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
  35 #include <asm/uaccess.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/mmu_context.h>
  38
  39 #ifdef CONFIG_KMOD
  40 #include <linux/kmod.h>
  41 #endif
  42
  43 /*
  44  * Here are the actual binaries that will be accepted:
  45  * add more with "register_binfmt()" if using modules...
  46  *
  47  * These are defined again for the 'real' modules if you are using a
  48  * module definition for these routines.
  49  */
  50
  51 static struct linux_binfmt *formats = (struct linux_binfmt *) NULL;
  52
  53 void __init binfmt_setup(void)
  54 {
  55 #ifdef CONFIG_BINFMT_MISC
  56         init_misc_binfmt();
  57 #endif
  58
  59 #ifdef CONFIG_BINFMT_ELF
  60         init_elf_binfmt();
  61 #endif
  62
  63 #ifdef CONFIG_BINFMT_ELF32
  64         init_elf32_binfmt();
  65 #endif
  66
  67 #ifdef CONFIG_BINFMT_AOUT
  68         init_aout_binfmt();
  69 #endif
  70
  71 #ifdef CONFIG_BINFMT_AOUT32
  72         init_aout32_binfmt();
  73 #endif
  74
  75 #ifdef CONFIG_BINFMT_EM86
  76         init_em86_binfmt();
  77 #endif
  78
  79         /* This cannot be configured out of the kernel */
  80         init_script_binfmt();
  81 }
  82
  83 int register_binfmt(struct linux_binfmt * fmt)
  84 {
  85         struct linux_binfmt ** tmp = &formats;
  86
  87         if (!fmt)
  88                 return -EINVAL;
  89         if (fmt->next)
  90                 return -EBUSY;
  91         while (*tmp) {
  92                 if (fmt == *tmp)
  93                         return -EBUSY;
  94                 tmp = &(*tmp)->next;
  95         }
  96         fmt->next = formats;
  97         formats = fmt;
  98         return 0;
  99 }
 100
 101 #ifdef CONFIG_MODULES
 102 int unregister_binfmt(struct linux_binfmt * fmt)
 103 {
 104         struct linux_binfmt ** tmp = &formats;
 105
 106         while (*tmp) {
 107                 if (fmt == *tmp) {
 108                         *tmp = fmt->next;
 109                         return 0;
 110                 }
 111                 tmp = &(*tmp)->next;
 112         }
 113         return -EINVAL;
 114 }
 115 #endif  /* CONFIG_MODULES */
 116
 117 /* N.B. Error returns must be < 0 */
 118 int open_dentry(struct dentry * dentry, int mode)
 119 {
 120         struct inode * inode = dentry->d_inode;
 121         struct file * f;
 122         int fd, error;
 123
 124         error = -EINVAL;
 125         if (!inode->i_op || !inode->i_op->default_file_ops)
 126                 goto out;
 127         fd = get_unused_fd();
 128         if (fd >= 0) {
 129                 error = -ENFILE;
 130                 f = get_empty_filp();
 131                 if (!f)
 132                         goto out_fd;
 133                 f->f_flags = mode;
 134                 f->f_mode = (mode+1) & O_ACCMODE;
 135                 f->f_dentry = dentry;
 136                 f->f_pos = 0;
 137                 f->f_reada = 0;
 138                 f->f_op = inode->i_op->default_file_ops;
 139                 if (f->f_op->open) {
 140                         error = f->f_op->open(inode,f);
 141                         if (error)
 142                                 goto out_filp;
 143                 }
 144                 fd_install(fd, f);
 145                 dget(dentry);
 146         }
 147         return fd;
 148
 149 out_filp:
 150         if (error > 0)
 151                 error = -EIO;
 152         put_filp(f);
 153 out_fd:
 154         put_unused_fd(fd);
 155 out:
 156         return error;
 157 }
 158
 159 /*
 160  * Note that a shared library must be both readable and executable due to
 161  * security reasons.
 162  *
 163  * Also note that we take the address to load from from the file itself.
 164  */
 165 asmlinkage int sys_uselib(const char * library)
 166 {
 167         int fd, retval;
 168         struct file * file;
 169         struct linux_binfmt * fmt;
 170
 171         lock_kernel();
 172         fd = sys_open(library, 0, 0);
 173         retval = fd;
 174         if (fd < 0)
 175                 goto out;
 176         file = fget(fd);
 177         retval = -ENOEXEC;
 178         if (file && file->f_dentry && file->f_op && file->f_op->read) {
 179                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 180                         int (*fn)(int) = fmt->load_shlib;
 181                         if (!fn)
 182                                 continue;
 183                         /* N.B. Should use file instead of fd */
 184                         retval = fn(fd);
 185                         if (retval != -ENOEXEC)
 186                                 break;
 187                 }
 188         }
 189         fput(file);
 190         sys_close(fd);
 191 out:
 192         unlock_kernel();
 193         return retval;
 194 }
 195
 196 /*
 197  * count() counts the number of arguments/envelopes
 198  */
 199 static int count(char ** argv)
 200 {
 201         int i = 0;
 202
 203         if (argv != NULL) {
 204                 for (;;) {
 205                         char * p;
 206                         int error;
 207
 208                         error = get_user(p,argv);
 209                         if (error)
 210                                 return error;
 211                         if (!p)
 212                                 break;
 213                         argv++;
 214                         i++;
 215                 }
 216         }
 217         return i;
 218 }
 219
 220 /*
 221  * 'copy_strings()' copies argument/envelope strings from user
 222  * memory to free pages in kernel mem. These are in a format ready
 223  * to be put directly into the top of new user memory.
 224  */
 225 int copy_strings(int argc,char ** argv, struct linux_binprm *bprm)
 226 {
 227         while (argc-- > 0) {
 228                 char *str;
 229                 int len;
 230                 unsigned long pos;
 231
 232                 if (get_user(str, argv+argc) || !str || !(len = strlen_user(str)))
 233                         return -EFAULT;
 234                 if (bprm->p < len)
 235                         return -E2BIG;
 236
 237                 bprm->p -= len;
 238                 /* XXX: add architecture specific overflow check here. */
 239
 240                 pos = bprm->p;
 241                 while (len) {
 242                         char *pag;
 243                         int offset, bytes_to_copy;
 244
 245                         offset = pos % PAGE_SIZE;
 246                         if (!(pag = (char *) bprm->page[pos/PAGE_SIZE]) &&
 247                             !(pag = (char *) bprm->page[pos/PAGE_SIZE] =
 248                               (unsigned long *) get_free_page(GFP_USER)))
 249                                 return -ENOMEM;
 250
 251                         bytes_to_copy = PAGE_SIZE - offset;
 252                         if (bytes_to_copy > len)
 253                                 bytes_to_copy = len;
 254                         if (copy_from_user(pag + offset, str, bytes_to_copy))
 255                                 return -EFAULT;
 256
 257                         pos += bytes_to_copy;
 258                         str += bytes_to_copy;
 259                         len -= bytes_to_copy;
 260                 }
 261         }
 262         return 0;
 263 }
 264
 265 /*
 266  * Like copy_strings, but get argv and its values from kernel memory.
 267  */
 268 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 269 {
 270         int r;
 271         mm_segment_t oldfs = get_fs();
 272         set_fs(KERNEL_DS);
 273         r = copy_strings(argc, argv, bprm);
 274         set_fs(oldfs);
 275         return r;
 276 }
 277
 278 int setup_arg_pages(struct linux_binprm *bprm)
 279 {
 280         unsigned long stack_base;
 281         struct vm_area_struct *mpnt;
 282         int i;
 283
 284         stack_base = STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE;
 285
 286         bprm->p += stack_base;
 287         if (bprm->loader)
 288                 bprm->loader += stack_base;
 289         bprm->exec += stack_base;
 290
 291         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 292         if (!mpnt)
 293                 return -ENOMEM;
 294
 295         {
 296                 mpnt->vm_mm = current->mm;
 297                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 298                 mpnt->vm_end = STACK_TOP;
 299                 mpnt->vm_page_prot = PAGE_COPY;
 300                 mpnt->vm_flags = VM_STACK_FLAGS;
 301                 mpnt->vm_ops = NULL;
 302                 mpnt->vm_offset = 0;
 303                 mpnt->vm_file = NULL;
 304                 mpnt->vm_pte = 0;
 305                 insert_vm_struct(current->mm, mpnt);
 306                 current->mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 307         }
 308
 309         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 310                 if (bprm->page[i]) {
 311                         current->mm->rss++;
 312                         put_dirty_page(current,bprm->page[i],stack_base);
 313                 }
 314                 stack_base += PAGE_SIZE;
 315         }
 316
 317         return 0;
 318 }
 319
 320 /*
 321  * Read in the complete executable. This is used for "-N" files
 322  * that aren't on a block boundary, and for files on filesystems
 323  * without get_block support.
 324  */
 325 int read_exec(struct dentry *dentry, unsigned long offset,
 326         char * addr, unsigned long count, int to_kmem)
 327 {
 328         struct file file;
 329         struct inode * inode = dentry->d_inode;
 330         int result = -ENOEXEC;
 331
 332         if (!inode->i_op || !inode->i_op->default_file_ops)
 333                 goto end_readexec;
 334         if (init_private_file(&file, dentry, 1))
 335                 goto end_readexec;
 336         if (!file.f_op->read)
 337                 goto close_readexec;
 338         if (file.f_op->llseek) {
 339                 if (file.f_op->llseek(&file,offset,0) != offset)
 340                         goto close_readexec;
 341         } else
 342                 file.f_pos = offset;
 343         if (to_kmem) {
 344                 mm_segment_t old_fs = get_fs();
 345                 set_fs(get_ds());
 346                 result = file.f_op->read(&file, addr, count, &file.f_pos);
 347                 set_fs(old_fs);
 348         } else {
 349                 result = verify_area(VERIFY_WRITE, addr, count);
 350                 if (result)
 351                         goto close_readexec;
 352                 result = file.f_op->read(&file, addr, count, &file.f_pos);
 353         }
 354 close_readexec:
 355         if (file.f_op->release)
 356                 file.f_op->release(inode,&file);
 357 end_readexec:
 358         return result;
 359 }
 360
 361 static int exec_mmap(void)
 362 {
 363         struct mm_struct * mm, * old_mm;
 364         int retval, nr;
 365
 366         if (atomic_read(&current->mm->count) == 1) {
 367                 flush_cache_mm(current->mm);
 368                 mm_release();
 369                 release_segments(current->mm);
 370                 exit_mmap(current->mm);
 371                 flush_tlb_mm(current->mm);
 372                 return 0;
 373         }
 374
 375         retval = -ENOMEM;
 376         mm = mm_alloc();
 377         if (!mm)
 378                 goto fail_nomem;
 379
 380         mm->cpu_vm_mask = (1UL << smp_processor_id());
 381         mm->total_vm = 0;
 382         mm->rss = 0;
 383         /*
 384          * Make sure we have a private ldt if needed ...
 385          */
 386         nr = current->tarray_ptr - &task[0];
 387         copy_segments(nr, current, mm);
 388
 389         old_mm = current->mm;
 390         current->mm = mm;
 391         retval = new_page_tables(current);
 392         if (retval)
 393                 goto fail_restore;
 394         activate_context(current);
 395         up(&mm->mmap_sem);
 396         mm_release();
 397         mmput(old_mm);
 398         return 0;
 399
 400         /*
 401          * Failure ... restore the prior mm_struct.
 402          */
 403 fail_restore:
 404         current->mm = old_mm;
 405         /* restore the ldt for this task */
 406         copy_segments(nr, current, NULL);
 407         release_segments(mm);
 408         kmem_cache_free(mm_cachep, mm);
 409
 410 fail_nomem:
 411         return retval;
 412 }
 413
 414 /*
 415  * This function makes sure the current process has its own signal table,
 416  * so that flush_signal_handlers can later reset the handlers without
 417  * disturbing other processes.  (Other processes might share the signal
 418  * table via the CLONE_SIGHAND option to clone().)
 419  */
 420
 421 static inline int make_private_signals(void)
 422 {
 423         struct signal_struct * newsig;
 424
 425         if (atomic_read(&current->sig->count) <= 1)
 426                 return 0;
 427         newsig = kmalloc(sizeof(*newsig), GFP_KERNEL);
 428         if (newsig == NULL)
 429                 return -ENOMEM;
 430         spin_lock_init(&newsig->siglock);
 431         atomic_set(&newsig->count, 1);
 432         memcpy(newsig->action, current->sig->action, sizeof(newsig->action));
 433         current->sig = newsig;
 434         return 0;
 435 }
 436
 437 /*
 438  * If make_private_signals() made a copy of the signal table, decrement the
 439  * refcount of the original table, and free it if necessary.
 440  * We don't do that in make_private_signals() so that we can back off
 441  * in flush_old_exec() if an error occurs after calling make_private_signals().
 442  */
 443
 444 static inline void release_old_signals(struct signal_struct * oldsig)
 445 {
 446         if (current->sig == oldsig)
 447                 return;
 448         if (atomic_dec_and_test(&oldsig->count))
 449                 kfree(oldsig);
 450 }
 451
 452 /*
 453  * These functions flushes out all traces of the currently running executable
 454  * so that a new one can be started
 455  */
 456
 457 static inline void flush_old_files(struct files_struct * files)
 458 {
 459         unsigned long j;
 460
 461         j = 0;
 462         for (;;) {
 463                 unsigned long set, i;
 464
 465                 i = j * __NFDBITS;
 466                 if (i >= files->max_fds)
 467                         break;
 468                 set = files->close_on_exec.fds_bits[j];
 469                 files->close_on_exec.fds_bits[j] = 0;
 470                 j++;
 471                 for ( ; set ; i++,set >>= 1) {
 472                         if (set & 1)
 473                                 sys_close(i);
 474                 }
 475         }
 476 }
 477
 478 int flush_old_exec(struct linux_binprm * bprm)
 479 {
 480         char * name;
 481         int i, ch, retval;
 482         struct signal_struct * oldsig;
 483
 484         /*
 485          * Make sure we have a private signal table
 486          */
 487         oldsig = current->sig;
 488         retval = make_private_signals();
 489         if (retval) goto flush_failed;
 490
 491         /*
 492          * Release all of the old mmap stuff
 493          */
 494         retval = exec_mmap();
 495         if (retval) goto mmap_failed;
 496
 497         /* This is the point of no return */
 498         release_old_signals(oldsig);
 499
 500         if (current->euid == current->uid && current->egid == current->gid)
 501                 current->dumpable = 1;
 502         name = bprm->filename;
 503         for (i=0; (ch = *(name++)) != '\0';) {
 504                 if (ch == '/')
 505                         i = 0;
 506                 else
 507                         if (i < 15)
 508                                 current->comm[i++] = ch;
 509         }
 510         current->comm[i] = '\0';
 511
 512         flush_thread();
 513
 514         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 515             permission(bprm->dentry->d_inode,MAY_READ))
 516                 current->dumpable = 0;
 517
 518         flush_signal_handlers(current);
 519         flush_old_files(current->files);
 520
 521         return 0;
 522
 523 mmap_failed:
 524         if (current->sig != oldsig)
 525                 kfree(current->sig);
 526 flush_failed:
 527         current->sig = oldsig;
 528         return retval;
 529 }
 530
 531 /*
 532  * We mustn't allow tracing of suid binaries, unless
 533  * the tracer has the capability to trace anything..
 534  */
 535 static inline int must_not_trace_exec(struct task_struct * p)
 536 {
 537         return (p->flags & PF_PTRACED) && !cap_raised(p->p_pptr->cap_effective, CAP_SYS_PTRACE);
 538 }
 539
 540 /*
 541  * Fill the binprm structure from the inode.
 542  * Check permissions, then read the first 512 bytes
 543  */
 544 int prepare_binprm(struct linux_binprm *bprm)
 545 {
 546         int mode;
 547         int retval,id_change,cap_raised;
 548         struct inode * inode = bprm->dentry->d_inode;
 549
 550         mode = inode->i_mode;
 551         if (!S_ISREG(mode))                     /* must be regular file */
 552                 return -EACCES;
 553         if (!(mode & 0111))                     /* with at least _one_ execute bit set */
 554                 return -EACCES;
 555         if (IS_NOEXEC(inode))                   /* FS mustn't be mounted noexec */
 556                 return -EACCES;
 557         if (!inode->i_sb)
 558                 return -EACCES;
 559         if ((retval = permission(inode, MAY_EXEC)) != 0)
 560                 return retval;
 561         /* better not execute files which are being written to */
 562         if (inode->i_writecount > 0)
 563                 return -ETXTBSY;
 564
 565         bprm->e_uid = current->euid;
 566         bprm->e_gid = current->egid;
 567         id_change = cap_raised = 0;
 568
 569         /* Set-uid? */
 570         if (mode & S_ISUID) {
 571                 bprm->e_uid = inode->i_uid;
 572                 if (bprm->e_uid != current->euid)
 573                         id_change = 1;
 574         }
 575
 576         /* Set-gid? */
 577         /*
 578          * If setgid is set but no group execute bit then this
 579          * is a candidate for mandatory locking, not a setgid
 580          * executable.
 581          */
 582         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 583                 bprm->e_gid = inode->i_gid;
 584                 if (!in_group_p(bprm->e_gid))
 585                         id_change = 1;
 586         }
 587
 588         /* We don't have VFS support for capabilities yet */
 589         cap_clear(bprm->cap_inheritable);
 590         cap_clear(bprm->cap_permitted);
 591         cap_clear(bprm->cap_effective);
 592
 593         /*  To support inheritance of root-permissions and suid-root
 594          *  executables under compatibility mode, we raise the
 595          *  effective and inherited bitmasks of the executable file
 596          *  (translation: we set the executable "capability dumb" and
 597          *  set the allowed set to maximum). We don't set any forced
 598          *  bits.
 599          *
 600          *  If only the real uid is 0, we only raise the inheritable
 601          *  bitmask of the executable file (translation: we set the
 602          *  allowed set to maximum and the application to "capability
 603          *  smart").
 604          */
 605
 606         if (!issecure(SECURE_NOROOT)) {
 607                 if (bprm->e_uid == 0 || current->uid == 0)
 608                         cap_set_full(bprm->cap_inheritable);
 609                 if (bprm->e_uid == 0)
 610                         cap_set_full(bprm->cap_effective);
 611         }
 612
 613         /* Only if pP' is _not_ a subset of pP, do we consider there
 614          * has been a capability related "change of capability".  In
 615          * such cases, we need to check that the elevation of
 616          * privilege does not go against other system constraints.
 617          * The new Permitted set is defined below -- see (***). */
 618         {
 619                 kernel_cap_t working =
 620                         cap_combine(bprm->cap_permitted,
 621                                     cap_intersect(bprm->cap_inheritable,
 622                                                   current->cap_inheritable));
 623                 if (!cap_issubset(working, current->cap_permitted)) {
 624                         cap_raised = 1;
 625                 }
 626         }
 627
 628         if (id_change || cap_raised) {
 629                 /* We can't suid-execute if we're sharing parts of the executable */
 630                 /* or if we're being traced (or if suid execs are not allowed)    */
 631                 /* (current->mm->count > 1 is ok, as we'll get a new mm anyway)   */
 632                 if (IS_NOSUID(inode)
 633                     || must_not_trace_exec(current)
 634                     || (atomic_read(&current->fs->count) > 1)
 635                     || (atomic_read(&current->sig->count) > 1)
 636                     || (atomic_read(&current->files->count) > 1)) {
 637                         if (id_change && !capable(CAP_SETUID))
 638                                 return -EPERM;
 639                         if (cap_raised && !capable(CAP_SETPCAP))
 640                                 return -EPERM;
 641                 }
 642         }
 643
 644         memset(bprm->buf,0,sizeof(bprm->buf));
 645         return read_exec(bprm->dentry,0,bprm->buf,128,1);
 646 }
 647
 648 /*
 649  * This function is used to produce the new IDs and capabilities
 650  * from the old ones and the file's capabilities.
 651  *
 652  * The formula used for evolving capabilities is:
 653  *
 654  *       pI' = pI
 655  * (***) pP' = fP | (fI & pI)
 656  *       pE' = pP' & fE          [NB. fE is 0 or ~0]
 657  *
 658  * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
 659  * ' indicates post-exec().
 660  */
 661
 662 void compute_creds(struct linux_binprm *bprm)
 663 {
 664         int new_permitted = cap_t(bprm->cap_permitted) |
 665                 (cap_t(bprm->cap_inheritable) &
 666                  cap_t(current->cap_inheritable));
 667
 668         /* For init, we want to retain the capabilities set
 669          * in the init_task struct. Thus we skip the usual
 670          * capability rules */
 671         if (current->pid != 1) {
 672                 cap_t(current->cap_permitted) = new_permitted;
 673                 cap_t(current->cap_effective) = new_permitted &
 674                                                 cap_t(bprm->cap_effective);
 675         }
 676
 677         /* AUD: Audit candidate if current->cap_effective is set */
 678
 679         current->suid = current->euid = current->fsuid = bprm->e_uid;
 680         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 681         if (current->euid != current->uid || current->egid != current->gid ||
 682             !cap_issubset(new_permitted, current->cap_permitted))
 683                 current->dumpable = 0;
 684 }
 685
 686
 687 void remove_arg_zero(struct linux_binprm *bprm)
 688 {
 689         if (bprm->argc) {
 690                 unsigned long offset;
 691                 char * page;
 692                 offset = bprm->p % PAGE_SIZE;
 693                 page = (char*)bprm->page[bprm->p/PAGE_SIZE];
 694                 while(bprm->p++,*(page+offset++))
 695                         if(offset==PAGE_SIZE){
 696                                 offset=0;
 697                                 page = (char*)bprm->page[bprm->p/PAGE_SIZE];
 698                         }
 699                 bprm->argc--;
 700         }
 701 }
 702
 703 /*
 704  * cycle the list of binary formats handler, until one recognizes the image
 705  */
 706 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 707 {
 708         int try,retval=0;
 709         struct linux_binfmt *fmt;
 710 #ifdef __alpha__
 711         /* handle /sbin/loader.. */
 712         {
 713             struct exec * eh = (struct exec *) bprm->buf;
 714             struct linux_binprm bprm_loader;
 715
 716             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 717                 (eh->fh.f_flags & 0x3000) == 0x3000)
 718             {
 719                 int i;
 720                 char * dynloader[] = { "/sbin/loader" };
 721                 struct dentry * dentry;
 722
 723                 dput(bprm->dentry);
 724                 bprm->dentry = NULL;
 725
 726                 bprm_loader.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 727                 for (i=0 ; i<MAX_ARG_PAGES ; i++)       /* clear page-table */
 728                     bprm_loader.page[i] = 0;
 729
 730                 dentry = open_namei(dynloader[0], 0, 0);
 731                 retval = PTR_ERR(dentry);
 732                 if (IS_ERR(dentry))
 733                         return retval;
 734                 bprm->dentry = dentry;
 735                 bprm->loader = bprm_loader.p;
 736                 retval = prepare_binprm(bprm);
 737                 if (retval<0)
 738                         return retval;
 739                 /* should call search_binary_handler recursively here,
 740                    but it does not matter */
 741             }
 742         }
 743 #endif
 744         for (try=0; try<2; try++) {
 745                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 746                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
 747                         if (!fn)
 748                                 continue;
 749                         retval = fn(bprm, regs);
 750                         if (retval >= 0) {
 751                                 if (bprm->dentry)
 752                                         dput(bprm->dentry);
 753                                 bprm->dentry = NULL;
 754                                 current->did_exec = 1;
 755                                 return retval;
 756                         }
 757                         if (retval != -ENOEXEC)
 758                                 break;
 759                         if (!bprm->dentry) /* We don't have the dentry anymore */
 760                                 return retval;
 761                 }
 762                 if (retval != -ENOEXEC) {
 763                         break;
 764 #ifdef CONFIG_KMOD
 765                 }else{
 766 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
 767                         char modname[20];
 768                         if (printable(bprm->buf[0]) &&
 769                             printable(bprm->buf[1]) &&
 770                             printable(bprm->buf[2]) &&
 771                             printable(bprm->buf[3]))
 772                                 break; /* -ENOEXEC */
 773                         sprintf(modname, "binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
 774                         request_module(modname);
 775 #endif
 776                 }
 777         }
 778         return retval;
 779 }
 780
 781
 782 /*
 783  * sys_execve() executes a new program.
 784  */
 785 int do_execve(char * filename, char ** argv, char ** envp, struct pt_regs * regs)
 786 {
 787         struct linux_binprm bprm;
 788         struct dentry * dentry;
 789         int retval;
 790         int i;
 791
 792         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 793         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
 794
 795         dentry = open_namei(filename, 0, 0);
 796         retval = PTR_ERR(dentry);
 797         if (IS_ERR(dentry))
 798                 return retval;
 799
 800         bprm.dentry = dentry;
 801         bprm.filename = filename;
 802         bprm.sh_bang = 0;
 803         bprm.loader = 0;
 804         bprm.exec = 0;
 805         if ((bprm.argc = count(argv)) < 0) {
 806                 dput(dentry);
 807                 return bprm.argc;
 808         }
 809
 810         if ((bprm.envc = count(envp)) < 0) {
 811                 dput(dentry);
 812                 return bprm.envc;
 813         }
 814
 815         retval = prepare_binprm(&bprm);
 816         if (retval < 0)
 817                 goto out;
 818
 819         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
 820         if (retval < 0)
 821                 goto out;
 822
 823         bprm.exec = bprm.p;
 824         retval = copy_strings(bprm.envc, envp, &bprm);
 825         if (retval < 0)
 826                 goto out;
 827
 828         retval = copy_strings(bprm.argc, argv, &bprm);
 829         if (retval < 0)
 830                 goto out;
 831
 832         retval = search_binary_handler(&bprm,regs);
 833         if (retval >= 0)
 834                 /* execve success */
 835                 return retval;
 836
 837 out:
 838         /* Something went wrong, return the inode and free the argument pages*/
 839         if (bprm.dentry)
 840                 dput(bprm.dentry);
 841
 842         /* Assumes that free_page() can take a NULL argument. */
 843         /* I hope this is ok for all architectures */
 844         for (i=0 ; i<MAX_ARG_PAGES ; i++)
 845                 free_page(bprm.page[i]);
 846
 847         return retval;
 848 }