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/config.h>
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/mman.h>
  29 #include <linux/a.out.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/module.h>
  42 #include <linux/namei.h>
  43 #include <linux/proc_fs.h>
  44 #include <linux/ptrace.h>
  45 #include <linux/mount.h>
  46 #include <linux/security.h>
  47 #include <linux/rmap-locking.h>
  48
  49 #include <asm/uaccess.h>
  50 #include <asm/pgalloc.h>
  51 #include <asm/mmu_context.h>
  52
  53 #ifdef CONFIG_KMOD
  54 #include <linux/kmod.h>
  55 #endif
  56
  57 int core_uses_pid;
  58 char core_pattern[65] = "core";
  59 /* The maximal length of core_pattern is also specified in sysctl.c */
  60
  61 static struct linux_binfmt *formats;
  62 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  63
  64 int register_binfmt(struct linux_binfmt * fmt)
  65 {
  66         struct linux_binfmt ** tmp = &formats;
  67
  68         if (!fmt)
  69                 return -EINVAL;
  70         if (fmt->next)
  71                 return -EBUSY;
  72         write_lock(&binfmt_lock);
  73         while (*tmp) {
  74                 if (fmt == *tmp) {
  75                         write_unlock(&binfmt_lock);
  76                         return -EBUSY;
  77                 }
  78                 tmp = &(*tmp)->next;
  79         }
  80         fmt->next = formats;
  81         formats = fmt;
  82         write_unlock(&binfmt_lock);
  83         return 0;
  84 }
  85
  86 int unregister_binfmt(struct linux_binfmt * fmt)
  87 {
  88         struct linux_binfmt ** tmp = &formats;
  89
  90         write_lock(&binfmt_lock);
  91         while (*tmp) {
  92                 if (fmt == *tmp) {
  93                         *tmp = fmt->next;
  94                         write_unlock(&binfmt_lock);
  95                         return 0;
  96                 }
  97                 tmp = &(*tmp)->next;
  98         }
  99         write_unlock(&binfmt_lock);
 100         return -EINVAL;
 101 }
 102
 103 static inline void put_binfmt(struct linux_binfmt * fmt)
 104 {
 105         module_put(fmt->module);
 106 }
 107
 108 /*
 109  * Note that a shared library must be both readable and executable due to
 110  * security reasons.
 111  *
 112  * Also note that we take the address to load from from the file itself.
 113  */
 114 asmlinkage long sys_uselib(const char __user * library)
 115 {
 116         struct file * file;
 117         struct nameidata nd;
 118         int error;
 119
 120         error = user_path_walk(library, &nd);
 121         if (error)
 122                 goto out;
 123
 124         error = -EINVAL;
 125         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 126                 goto exit;
 127
 128         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC);
 129         if (error)
 130                 goto exit;
 131
 132         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 133         error = PTR_ERR(file);
 134         if (IS_ERR(file))
 135                 goto out;
 136
 137         error = -ENOEXEC;
 138         if(file->f_op) {
 139                 struct linux_binfmt * fmt;
 140
 141                 read_lock(&binfmt_lock);
 142                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 143                         if (!fmt->load_shlib)
 144                                 continue;
 145                         if (!try_module_get(fmt->module))
 146                                 continue;
 147                         read_unlock(&binfmt_lock);
 148                         error = fmt->load_shlib(file);
 149                         read_lock(&binfmt_lock);
 150                         put_binfmt(fmt);
 151                         if (error != -ENOEXEC)
 152                                 break;
 153                 }
 154                 read_unlock(&binfmt_lock);
 155         }
 156         fput(file);
 157 out:
 158         return error;
 159 exit:
 160         path_release(&nd);
 161         goto out;
 162 }
 163
 164 /*
 165  * count() counts the number of strings in array ARGV.
 166  */
 167 static int count(char __user * __user * argv, int max)
 168 {
 169         int i = 0;
 170
 171         if (argv != NULL) {
 172                 for (;;) {
 173                         char __user * p;
 174
 175                         if (get_user(p, argv))
 176                                 return -EFAULT;
 177                         if (!p)
 178                                 break;
 179                         argv++;
 180                         if(++i > max)
 181                                 return -E2BIG;
 182                 }
 183         }
 184         return i;
 185 }
 186
 187 /*
 188  * 'copy_strings()' copies argument/environment strings from user
 189  * memory to free pages in kernel mem. These are in a format ready
 190  * to be put directly into the top of new user memory.
 191  */
 192 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 193 {
 194         struct page *kmapped_page = NULL;
 195         char *kaddr = NULL;
 196         int ret;
 197
 198         while (argc-- > 0) {
 199                 char __user *str;
 200                 int len;
 201                 unsigned long pos;
 202
 203                 if (get_user(str, argv+argc) ||
 204                                 !(len = strnlen_user(str, bprm->p))) {
 205                         ret = -EFAULT;
 206                         goto out;
 207                 }
 208
 209                 if (bprm->p < len)  {
 210                         ret = -E2BIG;
 211                         goto out;
 212                 }
 213
 214                 bprm->p -= len;
 215                 /* XXX: add architecture specific overflow check here. */
 216                 pos = bprm->p;
 217
 218                 while (len > 0) {
 219                         int i, new, err;
 220                         int offset, bytes_to_copy;
 221                         struct page *page;
 222
 223                         offset = pos % PAGE_SIZE;
 224                         i = pos/PAGE_SIZE;
 225                         page = bprm->page[i];
 226                         new = 0;
 227                         if (!page) {
 228                                 page = alloc_page(GFP_HIGHUSER);
 229                                 bprm->page[i] = page;
 230                                 if (!page) {
 231                                         ret = -ENOMEM;
 232                                         goto out;
 233                                 }
 234                                 new = 1;
 235                         }
 236
 237                         if (page != kmapped_page) {
 238                                 if (kmapped_page)
 239                                         kunmap(kmapped_page);
 240                                 kmapped_page = page;
 241                                 kaddr = kmap(kmapped_page);
 242                         }
 243                         if (new && offset)
 244                                 memset(kaddr, 0, offset);
 245                         bytes_to_copy = PAGE_SIZE - offset;
 246                         if (bytes_to_copy > len) {
 247                                 bytes_to_copy = len;
 248                                 if (new)
 249                                         memset(kaddr+offset+len, 0,
 250                                                 PAGE_SIZE-offset-len);
 251                         }
 252                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 253                         if (err) {
 254                                 ret = -EFAULT;
 255                                 goto out;
 256                         }
 257
 258                         pos += bytes_to_copy;
 259                         str += bytes_to_copy;
 260                         len -= bytes_to_copy;
 261                 }
 262         }
 263         ret = 0;
 264 out:
 265         if (kmapped_page)
 266                 kunmap(kmapped_page);
 267         return ret;
 268 }
 269
 270 /*
 271  * Like copy_strings, but get argv and its values from kernel memory.
 272  */
 273 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 274 {
 275         int r;
 276         mm_segment_t oldfs = get_fs();
 277         set_fs(KERNEL_DS);
 278         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 279         set_fs(oldfs);
 280         return r;
 281 }
 282
 283 #ifdef CONFIG_MMU
 284 /*
 285  * This routine is used to map in a page into an address space: needed by
 286  * execve() for the initial stack and environment pages.
 287  *
 288  * tsk->mmap_sem is held for writing.
 289  */
 290 void put_dirty_page(struct task_struct *tsk, struct page *page,
 291                         unsigned long address, pgprot_t prot)
 292 {
 293         pgd_t * pgd;
 294         pmd_t * pmd;
 295         pte_t * pte;
 296         struct pte_chain *pte_chain;
 297
 298         if (page_count(page) != 1)
 299                 printk(KERN_ERR "mem_map disagrees with %p at %08lx\n",
 300                                 page, address);
 301
 302         pgd = pgd_offset(tsk->mm, address);
 303         pte_chain = pte_chain_alloc(GFP_KERNEL);
 304         if (!pte_chain)
 305                 goto out_sig;
 306         spin_lock(&tsk->mm->page_table_lock);
 307         pmd = pmd_alloc(tsk->mm, pgd, address);
 308         if (!pmd)
 309                 goto out;
 310         pte = pte_alloc_map(tsk->mm, pmd, address);
 311         if (!pte)
 312                 goto out;
 313         if (!pte_none(*pte)) {
 314                 pte_unmap(pte);
 315                 goto out;
 316         }
 317         lru_cache_add_active(page);
 318         flush_dcache_page(page);
 319         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(page, prot))));
 320         pte_chain = page_add_rmap(page, pte, pte_chain);
 321         pte_unmap(pte);
 322         tsk->mm->rss++;
 323         spin_unlock(&tsk->mm->page_table_lock);
 324
 325         /* no need for flush_tlb */
 326         pte_chain_free(pte_chain);
 327         return;
 328 out:
 329         spin_unlock(&tsk->mm->page_table_lock);
 330 out_sig:
 331         __free_page(page);
 332         force_sig(SIGKILL, tsk);
 333         pte_chain_free(pte_chain);
 334         return;
 335 }
 336
 337 int setup_arg_pages(struct linux_binprm *bprm)
 338 {
 339         unsigned long stack_base;
 340         struct vm_area_struct *mpnt;
 341         struct mm_struct *mm = current->mm;
 342         int i;
 343
 344 #ifdef CONFIG_STACK_GROWSUP
 345         /* Move the argument and environment strings to the bottom of the
 346          * stack space.
 347          */
 348         int offset, j;
 349         char *to, *from;
 350
 351         /* Start by shifting all the pages down */
 352         i = 0;
 353         for (j = 0; j < MAX_ARG_PAGES; j++) {
 354                 struct page *page = bprm->page[j];
 355                 if (!page)
 356                         continue;
 357                 bprm->page[i++] = page;
 358         }
 359
 360         /* Now move them within their pages */
 361         offset = bprm->p % PAGE_SIZE;
 362         to = kmap(bprm->page[0]);
 363         for (j = 1; j < i; j++) {
 364                 memmove(to, to + offset, PAGE_SIZE - offset);
 365                 from = kmap(bprm->page[j]);
 366                 memcpy(to + PAGE_SIZE - offset, from, offset);
 367                 kunmap(bprm->page[j - 1]);
 368                 to = from;
 369         }
 370         memmove(to, to + offset, PAGE_SIZE - offset);
 371         kunmap(bprm->page[j - 1]);
 372
 373         /* Adjust bprm->p to point to the end of the strings. */
 374         bprm->p = PAGE_SIZE * i - offset;
 375         stack_base = STACK_TOP - current->rlim[RLIMIT_STACK].rlim_max;
 376         mm->arg_start = stack_base;
 377
 378         /* zero pages that were copied above */
 379         while (i < MAX_ARG_PAGES)
 380                 bprm->page[i++] = NULL;
 381 #else
 382         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 383         mm->arg_start = bprm->p + stack_base;
 384 #endif
 385
 386         bprm->p += stack_base;
 387         if (bprm->loader)
 388                 bprm->loader += stack_base;
 389         bprm->exec += stack_base;
 390
 391         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 392         if (!mpnt)
 393                 return -ENOMEM;
 394
 395         if (!vm_enough_memory((STACK_TOP - (PAGE_MASK & (unsigned long) bprm->p))>>PAGE_SHIFT)) {
 396                 kmem_cache_free(vm_area_cachep, mpnt);
 397                 return -ENOMEM;
 398         }
 399
 400         down_write(&mm->mmap_sem);
 401         {
 402                 mpnt->vm_mm = mm;
 403 #ifdef CONFIG_STACK_GROWSUP
 404                 mpnt->vm_start = stack_base;
 405                 mpnt->vm_end = PAGE_MASK &
 406                         (PAGE_SIZE - 1 + (unsigned long) bprm->p);
 407 #else
 408                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 409                 mpnt->vm_end = STACK_TOP;
 410 #endif
 411                 mpnt->vm_page_prot = protection_map[VM_STACK_FLAGS & 0x7];
 412                 mpnt->vm_flags = VM_STACK_FLAGS;
 413                 mpnt->vm_ops = NULL;
 414                 mpnt->vm_pgoff = 0;
 415                 mpnt->vm_file = NULL;
 416                 INIT_LIST_HEAD(&mpnt->shared);
 417                 mpnt->vm_private_data = (void *) 0;
 418                 insert_vm_struct(mm, mpnt);
 419                 mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 420         }
 421
 422         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 423                 struct page *page = bprm->page[i];
 424                 if (page) {
 425                         bprm->page[i] = NULL;
 426                         put_dirty_page(current, page, stack_base,
 427                                         mpnt->vm_page_prot);
 428                 }
 429                 stack_base += PAGE_SIZE;
 430         }
 431         up_write(&mm->mmap_sem);
 432
 433         return 0;
 434 }
 435
 436 #define free_arg_pages(bprm) do { } while (0)
 437
 438 #else
 439
 440 static inline void free_arg_pages(struct linux_binprm *bprm)
 441 {
 442         int i;
 443
 444         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 445                 if (bprm->page[i])
 446                 __free_page(bprm->page[i]);
 447                 bprm->page[i] = NULL;
 448         }
 449 }
 450
 451 #endif /* CONFIG_MMU */
 452
 453 struct file *open_exec(const char *name)
 454 {
 455         struct nameidata nd;
 456         int err = path_lookup(name, LOOKUP_FOLLOW, &nd);
 457         struct file *file = ERR_PTR(err);
 458
 459         if (!err) {
 460                 struct inode *inode = nd.dentry->d_inode;
 461                 file = ERR_PTR(-EACCES);
 462                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 463                     S_ISREG(inode->i_mode)) {
 464                         int err = permission(inode, MAY_EXEC);
 465                         if (!err && !(inode->i_mode & 0111))
 466                                 err = -EACCES;
 467                         file = ERR_PTR(err);
 468                         if (!err) {
 469                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 470                                 if (!IS_ERR(file)) {
 471                                         err = deny_write_access(file);
 472                                         if (err) {
 473                                                 fput(file);
 474                                                 file = ERR_PTR(err);
 475                                         }
 476                                 }
 477 out:
 478                                 return file;
 479                         }
 480                 }
 481                 path_release(&nd);
 482         }
 483         goto out;
 484 }
 485
 486 int kernel_read(struct file *file, unsigned long offset,
 487         char *addr, unsigned long count)
 488 {
 489         mm_segment_t old_fs;
 490         loff_t pos = offset;
 491         int result;
 492
 493         old_fs = get_fs();
 494         set_fs(get_ds());
 495         /* The cast to a user pointer is valid due to the set_fs() */
 496         result = vfs_read(file, (void __user *)addr, count, &pos);
 497         set_fs(old_fs);
 498         return result;
 499 }
 500
 501 static int exec_mmap(struct mm_struct *mm)
 502 {
 503         struct task_struct *tsk;
 504         struct mm_struct * old_mm, *active_mm;
 505
 506         /* Add it to the list of mm's */
 507         spin_lock(&mmlist_lock);
 508         list_add(&mm->mmlist, &init_mm.mmlist);
 509         mmlist_nr++;
 510         spin_unlock(&mmlist_lock);
 511
 512         /* Notify parent that we're no longer interested in the old VM */
 513         tsk = current;
 514         old_mm = current->mm;
 515         mm_release(tsk, old_mm);
 516
 517         task_lock(tsk);
 518         active_mm = tsk->active_mm;
 519         tsk->mm = mm;
 520         tsk->active_mm = mm;
 521         activate_mm(active_mm, mm);
 522         task_unlock(tsk);
 523         if (old_mm) {
 524                 if (active_mm != old_mm) BUG();
 525                 mmput(old_mm);
 526                 return 0;
 527         }
 528         mmdrop(active_mm);
 529         return 0;
 530 }
 531
 532 /*
 533  * This function makes sure the current process has its own signal table,
 534  * so that flush_signal_handlers can later reset the handlers without
 535  * disturbing other processes.  (Other processes might share the signal
 536  * table via the CLONE_SIGHAND option to clone().)
 537  */
 538 static inline int de_thread(struct task_struct *tsk)
 539 {
 540         struct signal_struct *newsig, *oldsig = tsk->signal;
 541         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 542         spinlock_t *lock = &oldsighand->siglock;
 543         int count;
 544
 545         /*
 546          * If we don't share sighandlers, then we aren't sharing anything
 547          * and we can just re-use it all.
 548          */
 549         if (atomic_read(&oldsighand->count) <= 1)
 550                 return 0;
 551
 552         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 553         if (!newsighand)
 554                 return -ENOMEM;
 555
 556         spin_lock_init(&newsighand->siglock);
 557         atomic_set(&newsighand->count, 1);
 558         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
 559
 560         /*
 561          * See if we need to allocate a new signal structure
 562          */
 563         newsig = NULL;
 564         if (atomic_read(&oldsig->count) > 1) {
 565                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 566                 if (!newsig) {
 567                         kmem_cache_free(sighand_cachep, newsighand);
 568                         return -ENOMEM;
 569                 }
 570                 atomic_set(&newsig->count, 1);
 571                 newsig->group_exit = 0;
 572                 newsig->group_exit_code = 0;
 573                 newsig->group_exit_task = NULL;
 574                 newsig->group_stop_count = 0;
 575                 newsig->curr_target = NULL;
 576                 init_sigpending(&newsig->shared_pending);
 577         }
 578
 579         if (thread_group_empty(current))
 580                 goto no_thread_group;
 581
 582         /*
 583          * Kill all other threads in the thread group.
 584          * We must hold tasklist_lock to call zap_other_threads.
 585          */
 586         read_lock(&tasklist_lock);
 587         spin_lock_irq(lock);
 588         if (oldsig->group_exit) {
 589                 /*
 590                  * Another group action in progress, just
 591                  * return so that the signal is processed.
 592                  */
 593                 spin_unlock_irq(lock);
 594                 read_unlock(&tasklist_lock);
 595                 kmem_cache_free(sighand_cachep, newsighand);
 596                 if (newsig)
 597                         kmem_cache_free(signal_cachep, newsig);
 598                 return -EAGAIN;
 599         }
 600         oldsig->group_exit = 1;
 601         zap_other_threads(current);
 602         read_unlock(&tasklist_lock);
 603
 604         /*
 605          * Account for the thread group leader hanging around:
 606          */
 607         count = 2;
 608         if (current->pid == current->tgid)
 609                 count = 1;
 610         while (atomic_read(&oldsig->count) > count) {
 611                 oldsig->group_exit_task = current;
 612                 oldsig->notify_count = count;
 613                 __set_current_state(TASK_UNINTERRUPTIBLE);
 614                 spin_unlock_irq(lock);
 615                 schedule();
 616                 spin_lock_irq(lock);
 617         }
 618         spin_unlock_irq(lock);
 619
 620         /*
 621          * At this point all other threads have exited, all we have to
 622          * do is to wait for the thread group leader to become inactive,
 623          * and to assume its PID:
 624          */
 625         if (current->pid != current->tgid) {
 626                 struct task_struct *leader = current->group_leader, *parent;
 627                 struct dentry *proc_dentry1, *proc_dentry2;
 628                 unsigned long state, ptrace;
 629
 630                 /*
 631                  * Wait for the thread group leader to be a zombie.
 632                  * It should already be zombie at this point, most
 633                  * of the time.
 634                  */
 635                 while (leader->state != TASK_ZOMBIE)
 636                         yield();
 637
 638                 spin_lock(&leader->proc_lock);
 639                 spin_lock(&current->proc_lock);
 640                 proc_dentry1 = proc_pid_unhash(current);
 641                 proc_dentry2 = proc_pid_unhash(leader);
 642                 write_lock_irq(&tasklist_lock);
 643
 644                 if (leader->tgid != current->tgid)
 645                         BUG();
 646                 if (current->pid == current->tgid)
 647                         BUG();
 648                 /*
 649                  * An exec() starts a new thread group with the
 650                  * TGID of the previous thread group. Rehash the
 651                  * two threads with a switched PID, and release
 652                  * the former thread group leader:
 653                  */
 654                 ptrace = leader->ptrace;
 655                 parent = leader->parent;
 656
 657                 ptrace_unlink(current);
 658                 ptrace_unlink(leader);
 659                 remove_parent(current);
 660                 remove_parent(leader);
 661
 662                 switch_exec_pids(leader, current);
 663
 664                 current->parent = current->real_parent = leader->real_parent;
 665                 leader->parent = leader->real_parent = child_reaper;
 666                 current->group_leader = current;
 667                 leader->group_leader = leader;
 668
 669                 add_parent(current, current->parent);
 670                 add_parent(leader, leader->parent);
 671                 if (ptrace) {
 672                         current->ptrace = ptrace;
 673                         __ptrace_link(current, parent);
 674                 }
 675
 676                 list_del(&current->tasks);
 677                 list_add_tail(&current->tasks, &init_task.tasks);
 678                 current->exit_signal = SIGCHLD;
 679                 state = leader->state;
 680
 681                 write_unlock_irq(&tasklist_lock);
 682                 spin_unlock(&leader->proc_lock);
 683                 spin_unlock(&current->proc_lock);
 684                 proc_pid_flush(proc_dentry1);
 685                 proc_pid_flush(proc_dentry2);
 686
 687                 if (state != TASK_ZOMBIE)
 688                         BUG();
 689                 release_task(leader);
 690         }
 691
 692 no_thread_group:
 693
 694         write_lock_irq(&tasklist_lock);
 695         spin_lock(&oldsighand->siglock);
 696         spin_lock(&newsighand->siglock);
 697
 698         if (current == oldsig->curr_target)
 699                 oldsig->curr_target = next_thread(current);
 700         if (newsig)
 701                 current->signal = newsig;
 702         current->sighand = newsighand;
 703         init_sigpending(&current->pending);
 704         recalc_sigpending();
 705
 706         spin_unlock(&newsighand->siglock);
 707         spin_unlock(&oldsighand->siglock);
 708         write_unlock_irq(&tasklist_lock);
 709
 710         if (newsig && atomic_dec_and_test(&oldsig->count))
 711                 kmem_cache_free(signal_cachep, oldsig);
 712
 713         if (atomic_dec_and_test(&oldsighand->count))
 714                 kmem_cache_free(sighand_cachep, oldsighand);
 715
 716         if (!thread_group_empty(current))
 717                 BUG();
 718         if (current->tgid != current->pid)
 719                 BUG();
 720         return 0;
 721 }
 722
 723 /*
 724  * These functions flushes out all traces of the currently running executable
 725  * so that a new one can be started
 726  */
 727
 728 static inline void flush_old_files(struct files_struct * files)
 729 {
 730         long j = -1;
 731
 732         spin_lock(&files->file_lock);
 733         for (;;) {
 734                 unsigned long set, i;
 735
 736                 j++;
 737                 i = j * __NFDBITS;
 738                 if (i >= files->max_fds || i >= files->max_fdset)
 739                         break;
 740                 set = files->close_on_exec->fds_bits[j];
 741                 if (!set)
 742                         continue;
 743                 files->close_on_exec->fds_bits[j] = 0;
 744                 spin_unlock(&files->file_lock);
 745                 for ( ; set ; i++,set >>= 1) {
 746                         if (set & 1) {
 747                                 sys_close(i);
 748                         }
 749                 }
 750                 spin_lock(&files->file_lock);
 751
 752         }
 753         spin_unlock(&files->file_lock);
 754 }
 755
 756 int flush_old_exec(struct linux_binprm * bprm)
 757 {
 758         char * name;
 759         int i, ch, retval;
 760
 761         /*
 762          * Release all of the old mmap stuff
 763          */
 764         retval = exec_mmap(bprm->mm);
 765         if (retval)
 766                 goto out;
 767         /*
 768          * Make sure we have a private signal table and that
 769          * we are unassociated from the previous thread group.
 770          */
 771         retval = de_thread(current);
 772         if (retval)
 773                 goto out;
 774
 775         /* This is the point of no return */
 776
 777         current->sas_ss_sp = current->sas_ss_size = 0;
 778
 779         if (current->euid == current->uid && current->egid == current->gid)
 780                 current->mm->dumpable = 1;
 781         name = bprm->filename;
 782         for (i=0; (ch = *(name++)) != '\0';) {
 783                 if (ch == '/')
 784                         i = 0;
 785                 else
 786                         if (i < 15)
 787                                 current->comm[i++] = ch;
 788         }
 789         current->comm[i] = '\0';
 790
 791         flush_thread();
 792
 793         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 794             permission(bprm->file->f_dentry->d_inode,MAY_READ))
 795                 current->mm->dumpable = 0;
 796
 797         /* An exec changes our domain. We are no longer part of the thread
 798            group */
 799
 800         current->self_exec_id++;
 801
 802         flush_signal_handlers(current, 0);
 803         flush_old_files(current->files);
 804         exit_itimers(current);
 805
 806         return 0;
 807
 808 out:
 809         return retval;
 810 }
 811
 812 /*
 813  * We mustn't allow tracing of suid binaries, unless
 814  * the tracer has the capability to trace anything..
 815  */
 816 static inline int must_not_trace_exec(struct task_struct * p)
 817 {
 818         return (p->ptrace & PT_PTRACED) && !(p->ptrace & PT_PTRACE_CAP);
 819 }
 820
 821 /*
 822  * Fill the binprm structure from the inode.
 823  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 824  */
 825 int prepare_binprm(struct linux_binprm *bprm)
 826 {
 827         int mode;
 828         struct inode * inode = bprm->file->f_dentry->d_inode;
 829         int retval;
 830
 831         mode = inode->i_mode;
 832         /*
 833          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 834          * vfs_permission lets a non-executable through
 835          */
 836         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 837                 return -EACCES;
 838         if (bprm->file->f_op == NULL)
 839                 return -EACCES;
 840
 841         bprm->e_uid = current->euid;
 842         bprm->e_gid = current->egid;
 843
 844         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 845                 /* Set-uid? */
 846                 if (mode & S_ISUID)
 847                         bprm->e_uid = inode->i_uid;
 848
 849                 /* Set-gid? */
 850                 /*
 851                  * If setgid is set but no group execute bit then this
 852                  * is a candidate for mandatory locking, not a setgid
 853                  * executable.
 854                  */
 855                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
 856                         bprm->e_gid = inode->i_gid;
 857         }
 858
 859         /* fill in binprm security blob */
 860         retval = security_bprm_set(bprm);
 861         if (retval)
 862                 return retval;
 863
 864         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 865         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 866 }
 867
 868 /*
 869  * This function is used to produce the new IDs and capabilities
 870  * from the old ones and the file's capabilities.
 871  *
 872  * The formula used for evolving capabilities is:
 873  *
 874  *       pI' = pI
 875  * (***) pP' = (fP & X) | (fI & pI)
 876  *       pE' = pP' & fE          [NB. fE is 0 or ~0]
 877  *
 878  * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
 879  * ' indicates post-exec(), and X is the global 'cap_bset'.
 880  *
 881  */
 882
 883 void compute_creds(struct linux_binprm *bprm)
 884 {
 885         task_lock(current);
 886         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid) {
 887                 current->mm->dumpable = 0;
 888
 889                 if (must_not_trace_exec(current)
 890                     || atomic_read(&current->fs->count) > 1
 891                     || atomic_read(&current->files->count) > 1
 892                     || atomic_read(&current->sighand->count) > 1) {
 893                         if(!capable(CAP_SETUID)) {
 894                                 bprm->e_uid = current->uid;
 895                                 bprm->e_gid = current->gid;
 896                         }
 897                 }
 898         }
 899
 900         current->suid = current->euid = current->fsuid = bprm->e_uid;
 901         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 902
 903         task_unlock(current);
 904
 905         security_bprm_compute_creds(bprm);
 906 }
 907
 908 void remove_arg_zero(struct linux_binprm *bprm)
 909 {
 910         if (bprm->argc) {
 911                 unsigned long offset;
 912                 char * kaddr;
 913                 struct page *page;
 914
 915                 offset = bprm->p % PAGE_SIZE;
 916                 goto inside;
 917
 918                 while (bprm->p++, *(kaddr+offset++)) {
 919                         if (offset != PAGE_SIZE)
 920                                 continue;
 921                         offset = 0;
 922                         kunmap(page);
 923 inside:
 924                         page = bprm->page[bprm->p/PAGE_SIZE];
 925                         kaddr = kmap(page);
 926                 }
 927                 kunmap(page);
 928                 bprm->argc--;
 929         }
 930 }
 931
 932 /*
 933  * cycle the list of binary formats handler, until one recognizes the image
 934  */
 935 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 936 {
 937         int try,retval=0;
 938         struct linux_binfmt *fmt;
 939 #ifdef __alpha__
 940         /* handle /sbin/loader.. */
 941         {
 942             struct exec * eh = (struct exec *) bprm->buf;
 943
 944             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 945                 (eh->fh.f_flags & 0x3000) == 0x3000)
 946             {
 947                 struct file * file;
 948                 unsigned long loader;
 949
 950                 allow_write_access(bprm->file);
 951                 fput(bprm->file);
 952                 bprm->file = NULL;
 953
 954                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 955
 956                 file = open_exec("/sbin/loader");
 957                 retval = PTR_ERR(file);
 958                 if (IS_ERR(file))
 959                         return retval;
 960
 961                 /* Remember if the application is TASO.  */
 962                 bprm->sh_bang = eh->ah.entry < 0x100000000;
 963
 964                 bprm->file = file;
 965                 bprm->loader = loader;
 966                 retval = prepare_binprm(bprm);
 967                 if (retval<0)
 968                         return retval;
 969                 /* should call search_binary_handler recursively here,
 970                    but it does not matter */
 971             }
 972         }
 973 #endif
 974         retval = security_bprm_check(bprm);
 975         if (retval)
 976                 return retval;
 977
 978         /* kernel module loader fixup */
 979         /* so we don't try to load run modprobe in kernel space. */
 980         set_fs(USER_DS);
 981         for (try=0; try<2; try++) {
 982                 read_lock(&binfmt_lock);
 983                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 984                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
 985                         if (!fn)
 986                                 continue;
 987                         if (!try_module_get(fmt->module))
 988                                 continue;
 989                         read_unlock(&binfmt_lock);
 990                         retval = fn(bprm, regs);
 991                         if (retval >= 0) {
 992                                 put_binfmt(fmt);
 993                                 allow_write_access(bprm->file);
 994                                 if (bprm->file)
 995                                         fput(bprm->file);
 996                                 bprm->file = NULL;
 997                                 current->did_exec = 1;
 998                                 return retval;
 999                         }
1000                         read_lock(&binfmt_lock);
1001                         put_binfmt(fmt);
1002                         if (retval != -ENOEXEC)
1003                                 break;
1004                         if (!bprm->file) {
1005                                 read_unlock(&binfmt_lock);
1006                                 return retval;
1007                         }
1008                 }
1009                 read_unlock(&binfmt_lock);
1010                 if (retval != -ENOEXEC) {
1011                         break;
1012 #ifdef CONFIG_KMOD
1013                 }else{
1014 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1015                         if (printable(bprm->buf[0]) &&
1016                             printable(bprm->buf[1]) &&
1017                             printable(bprm->buf[2]) &&
1018                             printable(bprm->buf[3]))
1019                                 break; /* -ENOEXEC */
1020                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1021 #endif
1022                 }
1023         }
1024         return retval;
1025 }
1026
1027 /*
1028  * sys_execve() executes a new program.
1029  */
1030 int do_execve(char * filename,
1031         char __user *__user *argv,
1032         char __user *__user *envp,
1033         struct pt_regs * regs)
1034 {
1035         struct linux_binprm bprm;
1036         struct file *file;
1037         int retval;
1038         int i;
1039
1040         sched_balance_exec();
1041
1042         file = open_exec(filename);
1043
1044         retval = PTR_ERR(file);
1045         if (IS_ERR(file))
1046                 return retval;
1047
1048         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1049         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
1050
1051         bprm.file = file;
1052         bprm.filename = filename;
1053         bprm.sh_bang = 0;
1054         bprm.loader = 0;
1055         bprm.exec = 0;
1056         bprm.security = NULL;
1057         bprm.mm = mm_alloc();
1058         retval = -ENOMEM;
1059         if (!bprm.mm)
1060                 goto out_file;
1061
1062         retval = init_new_context(current, bprm.mm);
1063         if (retval < 0)
1064                 goto out_mm;
1065
1066         bprm.argc = count(argv, bprm.p / sizeof(void *));
1067         if ((retval = bprm.argc) < 0)
1068                 goto out_mm;
1069
1070         bprm.envc = count(envp, bprm.p / sizeof(void *));
1071         if ((retval = bprm.envc) < 0)
1072                 goto out_mm;
1073
1074         retval = security_bprm_alloc(&bprm);
1075         if (retval)
1076                 goto out;
1077
1078         retval = prepare_binprm(&bprm);
1079         if (retval < 0)
1080                 goto out;
1081
1082         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
1083         if (retval < 0)
1084                 goto out;
1085
1086         bprm.exec = bprm.p;
1087         retval = copy_strings(bprm.envc, envp, &bprm);
1088         if (retval < 0)
1089                 goto out;
1090
1091         retval = copy_strings(bprm.argc, argv, &bprm);
1092         if (retval < 0)
1093                 goto out;
1094
1095         retval = search_binary_handler(&bprm,regs);
1096         if (retval >= 0) {
1097                 free_arg_pages(&bprm);
1098
1099                 /* execve success */
1100                 security_bprm_free(&bprm);
1101                 return retval;
1102         }
1103
1104 out:
1105         /* Something went wrong, return the inode and free the argument pages*/
1106         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1107                 struct page * page = bprm.page[i];
1108                 if (page)
1109                         __free_page(page);
1110         }
1111
1112         if (bprm.security)
1113                 security_bprm_free(&bprm);
1114
1115 out_mm:
1116         mmdrop(bprm.mm);
1117
1118 out_file:
1119         if (bprm.file) {
1120                 allow_write_access(bprm.file);
1121                 fput(bprm.file);
1122         }
1123         return retval;
1124 }
1125
1126 int set_binfmt(struct linux_binfmt *new)
1127 {
1128         struct linux_binfmt *old = current->binfmt;
1129
1130         if (new) {
1131                 if (!try_module_get(new->module))
1132                         return -1;
1133         }
1134         current->binfmt = new;
1135         if (old)
1136                 module_put(old->module);
1137         return 0;
1138 }
1139
1140 #define CORENAME_MAX_SIZE 64
1141
1142 /* format_corename will inspect the pattern parameter, and output a
1143  * name into corename, which must have space for at least
1144  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1145  */
1146 void format_corename(char *corename, const char *pattern, long signr)
1147 {
1148         const char *pat_ptr = pattern;
1149         char *out_ptr = corename;
1150         char *const out_end = corename + CORENAME_MAX_SIZE;
1151         int rc;
1152         int pid_in_pattern = 0;
1153
1154         /* Repeat as long as we have more pattern to process and more output
1155            space */
1156         while (*pat_ptr) {
1157                 if (*pat_ptr != '%') {
1158                         if (out_ptr == out_end)
1159                                 goto out;
1160                         *out_ptr++ = *pat_ptr++;
1161                 } else {
1162                         switch (*++pat_ptr) {
1163                         case 0:
1164                                 goto out;
1165                         /* Double percent, output one percent */
1166                         case '%':
1167                                 if (out_ptr == out_end)
1168                                         goto out;
1169                                 *out_ptr++ = '%';
1170                                 break;
1171                         /* pid */
1172                         case 'p':
1173                                 pid_in_pattern = 1;
1174                                 rc = snprintf(out_ptr, out_end - out_ptr,
1175                                               "%d", current->tgid);
1176                                 if (rc > out_end - out_ptr)
1177                                         goto out;
1178                                 out_ptr += rc;
1179                                 break;
1180                         /* uid */
1181                         case 'u':
1182                                 rc = snprintf(out_ptr, out_end - out_ptr,
1183                                               "%d", current->uid);
1184                                 if (rc > out_end - out_ptr)
1185                                         goto out;
1186                                 out_ptr += rc;
1187                                 break;
1188                         /* gid */
1189                         case 'g':
1190                                 rc = snprintf(out_ptr, out_end - out_ptr,
1191                                               "%d", current->gid);
1192                                 if (rc > out_end - out_ptr)
1193                                         goto out;
1194                                 out_ptr += rc;
1195                                 break;
1196                         /* signal that caused the coredump */
1197                         case 's':
1198                                 rc = snprintf(out_ptr, out_end - out_ptr,
1199                                               "%ld", signr);
1200                                 if (rc > out_end - out_ptr)
1201                                         goto out;
1202                                 out_ptr += rc;
1203                                 break;
1204                         /* UNIX time of coredump */
1205                         case 't': {
1206                                 struct timeval tv;
1207                                 do_gettimeofday(&tv);
1208                                 rc = snprintf(out_ptr, out_end - out_ptr,
1209                                               "%lu", tv.tv_sec);
1210                                 if (rc > out_end - out_ptr)
1211                                         goto out;
1212                                 out_ptr += rc;
1213                                 break;
1214                         }
1215                         /* hostname */
1216                         case 'h':
1217                                 down_read(&uts_sem);
1218                                 rc = snprintf(out_ptr, out_end - out_ptr,
1219                                               "%s", system_utsname.nodename);
1220                                 up_read(&uts_sem);
1221                                 if (rc > out_end - out_ptr)
1222                                         goto out;
1223                                 out_ptr += rc;
1224                                 break;
1225                         /* executable */
1226                         case 'e':
1227                                 rc = snprintf(out_ptr, out_end - out_ptr,
1228                                               "%s", current->comm);
1229                                 if (rc > out_end - out_ptr)
1230                                         goto out;
1231                                 out_ptr += rc;
1232                                 break;
1233                         default:
1234                                 break;
1235                         }
1236                         ++pat_ptr;
1237                 }
1238         }
1239         /* Backward compatibility with core_uses_pid:
1240          *
1241          * If core_pattern does not include a %p (as is the default)
1242          * and core_uses_pid is set, then .%pid will be appended to
1243          * the filename */
1244         if (!pid_in_pattern
1245             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1246                 rc = snprintf(out_ptr, out_end - out_ptr,
1247                               ".%d", current->tgid);
1248                 if (rc > out_end - out_ptr)
1249                         goto out;
1250                 out_ptr += rc;
1251         }
1252       out:
1253         *out_ptr = 0;
1254 }
1255
1256 static void zap_threads (struct mm_struct *mm)
1257 {
1258         struct task_struct *g, *p;
1259
1260         read_lock(&tasklist_lock);
1261         do_each_thread(g,p)
1262                 if (mm == p->mm && p != current) {
1263                         force_sig_specific(SIGKILL, p);
1264                         mm->core_waiters++;
1265                 }
1266         while_each_thread(g,p);
1267
1268         read_unlock(&tasklist_lock);
1269 }
1270
1271 static void coredump_wait(struct mm_struct *mm)
1272 {
1273         DECLARE_COMPLETION(startup_done);
1274
1275         mm->core_waiters++; /* let other threads block */
1276         mm->core_startup_done = &startup_done;
1277
1278         /* give other threads a chance to run: */
1279         yield();
1280
1281         zap_threads(mm);
1282         if (--mm->core_waiters) {
1283                 up_write(&mm->mmap_sem);
1284                 wait_for_completion(&startup_done);
1285         } else
1286                 up_write(&mm->mmap_sem);
1287         BUG_ON(mm->core_waiters);
1288 }
1289
1290 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1291 {
1292         char corename[CORENAME_MAX_SIZE + 1];
1293         struct mm_struct *mm = current->mm;
1294         struct linux_binfmt * binfmt;
1295         struct inode * inode;
1296         struct file * file;
1297         int retval = 0;
1298
1299         lock_kernel();
1300         binfmt = current->binfmt;
1301         if (!binfmt || !binfmt->core_dump)
1302                 goto fail;
1303         down_write(&mm->mmap_sem);
1304         if (!mm->dumpable) {
1305                 up_write(&mm->mmap_sem);
1306                 goto fail;
1307         }
1308         mm->dumpable = 0;
1309         init_completion(&mm->core_done);
1310         current->signal->group_exit = 1;
1311         current->signal->group_exit_code = exit_code;
1312         coredump_wait(mm);
1313
1314         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1315                 goto fail_unlock;
1316
1317         format_corename(corename, core_pattern, signr);
1318         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW, 0600);
1319         if (IS_ERR(file))
1320                 goto fail_unlock;
1321         inode = file->f_dentry->d_inode;
1322         if (inode->i_nlink > 1)
1323                 goto close_fail;        /* multiple links - don't dump */
1324         if (d_unhashed(file->f_dentry))
1325                 goto close_fail;
1326
1327         if (!S_ISREG(inode->i_mode))
1328                 goto close_fail;
1329         if (!file->f_op)
1330                 goto close_fail;
1331         if (!file->f_op->write)
1332                 goto close_fail;
1333         if (do_truncate(file->f_dentry, 0) != 0)
1334                 goto close_fail;
1335
1336         retval = binfmt->core_dump(signr, regs, file);
1337
1338         current->signal->group_exit_code |= 0x80;
1339 close_fail:
1340         filp_close(file, NULL);
1341 fail_unlock:
1342         complete_all(&mm->core_done);
1343 fail:
1344         unlock_kernel();
1345         return retval;
1346 }