SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/proc_fs.h>
  47 #include <linux/mount.h>
  48 #include <linux/security.h>
  49 #include <linux/syscalls.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/tracehook.h>
  54 #include <linux/kmod.h>
  55 #include <linux/fsnotify.h>
  56 #include <linux/fs_struct.h>
  57 #include <linux/pipe_fs_i.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62 #include "internal.h"
  63
  64 int core_uses_pid;
  65 char core_pattern[CORENAME_MAX_SIZE] = "core";
  66 unsigned int core_pipe_limit;
  67 int suid_dumpable = 0;
  68
  69 /* The maximal length of core_pattern is also specified in sysctl.c */
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         if (!fmt)
  77                 return -EINVAL;
  78         write_lock(&binfmt_lock);
  79         insert ? list_add(&fmt->lh, &formats) :
  80                  list_add_tail(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82         return 0;
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 SYSCALL_DEFINE1(uselib, const char __user *, library)
 108 {
 109         struct file *file;
 110         char *tmp = getname(library);
 111         int error = PTR_ERR(tmp);
 112
 113         if (IS_ERR(tmp))
 114                 goto out;
 115
 116         file = do_filp_open(AT_FDCWD, tmp,
 117                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 118                                 MAY_READ | MAY_EXEC | MAY_OPEN);
 119         putname(tmp);
 120         error = PTR_ERR(file);
 121         if (IS_ERR(file))
 122                 goto out;
 123
 124         error = -EINVAL;
 125         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 126                 goto exit;
 127
 128         error = -EACCES;
 129         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 130                 goto exit;
 131
 132         fsnotify_open(file->f_path.dentry);
 133
 134         error = -ENOEXEC;
 135         if(file->f_op) {
 136                 struct linux_binfmt * fmt;
 137
 138                 read_lock(&binfmt_lock);
 139                 list_for_each_entry(fmt, &formats, lh) {
 140                         if (!fmt->load_shlib)
 141                                 continue;
 142                         if (!try_module_get(fmt->module))
 143                                 continue;
 144                         read_unlock(&binfmt_lock);
 145                         error = fmt->load_shlib(file);
 146                         read_lock(&binfmt_lock);
 147                         put_binfmt(fmt);
 148                         if (error != -ENOEXEC)
 149                                 break;
 150                 }
 151                 read_unlock(&binfmt_lock);
 152         }
 153 exit:
 154         fput(file);
 155 out:
 156         return error;
 157 }
 158
 159 #ifdef CONFIG_MMU
 160
 161 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 162                 int write)
 163 {
 164         struct page *page;
 165         int ret;
 166
 167 #ifdef CONFIG_STACK_GROWSUP
 168         if (write) {
 169                 ret = expand_stack_downwards(bprm->vma, pos);
 170                 if (ret < 0)
 171                         return NULL;
 172         }
 173 #endif
 174         ret = get_user_pages(current, bprm->mm, pos,
 175                         1, write, 1, &page, NULL);
 176         if (ret <= 0)
 177                 return NULL;
 178
 179         if (write) {
 180                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 181                 struct rlimit *rlim;
 182
 183                 /*
 184                  * We've historically supported up to 32 pages (ARG_MAX)
 185                  * of argument strings even with small stacks
 186                  */
 187                 if (size <= ARG_MAX)
 188                         return page;
 189
 190                 /*
 191                  * Limit to 1/4-th the stack size for the argv+env strings.
 192                  * This ensures that:
 193                  *  - the remaining binfmt code will not run out of stack space,
 194                  *  - the program will have a reasonable amount of stack left
 195                  *    to work from.
 196                  */
 197                 rlim = current->signal->rlim;
 198                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 199                         put_page(page);
 200                         return NULL;
 201                 }
 202         }
 203
 204         return page;
 205 }
 206
 207 static void put_arg_page(struct page *page)
 208 {
 209         put_page(page);
 210 }
 211
 212 static void free_arg_page(struct linux_binprm *bprm, int i)
 213 {
 214 }
 215
 216 static void free_arg_pages(struct linux_binprm *bprm)
 217 {
 218 }
 219
 220 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 221                 struct page *page)
 222 {
 223         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 224 }
 225
 226 static int __bprm_mm_init(struct linux_binprm *bprm)
 227 {
 228         int err;
 229         struct vm_area_struct *vma = NULL;
 230         struct mm_struct *mm = bprm->mm;
 231
 232         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 233         if (!vma)
 234                 return -ENOMEM;
 235
 236         down_write(&mm->mmap_sem);
 237         vma->vm_mm = mm;
 238
 239         /*
 240          * Place the stack at the largest stack address the architecture
 241          * supports. Later, we'll move this to an appropriate place. We don't
 242          * use STACK_TOP because that can depend on attributes which aren't
 243          * configured yet.
 244          */
 245         vma->vm_end = STACK_TOP_MAX;
 246         vma->vm_start = vma->vm_end - PAGE_SIZE;
 247         vma->vm_flags = VM_STACK_FLAGS;
 248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 249         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(char __user * __user * argv, int max)
 365 {
 366         int i = 0;
 367
 368         if (argv != NULL) {
 369                 for (;;) {
 370                         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                         cond_resched();
 380                 }
 381         }
 382         return i;
 383 }
 384
 385 /*
 386  * 'copy_strings()' copies argument/environment strings from the old
 387  * processes's memory to the new process's stack.  The call to get_user_pages()
 388  * ensures the destination page is created and not swapped out.
 389  */
 390 static int copy_strings(int argc, char __user * __user * argv,
 391                         struct linux_binprm *bprm)
 392 {
 393         struct page *kmapped_page = NULL;
 394         char *kaddr = NULL;
 395         unsigned long kpos = 0;
 396         int ret;
 397
 398         while (argc-- > 0) {
 399                 char __user *str;
 400                 int len;
 401                 unsigned long pos;
 402
 403                 if (get_user(str, argv+argc) ||
 404                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 405                         ret = -EFAULT;
 406                         goto out;
 407                 }
 408
 409                 if (!valid_arg_len(bprm, len)) {
 410                         ret = -E2BIG;
 411                         goto out;
 412                 }
 413
 414                 /* We're going to work our way backwords. */
 415                 pos = bprm->p;
 416                 str += len;
 417                 bprm->p -= len;
 418
 419                 while (len > 0) {
 420                         int offset, bytes_to_copy;
 421
 422                         offset = pos % PAGE_SIZE;
 423                         if (offset == 0)
 424                                 offset = PAGE_SIZE;
 425
 426                         bytes_to_copy = offset;
 427                         if (bytes_to_copy > len)
 428                                 bytes_to_copy = len;
 429
 430                         offset -= bytes_to_copy;
 431                         pos -= bytes_to_copy;
 432                         str -= bytes_to_copy;
 433                         len -= bytes_to_copy;
 434
 435                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 436                                 struct page *page;
 437
 438                                 page = get_arg_page(bprm, pos, 1);
 439                                 if (!page) {
 440                                         ret = -E2BIG;
 441                                         goto out;
 442                                 }
 443
 444                                 if (kmapped_page) {
 445                                         flush_kernel_dcache_page(kmapped_page);
 446                                         kunmap(kmapped_page);
 447                                         put_arg_page(kmapped_page);
 448                                 }
 449                                 kmapped_page = page;
 450                                 kaddr = kmap(kmapped_page);
 451                                 kpos = pos & PAGE_MASK;
 452                                 flush_arg_page(bprm, kpos, kmapped_page);
 453                         }
 454                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 455                                 ret = -EFAULT;
 456                                 goto out;
 457                         }
 458                 }
 459         }
 460         ret = 0;
 461 out:
 462         if (kmapped_page) {
 463                 flush_kernel_dcache_page(kmapped_page);
 464                 kunmap(kmapped_page);
 465                 put_arg_page(kmapped_page);
 466         }
 467         return ret;
 468 }
 469
 470 /*
 471  * Like copy_strings, but get argv and its values from kernel memory.
 472  */
 473 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 474 {
 475         int r;
 476         mm_segment_t oldfs = get_fs();
 477         set_fs(KERNEL_DS);
 478         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 479         set_fs(oldfs);
 480         return r;
 481 }
 482 EXPORT_SYMBOL(copy_strings_kernel);
 483
 484 #ifdef CONFIG_MMU
 485
 486 /*
 487  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 488  * the binfmt code determines where the new stack should reside, we shift it to
 489  * its final location.  The process proceeds as follows:
 490  *
 491  * 1) Use shift to calculate the new vma endpoints.
 492  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 493  *    arguments passed to subsequent functions are consistent.
 494  * 3) Move vma's page tables to the new range.
 495  * 4) Free up any cleared pgd range.
 496  * 5) Shrink the vma to cover only the new range.
 497  */
 498 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 499 {
 500         struct mm_struct *mm = vma->vm_mm;
 501         unsigned long old_start = vma->vm_start;
 502         unsigned long old_end = vma->vm_end;
 503         unsigned long length = old_end - old_start;
 504         unsigned long new_start = old_start - shift;
 505         unsigned long new_end = old_end - shift;
 506         struct mmu_gather *tlb;
 507
 508         BUG_ON(new_start > new_end);
 509
 510         /*
 511          * ensure there are no vmas between where we want to go
 512          * and where we are
 513          */
 514         if (vma != find_vma(mm, new_start))
 515                 return -EFAULT;
 516
 517         /*
 518          * cover the whole range: [new_start, old_end)
 519          */
 520         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 521                 return -ENOMEM;
 522
 523         /*
 524          * move the page tables downwards, on failure we rely on
 525          * process cleanup to remove whatever mess we made.
 526          */
 527         if (length != move_page_tables(vma, old_start,
 528                                        vma, new_start, length))
 529                 return -ENOMEM;
 530
 531         lru_add_drain();
 532         tlb = tlb_gather_mmu(mm, 0);
 533         if (new_end > old_start) {
 534                 /*
 535                  * when the old and new regions overlap clear from new_end.
 536                  */
 537                 free_pgd_range(tlb, new_end, old_end, new_end,
 538                         vma->vm_next ? vma->vm_next->vm_start : 0);
 539         } else {
 540                 /*
 541                  * otherwise, clean from old_start; this is done to not touch
 542                  * the address space in [new_end, old_start) some architectures
 543                  * have constraints on va-space that make this illegal (IA64) -
 544                  * for the others its just a little faster.
 545                  */
 546                 free_pgd_range(tlb, old_start, old_end, new_end,
 547                         vma->vm_next ? vma->vm_next->vm_start : 0);
 548         }
 549         tlb_finish_mmu(tlb, new_end, old_end);
 550
 551         /*
 552          * Shrink the vma to just the new range.  Always succeeds.
 553          */
 554         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 555
 556         return 0;
 557 }
 558
 559 #define EXTRA_STACK_VM_PAGES    20      /* random */
 560
 561 /*
 562  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 563  * the stack is optionally relocated, and some extra space is added.
 564  */
 565 int setup_arg_pages(struct linux_binprm *bprm,
 566                     unsigned long stack_top,
 567                     int executable_stack)
 568 {
 569         unsigned long ret;
 570         unsigned long stack_shift;
 571         struct mm_struct *mm = current->mm;
 572         struct vm_area_struct *vma = bprm->vma;
 573         struct vm_area_struct *prev = NULL;
 574         unsigned long vm_flags;
 575         unsigned long stack_base;
 576         unsigned long stack_size;
 577         unsigned long stack_expand;
 578         unsigned long rlim_stack;
 579
 580 #ifdef CONFIG_STACK_GROWSUP
 581         /* Limit stack size to 1GB */
 582         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 583         if (stack_base > (1 << 30))
 584                 stack_base = 1 << 30;
 585
 586         /* Make sure we didn't let the argument array grow too large. */
 587         if (vma->vm_end - vma->vm_start > stack_base)
 588                 return -ENOMEM;
 589
 590         stack_base = PAGE_ALIGN(stack_top - stack_base);
 591
 592         stack_shift = vma->vm_start - stack_base;
 593         mm->arg_start = bprm->p - stack_shift;
 594         bprm->p = vma->vm_end - stack_shift;
 595 #else
 596         stack_top = arch_align_stack(stack_top);
 597         stack_top = PAGE_ALIGN(stack_top);
 598         stack_shift = vma->vm_end - stack_top;
 599
 600         bprm->p -= stack_shift;
 601         mm->arg_start = bprm->p;
 602 #endif
 603
 604         if (bprm->loader)
 605                 bprm->loader -= stack_shift;
 606         bprm->exec -= stack_shift;
 607
 608         down_write(&mm->mmap_sem);
 609         vm_flags = VM_STACK_FLAGS;
 610
 611         /*
 612          * Adjust stack execute permissions; explicitly enable for
 613          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 614          * (arch default) otherwise.
 615          */
 616         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 617                 vm_flags |= VM_EXEC;
 618         else if (executable_stack == EXSTACK_DISABLE_X)
 619                 vm_flags &= ~VM_EXEC;
 620         vm_flags |= mm->def_flags;
 621
 622         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 623                         vm_flags);
 624         if (ret)
 625                 goto out_unlock;
 626         BUG_ON(prev != vma);
 627
 628         /* Move stack pages down in memory. */
 629         if (stack_shift) {
 630                 ret = shift_arg_pages(vma, stack_shift);
 631                 if (ret)
 632                         goto out_unlock;
 633         }
 634
 635         stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 636         stack_size = vma->vm_end - vma->vm_start;
 637         /*
 638          * Align this down to a page boundary as expand_stack
 639          * will align it up.
 640          */
 641         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 642 #ifdef CONFIG_STACK_GROWSUP
 643         if (stack_size + stack_expand > rlim_stack)
 644                 stack_base = vma->vm_start + rlim_stack;
 645         else
 646                 stack_base = vma->vm_end + stack_expand;
 647 #else
 648         if (stack_size + stack_expand > rlim_stack)
 649                 stack_base = vma->vm_end - rlim_stack;
 650         else
 651                 stack_base = vma->vm_start - stack_expand;
 652 #endif
 653         ret = expand_stack(vma, stack_base);
 654         if (ret)
 655                 ret = -EFAULT;
 656
 657 out_unlock:
 658         up_write(&mm->mmap_sem);
 659         return ret;
 660 }
 661 EXPORT_SYMBOL(setup_arg_pages);
 662
 663 #endif /* CONFIG_MMU */
 664
 665 struct file *open_exec(const char *name)
 666 {
 667         struct file *file;
 668         int err;
 669
 670         file = do_filp_open(AT_FDCWD, name,
 671                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 672                                 MAY_EXEC | MAY_OPEN);
 673         if (IS_ERR(file))
 674                 goto out;
 675
 676         err = -EACCES;
 677         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 678                 goto exit;
 679
 680         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 681                 goto exit;
 682
 683         fsnotify_open(file->f_path.dentry);
 684
 685         err = deny_write_access(file);
 686         if (err)
 687                 goto exit;
 688
 689 out:
 690         return file;
 691
 692 exit:
 693         fput(file);
 694         return ERR_PTR(err);
 695 }
 696 EXPORT_SYMBOL(open_exec);
 697
 698 int kernel_read(struct file *file, loff_t offset,
 699                 char *addr, unsigned long count)
 700 {
 701         mm_segment_t old_fs;
 702         loff_t pos = offset;
 703         int result;
 704
 705         old_fs = get_fs();
 706         set_fs(get_ds());
 707         /* The cast to a user pointer is valid due to the set_fs() */
 708         result = vfs_read(file, (void __user *)addr, count, &pos);
 709         set_fs(old_fs);
 710         return result;
 711 }
 712
 713 EXPORT_SYMBOL(kernel_read);
 714
 715 static int exec_mmap(struct mm_struct *mm)
 716 {
 717         struct task_struct *tsk;
 718         struct mm_struct * old_mm, *active_mm;
 719
 720         /* Notify parent that we're no longer interested in the old VM */
 721         tsk = current;
 722         old_mm = current->mm;
 723         sync_mm_rss(tsk, old_mm);
 724         mm_release(tsk, old_mm);
 725
 726         if (old_mm) {
 727                 /*
 728                  * Make sure that if there is a core dump in progress
 729                  * for the old mm, we get out and die instead of going
 730                  * through with the exec.  We must hold mmap_sem around
 731                  * checking core_state and changing tsk->mm.
 732                  */
 733                 down_read(&old_mm->mmap_sem);
 734                 if (unlikely(old_mm->core_state)) {
 735                         up_read(&old_mm->mmap_sem);
 736                         return -EINTR;
 737                 }
 738         }
 739         task_lock(tsk);
 740         active_mm = tsk->active_mm;
 741         tsk->mm = mm;
 742         tsk->active_mm = mm;
 743         activate_mm(active_mm, mm);
 744         task_unlock(tsk);
 745         arch_pick_mmap_layout(mm);
 746         if (old_mm) {
 747                 up_read(&old_mm->mmap_sem);
 748                 BUG_ON(active_mm != old_mm);
 749                 mm_update_next_owner(old_mm);
 750                 mmput(old_mm);
 751                 return 0;
 752         }
 753         mmdrop(active_mm);
 754         return 0;
 755 }
 756
 757 /*
 758  * This function makes sure the current process has its own signal table,
 759  * so that flush_signal_handlers can later reset the handlers without
 760  * disturbing other processes.  (Other processes might share the signal
 761  * table via the CLONE_SIGHAND option to clone().)
 762  */
 763 static int de_thread(struct task_struct *tsk)
 764 {
 765         struct signal_struct *sig = tsk->signal;
 766         struct sighand_struct *oldsighand = tsk->sighand;
 767         spinlock_t *lock = &oldsighand->siglock;
 768         int count;
 769
 770         if (thread_group_empty(tsk))
 771                 goto no_thread_group;
 772
 773         /*
 774          * Kill all other threads in the thread group.
 775          */
 776         spin_lock_irq(lock);
 777         if (signal_group_exit(sig)) {
 778                 /*
 779                  * Another group action in progress, just
 780                  * return so that the signal is processed.
 781                  */
 782                 spin_unlock_irq(lock);
 783                 return -EAGAIN;
 784         }
 785         sig->group_exit_task = tsk;
 786         zap_other_threads(tsk);
 787
 788         /* Account for the thread group leader hanging around: */
 789         count = thread_group_leader(tsk) ? 1 : 2;
 790         sig->notify_count = count;
 791         while (atomic_read(&sig->count) > count) {
 792                 __set_current_state(TASK_UNINTERRUPTIBLE);
 793                 spin_unlock_irq(lock);
 794                 schedule();
 795                 spin_lock_irq(lock);
 796         }
 797         spin_unlock_irq(lock);
 798
 799         /*
 800          * At this point all other threads have exited, all we have to
 801          * do is to wait for the thread group leader to become inactive,
 802          * and to assume its PID:
 803          */
 804         if (!thread_group_leader(tsk)) {
 805                 struct task_struct *leader = tsk->group_leader;
 806
 807                 sig->notify_count = -1; /* for exit_notify() */
 808                 for (;;) {
 809                         write_lock_irq(&tasklist_lock);
 810                         if (likely(leader->exit_state))
 811                                 break;
 812                         __set_current_state(TASK_UNINTERRUPTIBLE);
 813                         write_unlock_irq(&tasklist_lock);
 814                         schedule();
 815                 }
 816
 817                 /*
 818                  * The only record we have of the real-time age of a
 819                  * process, regardless of execs it's done, is start_time.
 820                  * All the past CPU time is accumulated in signal_struct
 821                  * from sister threads now dead.  But in this non-leader
 822                  * exec, nothing survives from the original leader thread,
 823                  * whose birth marks the true age of this process now.
 824                  * When we take on its identity by switching to its PID, we
 825                  * also take its birthdate (always earlier than our own).
 826                  */
 827                 tsk->start_time = leader->start_time;
 828
 829                 BUG_ON(!same_thread_group(leader, tsk));
 830                 BUG_ON(has_group_leader_pid(tsk));
 831                 /*
 832                  * An exec() starts a new thread group with the
 833                  * TGID of the previous thread group. Rehash the
 834                  * two threads with a switched PID, and release
 835                  * the former thread group leader:
 836                  */
 837
 838                 /* Become a process group leader with the old leader's pid.
 839                  * The old leader becomes a thread of the this thread group.
 840                  * Note: The old leader also uses this pid until release_task
 841                  *       is called.  Odd but simple and correct.
 842                  */
 843                 detach_pid(tsk, PIDTYPE_PID);
 844                 tsk->pid = leader->pid;
 845                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 846                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 847                 transfer_pid(leader, tsk, PIDTYPE_SID);
 848
 849                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 850                 list_replace_init(&leader->sibling, &tsk->sibling);
 851
 852                 tsk->group_leader = tsk;
 853                 leader->group_leader = tsk;
 854
 855                 tsk->exit_signal = SIGCHLD;
 856
 857                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 858                 leader->exit_state = EXIT_DEAD;
 859                 write_unlock_irq(&tasklist_lock);
 860
 861                 release_task(leader);
 862         }
 863
 864         sig->group_exit_task = NULL;
 865         sig->notify_count = 0;
 866
 867 no_thread_group:
 868         if (current->mm)
 869                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 870
 871         exit_itimers(sig);
 872         flush_itimer_signals();
 873
 874         if (atomic_read(&oldsighand->count) != 1) {
 875                 struct sighand_struct *newsighand;
 876                 /*
 877                  * This ->sighand is shared with the CLONE_SIGHAND
 878                  * but not CLONE_THREAD task, switch to the new one.
 879                  */
 880                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 881                 if (!newsighand)
 882                         return -ENOMEM;
 883
 884                 atomic_set(&newsighand->count, 1);
 885                 memcpy(newsighand->action, oldsighand->action,
 886                        sizeof(newsighand->action));
 887
 888                 write_lock_irq(&tasklist_lock);
 889                 spin_lock(&oldsighand->siglock);
 890                 rcu_assign_pointer(tsk->sighand, newsighand);
 891                 spin_unlock(&oldsighand->siglock);
 892                 write_unlock_irq(&tasklist_lock);
 893
 894                 __cleanup_sighand(oldsighand);
 895         }
 896
 897         BUG_ON(!thread_group_leader(tsk));
 898         return 0;
 899 }
 900
 901 /*
 902  * These functions flushes out all traces of the currently running executable
 903  * so that a new one can be started
 904  */
 905 static void flush_old_files(struct files_struct * files)
 906 {
 907         long j = -1;
 908         struct fdtable *fdt;
 909
 910         spin_lock(&files->file_lock);
 911         for (;;) {
 912                 unsigned long set, i;
 913
 914                 j++;
 915                 i = j * __NFDBITS;
 916                 fdt = files_fdtable(files);
 917                 if (i >= fdt->max_fds)
 918                         break;
 919                 set = fdt->close_on_exec->fds_bits[j];
 920                 if (!set)
 921                         continue;
 922                 fdt->close_on_exec->fds_bits[j] = 0;
 923                 spin_unlock(&files->file_lock);
 924                 for ( ; set ; i++,set >>= 1) {
 925                         if (set & 1) {
 926                                 sys_close(i);
 927                         }
 928                 }
 929                 spin_lock(&files->file_lock);
 930
 931         }
 932         spin_unlock(&files->file_lock);
 933 }
 934
 935 char *get_task_comm(char *buf, struct task_struct *tsk)
 936 {
 937         /* buf must be at least sizeof(tsk->comm) in size */
 938         task_lock(tsk);
 939         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 940         task_unlock(tsk);
 941         return buf;
 942 }
 943
 944 void set_task_comm(struct task_struct *tsk, char *buf)
 945 {
 946         task_lock(tsk);
 947
 948         /*
 949          * Threads may access current->comm without holding
 950          * the task lock, so write the string carefully.
 951          * Readers without a lock may see incomplete new
 952          * names but are safe from non-terminating string reads.
 953          */
 954         memset(tsk->comm, 0, TASK_COMM_LEN);
 955         wmb();
 956         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 957         task_unlock(tsk);
 958         perf_event_comm(tsk);
 959 }
 960
 961 int flush_old_exec(struct linux_binprm * bprm)
 962 {
 963         int retval;
 964
 965         /*
 966          * Make sure we have a private signal table and that
 967          * we are unassociated from the previous thread group.
 968          */
 969         retval = de_thread(current);
 970         if (retval)
 971                 goto out;
 972
 973         set_mm_exe_file(bprm->mm, bprm->file);
 974
 975         /*
 976          * Release all of the old mmap stuff
 977          */
 978         retval = exec_mmap(bprm->mm);
 979         if (retval)
 980                 goto out;
 981
 982         bprm->mm = NULL;                /* We're using it now */
 983
 984         current->flags &= ~PF_RANDOMIZE;
 985         flush_thread();
 986         current->personality &= ~bprm->per_clear;
 987
 988         return 0;
 989
 990 out:
 991         return retval;
 992 }
 993 EXPORT_SYMBOL(flush_old_exec);
 994
 995 void setup_new_exec(struct linux_binprm * bprm)
 996 {
 997         int i, ch;
 998         char * name;
 999         char tcomm[sizeof(current->comm)];
1000
1001         arch_pick_mmap_layout(current->mm);
1002
1003         /* This is the point of no return */
1004         current->sas_ss_sp = current->sas_ss_size = 0;
1005
1006         if (current_euid() == current_uid() && current_egid() == current_gid())
1007                 set_dumpable(current->mm, 1);
1008         else
1009                 set_dumpable(current->mm, suid_dumpable);
1010
1011         name = bprm->filename;
1012
1013         /* Copies the binary name from after last slash */
1014         for (i=0; (ch = *(name++)) != '\0';) {
1015                 if (ch == '/')
1016                         i = 0; /* overwrite what we wrote */
1017                 else
1018                         if (i < (sizeof(tcomm) - 1))
1019                                 tcomm[i++] = ch;
1020         }
1021         tcomm[i] = '\0';
1022         set_task_comm(current, tcomm);
1023
1024         /* Set the new mm task size. We have to do that late because it may
1025          * depend on TIF_32BIT which is only updated in flush_thread() on
1026          * some architectures like powerpc
1027          */
1028         current->mm->task_size = TASK_SIZE;
1029
1030         /* install the new credentials */
1031         if (bprm->cred->uid != current_euid() ||
1032             bprm->cred->gid != current_egid()) {
1033                 current->pdeath_signal = 0;
1034         } else if (file_permission(bprm->file, MAY_READ) ||
1035                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1036                 set_dumpable(current->mm, suid_dumpable);
1037         }
1038
1039         /*
1040          * Flush performance counters when crossing a
1041          * security domain:
1042          */
1043         if (!get_dumpable(current->mm))
1044                 perf_event_exit_task(current);
1045
1046         /* An exec changes our domain. We are no longer part of the thread
1047            group */
1048
1049         current->self_exec_id++;
1050
1051         flush_signal_handlers(current, 0);
1052         flush_old_files(current->files);
1053 }
1054 EXPORT_SYMBOL(setup_new_exec);
1055
1056 /*
1057  * Prepare credentials and lock ->cred_guard_mutex.
1058  * install_exec_creds() commits the new creds and drops the lock.
1059  * Or, if exec fails before, free_bprm() should release ->cred and
1060  * and unlock.
1061  */
1062 int prepare_bprm_creds(struct linux_binprm *bprm)
1063 {
1064         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1065                 return -ERESTARTNOINTR;
1066
1067         bprm->cred = prepare_exec_creds();
1068         if (likely(bprm->cred))
1069                 return 0;
1070
1071         mutex_unlock(&current->cred_guard_mutex);
1072         return -ENOMEM;
1073 }
1074
1075 void free_bprm(struct linux_binprm *bprm)
1076 {
1077         free_arg_pages(bprm);
1078         if (bprm->cred) {
1079                 mutex_unlock(&current->cred_guard_mutex);
1080                 abort_creds(bprm->cred);
1081         }
1082         kfree(bprm);
1083 }
1084
1085 /*
1086  * install the new credentials for this executable
1087  */
1088 void install_exec_creds(struct linux_binprm *bprm)
1089 {
1090         security_bprm_committing_creds(bprm);
1091
1092         commit_creds(bprm->cred);
1093         bprm->cred = NULL;
1094         /*
1095          * cred_guard_mutex must be held at least to this point to prevent
1096          * ptrace_attach() from altering our determination of the task's
1097          * credentials; any time after this it may be unlocked.
1098          */
1099         security_bprm_committed_creds(bprm);
1100         mutex_unlock(&current->cred_guard_mutex);
1101 }
1102 EXPORT_SYMBOL(install_exec_creds);
1103
1104 /*
1105  * determine how safe it is to execute the proposed program
1106  * - the caller must hold current->cred_guard_mutex to protect against
1107  *   PTRACE_ATTACH
1108  */
1109 int check_unsafe_exec(struct linux_binprm *bprm)
1110 {
1111         struct task_struct *p = current, *t;
1112         unsigned n_fs;
1113         int res = 0;
1114
1115         bprm->unsafe = tracehook_unsafe_exec(p);
1116
1117         n_fs = 1;
1118         write_lock(&p->fs->lock);
1119         rcu_read_lock();
1120         for (t = next_thread(p); t != p; t = next_thread(t)) {
1121                 if (t->fs == p->fs)
1122                         n_fs++;
1123         }
1124         rcu_read_unlock();
1125
1126         if (p->fs->users > n_fs) {
1127                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1128         } else {
1129                 res = -EAGAIN;
1130                 if (!p->fs->in_exec) {
1131                         p->fs->in_exec = 1;
1132                         res = 1;
1133                 }
1134         }
1135         write_unlock(&p->fs->lock);
1136
1137         return res;
1138 }
1139
1140 /*
1141  * Fill the binprm structure from the inode.
1142  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1143  *
1144  * This may be called multiple times for binary chains (scripts for example).
1145  */
1146 int prepare_binprm(struct linux_binprm *bprm)
1147 {
1148         umode_t mode;
1149         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1150         int retval;
1151
1152         mode = inode->i_mode;
1153         if (bprm->file->f_op == NULL)
1154                 return -EACCES;
1155
1156         /* clear any previous set[ug]id data from a previous binary */
1157         bprm->cred->euid = current_euid();
1158         bprm->cred->egid = current_egid();
1159
1160         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1161                 /* Set-uid? */
1162                 if (mode & S_ISUID) {
1163                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1164                         bprm->cred->euid = inode->i_uid;
1165                 }
1166
1167                 /* Set-gid? */
1168                 /*
1169                  * If setgid is set but no group execute bit then this
1170                  * is a candidate for mandatory locking, not a setgid
1171                  * executable.
1172                  */
1173                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1174                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1175                         bprm->cred->egid = inode->i_gid;
1176                 }
1177         }
1178
1179         /* fill in binprm security blob */
1180         retval = security_bprm_set_creds(bprm);
1181         if (retval)
1182                 return retval;
1183         bprm->cred_prepared = 1;
1184
1185         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1186         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1187 }
1188
1189 EXPORT_SYMBOL(prepare_binprm);
1190
1191 /*
1192  * Arguments are '\0' separated strings found at the location bprm->p
1193  * points to; chop off the first by relocating brpm->p to right after
1194  * the first '\0' encountered.
1195  */
1196 int remove_arg_zero(struct linux_binprm *bprm)
1197 {
1198         int ret = 0;
1199         unsigned long offset;
1200         char *kaddr;
1201         struct page *page;
1202
1203         if (!bprm->argc)
1204                 return 0;
1205
1206         do {
1207                 offset = bprm->p & ~PAGE_MASK;
1208                 page = get_arg_page(bprm, bprm->p, 0);
1209                 if (!page) {
1210                         ret = -EFAULT;
1211                         goto out;
1212                 }
1213                 kaddr = kmap_atomic(page, KM_USER0);
1214
1215                 for (; offset < PAGE_SIZE && kaddr[offset];
1216                                 offset++, bprm->p++)
1217                         ;
1218
1219                 kunmap_atomic(kaddr, KM_USER0);
1220                 put_arg_page(page);
1221
1222                 if (offset == PAGE_SIZE)
1223                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1224         } while (offset == PAGE_SIZE);
1225
1226         bprm->p++;
1227         bprm->argc--;
1228         ret = 0;
1229
1230 out:
1231         return ret;
1232 }
1233 EXPORT_SYMBOL(remove_arg_zero);
1234
1235 /*
1236  * cycle the list of binary formats handler, until one recognizes the image
1237  */
1238 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1239 {
1240         unsigned int depth = bprm->recursion_depth;
1241         int try,retval;
1242         struct linux_binfmt *fmt;
1243
1244         retval = security_bprm_check(bprm);
1245         if (retval)
1246                 return retval;
1247
1248         /* kernel module loader fixup */
1249         /* so we don't try to load run modprobe in kernel space. */
1250         set_fs(USER_DS);
1251
1252         retval = audit_bprm(bprm);
1253         if (retval)
1254                 return retval;
1255
1256         retval = -ENOENT;
1257         for (try=0; try<2; try++) {
1258                 read_lock(&binfmt_lock);
1259                 list_for_each_entry(fmt, &formats, lh) {
1260                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1261                         if (!fn)
1262                                 continue;
1263                         if (!try_module_get(fmt->module))
1264                                 continue;
1265                         read_unlock(&binfmt_lock);
1266                         retval = fn(bprm, regs);
1267                         /*
1268                          * Restore the depth counter to its starting value
1269                          * in this call, so we don't have to rely on every
1270                          * load_binary function to restore it on return.
1271                          */
1272                         bprm->recursion_depth = depth;
1273                         if (retval >= 0) {
1274                                 if (depth == 0)
1275                                         tracehook_report_exec(fmt, bprm, regs);
1276                                 put_binfmt(fmt);
1277                                 allow_write_access(bprm->file);
1278                                 if (bprm->file)
1279                                         fput(bprm->file);
1280                                 bprm->file = NULL;
1281                                 current->did_exec = 1;
1282                                 proc_exec_connector(current);
1283                                 return retval;
1284                         }
1285                         read_lock(&binfmt_lock);
1286                         put_binfmt(fmt);
1287                         if (retval != -ENOEXEC || bprm->mm == NULL)
1288                                 break;
1289                         if (!bprm->file) {
1290                                 read_unlock(&binfmt_lock);
1291                                 return retval;
1292                         }
1293                 }
1294                 read_unlock(&binfmt_lock);
1295                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1296                         break;
1297 #ifdef CONFIG_MODULES
1298                 } else {
1299 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1300                         if (printable(bprm->buf[0]) &&
1301                             printable(bprm->buf[1]) &&
1302                             printable(bprm->buf[2]) &&
1303                             printable(bprm->buf[3]))
1304                                 break; /* -ENOEXEC */
1305                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1306 #endif
1307                 }
1308         }
1309         return retval;
1310 }
1311
1312 EXPORT_SYMBOL(search_binary_handler);
1313
1314 /*
1315  * sys_execve() executes a new program.
1316  */
1317 int do_execve(char * filename,
1318         char __user *__user *argv,
1319         char __user *__user *envp,
1320         struct pt_regs * regs)
1321 {
1322         struct linux_binprm *bprm;
1323         struct file *file;
1324         struct files_struct *displaced;
1325         bool clear_in_exec;
1326         int retval;
1327
1328         retval = unshare_files(&displaced);
1329         if (retval)
1330                 goto out_ret;
1331
1332         retval = -ENOMEM;
1333         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1334         if (!bprm)
1335                 goto out_files;
1336
1337         retval = prepare_bprm_creds(bprm);
1338         if (retval)
1339                 goto out_free;
1340
1341         retval = check_unsafe_exec(bprm);
1342         if (retval < 0)
1343                 goto out_free;
1344         clear_in_exec = retval;
1345         current->in_execve = 1;
1346
1347         file = open_exec(filename);
1348         retval = PTR_ERR(file);
1349         if (IS_ERR(file))
1350                 goto out_unmark;
1351
1352         sched_exec();
1353
1354         bprm->file = file;
1355         bprm->filename = filename;
1356         bprm->interp = filename;
1357
1358         retval = bprm_mm_init(bprm);
1359         if (retval)
1360                 goto out_file;
1361
1362         bprm->argc = count(argv, MAX_ARG_STRINGS);
1363         if ((retval = bprm->argc) < 0)
1364                 goto out;
1365
1366         bprm->envc = count(envp, MAX_ARG_STRINGS);
1367         if ((retval = bprm->envc) < 0)
1368                 goto out;
1369
1370         retval = prepare_binprm(bprm);
1371         if (retval < 0)
1372                 goto out;
1373
1374         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1375         if (retval < 0)
1376                 goto out;
1377
1378         bprm->exec = bprm->p;
1379         retval = copy_strings(bprm->envc, envp, bprm);
1380         if (retval < 0)
1381                 goto out;
1382
1383         retval = copy_strings(bprm->argc, argv, bprm);
1384         if (retval < 0)
1385                 goto out;
1386
1387         current->flags &= ~PF_KTHREAD;
1388         retval = search_binary_handler(bprm,regs);
1389         if (retval < 0)
1390                 goto out;
1391
1392         current->stack_start = current->mm->start_stack;
1393
1394         /* execve succeeded */
1395         current->fs->in_exec = 0;
1396         current->in_execve = 0;
1397         acct_update_integrals(current);
1398         free_bprm(bprm);
1399         if (displaced)
1400                 put_files_struct(displaced);
1401         return retval;
1402
1403 out:
1404         if (bprm->mm)
1405                 mmput (bprm->mm);
1406
1407 out_file:
1408         if (bprm->file) {
1409                 allow_write_access(bprm->file);
1410                 fput(bprm->file);
1411         }
1412
1413 out_unmark:
1414         if (clear_in_exec)
1415                 current->fs->in_exec = 0;
1416         current->in_execve = 0;
1417
1418 out_free:
1419         free_bprm(bprm);
1420
1421 out_files:
1422         if (displaced)
1423                 reset_files_struct(displaced);
1424 out_ret:
1425         return retval;
1426 }
1427
1428 void set_binfmt(struct linux_binfmt *new)
1429 {
1430         struct mm_struct *mm = current->mm;
1431
1432         if (mm->binfmt)
1433                 module_put(mm->binfmt->module);
1434
1435         mm->binfmt = new;
1436         if (new)
1437                 __module_get(new->module);
1438 }
1439
1440 EXPORT_SYMBOL(set_binfmt);
1441
1442 /* format_corename will inspect the pattern parameter, and output a
1443  * name into corename, which must have space for at least
1444  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1445  */
1446 static int format_corename(char *corename, long signr)
1447 {
1448         const struct cred *cred = current_cred();
1449         const char *pat_ptr = core_pattern;
1450         int ispipe = (*pat_ptr == '|');
1451         char *out_ptr = corename;
1452         char *const out_end = corename + CORENAME_MAX_SIZE;
1453         int rc;
1454         int pid_in_pattern = 0;
1455
1456         /* Repeat as long as we have more pattern to process and more output
1457            space */
1458         while (*pat_ptr) {
1459                 if (*pat_ptr != '%') {
1460                         if (out_ptr == out_end)
1461                                 goto out;
1462                         *out_ptr++ = *pat_ptr++;
1463                 } else {
1464                         switch (*++pat_ptr) {
1465                         case 0:
1466                                 goto out;
1467                         /* Double percent, output one percent */
1468                         case '%':
1469                                 if (out_ptr == out_end)
1470                                         goto out;
1471                                 *out_ptr++ = '%';
1472                                 break;
1473                         /* pid */
1474                         case 'p':
1475                                 pid_in_pattern = 1;
1476                                 rc = snprintf(out_ptr, out_end - out_ptr,
1477                                               "%d", task_tgid_vnr(current));
1478                                 if (rc > out_end - out_ptr)
1479                                         goto out;
1480                                 out_ptr += rc;
1481                                 break;
1482                         /* uid */
1483                         case 'u':
1484                                 rc = snprintf(out_ptr, out_end - out_ptr,
1485                                               "%d", cred->uid);
1486                                 if (rc > out_end - out_ptr)
1487                                         goto out;
1488                                 out_ptr += rc;
1489                                 break;
1490                         /* gid */
1491                         case 'g':
1492                                 rc = snprintf(out_ptr, out_end - out_ptr,
1493                                               "%d", cred->gid);
1494                                 if (rc > out_end - out_ptr)
1495                                         goto out;
1496                                 out_ptr += rc;
1497                                 break;
1498                         /* signal that caused the coredump */
1499                         case 's':
1500                                 rc = snprintf(out_ptr, out_end - out_ptr,
1501                                               "%ld", signr);
1502                                 if (rc > out_end - out_ptr)
1503                                         goto out;
1504                                 out_ptr += rc;
1505                                 break;
1506                         /* UNIX time of coredump */
1507                         case 't': {
1508                                 struct timeval tv;
1509                                 do_gettimeofday(&tv);
1510                                 rc = snprintf(out_ptr, out_end - out_ptr,
1511                                               "%lu", tv.tv_sec);
1512                                 if (rc > out_end - out_ptr)
1513                                         goto out;
1514                                 out_ptr += rc;
1515                                 break;
1516                         }
1517                         /* hostname */
1518                         case 'h':
1519                                 down_read(&uts_sem);
1520                                 rc = snprintf(out_ptr, out_end - out_ptr,
1521                                               "%s", utsname()->nodename);
1522                                 up_read(&uts_sem);
1523                                 if (rc > out_end - out_ptr)
1524                                         goto out;
1525                                 out_ptr += rc;
1526                                 break;
1527                         /* executable */
1528                         case 'e':
1529                                 rc = snprintf(out_ptr, out_end - out_ptr,
1530                                               "%s", current->comm);
1531                                 if (rc > out_end - out_ptr)
1532                                         goto out;
1533                                 out_ptr += rc;
1534                                 break;
1535                         /* core limit size */
1536                         case 'c':
1537                                 rc = snprintf(out_ptr, out_end - out_ptr,
1538                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1539                                 if (rc > out_end - out_ptr)
1540                                         goto out;
1541                                 out_ptr += rc;
1542                                 break;
1543                         default:
1544                                 break;
1545                         }
1546                         ++pat_ptr;
1547                 }
1548         }
1549         /* Backward compatibility with core_uses_pid:
1550          *
1551          * If core_pattern does not include a %p (as is the default)
1552          * and core_uses_pid is set, then .%pid will be appended to
1553          * the filename. Do not do this for piped commands. */
1554         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1555                 rc = snprintf(out_ptr, out_end - out_ptr,
1556                               ".%d", task_tgid_vnr(current));
1557                 if (rc > out_end - out_ptr)
1558                         goto out;
1559                 out_ptr += rc;
1560         }
1561 out:
1562         *out_ptr = 0;
1563         return ispipe;
1564 }
1565
1566 static int zap_process(struct task_struct *start)
1567 {
1568         struct task_struct *t;
1569         int nr = 0;
1570
1571         start->signal->flags = SIGNAL_GROUP_EXIT;
1572         start->signal->group_stop_count = 0;
1573
1574         t = start;
1575         do {
1576                 if (t != current && t->mm) {
1577                         sigaddset(&t->pending.signal, SIGKILL);
1578                         signal_wake_up(t, 1);
1579                         nr++;
1580                 }
1581         } while_each_thread(start, t);
1582
1583         return nr;
1584 }
1585
1586 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1587                                 struct core_state *core_state, int exit_code)
1588 {
1589         struct task_struct *g, *p;
1590         unsigned long flags;
1591         int nr = -EAGAIN;
1592
1593         spin_lock_irq(&tsk->sighand->siglock);
1594         if (!signal_group_exit(tsk->signal)) {
1595                 mm->core_state = core_state;
1596                 tsk->signal->group_exit_code = exit_code;
1597                 nr = zap_process(tsk);
1598         }
1599         spin_unlock_irq(&tsk->sighand->siglock);
1600         if (unlikely(nr < 0))
1601                 return nr;
1602
1603         if (atomic_read(&mm->mm_users) == nr + 1)
1604                 goto done;
1605         /*
1606          * We should find and kill all tasks which use this mm, and we should
1607          * count them correctly into ->nr_threads. We don't take tasklist
1608          * lock, but this is safe wrt:
1609          *
1610          * fork:
1611          *      None of sub-threads can fork after zap_process(leader). All
1612          *      processes which were created before this point should be
1613          *      visible to zap_threads() because copy_process() adds the new
1614          *      process to the tail of init_task.tasks list, and lock/unlock
1615          *      of ->siglock provides a memory barrier.
1616          *
1617          * do_exit:
1618          *      The caller holds mm->mmap_sem. This means that the task which
1619          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1620          *      its ->mm.
1621          *
1622          * de_thread:
1623          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1624          *      we must see either old or new leader, this does not matter.
1625          *      However, it can change p->sighand, so lock_task_sighand(p)
1626          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1627          *      it can't fail.
1628          *
1629          *      Note also that "g" can be the old leader with ->mm == NULL
1630          *      and already unhashed and thus removed from ->thread_group.
1631          *      This is OK, __unhash_process()->list_del_rcu() does not
1632          *      clear the ->next pointer, we will find the new leader via
1633          *      next_thread().
1634          */
1635         rcu_read_lock();
1636         for_each_process(g) {
1637                 if (g == tsk->group_leader)
1638                         continue;
1639                 if (g->flags & PF_KTHREAD)
1640                         continue;
1641                 p = g;
1642                 do {
1643                         if (p->mm) {
1644                                 if (unlikely(p->mm == mm)) {
1645                                         lock_task_sighand(p, &flags);
1646                                         nr += zap_process(p);
1647                                         unlock_task_sighand(p, &flags);
1648                                 }
1649                                 break;
1650                         }
1651                 } while_each_thread(g, p);
1652         }
1653         rcu_read_unlock();
1654 done:
1655         atomic_set(&core_state->nr_threads, nr);
1656         return nr;
1657 }
1658
1659 static int coredump_wait(int exit_code, struct core_state *core_state)
1660 {
1661         struct task_struct *tsk = current;
1662         struct mm_struct *mm = tsk->mm;
1663         struct completion *vfork_done;
1664         int core_waiters;
1665
1666         init_completion(&core_state->startup);
1667         core_state->dumper.task = tsk;
1668         core_state->dumper.next = NULL;
1669         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1670         up_write(&mm->mmap_sem);
1671
1672         if (unlikely(core_waiters < 0))
1673                 goto fail;
1674
1675         /*
1676          * Make sure nobody is waiting for us to release the VM,
1677          * otherwise we can deadlock when we wait on each other
1678          */
1679         vfork_done = tsk->vfork_done;
1680         if (vfork_done) {
1681                 tsk->vfork_done = NULL;
1682                 complete(vfork_done);
1683         }
1684
1685         if (core_waiters)
1686                 wait_for_completion(&core_state->startup);
1687 fail:
1688         return core_waiters;
1689 }
1690
1691 static void coredump_finish(struct mm_struct *mm)
1692 {
1693         struct core_thread *curr, *next;
1694         struct task_struct *task;
1695
1696         next = mm->core_state->dumper.next;
1697         while ((curr = next) != NULL) {
1698                 next = curr->next;
1699                 task = curr->task;
1700                 /*
1701                  * see exit_mm(), curr->task must not see
1702                  * ->task == NULL before we read ->next.
1703                  */
1704                 smp_mb();
1705                 curr->task = NULL;
1706                 wake_up_process(task);
1707         }
1708
1709         mm->core_state = NULL;
1710 }
1711
1712 /*
1713  * set_dumpable converts traditional three-value dumpable to two flags and
1714  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1715  * these bits are not changed atomically.  So get_dumpable can observe the
1716  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1717  * return either old dumpable or new one by paying attention to the order of
1718  * modifying the bits.
1719  *
1720  * dumpable |   mm->flags (binary)
1721  * old  new | initial interim  final
1722  * ---------+-----------------------
1723  *  0    1  |   00      01      01
1724  *  0    2  |   00      10(*)   11
1725  *  1    0  |   01      00      00
1726  *  1    2  |   01      11      11
1727  *  2    0  |   11      10(*)   00
1728  *  2    1  |   11      11      01
1729  *
1730  * (*) get_dumpable regards interim value of 10 as 11.
1731  */
1732 void set_dumpable(struct mm_struct *mm, int value)
1733 {
1734         switch (value) {
1735         case 0:
1736                 clear_bit(MMF_DUMPABLE, &mm->flags);
1737                 smp_wmb();
1738                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1739                 break;
1740         case 1:
1741                 set_bit(MMF_DUMPABLE, &mm->flags);
1742                 smp_wmb();
1743                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1744                 break;
1745         case 2:
1746                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1747                 smp_wmb();
1748                 set_bit(MMF_DUMPABLE, &mm->flags);
1749                 break;
1750         }
1751 }
1752
1753 int get_dumpable(struct mm_struct *mm)
1754 {
1755         int ret;
1756
1757         ret = mm->flags & 0x3;
1758         return (ret >= 2) ? 2 : ret;
1759 }
1760
1761 static void wait_for_dump_helpers(struct file *file)
1762 {
1763         struct pipe_inode_info *pipe;
1764
1765         pipe = file->f_path.dentry->d_inode->i_pipe;
1766
1767         pipe_lock(pipe);
1768         pipe->readers++;
1769         pipe->writers--;
1770
1771         while ((pipe->readers > 1) && (!signal_pending(current))) {
1772                 wake_up_interruptible_sync(&pipe->wait);
1773                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1774                 pipe_wait(pipe);
1775         }
1776
1777         pipe->readers--;
1778         pipe->writers++;
1779         pipe_unlock(pipe);
1780
1781 }
1782
1783
1784 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1785 {
1786         struct core_state core_state;
1787         char corename[CORENAME_MAX_SIZE + 1];
1788         struct mm_struct *mm = current->mm;
1789         struct linux_binfmt * binfmt;
1790         struct inode * inode;
1791         const struct cred *old_cred;
1792         struct cred *cred;
1793         int retval = 0;
1794         int flag = 0;
1795         int ispipe = 0;
1796         char **helper_argv = NULL;
1797         int helper_argc = 0;
1798         int dump_count = 0;
1799         static atomic_t core_dump_count = ATOMIC_INIT(0);
1800         struct coredump_params cprm = {
1801                 .signr = signr,
1802                 .regs = regs,
1803                 .limit = current->signal->rlim[RLIMIT_CORE].rlim_cur,
1804         };
1805
1806         audit_core_dumps(signr);
1807
1808         binfmt = mm->binfmt;
1809         if (!binfmt || !binfmt->core_dump)
1810                 goto fail;
1811
1812         cred = prepare_creds();
1813         if (!cred) {
1814                 retval = -ENOMEM;
1815                 goto fail;
1816         }
1817
1818         down_write(&mm->mmap_sem);
1819         /*
1820          * If another thread got here first, or we are not dumpable, bail out.
1821          */
1822         if (mm->core_state || !get_dumpable(mm)) {
1823                 up_write(&mm->mmap_sem);
1824                 put_cred(cred);
1825                 goto fail;
1826         }
1827
1828         /*
1829          *      We cannot trust fsuid as being the "true" uid of the
1830          *      process nor do we know its entire history. We only know it
1831          *      was tainted so we dump it as root in mode 2.
1832          */
1833         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1834                 flag = O_EXCL;          /* Stop rewrite attacks */
1835                 cred->fsuid = 0;        /* Dump root private */
1836         }
1837
1838         retval = coredump_wait(exit_code, &core_state);
1839         if (retval < 0) {
1840                 put_cred(cred);
1841                 goto fail;
1842         }
1843
1844         old_cred = override_creds(cred);
1845
1846         /*
1847          * Clear any false indication of pending signals that might
1848          * be seen by the filesystem code called to write the core file.
1849          */
1850         clear_thread_flag(TIF_SIGPENDING);
1851
1852         /*
1853          * lock_kernel() because format_corename() is controlled by sysctl, which
1854          * uses lock_kernel()
1855          */
1856         lock_kernel();
1857         ispipe = format_corename(corename, signr);
1858         unlock_kernel();
1859
1860         if ((!ispipe) && (cprm.limit < binfmt->min_coredump))
1861                 goto fail_unlock;
1862
1863         if (ispipe) {
1864                 if (cprm.limit == 0) {
1865                         /*
1866                          * Normally core limits are irrelevant to pipes, since
1867                          * we're not writing to the file system, but we use
1868                          * cprm.limit of 0 here as a speacial value. Any
1869                          * non-zero limit gets set to RLIM_INFINITY below, but
1870                          * a limit of 0 skips the dump.  This is a consistent
1871                          * way to catch recursive crashes.  We can still crash
1872                          * if the core_pattern binary sets RLIM_CORE =  !0
1873                          * but it runs as root, and can do lots of stupid things
1874                          * Note that we use task_tgid_vnr here to grab the pid
1875                          * of the process group leader.  That way we get the
1876                          * right pid if a thread in a multi-threaded
1877                          * core_pattern process dies.
1878                          */
1879                         printk(KERN_WARNING
1880                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1881                                 task_tgid_vnr(current), current->comm);
1882                         printk(KERN_WARNING "Aborting core\n");
1883                         goto fail_unlock;
1884                 }
1885
1886                 dump_count = atomic_inc_return(&core_dump_count);
1887                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1888                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1889                                task_tgid_vnr(current), current->comm);
1890                         printk(KERN_WARNING "Skipping core dump\n");
1891                         goto fail_dropcount;
1892                 }
1893
1894                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1895                 if (!helper_argv) {
1896                         printk(KERN_WARNING "%s failed to allocate memory\n",
1897                                __func__);
1898                         goto fail_dropcount;
1899                 }
1900
1901                 cprm.limit = RLIM_INFINITY;
1902
1903                 /* SIGPIPE can happen, but it's just never processed */
1904                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1905                                 &cprm.file)) {
1906                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1907                                corename);
1908                         goto fail_dropcount;
1909                 }
1910         } else
1911                 cprm.file = filp_open(corename,
1912                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1913                                  0600);
1914         if (IS_ERR(cprm.file))
1915                 goto fail_dropcount;
1916         inode = cprm.file->f_path.dentry->d_inode;
1917         if (inode->i_nlink > 1)
1918                 goto close_fail;        /* multiple links - don't dump */
1919         if (!ispipe && d_unhashed(cprm.file->f_path.dentry))
1920                 goto close_fail;
1921
1922         /* AK: actually i see no reason to not allow this for named pipes etc.,
1923            but keep the previous behaviour for now. */
1924         if (!ispipe && !S_ISREG(inode->i_mode))
1925                 goto close_fail;
1926         /*
1927          * Dont allow local users get cute and trick others to coredump
1928          * into their pre-created files:
1929          */
1930         if (inode->i_uid != current_fsuid())
1931                 goto close_fail;
1932         if (!cprm.file->f_op)
1933                 goto close_fail;
1934         if (!cprm.file->f_op->write)
1935                 goto close_fail;
1936         if (!ispipe &&
1937             do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0)
1938                 goto close_fail;
1939
1940         retval = binfmt->core_dump(&cprm);
1941
1942         if (retval)
1943                 current->signal->group_exit_code |= 0x80;
1944 close_fail:
1945         if (ispipe && core_pipe_limit)
1946                 wait_for_dump_helpers(cprm.file);
1947         filp_close(cprm.file, NULL);
1948 fail_dropcount:
1949         if (dump_count)
1950                 atomic_dec(&core_dump_count);
1951 fail_unlock:
1952         if (helper_argv)
1953                 argv_free(helper_argv);
1954
1955         revert_creds(old_cred);
1956         put_cred(cred);
1957         coredump_finish(mm);
1958 fail:
1959         return;
1960 }