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