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 #define __NO_VERSION__
  38 #include <linux/module.h>
  39
  40 #include <asm/uaccess.h>
  41 #include <asm/pgalloc.h>
  42 #include <asm/mmu_context.h>
  43
  44 #ifdef CONFIG_KMOD
  45 #include <linux/kmod.h>
  46 #endif
  47
  48 static struct linux_binfmt *formats;
  49 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  50
  51 int register_binfmt(struct linux_binfmt * fmt)
  52 {
  53         struct linux_binfmt ** tmp = &formats;
  54
  55         if (!fmt)
  56                 return -EINVAL;
  57         if (fmt->next)
  58                 return -EBUSY;
  59         write_lock(&binfmt_lock);
  60         while (*tmp) {
  61                 if (fmt == *tmp) {
  62                         write_unlock(&binfmt_lock);
  63                         return -EBUSY;
  64                 }
  65                 tmp = &(*tmp)->next;
  66         }
  67         fmt->next = formats;
  68         formats = fmt;
  69         write_unlock(&binfmt_lock);
  70         return 0;
  71 }
  72
  73 int unregister_binfmt(struct linux_binfmt * fmt)
  74 {
  75         struct linux_binfmt ** tmp = &formats;
  76
  77         write_lock(&binfmt_lock);
  78         while (*tmp) {
  79                 if (fmt == *tmp) {
  80                         *tmp = fmt->next;
  81                         write_unlock(&binfmt_lock);
  82                         return 0;
  83                 }
  84                 tmp = &(*tmp)->next;
  85         }
  86         write_unlock(&binfmt_lock);
  87         return -EINVAL;
  88 }
  89
  90 static inline void put_binfmt(struct linux_binfmt * fmt)
  91 {
  92         if (fmt->module)
  93                 __MOD_DEC_USE_COUNT(fmt->module);
  94 }
  95
  96 /*
  97  * Note that a shared library must be both readable and executable due to
  98  * security reasons.
  99  *
 100  * Also note that we take the address to load from from the file itself.
 101  */
 102 asmlinkage long sys_uselib(const char * library)
 103 {
 104         struct file * file;
 105         struct nameidata nd;
 106         int error;
 107
 108         error = user_path_walk(library, &nd);
 109         if (error)
 110                 goto out;
 111
 112         error = -EINVAL;
 113         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 114                 goto exit;
 115
 116         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC);
 117         if (error)
 118                 goto exit;
 119
 120         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 121         error = PTR_ERR(file);
 122         if (IS_ERR(file))
 123                 goto out;
 124
 125         error = -ENOEXEC;
 126         if(file->f_op && file->f_op->read) {
 127                 struct linux_binfmt * fmt;
 128
 129                 read_lock(&binfmt_lock);
 130                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 131                         if (!fmt->load_shlib)
 132                                 continue;
 133                         if (!try_inc_mod_count(fmt->module))
 134                                 continue;
 135                         read_unlock(&binfmt_lock);
 136                         error = fmt->load_shlib(file);
 137                         read_lock(&binfmt_lock);
 138                         put_binfmt(fmt);
 139                         if (error != -ENOEXEC)
 140                                 break;
 141                 }
 142                 read_unlock(&binfmt_lock);
 143         }
 144         fput(file);
 145 out:
 146         return error;
 147 exit:
 148         path_release(&nd);
 149         goto out;
 150 }
 151
 152 /*
 153  * count() counts the number of arguments/envelopes
 154  */
 155 static int count(char ** argv, int max)
 156 {
 157         int i = 0;
 158
 159         if (argv != NULL) {
 160                 for (;;) {
 161                         char * p;
 162                         int error;
 163
 164                         error = get_user(p,argv);
 165                         if (error)
 166                                 return error;
 167                         if (!p)
 168                                 break;
 169                         argv++;
 170                         if(++i > max)
 171                                 return -E2BIG;
 172                 }
 173         }
 174         return i;
 175 }
 176
 177 /*
 178  * 'copy_strings()' copies argument/envelope strings from user
 179  * memory to free pages in kernel mem. These are in a format ready
 180  * to be put directly into the top of new user memory.
 181  */
 182 int copy_strings(int argc,char ** argv, struct linux_binprm *bprm)
 183 {
 184         while (argc-- > 0) {
 185                 char *str;
 186                 int len;
 187                 unsigned long pos;
 188
 189                 if (get_user(str, argv+argc) || !str || !(len = strnlen_user(str, bprm->p)))
 190                         return -EFAULT;
 191                 if (bprm->p < len)
 192                         return -E2BIG;
 193
 194                 bprm->p -= len;
 195                 /* XXX: add architecture specific overflow check here. */
 196
 197                 pos = bprm->p;
 198                 while (len > 0) {
 199                         char *kaddr;
 200                         int i, new, err;
 201                         struct page *page;
 202                         int offset, bytes_to_copy;
 203
 204                         offset = pos % PAGE_SIZE;
 205                         i = pos/PAGE_SIZE;
 206                         page = bprm->page[i];
 207                         new = 0;
 208                         if (!page) {
 209                                 page = alloc_page(GFP_HIGHUSER);
 210                                 bprm->page[i] = page;
 211                                 if (!page)
 212                                         return -ENOMEM;
 213                                 new = 1;
 214                         }
 215                         kaddr = kmap(page);
 216
 217                         if (new && offset)
 218                                 memset(kaddr, 0, offset);
 219                         bytes_to_copy = PAGE_SIZE - offset;
 220                         if (bytes_to_copy > len) {
 221                                 bytes_to_copy = len;
 222                                 if (new)
 223                                         memset(kaddr+offset+len, 0, PAGE_SIZE-offset-len);
 224                         }
 225                         err = copy_from_user(kaddr + offset, str, bytes_to_copy);
 226                         kunmap(page);
 227
 228                         if (err)
 229                                 return -EFAULT;
 230
 231                         pos += bytes_to_copy;
 232                         str += bytes_to_copy;
 233                         len -= bytes_to_copy;
 234                 }
 235         }
 236         return 0;
 237 }
 238
 239 /*
 240  * Like copy_strings, but get argv and its values from kernel memory.
 241  */
 242 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 243 {
 244         int r;
 245         mm_segment_t oldfs = get_fs();
 246         set_fs(KERNEL_DS);
 247         r = copy_strings(argc, argv, bprm);
 248         set_fs(oldfs);
 249         return r;
 250 }
 251
 252 /*
 253  * This routine is used to map in a page into an address space: needed by
 254  * execve() for the initial stack and environment pages.
 255  */
 256 void put_dirty_page(struct task_struct * tsk, struct page *page, unsigned long address)
 257 {
 258         pgd_t * pgd;
 259         pmd_t * pmd;
 260         pte_t * pte;
 261
 262         if (page_count(page) != 1)
 263                 printk("mem_map disagrees with %p at %08lx\n", page, address);
 264         pgd = pgd_offset(tsk->mm, address);
 265         pmd = pmd_alloc(pgd, address);
 266         if (!pmd) {
 267                 __free_page(page);
 268                 force_sig(SIGKILL, tsk);
 269                 return;
 270         }
 271         pte = pte_alloc(pmd, address);
 272         if (!pte) {
 273                 __free_page(page);
 274                 force_sig(SIGKILL, tsk);
 275                 return;
 276         }
 277         if (!pte_none(*pte)) {
 278                 pte_ERROR(*pte);
 279                 __free_page(page);
 280                 return;
 281         }
 282         flush_dcache_page(page);
 283         flush_page_to_ram(page);
 284         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(page, PAGE_COPY))));
 285 /* no need for flush_tlb */
 286 }
 287
 288 int setup_arg_pages(struct linux_binprm *bprm)
 289 {
 290         unsigned long stack_base;
 291         struct vm_area_struct *mpnt;
 292         int i;
 293
 294         stack_base = STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE;
 295
 296         bprm->p += stack_base;
 297         if (bprm->loader)
 298                 bprm->loader += stack_base;
 299         bprm->exec += stack_base;
 300
 301         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 302         if (!mpnt)
 303                 return -ENOMEM;
 304
 305         down(&current->mm->mmap_sem);
 306         {
 307                 mpnt->vm_mm = current->mm;
 308                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 309                 mpnt->vm_end = STACK_TOP;
 310                 mpnt->vm_page_prot = PAGE_COPY;
 311                 mpnt->vm_flags = VM_STACK_FLAGS;
 312                 mpnt->vm_ops = NULL;
 313                 mpnt->vm_pgoff = 0;
 314                 mpnt->vm_file = NULL;
 315                 mpnt->vm_private_data = (void *) 0;
 316                 spin_lock(&current->mm->page_table_lock);
 317                 insert_vm_struct(current->mm, mpnt);
 318                 spin_unlock(&current->mm->page_table_lock);
 319                 current->mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 320         }
 321
 322         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 323                 struct page *page = bprm->page[i];
 324                 if (page) {
 325                         bprm->page[i] = NULL;
 326                         current->mm->rss++;
 327                         put_dirty_page(current,page,stack_base);
 328                 }
 329                 stack_base += PAGE_SIZE;
 330         }
 331         up(&current->mm->mmap_sem);
 332
 333         return 0;
 334 }
 335
 336 struct file *open_exec(const char *name)
 337 {
 338         struct nameidata nd;
 339         struct inode *inode;
 340         struct file *file;
 341         int err = 0;
 342
 343         if (path_init(name, LOOKUP_FOLLOW|LOOKUP_POSITIVE, &nd))
 344                 err = path_walk(name, &nd);
 345         file = ERR_PTR(err);
 346         if (!err) {
 347                 inode = nd.dentry->d_inode;
 348                 file = ERR_PTR(-EACCES);
 349                 if (!IS_NOEXEC(inode) && S_ISREG(inode->i_mode)) {
 350                         int err = permission(inode, MAY_EXEC);
 351                         file = ERR_PTR(err);
 352                         if (!err) {
 353                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 354                                 if (!IS_ERR(file)) {
 355                                         err = deny_write_access(file);
 356                                         if (err) {
 357                                                 fput(file);
 358                                                 file = ERR_PTR(err);
 359                                         }
 360                                 }
 361 out:
 362                                 return file;
 363                         }
 364                 }
 365                 path_release(&nd);
 366         }
 367         goto out;
 368 }
 369
 370 int kernel_read(struct file *file, unsigned long offset,
 371         char * addr, unsigned long count)
 372 {
 373         mm_segment_t old_fs;
 374         loff_t pos = offset;
 375         int result = -ENOSYS;
 376
 377         if (!file->f_op->read)
 378                 goto fail;
 379         old_fs = get_fs();
 380         set_fs(get_ds());
 381         result = file->f_op->read(file, addr, count, &pos);
 382         set_fs(old_fs);
 383 fail:
 384         return result;
 385 }
 386
 387 static int exec_mmap(void)
 388 {
 389         struct mm_struct * mm, * old_mm;
 390
 391         old_mm = current->mm;
 392         if (old_mm && atomic_read(&old_mm->mm_users) == 1) {
 393                 flush_cache_mm(old_mm);
 394                 mm_release();
 395                 exit_mmap(old_mm);
 396                 flush_tlb_mm(old_mm);
 397                 return 0;
 398         }
 399
 400         mm = mm_alloc();
 401         if (mm) {
 402                 struct mm_struct *active_mm = current->active_mm;
 403
 404                 if (init_new_context(current, mm)) {
 405                         mmdrop(mm);
 406                         return -ENOMEM;
 407                 }
 408                 task_lock(current);
 409                 current->mm = mm;
 410                 current->active_mm = mm;
 411                 task_unlock(current);
 412                 activate_mm(active_mm, mm);
 413                 mm_release();
 414                 if (old_mm) {
 415                         if (active_mm != old_mm) BUG();
 416                         mmput(old_mm);
 417                         return 0;
 418                 }
 419                 mmdrop(active_mm);
 420                 return 0;
 421         }
 422         return -ENOMEM;
 423 }
 424
 425 /*
 426  * This function makes sure the current process has its own signal table,
 427  * so that flush_signal_handlers can later reset the handlers without
 428  * disturbing other processes.  (Other processes might share the signal
 429  * table via the CLONE_SIGNAL option to clone().)
 430  */
 431
 432 static inline int make_private_signals(void)
 433 {
 434         struct signal_struct * newsig;
 435
 436         if (atomic_read(&current->sig->count) <= 1)
 437                 return 0;
 438         newsig = kmem_cache_alloc(sigact_cachep, GFP_KERNEL);
 439         if (newsig == NULL)
 440                 return -ENOMEM;
 441         spin_lock_init(&newsig->siglock);
 442         atomic_set(&newsig->count, 1);
 443         memcpy(newsig->action, current->sig->action, sizeof(newsig->action));
 444         spin_lock_irq(&current->sigmask_lock);
 445         current->sig = newsig;
 446         spin_unlock_irq(&current->sigmask_lock);
 447         return 0;
 448 }
 449
 450 /*
 451  * If make_private_signals() made a copy of the signal table, decrement the
 452  * refcount of the original table, and free it if necessary.
 453  * We don't do that in make_private_signals() so that we can back off
 454  * in flush_old_exec() if an error occurs after calling make_private_signals().
 455  */
 456
 457 static inline void release_old_signals(struct signal_struct * oldsig)
 458 {
 459         if (current->sig == oldsig)
 460                 return;
 461         if (atomic_dec_and_test(&oldsig->count))
 462                 kmem_cache_free(sigact_cachep, oldsig);
 463 }
 464
 465 /*
 466  * These functions flushes out all traces of the currently running executable
 467  * so that a new one can be started
 468  */
 469
 470 static inline void flush_old_files(struct files_struct * files)
 471 {
 472         long j = -1;
 473
 474         write_lock(&files->file_lock);
 475         for (;;) {
 476                 unsigned long set, i;
 477
 478                 j++;
 479                 i = j * __NFDBITS;
 480                 if (i >= files->max_fds || i >= files->max_fdset)
 481                         break;
 482                 set = files->close_on_exec->fds_bits[j];
 483                 if (!set)
 484                         continue;
 485                 files->close_on_exec->fds_bits[j] = 0;
 486                 write_unlock(&files->file_lock);
 487                 for ( ; set ; i++,set >>= 1) {
 488                         if (set & 1) {
 489                                 sys_close(i);
 490                         }
 491                 }
 492                 write_lock(&files->file_lock);
 493
 494         }
 495         write_unlock(&files->file_lock);
 496 }
 497
 498 /*
 499  * An execve() will automatically "de-thread" the process.
 500  * Note: we don't have to hold the tasklist_lock to test
 501  * whether we migth need to do this. If we're not part of
 502  * a thread group, there is no way we can become one
 503  * dynamically. And if we are, we only need to protect the
 504  * unlink - even if we race with the last other thread exit,
 505  * at worst the list_del_init() might end up being a no-op.
 506  */
 507 static inline void de_thread(struct task_struct *tsk)
 508 {
 509         if (!list_empty(&tsk->thread_group)) {
 510                 write_lock_irq(&tasklist_lock);
 511                 list_del_init(&tsk->thread_group);
 512                 write_unlock_irq(&tasklist_lock);
 513         }
 514
 515         /* Minor oddity: this might stay the same. */
 516         tsk->tgid = tsk->pid;
 517 }
 518
 519 int flush_old_exec(struct linux_binprm * bprm)
 520 {
 521         char * name;
 522         int i, ch, retval;
 523         struct signal_struct * oldsig;
 524
 525         /*
 526          * Make sure we have a private signal table
 527          */
 528         oldsig = current->sig;
 529         retval = make_private_signals();
 530         if (retval) goto flush_failed;
 531
 532         /*
 533          * Release all of the old mmap stuff
 534          */
 535         retval = exec_mmap();
 536         if (retval) goto mmap_failed;
 537
 538         /* This is the point of no return */
 539         release_old_signals(oldsig);
 540
 541         current->sas_ss_sp = current->sas_ss_size = 0;
 542
 543         if (current->euid == current->uid && current->egid == current->gid)
 544                 current->dumpable = 1;
 545         name = bprm->filename;
 546         for (i=0; (ch = *(name++)) != '\0';) {
 547                 if (ch == '/')
 548                         i = 0;
 549                 else
 550                         if (i < 15)
 551                                 current->comm[i++] = ch;
 552         }
 553         current->comm[i] = '\0';
 554
 555         flush_thread();
 556
 557         de_thread(current);
 558
 559         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 560             permission(bprm->file->f_dentry->d_inode,MAY_READ))
 561                 current->dumpable = 0;
 562
 563         /* An exec changes our domain. We are no longer part of the thread
 564            group */
 565
 566         current->self_exec_id++;
 567
 568         flush_signal_handlers(current);
 569         flush_old_files(current->files);
 570
 571         return 0;
 572
 573 mmap_failed:
 574 flush_failed:
 575         spin_lock_irq(&current->sigmask_lock);
 576         if (current->sig != oldsig)
 577                 kfree(current->sig);
 578         current->sig = oldsig;
 579         spin_unlock_irq(&current->sigmask_lock);
 580         return retval;
 581 }
 582
 583 /*
 584  * We mustn't allow tracing of suid binaries, unless
 585  * the tracer has the capability to trace anything..
 586  */
 587 static inline int must_not_trace_exec(struct task_struct * p)
 588 {
 589         return (p->ptrace & PT_PTRACED) && !cap_raised(p->p_pptr->cap_effective, CAP_SYS_PTRACE);
 590 }
 591
 592 /*
 593  * Fill the binprm structure from the inode.
 594  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 595  */
 596 int prepare_binprm(struct linux_binprm *bprm)
 597 {
 598         int mode;
 599         int id_change,cap_raised;
 600         struct inode * inode = bprm->file->f_dentry->d_inode;
 601
 602         mode = inode->i_mode;
 603         /* Huh? We had already checked for MAY_EXEC, WTF do we check this? */
 604         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 605                 return -EACCES;
 606         if (bprm->file->f_op == NULL)
 607                 return -EACCES;
 608
 609         bprm->e_uid = current->euid;
 610         bprm->e_gid = current->egid;
 611         id_change = cap_raised = 0;
 612
 613         /* Set-uid? */
 614         if (mode & S_ISUID) {
 615                 bprm->e_uid = inode->i_uid;
 616                 if (bprm->e_uid != current->euid)
 617                         id_change = 1;
 618         }
 619
 620         /* Set-gid? */
 621         /*
 622          * If setgid is set but no group execute bit then this
 623          * is a candidate for mandatory locking, not a setgid
 624          * executable.
 625          */
 626         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 627                 bprm->e_gid = inode->i_gid;
 628                 if (!in_group_p(bprm->e_gid))
 629                         id_change = 1;
 630         }
 631
 632         /* We don't have VFS support for capabilities yet */
 633         cap_clear(bprm->cap_inheritable);
 634         cap_clear(bprm->cap_permitted);
 635         cap_clear(bprm->cap_effective);
 636
 637         /*  To support inheritance of root-permissions and suid-root
 638          *  executables under compatibility mode, we raise all three
 639          *  capability sets for the file.
 640          *
 641          *  If only the real uid is 0, we only raise the inheritable
 642          *  and permitted sets of the executable file.
 643          */
 644
 645         if (!issecure(SECURE_NOROOT)) {
 646                 if (bprm->e_uid == 0 || current->uid == 0) {
 647                         cap_set_full(bprm->cap_inheritable);
 648                         cap_set_full(bprm->cap_permitted);
 649                 }
 650                 if (bprm->e_uid == 0)
 651                         cap_set_full(bprm->cap_effective);
 652         }
 653
 654         /* Only if pP' is _not_ a subset of pP, do we consider there
 655          * has been a capability related "change of capability".  In
 656          * such cases, we need to check that the elevation of
 657          * privilege does not go against other system constraints.
 658          * The new Permitted set is defined below -- see (***). */
 659         {
 660                 kernel_cap_t permitted, working;
 661
 662                 permitted = cap_intersect(bprm->cap_permitted, cap_bset);
 663                 working = cap_intersect(bprm->cap_inheritable,
 664                                         current->cap_inheritable);
 665                 working = cap_combine(permitted, working);
 666                 if (!cap_issubset(working, current->cap_permitted)) {
 667                         cap_raised = 1;
 668                 }
 669         }
 670
 671         if (id_change || cap_raised) {
 672                 /* We can't suid-execute if we're sharing parts of the executable */
 673                 /* or if we're being traced (or if suid execs are not allowed)    */
 674                 /* (current->mm->mm_users > 1 is ok, as we'll get a new mm anyway)   */
 675                 if (IS_NOSUID(inode)
 676                     || must_not_trace_exec(current)
 677                     || (atomic_read(&current->fs->count) > 1)
 678                     || (atomic_read(&current->sig->count) > 1)
 679                     || (atomic_read(&current->files->count) > 1)) {
 680                         if (id_change && !capable(CAP_SETUID))
 681                                 return -EPERM;
 682                         if (cap_raised && !capable(CAP_SETPCAP))
 683                                 return -EPERM;
 684                 }
 685         }
 686
 687         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 688         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 689 }
 690
 691 /*
 692  * This function is used to produce the new IDs and capabilities
 693  * from the old ones and the file's capabilities.
 694  *
 695  * The formula used for evolving capabilities is:
 696  *
 697  *       pI' = pI
 698  * (***) pP' = (fP & X) | (fI & pI)
 699  *       pE' = pP' & fE          [NB. fE is 0 or ~0]
 700  *
 701  * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
 702  * ' indicates post-exec(), and X is the global 'cap_bset'.
 703  */
 704
 705 void compute_creds(struct linux_binprm *bprm)
 706 {
 707         kernel_cap_t new_permitted, working;
 708
 709         new_permitted = cap_intersect(bprm->cap_permitted, cap_bset);
 710         working = cap_intersect(bprm->cap_inheritable,
 711                                 current->cap_inheritable);
 712         new_permitted = cap_combine(new_permitted, working);
 713
 714         /* For init, we want to retain the capabilities set
 715          * in the init_task struct. Thus we skip the usual
 716          * capability rules */
 717         if (current->pid != 1) {
 718                 current->cap_permitted = new_permitted;
 719                 current->cap_effective =
 720                         cap_intersect(new_permitted, bprm->cap_effective);
 721         }
 722
 723         /* AUD: Audit candidate if current->cap_effective is set */
 724
 725         current->suid = current->euid = current->fsuid = bprm->e_uid;
 726         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 727         if (current->euid != current->uid || current->egid != current->gid ||
 728             !cap_issubset(new_permitted, current->cap_permitted))
 729                 current->dumpable = 0;
 730
 731         current->keep_capabilities = 0;
 732 }
 733
 734
 735 void remove_arg_zero(struct linux_binprm *bprm)
 736 {
 737         if (bprm->argc) {
 738                 unsigned long offset;
 739                 char * kaddr;
 740                 struct page *page;
 741
 742                 offset = bprm->p % PAGE_SIZE;
 743                 goto inside;
 744
 745                 while (bprm->p++, *(kaddr+offset++)) {
 746                         if (offset != PAGE_SIZE)
 747                                 continue;
 748                         offset = 0;
 749                         kunmap(page);
 750 inside:
 751                         page = bprm->page[bprm->p/PAGE_SIZE];
 752                         kaddr = kmap(page);
 753                 }
 754                 kunmap(page);
 755                 bprm->argc--;
 756         }
 757 }
 758
 759 /*
 760  * cycle the list of binary formats handler, until one recognizes the image
 761  */
 762 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 763 {
 764         int try,retval=0;
 765         struct linux_binfmt *fmt;
 766 #ifdef __alpha__
 767         /* handle /sbin/loader.. */
 768         {
 769             struct exec * eh = (struct exec *) bprm->buf;
 770             struct linux_binprm bprm_loader;
 771
 772             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 773                 (eh->fh.f_flags & 0x3000) == 0x3000)
 774             {
 775                 int i;
 776                 char * dynloader[] = { "/sbin/loader" };
 777                 struct file * file;
 778
 779                 allow_write_access(bprm->file);
 780                 fput(bprm->file);
 781                 bprm->file = NULL;
 782
 783                 bprm_loader.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 784                 for (i = 0 ; i < MAX_ARG_PAGES ; i++)   /* clear page-table */
 785                     bprm_loader.page[i] = NULL;
 786
 787                 file = open_exec(dynloader[0]);
 788                 retval = PTR_ERR(file);
 789                 if (IS_ERR(file))
 790                         return retval;
 791                 bprm->file = file;
 792                 bprm->loader = bprm_loader.p;
 793                 retval = prepare_binprm(bprm);
 794                 if (retval<0)
 795                         return retval;
 796                 /* should call search_binary_handler recursively here,
 797                    but it does not matter */
 798             }
 799         }
 800 #endif
 801         for (try=0; try<2; try++) {
 802                 read_lock(&binfmt_lock);
 803                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 804                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
 805                         if (!fn)
 806                                 continue;
 807                         if (!try_inc_mod_count(fmt->module))
 808                                 continue;
 809                         read_unlock(&binfmt_lock);
 810                         retval = fn(bprm, regs);
 811                         if (retval >= 0) {
 812                                 put_binfmt(fmt);
 813                                 allow_write_access(bprm->file);
 814                                 if (bprm->file)
 815                                         fput(bprm->file);
 816                                 bprm->file = NULL;
 817                                 current->did_exec = 1;
 818                                 return retval;
 819                         }
 820                         read_lock(&binfmt_lock);
 821                         put_binfmt(fmt);
 822                         if (retval != -ENOEXEC)
 823                                 break;
 824                         if (!bprm->file) {
 825                                 read_unlock(&binfmt_lock);
 826                                 return retval;
 827                         }
 828                 }
 829                 read_unlock(&binfmt_lock);
 830                 if (retval != -ENOEXEC) {
 831                         break;
 832 #ifdef CONFIG_KMOD
 833                 }else{
 834 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
 835                         char modname[20];
 836                         if (printable(bprm->buf[0]) &&
 837                             printable(bprm->buf[1]) &&
 838                             printable(bprm->buf[2]) &&
 839                             printable(bprm->buf[3]))
 840                                 break; /* -ENOEXEC */
 841                         sprintf(modname, "binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
 842                         request_module(modname);
 843 #endif
 844                 }
 845         }
 846         return retval;
 847 }
 848
 849
 850 /*
 851  * sys_execve() executes a new program.
 852  */
 853 int do_execve(char * filename, char ** argv, char ** envp, struct pt_regs * regs)
 854 {
 855         struct linux_binprm bprm;
 856         struct file *file;
 857         int retval;
 858         int i;
 859
 860         file = open_exec(filename);
 861
 862         retval = PTR_ERR(file);
 863         if (IS_ERR(file))
 864                 return retval;
 865
 866         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 867         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
 868
 869         bprm.file = file;
 870         bprm.filename = filename;
 871         bprm.sh_bang = 0;
 872         bprm.loader = 0;
 873         bprm.exec = 0;
 874         if ((bprm.argc = count(argv, bprm.p / sizeof(void *))) < 0) {
 875                 allow_write_access(file);
 876                 fput(file);
 877                 return bprm.argc;
 878         }
 879
 880         if ((bprm.envc = count(envp, bprm.p / sizeof(void *))) < 0) {
 881                 allow_write_access(file);
 882                 fput(file);
 883                 return bprm.envc;
 884         }
 885
 886         retval = prepare_binprm(&bprm);
 887         if (retval < 0)
 888                 goto out;
 889
 890         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
 891         if (retval < 0)
 892                 goto out;
 893
 894         bprm.exec = bprm.p;
 895         retval = copy_strings(bprm.envc, envp, &bprm);
 896         if (retval < 0)
 897                 goto out;
 898
 899         retval = copy_strings(bprm.argc, argv, &bprm);
 900         if (retval < 0)
 901                 goto out;
 902
 903         retval = search_binary_handler(&bprm,regs);
 904         if (retval >= 0)
 905                 /* execve success */
 906                 return retval;
 907
 908 out:
 909         /* Something went wrong, return the inode and free the argument pages*/
 910         allow_write_access(bprm.file);
 911         if (bprm.file)
 912                 fput(bprm.file);
 913
 914         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 915                 struct page * page = bprm.page[i];
 916                 if (page)
 917                         __free_page(page);
 918         }
 919
 920         return retval;
 921 }
 922
 923 void set_binfmt(struct linux_binfmt *new)
 924 {
 925         struct linux_binfmt *old = current->binfmt;
 926         if (new && new->module)
 927                 __MOD_INC_USE_COUNT(new->module);
 928         current->binfmt = new;
 929         if (old && old->module)
 930                 __MOD_DEC_USE_COUNT(old->module);
 931 }
 932
 933 int do_coredump(long signr, struct pt_regs * regs)
 934 {
 935         struct linux_binfmt * binfmt;
 936         char corename[6+sizeof(current->comm)];
 937         struct file * file;
 938         struct inode * inode;
 939
 940         lock_kernel();
 941         binfmt = current->binfmt;
 942         if (!binfmt || !binfmt->core_dump)
 943                 goto fail;
 944         if (!current->dumpable || atomic_read(&current->mm->mm_users) != 1)
 945                 goto fail;
 946         current->dumpable = 0;
 947         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
 948                 goto fail;
 949
 950         memcpy(corename,"core.", 5);
 951 #if 0
 952         memcpy(corename+5,current->comm,sizeof(current->comm));
 953 #else
 954         corename[4] = '\0';
 955 #endif
 956         file = filp_open(corename, O_CREAT | 2 | O_TRUNC | O_NOFOLLOW, 0600);
 957         if (IS_ERR(file))
 958                 goto fail;
 959         inode = file->f_dentry->d_inode;
 960         if (inode->i_nlink > 1)
 961                 goto close_fail;        /* multiple links - don't dump */
 962
 963         if (!S_ISREG(inode->i_mode))
 964                 goto close_fail;
 965         if (!file->f_op)
 966                 goto close_fail;
 967         if (!file->f_op->write)
 968                 goto close_fail;
 969         if (!binfmt->core_dump(signr, regs, file))
 970                 goto close_fail;
 971         unlock_kernel();
 972         filp_close(file, NULL);
 973         return 1;
 974
 975 close_fail:
 976         filp_close(file, NULL);
 977 fail:
 978         unlock_kernel();
 979         return 0;
 980 }