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         nd.intent.open.flags = O_RDONLY;
 121         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 122         if (error)
 123                 goto out;
 124
 125         error = -EINVAL;
 126         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 127                 goto exit;
 128
 129         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 130         if (error)
 131                 goto exit;
 132
 133         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 134         error = PTR_ERR(file);
 135         if (IS_ERR(file))
 136                 goto out;
 137
 138         error = -ENOEXEC;
 139         if(file->f_op) {
 140                 struct linux_binfmt * fmt;
 141
 142                 read_lock(&binfmt_lock);
 143                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 144                         if (!fmt->load_shlib)
 145                                 continue;
 146                         if (!try_module_get(fmt->module))
 147                                 continue;
 148                         read_unlock(&binfmt_lock);
 149                         error = fmt->load_shlib(file);
 150                         read_lock(&binfmt_lock);
 151                         put_binfmt(fmt);
 152                         if (error != -ENOEXEC)
 153                                 break;
 154                 }
 155                 read_unlock(&binfmt_lock);
 156         }
 157         fput(file);
 158 out:
 159         return error;
 160 exit:
 161         path_release(&nd);
 162         goto out;
 163 }
 164
 165 /*
 166  * count() counts the number of strings in array ARGV.
 167  */
 168 static int count(char __user * __user * argv, int max)
 169 {
 170         int i = 0;
 171
 172         if (argv != NULL) {
 173                 for (;;) {
 174                         char __user * p;
 175
 176                         if (get_user(p, argv))
 177                                 return -EFAULT;
 178                         if (!p)
 179                                 break;
 180                         argv++;
 181                         if(++i > max)
 182                                 return -E2BIG;
 183                 }
 184         }
 185         return i;
 186 }
 187
 188 /*
 189  * 'copy_strings()' copies argument/environment strings from user
 190  * memory to free pages in kernel mem. These are in a format ready
 191  * to be put directly into the top of new user memory.
 192  */
 193 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 194 {
 195         struct page *kmapped_page = NULL;
 196         char *kaddr = NULL;
 197         int ret;
 198
 199         while (argc-- > 0) {
 200                 char __user *str;
 201                 int len;
 202                 unsigned long pos;
 203
 204                 if (get_user(str, argv+argc) ||
 205                                 !(len = strnlen_user(str, bprm->p))) {
 206                         ret = -EFAULT;
 207                         goto out;
 208                 }
 209
 210                 if (bprm->p < len)  {
 211                         ret = -E2BIG;
 212                         goto out;
 213                 }
 214
 215                 bprm->p -= len;
 216                 /* XXX: add architecture specific overflow check here. */
 217                 pos = bprm->p;
 218
 219                 while (len > 0) {
 220                         int i, new, err;
 221                         int offset, bytes_to_copy;
 222                         struct page *page;
 223
 224                         offset = pos % PAGE_SIZE;
 225                         i = pos/PAGE_SIZE;
 226                         page = bprm->page[i];
 227                         new = 0;
 228                         if (!page) {
 229                                 page = alloc_page(GFP_HIGHUSER);
 230                                 bprm->page[i] = page;
 231                                 if (!page) {
 232                                         ret = -ENOMEM;
 233                                         goto out;
 234                                 }
 235                                 new = 1;
 236                         }
 237
 238                         if (page != kmapped_page) {
 239                                 if (kmapped_page)
 240                                         kunmap(kmapped_page);
 241                                 kmapped_page = page;
 242                                 kaddr = kmap(kmapped_page);
 243                         }
 244                         if (new && offset)
 245                                 memset(kaddr, 0, offset);
 246                         bytes_to_copy = PAGE_SIZE - offset;
 247                         if (bytes_to_copy > len) {
 248                                 bytes_to_copy = len;
 249                                 if (new)
 250                                         memset(kaddr+offset+len, 0,
 251                                                 PAGE_SIZE-offset-len);
 252                         }
 253                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 254                         if (err) {
 255                                 ret = -EFAULT;
 256                                 goto out;
 257                         }
 258
 259                         pos += bytes_to_copy;
 260                         str += bytes_to_copy;
 261                         len -= bytes_to_copy;
 262                 }
 263         }
 264         ret = 0;
 265 out:
 266         if (kmapped_page)
 267                 kunmap(kmapped_page);
 268         return ret;
 269 }
 270
 271 /*
 272  * Like copy_strings, but get argv and its values from kernel memory.
 273  */
 274 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 275 {
 276         int r;
 277         mm_segment_t oldfs = get_fs();
 278         set_fs(KERNEL_DS);
 279         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 280         set_fs(oldfs);
 281         return r;
 282 }
 283
 284 #ifdef CONFIG_MMU
 285 /*
 286  * This routine is used to map in a page into an address space: needed by
 287  * execve() for the initial stack and environment pages.
 288  *
 289  * tsk->mmap_sem is held for writing.
 290  */
 291 void put_dirty_page(struct task_struct *tsk, struct page *page,
 292                         unsigned long address, pgprot_t prot)
 293 {
 294         pgd_t * pgd;
 295         pmd_t * pmd;
 296         pte_t * pte;
 297         struct pte_chain *pte_chain;
 298
 299         if (page_count(page) != 1)
 300                 printk(KERN_ERR "mem_map disagrees with %p at %08lx\n",
 301                                 page, address);
 302
 303         pgd = pgd_offset(tsk->mm, address);
 304         pte_chain = pte_chain_alloc(GFP_KERNEL);
 305         if (!pte_chain)
 306                 goto out_sig;
 307         spin_lock(&tsk->mm->page_table_lock);
 308         pmd = pmd_alloc(tsk->mm, pgd, address);
 309         if (!pmd)
 310                 goto out;
 311         pte = pte_alloc_map(tsk->mm, pmd, address);
 312         if (!pte)
 313                 goto out;
 314         if (!pte_none(*pte)) {
 315                 pte_unmap(pte);
 316                 goto out;
 317         }
 318         lru_cache_add_active(page);
 319         flush_dcache_page(page);
 320         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(page, prot))));
 321         pte_chain = page_add_rmap(page, pte, pte_chain);
 322         pte_unmap(pte);
 323         tsk->mm->rss++;
 324         spin_unlock(&tsk->mm->page_table_lock);
 325
 326         /* no need for flush_tlb */
 327         pte_chain_free(pte_chain);
 328         return;
 329 out:
 330         spin_unlock(&tsk->mm->page_table_lock);
 331 out_sig:
 332         __free_page(page);
 333         force_sig(SIGKILL, tsk);
 334         pte_chain_free(pte_chain);
 335         return;
 336 }
 337
 338 int setup_arg_pages(struct linux_binprm *bprm)
 339 {
 340         unsigned long stack_base;
 341         struct vm_area_struct *mpnt;
 342         struct mm_struct *mm = current->mm;
 343         int i;
 344
 345 #ifdef CONFIG_STACK_GROWSUP
 346         /* Move the argument and environment strings to the bottom of the
 347          * stack space.
 348          */
 349         int offset, j;
 350         char *to, *from;
 351
 352         /* Start by shifting all the pages down */
 353         i = 0;
 354         for (j = 0; j < MAX_ARG_PAGES; j++) {
 355                 struct page *page = bprm->page[j];
 356                 if (!page)
 357                         continue;
 358                 bprm->page[i++] = page;
 359         }
 360
 361         /* Now move them within their pages */
 362         offset = bprm->p % PAGE_SIZE;
 363         to = kmap(bprm->page[0]);
 364         for (j = 1; j < i; j++) {
 365                 memmove(to, to + offset, PAGE_SIZE - offset);
 366                 from = kmap(bprm->page[j]);
 367                 memcpy(to + PAGE_SIZE - offset, from, offset);
 368                 kunmap(bprm->page[j - 1]);
 369                 to = from;
 370         }
 371         memmove(to, to + offset, PAGE_SIZE - offset);
 372         kunmap(bprm->page[j - 1]);
 373
 374         /* Adjust bprm->p to point to the end of the strings. */
 375         bprm->p = PAGE_SIZE * i - offset;
 376         stack_base = STACK_TOP - current->rlim[RLIMIT_STACK].rlim_max;
 377         mm->arg_start = stack_base;
 378
 379         /* zero pages that were copied above */
 380         while (i < MAX_ARG_PAGES)
 381                 bprm->page[i++] = NULL;
 382 #else
 383         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 384         mm->arg_start = bprm->p + stack_base;
 385 #endif
 386
 387         bprm->p += stack_base;
 388         if (bprm->loader)
 389                 bprm->loader += stack_base;
 390         bprm->exec += stack_base;
 391
 392         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 393         if (!mpnt)
 394                 return -ENOMEM;
 395
 396         if (security_vm_enough_memory((STACK_TOP - (PAGE_MASK & (unsigned long) bprm->p))>>PAGE_SHIFT)) {
 397                 kmem_cache_free(vm_area_cachep, mpnt);
 398                 return -ENOMEM;
 399         }
 400
 401         down_write(&mm->mmap_sem);
 402         {
 403                 mpnt->vm_mm = mm;
 404 #ifdef CONFIG_STACK_GROWSUP
 405                 mpnt->vm_start = stack_base;
 406                 mpnt->vm_end = PAGE_MASK &
 407                         (PAGE_SIZE - 1 + (unsigned long) bprm->p);
 408 #else
 409                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 410                 mpnt->vm_end = STACK_TOP;
 411 #endif
 412                 mpnt->vm_page_prot = protection_map[VM_STACK_FLAGS & 0x7];
 413                 mpnt->vm_flags = VM_STACK_FLAGS;
 414                 mpnt->vm_ops = NULL;
 415                 mpnt->vm_pgoff = 0;
 416                 mpnt->vm_file = NULL;
 417                 INIT_LIST_HEAD(&mpnt->shared);
 418                 mpnt->vm_private_data = (void *) 0;
 419                 insert_vm_struct(mm, mpnt);
 420                 mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 421         }
 422
 423         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 424                 struct page *page = bprm->page[i];
 425                 if (page) {
 426                         bprm->page[i] = NULL;
 427                         put_dirty_page(current, page, stack_base,
 428                                         mpnt->vm_page_prot);
 429                 }
 430                 stack_base += PAGE_SIZE;
 431         }
 432         up_write(&mm->mmap_sem);
 433
 434         return 0;
 435 }
 436
 437 #define free_arg_pages(bprm) do { } while (0)
 438
 439 #else
 440
 441 static inline void free_arg_pages(struct linux_binprm *bprm)
 442 {
 443         int i;
 444
 445         for (i = 0; i < MAX_ARG_PAGES; i++) {
 446                 if (bprm->page[i])
 447                         __free_page(bprm->page[i]);
 448                 bprm->page[i] = NULL;
 449         }
 450 }
 451
 452 #endif /* CONFIG_MMU */
 453
 454 struct file *open_exec(const char *name)
 455 {
 456         struct nameidata nd;
 457         int err = path_lookup(name, LOOKUP_FOLLOW, &nd);
 458         struct file *file = ERR_PTR(err);
 459
 460         if (!err) {
 461                 struct inode *inode = nd.dentry->d_inode;
 462                 file = ERR_PTR(-EACCES);
 463                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 464                     S_ISREG(inode->i_mode)) {
 465                         int err = permission(inode, MAY_EXEC, &nd);
 466                         if (!err && !(inode->i_mode & 0111))
 467                                 err = -EACCES;
 468                         file = ERR_PTR(err);
 469                         if (!err) {
 470                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 471                                 if (!IS_ERR(file)) {
 472                                         err = deny_write_access(file);
 473                                         if (err) {
 474                                                 fput(file);
 475                                                 file = ERR_PTR(err);
 476                                         }
 477                                 }
 478 out:
 479                                 return file;
 480                         }
 481                 }
 482                 path_release(&nd);
 483         }
 484         goto out;
 485 }
 486
 487 int kernel_read(struct file *file, unsigned long offset,
 488         char *addr, unsigned long count)
 489 {
 490         mm_segment_t old_fs;
 491         loff_t pos = offset;
 492         int result;
 493
 494         old_fs = get_fs();
 495         set_fs(get_ds());
 496         /* The cast to a user pointer is valid due to the set_fs() */
 497         result = vfs_read(file, (void __user *)addr, count, &pos);
 498         set_fs(old_fs);
 499         return result;
 500 }
 501
 502 static int exec_mmap(struct mm_struct *mm)
 503 {
 504         struct task_struct *tsk;
 505         struct mm_struct * old_mm, *active_mm;
 506
 507         /* Add it to the list of mm's */
 508         spin_lock(&mmlist_lock);
 509         list_add(&mm->mmlist, &init_mm.mmlist);
 510         mmlist_nr++;
 511         spin_unlock(&mmlist_lock);
 512
 513         /* Notify parent that we're no longer interested in the old VM */
 514         tsk = current;
 515         old_mm = current->mm;
 516         mm_release(tsk, old_mm);
 517
 518         task_lock(tsk);
 519         active_mm = tsk->active_mm;
 520         tsk->mm = mm;
 521         tsk->active_mm = mm;
 522         activate_mm(active_mm, mm);
 523         task_unlock(tsk);
 524         if (old_mm) {
 525                 if (active_mm != old_mm) BUG();
 526                 mmput(old_mm);
 527                 return 0;
 528         }
 529         mmdrop(active_mm);
 530         return 0;
 531 }
 532
 533 /*
 534  * This function makes sure the current process has its own signal table,
 535  * so that flush_signal_handlers can later reset the handlers without
 536  * disturbing other processes.  (Other processes might share the signal
 537  * table via the CLONE_SIGHAND option to clone().)
 538  */
 539 static inline int de_thread(struct task_struct *tsk)
 540 {
 541         struct signal_struct *newsig, *oldsig = tsk->signal;
 542         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 543         spinlock_t *lock = &oldsighand->siglock;
 544         int count;
 545
 546         /*
 547          * If we don't share sighandlers, then we aren't sharing anything
 548          * and we can just re-use it all.
 549          */
 550         if (atomic_read(&oldsighand->count) <= 1)
 551                 return 0;
 552
 553         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 554         if (!newsighand)
 555                 return -ENOMEM;
 556
 557         spin_lock_init(&newsighand->siglock);
 558         atomic_set(&newsighand->count, 1);
 559         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
 560
 561         /*
 562          * See if we need to allocate a new signal structure
 563          */
 564         newsig = NULL;
 565         if (atomic_read(&oldsig->count) > 1) {
 566                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 567                 if (!newsig) {
 568                         kmem_cache_free(sighand_cachep, newsighand);
 569                         return -ENOMEM;
 570                 }
 571                 atomic_set(&newsig->count, 1);
 572                 newsig->group_exit = 0;
 573                 newsig->group_exit_code = 0;
 574                 newsig->group_exit_task = NULL;
 575                 newsig->group_stop_count = 0;
 576                 newsig->curr_target = NULL;
 577                 init_sigpending(&newsig->shared_pending);
 578         }
 579
 580         if (thread_group_empty(current))
 581                 goto no_thread_group;
 582
 583         /*
 584          * Kill all other threads in the thread group.
 585          * We must hold tasklist_lock to call zap_other_threads.
 586          */
 587         read_lock(&tasklist_lock);
 588         spin_lock_irq(lock);
 589         if (oldsig->group_exit) {
 590                 /*
 591                  * Another group action in progress, just
 592                  * return so that the signal is processed.
 593                  */
 594                 spin_unlock_irq(lock);
 595                 read_unlock(&tasklist_lock);
 596                 kmem_cache_free(sighand_cachep, newsighand);
 597                 if (newsig)
 598                         kmem_cache_free(signal_cachep, newsig);
 599                 return -EAGAIN;
 600         }
 601         oldsig->group_exit = 1;
 602         zap_other_threads(current);
 603         read_unlock(&tasklist_lock);
 604
 605         /*
 606          * Account for the thread group leader hanging around:
 607          */
 608         count = 2;
 609         if (current->pid == current->tgid)
 610                 count = 1;
 611         while (atomic_read(&oldsig->count) > count) {
 612                 oldsig->group_exit_task = current;
 613                 oldsig->notify_count = count;
 614                 __set_current_state(TASK_UNINTERRUPTIBLE);
 615                 spin_unlock_irq(lock);
 616                 schedule();
 617                 spin_lock_irq(lock);
 618         }
 619         spin_unlock_irq(lock);
 620
 621         /*
 622          * At this point all other threads have exited, all we have to
 623          * do is to wait for the thread group leader to become inactive,
 624          * and to assume its PID:
 625          */
 626         if (current->pid != current->tgid) {
 627                 struct task_struct *leader = current->group_leader, *parent;
 628                 struct dentry *proc_dentry1, *proc_dentry2;
 629                 unsigned long state, ptrace;
 630
 631                 /*
 632                  * Wait for the thread group leader to be a zombie.
 633                  * It should already be zombie at this point, most
 634                  * of the time.
 635                  */
 636                 while (leader->state != TASK_ZOMBIE)
 637                         yield();
 638
 639                 spin_lock(&leader->proc_lock);
 640                 spin_lock(&current->proc_lock);
 641                 proc_dentry1 = proc_pid_unhash(current);
 642                 proc_dentry2 = proc_pid_unhash(leader);
 643                 write_lock_irq(&tasklist_lock);
 644
 645                 if (leader->tgid != current->tgid)
 646                         BUG();
 647                 if (current->pid == current->tgid)
 648                         BUG();
 649                 /*
 650                  * An exec() starts a new thread group with the
 651                  * TGID of the previous thread group. Rehash the
 652                  * two threads with a switched PID, and release
 653                  * the former thread group leader:
 654                  */
 655                 ptrace = leader->ptrace;
 656                 parent = leader->parent;
 657
 658                 ptrace_unlink(current);
 659                 ptrace_unlink(leader);
 660                 remove_parent(current);
 661                 remove_parent(leader);
 662
 663                 switch_exec_pids(leader, current);
 664
 665                 current->parent = current->real_parent = leader->real_parent;
 666                 leader->parent = leader->real_parent = child_reaper;
 667                 current->group_leader = current;
 668                 leader->group_leader = leader;
 669
 670                 add_parent(current, current->parent);
 671                 add_parent(leader, leader->parent);
 672                 if (ptrace) {
 673                         current->ptrace = ptrace;
 674                         __ptrace_link(current, parent);
 675                 }
 676
 677                 list_del(&current->tasks);
 678                 list_add_tail(&current->tasks, &init_task.tasks);
 679                 current->exit_signal = SIGCHLD;
 680                 state = leader->state;
 681
 682                 write_unlock_irq(&tasklist_lock);
 683                 spin_unlock(&leader->proc_lock);
 684                 spin_unlock(&current->proc_lock);
 685                 proc_pid_flush(proc_dentry1);
 686                 proc_pid_flush(proc_dentry2);
 687
 688                 if (state != TASK_ZOMBIE)
 689                         BUG();
 690                 release_task(leader);
 691         }
 692
 693 no_thread_group:
 694
 695         write_lock_irq(&tasklist_lock);
 696         spin_lock(&oldsighand->siglock);
 697         spin_lock(&newsighand->siglock);
 698
 699         if (current == oldsig->curr_target)
 700                 oldsig->curr_target = next_thread(current);
 701         if (newsig)
 702                 current->signal = newsig;
 703         current->sighand = newsighand;
 704         init_sigpending(&current->pending);
 705         recalc_sigpending();
 706
 707         spin_unlock(&newsighand->siglock);
 708         spin_unlock(&oldsighand->siglock);
 709         write_unlock_irq(&tasklist_lock);
 710
 711         if (newsig && atomic_dec_and_test(&oldsig->count))
 712                 kmem_cache_free(signal_cachep, oldsig);
 713
 714         if (atomic_dec_and_test(&oldsighand->count))
 715                 kmem_cache_free(sighand_cachep, oldsighand);
 716
 717         if (!thread_group_empty(current))
 718                 BUG();
 719         if (current->tgid != current->pid)
 720                 BUG();
 721         return 0;
 722 }
 723
 724 /*
 725  * These functions flushes out all traces of the currently running executable
 726  * so that a new one can be started
 727  */
 728
 729 static inline void flush_old_files(struct files_struct * files)
 730 {
 731         long j = -1;
 732
 733         spin_lock(&files->file_lock);
 734         for (;;) {
 735                 unsigned long set, i;
 736
 737                 j++;
 738                 i = j * __NFDBITS;
 739                 if (i >= files->max_fds || i >= files->max_fdset)
 740                         break;
 741                 set = files->close_on_exec->fds_bits[j];
 742                 if (!set)
 743                         continue;
 744                 files->close_on_exec->fds_bits[j] = 0;
 745                 spin_unlock(&files->file_lock);
 746                 for ( ; set ; i++,set >>= 1) {
 747                         if (set & 1) {
 748                                 sys_close(i);
 749                         }
 750                 }
 751                 spin_lock(&files->file_lock);
 752
 753         }
 754         spin_unlock(&files->file_lock);
 755 }
 756
 757 int flush_old_exec(struct linux_binprm * bprm)
 758 {
 759         char * name;
 760         int i, ch, retval;
 761
 762         /*
 763          * Make sure we have a private signal table and that
 764          * we are unassociated from the previous thread group.
 765          */
 766         retval = de_thread(current);
 767         if (retval)
 768                 goto out;
 769
 770         /*
 771          * Release all of the old mmap stuff
 772          */
 773         retval = exec_mmap(bprm->mm);
 774         if (retval)
 775                 goto out;
 776
 777         bprm->mm = NULL;                /* We're using it now */
 778
 779         /* This is the point of no return */
 780
 781         current->sas_ss_sp = current->sas_ss_size = 0;
 782
 783         if (current->euid == current->uid && current->egid == current->gid)
 784                 current->mm->dumpable = 1;
 785         name = bprm->filename;
 786         for (i=0; (ch = *(name++)) != '\0';) {
 787                 if (ch == '/')
 788                         i = 0;
 789                 else
 790                         if (i < 15)
 791                                 current->comm[i++] = ch;
 792         }
 793         current->comm[i] = '\0';
 794
 795         flush_thread();
 796
 797         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 798             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL))
 799                 current->mm->dumpable = 0;
 800
 801         /* An exec changes our domain. We are no longer part of the thread
 802            group */
 803
 804         current->self_exec_id++;
 805
 806         flush_signal_handlers(current, 0);
 807         flush_old_files(current->files);
 808         exit_itimers(current);
 809
 810         return 0;
 811
 812 out:
 813         return retval;
 814 }
 815
 816 /*
 817  * We mustn't allow tracing of suid binaries, unless
 818  * the tracer has the capability to trace anything..
 819  */
 820 static inline int must_not_trace_exec(struct task_struct * p)
 821 {
 822         return (p->ptrace & PT_PTRACED) && !(p->ptrace & PT_PTRACE_CAP);
 823 }
 824
 825 /*
 826  * Fill the binprm structure from the inode.
 827  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 828  */
 829 int prepare_binprm(struct linux_binprm *bprm)
 830 {
 831         int mode;
 832         struct inode * inode = bprm->file->f_dentry->d_inode;
 833         int retval;
 834
 835         mode = inode->i_mode;
 836         /*
 837          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 838          * vfs_permission lets a non-executable through
 839          */
 840         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 841                 return -EACCES;
 842         if (bprm->file->f_op == NULL)
 843                 return -EACCES;
 844
 845         bprm->e_uid = current->euid;
 846         bprm->e_gid = current->egid;
 847
 848         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 849                 /* Set-uid? */
 850                 if (mode & S_ISUID)
 851                         bprm->e_uid = inode->i_uid;
 852
 853                 /* Set-gid? */
 854                 /*
 855                  * If setgid is set but no group execute bit then this
 856                  * is a candidate for mandatory locking, not a setgid
 857                  * executable.
 858                  */
 859                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
 860                         bprm->e_gid = inode->i_gid;
 861         }
 862
 863         /* fill in binprm security blob */
 864         retval = security_bprm_set(bprm);
 865         if (retval)
 866                 return retval;
 867
 868         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 869         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 870 }
 871
 872 /*
 873  * This function is used to produce the new IDs and capabilities
 874  * from the old ones and the file's capabilities.
 875  *
 876  * The formula used for evolving capabilities is:
 877  *
 878  *       pI' = pI
 879  * (***) pP' = (fP & X) | (fI & pI)
 880  *       pE' = pP' & fE          [NB. fE is 0 or ~0]
 881  *
 882  * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
 883  * ' indicates post-exec(), and X is the global 'cap_bset'.
 884  *
 885  */
 886
 887 void compute_creds(struct linux_binprm *bprm)
 888 {
 889         task_lock(current);
 890         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid) {
 891                 current->mm->dumpable = 0;
 892
 893                 if (must_not_trace_exec(current)
 894                     || atomic_read(&current->fs->count) > 1
 895                     || atomic_read(&current->files->count) > 1
 896                     || atomic_read(&current->sighand->count) > 1) {
 897                         if(!capable(CAP_SETUID)) {
 898                                 bprm->e_uid = current->uid;
 899                                 bprm->e_gid = current->gid;
 900                         }
 901                 }
 902         }
 903
 904         current->suid = current->euid = current->fsuid = bprm->e_uid;
 905         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 906
 907         task_unlock(current);
 908
 909         security_bprm_compute_creds(bprm);
 910 }
 911
 912 void remove_arg_zero(struct linux_binprm *bprm)
 913 {
 914         if (bprm->argc) {
 915                 unsigned long offset;
 916                 char * kaddr;
 917                 struct page *page;
 918
 919                 offset = bprm->p % PAGE_SIZE;
 920                 goto inside;
 921
 922                 while (bprm->p++, *(kaddr+offset++)) {
 923                         if (offset != PAGE_SIZE)
 924                                 continue;
 925                         offset = 0;
 926                         kunmap_atomic(kaddr, KM_USER0);
 927 inside:
 928                         page = bprm->page[bprm->p/PAGE_SIZE];
 929                         kaddr = kmap_atomic(page, KM_USER0);
 930                 }
 931                 kunmap_atomic(kaddr, KM_USER0);
 932                 bprm->argc--;
 933         }
 934 }
 935
 936 /*
 937  * cycle the list of binary formats handler, until one recognizes the image
 938  */
 939 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 940 {
 941         int try,retval=0;
 942         struct linux_binfmt *fmt;
 943 #ifdef __alpha__
 944         /* handle /sbin/loader.. */
 945         {
 946             struct exec * eh = (struct exec *) bprm->buf;
 947
 948             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 949                 (eh->fh.f_flags & 0x3000) == 0x3000)
 950             {
 951                 struct file * file;
 952                 unsigned long loader;
 953
 954                 allow_write_access(bprm->file);
 955                 fput(bprm->file);
 956                 bprm->file = NULL;
 957
 958                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 959
 960                 file = open_exec("/sbin/loader");
 961                 retval = PTR_ERR(file);
 962                 if (IS_ERR(file))
 963                         return retval;
 964
 965                 /* Remember if the application is TASO.  */
 966                 bprm->sh_bang = eh->ah.entry < 0x100000000;
 967
 968                 bprm->file = file;
 969                 bprm->loader = loader;
 970                 retval = prepare_binprm(bprm);
 971                 if (retval<0)
 972                         return retval;
 973                 /* should call search_binary_handler recursively here,
 974                    but it does not matter */
 975             }
 976         }
 977 #endif
 978         retval = security_bprm_check(bprm);
 979         if (retval)
 980                 return retval;
 981
 982         /* kernel module loader fixup */
 983         /* so we don't try to load run modprobe in kernel space. */
 984         set_fs(USER_DS);
 985         for (try=0; try<2; try++) {
 986                 read_lock(&binfmt_lock);
 987                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 988                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
 989                         if (!fn)
 990                                 continue;
 991                         if (!try_module_get(fmt->module))
 992                                 continue;
 993                         read_unlock(&binfmt_lock);
 994                         retval = fn(bprm, regs);
 995                         if (retval >= 0) {
 996                                 put_binfmt(fmt);
 997                                 allow_write_access(bprm->file);
 998                                 if (bprm->file)
 999                                         fput(bprm->file);
1000                                 bprm->file = NULL;
1001                                 current->did_exec = 1;
1002                                 return retval;
1003                         }
1004                         read_lock(&binfmt_lock);
1005                         put_binfmt(fmt);
1006                         if (retval != -ENOEXEC || bprm->mm == NULL)
1007                                 break;
1008                         if (!bprm->file) {
1009                                 read_unlock(&binfmt_lock);
1010                                 return retval;
1011                         }
1012                 }
1013                 read_unlock(&binfmt_lock);
1014                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1015                         break;
1016 #ifdef CONFIG_KMOD
1017                 }else{
1018 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1019                         if (printable(bprm->buf[0]) &&
1020                             printable(bprm->buf[1]) &&
1021                             printable(bprm->buf[2]) &&
1022                             printable(bprm->buf[3]))
1023                                 break; /* -ENOEXEC */
1024                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1025 #endif
1026                 }
1027         }
1028         return retval;
1029 }
1030
1031 /*
1032  * sys_execve() executes a new program.
1033  */
1034 int do_execve(char * filename,
1035         char __user *__user *argv,
1036         char __user *__user *envp,
1037         struct pt_regs * regs)
1038 {
1039         struct linux_binprm bprm;
1040         struct file *file;
1041         int retval;
1042
1043         sched_balance_exec();
1044
1045         file = open_exec(filename);
1046
1047         retval = PTR_ERR(file);
1048         if (IS_ERR(file))
1049                 return retval;
1050
1051         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1052         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
1053
1054         bprm.file = file;
1055         bprm.filename = filename;
1056         bprm.sh_bang = 0;
1057         bprm.loader = 0;
1058         bprm.exec = 0;
1059         bprm.security = NULL;
1060         bprm.mm = mm_alloc();
1061         retval = -ENOMEM;
1062         if (!bprm.mm)
1063                 goto out_file;
1064
1065         retval = init_new_context(current, bprm.mm);
1066         if (retval < 0)
1067                 goto out_mm;
1068
1069         bprm.argc = count(argv, bprm.p / sizeof(void *));
1070         if ((retval = bprm.argc) < 0)
1071                 goto out_mm;
1072
1073         bprm.envc = count(envp, bprm.p / sizeof(void *));
1074         if ((retval = bprm.envc) < 0)
1075                 goto out_mm;
1076
1077         retval = security_bprm_alloc(&bprm);
1078         if (retval)
1079                 goto out;
1080
1081         retval = prepare_binprm(&bprm);
1082         if (retval < 0)
1083                 goto out;
1084
1085         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
1086         if (retval < 0)
1087                 goto out;
1088
1089         bprm.exec = bprm.p;
1090         retval = copy_strings(bprm.envc, envp, &bprm);
1091         if (retval < 0)
1092                 goto out;
1093
1094         retval = copy_strings(bprm.argc, argv, &bprm);
1095         if (retval < 0)
1096                 goto out;
1097
1098         retval = search_binary_handler(&bprm,regs);
1099         if (retval >= 0) {
1100                 free_arg_pages(&bprm);
1101
1102                 /* execve success */
1103                 security_bprm_free(&bprm);
1104                 return retval;
1105         }
1106
1107 out:
1108         /* Something went wrong, return the inode and free the argument pages*/
1109         free_arg_pages(&bprm);
1110
1111         if (bprm.security)
1112                 security_bprm_free(&bprm);
1113
1114 out_mm:
1115         if (bprm.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         struct task_struct *tsk = current;
1260         struct completion *vfork_done = tsk->vfork_done;
1261
1262         /*
1263          * Make sure nobody is waiting for us to release the VM,
1264          * otherwise we can deadlock when we wait on each other
1265          */
1266         if (vfork_done) {
1267                 tsk->vfork_done = NULL;
1268                 complete(vfork_done);
1269         }
1270
1271         read_lock(&tasklist_lock);
1272         do_each_thread(g,p)
1273                 if (mm == p->mm && p != tsk) {
1274                         force_sig_specific(SIGKILL, p);
1275                         mm->core_waiters++;
1276                 }
1277         while_each_thread(g,p);
1278
1279         read_unlock(&tasklist_lock);
1280 }
1281
1282 static void coredump_wait(struct mm_struct *mm)
1283 {
1284         DECLARE_COMPLETION(startup_done);
1285
1286         mm->core_waiters++; /* let other threads block */
1287         mm->core_startup_done = &startup_done;
1288
1289         /* give other threads a chance to run: */
1290         yield();
1291
1292         zap_threads(mm);
1293         if (--mm->core_waiters) {
1294                 up_write(&mm->mmap_sem);
1295                 wait_for_completion(&startup_done);
1296         } else
1297                 up_write(&mm->mmap_sem);
1298         BUG_ON(mm->core_waiters);
1299 }
1300
1301 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1302 {
1303         char corename[CORENAME_MAX_SIZE + 1];
1304         struct mm_struct *mm = current->mm;
1305         struct linux_binfmt * binfmt;
1306         struct inode * inode;
1307         struct file * file;
1308         int retval = 0;
1309
1310         lock_kernel();
1311         binfmt = current->binfmt;
1312         if (!binfmt || !binfmt->core_dump)
1313                 goto fail;
1314         down_write(&mm->mmap_sem);
1315         if (!mm->dumpable) {
1316                 up_write(&mm->mmap_sem);
1317                 goto fail;
1318         }
1319         mm->dumpable = 0;
1320         init_completion(&mm->core_done);
1321         current->signal->group_exit = 1;
1322         current->signal->group_exit_code = exit_code;
1323         coredump_wait(mm);
1324
1325         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1326                 goto fail_unlock;
1327
1328         format_corename(corename, core_pattern, signr);
1329         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW, 0600);
1330         if (IS_ERR(file))
1331                 goto fail_unlock;
1332         inode = file->f_dentry->d_inode;
1333         if (inode->i_nlink > 1)
1334                 goto close_fail;        /* multiple links - don't dump */
1335         if (d_unhashed(file->f_dentry))
1336                 goto close_fail;
1337
1338         if (!S_ISREG(inode->i_mode))
1339                 goto close_fail;
1340         if (!file->f_op)
1341                 goto close_fail;
1342         if (!file->f_op->write)
1343                 goto close_fail;
1344         if (do_truncate(file->f_dentry, 0) != 0)
1345                 goto close_fail;
1346
1347         retval = binfmt->core_dump(signr, regs, file);
1348
1349         current->signal->group_exit_code |= 0x80;
1350 close_fail:
1351         filp_close(file, NULL);
1352 fail_unlock:
1353         complete_all(&mm->core_done);
1354 fail:
1355         unlock_kernel();
1356         return retval;
1357 }