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->f_path.dentry);
 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(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 /*
 560  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 561  * the stack is optionally relocated, and some extra space is added.
 562  */
 563 int setup_arg_pages(struct linux_binprm *bprm,
 564                     unsigned long stack_top,
 565                     int executable_stack)
 566 {
 567         unsigned long ret;
 568         unsigned long stack_shift;
 569         struct mm_struct *mm = current->mm;
 570         struct vm_area_struct *vma = bprm->vma;
 571         struct vm_area_struct *prev = NULL;
 572         unsigned long vm_flags;
 573         unsigned long stack_base;
 574         unsigned long stack_size;
 575         unsigned long stack_expand;
 576         unsigned long rlim_stack;
 577
 578 #ifdef CONFIG_STACK_GROWSUP
 579         /* Limit stack size to 1GB */
 580         stack_base = rlimit_max(RLIMIT_STACK);
 581         if (stack_base > (1 << 30))
 582                 stack_base = 1 << 30;
 583
 584         /* Make sure we didn't let the argument array grow too large. */
 585         if (vma->vm_end - vma->vm_start > stack_base)
 586                 return -ENOMEM;
 587
 588         stack_base = PAGE_ALIGN(stack_top - stack_base);
 589
 590         stack_shift = vma->vm_start - stack_base;
 591         mm->arg_start = bprm->p - stack_shift;
 592         bprm->p = vma->vm_end - stack_shift;
 593 #else
 594         stack_top = arch_align_stack(stack_top);
 595         stack_top = PAGE_ALIGN(stack_top);
 596         stack_shift = vma->vm_end - stack_top;
 597
 598         bprm->p -= stack_shift;
 599         mm->arg_start = bprm->p;
 600 #endif
 601
 602         if (bprm->loader)
 603                 bprm->loader -= stack_shift;
 604         bprm->exec -= stack_shift;
 605
 606         down_write(&mm->mmap_sem);
 607         vm_flags = VM_STACK_FLAGS;
 608
 609         /*
 610          * Adjust stack execute permissions; explicitly enable for
 611          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 612          * (arch default) otherwise.
 613          */
 614         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 615                 vm_flags |= VM_EXEC;
 616         else if (executable_stack == EXSTACK_DISABLE_X)
 617                 vm_flags &= ~VM_EXEC;
 618         vm_flags |= mm->def_flags;
 619         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 620
 621         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 622                         vm_flags);
 623         if (ret)
 624                 goto out_unlock;
 625         BUG_ON(prev != vma);
 626
 627         /* Move stack pages down in memory. */
 628         if (stack_shift) {
 629                 ret = shift_arg_pages(vma, stack_shift);
 630                 if (ret)
 631                         goto out_unlock;
 632         }
 633
 634         /* mprotect_fixup is overkill to remove the temporary stack flags */
 635         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 636
 637         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 638         stack_size = vma->vm_end - vma->vm_start;
 639         /*
 640          * Align this down to a page boundary as expand_stack
 641          * will align it up.
 642          */
 643         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 644 #ifdef CONFIG_STACK_GROWSUP
 645         if (stack_size + stack_expand > rlim_stack)
 646                 stack_base = vma->vm_start + rlim_stack;
 647         else
 648                 stack_base = vma->vm_end + stack_expand;
 649 #else
 650         if (stack_size + stack_expand > rlim_stack)
 651                 stack_base = vma->vm_end - rlim_stack;
 652         else
 653                 stack_base = vma->vm_start - stack_expand;
 654 #endif
 655         ret = expand_stack(vma, stack_base);
 656         if (ret)
 657                 ret = -EFAULT;
 658
 659 out_unlock:
 660         up_write(&mm->mmap_sem);
 661         return ret;
 662 }
 663 EXPORT_SYMBOL(setup_arg_pages);
 664
 665 #endif /* CONFIG_MMU */
 666
 667 struct file *open_exec(const char *name)
 668 {
 669         struct file *file;
 670         int err;
 671
 672         file = do_filp_open(AT_FDCWD, name,
 673                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
 674                                 MAY_EXEC | MAY_OPEN);
 675         if (IS_ERR(file))
 676                 goto out;
 677
 678         err = -EACCES;
 679         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 680                 goto exit;
 681
 682         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 683                 goto exit;
 684
 685         fsnotify_open(file->f_path.dentry);
 686
 687         err = deny_write_access(file);
 688         if (err)
 689                 goto exit;
 690
 691 out:
 692         return file;
 693
 694 exit:
 695         fput(file);
 696         return ERR_PTR(err);
 697 }
 698 EXPORT_SYMBOL(open_exec);
 699
 700 int kernel_read(struct file *file, loff_t offset,
 701                 char *addr, unsigned long count)
 702 {
 703         mm_segment_t old_fs;
 704         loff_t pos = offset;
 705         int result;
 706
 707         old_fs = get_fs();
 708         set_fs(get_ds());
 709         /* The cast to a user pointer is valid due to the set_fs() */
 710         result = vfs_read(file, (void __user *)addr, count, &pos);
 711         set_fs(old_fs);
 712         return result;
 713 }
 714
 715 EXPORT_SYMBOL(kernel_read);
 716
 717 static int exec_mmap(struct mm_struct *mm)
 718 {
 719         struct task_struct *tsk;
 720         struct mm_struct * old_mm, *active_mm;
 721
 722         /* Notify parent that we're no longer interested in the old VM */
 723         tsk = current;
 724         old_mm = current->mm;
 725         sync_mm_rss(tsk, old_mm);
 726         mm_release(tsk, old_mm);
 727
 728         if (old_mm) {
 729                 /*
 730                  * Make sure that if there is a core dump in progress
 731                  * for the old mm, we get out and die instead of going
 732                  * through with the exec.  We must hold mmap_sem around
 733                  * checking core_state and changing tsk->mm.
 734                  */
 735                 down_read(&old_mm->mmap_sem);
 736                 if (unlikely(old_mm->core_state)) {
 737                         up_read(&old_mm->mmap_sem);
 738                         return -EINTR;
 739                 }
 740         }
 741         task_lock(tsk);
 742         active_mm = tsk->active_mm;
 743         tsk->mm = mm;
 744         tsk->active_mm = mm;
 745         activate_mm(active_mm, mm);
 746         task_unlock(tsk);
 747         arch_pick_mmap_layout(mm);
 748         if (old_mm) {
 749                 up_read(&old_mm->mmap_sem);
 750                 BUG_ON(active_mm != old_mm);
 751                 mm_update_next_owner(old_mm);
 752                 mmput(old_mm);
 753                 return 0;
 754         }
 755         mmdrop(active_mm);
 756         return 0;
 757 }
 758
 759 /*
 760  * This function makes sure the current process has its own signal table,
 761  * so that flush_signal_handlers can later reset the handlers without
 762  * disturbing other processes.  (Other processes might share the signal
 763  * table via the CLONE_SIGHAND option to clone().)
 764  */
 765 static int de_thread(struct task_struct *tsk)
 766 {
 767         struct signal_struct *sig = tsk->signal;
 768         struct sighand_struct *oldsighand = tsk->sighand;
 769         spinlock_t *lock = &oldsighand->siglock;
 770
 771         if (thread_group_empty(tsk))
 772                 goto no_thread_group;
 773
 774         /*
 775          * Kill all other threads in the thread group.
 776          */
 777         spin_lock_irq(lock);
 778         if (signal_group_exit(sig)) {
 779                 /*
 780                  * Another group action in progress, just
 781                  * return so that the signal is processed.
 782                  */
 783                 spin_unlock_irq(lock);
 784                 return -EAGAIN;
 785         }
 786
 787         sig->group_exit_task = tsk;
 788         sig->notify_count = zap_other_threads(tsk);
 789         if (!thread_group_leader(tsk))
 790                 sig->notify_count--;
 791
 792         while (sig->notify_count) {
 793                 __set_current_state(TASK_UNINTERRUPTIBLE);
 794                 spin_unlock_irq(lock);
 795                 schedule();
 796                 spin_lock_irq(lock);
 797         }
 798         spin_unlock_irq(lock);
 799
 800         /*
 801          * At this point all other threads have exited, all we have to
 802          * do is to wait for the thread group leader to become inactive,
 803          * and to assume its PID:
 804          */
 805         if (!thread_group_leader(tsk)) {
 806                 struct task_struct *leader = tsk->group_leader;
 807
 808                 sig->notify_count = -1; /* for exit_notify() */
 809                 for (;;) {
 810                         write_lock_irq(&tasklist_lock);
 811                         if (likely(leader->exit_state))
 812                                 break;
 813                         __set_current_state(TASK_UNINTERRUPTIBLE);
 814                         write_unlock_irq(&tasklist_lock);
 815                         schedule();
 816                 }
 817
 818                 /*
 819                  * The only record we have of the real-time age of a
 820                  * process, regardless of execs it's done, is start_time.
 821                  * All the past CPU time is accumulated in signal_struct
 822                  * from sister threads now dead.  But in this non-leader
 823                  * exec, nothing survives from the original leader thread,
 824                  * whose birth marks the true age of this process now.
 825                  * When we take on its identity by switching to its PID, we
 826                  * also take its birthdate (always earlier than our own).
 827                  */
 828                 tsk->start_time = leader->start_time;
 829
 830                 BUG_ON(!same_thread_group(leader, tsk));
 831                 BUG_ON(has_group_leader_pid(tsk));
 832                 /*
 833                  * An exec() starts a new thread group with the
 834                  * TGID of the previous thread group. Rehash the
 835                  * two threads with a switched PID, and release
 836                  * the former thread group leader:
 837                  */
 838
 839                 /* Become a process group leader with the old leader's pid.
 840                  * The old leader becomes a thread of the this thread group.
 841                  * Note: The old leader also uses this pid until release_task
 842                  *       is called.  Odd but simple and correct.
 843                  */
 844                 detach_pid(tsk, PIDTYPE_PID);
 845                 tsk->pid = leader->pid;
 846                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 847                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 848                 transfer_pid(leader, tsk, PIDTYPE_SID);
 849
 850                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 851                 list_replace_init(&leader->sibling, &tsk->sibling);
 852
 853                 tsk->group_leader = tsk;
 854                 leader->group_leader = tsk;
 855
 856                 tsk->exit_signal = SIGCHLD;
 857
 858                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 859                 leader->exit_state = EXIT_DEAD;
 860                 write_unlock_irq(&tasklist_lock);
 861
 862                 release_task(leader);
 863         }
 864
 865         sig->group_exit_task = NULL;
 866         sig->notify_count = 0;
 867
 868 no_thread_group:
 869         if (current->mm)
 870                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 871
 872         exit_itimers(sig);
 873         flush_itimer_signals();
 874
 875         if (atomic_read(&oldsighand->count) != 1) {
 876                 struct sighand_struct *newsighand;
 877                 /*
 878                  * This ->sighand is shared with the CLONE_SIGHAND
 879                  * but not CLONE_THREAD task, switch to the new one.
 880                  */
 881                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 882                 if (!newsighand)
 883                         return -ENOMEM;
 884
 885                 atomic_set(&newsighand->count, 1);
 886                 memcpy(newsighand->action, oldsighand->action,
 887                        sizeof(newsighand->action));
 888
 889                 write_lock_irq(&tasklist_lock);
 890                 spin_lock(&oldsighand->siglock);
 891                 rcu_assign_pointer(tsk->sighand, newsighand);
 892                 spin_unlock(&oldsighand->siglock);
 893                 write_unlock_irq(&tasklist_lock);
 894
 895                 __cleanup_sighand(oldsighand);
 896         }
 897
 898         BUG_ON(!thread_group_leader(tsk));
 899         return 0;
 900 }
 901
 902 /*
 903  * These functions flushes out all traces of the currently running executable
 904  * so that a new one can be started
 905  */
 906 static void flush_old_files(struct files_struct * files)
 907 {
 908         long j = -1;
 909         struct fdtable *fdt;
 910
 911         spin_lock(&files->file_lock);
 912         for (;;) {
 913                 unsigned long set, i;
 914
 915                 j++;
 916                 i = j * __NFDBITS;
 917                 fdt = files_fdtable(files);
 918                 if (i >= fdt->max_fds)
 919                         break;
 920                 set = fdt->close_on_exec->fds_bits[j];
 921                 if (!set)
 922                         continue;
 923                 fdt->close_on_exec->fds_bits[j] = 0;
 924                 spin_unlock(&files->file_lock);
 925                 for ( ; set ; i++,set >>= 1) {
 926                         if (set & 1) {
 927                                 sys_close(i);
 928                         }
 929                 }
 930                 spin_lock(&files->file_lock);
 931
 932         }
 933         spin_unlock(&files->file_lock);
 934 }
 935
 936 char *get_task_comm(char *buf, struct task_struct *tsk)
 937 {
 938         /* buf must be at least sizeof(tsk->comm) in size */
 939         task_lock(tsk);
 940         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 941         task_unlock(tsk);
 942         return buf;
 943 }
 944
 945 void set_task_comm(struct task_struct *tsk, char *buf)
 946 {
 947         task_lock(tsk);
 948
 949         /*
 950          * Threads may access current->comm without holding
 951          * the task lock, so write the string carefully.
 952          * Readers without a lock may see incomplete new
 953          * names but are safe from non-terminating string reads.
 954          */
 955         memset(tsk->comm, 0, TASK_COMM_LEN);
 956         wmb();
 957         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 958         task_unlock(tsk);
 959         perf_event_comm(tsk);
 960 }
 961
 962 int flush_old_exec(struct linux_binprm * bprm)
 963 {
 964         int retval;
 965
 966         /*
 967          * Make sure we have a private signal table and that
 968          * we are unassociated from the previous thread group.
 969          */
 970         retval = de_thread(current);
 971         if (retval)
 972                 goto out;
 973
 974         set_mm_exe_file(bprm->mm, bprm->file);
 975
 976         /*
 977          * Release all of the old mmap stuff
 978          */
 979         retval = exec_mmap(bprm->mm);
 980         if (retval)
 981                 goto out;
 982
 983         bprm->mm = NULL;                /* We're using it now */
 984
 985         current->flags &= ~PF_RANDOMIZE;
 986         flush_thread();
 987         current->personality &= ~bprm->per_clear;
 988
 989         return 0;
 990
 991 out:
 992         return retval;
 993 }
 994 EXPORT_SYMBOL(flush_old_exec);
 995
 996 void setup_new_exec(struct linux_binprm * bprm)
 997 {
 998         int i, ch;
 999         char * name;
1000         char tcomm[sizeof(current->comm)];
1001
1002         arch_pick_mmap_layout(current->mm);
1003
1004         /* This is the point of no return */
1005         current->sas_ss_sp = current->sas_ss_size = 0;
1006
1007         if (current_euid() == current_uid() && current_egid() == current_gid())
1008                 set_dumpable(current->mm, 1);
1009         else
1010                 set_dumpable(current->mm, suid_dumpable);
1011
1012         name = bprm->filename;
1013
1014         /* Copies the binary name from after last slash */
1015         for (i=0; (ch = *(name++)) != '\0';) {
1016                 if (ch == '/')
1017                         i = 0; /* overwrite what we wrote */
1018                 else
1019                         if (i < (sizeof(tcomm) - 1))
1020                                 tcomm[i++] = ch;
1021         }
1022         tcomm[i] = '\0';
1023         set_task_comm(current, tcomm);
1024
1025         /* Set the new mm task size. We have to do that late because it may
1026          * depend on TIF_32BIT which is only updated in flush_thread() on
1027          * some architectures like powerpc
1028          */
1029         current->mm->task_size = TASK_SIZE;
1030
1031         /* install the new credentials */
1032         if (bprm->cred->uid != current_euid() ||
1033             bprm->cred->gid != current_egid()) {
1034                 current->pdeath_signal = 0;
1035         } else if (file_permission(bprm->file, MAY_READ) ||
1036                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1037                 set_dumpable(current->mm, suid_dumpable);
1038         }
1039
1040         /*
1041          * Flush performance counters when crossing a
1042          * security domain:
1043          */
1044         if (!get_dumpable(current->mm))
1045                 perf_event_exit_task(current);
1046
1047         /* An exec changes our domain. We are no longer part of the thread
1048            group */
1049
1050         current->self_exec_id++;
1051
1052         flush_signal_handlers(current, 0);
1053         flush_old_files(current->files);
1054 }
1055 EXPORT_SYMBOL(setup_new_exec);
1056
1057 /*
1058  * Prepare credentials and lock ->cred_guard_mutex.
1059  * install_exec_creds() commits the new creds and drops the lock.
1060  * Or, if exec fails before, free_bprm() should release ->cred and
1061  * and unlock.
1062  */
1063 int prepare_bprm_creds(struct linux_binprm *bprm)
1064 {
1065         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1066                 return -ERESTARTNOINTR;
1067
1068         bprm->cred = prepare_exec_creds();
1069         if (likely(bprm->cred))
1070                 return 0;
1071
1072         mutex_unlock(&current->cred_guard_mutex);
1073         return -ENOMEM;
1074 }
1075
1076 void free_bprm(struct linux_binprm *bprm)
1077 {
1078         free_arg_pages(bprm);
1079         if (bprm->cred) {
1080                 mutex_unlock(&current->cred_guard_mutex);
1081                 abort_creds(bprm->cred);
1082         }
1083         kfree(bprm);
1084 }
1085
1086 /*
1087  * install the new credentials for this executable
1088  */
1089 void install_exec_creds(struct linux_binprm *bprm)
1090 {
1091         security_bprm_committing_creds(bprm);
1092
1093         commit_creds(bprm->cred);
1094         bprm->cred = NULL;
1095         /*
1096          * cred_guard_mutex must be held at least to this point to prevent
1097          * ptrace_attach() from altering our determination of the task's
1098          * credentials; any time after this it may be unlocked.
1099          */
1100         security_bprm_committed_creds(bprm);
1101         mutex_unlock(&current->cred_guard_mutex);
1102 }
1103 EXPORT_SYMBOL(install_exec_creds);
1104
1105 /*
1106  * determine how safe it is to execute the proposed program
1107  * - the caller must hold current->cred_guard_mutex to protect against
1108  *   PTRACE_ATTACH
1109  */
1110 int check_unsafe_exec(struct linux_binprm *bprm)
1111 {
1112         struct task_struct *p = current, *t;
1113         unsigned n_fs;
1114         int res = 0;
1115
1116         bprm->unsafe = tracehook_unsafe_exec(p);
1117
1118         n_fs = 1;
1119         write_lock(&p->fs->lock);
1120         rcu_read_lock();
1121         for (t = next_thread(p); t != p; t = next_thread(t)) {
1122                 if (t->fs == p->fs)
1123                         n_fs++;
1124         }
1125         rcu_read_unlock();
1126
1127         if (p->fs->users > n_fs) {
1128                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1129         } else {
1130                 res = -EAGAIN;
1131                 if (!p->fs->in_exec) {
1132                         p->fs->in_exec = 1;
1133                         res = 1;
1134                 }
1135         }
1136         write_unlock(&p->fs->lock);
1137
1138         return res;
1139 }
1140
1141 /*
1142  * Fill the binprm structure from the inode.
1143  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1144  *
1145  * This may be called multiple times for binary chains (scripts for example).
1146  */
1147 int prepare_binprm(struct linux_binprm *bprm)
1148 {
1149         umode_t mode;
1150         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1151         int retval;
1152
1153         mode = inode->i_mode;
1154         if (bprm->file->f_op == NULL)
1155                 return -EACCES;
1156
1157         /* clear any previous set[ug]id data from a previous binary */
1158         bprm->cred->euid = current_euid();
1159         bprm->cred->egid = current_egid();
1160
1161         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1162                 /* Set-uid? */
1163                 if (mode & S_ISUID) {
1164                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1165                         bprm->cred->euid = inode->i_uid;
1166                 }
1167
1168                 /* Set-gid? */
1169                 /*
1170                  * If setgid is set but no group execute bit then this
1171                  * is a candidate for mandatory locking, not a setgid
1172                  * executable.
1173                  */
1174                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1175                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1176                         bprm->cred->egid = inode->i_gid;
1177                 }
1178         }
1179
1180         /* fill in binprm security blob */
1181         retval = security_bprm_set_creds(bprm);
1182         if (retval)
1183                 return retval;
1184         bprm->cred_prepared = 1;
1185
1186         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1187         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1188 }
1189
1190 EXPORT_SYMBOL(prepare_binprm);
1191
1192 /*
1193  * Arguments are '\0' separated strings found at the location bprm->p
1194  * points to; chop off the first by relocating brpm->p to right after
1195  * the first '\0' encountered.
1196  */
1197 int remove_arg_zero(struct linux_binprm *bprm)
1198 {
1199         int ret = 0;
1200         unsigned long offset;
1201         char *kaddr;
1202         struct page *page;
1203
1204         if (!bprm->argc)
1205                 return 0;
1206
1207         do {
1208                 offset = bprm->p & ~PAGE_MASK;
1209                 page = get_arg_page(bprm, bprm->p, 0);
1210                 if (!page) {
1211                         ret = -EFAULT;
1212                         goto out;
1213                 }
1214                 kaddr = kmap_atomic(page, KM_USER0);
1215
1216                 for (; offset < PAGE_SIZE && kaddr[offset];
1217                                 offset++, bprm->p++)
1218                         ;
1219
1220                 kunmap_atomic(kaddr, KM_USER0);
1221                 put_arg_page(page);
1222
1223                 if (offset == PAGE_SIZE)
1224                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1225         } while (offset == PAGE_SIZE);
1226
1227         bprm->p++;
1228         bprm->argc--;
1229         ret = 0;
1230
1231 out:
1232         return ret;
1233 }
1234 EXPORT_SYMBOL(remove_arg_zero);
1235
1236 /*
1237  * cycle the list of binary formats handler, until one recognizes the image
1238  */
1239 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1240 {
1241         unsigned int depth = bprm->recursion_depth;
1242         int try,retval;
1243         struct linux_binfmt *fmt;
1244
1245         retval = security_bprm_check(bprm);
1246         if (retval)
1247                 return retval;
1248
1249         /* kernel module loader fixup */
1250         /* so we don't try to load run modprobe in kernel space. */
1251         set_fs(USER_DS);
1252
1253         retval = audit_bprm(bprm);
1254         if (retval)
1255                 return retval;
1256
1257         retval = -ENOENT;
1258         for (try=0; try<2; try++) {
1259                 read_lock(&binfmt_lock);
1260                 list_for_each_entry(fmt, &formats, lh) {
1261                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1262                         if (!fn)
1263                                 continue;
1264                         if (!try_module_get(fmt->module))
1265                                 continue;
1266                         read_unlock(&binfmt_lock);
1267                         retval = fn(bprm, regs);
1268                         /*
1269                          * Restore the depth counter to its starting value
1270                          * in this call, so we don't have to rely on every
1271                          * load_binary function to restore it on return.
1272                          */
1273                         bprm->recursion_depth = depth;
1274                         if (retval >= 0) {
1275                                 if (depth == 0)
1276                                         tracehook_report_exec(fmt, bprm, regs);
1277                                 put_binfmt(fmt);
1278                                 allow_write_access(bprm->file);
1279                                 if (bprm->file)
1280                                         fput(bprm->file);
1281                                 bprm->file = NULL;
1282                                 current->did_exec = 1;
1283                                 proc_exec_connector(current);
1284                                 return retval;
1285                         }
1286                         read_lock(&binfmt_lock);
1287                         put_binfmt(fmt);
1288                         if (retval != -ENOEXEC || bprm->mm == NULL)
1289                                 break;
1290                         if (!bprm->file) {
1291                                 read_unlock(&binfmt_lock);
1292                                 return retval;
1293                         }
1294                 }
1295                 read_unlock(&binfmt_lock);
1296                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1297                         break;
1298 #ifdef CONFIG_MODULES
1299                 } else {
1300 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1301                         if (printable(bprm->buf[0]) &&
1302                             printable(bprm->buf[1]) &&
1303                             printable(bprm->buf[2]) &&
1304                             printable(bprm->buf[3]))
1305                                 break; /* -ENOEXEC */
1306                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1307 #endif
1308                 }
1309         }
1310         return retval;
1311 }
1312
1313 EXPORT_SYMBOL(search_binary_handler);
1314
1315 /*
1316  * sys_execve() executes a new program.
1317  */
1318 int do_execve(char * filename,
1319         char __user *__user *argv,
1320         char __user *__user *envp,
1321         struct pt_regs * regs)
1322 {
1323         struct linux_binprm *bprm;
1324         struct file *file;
1325         struct files_struct *displaced;
1326         bool clear_in_exec;
1327         int retval;
1328
1329         retval = unshare_files(&displaced);
1330         if (retval)
1331                 goto out_ret;
1332
1333         retval = -ENOMEM;
1334         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1335         if (!bprm)
1336                 goto out_files;
1337
1338         retval = prepare_bprm_creds(bprm);
1339         if (retval)
1340                 goto out_free;
1341
1342         retval = check_unsafe_exec(bprm);
1343         if (retval < 0)
1344                 goto out_free;
1345         clear_in_exec = retval;
1346         current->in_execve = 1;
1347
1348         file = open_exec(filename);
1349         retval = PTR_ERR(file);
1350         if (IS_ERR(file))
1351                 goto out_unmark;
1352
1353         sched_exec();
1354
1355         bprm->file = file;
1356         bprm->filename = filename;
1357         bprm->interp = filename;
1358
1359         retval = bprm_mm_init(bprm);
1360         if (retval)
1361                 goto out_file;
1362
1363         bprm->argc = count(argv, MAX_ARG_STRINGS);
1364         if ((retval = bprm->argc) < 0)
1365                 goto out;
1366
1367         bprm->envc = count(envp, MAX_ARG_STRINGS);
1368         if ((retval = bprm->envc) < 0)
1369                 goto out;
1370
1371         retval = prepare_binprm(bprm);
1372         if (retval < 0)
1373                 goto out;
1374
1375         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1376         if (retval < 0)
1377                 goto out;
1378
1379         bprm->exec = bprm->p;
1380         retval = copy_strings(bprm->envc, envp, bprm);
1381         if (retval < 0)
1382                 goto out;
1383
1384         retval = copy_strings(bprm->argc, argv, bprm);
1385         if (retval < 0)
1386                 goto out;
1387
1388         current->flags &= ~PF_KTHREAD;
1389         retval = search_binary_handler(bprm,regs);
1390         if (retval < 0)
1391                 goto out;
1392
1393         /* execve succeeded */
1394         current->fs->in_exec = 0;
1395         current->in_execve = 0;
1396         acct_update_integrals(current);
1397         free_bprm(bprm);
1398         if (displaced)
1399                 put_files_struct(displaced);
1400         return retval;
1401
1402 out:
1403         if (bprm->mm)
1404                 mmput (bprm->mm);
1405
1406 out_file:
1407         if (bprm->file) {
1408                 allow_write_access(bprm->file);
1409                 fput(bprm->file);
1410         }
1411
1412 out_unmark:
1413         if (clear_in_exec)
1414                 current->fs->in_exec = 0;
1415         current->in_execve = 0;
1416
1417 out_free:
1418         free_bprm(bprm);
1419
1420 out_files:
1421         if (displaced)
1422                 reset_files_struct(displaced);
1423 out_ret:
1424         return retval;
1425 }
1426
1427 void set_binfmt(struct linux_binfmt *new)
1428 {
1429         struct mm_struct *mm = current->mm;
1430
1431         if (mm->binfmt)
1432                 module_put(mm->binfmt->module);
1433
1434         mm->binfmt = new;
1435         if (new)
1436                 __module_get(new->module);
1437 }
1438
1439 EXPORT_SYMBOL(set_binfmt);
1440
1441 /* format_corename will inspect the pattern parameter, and output a
1442  * name into corename, which must have space for at least
1443  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1444  */
1445 static int format_corename(char *corename, long signr)
1446 {
1447         const struct cred *cred = current_cred();
1448         const char *pat_ptr = core_pattern;
1449         int ispipe = (*pat_ptr == '|');
1450         char *out_ptr = corename;
1451         char *const out_end = corename + CORENAME_MAX_SIZE;
1452         int rc;
1453         int pid_in_pattern = 0;
1454
1455         /* Repeat as long as we have more pattern to process and more output
1456            space */
1457         while (*pat_ptr) {
1458                 if (*pat_ptr != '%') {
1459                         if (out_ptr == out_end)
1460                                 goto out;
1461                         *out_ptr++ = *pat_ptr++;
1462                 } else {
1463                         switch (*++pat_ptr) {
1464                         case 0:
1465                                 goto out;
1466                         /* Double percent, output one percent */
1467                         case '%':
1468                                 if (out_ptr == out_end)
1469                                         goto out;
1470                                 *out_ptr++ = '%';
1471                                 break;
1472                         /* pid */
1473                         case 'p':
1474                                 pid_in_pattern = 1;
1475                                 rc = snprintf(out_ptr, out_end - out_ptr,
1476                                               "%d", task_tgid_vnr(current));
1477                                 if (rc > out_end - out_ptr)
1478                                         goto out;
1479                                 out_ptr += rc;
1480                                 break;
1481                         /* uid */
1482                         case 'u':
1483                                 rc = snprintf(out_ptr, out_end - out_ptr,
1484                                               "%d", cred->uid);
1485                                 if (rc > out_end - out_ptr)
1486                                         goto out;
1487                                 out_ptr += rc;
1488                                 break;
1489                         /* gid */
1490                         case 'g':
1491                                 rc = snprintf(out_ptr, out_end - out_ptr,
1492                                               "%d", cred->gid);
1493                                 if (rc > out_end - out_ptr)
1494                                         goto out;
1495                                 out_ptr += rc;
1496                                 break;
1497                         /* signal that caused the coredump */
1498                         case 's':
1499                                 rc = snprintf(out_ptr, out_end - out_ptr,
1500                                               "%ld", signr);
1501                                 if (rc > out_end - out_ptr)
1502                                         goto out;
1503                                 out_ptr += rc;
1504                                 break;
1505                         /* UNIX time of coredump */
1506                         case 't': {
1507                                 struct timeval tv;
1508                                 do_gettimeofday(&tv);
1509                                 rc = snprintf(out_ptr, out_end - out_ptr,
1510                                               "%lu", tv.tv_sec);
1511                                 if (rc > out_end - out_ptr)
1512                                         goto out;
1513                                 out_ptr += rc;
1514                                 break;
1515                         }
1516                         /* hostname */
1517                         case 'h':
1518                                 down_read(&uts_sem);
1519                                 rc = snprintf(out_ptr, out_end - out_ptr,
1520                                               "%s", utsname()->nodename);
1521                                 up_read(&uts_sem);
1522                                 if (rc > out_end - out_ptr)
1523                                         goto out;
1524                                 out_ptr += rc;
1525                                 break;
1526                         /* executable */
1527                         case 'e':
1528                                 rc = snprintf(out_ptr, out_end - out_ptr,
1529                                               "%s", current->comm);
1530                                 if (rc > out_end - out_ptr)
1531                                         goto out;
1532                                 out_ptr += rc;
1533                                 break;
1534                         /* core limit size */
1535                         case 'c':
1536                                 rc = snprintf(out_ptr, out_end - out_ptr,
1537                                               "%lu", rlimit(RLIMIT_CORE));
1538                                 if (rc > out_end - out_ptr)
1539                                         goto out;
1540                                 out_ptr += rc;
1541                                 break;
1542                         default:
1543                                 break;
1544                         }
1545                         ++pat_ptr;
1546                 }
1547         }
1548         /* Backward compatibility with core_uses_pid:
1549          *
1550          * If core_pattern does not include a %p (as is the default)
1551          * and core_uses_pid is set, then .%pid will be appended to
1552          * the filename. Do not do this for piped commands. */
1553         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1554                 rc = snprintf(out_ptr, out_end - out_ptr,
1555                               ".%d", task_tgid_vnr(current));
1556                 if (rc > out_end - out_ptr)
1557                         goto out;
1558                 out_ptr += rc;
1559         }
1560 out:
1561         *out_ptr = 0;
1562         return ispipe;
1563 }
1564
1565 static int zap_process(struct task_struct *start, int exit_code)
1566 {
1567         struct task_struct *t;
1568         int nr = 0;
1569
1570         start->signal->flags = SIGNAL_GROUP_EXIT;
1571         start->signal->group_exit_code = exit_code;
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                 nr = zap_process(tsk, exit_code);
1597         }
1598         spin_unlock_irq(&tsk->sighand->siglock);
1599         if (unlikely(nr < 0))
1600                 return nr;
1601
1602         if (atomic_read(&mm->mm_users) == nr + 1)
1603                 goto done;
1604         /*
1605          * We should find and kill all tasks which use this mm, and we should
1606          * count them correctly into ->nr_threads. We don't take tasklist
1607          * lock, but this is safe wrt:
1608          *
1609          * fork:
1610          *      None of sub-threads can fork after zap_process(leader). All
1611          *      processes which were created before this point should be
1612          *      visible to zap_threads() because copy_process() adds the new
1613          *      process to the tail of init_task.tasks list, and lock/unlock
1614          *      of ->siglock provides a memory barrier.
1615          *
1616          * do_exit:
1617          *      The caller holds mm->mmap_sem. This means that the task which
1618          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1619          *      its ->mm.
1620          *
1621          * de_thread:
1622          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1623          *      we must see either old or new leader, this does not matter.
1624          *      However, it can change p->sighand, so lock_task_sighand(p)
1625          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1626          *      it can't fail.
1627          *
1628          *      Note also that "g" can be the old leader with ->mm == NULL
1629          *      and already unhashed and thus removed from ->thread_group.
1630          *      This is OK, __unhash_process()->list_del_rcu() does not
1631          *      clear the ->next pointer, we will find the new leader via
1632          *      next_thread().
1633          */
1634         rcu_read_lock();
1635         for_each_process(g) {
1636                 if (g == tsk->group_leader)
1637                         continue;
1638                 if (g->flags & PF_KTHREAD)
1639                         continue;
1640                 p = g;
1641                 do {
1642                         if (p->mm) {
1643                                 if (unlikely(p->mm == mm)) {
1644                                         lock_task_sighand(p, &flags);
1645                                         nr += zap_process(p, exit_code);
1646                                         unlock_task_sighand(p, &flags);
1647                                 }
1648                                 break;
1649                         }
1650                 } while_each_thread(g, p);
1651         }
1652         rcu_read_unlock();
1653 done:
1654         atomic_set(&core_state->nr_threads, nr);
1655         return nr;
1656 }
1657
1658 static int coredump_wait(int exit_code, struct core_state *core_state)
1659 {
1660         struct task_struct *tsk = current;
1661         struct mm_struct *mm = tsk->mm;
1662         struct completion *vfork_done;
1663         int core_waiters = -EBUSY;
1664
1665         init_completion(&core_state->startup);
1666         core_state->dumper.task = tsk;
1667         core_state->dumper.next = NULL;
1668
1669         down_write(&mm->mmap_sem);
1670         if (!mm->core_state)
1671                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1672         up_write(&mm->mmap_sem);
1673
1674         if (unlikely(core_waiters < 0))
1675                 goto fail;
1676
1677         /*
1678          * Make sure nobody is waiting for us to release the VM,
1679          * otherwise we can deadlock when we wait on each other
1680          */
1681         vfork_done = tsk->vfork_done;
1682         if (vfork_done) {
1683                 tsk->vfork_done = NULL;
1684                 complete(vfork_done);
1685         }
1686
1687         if (core_waiters)
1688                 wait_for_completion(&core_state->startup);
1689 fail:
1690         return core_waiters;
1691 }
1692
1693 static void coredump_finish(struct mm_struct *mm)
1694 {
1695         struct core_thread *curr, *next;
1696         struct task_struct *task;
1697
1698         next = mm->core_state->dumper.next;
1699         while ((curr = next) != NULL) {
1700                 next = curr->next;
1701                 task = curr->task;
1702                 /*
1703                  * see exit_mm(), curr->task must not see
1704                  * ->task == NULL before we read ->next.
1705                  */
1706                 smp_mb();
1707                 curr->task = NULL;
1708                 wake_up_process(task);
1709         }
1710
1711         mm->core_state = NULL;
1712 }
1713
1714 /*
1715  * set_dumpable converts traditional three-value dumpable to two flags and
1716  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1717  * these bits are not changed atomically.  So get_dumpable can observe the
1718  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1719  * return either old dumpable or new one by paying attention to the order of
1720  * modifying the bits.
1721  *
1722  * dumpable |   mm->flags (binary)
1723  * old  new | initial interim  final
1724  * ---------+-----------------------
1725  *  0    1  |   00      01      01
1726  *  0    2  |   00      10(*)   11
1727  *  1    0  |   01      00      00
1728  *  1    2  |   01      11      11
1729  *  2    0  |   11      10(*)   00
1730  *  2    1  |   11      11      01
1731  *
1732  * (*) get_dumpable regards interim value of 10 as 11.
1733  */
1734 void set_dumpable(struct mm_struct *mm, int value)
1735 {
1736         switch (value) {
1737         case 0:
1738                 clear_bit(MMF_DUMPABLE, &mm->flags);
1739                 smp_wmb();
1740                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1741                 break;
1742         case 1:
1743                 set_bit(MMF_DUMPABLE, &mm->flags);
1744                 smp_wmb();
1745                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1746                 break;
1747         case 2:
1748                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1749                 smp_wmb();
1750                 set_bit(MMF_DUMPABLE, &mm->flags);
1751                 break;
1752         }
1753 }
1754
1755 static int __get_dumpable(unsigned long mm_flags)
1756 {
1757         int ret;
1758
1759         ret = mm_flags & MMF_DUMPABLE_MASK;
1760         return (ret >= 2) ? 2 : ret;
1761 }
1762
1763 int get_dumpable(struct mm_struct *mm)
1764 {
1765         return __get_dumpable(mm->flags);
1766 }
1767
1768 static void wait_for_dump_helpers(struct file *file)
1769 {
1770         struct pipe_inode_info *pipe;
1771
1772         pipe = file->f_path.dentry->d_inode->i_pipe;
1773
1774         pipe_lock(pipe);
1775         pipe->readers++;
1776         pipe->writers--;
1777
1778         while ((pipe->readers > 1) && (!signal_pending(current))) {
1779                 wake_up_interruptible_sync(&pipe->wait);
1780                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1781                 pipe_wait(pipe);
1782         }
1783
1784         pipe->readers--;
1785         pipe->writers++;
1786         pipe_unlock(pipe);
1787
1788 }
1789
1790
1791 /*
1792  * uhm_pipe_setup
1793  * helper function to customize the process used
1794  * to collect the core in userspace.  Specifically
1795  * it sets up a pipe and installs it as fd 0 (stdin)
1796  * for the process.  Returns 0 on success, or
1797  * PTR_ERR on failure.
1798  * Note that it also sets the core limit to 1.  This
1799  * is a special value that we use to trap recursive
1800  * core dumps
1801  */
1802 static int umh_pipe_setup(struct subprocess_info *info)
1803 {
1804         struct file *rp, *wp;
1805         struct fdtable *fdt;
1806         struct coredump_params *cp = (struct coredump_params *)info->data;
1807         struct files_struct *cf = current->files;
1808
1809         wp = create_write_pipe(0);
1810         if (IS_ERR(wp))
1811                 return PTR_ERR(wp);
1812
1813         rp = create_read_pipe(wp, 0);
1814         if (IS_ERR(rp)) {
1815                 free_write_pipe(wp);
1816                 return PTR_ERR(rp);
1817         }
1818
1819         cp->file = wp;
1820
1821         sys_close(0);
1822         fd_install(0, rp);
1823         spin_lock(&cf->file_lock);
1824         fdt = files_fdtable(cf);
1825         FD_SET(0, fdt->open_fds);
1826         FD_CLR(0, fdt->close_on_exec);
1827         spin_unlock(&cf->file_lock);
1828
1829         /* and disallow core files too */
1830         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1831
1832         return 0;
1833 }
1834
1835 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1836 {
1837         struct core_state core_state;
1838         char corename[CORENAME_MAX_SIZE + 1];
1839         struct mm_struct *mm = current->mm;
1840         struct linux_binfmt * binfmt;
1841         const struct cred *old_cred;
1842         struct cred *cred;
1843         int retval = 0;
1844         int flag = 0;
1845         int ispipe;
1846         static atomic_t core_dump_count = ATOMIC_INIT(0);
1847         struct coredump_params cprm = {
1848                 .signr = signr,
1849                 .regs = regs,
1850                 .limit = rlimit(RLIMIT_CORE),
1851                 /*
1852                  * We must use the same mm->flags while dumping core to avoid
1853                  * inconsistency of bit flags, since this flag is not protected
1854                  * by any locks.
1855                  */
1856                 .mm_flags = mm->flags,
1857         };
1858
1859         audit_core_dumps(signr);
1860
1861         binfmt = mm->binfmt;
1862         if (!binfmt || !binfmt->core_dump)
1863                 goto fail;
1864         if (!__get_dumpable(cprm.mm_flags))
1865                 goto fail;
1866
1867         cred = prepare_creds();
1868         if (!cred)
1869                 goto fail;
1870         /*
1871          *      We cannot trust fsuid as being the "true" uid of the
1872          *      process nor do we know its entire history. We only know it
1873          *      was tainted so we dump it as root in mode 2.
1874          */
1875         if (__get_dumpable(cprm.mm_flags) == 2) {
1876                 /* Setuid core dump mode */
1877                 flag = O_EXCL;          /* Stop rewrite attacks */
1878                 cred->fsuid = 0;        /* Dump root private */
1879         }
1880
1881         retval = coredump_wait(exit_code, &core_state);
1882         if (retval < 0)
1883                 goto fail_creds;
1884
1885         old_cred = override_creds(cred);
1886
1887         /*
1888          * Clear any false indication of pending signals that might
1889          * be seen by the filesystem code called to write the core file.
1890          */
1891         clear_thread_flag(TIF_SIGPENDING);
1892
1893         ispipe = format_corename(corename, signr);
1894
1895         if (ispipe) {
1896                 int dump_count;
1897                 char **helper_argv;
1898
1899                 if (cprm.limit == 1) {
1900                         /*
1901                          * Normally core limits are irrelevant to pipes, since
1902                          * we're not writing to the file system, but we use
1903                          * cprm.limit of 1 here as a speacial value. Any
1904                          * non-1 limit gets set to RLIM_INFINITY below, but
1905                          * a limit of 0 skips the dump.  This is a consistent
1906                          * way to catch recursive crashes.  We can still crash
1907                          * if the core_pattern binary sets RLIM_CORE =  !1
1908                          * but it runs as root, and can do lots of stupid things
1909                          * Note that we use task_tgid_vnr here to grab the pid
1910                          * of the process group leader.  That way we get the
1911                          * right pid if a thread in a multi-threaded
1912                          * core_pattern process dies.
1913                          */
1914                         printk(KERN_WARNING
1915                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1916                                 task_tgid_vnr(current), current->comm);
1917                         printk(KERN_WARNING "Aborting core\n");
1918                         goto fail_unlock;
1919                 }
1920                 cprm.limit = RLIM_INFINITY;
1921
1922                 dump_count = atomic_inc_return(&core_dump_count);
1923                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1924                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1925                                task_tgid_vnr(current), current->comm);
1926                         printk(KERN_WARNING "Skipping core dump\n");
1927                         goto fail_dropcount;
1928                 }
1929
1930                 helper_argv = argv_split(GFP_KERNEL, corename+1, NULL);
1931                 if (!helper_argv) {
1932                         printk(KERN_WARNING "%s failed to allocate memory\n",
1933                                __func__);
1934                         goto fail_dropcount;
1935                 }
1936
1937                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
1938                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
1939                                         NULL, &cprm);
1940                 argv_free(helper_argv);
1941                 if (retval) {
1942                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1943                                corename);
1944                         goto close_fail;
1945                 }
1946         } else {
1947                 struct inode *inode;
1948
1949                 if (cprm.limit < binfmt->min_coredump)
1950                         goto fail_unlock;
1951
1952                 cprm.file = filp_open(corename,
1953                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1954                                  0600);
1955                 if (IS_ERR(cprm.file))
1956                         goto fail_unlock;
1957
1958                 inode = cprm.file->f_path.dentry->d_inode;
1959                 if (inode->i_nlink > 1)
1960                         goto close_fail;
1961                 if (d_unhashed(cprm.file->f_path.dentry))
1962                         goto close_fail;
1963                 /*
1964                  * AK: actually i see no reason to not allow this for named
1965                  * pipes etc, but keep the previous behaviour for now.
1966                  */
1967                 if (!S_ISREG(inode->i_mode))
1968                         goto close_fail;
1969                 /*
1970                  * Dont allow local users get cute and trick others to coredump
1971                  * into their pre-created files.
1972                  */
1973                 if (inode->i_uid != current_fsuid())
1974                         goto close_fail;
1975                 if (!cprm.file->f_op || !cprm.file->f_op->write)
1976                         goto close_fail;
1977                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
1978                         goto close_fail;
1979         }
1980
1981         retval = binfmt->core_dump(&cprm);
1982         if (retval)
1983                 current->signal->group_exit_code |= 0x80;
1984
1985         if (ispipe && core_pipe_limit)
1986                 wait_for_dump_helpers(cprm.file);
1987 close_fail:
1988         if (cprm.file)
1989                 filp_close(cprm.file, NULL);
1990 fail_dropcount:
1991         if (ispipe)
1992                 atomic_dec(&core_dump_count);
1993 fail_unlock:
1994         coredump_finish(mm);
1995         revert_creds(old_cred);
1996 fail_creds:
1997         put_cred(cred);
1998 fail:
1999         return;
2000 }