fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/config.h>
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/mman.h>
  29 #include <linux/a.out.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #define __NO_VERSION__
  38 #include <linux/module.h>
  39
  40 #include <asm/uaccess.h>
  41 #include <asm/pgalloc.h>
  42 #include <asm/mmu_context.h>
  43
  44 #ifdef CONFIG_KMOD
  45 #include <linux/kmod.h>
  46 #endif
  47
  48 static struct linux_binfmt *formats;
  49 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  50
  51 int register_binfmt(struct linux_binfmt * fmt)
  52 {
  53         struct linux_binfmt ** tmp = &formats;
  54
  55         if (!fmt)
  56                 return -EINVAL;
  57         if (fmt->next)
  58                 return -EBUSY;
  59         write_lock(&binfmt_lock);
  60         while (*tmp) {
  61                 if (fmt == *tmp) {
  62                         write_unlock(&binfmt_lock);
  63                         return -EBUSY;
  64                 }
  65                 tmp = &(*tmp)->next;
  66         }
  67         fmt->next = formats;
  68         formats = fmt;
  69         write_unlock(&binfmt_lock);
  70         return 0;
  71 }
  72
  73 int unregister_binfmt(struct linux_binfmt * fmt)
  74 {
  75         struct linux_binfmt ** tmp = &formats;
  76
  77         write_lock(&binfmt_lock);
  78         while (*tmp) {
  79                 if (fmt == *tmp) {
  80                         *tmp = fmt->next;
  81                         write_unlock(&binfmt_lock);
  82                         return 0;
  83                 }
  84                 tmp = &(*tmp)->next;
  85         }
  86         write_unlock(&binfmt_lock);
  87         return -EINVAL;
  88 }
  89
  90 static inline void put_binfmt(struct linux_binfmt * fmt)
  91 {
  92         if (fmt->module)
  93                 __MOD_DEC_USE_COUNT(fmt->module);
  94 }
  95
  96 /*
  97  * Note that a shared library must be both readable and executable due to
  98  * security reasons.
  99  *
 100  * Also note that we take the address to load from from the file itself.
 101  */
 102 asmlinkage long sys_uselib(const char * library)
 103 {
 104         struct file * file;
 105         struct nameidata nd;
 106         int error;
 107
 108         error = user_path_walk(library, &nd);
 109         if (error)
 110                 goto out;
 111
 112         error = -EINVAL;
 113         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 114                 goto exit;
 115
 116         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC);
 117         if (error)
 118                 goto exit;
 119
 120         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 121         error = PTR_ERR(file);
 122         if (IS_ERR(file))
 123                 goto out;
 124
 125         error = -ENOEXEC;
 126         if(file->f_op && file->f_op->read) {
 127                 struct linux_binfmt * fmt;
 128
 129                 read_lock(&binfmt_lock);
 130                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 131                         if (!fmt->load_shlib)
 132                                 continue;
 133                         if (!try_inc_mod_count(fmt->module))
 134                                 continue;
 135                         read_unlock(&binfmt_lock);
 136                         error = fmt->load_shlib(file);
 137                         read_lock(&binfmt_lock);
 138                         put_binfmt(fmt);
 139                         if (error != -ENOEXEC)
 140                                 break;
 141                 }
 142                 read_unlock(&binfmt_lock);
 143         }
 144         fput(file);
 145 out:
 146         return error;
 147 exit:
 148         path_release(&nd);
 149         goto out;
 150 }
 151
 152 /*
 153  * count() counts the number of arguments/envelopes
 154  */
 155 static int count(char ** argv, int max)
 156 {
 157         int i = 0;
 158
 159         if (argv != NULL) {
 160                 for (;;) {
 161                         char * p;
 162                         int error;
 163
 164                         error = get_user(p,argv);
 165                         if (error)
 166                                 return error;
 167                         if (!p)
 168                                 break;
 169                         argv++;
 170                         if(++i > max)
 171                                 return -E2BIG;
 172                 }
 173         }
 174         return i;
 175 }
 176
 177 /*
 178  * 'copy_strings()' copies argument/envelope strings from user
 179  * memory to free pages in kernel mem. These are in a format ready
 180  * to be put directly into the top of new user memory.
 181  */
 182 int copy_strings(int argc,char ** argv, struct linux_binprm *bprm)
 183 {
 184         while (argc-- > 0) {
 185                 char *str;
 186                 int len;
 187                 unsigned long pos;
 188
 189                 if (get_user(str, argv+argc) || !str || !(len = strnlen_user(str, bprm->p)))
 190                         return -EFAULT;
 191                 if (bprm->p < len)
 192                         return -E2BIG;
 193
 194                 bprm->p -= len;
 195                 /* XXX: add architecture specific overflow check here. */
 196
 197                 pos = bprm->p;
 198                 while (len > 0) {
 199                         char *kaddr;
 200                         int i, new, err;
 201                         struct page *page;
 202                         int offset, bytes_to_copy;
 203
 204                         offset = pos % PAGE_SIZE;
 205                         i = pos/PAGE_SIZE;
 206                         page = bprm->page[i];
 207                         new = 0;
 208                         if (!page) {
 209                                 page = alloc_page(GFP_HIGHUSER);
 210                                 bprm->page[i] = page;
 211                                 if (!page)
 212                                         return -ENOMEM;
 213                                 new = 1;
 214                         }
 215                         kaddr = (char *)kmap(page);
 216
 217                         if (new && offset)
 218                                 memset(kaddr, 0, offset);
 219                         bytes_to_copy = PAGE_SIZE - offset;
 220                         if (bytes_to_copy > len) {
 221                                 bytes_to_copy = len;
 222                                 if (new)
 223                                         memset(kaddr+offset+len, 0, PAGE_SIZE-offset-len);
 224                         }
 225                         err = copy_from_user(kaddr + offset, str, bytes_to_copy);
 226                         flush_dcache_page(page);
 227                         flush_page_to_ram(page);
 228                         kunmap(page);
 229
 230                         if (err)
 231                                 return -EFAULT;
 232
 233                         pos += bytes_to_copy;
 234                         str += bytes_to_copy;
 235                         len -= bytes_to_copy;
 236                 }
 237         }
 238         return 0;
 239 }
 240
 241 /*
 242  * Like copy_strings, but get argv and its values from kernel memory.
 243  */
 244 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 245 {
 246         int r;
 247         mm_segment_t oldfs = get_fs();
 248         set_fs(KERNEL_DS);
 249         r = copy_strings(argc, argv, bprm);
 250         set_fs(oldfs);
 251         return r;
 252 }
 253
 254 /*
 255  * This routine is used to map in a page into an address space: needed by
 256  * execve() for the initial stack and environment pages.
 257  */
 258 void put_dirty_page(struct task_struct * tsk, struct page *page, unsigned long address)
 259 {
 260         pgd_t * pgd;
 261         pmd_t * pmd;
 262         pte_t * pte;
 263
 264         if (page_count(page) != 1)
 265                 printk("mem_map disagrees with %p at %08lx\n", page, address);
 266         pgd = pgd_offset(tsk->mm, address);
 267         pmd = pmd_alloc(pgd, address);
 268         if (!pmd) {
 269                 __free_page(page);
 270                 force_sig(SIGKILL, tsk);
 271                 return;
 272         }
 273         pte = pte_alloc(pmd, address);
 274         if (!pte) {
 275                 __free_page(page);
 276                 force_sig(SIGKILL, tsk);
 277                 return;
 278         }
 279         if (!pte_none(*pte)) {
 280                 pte_ERROR(*pte);
 281                 __free_page(page);
 282                 return;
 283         }
 284         flush_page_to_ram(page);
 285         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(page, PAGE_COPY))));
 286 /* no need for flush_tlb */
 287 }
 288
 289 int setup_arg_pages(struct linux_binprm *bprm)
 290 {
 291         unsigned long stack_base;
 292         struct vm_area_struct *mpnt;
 293         int i;
 294
 295         stack_base = STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE;
 296
 297         bprm->p += stack_base;
 298         if (bprm->loader)
 299                 bprm->loader += stack_base;
 300         bprm->exec += stack_base;
 301
 302         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 303         if (!mpnt)
 304                 return -ENOMEM;
 305
 306         down(&current->mm->mmap_sem);
 307         {
 308                 mpnt->vm_mm = current->mm;
 309                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 310                 mpnt->vm_end = STACK_TOP;
 311                 mpnt->vm_page_prot = PAGE_COPY;
 312                 mpnt->vm_flags = VM_STACK_FLAGS;
 313                 mpnt->vm_ops = NULL;
 314                 mpnt->vm_pgoff = 0;
 315                 mpnt->vm_file = NULL;
 316                 mpnt->vm_private_data = (void *) 0;
 317                 vmlist_modify_lock(current->mm);
 318                 insert_vm_struct(current->mm, mpnt);
 319                 vmlist_modify_unlock(current->mm);
 320                 current->mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 321         }
 322
 323         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 324                 struct page *page = bprm->page[i];
 325                 if (page) {
 326                         bprm->page[i] = NULL;
 327                         current->mm->rss++;
 328                         put_dirty_page(current,page,stack_base);
 329                 }
 330                 stack_base += PAGE_SIZE;
 331         }
 332         up(&current->mm->mmap_sem);
 333
 334         return 0;
 335 }
 336
 337 struct file *open_exec(const char *name)
 338 {
 339         struct nameidata nd;
 340         struct inode *inode;
 341         struct file *file;
 342         int err = 0;
 343
 344         if (path_init(name, LOOKUP_FOLLOW|LOOKUP_POSITIVE, &nd))
 345                 err = path_walk(name, &nd);
 346         file = ERR_PTR(err);
 347         if (!err) {
 348                 inode = nd.dentry->d_inode;
 349                 file = ERR_PTR(-EACCES);
 350                 if (!IS_NOEXEC(inode) && S_ISREG(inode->i_mode)) {
 351                         int err = permission(inode, MAY_EXEC);
 352                         file = ERR_PTR(err);
 353                         if (!err) {
 354                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 355                                 if (!IS_ERR(file)) {
 356                                         err = deny_write_access(file);
 357                                         if (err) {
 358                                                 fput(file);
 359                                                 file = ERR_PTR(err);
 360                                         }
 361                                 }
 362 out:
 363                                 return file;
 364                         }
 365                 }
 366                 path_release(&nd);
 367         }
 368         goto out;
 369 }
 370
 371 int kernel_read(struct file *file, unsigned long offset,
 372         char * addr, unsigned long count)
 373 {
 374         mm_segment_t old_fs;
 375         loff_t pos = offset;
 376         int result = -ENOSYS;
 377
 378         if (!file->f_op->read)
 379                 goto fail;
 380         old_fs = get_fs();
 381         set_fs(get_ds());
 382         result = file->f_op->read(file, addr, count, &pos);
 383         set_fs(old_fs);
 384 fail:
 385         return result;
 386 }
 387
 388 static int exec_mmap(void)
 389 {
 390         struct mm_struct * mm, * old_mm;
 391
 392         old_mm = current->mm;
 393         if (old_mm && atomic_read(&old_mm->mm_users) == 1) {
 394                 flush_cache_mm(old_mm);
 395                 mm_release();
 396                 exit_mmap(old_mm);
 397                 flush_tlb_mm(old_mm);
 398                 return 0;
 399         }
 400
 401         mm = mm_alloc();
 402         if (mm) {
 403                 struct mm_struct *active_mm = current->active_mm;
 404
 405                 if (init_new_context(current, mm)) {
 406                         mmdrop(mm);
 407                         return -ENOMEM;
 408                 }
 409                 task_lock(current);
 410                 current->mm = mm;
 411                 current->active_mm = mm;
 412                 task_unlock(current);
 413                 activate_mm(active_mm, mm);
 414                 mm_release();
 415                 if (old_mm) {
 416                         if (active_mm != old_mm) BUG();
 417                         mmput(old_mm);
 418                         return 0;
 419                 }
 420                 mmdrop(active_mm);
 421                 return 0;
 422         }
 423         return -ENOMEM;
 424 }
 425
 426 /*
 427  * This function makes sure the current process has its own signal table,
 428  * so that flush_signal_handlers can later reset the handlers without
 429  * disturbing other processes.  (Other processes might share the signal
 430  * table via the CLONE_SIGNAL option to clone().)
 431  */
 432
 433 static inline int make_private_signals(void)
 434 {
 435         struct signal_struct * newsig;
 436
 437         if (atomic_read(&current->sig->count) <= 1)
 438                 return 0;
 439         newsig = kmem_cache_alloc(sigact_cachep, GFP_KERNEL);
 440         if (newsig == NULL)
 441                 return -ENOMEM;
 442         spin_lock_init(&newsig->siglock);
 443         atomic_set(&newsig->count, 1);
 444         memcpy(newsig->action, current->sig->action, sizeof(newsig->action));
 445         spin_lock_irq(&current->sigmask_lock);
 446         current->sig = newsig;
 447         spin_unlock_irq(&current->sigmask_lock);
 448         return 0;
 449 }
 450
 451 /*
 452  * If make_private_signals() made a copy of the signal table, decrement the
 453  * refcount of the original table, and free it if necessary.
 454  * We don't do that in make_private_signals() so that we can back off
 455  * in flush_old_exec() if an error occurs after calling make_private_signals().
 456  */
 457
 458 static inline void release_old_signals(struct signal_struct * oldsig)
 459 {
 460         if (current->sig == oldsig)
 461                 return;
 462         if (atomic_dec_and_test(&oldsig->count))
 463                 kmem_cache_free(sigact_cachep, oldsig);
 464 }
 465
 466 /*
 467  * These functions flushes out all traces of the currently running executable
 468  * so that a new one can be started
 469  */
 470
 471 static inline void flush_old_files(struct files_struct * files)
 472 {
 473         long j = -1;
 474
 475         write_lock(&files->file_lock);
 476         for (;;) {
 477                 unsigned long set, i;
 478
 479                 j++;
 480                 i = j * __NFDBITS;
 481                 if (i >= files->max_fds || i >= files->max_fdset)
 482                         break;
 483                 set = files->close_on_exec->fds_bits[j];
 484                 if (!set)
 485                         continue;
 486                 files->close_on_exec->fds_bits[j] = 0;
 487                 write_unlock(&files->file_lock);
 488                 for ( ; set ; i++,set >>= 1) {
 489                         if (set & 1) {
 490                                 sys_close(i);
 491                         }
 492                 }
 493                 write_lock(&files->file_lock);
 494
 495         }
 496         write_unlock(&files->file_lock);
 497 }
 498
 499 /*
 500  * An execve() will automatically "de-thread" the process.
 501  * Note: we don't have to hold the tasklist_lock to test
 502  * whether we migth need to do this. If we're not part of
 503  * a thread group, there is no way we can become one
 504  * dynamically. And if we are, we only need to protect the
 505  * unlink - even if we race with the last other thread exit,
 506  * at worst the list_del_init() might end up being a no-op.
 507  */
 508 static inline void de_thread(struct task_struct *tsk)
 509 {
 510         if (!list_empty(&tsk->thread_group)) {
 511                 write_lock_irq(&tasklist_lock);
 512                 list_del_init(&tsk->thread_group);
 513                 write_unlock_irq(&tasklist_lock);
 514         }
 515
 516         /* Minor oddity: this might stay the same. */
 517         tsk->tgid = tsk->pid;
 518 }
 519
 520 int flush_old_exec(struct linux_binprm * bprm)
 521 {
 522         char * name;
 523         int i, ch, retval;
 524         struct signal_struct * oldsig;
 525
 526         /*
 527          * Make sure we have a private signal table
 528          */
 529         oldsig = current->sig;
 530         retval = make_private_signals();
 531         if (retval) goto flush_failed;
 532
 533         /*
 534          * Release all of the old mmap stuff
 535          */
 536         retval = exec_mmap();
 537         if (retval) goto mmap_failed;
 538
 539         /* This is the point of no return */
 540         release_old_signals(oldsig);
 541
 542         current->sas_ss_sp = current->sas_ss_size = 0;
 543
 544         if (current->euid == current->uid && current->egid == current->gid)
 545                 current->dumpable = 1;
 546         name = bprm->filename;
 547         for (i=0; (ch = *(name++)) != '\0';) {
 548                 if (ch == '/')
 549                         i = 0;
 550                 else
 551                         if (i < 15)
 552                                 current->comm[i++] = ch;
 553         }
 554         current->comm[i] = '\0';
 555
 556         flush_thread();
 557
 558         de_thread(current);
 559
 560         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 561             permission(bprm->file->f_dentry->d_inode,MAY_READ))
 562                 current->dumpable = 0;
 563
 564         /* An exec changes our domain. We are no longer part of the thread
 565            group */
 566
 567         current->self_exec_id++;
 568
 569         flush_signal_handlers(current);
 570         flush_old_files(current->files);
 571
 572         return 0;
 573
 574 mmap_failed:
 575 flush_failed:
 576         spin_lock_irq(&current->sigmask_lock);
 577         if (current->sig != oldsig)
 578                 kfree(current->sig);
 579         current->sig = oldsig;
 580         spin_unlock_irq(&current->sigmask_lock);
 581         return retval;
 582 }
 583
 584 /*
 585  * We mustn't allow tracing of suid binaries, unless
 586  * the tracer has the capability to trace anything..
 587  */
 588 static inline int must_not_trace_exec(struct task_struct * p)
 589 {
 590         return (p->ptrace & PT_PTRACED) && !cap_raised(p->p_pptr->cap_effective, CAP_SYS_PTRACE);
 591 }
 592
 593 /*
 594  * Fill the binprm structure from the inode.
 595  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 596  */
 597 int prepare_binprm(struct linux_binprm *bprm)
 598 {
 599         int mode;
 600         int id_change,cap_raised;
 601         struct inode * inode = bprm->file->f_dentry->d_inode;
 602
 603         mode = inode->i_mode;
 604         /* Huh? We had already checked for MAY_EXEC, WTF do we check this? */
 605         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 606                 return -EACCES;
 607
 608         bprm->e_uid = current->euid;
 609         bprm->e_gid = current->egid;
 610         id_change = cap_raised = 0;
 611
 612         /* Set-uid? */
 613         if (mode & S_ISUID) {
 614                 bprm->e_uid = inode->i_uid;
 615                 if (bprm->e_uid != current->euid)
 616                         id_change = 1;
 617         }
 618
 619         /* Set-gid? */
 620         /*
 621          * If setgid is set but no group execute bit then this
 622          * is a candidate for mandatory locking, not a setgid
 623          * executable.
 624          */
 625         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 626                 bprm->e_gid = inode->i_gid;
 627                 if (!in_group_p(bprm->e_gid))
 628                         id_change = 1;
 629         }
 630
 631         /* We don't have VFS support for capabilities yet */
 632         cap_clear(bprm->cap_inheritable);
 633         cap_clear(bprm->cap_permitted);
 634         cap_clear(bprm->cap_effective);
 635
 636         /*  To support inheritance of root-permissions and suid-root
 637          *  executables under compatibility mode, we raise all three
 638          *  capability sets for the file.
 639          *
 640          *  If only the real uid is 0, we only raise the inheritable
 641          *  and permitted sets of the executable file.
 642          */
 643
 644         if (!issecure(SECURE_NOROOT)) {
 645                 if (bprm->e_uid == 0 || current->uid == 0) {
 646                         cap_set_full(bprm->cap_inheritable);
 647                         cap_set_full(bprm->cap_permitted);
 648                 }
 649                 if (bprm->e_uid == 0)
 650                         cap_set_full(bprm->cap_effective);
 651         }
 652
 653         /* Only if pP' is _not_ a subset of pP, do we consider there
 654          * has been a capability related "change of capability".  In
 655          * such cases, we need to check that the elevation of
 656          * privilege does not go against other system constraints.
 657          * The new Permitted set is defined below -- see (***). */
 658         {
 659                 kernel_cap_t permitted, working;
 660
 661                 permitted = cap_intersect(bprm->cap_permitted, cap_bset);
 662                 working = cap_intersect(bprm->cap_inheritable,
 663                                         current->cap_inheritable);
 664                 working = cap_combine(permitted, working);
 665                 if (!cap_issubset(working, current->cap_permitted)) {
 666                         cap_raised = 1;
 667                 }
 668         }
 669
 670         if (id_change || cap_raised) {
 671                 /* We can't suid-execute if we're sharing parts of the executable */
 672                 /* or if we're being traced (or if suid execs are not allowed)    */
 673                 /* (current->mm->mm_users > 1 is ok, as we'll get a new mm anyway)   */
 674                 if (IS_NOSUID(inode)
 675                     || must_not_trace_exec(current)
 676                     || (atomic_read(&current->fs->count) > 1)
 677                     || (atomic_read(&current->sig->count) > 1)
 678                     || (atomic_read(&current->files->count) > 1)) {
 679                         if (id_change && !capable(CAP_SETUID))
 680                                 return -EPERM;
 681                         if (cap_raised && !capable(CAP_SETPCAP))
 682                                 return -EPERM;
 683                 }
 684         }
 685
 686         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 687         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 688 }
 689
 690 /*
 691  * This function is used to produce the new IDs and capabilities
 692  * from the old ones and the file's capabilities.
 693  *
 694  * The formula used for evolving capabilities is:
 695  *
 696  *       pI' = pI
 697  * (***) pP' = (fP & X) | (fI & pI)
 698  *       pE' = pP' & fE          [NB. fE is 0 or ~0]
 699  *
 700  * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
 701  * ' indicates post-exec(), and X is the global 'cap_bset'.
 702  */
 703
 704 void compute_creds(struct linux_binprm *bprm)
 705 {
 706         kernel_cap_t new_permitted, working;
 707
 708         new_permitted = cap_intersect(bprm->cap_permitted, cap_bset);
 709         working = cap_intersect(bprm->cap_inheritable,
 710                                 current->cap_inheritable);
 711         new_permitted = cap_combine(new_permitted, working);
 712
 713         /* For init, we want to retain the capabilities set
 714          * in the init_task struct. Thus we skip the usual
 715          * capability rules */
 716         if (current->pid != 1) {
 717                 current->cap_permitted = new_permitted;
 718                 current->cap_effective =
 719                         cap_intersect(new_permitted, bprm->cap_effective);
 720         }
 721
 722         /* AUD: Audit candidate if current->cap_effective is set */
 723
 724         current->suid = current->euid = current->fsuid = bprm->e_uid;
 725         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 726         if (current->euid != current->uid || current->egid != current->gid ||
 727             !cap_issubset(new_permitted, current->cap_permitted))
 728                 current->dumpable = 0;
 729
 730         current->keep_capabilities = 0;
 731 }
 732
 733
 734 void remove_arg_zero(struct linux_binprm *bprm)
 735 {
 736         if (bprm->argc) {
 737                 unsigned long offset;
 738                 char * kaddr;
 739                 struct page *page;
 740
 741                 offset = bprm->p % PAGE_SIZE;
 742                 goto inside;
 743
 744                 while (bprm->p++, *(kaddr+offset++)) {
 745                         if (offset != PAGE_SIZE)
 746                                 continue;
 747                         offset = 0;
 748                         kunmap(page);
 749 inside:
 750                         page = bprm->page[bprm->p/PAGE_SIZE];
 751                         kaddr = (char *)kmap(page);
 752                 }
 753                 kunmap(page);
 754                 bprm->argc--;
 755         }
 756 }
 757
 758 /*
 759  * cycle the list of binary formats handler, until one recognizes the image
 760  */
 761 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
 762 {
 763         int try,retval=0;
 764         struct linux_binfmt *fmt;
 765 #ifdef __alpha__
 766         /* handle /sbin/loader.. */
 767         {
 768             struct exec * eh = (struct exec *) bprm->buf;
 769             struct linux_binprm bprm_loader;
 770
 771             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
 772                 (eh->fh.f_flags & 0x3000) == 0x3000)
 773             {
 774                 int i;
 775                 char * dynloader[] = { "/sbin/loader" };
 776                 struct file * file;
 777
 778                 allow_write_access(bprm->file);
 779                 fput(bprm->file);
 780                 bprm->file = NULL;
 781
 782                 bprm_loader.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 783                 for (i = 0 ; i < MAX_ARG_PAGES ; i++)   /* clear page-table */
 784                     bprm_loader.page[i] = NULL;
 785
 786                 file = open_exec(dynloader[0]);
 787                 retval = PTR_ERR(file);
 788                 if (IS_ERR(file))
 789                         return retval;
 790                 bprm->file = file;
 791                 bprm->loader = bprm_loader.p;
 792                 retval = prepare_binprm(bprm);
 793                 if (retval<0)
 794                         return retval;
 795                 /* should call search_binary_handler recursively here,
 796                    but it does not matter */
 797             }
 798         }
 799 #endif
 800         for (try=0; try<2; try++) {
 801                 read_lock(&binfmt_lock);
 802                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 803                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
 804                         if (!fn)
 805                                 continue;
 806                         if (!try_inc_mod_count(fmt->module))
 807                                 continue;
 808                         read_unlock(&binfmt_lock);
 809                         retval = fn(bprm, regs);
 810                         if (retval >= 0) {
 811                                 put_binfmt(fmt);
 812                                 allow_write_access(bprm->file);
 813                                 if (bprm->file)
 814                                         fput(bprm->file);
 815                                 bprm->file = NULL;
 816                                 current->did_exec = 1;
 817                                 return retval;
 818                         }
 819                         read_lock(&binfmt_lock);
 820                         put_binfmt(fmt);
 821                         if (retval != -ENOEXEC)
 822                                 break;
 823                         if (!bprm->file) {
 824                                 read_unlock(&binfmt_lock);
 825                                 return retval;
 826                         }
 827                 }
 828                 read_unlock(&binfmt_lock);
 829                 if (retval != -ENOEXEC) {
 830                         break;
 831 #ifdef CONFIG_KMOD
 832                 }else{
 833 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
 834                         char modname[20];
 835                         if (printable(bprm->buf[0]) &&
 836                             printable(bprm->buf[1]) &&
 837                             printable(bprm->buf[2]) &&
 838                             printable(bprm->buf[3]))
 839                                 break; /* -ENOEXEC */
 840                         sprintf(modname, "binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
 841                         request_module(modname);
 842 #endif
 843                 }
 844         }
 845         return retval;
 846 }
 847
 848
 849 /*
 850  * sys_execve() executes a new program.
 851  */
 852 int do_execve(char * filename, char ** argv, char ** envp, struct pt_regs * regs)
 853 {
 854         struct linux_binprm bprm;
 855         struct file *file;
 856         int retval;
 857         int i;
 858
 859         file = open_exec(filename);
 860
 861         retval = PTR_ERR(file);
 862         if (IS_ERR(file))
 863                 return retval;
 864
 865         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
 866         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
 867
 868         bprm.file = file;
 869         bprm.filename = filename;
 870         bprm.sh_bang = 0;
 871         bprm.loader = 0;
 872         bprm.exec = 0;
 873         if ((bprm.argc = count(argv, bprm.p / sizeof(void *))) < 0) {
 874                 allow_write_access(file);
 875                 fput(file);
 876                 return bprm.argc;
 877         }
 878
 879         if ((bprm.envc = count(envp, bprm.p / sizeof(void *))) < 0) {
 880                 allow_write_access(file);
 881                 fput(file);
 882                 return bprm.envc;
 883         }
 884
 885         retval = prepare_binprm(&bprm);
 886         if (retval < 0)
 887                 goto out;
 888
 889         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
 890         if (retval < 0)
 891                 goto out;
 892
 893         bprm.exec = bprm.p;
 894         retval = copy_strings(bprm.envc, envp, &bprm);
 895         if (retval < 0)
 896                 goto out;
 897
 898         retval = copy_strings(bprm.argc, argv, &bprm);
 899         if (retval < 0)
 900                 goto out;
 901
 902         retval = search_binary_handler(&bprm,regs);
 903         if (retval >= 0)
 904                 /* execve success */
 905                 return retval;
 906
 907 out:
 908         /* Something went wrong, return the inode and free the argument pages*/
 909         allow_write_access(bprm.file);
 910         if (bprm.file)
 911                 fput(bprm.file);
 912
 913         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 914                 struct page * page = bprm.page[i];
 915                 if (page)
 916                         __free_page(page);
 917         }
 918
 919         return retval;
 920 }
 921
 922 void set_binfmt(struct linux_binfmt *new)
 923 {
 924         struct linux_binfmt *old = current->binfmt;
 925         if (new && new->module)
 926                 __MOD_INC_USE_COUNT(new->module);
 927         current->binfmt = new;
 928         if (old && old->module)
 929                 __MOD_DEC_USE_COUNT(old->module);
 930 }
 931
 932 int do_coredump(long signr, struct pt_regs * regs)
 933 {
 934         struct linux_binfmt * binfmt;
 935         char corename[6+sizeof(current->comm)];
 936         struct file * file;
 937         struct inode * inode;
 938
 939         lock_kernel();
 940         binfmt = current->binfmt;
 941         if (!binfmt || !binfmt->core_dump)
 942                 goto fail;
 943         if (!current->dumpable || atomic_read(&current->mm->mm_users) != 1)
 944                 goto fail;
 945         current->dumpable = 0;
 946         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
 947                 goto fail;
 948
 949         memcpy(corename,"core.", 5);
 950 #if 0
 951         memcpy(corename+5,current->comm,sizeof(current->comm));
 952 #else
 953         corename[4] = '\0';
 954 #endif
 955         file = filp_open(corename, O_CREAT | 2 | O_TRUNC | O_NOFOLLOW, 0600);
 956         if (IS_ERR(file))
 957                 goto fail;
 958         inode = file->f_dentry->d_inode;
 959         if (inode->i_nlink > 1)
 960                 goto close_fail;        /* multiple links - don't dump */
 961
 962         if (!S_ISREG(inode->i_mode))
 963                 goto close_fail;
 964         if (!file->f_op)
 965                 goto close_fail;
 966         if (!file->f_op->write)
 967                 goto close_fail;
 968         if (!binfmt->core_dump(signr, regs, file))
 969                 goto close_fail;
 970         unlock_kernel();
 971         filp_close(file, NULL);
 972         return 1;
 973
 974 close_fail:
 975         filp_close(file, NULL);
 976 fail:
 977         unlock_kernel();
 978         return 0;
 979 }