
1180         n_fs = 1;
1181         spin_lock(&p->fs->lock);
1182         rcu_read_lock();
1183         for (t = next_thread(p); t != p; t = next_thread(t)) {
1184                 if (t->fs == p->fs)
1185                         n_fs++;
1186         }
1187         rcu_read_unlock();
1188
1189         if (p->fs->users > n_fs) {
1190                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1191         } else {
1192                 res = -EAGAIN;
1193                 if (!p->fs->in_exec) {
1194                         p->fs->in_exec = 1;
1195                         res = 1;
1196                 }
1197         }
1198         spin_unlock(&p->fs->lock);
1199
1200         return res;
1201 }
1202
1203 /*
1204  * Fill the binprm structure from the inode.
1205  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1206  *
1207  * This may be called multiple times for binary chains (scripts for example).
1208  */
1209 int prepare_binprm(struct linux_binprm *bprm)
1210 {
1211         umode_t mode;
1212         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1213         int retval;
1214
1215         mode = inode->i_mode;
1216         if (bprm->file->f_op == NULL)
1217                 return -EACCES;
1218
1219         /* clear any previous set[ug]id data from a previous binary */
1220         bprm->cred->euid = current_euid();
1221         bprm->cred->egid = current_egid();
1222
1223         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1224                 /* Set-uid? */
1225                 if (mode & S_ISUID) {
1226                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1227                         bprm->cred->euid = inode->i_uid;
1228                 }
1229
1230                 /* Set-gid? */
1231                 /*
1232                  * If setgid is set but no group execute bit then this
1233                  * is a candidate for mandatory locking, not a setgid
1234                  * executable.
1235                  */
1236                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1237                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1238                         bprm->cred->egid = inode->i_gid;
1239                 }
1240         }
1241
1242         /* fill in binprm security blob */
1243         retval = security_bprm_set_creds(bprm);
1244         if (retval)
1245                 return retval;
1246         bprm->cred_prepared = 1;
1247
1248         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1249         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1250 }
1251
1252 EXPORT_SYMBOL(prepare_binprm);
1253
1254 /*
1255  * Arguments are '\0' separated strings found at the location bprm->p
1256  * points to; chop off the first by relocating brpm->p to right after
1257  * the first '\0' encountered.
1258  */
1259 int remove_arg_zero(struct linux_binprm *bprm)
1260 {
1261         int ret = 0;
1262         unsigned long offset;
1263         char *kaddr;
1264         struct page *page;
1265
1266         if (!bprm->argc)
1267                 return 0;
1268
1269         do {
1270                 offset = bprm->p & ~PAGE_MASK;
1271                 page = get_arg_page(bprm, bprm->p, 0);
1272                 if (!page) {
1273                         ret = -EFAULT;
1274                         goto out;
1275                 }
1276                 kaddr = kmap_atomic(page, KM_USER0);
1277
1278                 for (; offset < PAGE_SIZE && kaddr[offset];
1279                                 offset++, bprm->p++)
1280                         ;
1281
1282                 kunmap_atomic(kaddr, KM_USER0);
1283                 put_arg_page(page);
1284
1285                 if (offset == PAGE_SIZE)
1286                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1287         } while (offset == PAGE_SIZE);
1288
1289         bprm->p++;
1290         bprm->argc--;
1291         ret = 0;
1292
1293 out:
1294         return ret;
1295 }
1296 EXPORT_SYMBOL(remove_arg_zero);
1297
1298 /*
1299  * cycle the list of binary formats handler, until one recognizes the image
1300  */
1301 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1302 {
1303         unsigned int depth = bprm->recursion_depth;
1304         int try,retval;
1305         struct linux_binfmt *fmt;
1306
1307         retval = security_bprm_check(bprm);
1308         if (retval)
1309                 return retval;
1310
1311         /* kernel module loader fixup */
1312         /* so we don't try to load run modprobe in kernel space. */
1313         set_fs(USER_DS);
1314
1315         retval = audit_bprm(bprm);
1316         if (retval)
1317                 return retval;
1318
1319         retval = -ENOENT;
1320         for (try=0; try<2; try++) {
1321                 read_lock(&binfmt_lock);
1322                 list_for_each_entry(fmt, &formats, lh) {
1323                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1324                         if (!fn)
1325                                 continue;
1326                         if (!try_module_get(fmt->module))
1327                                 continue;
1328                         read_unlock(&binfmt_lock);
1329                         retval = fn(bprm, regs);
1330                         /*
1331                          * Restore the depth counter to its starting value
1332                          * in this call, so we don't have to rely on every
1333                          * load_binary function to restore it on return.
1334                          */
1335                         bprm->recursion_depth = depth;
1336                         if (retval >= 0) {
1337                                 if (depth == 0)
1338                                         tracehook_report_exec(fmt, bprm, regs);
1339                                 put_binfmt(fmt);
1340                                 allow_write_access(bprm->file);
1341                                 if (bprm->file)
1342                                         fput(bprm->file);
1343                                 bprm->file = NULL;
1344                                 current->did_exec = 1;
1345                                 proc_exec_connector(current);
1346                                 return retval;
1347                         }
1348                         read_lock(&binfmt_lock);
1349                         put_binfmt(fmt);
1350                         if (retval != -ENOEXEC || bprm->mm == NULL)
1351                                 break;
1352                         if (!bprm->file) {
1353                                 read_unlock(&binfmt_lock);
1354                                 return retval;
1355                         }
1356                 }
1357                 read_unlock(&binfmt_lock);
1358                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1359                         break;
1360 #ifdef CONFIG_MODULES
1361                 } else {
1362 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1363                         if (printable(bprm->buf[0]) &&
1364                             printable(bprm->buf[1]) &&
1365                             printable(bprm->buf[2]) &&
1366                             printable(bprm->buf[3]))
1367                                 break; /* -ENOEXEC */
1368                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1369 #endif
1370                 }
1371         }
1372         return retval;
1373 }
1374
1375 EXPORT_SYMBOL(search_binary_handler);
1376
1377 /*
1378  * sys_execve() executes a new program.
1379  */
1380 int do_execve(const char * filename,
1381         const char __user *const __user *argv,
1382         const char __user *const __user *envp,
1383         struct pt_regs * regs)
1384 {
1385         struct linux_binprm *bprm;
1386         struct file *file;
1387         struct files_struct *displaced;
1388         bool clear_in_exec;
1389         int retval;
1390
1391         retval = unshare_files(&displaced);
1392         if (retval)
1393                 goto out_ret;
1394
1395         retval = -ENOMEM;
1396         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1397         if (!bprm)
1398                 goto out_files;
1399
1400         retval = prepare_bprm_creds(bprm);
1401         if (retval)
1402                 goto out_free;
1403
1404         retval = check_unsafe_exec(bprm);
1405         if (retval < 0)
1406                 goto out_free;
1407         clear_in_exec = retval;
1408         current->in_execve = 1;
1409
1410         file = open_exec(filename);
1411         retval = PTR_ERR(file);
1412         if (IS_ERR(file))
1413                 goto out_unmark;
1414
1415         sched_exec();
1416
1417         bprm->file = file;
1418         bprm->filename = filename;
1419         bprm->interp = filename;
1420
1421         retval = bprm_mm_init(bprm);
1422         if (retval)
1423                 goto out_file;
1424
1425         bprm->argc = count(argv, MAX_ARG_STRINGS);
1426         if ((retval = bprm->argc) < 0)
1427                 goto out;
1428
1429         bprm->envc = count(envp, MAX_ARG_STRINGS);
1430         if ((retval = bprm->envc) < 0)
1431                 goto out;
1432
1433         retval = prepare_binprm(bprm);
1434         if (retval < 0)
1435                 goto out;
1436
1437         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1438         if (retval < 0)
1439                 goto out;
1440
1441         bprm->exec = bprm->p;
1442         retval = copy_strings(bprm->envc, envp, bprm);
1443         if (retval < 0)
1444                 goto out;
1445
1446         retval = copy_strings(bprm->argc, argv, bprm);
1447         if (retval < 0)
1448                 goto out;
1449
1450         retval = search_binary_handler(bprm,regs);
1451         if (retval < 0)
1452                 goto out;
1453
1454         /* execve succeeded */
1455         current->fs->in_exec = 0;
1456         current->in_execve = 0;
1457         acct_update_integrals(current);
1458         free_bprm(bprm);
1459         if (displaced)
1460                 put_files_struct(displaced);
1461         return retval;
1462
1463 out:
1464         if (bprm->mm) {
1465                 acct_arg_size(bprm, 0);
1466                 mmput(bprm->mm);
1467         }
1468
1469 out_file:
1470         if (bprm->file) {
1471                 allow_write_access(bprm->file);
1472                 fput(bprm->file);
1473         }
1474
1475 out_unmark:
1476         if (clear_in_exec)
1477                 current->fs->in_exec = 0;
1478         current->in_execve = 0;
1479
1480 out_free:
1481         free_bprm(bprm);
1482
1483 out_files:
1484         if (displaced)
1485                 reset_files_struct(displaced);
1486 out_ret:
1487         return retval;
1488 }
1489
1490 void set_binfmt(struct linux_binfmt *new)
1491 {
1492         struct mm_struct *mm = current->mm;
1493
1494         if (mm->binfmt)
1495                 module_put(mm->binfmt->module);
1496
1497         mm->binfmt = new;
1498         if (new)
1499                 __module_get(new->module);
1500 }
1501
1502 EXPORT_SYMBOL(set_binfmt);
1503
1504 static int expand_corename(struct core_name *cn)
1505 {
1506         char *old_corename = cn->corename;
1507
1508         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1509         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1510
1511         if (!cn->corename) {
1512                 kfree(old_corename);
1513                 return -ENOMEM;
1514         }
1515
1516         return 0;
1517 }
1518
1519 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1520 {
1521         char *cur;
1522         int need;
1523         int ret;
1524         va_list arg;
1525
1526         va_start(arg, fmt);
1527         need = vsnprintf(NULL, 0, fmt, arg);
1528         va_end(arg);
1529
1530         if (likely(need < cn->size - cn->used - 1))
1531                 goto out_printf;
1532
1533         ret = expand_corename(cn);
1534         if (ret)
1535                 goto expand_fail;
1536
1537 out_printf:
1538         cur = cn->corename + cn->used;
1539         va_start(arg, fmt);
1540         vsnprintf(cur, need + 1, fmt, arg);
1541         va_end(arg);
1542         cn->used += need;
1543         return 0;
1544
1545 expand_fail:
1546         return ret;
1547 }
1548
1549 /* format_corename will inspect the pattern parameter, and output a
1550  * name into corename, which must have space for at least
1551  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1552  */
1553 static int format_corename(struct core_name *cn, long signr)
1554 {
1555         const struct cred *cred = current_cred();
1556         const char *pat_ptr = core_pattern;
1557         int ispipe = (*pat_ptr == '|');
1558         int pid_in_pattern = 0;
1559         int err = 0;
1560
1561         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1562         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1563         cn->used = 0;
1564
1565         if (!cn->corename)
1566                 return -ENOMEM;
1567
1568         /* Repeat as long as we have more pattern to process and more output
1569            space */
1570         while (*pat_ptr) {
1571                 if (*pat_ptr != '%') {
1572                         if (*pat_ptr == 0)
1573                                 goto out;
1574                         err = cn_printf(cn, "%c", *pat_ptr++);
1575                 } else {
1576                         switch (*++pat_ptr) {
1577                         /* single % at the end, drop that */
1578                         case 0:
1579                                 goto out;
1580                         /* Double percent, output one percent */
1581                         case '%':
1582                                 err = cn_printf(cn, "%c", '%');
1583                                 break;
1584                         /* pid */
1585                         case 'p':
1586                                 pid_in_pattern = 1;
1587                                 err = cn_printf(cn, "%d",
1588                                               task_tgid_vnr(current));
1589                                 break;
1590                         /* uid */
1591                         case 'u':
1592                                 err = cn_printf(cn, "%d", cred->uid);
1593                                 break;
1594                         /* gid */
1595                         case 'g':
1596                                 err = cn_printf(cn, "%d", cred->gid);
1597                                 break;
1598                         /* signal that caused the coredump */
1599                         case 's':
1600                                 err = cn_printf(cn, "%ld", signr);
1601                                 break;
1602                         /* UNIX time of coredump */
1603                         case 't': {
1604                                 struct timeval tv;
1605                                 do_gettimeofday(&tv);
1606                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1607                                 break;
1608                         }
1609                         /* hostname */
1610                         case 'h':
1611                                 down_read(&uts_sem);
1612                                 err = cn_printf(cn, "%s",
1613                                               utsname()->nodename);
1614                                 up_read(&uts_sem);
1615                                 break;
1616                         /* executable */
1617                         case 'e':
1618                                 err = cn_printf(cn, "%s", current->comm);
1619                                 break;
1620                         /* core limit size */
1621                         case 'c':
1622                                 err = cn_printf(cn, "%lu",
1623                                               rlimit(RLIMIT_CORE));
1624                                 break;
1625                         default:
1626                                 break;
1627                         }
1628                         ++pat_ptr;
1629                 }
1630
1631                 if (err)
1632                         return err;
1633         }
1634
1635         /* Backward compatibility with core_uses_pid:
1636          *
1637          * If core_pattern does not include a %p (as is the default)
1638          * and core_uses_pid is set, then .%pid will be appended to
1639          * the filename. Do not do this for piped commands. */
1640         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1641                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1642                 if (err)
1643                         return err;
1644         }
1645 out:
1646         return ispipe;
1647 }
1648
1649 static int zap_process(struct task_struct *start, int exit_code)
1650 {
1651         struct task_struct *t;
1652         int nr = 0;
1653
1654         start->signal->flags = SIGNAL_GROUP_EXIT;
1655         start->signal->group_exit_code = exit_code;
1656         start->signal->group_stop_count = 0;
1657
1658         t = start;
1659         do {
1660                 if (t != current && t->mm) {
1661                         sigaddset(&t->pending.signal, SIGKILL);
1662                         signal_wake_up(t, 1);
1663                         nr++;
1664                 }
1665         } while_each_thread(start, t);
1666
1667         return nr;
1668 }
1669
1670 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1671                                 struct core_state *core_state, int exit_code)
1672 {
1673         struct task_struct *g, *p;
1674         unsigned long flags;
1675         int nr = -EAGAIN;
1676
1677         spin_lock_irq(&tsk->sighand->siglock);
1678         if (!signal_group_exit(tsk->signal)) {
1679                 mm->core_state = core_state;
1680                 nr = zap_process(tsk, exit_code);
1681         }
1682         spin_unlock_irq(&tsk->sighand->siglock);
1683         if (unlikely(nr < 0))
1684                 return nr;
1685
1686         if (atomic_read(&mm->mm_users) == nr + 1)
1687                 goto done;
1688         /*
1689          * We should find and kill all tasks which use this mm, and we should
1690          * count them correctly into ->nr_threads. We don't take tasklist
1691          * lock, but this is safe wrt:
1692          *
1693          * fork:
1694          *      None of sub-threads can fork after zap_process(leader). All
1695          *      processes which were created before this point should be
1696          *      visible to zap_threads() because copy_process() adds the new
1697          *      process to the tail of init_task.tasks list, and lock/unlock
1698          *      of ->siglock provides a memory barrier.
1699          *
1700          * do_exit:
1701          *      The caller holds mm->mmap_sem. This means that the task which
1702          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1703          *      its ->mm.
1704          *
1705          * de_thread:
1706          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1707          *      we must see either old or new leader, this does not matter.
1708          *      However, it can change p->sighand, so lock_task_sighand(p)
1709          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1710          *      it can't fail.
1711          *
1712          *      Note also that "g" can be the old leader with ->mm == NULL
1713          *      and already unhashed and thus removed from ->thread_group.
1714          *      This is OK, __unhash_process()->list_del_rcu() does not
1715          *      clear the ->next pointer, we will find the new leader via
1716          *      next_thread().
1717          */
1718         rcu_read_lock();
1719         for_each_process(g) {
1720                 if (g == tsk->group_leader)
1721                         continue;
1722                 if (g->flags & PF_KTHREAD)
1723                         continue;
1724                 p = g;
1725                 do {
1726                         if (p->mm) {
1727                                 if (unlikely(p->mm == mm)) {
1728                                         lock_task_sighand(p, &flags);
1729                                         nr += zap_process(p, exit_code);
1730                                         unlock_task_sighand(p, &flags);
1731                                 }
1732                                 break;
1733                         }
1734                 } while_each_thread(g, p);
1735         }
1736         rcu_read_unlock();
1737 done:
1738         atomic_set(&core_state->nr_threads, nr);
1739         return nr;
1740 }
1741
1742 static int coredump_wait(int exit_code, struct core_state *core_state)
1743 {
1744         struct task_struct *tsk = current;
1745         struct mm_struct *mm = tsk->mm;
1746         struct completion *vfork_done;
1747         int core_waiters = -EBUSY;
1748
1749         init_completion(&core_state->startup);
1750         core_state->dumper.task = tsk;
1751         core_state->dumper.next = NULL;
1752
1753         down_write(&mm->mmap_sem);
1754         if (!mm->core_state)
1755                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1756         up_write(&mm->mmap_sem);
1757
1758         if (unlikely(core_waiters < 0))
1759                 goto fail;
1760
1761         /*
1762          * Make sure nobody is waiting for us to release the VM,
1763          * otherwise we can deadlock when we wait on each other
1764          */
1765         vfork_done = tsk->vfork_done;
1766         if (vfork_done) {
1767                 tsk->vfork_done = NULL;
1768                 complete(vfork_done);
1769         }
1770
1771         if (core_waiters)
1772                 wait_for_completion(&core_state->startup);
1773 fail:
1774         return core_waiters;
1775 }
1776
1777 static void coredump_finish(struct mm_struct *mm)
1778 {
1779         struct core_thread *curr, *next;
1780         struct task_struct *task;
1781
1782         next = mm->core_state->dumper.next;
1783         while ((curr = next) != NULL) {
1784                 next = curr->next;
1785                 task = curr->task;
1786                 /*
1787                  * see exit_mm(), curr->task must not see
1788                  * ->task == NULL before we read ->next.
1789                  */
1790                 smp_mb();
1791                 curr->task = NULL;
1792                 wake_up_process(task);
1793         }
1794
1795         mm->core_state = NULL;
1796 }
1797
1798 /*
1799  * set_dumpable converts traditional three-value dumpable to two flags and
1800  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1801  * these bits are not changed atomically.  So get_dumpable can observe the
1802  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1803  * return either old dumpable or new one by paying attention to the order of
1804  * modifying the bits.
1805  *
1806  * dumpable |   mm->flags (binary)
1807  * old  new | initial interim  final
1808  * ---------+-----------------------
1809  *  0    1  |   00      01      01
1810  *  0    2  |   00      10(*)   11
1811  *  1    0  |   01      00      00
1812  *  1    2  |   01      11      11
1813  *  2    0  |   11      10(*)   00
1814  *  2    1  |   11      11      01
1815  *
1816  * (*) get_dumpable regards interim value of 10 as 11.
1817  */
1818 void set_dumpable(struct mm_struct *mm, int value)
1819 {
1820         switch (value) {
1821         case 0:
1822                 clear_bit(MMF_DUMPABLE, &mm->flags);
1823                 smp_wmb();
1824                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1825                 break;
1826         case 1:
1827                 set_bit(MMF_DUMPABLE, &mm->flags);
1828                 smp_wmb();
1829                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1830                 break;
1831         case 2:
1832                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1833                 smp_wmb();
1834                 set_bit(MMF_DUMPABLE, &mm->flags);
1835                 break;
1836         }
1837 }
1838
1839 static int __get_dumpable(unsigned long mm_flags)
1840 {
1841         int ret;
1842
1843         ret = mm_flags & MMF_DUMPABLE_MASK;
1844         return (ret >= 2) ? 2 : ret;
1845 }
1846
1847 int get_dumpable(struct mm_struct *mm)
1848 {
1849         return __get_dumpable(mm->flags);
1850 }
1851
1852 static void wait_for_dump_helpers(struct file *file)
1853 {
1854         struct pipe_inode_info *pipe;
1855
1856         pipe = file->f_path.dentry->d_inode->i_pipe;
1857
1858         pipe_lock(pipe);
1859         pipe->readers++;
1860         pipe->writers--;
1861
1862         while ((pipe->readers > 1) && (!signal_pending(current))) {
1863                 wake_up_interruptible_sync(&pipe->wait);
1864                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1865                 pipe_wait(pipe);
1866         }
1867
1868         pipe->readers--;
1869         pipe->writers++;
1870         pipe_unlock(pipe);
1871
1872 }
1873
1874
1875 /*
1876  * uhm_pipe_setup
1877  * helper function to customize the process used
1878  * to collect the core in userspace.  Specifically
1879  * it sets up a pipe and installs it as fd 0 (stdin)
1880  * for the process.  Returns 0 on success, or
1881  * PTR_ERR on failure.
1882  * Note that it also sets the core limit to 1.  This
1883  * is a special value that we use to trap recursive
1884  * core dumps
1885  */
1886 static int umh_pipe_setup(struct subprocess_info *info)
1887 {
1888         struct file *rp, *wp;
1889         struct fdtable *fdt;
1890         struct coredump_params *cp = (struct coredump_params *)info->data;
1891         struct files_struct *cf = current->files;
1892
1893         wp = create_write_pipe(0);
1894         if (IS_ERR(wp))
1895                 return PTR_ERR(wp);
1896
1897         rp = create_read_pipe(wp, 0);
1898         if (IS_ERR(rp)) {
1899                 free_write_pipe(wp);
1900                 return PTR_ERR(rp);
1901         }
1902
1903         cp->file = wp;
1904
1905         sys_close(0);
1906         fd_install(0, rp);
1907         spin_lock(&cf->file_lock);
1908         fdt = files_fdtable(cf);
1909         FD_SET(0, fdt->open_fds);
1910         FD_CLR(0, fdt->close_on_exec);
1911         spin_unlock(&cf->file_lock);
1912
1913         /* and disallow core files too */
1914         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1915
1916         return 0;
1917 }
1918
1919 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1920 {
1921         struct core_state core_state;
1922         struct core_name cn;
1923         struct mm_struct *mm = current->mm;
1924         struct linux_binfmt * binfmt;
1925         const struct cred *old_cred;
1926         struct cred *cred;
1927         int retval = 0;
1928         int flag = 0;
1929         int ispipe;
1930         static atomic_t core_dump_count = ATOMIC_INIT(0);
1931         struct coredump_params cprm = {
1932                 .signr = signr,
1933                 .regs = regs,
1934                 .limit = rlimit(RLIMIT_CORE),
1935                 /*
1936                  * We must use the same mm->flags while dumping core to avoid
1937                  * inconsistency of bit flags, since this flag is not protected
1938                  * by any locks.
1939                  */
1940                 .mm_flags = mm->flags,
1941         };
1942
1943         audit_core_dumps(signr);
1944
1945         binfmt = mm->binfmt;
1946         if (!binfmt || !binfmt->core_dump)
1947                 goto fail;
1948         if (!__get_dumpable(cprm.mm_flags))
1949                 goto fail;
1950
1951         cred = prepare_creds();
1952         if (!cred)
1953                 goto fail;
1954         /*
1955          *      We cannot trust fsuid as being the "true" uid of the
1956          *      process nor do we know its entire history. We only know it
1957          *      was tainted so we dump it as root in mode 2.
1958          */
1959         if (__get_dumpable(cprm.mm_flags) == 2) {
1960                 /* Setuid core dump mode */
1961                 flag = O_EXCL;          /* Stop rewrite attacks */
1962                 cred->fsuid = 0;        /* Dump root private */
1963         }
1964
1965         retval = coredump_wait(exit_code, &core_state);
1966         if (retval < 0)
1967                 goto fail_creds;
1968
1969         old_cred = override_creds(cred);
1970
1971         /*
1972          * Clear any false indication of pending signals that might
1973          * be seen by the filesystem code called to write the core file.
1974          */
1975         clear_thread_flag(TIF_SIGPENDING);
1976
1977         ispipe = format_corename(&cn, signr);
1978
1979         if (ispipe == -ENOMEM) {
1980                 printk(KERN_WARNING "format_corename failed\n");
1981                 printk(KERN_WARNING "Aborting core\n");
1982                 goto fail_corename;
1983         }
1984
1985         if (ispipe) {
1986                 int dump_count;
1987                 char **helper_argv;
1988
1989                 if (cprm.limit == 1) {
1990                         /*
1991                          * Normally core limits are irrelevant to pipes, since
1992                          * we're not writing to the file system, but we use
1993                          * cprm.limit of 1 here as a speacial value. Any
1994                          * non-1 limit gets set to RLIM_INFINITY below, but
1995                          * a limit of 0 skips the dump.  This is a consistent
1996                          * way to catch recursive crashes.  We can still crash
1997                          * if the core_pattern binary sets RLIM_CORE =  !1
1998                          * but it runs as root, and can do lots of stupid things
1999                          * Note that we use task_tgid_vnr here to grab the pid
2000                          * of the process group leader.  That way we get the
2001                          * right pid if a thread in a multi-threaded
2002                          * core_pattern process dies.
2003                          */
2004                         printk(KERN_WARNING
2005                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2006                                 task_tgid_vnr(current), current->comm);
2007                         printk(KERN_WARNING "Aborting core\n");
2008                         goto fail_unlock;
2009                 }
2010                 cprm.limit = RLIM_INFINITY;
2011
2012                 dump_count = atomic_inc_return(&core_dump_count);
2013                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2014                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2015                                task_tgid_vnr(current), current->comm);
2016                         printk(KERN_WARNING "Skipping core dump\n");
2017                         goto fail_dropcount;
2018                 }
2019
2020                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2021                 if (!helper_argv) {
2022                         printk(KERN_WARNING "%s failed to allocate memory\n",
2023                                __func__);
2024                         goto fail_dropcount;
2025                 }
2026
2027                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2028                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2029                                         NULL, &cprm);
2030                 argv_free(helper_argv);
2031                 if (retval) {
2032                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2033                                cn.corename);
2034                         goto close_fail;
2035                 }
2036         } else {
2037                 struct inode *inode;
2038
2039                 if (cprm.limit < binfmt->min_coredump)
2040                         goto fail_unlock;
2041
2042                 cprm.file = filp_open(cn.corename,
2043                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2044                                  0600);
2045                 if (IS_ERR(cprm.file))
2046                         goto fail_unlock;
2047
2048                 inode = cprm.file->f_path.dentry->d_inode;
2049                 if (inode->i_nlink > 1)
2050                         goto close_fail;
2051                 if (d_unhashed(cprm.file->f_path.dentry))
2052                         goto close_fail;
2053                 /*
2054                  * AK: actually i see no reason to not allow this for named
2055                  * pipes etc, but keep the previous behaviour for now.
2056                  */
2057                 if (!S_ISREG(inode->i_mode))
2058                         goto close_fail;
2059                 /*
2060                  * Dont allow local users get cute and trick others to coredump
2061                  * into their pre-created files.
2062                  */
2063                 if (inode->i_uid != current_fsuid())
2064                         goto close_fail;
2065                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2066                         goto close_fail;
2067                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2068                         goto close_fail;
2069         }
2070
2071         retval = binfmt->core_dump(&cprm);
2072         if (retval)
2073                 current->signal->group_exit_code |= 0x80;
2074
2075         if (ispipe && core_pipe_limit)
2076                 wait_for_dump_helpers(cprm.file);
2077 close_fail:
2078         if (cprm.file)
2079                 filp_close(cprm.file, NULL);
2080 fail_dropcount:
2081         if (ispipe)
2082                 atomic_dec(&core_dump_count);
2083 fail_unlock:
2084         kfree(cn.corename);
2085 fail_corename:
2086         coredump_finish(mm);
2087         revert_creds(old_cred);
2088 fail_creds:
2089         put_cred(cred);
2090 fail:
2091         return;
2092 }
2093
2094 /*
2095  * Core dumping helper functions.  These are the only things you should
2096  * do on a core-file: use only these functions to write out all the
2097  * necessary info.
2098  */
2099 int dump_write(struct file *file, const void *addr, int nr)
2100 {
2101         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2102 }
2103 EXPORT_SYMBOL(dump_write);
2104
2105 int dump_seek(struct file *file, loff_t off)
2106 {
2107         int ret = 1;
2108
2109         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2110                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2111                         return 0;
2112         } else {
2113                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2114
2115                 if (!buf)
2116                         return 0;
2117                 while (off > 0) {
2118                         unsigned long n = off;
2119
2120                         if (n > PAGE_SIZE)
2121                                 n = PAGE_SIZE;
2122                         if (!dump_write(file, buf, n)) {
2123                                 ret = 0;
2124                                 break;
2125                         }
2126                         off -= n;
2127                 }
2128                 free_page((unsigned long)buf);
2129         }
2130         return ret;
2131 }
2132 EXPORT_SYMBOL(dump_seek);