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