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_JAVA
  76         init_java_binfmt();
  77 #endif
  78
  79 #ifdef CONFIG_BINFMT_EM86
  80         init_em86_binfmt();
  81 #endif
  82
  83         /* This cannot be configured out of the kernel */
  84         init_script_binfmt();
  85 }
  86
  87 int register_binfmt(struct linux_binfmt * fmt)
  88 {
  89         struct linux_binfmt ** tmp = &formats;
  90
  91         if (!fmt)
  92                 return -EINVAL;
  93         if (fmt->next)
  94                 return -EBUSY;
  95         while (*tmp) {
  96                 if (fmt == *tmp)
  97                         return -EBUSY;
  98                 tmp = &(*tmp)->next;
  99         }
 100         fmt->next = formats;
 101         formats = fmt;
 102         return 0;
 103 }
 104
 105 #ifdef CONFIG_MODULES
 106 int unregister_binfmt(struct linux_binfmt * fmt)
 107 {
 108         struct linux_binfmt ** tmp = &formats;
 109
 110         while (*tmp) {
 111                 if (fmt == *tmp) {
 112                         *tmp = fmt->next;
 113                         return 0;
 114                 }
 115                 tmp = &(*tmp)->next;
 116         }
 117         return -EINVAL;
 118 }
 119 #endif  /* CONFIG_MODULES */
 120
 121 /* N.B. Error returns must be < 0 */
 122 int open_dentry(struct dentry * dentry, int mode)
 123 {
 124         struct inode * inode = dentry->d_inode;
 125         struct file * f;
 126         int fd, error;
 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                 fd_install(fd, f);
 149                 dget(dentry);
 150         }
 151         return fd;
 152
 153 out_filp:
 154         if (error > 0)
 155                 error = -EIO;
 156         put_filp(f);
 157 out_fd:
 158         put_unused_fd(fd);
 159 out:
 160         return error;
 161 }
 162
 163 /*
 164  * Note that a shared library must be both readable and executable due to
 165  * security reasons.
 166  *
 167  * Also note that we take the address to load from from the file itself.
 168  */
 169 asmlinkage int sys_uselib(const char * library)
 170 {
 171         int fd, retval;
 172         struct file * file;
 173         struct linux_binfmt * fmt;
 174
 175         lock_kernel();
 176         fd = sys_open(library, 0, 0);
 177         retval = fd;
 178         if (fd < 0)
 179                 goto out;
 180         file = fget(fd);
 181         retval = -ENOEXEC;
 182         if (file && file->f_dentry && file->f_op && file->f_op->read) {
 183                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 184                         int (*fn)(int) = fmt->load_shlib;
 185                         if (!fn)
 186                                 continue;
 187                         /* N.B. Should use file instead of fd */
 188                         retval = fn(fd);
 189                         if (retval != -ENOEXEC)
 190                                 break;
 191                 }
 192         }
 193         fput(file);
 194         sys_close(fd);
 195 out:
 196         unlock_kernel();
 197         return retval;
 198 }
 199
 200 /*
 201  * count() counts the number of arguments/envelopes
 202  */
 203 static int count(char ** argv)
 204 {
 205         int i = 0;
 206
 207         if (argv != NULL) {
 208                 for (;;) {
 209                         char * p;
 210                         int error;
 211
 212                         error = get_user(p,argv);
 213                         if (error)
 214                                 return error;
 215                         if (!p)
 216                                 break;
 217                         argv++;
 218                         i++;
 219                 }
 220         }
 221         return i;
 222 }
 223
 224 /*
 225  * 'copy_string()' copies argument/envelope strings from user
 226  * memory to free pages in kernel mem. These are in a format ready
 227  * to be put directly into the top of new user memory.
 228  *
 229  * Modified by TYT, 11/24/91 to add the from_kmem argument, which specifies
 230  * whether the string and the string array are from user or kernel segments:
 231  *
 232  * from_kmem     argv *        argv **
 233  *    0          user space    user space
 234  *    1          kernel space  user space
 235  *    2          kernel space  kernel space
 236  *
 237  * We do this by playing games with the fs segment register.  Since it
 238  * is expensive to load a segment register, we try to avoid calling
 239  * set_fs() unless we absolutely have to.
 240  */
 241 unsigned long copy_strings(int argc,char ** argv,unsigned long *page,
 242                 unsigned long p, int from_kmem)
 243 {
 244         char *str;
 245         mm_segment_t old_fs;
 246
 247         if (!p)
 248                 return 0;       /* bullet-proofing */
 249         old_fs = get_fs();
 250         if (from_kmem==2)
 251                 set_fs(KERNEL_DS);
 252         while (argc-- > 0) {
 253                 int len;
 254                 unsigned long pos;
 255
 256                 if (from_kmem == 1)
 257                         set_fs(KERNEL_DS);
 258                 get_user(str, argv+argc);
 259                 if (!str)
 260                         panic("VFS: argc is wrong");
 261                 if (from_kmem == 1)
 262                         set_fs(old_fs);
 263                 len = strlen_user(str); /* includes the '\0' */
 264                 if (p < len) {  /* this shouldn't happen - 128kB */
 265                         set_fs(old_fs);
 266                         return 0;
 267                 }
 268                 p -= len;
 269                 pos = p;
 270                 while (len) {
 271                         char *pag;
 272                         int offset, bytes_to_copy;
 273
 274                         offset = pos % PAGE_SIZE;
 275                         if (!(pag = (char *) page[pos/PAGE_SIZE]) &&
 276                             !(pag = (char *) page[pos/PAGE_SIZE] =
 277                               (unsigned long *) get_free_page(GFP_USER))) {
 278                                 if (from_kmem==2)
 279                                         set_fs(old_fs);
 280                                 return 0;
 281                         }
 282                         bytes_to_copy = PAGE_SIZE - offset;
 283                         if (bytes_to_copy > len)
 284                                 bytes_to_copy = len;
 285                         copy_from_user(pag + offset, str, bytes_to_copy);
 286                         pos += bytes_to_copy;
 287                         str += bytes_to_copy;
 288                         len -= bytes_to_copy;
 289                 }
 290         }
 291         if (from_kmem==2)
 292                 set_fs(old_fs);
 293         return p;
 294 }
 295
 296 unsigned long setup_arg_pages(unsigned long p, struct linux_binprm * bprm)
 297 {
 298         unsigned long stack_base;
 299         struct vm_area_struct *mpnt;
 300         int i;
 301
 302         stack_base = STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE;
 303
 304         p += stack_base;
 305         if (bprm->loader)
 306                 bprm->loader += stack_base;
 307         bprm->exec += stack_base;
 308
 309         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 310         if (mpnt) {
 311                 mpnt->vm_mm = current->mm;
 312                 mpnt->vm_start = PAGE_MASK & (unsigned long) p;
 313                 mpnt->vm_end = STACK_TOP;
 314                 mpnt->vm_page_prot = PAGE_COPY;
 315                 mpnt->vm_flags = VM_STACK_FLAGS;
 316                 mpnt->vm_ops = NULL;
 317                 mpnt->vm_offset = 0;
 318                 mpnt->vm_file = NULL;
 319                 mpnt->vm_pte = 0;
 320                 insert_vm_struct(current->mm, mpnt);
 321                 current->mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 322         }
 323
 324         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 325                 if (bprm->page[i]) {
 326                         current->mm->rss++;
 327                         put_dirty_page(current,bprm->page[i],stack_base);
 328                 }
 329                 stack_base += PAGE_SIZE;
 330         }
 331         return p;
 332 }
 333
 334 /*
 335  * Read in the complete executable. This is used for "-N" files
 336  * that aren't on a block boundary, and for files on filesystems
 337  * without bmap support.
 338  */
 339 int read_exec(struct dentry *dentry, unsigned long offset,
 340         char * addr, unsigned long count, int to_kmem)
 341 {
 342         struct file file;
 343         struct inode * inode = dentry->d_inode;
 344         int result = -ENOEXEC;
 345
 346         if (!inode->i_op || !inode->i_op->default_file_ops)
 347                 goto end_readexec;
 348         if (init_private_file(&file, dentry, 1))
 349                 goto end_readexec;
 350         if (!file.f_op->read)
 351                 goto close_readexec;
 352         if (file.f_op->llseek) {
 353                 if (file.f_op->llseek(&file,offset,0) != offset)
 354                         goto close_readexec;
 355         } else
 356                 file.f_pos = offset;
 357         if (to_kmem) {
 358                 mm_segment_t old_fs = get_fs();
 359                 set_fs(get_ds());
 360                 result = file.f_op->read(&file, addr, count, &file.f_pos);
 361                 set_fs(old_fs);
 362         } else {
 363                 result = verify_area(VERIFY_WRITE, addr, count);
 364                 if (result)
 365                         goto close_readexec;
 366                 result = file.f_op->read(&file, addr, count, &file.f_pos);
 367         }
 368 close_readexec:
 369         if (file.f_op->release)
 370                 file.f_op->release(inode,&file);
 371 end_readexec:
 372         return result;
 373 }
 374
 375 static int exec_mmap(void)
 376 {
 377         struct mm_struct * mm, * old_mm;
 378         int retval, nr;
 379
 380         if (atomic_read(&current->mm->count) == 1) {
 381                 flush_cache_mm(current->mm);
 382                 mm_release();
 383                 release_segments(current->mm);
 384                 exit_mmap(current->mm);
 385                 flush_tlb_mm(current->mm);
 386                 return 0;
 387         }
 388
 389         retval = -ENOMEM;
 390         mm = mm_alloc();
 391         if (!mm)
 392                 goto fail_nomem;
 393
 394         mm->cpu_vm_mask = (1UL << smp_processor_id());
 395         mm->total_vm = 0;
 396         mm->rss = 0;
 397         /*
 398          * Make sure we have a private ldt if needed ...
 399          */
 400         nr = current->tarray_ptr - &task[0];
 401         copy_segments(nr, current, mm);
 402
 403         old_mm = current->mm;
 404         current->mm = mm;
 405         retval = new_page_tables(current);
 406         if (retval)
 407                 goto fail_restore;
 408         activate_context(current);
 409         up(&mm->mmap_sem);
 410         mm_release();
 411         mmput(old_mm);
 412         return 0;
 413
 414         /*
 415          * Failure ... restore the prior mm_struct.
 416          */
 417 fail_restore:
 418         /* The pgd belongs to the parent ... don't free it! */
 419         mm->pgd = NULL;
 420         current->mm = old_mm;
 421         /* restore the ldt for this task */
 422         copy_segments(nr, current, NULL);
 423         mmput(mm);
 424
 425 fail_nomem:
 426         return retval;
 427 }
 428
 429 /*
 430  * This function makes sure the current process has its own signal table,
 431  * so that flush_signal_handlers can later reset the handlers without
 432  * disturbing other processes.  (Other processes might share the signal
 433  * table via the CLONE_SIGHAND option to clone().)
 434  */
 435
 436 static inline int make_private_signals(void)
 437 {
 438         struct signal_struct * newsig;
 439
 440         if (atomic_read(&current->sig->count) <= 1)
 441                 return 0;
 442         newsig = kmalloc(sizeof(*newsig), GFP_KERNEL);
 443         if (newsig == NULL)
 444                 return -ENOMEM;
 445         spin_lock_init(&newsig->siglock);
 446         atomic_set(&newsig->count, 1);
 447         memcpy(newsig->action, current->sig->action, sizeof(newsig->action));
 448         current->sig = newsig;
 449         return 0;
 450 }
 451
 452 /*
 453  * If make_private_signals() made a copy of the signal table, decrement the
 454  * refcount of the original table, and free it if necessary.
 455  * We don't do that in make_private_signals() so that we can back off
 456  * in flush_old_exec() if an error occurs after calling make_private_signals().
 457  */
 458
 459 static inline void release_old_signals(struct signal_struct * oldsig)
 460 {
 461         if (current->sig == oldsig)
 462                 return;
 463         if (atomic_dec_and_test(&oldsig->count))
 464                 kfree(oldsig);
 465 }
 466
 467 /*
 468  * These functions flushes out all traces of the currently running executable
 469  * so that a new one can be started
 470  */
 471
 472 static inline void flush_old_files(struct files_struct * files)
 473 {
 474         unsigned long j;
 475
 476         j = 0;
 477         for (;;) {
 478                 unsigned long set, i;
 479
 480                 i = j * __NFDBITS;
 481                 if (i >= files->max_fds)
 482                         break;
 483                 set = files->close_on_exec.fds_bits[j];
 484                 files->close_on_exec.fds_bits[j] = 0;
 485                 j++;
 486                 for ( ; set ; i++,set >>= 1) {
 487                         if (set & 1)
 488                                 sys_close(i);
 489                 }
 490         }
 491 }
 492
 493 int flush_old_exec(struct linux_binprm * bprm)
 494 {
 495         char * name;
 496         int i, ch, retval;
 497         struct signal_struct * oldsig;
 498
 499         /*
 500          * Make sure we have a private signal table
 501          */
 502         oldsig = current->sig;
 503         retval = make_private_signals();
 504         if (retval) goto flush_failed;
 505
 506         /*
 507          * Release all of the old mmap stuff
 508          */
 509         retval = exec_mmap();
 510         if (retval) goto mmap_failed;
 511
 512         /* This is the point of no return */
 513         release_old_signals(oldsig);
 514
 515         if (current->euid == current->uid && current->egid == current->gid)
 516                 current->dumpable = 1;
 517         name = bprm->filename;
 518         for (i=0; (ch = *(name++)) != '\0';) {
 519                 if (ch == '/')
 520                         i = 0;
 521                 else
 522                         if (i < 15)
 523                                 current->comm[i++] = ch;
 524         }
 525         current->comm[i] = '\0';
 526
 527         flush_thread();
 528
 529         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 530             permission(bprm->dentry->d_inode,MAY_READ))
 531                 current->dumpable = 0;
 532
 533         flush_signal_handlers(current);
 534         flush_old_files(current->files);
 535
 536         return 0;
 537
 538 mmap_failed:
 539         if (current->sig != oldsig)
 540                 kfree(current->sig);
 541 flush_failed:
 542         current->sig = oldsig;
 543         return retval;
 544 }
 545
 546 /*
 547  * We mustn't allow tracing of suid binaries, unless
 548  * the tracer has the capability to trace anything..
 549  */
 550 static inline int must_not_trace_exec(struct task_struct * p)
 551 {
 552         return (p->flags & PF_PTRACED) && !cap_raised(p->p_pptr->cap_effective, CAP_SYS_PTRACE);
 553 }
 554
 555 /*
 556  * Fill the binprm structure from the inode.
 557  * Check permissions, then read the first 512 bytes
 558  */
 559 int prepare_binprm(struct linux_binprm *bprm)
 560 {
 561         int mode;
 562         int retval,id_change,cap_raised;
 563         struct inode * inode = bprm->dentry->d_inode;
 564
 565         mode = inode->i_mode;
 566         if (!S_ISREG(mode))                     /* must be regular file */
 567                 return -EACCES;
 568         if (!(mode & 0111))                     /* with at least _one_ execute bit set */
 569                 return -EACCES;
 570         if (IS_NOEXEC(inode))                   /* FS mustn't be mounted noexec */
 571                 return -EACCES;
 572         if (!inode->i_sb)
 573                 return -EACCES;
 574         if ((retval = permission(inode, MAY_EXEC)) != 0)
 575                 return retval;
 576         /* better not execute files which are being written to */
 577         if (inode->i_writecount > 0)
 578                 return -ETXTBSY;
 579
 580         bprm->e_uid = current->euid;
 581         bprm->e_gid = current->egid;
 582         id_change = cap_raised = 0;
 583
 584         /* Set-uid? */
 585         if (mode & S_ISUID) {
 586                 bprm->e_uid = inode->i_uid;
 587                 if (bprm->e_uid != current->euid)
 588                         id_change = 1;
 589         }
 590
 591         /* Set-gid? */
 592         /*
 593          * If setgid is set but no group execute bit then this
 594          * is a candidate for mandatory locking, not a setgid
 595          * executable.
 596          */
 597         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 598                 bprm->e_gid = inode->i_gid;
 599                 if (!in_group_p(bprm->e_gid))
 600                         id_change = 1;
 601         }
 602
 603         /* We don't have VFS support for capabilities yet */
 604         cap_clear(bprm->cap_inheritable);
 605         cap_clear(bprm->cap_permitted);
 606         cap_clear(bprm->cap_effective);
 607
 608         /*  To support inheritance of root-permissions and suid-root
 609          *  executables under compatibility mode, we raise the
 610          *  effective and inherited bitmasks of the executable file
 611          *  (translation: we set the executable "capability dumb" and
 612          *  set the allowed set to maximum). We don't set any forced
 613          *  bits.
 614          *
 615          *  If only the real uid is 0, we only raise the inheritable
 616          *  bitmask of the executable file (translation: we set the
 617          *  allowed set to maximum and the application to "capability
 618          *  smart").
 619          */
 620
 621         if (!issecure(SECURE_NOROOT)) {
 622                 if (bprm->e_uid == 0 || current->uid == 0)
 623                         cap_set_full(bprm->cap_inheritable);
 624                 if (bprm->e_uid == 0)
 625                         cap_set_full(bprm->cap_effective);
 626         }
 627
 628         /* Only if pP' is _not_ a subset of pP, do we consider there
 629          * has been a capability related "change of capability".  In
 630          * such cases, we need to check that the elevation of
 631          * privilege does not go against other system constraints.
 632          * The new Permitted set is defined below -- see (***). */
 633         {
 634                 kernel_cap_t working =
 635                         cap_combine(bprm->cap_permitted,
 636                                     cap_intersect(bprm->cap_inheritable,
 637                                                   current->cap_inheritable));
 638                 if (!cap_issubset(working, current->cap_permitted)) {
 639                         cap_raised = 1;
 640                 }
 641         }
 642
 643         if (id_change || cap_raised) {
 644                 /* We can't suid-execute if we're sharing parts of the executable */
 645                 /* or if we're being traced (or if suid execs are not allowed)    */
 646                 /* (current->mm->count > 1 is ok, as we'll get a new mm anyway)   */
 647                 if (IS_NOSUID(inode)
 648                     || must_not_trace_exec(current)
 649                     || (atomic_read(&current->fs->count) > 1)
 650                     || (atomic_read(&current->sig->count) > 1)
 651                     || (atomic_read(&current->files->count) > 1)) {
 652                         if (id_change && !capable(CAP_SETUID))
 653                                 return -EPERM;
 654                         if (cap_raised && !capable(CAP_SETPCAP))
 655                                 return -EPERM;
 656                 }
 657         }
 658
 659         memset(bprm->buf,0,sizeof(bprm->buf));
 660         return read_exec(bprm->dentry,0,bprm->buf,128,1);
 661 }
 662
 663 /*
 664  * This function is used to produce the new IDs and capabilities
 665  * from the old ones and the file's capabilities.
 666  *
 667  * The formula used for evolving capabilities is:
 668  *
 669  *       pI' = pI
 670  * (***) pP' = fP | (fI & pI)
 671  *       pE' = pP' & fE          [NB. fE is 0 or ~0]
 672  *
 673  * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
 674  * ' indicates post-exec().
 675  */
 676
 677 void compute_creds(struct linux_binprm *bprm)
 678 {
 679         int new_permitted = cap_t(bprm->cap_permitted) |
 680                 (cap_t(bprm->cap_inheritable) &
 681                  cap_t(current->cap_inheritable));
 682
 683         /* For init, we want to retain the capabilities set
 684          * in the init_task struct. Thus we skip the usual
 685          * capability rules */
 686         if (current->pid != 1) {
 687                 cap_t(current->cap_permitted) = new_permitted;
 688                 cap_t(current->cap_effective) = new_permitted &
 689                                                 cap_t(bprm->cap_effective);
 690         }
 691
 692         /* AUD: Audit candidate if current->cap_effective is set */
 693
 694         current->suid = current->euid = current->fsuid = bprm->e_uid;
 695         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 696         if (current->euid != current->uid || current->egid != current->gid ||
 697             !cap_issubset(new_permitted, current->cap_permitted))
 698                 current->dumpable = 0;
 699 }
 700
 701
 702 void remove_arg_zero(struct linux_binprm *bprm)
 703 {
 704         if (bprm->argc) {
 705                 unsigned long offset;
 706                 char * page;
 707                 offset = bprm->p % PAGE_SIZE;
 708                 page = (char*)bprm->page[bprm->p/PAGE_SIZE];
 709                 while(bprm->p++,*(page+offset++))
 710                         if(offset==PAGE_SIZE){
 711                                 offset=0;
 712                                 page = (char*)bprm->page[bprm->p/PAGE_SIZE];
 713                         }
 714                 bprm->argc--;
 715         }
 716 }
 717
 718 /*
 719  * cycle the list of binary formats handler, until one recognizes the image
 720  */
 721 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 722 {
 723         int try,retval=0;
 724         struct linux_binfmt *fmt;
 725 #ifdef __alpha__
 726         /* handle /sbin/loader.. */
 727         {
 728             struct exec * eh = (struct exec *) bprm->buf;
 729             struct linux_binprm bprm_loader;
 730
 731             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 732                 (eh->fh.f_flags & 0x3000) == 0x3000)
 733             {
 734                 int i;
 735                 char * dynloader[] = { "/sbin/loader" };
 736                 struct dentry * dentry;
 737
 738                 dput(bprm->dentry);
 739                 bprm->dentry = NULL;
 740
 741                 bprm_loader.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 742                 for (i=0 ; i<MAX_ARG_PAGES ; i++)       /* clear page-table */
 743                     bprm_loader.page[i] = 0;
 744
 745                 dentry = open_namei(dynloader[0], 0, 0);
 746                 retval = PTR_ERR(dentry);
 747                 if (IS_ERR(dentry))
 748                         return retval;
 749                 bprm->dentry = dentry;
 750                 bprm->loader = bprm_loader.p;
 751                 retval = prepare_binprm(bprm);
 752                 if (retval<0)
 753                         return retval;
 754                 /* should call search_binary_handler recursively here,
 755                    but it does not matter */
 756             }
 757         }
 758 #endif
 759         for (try=0; try<2; try++) {
 760                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 761                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
 762                         if (!fn)
 763                                 continue;
 764                         retval = fn(bprm, regs);
 765                         if (retval >= 0) {
 766                                 if (bprm->dentry)
 767                                         dput(bprm->dentry);
 768                                 bprm->dentry = NULL;
 769                                 current->did_exec = 1;
 770                                 return retval;
 771                         }
 772                         if (retval != -ENOEXEC)
 773                                 break;
 774                         if (!bprm->dentry) /* We don't have the dentry anymore */
 775                                 return retval;
 776                 }
 777                 if (retval != -ENOEXEC) {
 778                         break;
 779 #ifdef CONFIG_KMOD
 780                 }else{
 781 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
 782                         char modname[20];
 783                         if (printable(bprm->buf[0]) &&
 784                             printable(bprm->buf[1]) &&
 785                             printable(bprm->buf[2]) &&
 786                             printable(bprm->buf[3]))
 787                                 break; /* -ENOEXEC */
 788                         sprintf(modname, "binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
 789                         request_module(modname);
 790 #endif
 791                 }
 792         }
 793         return retval;
 794 }
 795
 796
 797 /*
 798  * sys_execve() executes a new program.
 799  */
 800 int do_execve(char * filename, char ** argv, char ** envp, struct pt_regs * regs)
 801 {
 802         struct linux_binprm bprm;
 803         struct dentry * dentry;
 804         int retval;
 805         int i;
 806
 807         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 808         for (i=0 ; i<MAX_ARG_PAGES ; i++)       /* clear page-table */
 809                 bprm.page[i] = 0;
 810
 811         dentry = open_namei(filename, 0, 0);
 812         retval = PTR_ERR(dentry);
 813         if (IS_ERR(dentry))
 814                 return retval;
 815
 816         bprm.dentry = dentry;
 817         bprm.filename = filename;
 818         bprm.sh_bang = 0;
 819         bprm.java = 0;
 820         bprm.loader = 0;
 821         bprm.exec = 0;
 822         if ((bprm.argc = count(argv)) < 0) {
 823                 dput(dentry);
 824                 return bprm.argc;
 825         }
 826
 827         if ((bprm.envc = count(envp)) < 0) {
 828                 dput(dentry);
 829                 return bprm.envc;
 830         }
 831
 832         retval = prepare_binprm(&bprm);
 833
 834         if (retval >= 0) {
 835                 bprm.p = copy_strings(1, &bprm.filename, bprm.page, bprm.p, 2);
 836                 bprm.exec = bprm.p;
 837                 bprm.p = copy_strings(bprm.envc,envp,bprm.page,bprm.p,0);
 838                 bprm.p = copy_strings(bprm.argc,argv,bprm.page,bprm.p,0);
 839                 if (!bprm.p)
 840                         retval = -E2BIG;
 841         }
 842
 843         if (retval >= 0)
 844                 retval = search_binary_handler(&bprm,regs);
 845         if (retval >= 0)
 846                 /* execve success */
 847                 return retval;
 848
 849         /* Something went wrong, return the inode and free the argument pages*/
 850         if (bprm.dentry)
 851                 dput(bprm.dentry);
 852
 853         for (i=0 ; i<MAX_ARG_PAGES ; i++)
 854                 free_page(bprm.page[i]);
 855
 856         return retval;
 857 }